A2 Complex Analysis 2024

A2.
2: Complex Analysis
Dmitry Belyaev1
September 30, 2024
1
These notes are closely based on previous versions of the notes by Kevin
McGerty, Ben Green, Panos Papazoglou and Richard Earl. Many thanks to them
all. The syllabus of the course has been changed since 2024. Please contact Dmitry
Belyaev belyaev@maths.ox.ac.uk if you have any comments or corrections.
2
Contents
Foreword iii
0.1 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
0.2 Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Introduction v
0.3 Basic notations . . . . . . . . . . . . . . . . . . . . . . . . . . v
1 Complex differentiability 1
1.1 Complex differentiability . . . . . . . . . . . . . . . . . . . . . 2
1.2 Cauchy-Riemann equations . . . . . . . . . . . . . . . . . . . 4
1.2.1 Wirtinger derivatives . . . . . . . . . . . . . . . . . . . 8
1.3 Harmonic functions . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Power series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.1 Power series about other points . . . . . . . . . . . . . 14
1.4.2 The exponential and trigonometric functions . . . . . 14
1.4.3 Properties of exponential and trigonometric functions 16
1.4.4 Logarithms and powers . . . . . . . . . . . . . . . . . 17
1.5 Branch cuts and multifunctions . . . . . . . . . . . . . . . . . 20
2 Paths and Integration 27

2.1 Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Integration along a path . . . . . . . . . . . . . . . . . . . . . 30
2.3 Cauchy’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4 Deformation theorem and homotopy . . . . . . . . . . . . . . 45
2.5 Winding numbers . . . . . . . . . . . . . . . . . . . . . . . . . 48
3 Cauchy’s Formula and its applications 55

3.1 Cauchy’s Integral Formula . . . . . . . . . . . . . . . . . . . . 55
3.2 Applications of the Integral Formula . . . . . . . . . . . . . . 58
3.3 The identity theorem . . . . . . . . . . . . . . . . . . . . . . . 62
i
ii CONTENTS
3.4 Isolated singularities . . . . . . . . . . . . . . . . . . . . . . . 64

3.5 The argument principle . . . . . . . . . . . . . . . . . . . . . 73
3.6 Application of the Residue theorem . . . . . . . . . . . . . . . 76
3.6.1 On the computation of residues . . . . . . . . . . . . 77
3.6.2 Residue Calculus . . . . . . . . . . . . . . . . . . . . . 80
3.6.3 Jordan’s Lemma and applications . . . . . . . . . . . . 82
3.6.4 Summation of infinite series . . . . . . . . . . . . . . . 86
3.6.5 Keyhole contours . . . . . . . . . . . . . . . . . . . . . 88
Appendices 91
A On the homotopy and homology versions of Cauchy’s theo-

rem 95
B Remark on the Inverse Function Theorem 101

Foreword
The aim of this part of the course is to study functions f : C → C, asking

what it means for them to be differentiable, how to integrate them, and
looking at the applications of all this. We will see that complex differen-
tiation is only superficially similar to the real differentiation and complex
differentiable function have a lot of unique properties. One of the main
goals of this course is to explore a beautiful and very unusual theory of such
functions.
In these lecture notes there will be several parts with the danger sign in
the margin. This sign gives you a warning that this is a more tricky part
and it is easy to make a mistake if you don’t pay enough attention.
I will use the star in the margin to mark several remarks that I think
are very important and give valuable intuition about what is going on.
Finally, there will be several smaller bits that are not part of the syllabus
and not examinable. They will be explicitly marked as non-examinable
material and there will be a marginal sign.
0.1 Synopsis
Complex differentiation. Holomorphic functions. Cauchy-Riemann equa-
tions (different versions). Real and imaginary parts of a holomorphic func-
tion are harmonic. [2 lectures]
Recap on power series and differentiation of power series. Exponential
function and logarithm function. Fractional powers — examples of multi-
functions. The use of cuts as a method of defining a branch of a multifunc-
tion. [2 lectures]
Path integration. Winding numbers. Cauchy’s Theorem (partial proof
only). Homology form of Cauchy’s Theorem (sketch of proof only — stu-
dents referred to various texts for proof.) Fundamental Theorem of Calculus
in the path integral/holomorphic situation. [4 lectures]
iii
iv FOREWORD
Cauchy’s Integral formulae. Taylor expansion. Liouville’s Theorem.

Morera’s Theorem. Identity Theorem. [2 lectures]
Laurent’s expansion. Classification of isolated singularities. Calcula-
tion of principal parts, particularly residues. The argument principle and
applications. [3 lectures]
Residue Theorem. Evaluation of integrals by the method of residues
(straightforward examples only but to include the use of Jordan’s Lemma
and simple poles on contour of integration). [3 lectures]
0.2 Further reading

These lecture notes provide all the essential information for the course and
they follow lectures quite closely. On the other hand, there are many ex-
cellent textbooks that discuss the foundations of complex analysis in more
detail and (very importantly!) from slightly different perspectives. If you
are stuck with one of the arguments in these notes I advise you to consult
other sources that could explain it in a different way. There is a huge list of
Complex Analysis textbooks. I do not aim to provide a comprehensive list,
below are just a few textbooks that I like.
Lars V. Ahlfors. Complex analysis. Third Edition. New York: McGraw-
Hill, 1979.
Churchill, Ruel V.; Brown, James Ward Complex variables and applica-
tions. 4th ed. McGraw-Hill Book Company. X, 339 p. (1984).
Conway, John B. Functions of one complex variable. (English) Zbl
0277.30001 Graduate Texts in Mathematics. 11. New York-Heidelberg-
Berlin: Springer-Verlag. xi, 313 p. (1973).
Gamelin, Theodore W. Complex analysis. (English) Zbl 0978.30001 Un-
dergraduate Texts in Mathematics. New York, NY: Springer. xvi, 478 p.
(2001).
Needham, Tristan Visual complex analysis. 25th-anniversary edition.
With a new foreword by Roger Penrose. Oxford: Oxford University Press
675 p. (2023).
Introduction
0.3 Basic notations

This course runs in parallel with the Metric Spaces course. We will use some
notions and a few results from metric spaces.
We can identify C with the plane R2 by taking real and imaginary parts.
Thus we have the correspondence
z ∈ C ←→ (x, y) = (Re z, Im z) ∈ R2 .
In particular, this allows us to treat C as a metric space by introducing
the distance
p
d(z, w) = (Re z − Re w)2 + (Im z − Im w)2
which obviously can be written as
d(z, w) = |z − w|
where | · | is the modulus of a complex number.

Let us write down some basic properties of the modulus |z|. Recall that
iθ
e = cos θ + i sin θ when θ ∈ R. For now, we will take this as the definition
of eiθ , which is more-or-less what was done in Prelims Complex Analysis.
Later on, we will define the complex exponential function ez and link the
two concepts.
Lemma 0.3.1. Let z, w ∈ C. Then
1. |z|2 = z z̄, where z̄ is the complex conjugate of z;
2. If z = reiθ , where r ∈ [0, ∞) and θ ∈ R, then |z| = r;
3. |zw| = |z||w|.
v
vi INTRODUCTION
Proof. (1) If z = a + ib then z z̄ = (a + ib)(a − ib) = a2 + b2 .

(2) We have z = r cos θ + ir sin θ and so
p
|z| = r2 cos2 θ + r2 sin2 θ = r.
(3) One can calculate directly, writing z = a + ib and w = c + id. Alter-

natively, write z = reiθ , w = r′ eiα , and then observe that zw = rr′ ei(θ+α)
and use (2).
We will also use some basic topological notions that will be discussed
more extensively in the Metric Spaces.
Definition 0.3.2. We say that U ⊂ C is open if for every z ∈ U there is

some r > 0 such that an open ball B(z, ε) := {w ∈ C : |z − w| < ε} ⊂ U .
Any open set U containing z is called a neighbourhood of z.
In complex analysis, it is often convenient to work with connected open

sets, and these are called domains.
Definition 0.3.3. A connected open subset D ⊆ C of the complex plane

will be called a domain.
Chapter 1
Complex differentiability
Now we come to a crucial part in the course – the discussion of what it

means for a function f : C → C to be differentiable. We begin with a quick
refresher on limits, the material that may be found (in the real case) in
Prelims.
Suppose that a ∈ C, and that U is a neighbourhood of a. That is, U
contains some ball B(a, r), r > 0, but U itself need not be open. Suppose
that f : U \ {a} → C is a function: that is, f is defined on U , except not at
a. Then we say that limz→a f (z) = L if the following is true: for all ε > 0,
there is some δ > 0 such that if 0 < |z − a| < δ then |F (z) − L| < ε (and we
assume δ < η so that F is defined when |z − a| < δ).
Sometimes it is convenient to write f in terms of its real and imaginary
parts f (z) = u(z) + iv(z), where u and v are real-valued functions. From
the definition of the modulus it is easy to see that limz→a f (z) = L if and
only if limz→a u(z) = Re L and limz→a v(z) = Im L.
Remark 1.0.1. Similarly, z = x + iy → z0 = x0 + iy0 if and only if x → x0
and y → y0 . But you have to be very careful with the order of limits. In
general,
lim f (x + iy), lim lim f (x + iy) and lim lim f (x + iy)

x→x0 ,y→y0 x→x0 y→y0 y→y0 x→x0
are three different limits. If the first limit exists, then two others exist as
well and have the same value, but this is the only connection.
For example, let us consider f (z) = f (x + iy) = xy/(x + y) and z0 = 0.
It is easy to see that
lim lim f (x + iy) = lim lim f (x + iy) = 0

x→0 y→0 y→0 x→0
1
2 CHAPTER 1. COMPLEX DIFFERENTIABILITY
but
lim f (z)
z→0
does not exist.
1.1 Complex differentiability

With the relevant notions of limit having been recalled, we can give the
definition of a (complex) derivative. In fact, it is the same as the definition
of real derivative, but with complex numbers in place of reals.
Definition 1.1.1 (Complex differentiability). Let a ∈ C, and suppose that
f : U → C is a function, where U ⊂ C is an open set containing a. In
particular, f is defined on some ball B(a, r). Then we say that f is (complex)
differentiable at a if
f (z) − f (a)
lim
z→a z−a
′
exists. If the limit exists, we write f (a) for it and call this the derivative of
f at a.
Since we will be talking exclusively about functions on C, we just use the
terms differentiable/derivative and omit the word ‘complex’. The following
lemma collects the basic facts about derivatives. We omit the proof, which
is essentially identical to the real case.
Lemma 1.1.2. Let a ∈ C, let U be a neighbourhood of a and let f, g : U →
C.
1. (Sums, products) If f, g are differentiable at a then f + g and f g

are differentiable at a and (f + g)′ (a) = f ′ (a) + g ′ (a), (f g)′ (a) =
f ′ (a)g(a) + f (a)g ′ (a).
2. (Quotients) If f, g are differentiable at a and g(a) ̸= 0 then f /g is

differentiable at a and
f ′ (a)g(a) − f (a)g ′ (a)
(f /g)′ (a) = .
g(a)2
3. (Chain rule) If U and V are open subsets of C and f : V → U and

g : U → C are functions, where f is differentiable at a ∈ V and g is
differentiable at f (a) ∈ U , then g ◦ f is differentiable at a with
(g ◦ f )′ (a) = g ′ (f (a))f ′ (a).

1.1. COMPLEX DIFFERENTIABILITY 3
Just as in the real case, the basic rules of differentiation stated above
allow one to check that polynomial functions are differentiable: using the
product rule and induction one sees that z n has derivative nz n−1 for all
n ≥ 0 (as a constant obviously has derivative 0, and f (z) = z has derivative
1). Then by linearity it follows every polynomial is differentiable.
Just as in the real-variable case (Prelims Analysis II) one can formulate
complex differentiability in the following form, which is in fact the better
form to use in most instances including the proof of Lemma 1.1.2.
Lemma 1.1.3. Let a ∈ C, let U be a neighbourhood of a and let f : U → C.

Then f is differentiable at a, with derivative f ′ (a), if and only if we have
f (z) = f (a) + f ′ (a)(z − a) + ε(z)(z − a),
where ε(z) → 0 as z → a.
It is an easy exercise to check that this definition is indeed equivalent to

(really just a reformulation of) the previous one.
Finally, we give an important definition.
Definition 1.1.4 (Holomorphic function). Let U ⊆ C be an open set (for

example, a domain). Let f : U → C be a function. If f is complex differen-
tiable at every a ∈ U , we say that f is holomorphic on U .
Sometimes one says that f is holomorphic at a point a; this means that

there is some open set U containing a on which f is holomorphic.
Remark 1.1.5. Some authors prefer to use the term analytic instead of holo-
morphic. Strictly speaking, an analytic function is a function f that is
infinitely differentiable and its Taylor series converges to f . This looks like
a much stronger property, but later in this course we will see that these two
properties are equivalent.
This is the first time we can see something very special about complex
analysis. In real analysis, there are functions that are differentiable but not
twice differentiable, that are twice differentiable but not thrice etc. There are
even functions that are infinitely differentiable but not analytic. So in the
real analysis classes of differentiable, twice differentiable, thrice differen-
tiable, . . . , infinitely differentiable and analytic functions are all different.
In complex analysis they are all the same.
1.2 Cauchy-Riemann equations

A function from C to C may also be thought of as a function from R2 to R2 ,
and it is useful to study what differentiability means in this language and
compare it with our notion of complex differentiability.
Let z0 ∈ C, and let U be a neighbourhood of z0 . Let f : U → C be
a function. We abuse notation and identify C ∼ = R2 in the usual way, and
identify z0 with (x0 , y0 ) (thus z0 = x0 + iy0 ). Then (again with some abuse
of notation) we may think of U as an open subset of R2 and write f = (u, v),
where u, v : R2 → R (the letters u, v are quite traditional in this context,
and sometimes we call these the components of f ). Another way to think of
this is that f (z) = f (x + iy) = f (x, y) = u(x, y) + iv(x, y).
Example 1.2.1. Consider the function f (z) = z 2 (which is holomorphic on
all of C). Since (x + iy)2 = (x2 − y 2 ) + 2ixy, we see that the components of
f are given by u(x, y) = x2 − y 2 , v(x, y) = 2xy.
In the Metric Spaces course we have defined the notion of partial deriva-
tives of a function u : R2 → R
u(a1 + h, a2 ) − u(a1 , a2 )
∂x u(a) := lim
h→0 h
(if the limit exists) and
u(a1 , a2 + k) − u(a1 , a2 )
∂y u(a) := lim .
k→0 k
It is important to note that h, k in these limits are real. We define partial
derivatives of v in the same way.
An important fact is that if f is differentiable then these partial deriva-
tives do exist, and moreover, they are subject to a constraint.
Theorem 1.2.2 (Cauchy-Riemann equations). Let z0 ∈ C, let U be a neigh-
bourhood of z0 , and let f : U → C be a function which is complex differ-
entiable at z0 . Let u, v : R2 → R be the components of f . Then the four
partial derivatives ∂x u, ∂y u, ∂x v, ∂y v exist at z0 . Moreover, we have the
Cauchy-Riemann equations
∂x u = ∂y v, ∂x v = −∂y u, (1.2.1)
and f ′ (z0 ) = ∂x u(z0 ) + i∂x v(z0 ).

Remark 1.2.3. By the Cauchy-Riemann equations, there are in fact four
different expressions for f ′ (z0 ) using the partial derivatives.
1.2. CAUCHY-RIEMANN EQUATIONS 5
Remark 1.2.4. The important point to take away from Theorem 1.2.2 is that
a complex differentiable function is much more than simply a pair of real
differentiable functions. For instance, the function f (z) = Re z. For this
function u(x, y) = x and v(x, y) = 0. This function is as differentiable as
one could wish for from the real point of view, but it is not a complex differ-
entiable function since the Cauchy-Riemann equations fail to hold. Indeed,
∂x u = 1 ̸= 0 = ∂y v.
Exercise 1.2.5. Use the definition of complex differentiability to verify
directly that the function f (z) = Re z is not differentiable anywhere.
Proof. By the definition of complex differentiability
f (z) − f (z0 )
f ′ (z0 ) = lim .
z→z0 z − z0
In particular, this shows that the limit exists and is the same if we consider
a particular way in which z approaches z0 . First, we consider z = z0 + h
where h is real and h → 0. Then we can see that
u(x0 + h, y0 ) + iv(x0 + h, y0 ) − u(x0 , y0 ) − iv(x0 , y0 )
f ′ (z0 ) = lim
h→0 h
u(x0 + h, y0 ) − u(x0 , y0 ) v(x0 + h, y0 ) − v(x0 , y0 )
= lim + i lim
h→0 h h→0 h
= ∂x u(z0 ) + i∂x v(z0 ).
In the second line we used that h is real and that a limit of a complex
function exists if and only if limits of its real and imaginary parts exist
and are equal to the real and imaginary parts of the function’s limit. In
particular, this proves that ∂x u and ∂x v exist and
f ′ (z0 ) = ∂x u(z0 ) + i∂x v(z0 ).
Next, we consider a different way to approach z0 . Let us consider z =
z0 + ih where h is real and h → 0. As before
u(x0 , y0 + h) + iv(x0 , y0 + h) − u(x0 , y0 ) − iv(x0 , y0 )
f ′ (z0 ) = lim
h→0 ih
u(x0 , y0 + h) − u(x0 , y0 ) v(x0 , y0 + h) − v(x0 , y0 )
= −i lim + lim
h→0 h h→0 h
= −i∂y u(z0 ) + ∂y v(z0 ).
Here, we use essentially the same argument and the fact that 1/i = −i. This
proves that ∂y u and ∂y v exist and
f ′ (z0 ) = −i∂y u(z0 ) + ∂x v(z0 ).
Combining two different expressions for f ′ (z0 ) we get

∂x u(z0 ) + i∂x v(z0 ) = −i∂y u(z0 ) + ∂x v(z0 ).
Equating real and imaginary parts of both sides and using that the partial
derivatives are real we get the Cauchy-Riemann equations.
There is an alternative way of proving Theorem 1.2.2 which also explains

the difference between real and complex differentiability.
Alternative proof of Theorem 1.2.2. First, we show that the function f is

real-differentiable. We know that complex differentiability implies that
f (z) = f (z0 ) + f ′ (z0 )(z − z0 ) + ϵ(z)(z − z0 ), (1.2.2)
where ϵ(z) → 0 as z → z0 . Next, we want to rewrite this expression in real
terms, i.e. identifying C with R2 . Writing f ′ (z0 ) = reiθ = r cos θ + ir sin θ
we can rewrite the derivative term f ′ (z0 )(z − z0 ) as

r cos θ(x − x0 ) − r sin θ(y − y0 ) r cos θ −r sin θ x − x0
= .
r cos θ(y − y0 ) + r sin θ(x − x0 ) r sin θ r cos θ y − y0
(1.2.3)
Hence this term is just a linear transformation given by a simple matrix.
We can see that geometrically this linear transformation is very simple. It
is just a rotation by angle θ and scaling by factor r. This should not be
surprising since multiplication by a complex number reiθ is equivalent to
multiplying the modulus by r and adding θ to the argument. This is exactly
the scaling by r and rotation by θ.
Recall from the Matric Spaces course that a function f : R2 → R2 is real
differentiable at z0 = (x0 , y0 ) if there is a linear function L = dfz0 : R2 → R2
such that
f (z) = f (z0 ) + L(z − z0 ) + R(z)
where |R(z)|/|z − z0 | → 0 as z → z0 and L in the standard basis (that
conveniently coincide with our use of real and imaginary parts) is given by
the matrix
∂x u(x0 , y0 ) ∂y u(x0 , y0 )
∂x v(x0 , y0 ) ∂y v(x0 , y0 )
where ∂x u, ∂y u, ∂x v and ∂y v are partial derivatives of u and v.
Comparing this with formulas (1.2.2) and (1.2.3) we can see that f is
indeed real differentiable, partial derivatives of u and v exist and

∂x u(x0 , y0 ) ∂y u(x0 , y0 ) r cos θ −r sin θ
= .
∂x v(x0 , y0 ) ∂y v(x0 , y0 ) r sin θ r cos θ
This immediately gives the Cauchy-Riemann equations and that f ′ (z0 ) =

r cos θ + ir sin θ = ∂x u(x0 , y0 ) + i∂x v(x0 , y0 ).
Remark 1.2.6 (non-examinable). The second proof sheds some light on the
difference between real and complex differentiability. In both cases f (z) −
f (z0 ) can be well approximated by a linear function.But in one case it is real
linear and in the other case it is complex linear. Being complex linear is a
stronger property since the set of scalars is larger. So any complex linear
function is real linear, but not the other way round. A real matrix of any
complex linear transformation has a very special form as in (1.2.3).
The Cauchy-Riemann equations are essentially the only requirement for
complex differentiability.
Theorem 1.2.7. Suppose that U ⊆ C is open and that f : U → C is a
function. Let the components of f be (u, v), where u, v : R2 → R. Suppose
that all four partial derivatives ∂x u, ∂y u, ∂x v, ∂y v exist, are continuous in
U , and satisfy the Cauchy-Riemann equations. Then f is holomorphic on
U with derivative ∂x u + i∂x v.
Proof. The heavy lifting for this theorem has been done in the Metric Spaces
course. Theorem 1.3.1 from the Metric Spaces lecture notes tells us that the
existence of continuous partial derivatives implies that f is real differen-
tiable. So for any z0 ∈ U
f (z) = f (z0 ) + L(z − z0 ) + R(z)
where L = dfz0 is a real linear transformation with the matrix (in the stan-
dard basis)
∂x u(x0 , y0 ) ∂y u(x0 , y0 )
∂x v(x0 , y0 ) ∂y v(x0 , y0 )
and |R|/|z − z0 | → 0. Using the Cauchy-Riemann equation it can be rewrit-
ten as
∂x u(x0 , y0 ) −∂x v(x0 , y0 )
.
∂x v(x0 , y0 ) ∂x u(x0 , y0 )
As we have discussed in the second proof of Theorem 1.2.2 this matrix cor-
responds to the complex multiplication by ∂x u(x0 , y0 ) + i∂x v(x0 , y0 ). Hence
f (z) = f (z0 ) + (∂x u(x0 , y0 ) + i∂x v(x0 , y0 ))(z − z0 ) + R(z).
This is exactly the formula from Lemma 1.1.3 which shows that f is complex
differentiable and f ′ (z0 ) = ∂x u(x0 , y0 ) + i∂x v(x0 , y0 ).
The assumption that function is continuously differentiable is important.

Otherwise, we could have a function such that its partial derivatives satisfy
the Cauchy-Riemann ‘by coincidence’ and this is not enough to imply dif-
ferentiability.
Example 1.2.8. Let f (z) = xy/(x2 + y 2 ) for z ̸= 0 and define f (0) = 0.

Then u = f and v = 0. It is easy to see that all partial derivatives at z0 = 0
vanish (since f = 0 along both axes) so they satisfy the Cauchy-Riemann
equations there. On the other hand, this function is not differentiable at 0,
it is not even continuous there.
Remark 1.2.9. Complex conjugation is a very natural and useful transfor-

mation of C but it completely ruins differentiability. Let f be differentiable
at z0 and that f ′ (z0 ) ̸= 0. Define g(z) = f¯(z) and h(z) = f (z̄). Then g is
not differentiable at z0 and h is not differentiable at z̄0 . This can be checked
by direct computation of partial derivatives and checking that they do not
satisfy the Cauchy-Riemann equations.
The following basic fact will be established again later in the course in
a different way and in greater generality, but we will need it in this section
when establishing the basic properties of the exponential function.
Lemma 1.2.10. Suppose f : C → C is holomorphic and that f ′ is identically

zero. Then f is constant.
Proof. Let the components of f be (u, v). By Theorem 1.2.2, the partial
derivative ∂x u exists and is zero. This means that, for fixed y, the function
x 7→ u(x, y) is differentiable with derivative zero. By the real-variable version
of the lemma we are trying to prove (which is a simple consequence of the
mean value theorem) we see that u(x, y) is constant as a function of x,
for fixed y. Similarly, since ∂y u exists and is zero, u(x, y) is constant as
a function of y, for fixed x. Therefore, for arbitrary (x, y) and (x′ , y ′ ) we
have u(x, y) = u(x′ , y) = u(x′ , y ′ ), which means that u is constant. By an
identical argument, v is constant.
1.2.1 Wirtinger derivatives

Although we will not use them again in this course, we briefly mention
another way to state the Cauchy-Riemann equations.
Definition 1.2.11. Let f : U → C be a function with components (u, v),

and suppose that the partial derivatives of these exist. Then we define the
Wirtinger (partial) derivatives by
1 1 1
∂z f := (∂x f − i∂y f ) = (∂x − i∂y ) u + i (∂x − i∂y ) v
2 2 2
and
1 1 1
∂z̄ f := (∂x f + i∂y f ) = (∂x + i∂y ) u + i (∂x + i∂y ) v.
2 2 2
Lemma 1.2.12. Let U be an open subset of C and let f : U → C. Then f
satisfies the Cauchy-Riemann equations if and only if ∂z̄ f = 0, moreover, in
this case f ′ = ∂z f .
Proof. Straightforward calculation.
Remark 1.2.13 (Non-examinable). Wirtinger derivatives are very natural.

Let us write x = (z + z̄)/2 and y = (z − z̄)/(2i) and pretend for a second that
we change the variables1 from (x, y) to (z, z̄). Then we get Wirtinger deriva-
tives from the change of variables formula.
Using this ‘change of variables’ any function of x and y can be written
as a function of z and z̄. Then the Cauchy-Riemann equation could be
informally interpreted as the fact that f depends on z but not on z̄ and
f ′ = ∂z f . Although Wirtinger derivatives are not partial derivatives with
respect to z and z̄, they behave like this for all practical purposes. For
example, let us consider f (z) = x2 + y 2 = |z|2 = z z̄. A simple computation
shows that
1
∂z z z̄ = ∂z f = (2x − i2y) = z̄
2
and
1
∂z̄ z z̄ = ∂z̄ f = (2x + i2y) = z.
2
In light of this, Remark 1.2.9 becomes very natural. If f is differentiable
and g = f¯ then
1 1
∂z̄ g = (∂x g + i∂y g) = ∂x f − i∂y f = ∂z f = f¯′ .
2 2
So, unless f ′ = 0, function g is not holomorphic. On the other hand, it satis-
fies a different equation ∂z g = 0. Such functions are called anti-holomorphic
functions.
Finally, it is very important to remember that although this remark gives
valuable intuition, it is not rigorous and can not be used in proofs.
1
This is not really true since as complex variables z and z̄ are not independent variables
1.3 Harmonic functions

In this brief section, we introduce the notion of a harmonic function and the
relation of this concept to complex differentiability. We will return to this
in much more detail later in the course.
We begin with the basic definitions.
Definition 1.3.1. Suppose that u : R2 → R is a function on some open

set U ⊆ R2 which is twice differentiable (that is, the partial derivatives
themselves have partial derivatives). Then we define the Laplacian ∆u =
∂xx u + ∂yy u, where ∂xx u = ∂x (∂x u) = ∂x2 u and similarly for ∂yy u.
Definition 1.3.2. Suppose that u : R2 → R is a function on some open

set U ⊆ R2 which is twice differentiable. Then we say that u is harmonic if
∆u = 0.
The reason for introducing this notion here is the following important
result.
Theorem 1.3.3. Let U ⊆ C be open, and suppose that f : U → C is

holomorphic. Let the components of f be (u, v), and suppose that they are
both twice continuously differentiable. Then u and v are harmonic.
Proof. From the Cauchy-Riemann equations,
∂xx u = ∂xy v (= ∂x ∂y v), ∂yy u = −∂yx v.
However, one knows (Prelims) that under the stated conditions we have the
symmetry property of partial derivatives
∂xy v = ∂yx v,
and the result follows.
Let us make some further comments on this result:
• We will show later in the course that a holomorphic function such as

f is in fact infinitely (complex) differentiable. (This is a rather re-
markable and important fact, not true at all in real-variable analysis.)
Therefore the assumption that u, v be twice differentiable is automat-
ically satisfied and can be omitted in the statement of the theorem,
once one has established that result later in the course.
1.4. POWER SERIES 11
• The symmetry of mixed partial derivatives means that we can factorise

∆ = (∂x − i∂y )(∂x + i∂y ). So in terms of Wirtinger derivatives we can
write ∆ = 4∂z ∂z̄ .
• If u, v : R2 → R are harmonic functions such that f (z) = u(z) + iv(z)

is holomorphic, we say that u and v are harmonic conjugates.
• [Non-examinable] It can be shown that if u is harmonic in a simply

connected domain i.e. a domain without holes (we will discuss it in
more detail later), then it has a harmonic conjugate v and so it is a
real part of a holomorphic functions.
1.4 Power series

In this section we look at the power series of a complex variable. Much of
the theory parallels the real-variable theory as seen in Prelims Analysis II
and the proofs go over verbatim. For the most part we will omit them.
A (formal) power series is really just a sequence (an )∞ n=0 of complex num-
bers,
P∞ but we call it a power series because we are interested in understanding
a z n . A priori, however, this sum may not converge for even a single
n=0 n
nonzero z; nonetheless, it is conventional to write ∞ n
P
n=0 an z , rather than
be technically formal and correct and refer to the sequence (an )∞ n=0 .
P∞ n converges at a point z if the
We say that a power series a
n=0 n z
Pk n tends to a limit as k → ∞. For such z,
sequence
P∞ of partial sums n=0 a n z
n
n=0 an z makes sense as an actual complex number.
Definition 1.4.1 (Radius of convergence). Let ∞ n

P
n=0 an z be a power se-
ries, and let S be the set of z ∈ C at which it converges. The radius of
convergence of the power series is sup{|z| : z ∈ S}, or ∞ if the set S is
unbounded. Note that S is always nonempty since 0 ∈ S.
The following result is mostly, but not entirely, in Prelims Analysis I.

We will prove it again, albeit at a moderately high speed.
Proposition 1.4.2. Let n=0 an z n be a power series, let S be the subset

P
of C on which it converges and let R be its radius of convergence. Then we
have
B(0, R) ⊆ S ⊆ B̄(0, R). (1.4.1)
The series converges absolutely on B(0, R) and if 0 ≤ r < R then it converges

uniformly on B̄(0, r). Moreover, we have
1
= lim sup |an |1/n . (1.4.2)
R n
Remark. The statement is uncontroversial when 0 < R < ∞. Suitably

interpreted, the proposition makes sense when R = 0 and R = ∞ as well,
and we consider the statement to include these cases:
• When R = 0, one should take B(0, R) = ∅ and B̄(0, R) = {0}, so
(1.4.1) is the statement that S ⊆ {0} in this case (which is trivial).
Statement (1.4.2) should be taken to mean that lim supn |an |1/n = ∞
(which is not so trivial).
• When R = ∞, one should take B(0, R) = B̄(0, R) = C, so (1.4.1) is

the statement that S = C. Statement (1.4.2) should be taken to mean
that limn→∞ |an |1/n = 0.
Proof. We begin with (1.4.1), which was essentially proven in Prelims. The
containment S ⊆ B̄(0, R) is immediate from the definition of the radius of
convergence (even when R = ∞). The other containment B(0, R) ⊆ S, as
well as the statement that the series converges absolutely on B(0, R), are
both consequences of the statement that the series converges
S uniformly on
B̄(0, r) when 0 ≤ r < R. This is because B(0, R) = r<R B̄(0, r) (this is
also true when R = ∞).
Let us, then, prove
P∞this statement. By definition of R, there is some w,
n
|w| > r, such that n=0 an w converges. In particular, the terms of the
sum are bounded: |an wn | ≤ M for some M . But then if |z| ≤ r we have
z r
|an z n | = |an wn || |n ≤ M | |n .
w w
The geometric series n | wrP |n converges, since |w| > r. Therefore, by the
P
Weierstass test (for series) ∞ n
n=0 an z converges uniformly for |z| ≤ r.
Now we turn to the formula (1.4.2), which is not always covered in
Prelims. Suppose the radius of convergence is R. Let 0 ≤ r < R. By the
above, there is some w, |w| > r, such that |an wn | ≤ M for all n. We may
clearly assume that M ̸= 0. Taking nth roots gives |an |1/n |w| ≤ M 1/n . Since
1
M 1/n → 1 as n → ∞, this implies that lim supn |an |1/n ≤ |w| < 1r . Since
r < R was arbitrary, it follows that lim supn |an |1/n ≤ R1 . (This is perfectly
legitimate when R = ∞ as well, with the interpretation that R1 = 0 in this
case.)
In the other direction, suppose that lim supn |an |1/n = L and that L ∈
(0, ∞). If L′ > L, this means that |an |1/n ≤ L′ for all sufficiently large n.
Therefore |an z n | ≤ |L′ z|nP(for sufficiently large n), and by the geometric
series formula the series n an z n converges provided |z| < L1′ . Therefore
R ≥ L1′ . Since L′ > L was arbitrary, R ≥ L1 , that is to say lim supn |an |1/n ≥
1
R.
The argument is valid with minimal changes when L = 0; we have
shown that R > L1′ for all L′ > 0, and so R = ∞, and so the inequality
lim supn |an |1/n ≥ R1 remains true (with the interpretation discussed above).
When L = ∞, the inequality is vacuously true. Putting all this together
concludes the proof.
The next lemma is about sums and products of power series.
Lemma 1.4.3. Let ∞

P n
P∞ n
n=0 an z and n=0 bn z be power series with radii of
convergence R1 and R2 respectively. For |z| < min(R1 , R2 ), write s(z), t(z)
for the functions to which these series converge.
1. The power series ∞ n

P
n=0 (an + bn )z converges in |z| < min(R1 , R2 ), to
s(z) + t(z).
2. The power series ∞ n

P P
n=0 ( k+l=n ak bl )z converges in |z| < min(R1 , R2 ),
to s(z)t(z).
Proof. See Prelims Analysis I Problem Sheet 7 (for the real variable case;
the complex case is the same). Note that min(R1 , R2 ) is only a lower bound
for the radius of convergence in each case – it is easy to find examples where
the actual radius of convergence of the sum or product is strictly larger than
this.
Next, we differentiate power series term by term.

P∞ n be
Proposition 1.4.4 (Differentiation of power series). Let n=0 an z
a power series, with the radius of convergence R. Let s(z) be the func-
tion
P∞ to which this series converges on B(0, R). Then power series t(z) =
na z n−1 also has radius of convergence R and on B(0, R) the power
n=1 n
series s is complex differentiable with s′ (z) = t(z). In particular, a power
series is infinitely complex-differentiable within its radius of convergence.
Proof. This is proved in the real variable case Prelims Analysis II (see The-
orem 8.16 in the current lecture notes); the proof adapts to the complex
case with trivial changes.
1.4.1 Power series about other points

We conclude with some remarks about power series about points z0 other
than 0, which come up frequently in complex analysis.
Such power series are functions given by an expression of the form
∞
X
f (z) = an (z − z0 )n .
n=0
All the results we have shown above immediately extend to these more
general power series, since if
∞
X
g(z) = an z n ,
n=0
then the function f is obtained from g simply by composing with the trans-
lation z 7→ z − z0 . In particular, the chain rule shows that
∞
X
′
f (z) = nan (z − z0 )n−1 .
n=1
Remark 1.4.5. Let f (z) = ∞ n

P
n=0 an (z − z0 ) be a function given by a power
series with positive radiusP of convergence R. Since the radius of convergence
for the derivative f ′ (z) = nan (z −z0 )n−1 is at least R > 0, it shows that f ′
is also holomorphic in B(z0 , R). By induction, all derivatives of f . Moreover,
f (z0 ) = a0 , f ′ (z0 ) = a1 , etc, so
∞
X f (n) (z0 )
f (z) = (z − z0 )n ,
n!
n=0
so f is given by its Taylor series. Such functions are called analytic. Theorem
1.4.4 proves that analytic functions are holomorphic. Later in the course we
will see that all holomorphic functions are analytic.
1.4.2 The exponential and trigonometric functions

With the basic facts about complex power series under our belt, we can
define some of the most important functions in mathematics as functions of
a complex variable.
Example 1.4.6. The functions

∞
X zn
ez = exp(z) = ,
n!
n=0
∞
X z 2n
cos(z) = (−1)n ,
(2n)!
n=0
and
∞
X z 2n+1
sin(z) = (−1)n
(2n + 1)!
n=0
are all holomorphic on all of C and their derivatives are given by term-by-
term differentiation of the series. In particular,
exp′ = exp, cos′ = − sin, sin′ = cos .
Also
eiz = exp(iz) = cos z + i sin z.
Clearly, when z is real these formulas coincide with definitions of real

exp, sin and cos from Prelims. This is a very natural way to extend a
function from R to C. Not all functions that are infinitely differentiable in
R can be extended to the complex plane.
Example 1.4.7. Consider f (x) = 1/(1 + x2 ). This is a perfectly nice real

function. It is real analytic and for every x0 ∈ R there is r = r(x0 ) > 0 such
that the Taylor series of f at x0 converges to f in B(x0 , r). In particular, these
power series can be used to extend f to some region in C. Of course, in this
particular case, there is a much simpler way to extend our function: we can
just write f (z) = 1/(1 + z 2 ). This function is well defined and holomorphic
in C \ {i, −i}
Remark 1.4.8 (Non-examinable). Sometimes, extensions are not so simple

and it is not obvious that there is a natural maximal domain where the
extension exists and that extensions are unique.There is a rich and impor-
tant theory of analytic extensions, but it is beyond the scope of this course.
Example 1.4.9 (Non-examinable). One of the most important examples of

analytic extension is the Riemann zeta function. For real x > 1 we define
∞
X 1
ζ(x) = .
nx
n=1
It is a standard fact that this series converges for all x > 0. With a little
bit of work2 one can show that this series converges if we replace x by a
complex number z with Re z > 1. This is not a power series, so it is not
obvious that ζ is holomorphic, but it can be shown that it is. Riemann has
shown that this function can be analytically extended to C \ {1}.
1.4.3 Properties of exponential and trigonometric functions

From the definition we can immediately see that
sin(−z) = − sin z, cos(−z) = cos z
and so
eiz + e−iz eiz − e−iz
cos z = , sin z = .
2 2i
We can also see that hyperbolic sine and cosine are very closely related to
sine and cosine. They all defined by similar formulas in terms of exp, the
only difference is in the factor i.
We can also properly understand Euler’s formula eiθ = cos θ + i sin θ for
θ ∈ R. Note also that
The exponential function also satisfies the following extremely important
property.
Proposition 1.4.10. We have exp(z + w) = exp(z) exp(w).
Proof. Fix a ∈ C, and consider the function f (z) = exp(z) exp(a − z).
Differentiating and using the product rule and chain rule, we see that
f ′ (z) = exp(z) exp(a − z) − exp(z) exp(a − z) = 0.
Therefore, by Lemma 1.2.10, f is constant. It follows that
f (z) = f (0) = exp(a),
that is to say
exp(z) exp(a − z) = exp(a).
Substituting a = z + w gives the stated result.
Corollary 1.4.11. For x, y ∈ R we have ex+iy = ex (cos y + i sin y).
2
In particular, we have not yet defined what nz means for complex z
1.4.4 Logarithms and powers

There are several ways to define the real logarithm, but arguably the most
natural one is to say that log is the inverse of exp. Since in the real case
exp is a monotone function that maps (−∞, ∞) to (0, ∞), the logarithm is
uniquely defined for all positive real numbers.
In C things become immediately more complicated. From Corollary
1.4.11 we can see that the range of exp is C \ {0}, but the function is not
one-to-one. If ew = z then ew+2πin = z for any n ∈ Z. In other words, if z
has a logarithm then it has infinitely many3 . It turns out that there is no
canonical choice of logarithm and, worse still, there is no way to define the
logarithm as a holomorphic function on all of C \ {0}. We will pay closer at-
tention to these points in the coming lectures, but for now we record the
following positive results.
The good news is that Corollary 1.4.11 gives us a way to solve the equa-
tion exp(w) = z. By writing w = u + iv and z = |z|(cos θ + i sin θ) we
get
ew = eu (cos(v) + i sin(v)) = r(cos θ + i sin θ) = z
This implies exp(u) = |z|, cos(v) = cos θ and sin(v) = sin θ. Since u is real
we must have u = log |z| and the last two equations imply that that v − θ
must be an integer multiple of 2π.
At this point, there are two possible ways to continue. One is to accept
that the equation exp(w) = z has infinitely many solutions and define the
logarithm as a multifunction (this is like a function but it could have many
values) or to choose one of the solutions and to declare it to be the ‘right
one’. We will explore both possibilities but we will start with the second
one.
First, we will need a simple version of the Inverse Function Theorem.
Later in this course, we will have the full version.
Proposition 1.4.12 (Simplified Inverse function Theorem). Let f : U → C

be a holomorphic function. Let z0 be a point in U and assume that there is
V ⊂ U containing z0 such that f is one-to-one on V → f (V ) = W and the
inverse function g = f −1 : W → V is continuous. Then g is differentiable
at w0 = f (z0 ) and g ′ (w0 ) = 1/f ′ (z0 ).
Remark 1.4.13. The main difference with the usual inverse function theorem
is that here we assume the existence and continuity of the inverse function.
3
This is exactly the same problem that we encounter trying to define sin−1 and cos−1
in R.
In Prelims we have shown that if f (x) is continuously differentiable and

f ′ (x0 ) ̸= 0, then f is locally monotone and this almost immediately implied
that f is locally invertible and the inverse is continuous. The main problem
is that C can not be ordered and monotonicity makes no sense. Eventually,
we will overcome this problem, but it will happen much later in this course.
The good news is that for many functions it is not too difficult to prove
that the inverse exists and is continuous, so Proposition 1.4.12 gives that
the inverse is holomorphic.
Proof. The proof is straightforward and essentially the same as in Prelims.

Let us consider f (z) = w → w0 = f (z0 ) and write
g(w) − g(w0 ) z − z0 1
= → ′ .
w − w0 f (z) − f (z0 ) f (z0 )
Here we used that z → z0 which follows from the continuity of g.
Proposition 1.4.14. Slightly abusing notations we define D = C \ (−∞, 0].

That is, D is the complex plane minus the negative real axis (and 0). Define
the function Log : D → C as follows: if z = |z|eiθ with θ ∈ (−π, π] then set
Log(z) := log |z| + iθ.
Then Log is holomorphic on D and

1
Log′ (z) = .
z
Remark 1.4.15. It might look like a circular argument since we define Log
in terms of log. But in fact, log is a real function that we have defined in
Prelims. So essentially, we define the complex logarithm in terms of the real
logarithm.
Definition 1.4.16. The values θ ∈ (−π, π] such that z = |z|eiθ is called

the principle value of the argument of z and the function Log defined in
Proposition 1.4.14 is called the principle value of the logarithm.
Remark 1.4.17. The notation Log is not universal. Many authors use log
for all versions of the logarithm. We will always use Log for the principle
value and other notations for other versions.
Proof of Proposition 1.4.14. We are going to apply Proposition 1.4.12 and

in order to do so we just have to show that Log is a continuous inverse of
exp. It is an inverse by construction, so we just need to prove continuity.
We know that |z| and log are continuous, so the real part of Log is
continuous, so it remains to show that the argument θ = θ(z) is a continuous
function of z.
By the law of cosines
|z|2 + |z + h|2 − |h|2

cos(θ(z + h) − θ(z)) = .
2|z||z + h|
It is easy to see that for all |z| > 0 this converges to 1 as h → 0. Since z ∈ D
and θ ∈ (−π, π) this implies that θ(z + h) → θ(z). This completes the proof
that Log is continuous.
Note, that it is important that we exclude the negative real line from D.
If we were to include it and define the argument to be π there, then θ would
not be continuous. For example, θ(−1) = π but θ(−1−iϵ) → −π as ϵ → 0+.
By Proposition 1.4.12 Log is differentiable and its inverse is 1/ exp′ (Log(z)) =
1/ exp(Log(z)) = 1/z. Since this argument works for any z ∈ D, this proves
that Log is holomorphic in D and its derivative is 1/z.
Remark 1.4.18. This argument is essentially the same as the proof of the
Inverse Function Theorem from Prelims. In a similar way one can prove a
complex version of this theorem. The main difficulty is in proving that if
f ′ ̸= 0 then the function locally is one-to-one and there is a local inverse. In
Prelims, we have used monotonicity for this, but complex numbers are not
ordered and there is no way to generalise this argument. The result is still
correct, but we will not prove it here.
Remark 1.4.19. There is yet another possibility to prove that Log is holo-
morphic. We can write Log = u + iv where u = (1/2) log(x2 + y 2 ) and
v = arctan(y/x) and check that u and v are continuously differentiable and
satisfy the Cauchy-Riemann equations. Then, by Theorem 1.2.7, Log is
holomorphic. The downside is that the expression for v is correct only for
x > 0, for other parts of D one has to write other similar expressions. This
argument is not very difficult, but a bit has too many technicalities.
Remark 1.4.20. Finally, there is a completely different approach to the def-
inition of complex
Rz logarithm. As in R we can define the logarithm as the
integral of 1 1/w dw, but this requires the notion of the integral and there
are other subtleties so we do not use this approach. On the other hand, by
the end of this course you will see that this approach makes perfect sense.
Since we now have a version of the logarithm, we can define complex
powers as well:
Definition 1.4.21. Let α, z ∈ C and z ̸= 0. Then we define the principle

value of z α as exp(α Log(z)).
It is important to note that these versions of the logarithm and power

inherit many properties from their real counterparts, but not all of them.
For example
−2πi 4πi
Log(e4πi/3 ) = Log(e−2πi/3 ) = ̸= = Log(e2πi/3 ) + Log(e2πi/3 ).
3 3
Similarly, √
√
r
1 1 1
= −1 = i ̸= −i = = √ .
−1 i −1
1.5 Branch cuts and multifunctions

Many functions like logarithm, roots, and inverse trigonometric functions are
defined as functions that are inverse of functions that are not injective. So,
potentially, they could have many values. On the other hand, our definition
of a function does not allow this. One way to overcome this issue is to extend
the definition.
Definition 1.5.1. A multi-valued function or multifunction on a subset

U ⊆ C is a map f : U → P(C) assigning to each point in U a subset4 of the
complex numbers.
Sometimes, like in the previous section with the principle value of the
logarithm or when we choose the positive square root of a positive real
number, we would like to make a unique choice of a value. This is formalized
in the following definition:
Definition 1.5.2. Let : U → P(C) be a multifunction. A branch of f on a

subset V ⊆ U is a function g : V → C such that g(z) ∈ f (z), for all z ∈ V .
If g is continuous (or holomorphic) on V we refer to it as a continuous,
(respectively holomorphic) branch of f .
Definition 1.5.3. Suppose that f : U → P(C) is a multi-valued function

defined on an open subset U of C. We say that z0 ∈ U is not a branch
point of f if there is an open disk5 D ⊆ U containing z0 such that there is
4
We use the notation P(X) to denote the power set of X, that is, the set of all subsets
of X.
5
In fact any simply connected domain – see our discussion of the homotopy form of
Cauchy’s theorem.
1.5. BRANCH CUTS AND MULTIFUNCTIONS 21
a holomorphic branch of f defined on D\{z0 }. We say z0 is a branch point

otherwise. When C\U is bounded, we say that f does not have a branch
point at ∞ if there is a holomorphic branch of f defined on C\B(0, R) ⊆ U
for some R > 0. Otherwise, we say that ∞ is a branch point of f .
Branch cut for a multifunction f is a collection of curves in the plane
on whose complement we can pick a holomorphic branch of f . Thus branch
cuts must contain all the branch points.
Remark 1.5.4. In order to distinguish between multifunctions and functions,

it is sometimes useful to introduce some notation: if we wish to consider
z 7→ z 1/2 as a multifunction, then to emphasize that we mean a multifunction
we will write [z 1/2 ]. Thus [z 1/2 ] = {w ∈ C : w2 = z}. Similarly we write
[log(z)] = {w ∈ C : ew = z}. This is not a uniform convention in the
subject, quite often depending on the context z 1/2 might be a multifunction
or a branch of a multifunction or even the ‘main’ branch.
Example 1.5.5. Consider the square root ‘function’ f (z) = z 1/2 . Unlike
the case of real numbers, every complex number has a square root, but just
as for the real numbers, there are two possibilities unless z = 0. Indeed if
z = x + iy and w = u + iv has w2 = z we see that
u2 − v 2 = x; 2uv = y,
and so p p
2 x+ x2 + y 2 2 −x + x2 + y 2
u = ,v = .
2 2
where the requirement that u2 , v 2 are non-negative determines the signs.
Hence taking square roots we obtain the two possible solutions for w satis-
fying w2 = z. (Note it looks like there are four possible sign combinations
in the above, however the requirement that 2uv = y means the sign of u
determines that of v.) In polars it looks simpler: if z = reiθ then w =
√ iθ/2
re . The tricky part is that we have to consider all possible values
of θ. Note that is θ is some possible value of the argument, then all other
possible value are of the form θ + 2πn, n ∈ Z, so all possible values of w are
√ iθ/2 iπn
re e . The last factor takes one of two values 1 and −1 depending on
√
whether n is even or odd. So all solutions are ± reiθ/2 .
To make this a single-valued function we have to make of choice of θ.
For example, we can do it by requiring θ ∈ [0, 2π). This creates a branch in
C, but it is discontinuous along [0, ∞). On the other hand, it is continuous
and in fact holomorphic in a smaller domain D = C \ [0, ∞). In this case
[0, ∞) is the corresponding branch cut.
Example 1.5.6. Another important example of a multi-valued function

which we have already discussed is the complex logarithm: as a multifunc-
tion we have [Log(z)] = {log(|z|) + i(θ + 2nπ) : n ∈ Z} where z = |z|eiθ . It
is important to note, that θ is not defined uniquely, there are many possible
values of θ such that z = |z|eiθ , but they all different by integer multiples of
2π, so the set of values [Log(z)] does not depend on the choice of θ. To ob-
tain a branch of the multifunction we must make a choice of argument func-
tion arg : C → R which is continuous in some (smaller) domain D. Given
this choice, we may define a branch
Log(z) = log(|z|) + i arg(z), z∈D
which is a continuous function in D. This is exactly how we have constructed

the principal value of the logarithm in Proposition 1.4.14. It corresponds to
the branch cut [−∞, 0].
Instead, we can choose θ differently, for example, θ ∈ [0, 2π). This
could correspond to a branch cut [0, ∞) and the corresponding branch of the
logarithm will be holomorphic in D = C\[0, ∞]. The proof of holomorphicity
of this branch is exactly the same as the proof of Proposition 1.4.14.
Example 1.5.7. Another important class of examples of multifunctions are
the complex power multifunctions z 7→ [z α ] where α ∈ C: These are given
by
z 7→ exp(α · [log(z)]) = {exp(α · w) : w ∈ C, ew = z}
Note this includes the square root multifunction we discussed above, which
can be defined without the use of the exponential function. Indeed if α =
m/n is rational, m ∈ Z, n ∈ Z>0 , then [z α ] = {w ∈ C : wm = z n }. For
α ∈ C\Q however, we can only define [z α ] using the exponential function.
Clearly from its definition, anytime we choose a branch L(z) of [log(z)] we
obtain a corresponding branch exp(α.L(z)) of [z α ]. If we pick L(z) to be the
principal branch of [log(z)], then the corresponding branch of [z α ] is called
the principal branch of [z α ].
Multivalued logarithm and power functions share a lot of nice properties
with their real counterparts.
Proposition 1.5.8. For α, z, w ∈ C with z, w ̸= 0
[log z] + [log w] = [log zw],
and
[z α ][wα ] = [(zw)α ].
Proof. Let θ and ϕ be some values of arg(z) and arg(w), then [log z] =
{log |z| + iθ + 2πin, n ∈ Z} and [log w] = {log |w| + iϕ + 2πik, k ∈ Z}.
Adding two sets term-by-term we get
[log z] + [log w] ={log |z| + log |w| + i(θ + ϕ) + 2πi(k + n), k, n ∈ Z}

={log |zw| + i(θ + ϕ) + 2πim, m ∈ Z} = [log(zw)].
Similarly,
[z α ] = {|z|α exp(iαθ + iα2πin)}
[wα ] = {|w|α exp(iαϕ + iα2πik)}
and multiplying them term-by-term we get
[z α ][wα ] ={|zw|α exp(iα(θ + ϕ) + iα2πi(n + k), k, n ∈ Z}

={|zw|α exp(iα(θ + ϕ) + iα2πim, m ∈ Z} = [(zw)α ].
Remark 1.5.9. Not all properties of real functions are valid for complex
multifunctions. For example, in general,
[z α ][z β ] ̸= [z α+β ].
This can be seen by considering
[11/2 ][11/2 ] = {−1, 1}{−1, 1} = {−1, 1} =

̸ {1} = [11 ].
Some operations with multifunctions are a bit counter-intuitive, so one

has to be very careful. One such example is given by the following exercise.
Exercise 1.5.10. Let z be any non-zero complex number. Then [log(−z)2 ] =
[log z 2 ], by Proposition 1.5.8 this implies [log(−z)] + [log(−z)] = [log z] +
[log z], so 2[log(−z)] = 2[log z] and [log(−z)] = [log z].
The last conclusion is clearly wrong since
[log z] = {log |z| + iθ + 2πik, k ∈ Z}

[log(−z)] = {log |z| + iθ + πi + 2πik, k ∈ Z}.
Where is a mistake in this argument?

Example 1.5.11. Let F (z) be the multi-function
[(1 + z)α ] = {exp(α · w) : w ∈ C, exp(w) = 1 + z}.
Using Log(z) the principal branch of [log(z)] we obtain a branch f (z) of

[(1 + z)α ] given by f (z) = exp(α · Log(1 + z)). Let αk = k!
1

α · (α − 1) . . . (α −
k + 1). We want to show that a version of the binomial theorem holds for
this branch of the multifunction [(1 + z)α ]. Let
∞
X α k
s(z) = z ,
k
k=0
By the ratio test, s(z) has a radius of convergence equal to 1, so that s(z)
defines a holomorphic function in B(0, 1). Moreover, you can check using
the properties of power series established in a previous section, that within
B(0, 1), s(z) satisfies (1 + z)s′ (z) = α · s(z).
Now f (z) is defined on C\(−∞, −1), and hence on all of B(0, 1). More-
over f ′ (z) = α/(1+z). We claim that within the open ball B(0, 1) the power
series s(z) = ∞ α k
P
n=0 k z coincides with f (z). Indeed if we set
g(z) = s(z) exp(−f (z))
then g(z) is holomorphic for every z ∈ B(0, 1) and by the chain rule
g ′ (z) = (s′ (z) − s(z)f ′ (z)) exp(−f (z)) = 0
since s′ (z) = α·s(z)

1+z . Also g(0) = 1 so g is constant and s(z) = f (z).
Here we use the fact that if a holomorphic function g has g ′ (z) = 0 on
B(0, 1) then it is constant. We have already proven this for C and in fact the
same proof applies to B(0, 1). Indeed, as we saw in the case of C, if g ′ (z) = 0
for all z then g is constant on any vertical and horizontal segment, which
clearly implies that g is constant on B(0, 1). We note that this follows also
from the following general result that we will prove soon: if a holomorphic
function g has g ′ (z) = 0 for all z in a domain U , then g is constant on U .
Example 1.5.12. A more interesting example is the function f (z) = [(z 2 −
1)1/2 ]. Using the principal branch of the square root function, we obtain a
branch f1 of f on the complement of E = {z ∈ C : z 2 − 1 ∈ (−∞, 0]}, which
one calculates is equal to [−1, 1] ∪ iR.
Another approach is to write [(z 2 − 1)1/2 ] = [(z − 1)1/2 ][(z + 1)1/2 ] and
to define branches of both factors. For both factors we will use the branch
of z 1/2 with the branch cut (−∞, 0] which corresponds to using arg(z) ∈
(−π, π). This gives the branch
f (z) = |z − 1|1/2 |z + 1|1/2 exp(iθ1 /2) exp(iθ2 /2)
where θ1 and θ2 are arguments of z − 1 and z + 1 (see Figure 1.5.12). The

first factor uses the branch cut (−∞, 1] and the second (−∞, −1], so the
product has a branch cut (−∞, 1].
Figure 1.1: Defining the root of z 2 − 1.
However let us examine the behaviour of the product: If z crosses the

negative real axis at Im (z) < −1 then θ1 and θ2 both jump by 2π, so that
(θ1 +θ2 )/2 jumps by 2π, and hence exp((θ1 +θ2 )/2) is in fact continuous.
√ On
the other hand, if we cross the segment (−1, 1) then only the factor z − 1
switches sign, so our branch is discontinuous there. In a sense this means
that jumps across (−∞, −1) cancel out and so the branch is continuous and
holomorphic outside of the cut [−1, 1].
Chapter 2
Paths and Integration
Paths will play a crucial role in our development of the theory of complex
differentiable functions. In this section, we review the notion of a path and
define the integral of a continuous function along a path.
2.1 Paths
Recall that a path in the complex plane is a continuous function γ : [a, b] →
C. A path is said to be closed if γ(a) = γ(b). The image of a path γ is
{z ∈ C : z = γ(t), some t ∈ [a, b]}.
In many cases, we will abuse notations and denote the image by γ as well.
Usually, the meaning is clear from the context. In a few cases where it is
important to distinguish between the path and its image, we will denote the
image by γ ∗ .
Although for some purposes it suffices to assume that γ is continuous, in
order to make sense of the integral along a path we will require our paths to
be (at least piecewise) differentiable. We thus need to define what we mean
for a path to be differentiable:
Definition 2.1.1. We will say that a path γ : [a, b] → C is differentiable

if its real and imaginary parts are differentiable as real-valued functions.
Equivalently, γ is differentiable at t0 ∈ [a, b] if
γ(t) − γ(t0 )
lim
t→t0 t − t0
27
28 CHAPTER 2. PATHS AND INTEGRATION
exists, and then we denote this limit as γ ′ (t0 ). (If t = a or b then we interpret
the above as a one-sided limit.) We say that a path is C 1 if it is differentiable
and its derivative γ ′ (t) is continuous.
We will say a path is piecewise C 1 if it is continuous on [a, b] and the
interval [a, b] can be divided into subintervals on each of which γ is C 1 . That
is, there is a finite sequence a = a0 < a1 < . . . < am = b such that γ|[ai ,ai+1 ]
is C 1 . Thus in particular, the left-hand and right-hand derivatives of γ at
ai (1 ≤ i ≤ m − 1) may not be equal.
For any path γ : [a, b] → C we define the opposite path γ − by γ − :
[a, b] → C, γ − (t) = γ(b − t).
If γ1 : [a, b] → C and γ2 : [c, d] → C are two paths such that γ1 (b) = γ2 (c)
then they can be concatenated to give a path γ1 ⋆ γ2 which traverses first γ1
and then γ2 . Formally γ1 ⋆ γ2 : [a, b + d − c] → C where

γ1 (t), t≤b
γ1 ⋆ γ2 (t) =
γ2 (t − b + c), t≥b
So a piecewise C 1 path is precisely a finite concatenation of C 1 paths.
Remark 2.1.2. Note that a C 1 path may not have a well-defined tangent
at every point: if γ : [a, b] → C is a path and γ ′ (t) ̸= 0, then the line
{γ(t) + sγ ′ (t) : s ∈ R} is tangent to γ ∗ , however, if γ ′ (t) = 0, the image of
γ may have no tangent line there. Indeed consider the example of γ : [−1, 1] →
C given by
2
t −1 ≤ t ≤ 0
γ(t) =
it2 0 ≤ t ≤ 1.
Since γ ′ (0) = 0 the path is C 1 , even though it is clear there is no tangent

line to the image of γ at 0. This shows that the image of a C 1 path can
have a ‘corner’, in fact, it
If s : [a, b] → [c, d] is a differentiable map, then we have the following
version of the chain rule, which is proved in exactly the same way as the
real-valued case. It will be crucial in our definition of the integral of functions
f : C → C along paths.
Lemma 2.1.3. Let γ : [c, d] → C and s : [a, b] → [c, d] and suppose that s
is differentiable at t0 and γ is differentiable at s0 = s(t0 ). Then γ ◦ s is
differentiable at t0 with derivative
(γ ◦ s)′ (t0 ) = s′ (t0 ) · γ ′ (s(t0 )).

2.1. PATHS 29
Proof. Let ϵ : [c, d] → C be given by ϵ(s0 ) = 0 and
γ(x) = γ(s0 ) + γ ′ (s0 )(x − s0 ) + (x − s0 )ϵ(x),
(so that this equation holds for all x ∈ [c, d]), then ϵ(x) → 0 as x → s0 by
the definition of γ ′ (s0 ), i.e. ϵ is continuous at t0 . Substituting x = s(t) into
this we see that for all t ̸= t0 we have
γ(s(t)) − γ(s0 ) s(t) − s(t0 ) ′

= γ (s(t)) + ϵ(s(t)) .
t − t0 t − t0
Now s(t) is continuous at t0 since it is differentiable there hence ϵ(s(t)) → 0

as t → t0 , thus taking the limit as t → t0 we see that
(γ ◦ s)′ (t0 ) = s′ (t0 )(γ ′ (s0 ) + 0) = s′ (t0 )γ ′ (s(t0 )),
as required.
Definition 2.1.4. Let ϕ : [a, b] → [c, d] be continuously differentiable with

ϕ(a) = c and ϕ(b) = d, and let γ : [c, d] → C be a C 1 -path, then setting
γ̃ = γ ◦ ϕ, by Lemma 2.1.3 we see that γ̃ : [a, b] → C is again a C 1 -path with
the same image as γ and we say that γ̃ is a reparametrization of γ.
Definition 2.1.5. We will say two parametrized paths γ1 : [a, b] → C and

γ2 : [c, d] → C are equivalent if there is a continuously differentiable bijective
function s : [a, b] → [c, d] such that s′ (t) > 0 for all t ∈ [a, b] and γ1 = γ2 ◦ s.
It is straightforward to check that equivalence is indeed an equivalence rela-
tion on parametrized paths, and we will call the equivalence classes oriented
curves in the complex plane. We denote the equivalence class of γ by [γ].
The condition that s′ (t) > 0 ensures that the path is traversed in the same
direction for each of the parametrizations γ1 and γ2 . Moreover, γ1 is piece-
wise C 1 if and only if γ2 is.
Remark 2.1.6. Note that if γ : [a, b] → C is piecewise C 1 , then by choosing a

reparametrization by a function ψ : [a, b] → [a, b] which is strictly increasing
and has vanishing derivative at the points where γ fails to be C 1 , we can
replace γ by γ̃ = γ ◦ ψ to obtain a C 1 path with the same image. For this rea-
son, some texts insist that C 1 paths have everywhere non-vanishing deriva-
tive.
In this course we will not insist on this. Indeed sometimes it is convenient
to consider a constant path, that is a path γ : [a, b] → C such that γ(t) = z0
for all t ∈ [a, b] (and hence γ ′ (t) = 0 for all t ∈ [a, b]).
Example 2.1.7. The most basic example of a closed curve is a circle: If

z0 ∈ C and r > 0 then the path z(t) = z0 + re2πit (for t ∈ [0, 1]) is the simple
closed path with positive orientation encircling z0 with radius r. The path
z̃(t) = z0 + re−2πit is the simple closed path encircling z0 with radius r and
negative orientation.
Another useful path is a line segment: if a, b ∈ C then we denote by
γ[a,b] : [0, 1] → C the path given by t 7→ a + t(b − a) = (1 − t)a + tb traverses
the line segment from a to b. We denote the corresponding oriented curve
by [a, b] (which is consistent with the notation for an interval in the real
line). One of the simplest classes of closed paths is triangles: given three
points a, b, c, we define the triangle, or triangular path, associated to them,
to be the concatenation of the associated line segments, that is γa,b,c =
γ[a,b] ⋆ γ[b,c] ⋆ γ[c,a] .
2.2 Integration along a path

To define the integral of a complex-valued function along a path, we first
need to define the integral of functions f : [a, b] → C on a closed interval
[a, b] taking values in C. Last year in Analysis III the Riemann integral was
defined for a function on a closed interval [a, b] taking values in R, but it is
easy to extend this to functions taking values in C: Indeed we may write
f (t) = u(t) + iv(t) where u, v are functions on [a, b] taking real values. Then
we say that f is Riemann integrable if both u and v are, and we define:
Z b Z b Z b
F (t)dt = u(t)dt + i v(t)dt
a a a
It is easy to check that the integral is then complex linear, that is, if f1 , f2
are complex-valued Riemann integrable functions on [a, b], and α, β ∈ C,
then αf1 + βf2 is Riemann integrable and
Z b Z b Z b
(αf1 + βf2 )dt = α f1 dt + β f2 dt.
a a a
Note that if f is continuous, then its real and imaginary parts are also con-
tinuous, and so, in particular, Riemann integrable1 . The class of Riemann
integrable (real or complex-valued) functions on a closed interval is however
slightly larger than the class of continuous functions, and this will be useful
to us at certain points. In particular, we have the following:
1
It is clear this definition extends to give a notion of the integral of a function f : [a, b] →
n
R – we say f is integrable if each of its components is, and then define the integral to be
the vector given by the integrals of each component function.
2.2. INTEGRATION ALONG A PATH 31
Lemma 2.2.1. Let [a, b] be a closed interval and S ⊂ [a, b] a finite set. If f
is a bounded continuous function (taking real or complex values) on [a, b]\S
then it is Riemann integrable on [a, b].
Proof. The case of complex-valued functions follows from the real case by
taking real and imaginary parts. For the case of a function f : [a, b]\S → R,
let a = x0 < x1 < x2 < . . . < xk = b be any partition of [a, b] which includes
the elements of S. Then on each open interval (xi , xi+1 ) the function f is
bounded and continuous, and hence integrable. We may therefore set
Z b Z x1 Z x2 Z xk
f (t)dt = f (t)dt + f (t)dt + . . . + f (t)dt
a x0 x1 xk−1
Rb
The standard additivity properties of the integral then show that a f (t)dt
is independent of any choices.
Remark 2.2.2. Note that normally when one speaks of a function f being
integrable on an interval [a, b] one assumes that f is defined on all of [a, b].
However, if we change the value of a Riemann integrable function f at a
finite set of points, then the resulting function is still Riemann integrable
and its integral is the same. Thus if one prefers the function f in the previous
lemma to be defined on all of [a, b] one can define f to take any values at all
on the finite set S.
Lemma 2.2.3. Suppose that f : [a, b] → C is a complex-valued function.
Then we have Z b Z b
f (t)dt ≤ |f (t)|dt.
a a
√
Proof. First note that if f (t) = u(t) + iv(t) then |f (t)| = u2 + v 2 so that
b
if f is integrable |f (t)| is also2 . We may write a f (t)dt = reiθ , where
R
r ∈ [0, ∞) and θ ∈ [0, 2π). Now taking the components of f in the direction
of eiθ and ei(θ+π/2) = ieiθ , we may write f (t) = ũ(t)eiθ + iṽ(t)eiθ . Then by
Rb Rb
our choice of θ we have a f (t)dt = eiθ a ũ(t)dt, and so
Z b Z b Z b Z b
f (t)dt = ũ(t)dt ≤ |ũ(t)|dt ≤ |f (t)|dt,
a a a a
where in the first inequality we used the triangle inequality for the Riemann
integral of real-valued functions.
2
The simplest way to see this is to use that fact that if ϕ is continuous and f is Riemann
integrable, then ϕ ◦ f is Riemann integrable.
We are now ready to define the integral of a function f : C → C along a

piecewise-C 1 curve.
Definition 2.2.4. If γ : [a, b] → C is a piecewise-C 1 path and f : C → C,

then we define the integral of f along γ to be
Z Z b
f (z)dz = f (γ(t))γ ′ (t)dt.
γ a
In order for this integral to exist in the sense we have defined, we have
seen that it suffices for the functions f (γ(t)) and γ ′ (t) to be bounded and
continuous at all but finitely many t. Our definition of a piecewise C 1 -path
ensures that γ ′ (t) is bounded and continuous away from finitely many points
(the boundedness follows from the existence of the left and right-hand lim-
its at points of discontinuity of γ ′ (t)). For most of our applications, the
function f will be continuous on the whole image γ ∗ of γ, but it will occa-
sionally be useful to weaken this to allow f (γ(t)) finitely many (bounded)
discontinuities.
Lemma 2.2.5. If γ : [a, b] → C be a piecewise C 1 path and γ̃ : [c, d] → C is

an equivalent path, then for any continuous function f : C → C we have
Z Z
f (z)dz = f (z)dz.
γ γ̃
In particular, the integral only depends on the oriented curve [γ].
Proof. Since γ̃ is equivalent to γ there is a continuously differentiable func-

tion s : [c, d] → [a, b] with s(c) = a, s(d) = b and s′ (t) > 0 for all t ∈ [c, d].
Suppose first that γ is C 1 . Then by the chain rule, we have
Z Z d
f (z)dz = f (γ(s(t)))(γ ◦ s)′ (t)dt
γ̃ c
Z d
= f (γ(s(t))γ ′ (s(t))s′ (t)dt
c
Z b
= f (γ(s))γ ′ (s)ds
Za
= f (z)dz.
γ
where in the second last equality we used the change of variables formula.
If a = x0 < x1 < . . . < xn = b is a decomposition of [a, b] into subintervals
such that γ is C 1 on [xi , xi+1 ] for 1 ≤ i ≤ n − 1 then since s is a continuous

increasing bijection, we have a corresponding decomposition of [c, d] given
by the points s−1 (x0 ) < . . . < s−1 (xn ), and we have
Z Z d
f (z)dz = f (γ(s(t))γ ′ (s(t))s′ (t)dt
γ̃ c
n−1
X Z s−1 (xi+1 )
= f (γ(s(t))γ ′ (s(t))s′ (t)dt
i=0 s−1 (xi )
n−1
X Z xi+1
= f (γ(x))γ ′ (x)dx
i=0 xi
Z b Z
= f (γ(x))γ ′ (x)dx = f (z)dz.
a γ
where the third equality follows from the case of C 1 paths established above.
Definition 2.2.6. If γ : [a, b] → C is a C 1 path then we define the length of

γ to be
Z b
ℓ(γ) = |γ ′ (t)|dt.
a
Using the chain rule as we did to show that the integrals of a function
f : C → C along equivalent paths are equal, one can check that the length of
a parametrized path is also constant on equivalence classes of paths, so, in
fact, the above defines a length function for oriented curves. The definition
extends in an obvious way to give a notion of length for piecewise C 1 -paths.
More generally, one can define the integral with respect to arc-length of a
function f : U → C such that γ ∗ ⊆ U to be
Z Z b
f (z)|dz| = f (γ(t))|γ ′ (t)|dt.
γ a
This integral is invariant with respect to C 1 reparametrizations s : [c, d] →

[a, b] if we require s′ (t) ̸= 0 for all t ∈ [c, d] (the condition s′ (t) > 0 is not
necessary because of this integral takes the modulus of γ ′ (t)). In particular
ℓ(γ) = ℓ(γ − ).
Remark 2.2.7 (Non-examinable). It is possible to relax the assumption that

γ is (piecewise) C 1 and replace it with the assumption that it has a finite
length. The length can be defined using partitions (like in the definition of
the Riemann integral) so we don’t even need γ to be differentiable. Such curves
are called rectifiable. With minimal modifications, everything we do in this
course can be done for rectifiable curves.
The integration of functions along piecewise smooth paths has many of
the properties that the integral of real-valued functions along an interval
possesses. We record some of the most standard of these:
Proposition 2.2.8. Let f, g : U → C be continuous functions on an open
subset U ⊆ C and γ, η : [a, b] → C be piecewise-C 1 paths whose images lie in
U . Then we have the following:
1. (Linearity): For α, β ∈ C,
Z Z Z
(αf (z) + βg(z))dz = α f (z)dz + β g(z)dz.
γ γ γ
2. (Orientation): If γ − denotes the opposite path to γ then

Z Z
f (z)dz = − f (z)dz.
γ γ−
3. (Additivity): If γ ⋆ η is the concatenation of the paths γ, η in U , we

have Z Z Z
f (z)dz = f (z)dz + f (z)dz.
γ⋆η γ η
4. (Estimation Lemma.) We have

Z
f (z)dz ≤ sup |f (z)|ℓ(γ).
γ z∈γ ∗
Proof. Since f, g are continuous, and γ, η are piecewise C 1 , all the integrals
in the statement are well-defined: the functions f (γ(t))γ ′ (t), f (η(t))η ′ (t),
g(γ(t))γ ′ (t) and g(η(t))η ′ (t) are all Riemann integrable. It is easy to see that
one can reduce these claims to the case where γ is smooth. The first claim
is immediate from the linearity of the Riemann integral, while the second
claim follows from the definitions and the fact that (γ − )′ (t) = −γ ′ (a + b − t).
The third follows immediately for the corresponding additivity property of
Riemann integrable functions.
For the fourth part, first note that γ([a, b]) is compact in C since it is the
image of the compact set [a, b] under a continuous map. It follows that the
function |f | is bounded on this set so that supz∈γ([a,b]) |f (z)| exists. Thus

we have
Z Z b
f (z)dz = f (γ(t))γ ′ (t)dt
γ a
Z b
≤ |f (γ(t))||γ ′ (t)|dt
a
Z b
≤ sup |f (z)| |γ ′ (t)|dt
z∈γ ∗ a
= sup |f (z)|ℓ(γ).
z∈γ ∗
where for the first inequality we use the triangle inequality for complex-
valued functions as in Lemma 2.2.3 and the positivity of the Riemann inte-
gral for the second inequality.
Remark 2.2.9. We give part (4) of the above proposition a name (the “esti-
mation lemma”) because it will be very useful later in the course. We will
give one important application of it now:
Proposition 2.2.10. Let fn : U → C be a sequence of continuous functions

on an open subset U of the complex plane. Suppose that γ : [a, b] → C is
a path whose image is contained in U . If (fn ) converges uniformly to a
function f on the image of γ then
Z Z
fn (z)dz → f (z)dz.
γ γ
Proof. We have
Z Z Z
f (z)dz − fn (z)dz = (f (z) − fn (z))dz
γ γ γ
≤ sup {|f (z) − fn (z)|}ℓ(γ),
z∈γ ∗
by the estimation lemma. Since we are assuming that fn tends to f uni-

formly on γ ∗ we have sup{|f (z) − fn (z)| : z ∈ γ ∗ } → 0 as n → ∞ which
implies the result.
Definition 2.2.11. Let U ⊆ C be an open set and let f : U → C be a

continuous function. If there exists a differentiable function F : U → C with
F ′ (z) = f (z) then we say F is a primitive for f on U .
We will need a version of the chain rule for the composition of a complex
with a real function:
Lemma 2.2.12. Let U be an open subset of C and let f : U → C be a

holomorphic function. If γ : [a, b] → U is a (piecewise) C 1 -path, then f (γ(t))
is differentiable at any t where γ is differentiable and
d
(f (γ(t))) = f ′ (γ(t)) · γ ′ (t)
dt
Proof. Assume that γ is differentiable at t0 ∈ [a, b] and let z0 = γ(t0 ) ∈ U .

By definition of f ′ , there is a function ϵ(z) such that
f (z) = f (z0 ) + f ′ (z0 )(z − z0 ) + ϵ(z)(z − z0 )
where ϵ(z) → 0 = ϵ(z0 ) as z → z0 . But then
f (γ(t)) − f (γ(t0 )) γ(t) − γ(t0 ) γ(t) − γ(t0 )

= f ′ (z0 ) · + ϵ(γ(t)) · .
t − t0 t − t0 t − t0
But now consider the two terms on the right-hand side of this expression:
the first term, as t → t0 tends to f ′ (z0 )(γ ′ (t0 )). On the other hand, for
the second term, since γ(t)−γ(t t−t0
0)
tends to γ ′ (t0 ) as t tends to t0 , we see that
γ(t) − γ(t0 )/(t − t0 ) is bounded as t → t0 , while since γ(t) is continuous
at t0 since it is differentiable there, ϵ(γ(t)) → ϵ(γ(t0 )) = ϵ(z0 ) = 0. It
follows that the second term tends to zero, so that the left-hand side tends
to f ′ (z0 )(γ ′ (t0 )) as t → t0 , as required.
The fundamental theorem of calculus has the following important con-

sequence3 :
Theorem 2.2.13. (Fundamental theorem of Calculus): Let U ⊆ C be open

and let f : U → C be a continuous function. If F : U → C is a primitive for
f and γ : [a, b] → U is a piecewise C 1 path in U then we have
Z
f (z)dz = F (γ(b)) − F (γ(a)).
γ
In particular, the integral of such a function f around any closed path is

zero.
3
You should compare this to the existence of a potential in vector calculus.
Proof. First suppose that γ is C 1 . Then we have

Z Z Z b
′
f (z)dz = F (z)dz = F ′ (γ(t))γ ′ (t)dt
γ γ a
Z b
d
= (F ◦ γ)(t)dt
a dt
= F (γ(b)) − F (γ(a)),
where in the second line we used the chain rule (lemma 2.2.12) and in the last
line we used the Fundamental Theorem of Calculus from Prelims analysis
on the real and imaginary parts of F ◦ γ.
If γ is only4 piecewise C 1 , then take a partition a = a0 < a1 < . . . <
ak = b such that γ is C 1 on [ai , ai+1 ] for each i ∈ {0, 1, . . . , k − 1}. Then we
obtain a telescoping sum:
Z Z b
f (z) = f (γ(t))γ ′ (t)dt
γ a
k−1 Z ai+1
X
= f (γ(t))γ ′ (t)dt
i=0 ai
k−1
X
= (F (γ(ai+1 )) − F (γ(ai )))
i=0
= F (γ(b)) − F (γ(a)),
Finally, since γ is closed precisely when γ(a) = γ(b) it follows immedi-

ately that the integral of f along a closed path is zero.
Remark 2.2.14. If f (z) has finitely many points of discontinuity S ⊂ U

but is bounded near them, and γ(t) ∈ S for only finitely many t, then
provided F is continuous and F ′ = f on U \S, the same proof shows that
the fundamental theorem still holds – one just needs to take a partition of
[a, b] to take account of those singularities along with the singularities of
γ ′ (t).
Theorem 2.2.13 already has an important consequence:
Corollary 2.2.15. Let U be a domain and let f : U → C be a function with

f ′ (z) = 0 for all z ∈ U . Then f is constant.
4
The reason we must be careful about this case is that the Fundamental Theorem of
Calculus only holds when the integrand is continuous.
Proof. Pick z0 ∈ U . It has been shown in the Metric Spaces course that an
open connected set of a normed space (in particular C) is path-connected
and in fact even polygonally connected, i.e. any two points of the set can
be connected by the concatenation of finitely many line segments. It follows
that any point w of U can be joined to z0 by a piecewise C 1 -path γ : [0, 1] →
U so that γ(0) = z0 and γ(1) = w. Then by Theorem 2.2.13 we see that
Z
f (w) − f (z0 ) = f ′ (z)dz = 0,
γ
so that f is constant as required.

The following theorem is a kind of converse to the fundamental theorem:
Theorem 2.2.16. If U is a domain (i.e. it is open and path connected)
and f : UR → C is a continuous function such that for any closed path in U
we have γ f (z)dz = 0, then f has a primitive.
Proof. Fix z0 in U , and for any z ∈ U set
Z
F (z) = f (z)dz.
γ
where γ : [a, b] → U with γ(a) = z0 and γ(b) = z.

We claim that F (z) is independent of the choice of γ. Indeed if γ1 , γ2
are two such paths, let γ = γ1 ⋆ γ2− be the path obtained by concatenating
γ1 and the opposite γ2− of γ2 (that is, γ traverses the path γ1 and then goes
backward along γ2 ). Then γ is a closed path and so, using Proposition 2.2.8
we have Z Z Z
0 = f (z)dz = f (z)dz + f (z)dz,
γ γ1 γ2−
R R R R
hence since γ − f (z)dz = − γ2 f (z)dz we see that γ1 f (z)dz = γ2 f (z)dz.
2
Next, we claim that F is differentiable with F ′ (z) = f (z). To see this,
fix w ∈ U and ϵ > 0 such that B(w, ϵ) ⊆ U and choose a path γ : [a, b] → U
from z0 to w. If z1 ∈ B(w, ϵ) ⊆ U , then the concatenation of γ with the
straight-line path s : [0, 1] → U given by s(t) = w + t(z − w) from w to z is
a path γ1 from z0 to z. It follows that
Z Z
F (z1 ) − F (w) = f (z)dz − f (z)dz
γ1 γ
Z Z Z
= f (z)dz + f (z)dz − f (z)dz
γ s γ
Z
= f (z)dz.
s
But then we have for z1 ̸= w

Z 1
F (z1 ) − F (w) 1
− f (w) = f (w + t(z1 − w))(z1 − w)dt − f (w)
z1 − w z1 − w 0
Z 1
= (f (w + t(z1 − w)) − f (w))dt
0
≤ sup |f (w + t(z1 − w)) − f (w)| → 0 as z1 → w
t∈[0,1]
as f is continuous at w. Thus F is differentiable at w with derivative F ′ (w) =

f (w) as claimed.
Remark 2.2.17. Note that any two primitives for a function f differ by a
constant: This follows immediately from Corollary 2.2.15, since if F1 and
F2 are two primitives, their difference (F1 − F2 ) has zero derivative.
The combination of all of these results means that a continuous function
has a primitive if and only if integrals around any closed path are equal to
zero. Let us compare this with the real case. Any continuous function on a
bounded interval is integrable and hence has a primitive. It is also easy to
see that the integral along any closed path is zero because in R closed paths
go back and forth and so cancel out. This is not the case in C as shown by
an example below.
Example 2.2.18. Let U = C \ {0} and f (z) = 1/z which is a continuous

function in U . Let γ(t) = eit : [0, 2π] → U be the unit circle. Then
Z Z 2π
1 it
f (z)dz = ie dt = 2πi ̸= 0.
γ 0 eit
This shows that 1/z does not have a primitive in U .
It is important to keep in mind that whether a function has a primitive

depends not only on the function itself but also on the domain. In particu-
lar, 1/z has a primitive in U = C \ (−∞, 0] since Log′ (z) = 1/z where Log
is the principle value of the logarithm.
The argument above never used that Log is the principal value, any
branch will work. If L is a branch of the logarithm in some domain U , then
its derivative there must be 1/z, hence 1/z has a primitive in U .
2.3 Cauchy’s theorem

In this section, we prove one of the most fundamental results about holo-
morphic functions. Essentially the rest of the course will be based on this
result. It has many different forms, here we will formulate some of them,
but full proofs are beyond the scope of this course, so we will only prove a
simpler version and give a brief indication of how the general result can be
proved.
Theorem 2.3.1 (Cauchy or Cauchy-Goursat Theorem). Let U ⊂ C be a

domain and γ be a curve such that it and all bounded components of C \ γ ∗
are inside U . Let f be a function holomorphic in U . Then
Z
f (z)dz = 0.
γ
Remark 2.3.2. Note that in this theorem we do not assume that γ is simple.
We will start by proving this result in a very simple case when γ = γa,b,c
is the boundary of a triangle.
Proof of Theorem 2.3.1 for triangles. Let T = T0 be a triangle with vertices

A, B and C. Let γ = γA,B,C be its boundary, Slightly abusing notations5 we
will write γ = ∂T . We assume that it is positively (i.e. counter clockwise)
oriented. This is not important since reversing orientation changes the sign
but we will show that the integral is equal to zero, so the sign is irrelevant.
We split T into four similar triangles Si , i = 1, . . . , 4. New vertices are
mid-points of edges. See Figure 2.3 for an illustration.
We claim that
Z X4 Z
f (z)dz = f (z)dz.
∂T i=1 ∂Si
The reason is very simple. The integral along the boundary of a triangle is
equal to the sum of integrals along its edges. Note that ‘new’ edges appear
twice with different orientations. For example, the edge [D, F ] appears in
the boundary of S1 and [F, D] in the boundary of S4 . Since they have the
opposite orientation, integrals along them will cancel out. The remaining
5
The main issue is that when we write γA,B,C we use the concatenation of sides in
the counter-clockwise order starting from [A, B], in particular, A is the beginning and
the end of the curve. When we write ∂T , then we do not specify the orientation and
the starting point. On the other hand, the standard orientation in complex analysis is
counter-clockwise and it is easy to see that the integral does not depend on the starting
point.
2.3. CAUCHY’S THEOREM 41
Figure 2.1: Splitting the triangle into four similar triangles.
six integrals will add up to the integral along the boundary of the original
triangle. R
Denote ∂T f = I and assume that I ̸= 0. In this case, there is one of Si
such that Z
1
f (z)dz ≥ |I|.
∂Si 4
We denote this smaller triangle by T1 and repeat the same process: split it
into four triangles and choose one with a large integral. This way we obtain
a sequence of triangles Tn such that Tn ⊂ Tn−1 ,
Z
1
f (z)dz ≥ n |I| (2.3.1)
∂Tn 4
and
ℓ(∂Tn ) = 2−n ℓ(∂T ). (2.3.2)
It is not hard to prove directly that there is z0 = ∩Tn . Alternatively, we
have a nested family of compacts in a metric space and their diameters go
to zero. By a result from the Metric Spaces course, their intersection is a
single point.
Function f is differentiable at z0 hence
f (z) = f (z0 ) + f ′ (z0 )(z − z0 ) + ϵ(z)(z − z0 )

where ϵ(z) → 0 as z → z0 . For any ϵ > 0 there is δ > 0 such that |ϵ(z)| < ϵ
for all z ∈ B(z0 , δ).
Next, we choose n large enough so that Tn ⊂ B(z0 , δ). Then
Z Z Z
′

f (z)dz = f (z0 ) + f (z0 )(z − z0 ) dz + ϵ(z)(z − z0 )dz.
∂Tn ∂Tn ∂Tn
The first integral vanishes since f (z0 ) + f ′ (z0 )(z − z0 ) is a linear function
which clearly has a primitive and so its integral along any closed contour is
zero. By the estimation lemma
Z
f (z)dz ≤ ϵℓ2 (∂Tn ).
∂Tn
Here we used that |z − z0 | ≤ ℓ(∂Tn ) since the distance between any two
points in a triangle is bounded by its perimeter.
Combining this with (2.3.1) and (2.3.2) we have
Z
1 1
|I| ≤ f (z)dz ≤ ϵℓ(∂T )
4n ∂Tn 4n
which leads to a contradiction if we choose ϵ < 1/ℓ(∂T ).
Next we show that for a large class of ‘nice’ domains our previous argu-
ment about triangular curves implies that the theorem is true for all curves.
We start with a couple of definitions.
Definition 2.3.3. Let X be a subset in C. We say that X is convex if for

each z, w ∈ U the line segment between z and w is contained in X. We say
that X is star-like if there is a point z0 ∈ X such that for every w ∈ X
the line segment [z0 , w] joining z0 and w lies in X. We will say that X is
star-like with respect to z0 in this case. Thus a convex subset is starlike
with respect to every point it contains.
Example 2.3.4. A disk (open or closed) is convex, as is a solid triangle or

rectangle. On the other hand a cross, such as [−2, 2]×[−1, 1]∪[−1, 1]×[−2, 2]
is star-like with respect to the origin, but is not convex. See
Proof of Theorem 2.3.1 for star-like domains. The proof proceeds similarly
to the proof of Theorem 2.2.16: by Theorem 2.2.13 it suffices to show that
f has a primitive in U . To show this, let z0 ∈ U be a point for which the
2.3. CAUCHY’S THEOREM 43
Figure 2.2: The disc is both convex and star-like, the cross is not convex
but star-like and the shape on the right is neither.
line segment from z0 to every z ∈ U lies in U . Let γz = z0 + t(z − z0 ) be a

parametrization of this curve, and define
Z
F (z) = f (ζ)dζ.
γz
We claim that F is a primitive for f on U . Indeed pick ϵ > 0 such that

B(z, ϵ) ⊆ U . Then if w ∈ B(z, ϵ) note that the triangle T with vertices
z0 , z, w lies entirely in U by the assumption that U is star-like
R with respect
to z0 . We have already proved that in this case that ∂T f (ζ)dζ = 0, and
hence if η(t) = w + t(z − w) is the straight-line path going from w to z (so
that ∂T is the concatenation of γw , η and γz− , see Figure 2.3) we have
F (z) − F (w)
Z
f (ζ)
− f (z) = dζ − f (z)
z−w η z−w
Z 1
= f (w + t(z − w))dt − f (z)
0
Z 1
= (f (w + t(z − w)) − f (z)dt
0
≤ sup |f (w + t(z − w)) − f (z)|,
t∈[0,1]
which, since f is continuous at w, tends to zero as w → z so that F ′ (z) =

f (z) as required.
Finally, we outline the strategy of how to prove Theorem 2.3.1 in its

stated form. This part is non-examinable.
Outline of the proof of Theorem 2.3.1. Non-examinable. First, let us assume

that γ is a polygonal curve i.e. a concatenation of a finite number of in-
tervals.If γ is not a simple curve then the corresponding polygon can be
Figure 2.3: In a star-like domain if z and w are close enough then the triangle
z, z0 , w is completely inside the domain.
decomposed
R P R into a finite number of simple polygons Si .It is easy to see that
γ = ∂Si . When γ is a simple curve, then the corresponding polygon
can be triangulated into triangles Si . R See P Figure
R As before, the integrals
along new sides cancel out and again γ = ∂Si Since we already know
.
that integrals along triangles are equal to zero, this immediately proves the
statement for polygonal curves.
Note, that the argument above is almost complete. The only non-trivial
part is the statement that any polygon can be triangulated. This sounds
obvious, and in fact it is not very difficult proof but it is a bit tricky to write
down rigorously with all details.
Next, let γ : [a, b] → C be a general curve. It is not very difficult to show
that similarly to the real case, a complex integral can be approximated by a
Riemann sum. Namely, if ti form a sufficiently fine partition and zi = γ(ti )
then Z
X
f (zi )(zi+1 − zi ) → f (z)dz
γ
as the mesh goes to 0. Clearly, an integral along the part of γ from ti to
ti+1 is close to f (zi )(zi+1 − zi ) (since the curve is almost an interval an the
function is almost a constant. This part is relatively easy to justify. It is a
bit harder to justify that the sum of errors
Z is still small and
XZ
f (z)dz → f (z)dz,
γi γ
where γi is an interval [zi , zi+1 ]. The sum above is equal to the integral of f
along the polygonal curve with vertices zi . As we have shown before, such
an integral is equal to 0, so its limit is also 0.
2.4. DEFORMATION THEOREM AND HOMOTOPY 45
Figure 2.4: Left: a non-simple polygon can be decomposed into simple

polygons. Right: a simple polygon can be decomposed into triangles.
2.4 Deformation theorem and homotopy

In this section we state the most general version of the Cauchy’s theorem.
Roughly speaking it states that if we continuously deform a curve then the
integral does not change. This is true even if the integral is not zero. This
should not be surprising given our previous discussion. If we move a curve
a little bit, then the difference between two curves is a small contour and
the function is analytic inside of it, so be the previous version of the Cauchy
theorem the difference in integrals is equal to the integral along this contour
which is zero. Making this argument rigorous is not extremely difficult but
goes beyond the scope of this course. So we will rigorously define and set
up everything but will not prove most of the statements in this section.
Definition 2.4.1. Suppose that U is an open set in C and a, b ∈ U and

that γ0 : [0, 1] → U and γ1 : [0, 1] → U are two paths in U such that γ0 (0) =
γ1 (0) = a and γ0 (1) = γ1 (1) = b. We say that γ0 and γ1 are homotopic in
U if there is a continuous function h : [0, 1] × [0, 1] → U such that
h(0, s) =a, h(1, s) = b

h(t, 0) =γ0 (t), h(t, 1) = γ1 (t).
One should think that h defines a family of curves γs (t) = h(t, s) that have
the same end-points and as s changes from 0 to 1 the curves continuously
deform from γ0 to γ1 .
For closed curves, the definition is a bit different. We do not require that
end-points stay the same but we require that all γs are closed curves.
We say that a closed curve γ is null homotopic in U is it is homotopic

to a constant path.
One can show that the relation “γ is homotopic to η” is an equivalence

relation, so that any path γ between a and b belongs to a unique equivalence
class, known as its homotopy class.
Definition 2.4.2. Suppose that U is a domain in C. We say that U is simply

connected if for every a, b ∈ U , any two paths from a to b are homotopic in
U . Equivalently, is any closed curve is null homotopic.
Remark 2.4.3. The fact that the two definitions are equivalent is non-trivial
but it is a topological fact that is beyond the scope of this course.
Example 2.4.4. Any convex or starlike domain is simply connected. The

argument is simple. If γ(t) is a closed curve and U is star-like with respect
to z0 , then the function
h(t, s) = sz0 + (1 − s)γ(t)
is a homotopy between γ and a constant path γ1 (t) = z0 .
Generally, showing that a curve is not null-homotopic or that a domain

is not simply connected is harder. Informally, it means that the domain has
holes. Alternatively, a domain in C is simply connected if its complement
has only one connected component. We will not prove this statement.
Example 2.4.5. An annulus
A(z0 , r, R) = {z ∈ C : 0 ≤ r < |z − z0 | < R ≤ ∞}
is not simply connected. Note that in this definition we allow r = 0 and

R = ∞. We will see that this is indeed the case later in the course.
Example 2.4.6. Let us consider a horseshoe domain and a contour inside it

as in Figure 2.4.6. If γ(t) = r(t) exp(iθ(t)) where θ(t) ∈ (−π, π). Let r0 > 0
be a point inside the domain. It is not hard to check that the function
h(t, s) = (sr0 + (1 − s)r(t)) exp((1 − s)θ(t))
is a homotopy between γ0 = γ and γ1 (t) = r0 . Essentially, we independently

contract the modulus to r0 and the argument to 0. This shows that the
domain is simply connected.
2.4. DEFORMATION THEOREM AND HOMOTOPY 47
Figure 2.5: A contour inside a horseshoe domain can be contracted to a

point along the dotted lines.
We are now ready to state our extension of Cauchy’s theorem. The proof
is given in the Appendices. The proof is non-examinable.
Theorem 2.4.7 (Homotopy Cauchy’s Theorem or Deformation Theorem).

Let U be a domain in C. Suppose that γ1 and γ2 are two paths in U that
are homotopic in U .6 Let f : U → C be a holomorphic function. Then
Z Z
f (z)dz = f (z)dz.
γ1 γ2
Remark 2.4.8. Notice that this theorem is really more general than the
previous versions of Cauchy’s theorem we have seen – in the case where
a holomorphic function f : U → C has a primitive the conclusion of the
previous theorem is, of course, obvious from the Fundamental Theorem of
Calculus7 , and our previous formulations of Cauchy’s theorem were proved
by producing a primitive for f on U . One significance of the homotopy form
of Cauchy’s theorem is that it applies to domains U even when there is no
primitive for f on U .
Theorem 2.4.9 (Cauchy’s theorem for simply connected domains). Sup-

pose that U is a simply connected domain, let a, b ∈ U , and let f : U → C be
6
So they either have the same endpoint or both are closed
7
Indeed the hypothesis that the paths γ1 and γ2 are homotopic is irrelevant when f
has a primitive on U . If there is a primitive F then for any curve connecting a and b the
integral is equal to F (b) − F (a) and the integral along any contour is equal to 0.
a holomorphic function on U . Then if γ1 , γ2 are paths from a to b we have

Z Z
f (z)dz = f (z)dz.
γ1 γ2
R
In particular, if γ is a closed oriented curve we have γ f (z)dz = 0, and
hence any holomorphic function on U has a primitive.
Proof. Since U is simply connected, any two paths from from a to b are
homotopic, so we can apply Theorem 2.4.7. For the last part, in a simply
connected domain any closed path γ : [0, 1] → U , with γ(0) = R γ(1) = a
say, is homotopic to some constant path c(t) = z0 , and hence γ f (z)dz =
R
c f (z)dz = 0. The final assertion then follows from the Theorem 2.2.16.
Remark 2.4.10. Theorem 2.4.9 tells us that in a simply connected domain

any holomorphic function has a primitive. In fact, the converse is also
true. If any holomorphic function has a primitive then the domain is simply
connected.
Example 2.4.11. If U ⊆ C\{0} is simply connected, the previous theorem

shows that there is a holomorphic branch of [log(z)] defined on all of U
(since any primitive for f (z) = 1/z will be such a branch).
Example 2.4.12. Let us consider an annulus U and a curve γ which is ho-

motopic to a counter-clockwise oriented circle γr . Let f (z) = 1/z (see Figure
2.4.12 for
R an example). As we have discussed R before, f has no primitive in
U and γr f (z)dz = 2πi. By Theorem 2.4.7 γ f (z)dz = 2πi as well.
2.5 Winding numbers

In the present section we investigate the change in argument as we move
along a path. It will turn out to be a basic ingredient in computing integrals
around closed paths.
In more detail, suppose that γ : [0, 1] → C is a closed path which does not
pass through 0. We would like to give a rigorous definition of the number of
times γ “goes around the origin”. Roughly speaking, this will be the change
in argument arg(γ(t)), and therein lies the difficulty, since arg(z) cannot
be defined continuously on all of C\{0}. The next Proposition shows that
we can however always define the argument as a continuous function of the
parameter t ∈ [0, 1]:
2.5. WINDING NUMBERS 49
Figure 2.6: A contour is homotopic to a counter-clockwise oriented circle.
Proposition 2.5.1. Let γ : [0, 1] → C\{0} be a path. Then there is contin-

uous function a : [0, 1] → R such that
γ(t) = |γ(t)|e2πia(t) .
Moreover, if a and b are two such functions, then there exists n ∈ Z such
that a(t) = b(t) + n for all t ∈ [0, 1]. In particular, the a(t0 ) at any t0
uniquely determines a(t) for all t.
Proof. By replacing γ(t) with γ(t)/|γ(t)| we may assume that |γ(t)| = 1 for
all t. Since γ is continuous on a compact set,√it is uniformly continuous, so
that there is a δ > 0 such that |γ(s) − γ(t)| < 3 for any s, t with |s − t| < δ.
Choose an integer n > 0 such that n√ > 1/δ so that on each subinterval
[i/n, (i + 1)/n] we have |γ(s) − γ(t)| < 3/2. Now on any half-plane in C we
√ a continuous argument function, and if |z1 | = |z2 | = 1
may certainly define
and |z1 − z2 | < 3, then the angle between z1 and z2 is at most π/3. It
follows there exists a continuous functions ai : [j/n, (j + 1)/n] → R such
that γ(t) = e2πiaj (t) for t ∈ [j/n, (j + 1)/n] (since γ([j/n, (j + 1)/n]) must
lie in an arc of length at most 2π/3). Now since e2πiaj (j/n) = e2πiaj−1 (j/n)
aj−1 (j/n) and ai (j/n) differ by an integer. Thus we can successively adjust
the aj for j > 1 by an integer (as if γ(t) = e2πiaj (t) then γ(t) = e2πi(a(t)+n)
for any n ∈ Z) to obtain a continuous function a : [0, 1] → C such that
γ(t) = e2πia(t) as required. Finally, the uniqueness statement follows because
e2πi(a(t)−b(t)) = 1, hence a(t)−b(t) ∈ Z, and since [0, 1] is connected it follows
a(t) − b(t) is constant as required.
Definition 2.5.2. If γ : [0, 1] → C\{0} is a closed path and γ(t) = |γ(t)|e2πia(t)

as in the previous lemma, then since γ(0) = γ(1) we must have a(1) − a(0) ∈
Z. This integer is called the winding number I(γ, 0) of γ around 0. It is
uniquely determined by the path γ because the function a is unique up to
an integer. By translation, if γ is any closed path and z0 is not in the image
of γ, we may define the winding number I(γ, z0 ) of γ about z0 in the same
fashion. Explicitly, if γ is a closed path with z0 ∈/ γ ∗ then let t : C → C be
given by t(z) = z − z0 and define I(γ, z0 ) = I(t ◦ γ, 0). This quantity is also
called the index of z0 with respect to γ.
Remark 2.5.3. Note that if γ : [0, 1] → U where 0 ∈ / U and there exists a

holomorphic branch L : U → C of [log(z)] on U , then I(γ, 0) = 0. Indeed in
this case we may define a(t) = Im (L(γ(t))), and since γ(0) = γ(1) it follows
a(1) − a(0) = 0 as claimed. Note also that the definition of the winding
number only requires the closed path γ to be continuous, not piecewise C 1 .
Of course, as usual, we will mostly only be interested in piecewise C 1 paths,
as these are the ones along which we can integrate functions.
We now see that the winding number has a natural interpretation in
terms of path integrals: Note that if γ is piecewise C 1 then the function a(t)
is also piecewise C 1 , since any branch of the logarithm function is in fact
differentiable where it is defined, and a(t) is locally given as Im (log(γ(t))
for a suitable branch.
Lemma 2.5.4. Let γ be a piecewise C 1 closed path and z0 ∈ C a point not

in the image of γ. Then the winding number I(γ, z0 ) of γ around z0 is given
by Z
1 dz
I(γ, z0 ) = .
2πi γ z − z0
Proof. If γ : [0, 1] → C we may write γ(t) = z0 + r(t)e2πia(t) (where r(t) =

|γ(t) − z0 | > 0 is continuous and the existence of a(t) is guaranteed by
Proposition 2.5.1). Then we have
Z Z 1
dz 1
r′ (t) + 2πir(t)a′ (t) e2πia(t) dt

= 2πia(t)
γ z − z0 0 r(t)e
Z 1
= r′ (t)/r(t) + 2πia′ (t)dt = [log(r(t)) + 2πia(t)]10
0
= 2πi(a(1) − a(0)),
since r(1) = r(0) = |γ(0) − z0 |.

Let Γ : [0, 1] → C be any simple curve such that Γ(0) = 0 and Γ(t) → ∞
as t → 1. Let U = C \ Γ∗ be the plane without the curve. It can be shown
that the winding number of any closed curve in U around 0 is 0. A bit
later we will show that the winding number around z is a function which
is constant on each component of C \ γ ∗ . It is also easy to see that if |z|
is very large, then there is a line separating z and γ ∗ and so I(γ, z) = 0.
This means that the winding number in the unbounded component of C \ γ ∗
is equal to 0. We will present a different argument a bit later in Remark
2.5.8. Since Γ does not intersect γ and connects 0 and infinity, 0 lies in the
unbounded component of CR \ γ ∗ .
All of this proves that γ 1/z = 0 for any closed curve in such domain
U and so we can define a branch of [log] in such a domain. Note that all
branches that we have discussed before are of this type. For example, for
the principle value we use Γ∗ = (−∞, 0].
In fact, one can define logarithm this way. Given such a domain U
Theorem 2.2.16 implies that there is a primitive of 1/z. We can define this
primitive to be a branch of the logarithm.
The next Proposition will be useful not only for the study of winding
numbers. We first need a definition:
Definition 2.5.5. If f : U → C is a function on an open subset U of C,
z0 ∈ U there is an r > 0 with
then we say that f is analytic on U if for every P
∞ k
B(z0 , r) ⊆ U such that there is a power
P∞ series k=0 ak (z − z0 ) with radius
of convergence at least r and f (z) = k=0 ak (z − z0 )k . An analytic function
is holomorphic, as any power series is (infinitely) complex differentiable.
Proposition 2.5.6. Let U be an open set in C and let γ : [0, 1] → U be a
closed path. If f (z) is a continuous function on γ ∗ then the function
Z
1 f (z)
If (γ, w) = dz,
2πi γ z − w
is analytic in w.
In particular, if f (z) = 1 this shows that the function w 7→ I(γ, w) is a
continuous function on C\γ ∗ , and since it is integer-valued, it is constant
on the connected components of C\γ ∗ .
Proof. We wish to show that for each z0 ∈ / γ ∗ we can find a disk B(z0 , ϵ)
within which If (γ, w) is given by a power series in (w − z0 ). Translating if
necessary we may assume z0 = 0.
Now since C\γ ∗ is open, there is some r > 0 such that B(0, 2r) ∩ γ ∗ = ∅.
We claim that If (γ, w) is holomorphic in B(0, r). Indeed if w ∈ B(0, r)
and z ∈ γ ∗ it follows that |w/z| < 1/2. Moreover, since γ ∗ is compact,

M = sup{|f (z)| : z ∈ γ ∗ } is finite, and hence
M
|f (z)wn /z n+1 | = |f (z)||z|−1 |w/z|n <(1/2)n , ∀z ∈ γ ∗ .
2r
It follows from the Weierstrass M -test that the series
∞ ∞
X f (z)wn X f (z) f (z) f (z)
= (w/z)n = (1 − w/z)−1 =
z n+1 z z z−w
n=0 n=0
viewed as a function of z, converges uniformly on γ ∗ to f (z)/(z − w). Thus

for all w ∈ B(0, r) we have
Z ∞ Z
1 f (z)dz X 1 f (z)
If (γ, w) = = dz wn ,
2πi γ z − w 2πi γ z n+1
n=0
hence If (γ, w) is given by a power series in B(0, r) (and hence is also holo-
morphic there) as required. Finally, if f = 1, then since I1 (γ, z) = I(γ, z) is
integer-valued, it follows it must be constant on any connected component
of C\γ ∗ as required.
Remark 2.5.7. Note that since the coefficients of a power series centred at a
point z0 are given by its derivatives at that point, the proof above actually
also gives formulae for the derivatives of g(w) = If (γ, w) at z0 :
Z
n! f (z)dz
g (n) (z0 ) = .
2πi γ (z − z0 )n+1
Remark 2.5.8. If γ is a closed path then γ ∗ is compact and hence bounded.
Thus there is an R > 0 such that the connected set C\B(0, R) ∩ γ ∗ = ∅. It
follows that C\γ ∗ has exactly one unbounded connected component. Since
Z
dζ
≤ ℓ(γ) sup |1/(ζ − z)| → 0
γ ζ −z ζ∈γ ∗
as z → ∞ it follows that I(γ, z) = 0 on the unbounded component of C\γ ∗ .

Let γ be a simple closed curve. There is a theorem, known as the Jordan
Curve Theorem, that C \ γ ∗ has precisely one bounded and one unbounded
component. The bounded component is the interior 8 and the unbounded
component is the exterior of γ. Moreover, the winding number around any
point from the bounded component is either 1 or −1.
8
Not to be confused with the interior of a set. If γ is a simple curve then its (topological)
interior as defined in the Metric Spaces and elsewhere is empty. Usually, it is clear from
the context what is meant by ‘interior’. Occasionally, to avoid confusion, some authors
use the term ‘inside’ instead.
Definition 2.5.9. Let γ be as above. We say that γ is positively oriented

if its winding number around any point from the bounded component is 1.
Otherwise, it is negatively oriented. Equivalently, if w is a point from the
bounded component, then γ is positively oriented if
Z
1 dz
= 1.
2πi γ z − w
By Proposition 2.5.6 orientation does not depend on the choice of w.

Finally, we will need the following topological fact that we state without
proof.
Proposition 2.5.10. Let U be a domain, γ be a simple positively oriented

curve such that it and its interior are inside U and w be a point inside γ.
Let r > 0 be such that B(w, r) is inside γ and denote the positively oriented
circle of radius r around w by γr . Then γ is homotopic to γr inside U \ {w}.
Chapter 3
Cauchy’s Formula and its

applications
3.1 Cauchy’s Integral Formula

We are now almost ready to prove one of the most important consequences of
Cauchy’s theorem – the integral formula. This formula will have incredibly
powerful consequences and one of the main things that distinguish real and
complex analysis.
Theorem 3.1.1 (Cauchy’s Integral Formula). Suppose that f : U → C is a

holomorphic function on an open set U , w ∈ U and γ is a positively oriented
closed curve such that γ ∗ and the interior of γ are inside of U . Then for all
w that are inside of γ we have
Z
1 f (z)
f (w) = dz.
2πi γ z−w
Proof. Fix w inside γ. There is r0 such that B(w, r0 ) does not intersect
γ ∗ . In particular, it means that this disc is inside γ. Take any 0 < r < r0 .
By γr we denote the positively oriented circle of radius r around w. By
Proposition 2.5.10 γ is homotopic to γr inside U \ {w}. Since f (z)/(z − w)
is holomorphic in U \ {w}, then by the Deformation Theorem 2.4.7
Z Z
1 f (z) 1 f (z)
dz = dz.
2πi γ z−w 2πi γr z−w
55
56 CHAPTER 3. CAUCHY’S FORMULA AND ITS APPLICATIONS
The last integral can be rewritten as

f (z) − f (w)
Z Z Z
1 f (z) 1 f (w) dz
dz = dz + .
2πi γr z − w 2πi γr z−w 2πi γr z−w
f (z) − f (w)
Z
1
= dz + f (w)
2πi γr z−w
f (z) − f (w)
Z
1
= dz + f (w).
2πi γr z−w
Since f is complex differentiable at z = w, the term (f (z) − f (w))/(z − w)

is bounded as r → 0, so that by the estimation lemma its integral over γr
tends to zero. Thus as r → 0 the Rpath integral around γr tends to f (w).
But since it is also equal to (2πi)−1 γ f (z)/(z − w)dz, which is independent
of r, we conclude that it must in fact be equal to f (w). The result follows.
Remark 3.1.2. The same result holds for any oriented curve γ once we weight
/ γ∗,
the left-hand side by the winding number of a path around the point w ∈
provided that f is holomorphic on the inside of γ.
Cauchy’s integral formula has a different useful form that we state as a
corollary.
Corollary 3.1.3 (Cauchy Formula for multiple curves). Let U be a bounded
domain with piecewise C 1 boundary which has finitely many components
and f be a function holomorphic in the closure of U (this means that it
is holomorphic in some open domain that contains the closure of U ). We
parametrise each boundary component of U by a contour γi in such way
that iγi′ (t) is an inward normal. This means that the ‘outer’ boundary is
positively oriented (i.e. counter-clockwise) andR all ‘inner’
P R components are
negatively oriented (i.e. clockwise)1 . Denoting ∂U = γi we have
Z
f (z)dz = 0
∂U
and Z
1 f (z)
dz = f (w), w ∈ U.
2πi ∂U z−w
Remark 3.1.4. We assume that there are only finitely many boundary com-
ponents, but with some extra assumptions that various integrals and series
make sense this assumption can be relaxed.
1
Unfortunately, this orientation of the boundary is also called the positive orientation.
3.1. CAUCHY’S INTEGRAL FORMULA 57
Figure 3.1: A domain with two inner components. We add two cuts between
γ1 and γ2 and between γ2 and γ3 . We can apply the Cauchy Integral Formula
to a curve given by concatenation of γ1 , η1− , γ2,1 which is the part of γ2 from
B to C, η2−1 , γ3 , η2 , γ2,2 which is the part of γ2 from C to B and η1 .
Proof. The idea of the proof is simple. We add a few cuts that avoid w
and connect inner boundary components to each other and to the outer
boundary in such a way that U without these extra curves becomes simply
connected. See Figure 3.1 for details.
Let γ be the boundary of this domain. We can apply Theorem 3.1.1 to
γ and obtain that Z
1 f (z)
dz = f (w).
2πi γ z − w
New curve γ is made of boundary components γi in exactly the right ori-
entation and extra cuts ηi that appear twice with different orientations, to
their contributions to the integral cancel out. This completes the proof.
Remark 3.1.5. We prove the Cauchy Integral Formula using the Deformation
Theorem which we have not proved completely (although there is a proof
in the Appendix and most of Complex Analysis textbooks will contain a
proof). It is impossible to prove Cauchy’s Formula without the Deforma-
tion or very careful geometric/topological analysis of piecewise C 1 curves2 .
Alternatively, we can restrict the set of contours γ for which we prove the
theorem. For most applications, it is enough to consider only circular curves.
2
This is non-trivial, since, for example, they could have infinitely many points of self-
intersection or they could intersect a straight line at infinitely many points
Then the only thing that we have to prove is that integrals along two circles
are the same. The annular domain can be split into the union of star-like
domains for which we know the existence of a primitive, so these smaller
contour integrals vanish and this way we can prove the result for circles.
See Figure 3.1.5 for an illustration and some extra explanations.
Figure 3.2: An annular domain between two positively oriented circles γ1

and γ2 can be split into four sectors S1 , . . . , S4 . All domains Si are star-like
so integrals along their boundaries
R R vanish. Integrals along extra cuts cancel
out and we are left with γ1 f + γ − f = 0. This means that integrals along
both circles are equal.
3.2 Applications of the Integral Formula

Remark 3.2.1. Note that Cauchy’s integral formula can be interpreted as
saying the value of f (w) for w inside the circle is obtained as the “convolu-
tion” of f and the function 1/(z − w) on the boundary circle. Since the func-
tion 1/(z −w) is infinitely differentiable one can use this to show that f itself
is infinitely differentiable as we will shortly show. If you take the Integral
Transforms, you will see convolution play a crucial role in the theory of
transforms. In particular, the convolution of two functions often inherits
the “good” properties of either. We next show that in fact the formula
implies a strong version of Taylor’s Theorem.
Corollary 3.2.2. If f : U → C is holomorphic on an open set U , then for

any z0 ∈ U , f (z) is equal to its Taylor series at z0 and the Taylor series
converges on any open disk centred at z0 lying in U . Moreover the derivatives
3.2. APPLICATIONS OF THE INTEGRAL FORMULA 59
of f at z0 are given by
Z
n! f (z)
f (n) (z0 ) = dz. (3.2.1)
2πi γ(a,r) (z − z0 )n+1
For any a ∈ C, r ∈ R>0 with z0 ∈ B(a, r).
Proof. This follows immediately from the Integral formula, the proof of
Proposition 2.5.6, and Remark 2.5.7. The integral formulae of Equation
3.2.1 for the derivatives of f are also referred to as Cauchy’s Integral For-
mulae.
Definition 3.2.3. Recall that a function which is locally given by a power

series is said to be analytic. We have thus shown that any holomorphic
function is actually analytic, and from now on we may use the terms inter-
changeably (as you may notice is common practice in many textbooks).
One famous application of the Integral formula is known as Liouville’s

theorem, which will give an easy proof of the Fundamental Theorem of
Algebra3 . We say that a function f : C → C is entire if it is complex
differentiable on the whole complex plane.
Theorem 3.2.4 (Liouville). Let f : C → C be an entire function. If f is

bounded then it is constant.
Proof. Suppose that |f (z)| ≤ M for all z ∈ C. Let γR (t) = Re2πit be the
circular path centred at the origin with radius R. Then for R > |w| the
integral formula shows
Z
1 1 1
|f (w) − f (0)| = f (z) − dz
2πi γR z−w z
Z
1 wf (z)
= dz
2π γR z(z − w)
2πR wf (z)
≤ sup |
2π z:|z|=R z(z − w)
M |w| M |w|
≤R = ,
R(R − |w|) R − |w|
3
Which, when it comes down to it, isn’t really a theorem in algebra. The most “al-
gebraic” proof of that I know uses Galois theory, which you can learn about in Part B.
But it also uses the fact that any real polynomial of odd degree has a real root. This is
a simple corollary of the intermediate value theorem, which in its turn depends on the
completeness of R which is not an algebraic property at all.
Thus letting R → ∞ we see that |f (w) − f (0)| = 0, so that f is constant an

required.
Remark 3.2.5. Liouville’s theorem is one more manifestation of the unique

properties of holomorphic functions. First of all, there is nothing like this
in real analysis. For example, f (x) = 1/(1 + x2 ) is real-analytic in the en-
tire R, but it is bounded. Secondly, this is one of the first examples of a
dichotomy which often appears in complex analysis. In many cases objects
are either as good as they could be or as bad as they could be but nothing
in between. For example, here we see that an entire function f is either
constant or there is zn → ∞ such that f (zn ) → ∞. Later we will see that
the behaviour at infinity is even more interesting.
Theorem 3.2.6. Suppose that p(z) = nk=0 ak z k is a non-constant polyno-

P
mial where ak ∈ C and an ̸= 0. Then there is a z0 ∈ C for which p(z0 ) = 0.
Proof. By rescaling p we may assume that an = 1. If p(z) ̸= 0 for all z ∈ C

it follows that f (z) = 1/p(z) is an entire function (since p is clearly entire).
We claim that f is bounded. Indeed since it is continuous it is bounded on
any disc B̄(0, R), so it suffices to show that |f (z)| → 0 as z → ∞, that is,
to show that |p(z)| → ∞ as z → ∞. But we have
n−1 n−1 n−1

! !
X X ak X |ak |
|p(z)| = |z n + ak z k | = |z n | |1 + | ≥ |z n | 1 − .
z n−k |z|n−k
k=0 k=0 k=0
Since 1/|z|m → 0 as |z| → ∞ for any m ≥ 1 it follows that for sufficiently

Pn−1 |ak |
large |z|, say |z| ≥ R, we will have 1 − k=0 |z|n−k
≥ 1/2. Thus for |z| ≥ R
we have |p(z)| ≥ 12 |z|n . Since |z|n clearly tends to infinity as |z| tends to
infinity, it follows |p(z)| → ∞ as required.
Remark 3.2.7. The crucial point of the above proof is that one term of the
polynomial – the leading term in this case– dominates the behaviour of the
polynomial for large values of z. All proofs of the fundamental theorem hinge
on essentially this point. Note that p(z0 ) = 0 if and only if p(z) = (z−z0 )q(z)
for a polynomial q(z), thus by induction on the degree we see that the
theorem implies that a polynomial over C factors into a product of degree
one polynomials.
We end this section with a kind of converse to Cauchy’s theorem:
3.2. APPLICATIONS OF THE INTEGRAL FORMULA 61
Theorem 3.2.8 (Morera). Suppose that f : U → C is a continuous function

R an open subset U ⊆ C. If for any closed path γ : [a, b] → U we have
on
γ f (z)dz = 0, then f is holomorphic.
Proof. By Theorem 2.2.16 we know that f has a primitive F : U → C. But

then F is holomorphic on U and so infinitely differentiable on U , thus in
particular f = F ′ is also holomorphic.
Remark 3.2.9. One can prove variants of the above theorem: If U is a star-
like domain for example, then our proof of Cauchy’s theorem for such do-
mains shows thatR f : U → C has a primitive (and hence will be differentiable
itself) provided
R T f (z)dz = 0 for every triangle in U . In fact, the assump-
tion that T f (z)dz = 0 for all triangles whose interior lies in U suffices to
imply f is holomorphic for any open subset U : To show f is holomorphic
on U , it suffices to show that f is holomorphic on B(a, r) for each open
disk B(a, r) ⊂ U . But this follows from the above as disks are star-like (in
fact convex). It follows that we can characterize the fact that f : U → C
is holomorphic on U by an integral condition: f : U → C is holomorphic if
R all triangles T which bound a solid triangle T with T ⊂ U ,
and only if for
the integral T f (z)dz = 0.
This characterization of the property of being holomorphic has some
important consequences. We first need a definition:
Definition 3.2.10. Let U be an open subset of C. If (fn ) is a sequence of

functions defined on U , we say fn → f uniformly on compacts if for every
compact subset K of U , the sequence (fn|K ) converges uniformly to f|K .
Note that in this case, f is continuous if the fn are: Indeed to see that f is
continuous at a ∈ U , note that since U is open, there is some r > 0 with
B(a, r) ⊆ U . But then K = B̄(a, r/2) ⊆ U and fn → f uniformly on K,
whence f is continuous on K, and so certainly it is continuous at a.
Example 3.2.11. Convergence of power series f (z) = ∞ n

P
k=0 an z is a ba-
sic example of convergence on compacts: if R is the radius of convergences
of f (z) the partial sums sn (z) of the power series B(0, R) converge uni-
formly on compacts in B(0, R). The convergence
P∞ n is not necessarily uniform
on B(0, R),Sas the example f (z) = n=0 z shows. Nevertheless, since
B(0, R) = r<R B̄(0, r) is the union of its compact subsets, many of the
good properties of the polynomial functions sn (z) are inherited by the power
series because the convergence is uniform on compact subsets.
Proposition 3.2.12. Suppose that U is a domain and the sequence of holo-

morphic functions fn : U → C converges to f : U → C uniformly on com-
pacts in U . Then f is holomorphic.
Proof. Note by the above that f is continuous on U . Since the property

of being holomorphic is local, it suffices to show for each w ∈ U that there
is a ball B(w, r) ⊆ U within which f is holomorphic. Since U is open, for
any such w we may certainly find r > 0 such that B(w, r) ⊆ U . Then as
B(w, r) is convex, Cauchy’s theorem for a star-like domain shows that for
every
R closed path γ : [a, b] → B(w, r) whose image lies in B(w, r) we have
γ nf (z)dz = 0 for all n ∈ N.
∗
But γ = γ([a, b]) is a compact subset of U , hence fn → f uniformly on
∗
γ . It follows that Z Z
0= fn (z)dz → f (z)dz,
γ γ
so that the integral of f around any closed path in B(w, r) is zero. But then
Morera’s Theorem 3.2.8 shows that f is holomorphic.
3.3 The identity theorem

The fact that any complex differentiable function is in fact analytic has some
very surprising consequences – the most striking of which is perhaps cap-
tured by the “Identity theorem”. This says that if f, g are two holomorphic
functions defined on a domain U and we let S = {z ∈ U : f (z) = g(z)} be
the locus on which they are equal, then if S has a limit point in U it must
actually be all of U . Thus for example if there is a disk B(a, r) ⊆ U on
which f and g agree (no matter how small r is), then in fact they are equal
on all of U ! The key to the proof of the Identity theorem is the following
result on the zeros of a holomorphic function:
Proposition 3.3.1. Let U be an open set and suppose that g : U → C is

holomorphic on U . Let S = {z ∈ U : g(z) = 0}. If z0 ∈ S then either z0 is
isolated in S (so that g is non-zero in some disk about z0 except at z0 itself )
or g = 0 on a neighbourhood of z0 . In the former case there is a unique
integer k > 0 and holomorphic function g1 such that g(z) = (z − z0 )k g1 (z)
where g1 (z0 ) ̸= 0.
Proof. Pick any z0 ∈ U with g(z0 ) = 0. Since g is analytic at z0 , if we pick

3.3. THE IDENTITY THEOREM 63
r > 0 such that B̄(z0 , r) ⊆ U , then we may write

∞
X
g(z) = ck (z − z0 )k ,
k=0
for all z ∈ B(z0 , r) ⊆ U , where the coefficients ck are given as in Theorem

3.2.2. Now if ck = 0 for all k, it follows that g(z) = 0 for all z ∈ B(0, r).
Otherwise, we set k = min{n ∈ N : cn ̸= 0} (where since g(z0 ) = 0 we
have c0 = 0 so that k ≥ 1). Then if we let g1 (z) = (z − z0 )−k g(z), clearly
holomorphic on U \{z0 }, but since in B(z0 , r) we have we have
g1 (z) is P
g1 (z) = ∞ n
n=0 ck+n (z − z0 ) , it follows if we set g1 (z0 ) = ck ̸= 0 then g1
becomes a holomorphic function on all of U . Since g1 is continuous at z0
and g1 (z0 ) ̸= 0, there is an ϵ > 0 such that g1 (z) ̸= 0 for all z ∈ B(z0 , ϵ).
But (z − z0 )k vanishes only at z0 , hence it follows that g(z) = (z − z0 )k g1 (z)
is non-zero on B(a, ϵ)\{z0 }, so that z0 is isolated.
Finally, to see that k is unique, suppose that g(z) = (z − z0 )k g1 (z) =
(z−z0 )l g2 (z) say with g1 (z0 ) and g2 (z0 ) both nonzero. If k < l then g(z)/(z−
z0 )k = (z − z0 )l−k g2 (z) for all z ̸= z0 , hence as z → z0 we have g(z)/(z −
z0 )k → 0, which contradicts the assumption that g1 (z) ̸= 0. By symmetry,
we also cannot have k > l so k = l as required.
Remark 3.3.2. The integer k in the previous proposition is called the multi-
plicity of the zero of g at z = z0 (or sometimes the order of vanishing).
Theorem 3.3.3. (Identity theorem): Let U be a domain and suppose that

f1 , f2 are holomorphic functions defined on U . Then if S = {z ∈ U : f1 (z) =
f2 (z)} has a limit point in U , we must have S = U , that is f1 (z) = f2 (z)
for all z ∈ U .
Proof. Let g = f1 − f2 , so that S = g −1 ({0}). We must show that if S

has a limit point then S = U . Since g is clearly holomorphic in U , by
Proposition 3.3.1 we see that if z0 ∈ S then either z0 is an isolated point
of S or it lies in an open ball contained in S. It follows that S = V ∪ T
where T = {z ∈ S : z is isolated} and V = int(S) is open. But since g
is continuous, S = g −1 ({0}) is closed in U , thus V ∪ T is closed, and so
ClU (V ), the closure4 of V in U , lies in V ∪ T . However, by definition, no
limit point of V can lie in T so that ClU (V ) = V , and thus V is open and
closed in U . Since U is connected, it follows that V = ∅ or V = U . In the
4
I use the notation ClU (V ), as opposed to V̄ , to emphasize that I mean the closure of
V in U , not in C, that is, ClU (V ) is equal to the union of V with the limits points of V
which lie in U .
former case, all the zeros of g are isolated and S has no limit points. In the
latter case, V = S = U as required.
Remark 3.3.4. The requirement in the theorem that S have a limit point
lying in U is essential: If we take U = C\{0} and f1 = exp(1/z) − 1 and
f2 = 0, then the set S is just the points where f1 vanishes on U . Now the
zeros of f1 have a limit point at 0 ∈/ U since f1 (1/(2πin)) = 0 for all n ∈ N,
but certainly f1 is not identically zero on U !
3.4 Isolated singularities

We now wish to study singularities of holomorphic functions. We are mainly
interested in isolated singularities.
Definition 3.4.1. Let f : U → C be a function, where U is open. We say

that z0 ∈ Ū is a regular point of f if f is holomorphic at z0 . Otherwise, we
say that z0 is singular.
We say that z0 is an isolated singularity if f is holomorphic on B(z0 , r)\{z0 }
for some r > 0.
Remark 3.4.2. Note that the definition of the pole of order n is very similar
to the definition of the zero with multiplicity n.
Definition 3.4.3. A function on an open set U which has only isolated

singularities all of which are poles is called a meromorphic function on U .
(Thus, strictly speaking, it is a function only defined on the complement of
the poles in U .)
We will use the following classification of isolated singularities:
Definition 3.4.4. Let z0 be an isolated singularity of function f . We say

that z0 is
• An removable singularity if there is a function g holomorphic in B(z0 , r)

for some r > 0 such that f (z) = g(z) in B(z0 , r) \ {z0 }.
• A pole of order n if there is a function g holomorphic in B(z0 , r)

for some r > 0 such that g(z0 ) ̸= 0 and f (z) = (z − z0 )−n g(z) in
B(z0 , r) \ {z0 }.
• An essential singularity otherwise.

3.4. ISOLATED SINGULARITIES 65
Example 3.4.5. Let f (z) = z/z. This function is not even defined at z = 0
but it is equal to 1 otherwise and clearly holomorphic in C \ {0}. So z0 = 0
is an isolated singularity. This is clearly a removable singularity since g = 1
is holomorphic everywhere and equal to f outside of 0.
Examples of removable singularities can be more involved:
Example 3.4.6. Let f (z) = sin(z)/z. Again, it is easy to see that z0 = 0
is an isolated singularity. It is removable since the function
z2 z4
g(z) = 1 − + − ...
3! 5!
is entire and coincides with f outside of z0 .
In a similar way we can construct a pole.
Example 3.4.7. Let f (z) = sin(z)/z n+1 . This function has a pole of order
n at z0 = 0. It is easy to see that for z ̸= 0 we have f (z) = g(z)/z n where g
is the entire function from the previous example. Obviously g(0) = 1 ̸= 0.
A typical essential singularity is given by the following example.
Example 3.4.8. Let f (z) = sin(1/z). This function is holomorphic in
C\{0}. It is not too difficult to show that 0 is neither a removable singularity
nor a pole, so it must be an essential singularity. A bit later we will have
alternative characterizations of singularities that will be more suitable for
determining their types.
The main tool for studying singularities is the following theorem which
is a generalization of Taylor’s theorem.
Theorem 3.4.9 (Laurent’s Theorem). Suppose that 0 < r < R and
A = A(z0 , r, R) = {z : r < |z − z0 | < R}
is an annulus centred at z0 . If f : U → C is holomorphic on an open set U

which contains Ā, then there exist cn ∈ C such that
∞
X
f (z) = cn (z − z0 )n , ∀z ∈ A.
n=−∞
The series converges for all z ∈ A and it converges uniformly for all z ∈
A(z0 , r′ , R′ ) where r < r′ < R′ < R. The series is called the Laurent series
of f .
Moreover, the cn are unique and are given by the following formulae:
Z
1 f (z)
cn = dz,
2πi γs (z − z0 )n+1
where s ∈ [r, R] and for any s > 0 we set γs (t) = z0 + se2πit .
Remark 3.4.10. Since we have a formula for Laurent coefficients in terms of
f , it means that the Laurent expansion is unique. If two series converge to
the same function then they must coincide term-by-term.
Remark 3.4.11. By the Deformation Theorem 2.4.7 circular contour γs can
be replaced by any other curve homotopic to γs . In particular, by any other
positively oriented simple curve in A.
Remark 3.4.12. If f is holomorphic in B(z0 , R) then for n < 0 the integrand
in the formula for cn is holomorphic, hence cn = 0 for all n < 0. For n ≥ 0
formulas for cn are exactly the same as in Taylor’s theorem so in this case
the Laurent series is the same as the Taylor series.
Proof. By Corollary 3.1.3 for any w ∈ A we have

Z Z
1 f (z) 1 f (z)
f (w) = dz − dz.
2πi γR z − w 2πi γr z − w
Note that here both boundary components are counter-clockwise oriented
but in Corollary 3.1.3 the inner component is clockwise oriented. This is
compensated by the minus sign in front of the second integral.
But now the result follows in the same way as we showed holomorphic
functions were analytic: if we fix w, then, for |w| < R = |z| we have
∞
1 X
= wn /z n+1
z−w
n=0
and for an |w| < R − ϵ = |z| − ϵ the series converges uniformly z for any
ϵ > 0. It follows that
∞
f (z)wn X Z f (z)
Z Z X
f (z)
dz = n+1
dz = n+1
dz wn .
γR z − w γR z γR z
n=0 n≥0
for all w ∈ A. Similarly since for |z| = r < |w| we have5

−∞
1 X X
= z n /wn+1 = wn /z n+1 ,
w−z
n≥0 n=−1
5
Note the sign change.
again converging uniformly when |z| = r < |w| − ϵ for ϵ > 0, we see that
Z Z −∞ −∞ Z
f (z) X
n n+1
X f (z)
dz = f (z)w /z dz = dz wn .
γr w−z γr n=−1 γr z n+1
n=−1
Thus taking (cn )n∈Z as in the statement of the theorem, we see that
Z Z
1 f (z) 1 f (z) X
f (w) = dz − dz = cn z n ,
2πi γR z−w 2πi γr z−w
n∈Z
as required. To P see that the cn are unique, one checks using uniform con-
vergence that if n∈Z dn z n is any series expansion for f (z) on A, then the
dn must be given by the integral formulae above.
Finally, to see that the cn can be computed using any circular contour
γs , note that if r ≤ s1 < s2 ≤ R then f /(z − z0 )n+1 is holomorphic in
A hence by the Deformation Theorem 2.4.7 integrals are the same for all
circular contours.
R f (z)
Remark 3.4.13. Note that the above proof shows that the integral γR z−w dz
R f (z)
defines a holomorphic function of w in B(z0 , R), while γr z−w dz defines a
holomorphic function of w on C\B(z0 , r). Thus we have actually expressed
f (w) on A as the difference of two functions which are holomorphic on
B(z0 , R) and C\B̄(z0 , r) respectively.
It is possible to push the previous theorem to the limit and apply it to
a punctured disc.
Corollary 3.4.14 (Laurent series in a punctured disc). If f : U → C

is a holomorphic function and z0 is an isolated singularity, then f has a
Laurent expansion on a punctured disc B(z0 , R)\{z0 } for any R such that
B(z0 , R)\{z0 } ⊂ U .
Proof. Let us take some r0 such that 0 < r < R and apply Theorem 3.4.9
to A = A(z0 , r, R). Note that the coefficients cn can be written in terms of
integrals along γR , hence they do not depend on r. By sending r to zero we
can see that the Laurent series converges in
[
B(z0 , R) = A(z0 , r, R)
0<r<R
and uniformly in A(z0 , r, R′ ) for any 0 < r < R′ < R.

cn (z − z0 )n
P
Definition 3.4.15. Let z0 be an isolated singularity of f and
be its Laurent’s expansion. Its principal part of f at z0 is the sum of terms
with negative powers and denoted Pz0 f . Namely,
−1
X ∞
X
Pz0 f (z) = cn (z − z0 )n = c−n (z − z0 )−n .
−∞ 1
Proposition 3.4.16. The principal part of f at z0 converges on C \ {z0 }

and converges uniformly on C \ B(z0 , r).
Proof. This follows immediately from the proof of Theorem 3.4.9. In the
proof we have used that if f is holomorphic on A(z0 , r, R) then the principal
part converges uniformly on {z : |z − z0 | > r′ } for any r′ > r. Since in the
case of isolated singularities we can take r to be arbitrarily small, the claim
follows immediately.
Note that something very interesting happens with the term of order
n = −1. Theorem 3.4.9 tells us that the coefficient
Z
1
c−1 = f (z)dz.
2πi γs
This should not really be surprising. If we have a function f given by a

cn (z − z0 )n which converges uniformly in an annulus containing γs ,
P
series
then we can integrate it term by term. For all n ̸= −1 the powers (z − z0 )n
have a well defined primitive (z − z0 )n+1 /(n + 1) so they integrate to zero.
But as we have discussed before the integral of 1/(z − z0 ) along γs is 2πi so
Z Z X Z
1 1 n 1 c−1
f (z)dz = cn (z − z0 ) = dz = c−1 .
2πi γs 2πi γs 2πi γs z − z0
This observation motivates the following definition.
Definition 3.4.17. Let z0 be an isolated singularity of f . Then the residue

of f at 0 is defined as the coefficient c−1 of the Laurent expansion and
denoted by Resz0 f or Res(f, z0 ).
Laurent’s series allows us to characterize isolated singularities in terms

of the series coefficients.
Theorem 3.4.18 (CharacterizationP∞ of isolatednsingularities). Let z0 be an

isolated singularity of f . Let −∞ cn (z − z0 ) be its Laurent expansion.
Then z0 is
• A removable singularity if cn = 0 for all n < 0. Equivalently, the

principal part vanishes.
• A pole of order n is c−n ̸= 0 and ck = 0 for all k < −n. Equivalently,

the principal part is non-trivial but contains only a finite number of
non-zero terms.
• An essential singularities if there are arbitrary large n such that c−n =

0. Equivalently, the principal part contains infinitely many non-zero
terms.
Remark 3.4.19. This is very similar to the characterization of a zero of

multiplicity n. In this case we have a Taylor series ∞ − z0 )k with
P
a
n=0 k (z
ak = 0 for all k < n and an ̸= 0.
Remark 3.4.20. Some authors define types of singularities in terms of the
Laurent expansion.
Proof. If there is no principal part, then we have an ordinary power series

which converges to a function g which is analytic in some disc B(z0 , R)
around z0 . Clearly, f (z) − g(z) for all 0 < |z| < R, so z0 is removable.
If z0 is removable, then there is a function g which is holomorphic at
z0 and coincides with f in B(z0 , R) \ {z0 }. Hence their Laurent expansion
coefficients are equal, but for a holomorphic function g all coefficients with
negative n are given by integrals of a holomorphic function (z −z0 )−n−1 g(z),
but by Cauchy’s theorem such integrals vanish.
If the Laurent expansion is of the form
c−n
f (z) = + · · · + c0 + c1 (z − z0 ) + . . . ,
(z − z0 )n
with c−n ̸= 0 then f (z) = (z − z0 )−n g(z) where
g(z) = c−n + c−n+1 (z − z0 ) + . . .
which is analytic in some neighbourhood of z0 .

If z0 is a pole of order n, then there is a holomorphic g such that g(z0 ) ̸= 0
and f (z) = (z − z0 )−n g(z). Writing the Taylor series of g we see that the
Laurent expansion of f is of the form
g(z0 ) g ′ (z0 )
f (z) = + ....
(z − z0 )n (z − z0 )n−1
The last part follows trivially from the first two.

Finally, it is possible to characterize types of singularities by looking at

the local behaviour of f near z0 . We describe it in a series of three theorems.
Theorem 3.4.21 (Riemann’s removable singularity theorem). Suppose that

U is an open subset of C and z0 ∈ U and suppose that f : U \ {z0 } → C is
holomorphic. Then z0 is a removable singularity if and only if f is bounded
near z0 .
Proof. One direction is trivial. If z0 is removable, then there is holomorphic

g such that g(z) = f (z) on B(z0 , r) \ {z0 } for some r > 0. This proves that
f (z) = g(z) → g(z0 ) as z → z0 . In particular, f is bounded.
Now, let us assume that f is bounded and define h(z) by
(
(z − z0 )2 f (z), z ̸= z0 ;
h(z) =
0, z = z0
Then clearly h(z) is holomorphic on U \{z0 }, using the fact that f is and
standard rules for complex differentiability. On the other hand, at z = z0
we see directly that
h(z) − h(z0 )
= (z − z0 )f (z) → 0
z − z0
as z → z0 since f is bounded near z0 by assumption. It follows that h is
in fact holomorphic everywhere in U . But then if we chose r > 0 is such
that B̄(z0 , r) ⊂ U , then by Corollary 3.2.2 h(z) is equal to its Taylor series
centred at z0 , thus
X∞
h(z) = ak (z − z0 )k .
k=0
But
P∞ since we havek h(z0 ) = h′ (z
0 ) = 0 we see a0 = a1 = 0, and hence
k=0 ak+2 (z − z0 ) defines a holomorphic function in B(z0 , r). Since this
clearly agrees with f (z) on B(z0 , r)\{0}, we see that by redefining f (z0 ) =
a2 , we can extend f to a holomorphic function on all of U as required.
Remark 3.4.22. This is yet another reminder that complex analysis is very
different from the real one. The notions of isolated and removable singular-
ities make sense in R as well but there is nothing similar to Riemann’s
removable singularity theorem. Let us consider the following functions on
R \ {0}: f (x) = sign(x), g(x) = sin(1/x) and h(x) = |x|. All of them are
analytic outside of 0 and bounded in a neighbourhood of 0. Clearly, f has
left and right limits but they are different, so there is no way to define f (0)
so that it becomes continuous. Function g oscillates near 0, so it does not

have even one-sided limits. So the singularity is not removable. With h the
situation is a bit different, h has a limit at 0 and it can be made into a con-
tinuous function, but it will not be differentiable. So again, the singularity
is not removable. None of these scenarios is possible for complex functions.
Lemma 3.4.23. Let f be a holomorphic function in a neighbourhood of z0 .

Then z0 is a pole if and only if |f (z)| → ∞ as z → z0 . Moreover, in this
case, the function (
1/f (z), z ̸= z0 ;
h(z) =
0, z = z0
is holomorphic in a neighbourhood of z0 and the multiplicity of its zero at
z0 is equal to the order of the pole of f .
Proof. Let us assume that f has a pole of order n. Then by Theorem 3.4.18
it Laurent expansion is
c−n
f (z) = + · · · + c0 + c1 (z − z0 ) + . . . .
(z − z0 )n
This expansion can be rewritten as

1
(c−n + c−n+1 (z − z0 ) + . . . ) .
(z − z0 )n
The series converges to a function analytic at z0 which we denote by g.

Then f −1 = (z − z0 )n g −1 (z). Since g(z0 ) = c−n ̸= 0 the function h = 1/g is
holomorphic in a neighbourhood of z0 and h(z0 ) ̸= 0. This proves that 1/f
has a removable singularity and after removing this singularity it has a zero
of multiplicity n.
Let us assume that |f | → ∞. Then the function 1/f → 0 and by
Theorem 3.4.21 it extends to a holomorphic function h which has a zero at
z0 . We know that the zero must be of finite order, say n ≥ 1, so h(z) =
(z − z0 )n g(z) where g is holomorphic with g(z0 ) ̸= 0. Then f (z) = (z −
z0 )−n g −1 (z), so by the definition f has a pole of order n.
Remark 3.4.24. Theorems 3.4.21 and 3.4.23 show that removable singulari-
ties and poles are not that dissimilar. In both cases, f converges as z → z0
in one case to a finite limit in the other to infinity. By considering the
extended complex plane C∞ (also known as the Riemann sphere) the dis-
tinction completely disappears, in both cases f can be extended to a holo-
morphic C∞ -valued function.
At this stage, one might guess that the essential singularity covers all
other types of behaviour near z0 . But we already know that this is not quite
the case, for example, Theorem 3.4.21 shows that f can not stay bounded
without having a limit. The following theorem shows that the behaviour
near an essential singularity is quite peculiar: it must oscillate in the worst
imaginable way. This is yet another example showing the difference between
real and complex analysis. In complex analysis, function either has a limit
or fails to have a limit in the most spectacular way.
Theorem 3.4.25 (Casorati-Weierstrass or Weierstrass Theorem). Let U

be an open subset of C and let a ∈ U . Suppose that f : U \{a} → C is a
holomorphic function with an isolated essential singularity at a. Then for
all ρ > 0 with B(a, ρ) ⊆ U , the set f (B(a, ρ)\{a}) is dense in C, that is,
the closure of f (B(a, ρ)\{a}) is all of C.
Proof. Suppose, for the sake of a contradiction, that there is some ρ > 0
such that z0 ∈ C is not a limit point of f (B(a, ρ)\{a}). Then the function
g(z) = 1/(f (z)−z0 ) is bounded and non-vanishing on B(a, ρ)\{a}, and hence
by Riemann’s removable singularity theorem, it extends to a holomorphic
function on all of B(a, ρ). But then f (z) = z0 + 1/g(z) has at most a pole
at a which is a contradiction.
Remark 3.4.26. In fact, much more is true: Picard showed that if f has an
isolated essential singularity at z0 then in any open disk about z0 the function
f takes every complex value infinitely often with at most one exception. The
example of the function f (z) = exp(1/z), which has an essential singularity
at z = 0 shows that this result is best possible, since f (z) ̸= 0 for all z ̸= 0.
We conclude this section with one of the most useful theorems of the
course – it is extremely powerful as a method for computing integrals, as you
will see in this course and many others. We will discuss these applications
at the end of the course.
Theorem 3.4.27 (Residue theorem). Suppose that U is an open set in C

and γ is a simple closed that is contained in U together with its inside.
Suppose that f is holomorphic on U \ S where S is a finite set of isolated
singularities of f . We also assume that f has no singularities on γ ∗ that is
S ∩ γ ∗ = ∅. Then
Z
1 X
f (z)dz = I(γ, a)Resa (f )
2πi γ
a∈S
3.5. THE ARGUMENT PRINCIPLE 73
Proof. For each a ∈ S let Pa (f )(z) = −∞ n

P
n=−1 cn (a)(z − a) be the principal
part of f at a, a holomorphic function on C\{a}. Then by definition of
the difference f − Pa (f ) is holomorphic6 at a ∈ S, and thus g(z) =
Pa (f ), P
f (z) − a∈SR Pa (f ) is holomorphic on all of U . But then by Theorem 2.4.7
we see that γ g(z)dz = 0, so that
Z XZ
f (z)dz = Pa (f )(z)dz.
γ a∈S γ
Note that to apply the Homotopy Cauchy Theorem we use the fact that if
U contains γ and its inside then γ is null-homotopic.
By Proposition 3.4.16 the series Pa (f ) converges uniformly on γ ∗ so that
Z −∞
Z X ∞ Z
X c−n (a)dz
Pa (f )dz = cn (a)(z − a)n = n
γ γ n=−1 n=1γ (z − a)
Z
c−1 (a)dz
= = I(γ, a)Resa (f ),
γ z−a
since for n > 1 the function (z − a)−n has a primitive on C\{a}. The result
follows.
Remark 3.4.28. In practice, in applications of the residue theorem, the wind-

ing numbers I(γ, a) will be simple to compute in terms of the argument of
(z − a) – in fact most often they will be 0 or ±1 as we will usually apply the
theorem to integrals around simple closed curves.
3.5 The argument principle

Lemma 3.5.1. Suppose that f : U → C is meromorphic and has a zero of
order k or a pole of order k at z0 ∈ U . Then f ′ (z)/f (z) has a simple pole
at z0 with residue k or −k respectively.
Proof. If f (z) has a zero of order k we have f (z) = (z − z0 )k g(z) where g(z)
is holomorphic near z0 and g(z0 ) ̸= 0. It follows that
f ′ (z) k g ′ (z)
= + ,
f (z) z − z0 g(z)
6
This is a slight abuse of notations since strictly speaking f is not defined at a, so
f − Pa f is not even defined at a. On the other hand, it has a removable singularity there,
so we should remove this singularity and use the extended function instead.
and since g(z) ≠ 0 near z0 it follows g ′ (z)/g(z) is holomorphic near z0 , so

that the result follows. The case where f has a pole at z0 is similar.
Remark 3.5.2. Note that if U is an open set on which one can define a holo-
morphic branch L of [log(z)] then g(z) = L(f (z)) has g ′ (z) = f ′ (z)/f (z).
Thus integrating f ′ (z)/f (z) along a path γ will measure the change in ar-
gument around the origin of the path f (γ(t)). The residue theorem allows
us to relate this to the number of zeros and poles of f inside γ, as the next
theorem shows:
Theorem 3.5.3. (Argument principle): Suppose that f is meromorphic on

U and γ be a simple positively oriented contour such that the contour and its
inside are contained in U . We assume that f has no zeroes or poles on γ ∗ .
If N is the number of zeros (counted with multiplicity) and P is the number
of poles (again counted with multiplicity) of f inside γ then
f ′ (z)
Z
1
N −P = dz.
2πi γ f (z)
Moreover, this is the winding number of the path Γ = f ◦ γ about the origin.
Proof. It is easy to check that I(γ, z) is 1 if |z − a| ≤ 1 and is 0 otherwise.

Since Lemma 3.5.1 shows that f ′ (z)/f (z) has simple poles at the zeros and
poles of f with residues the corresponding orders the result immediately
from Theorem 3.4.27.
For the last part, note that the winding number of Γ(t) = f (γ(t)) about
zero is just
1
f ′ (z)
Z Z Z
1 1
dw = f ′ (γ(t))γ ′ (t)dt = dz.
f ◦γ w 0 f (γ(t)) γ f (z)
The argument principle is very useful – we use it here to establish some

important results.
Theorem 3.5.4 (Rouché). Suppose that f and g are holomorphic functions

on an open set U in C and γ be a simple contour that is contained inside of
U together with its interior. If |f (z)| > |g(z)| for all z ∈ γ ∗ then f and f + g
have the same change in argument around γ, and hence the same number of
zeros (counted with multiplicities) inside of γ.
3.5. THE ARGUMENT PRINCIPLE 75
Proof. Let γ(t) = a + re2πit be a parametrization of the boundary circle of

B(a, r). We need to show that h = (f + g)/f = 1 + g/f has the same total
number of zeros as poles inside of γ. Note that the assumption |f (z)| > |g(z)|
implies that h has no zeroes or poles on γ ∗ . But by the argument principle,
the difference between the number of zeroes and poles is equal to the winding
number of Γ(t) = h(γ(t)) about zero. Since, by assumption, for z ∈ γ ∗ we
have |g(z)| < |f (z)| and so |g(z)/f (z)| < 1, the image of Γ lies entirely
in B(1, 1) and thus in the half-plane {z : Re (z) > 0}. Hence picking the
principal branch of [log] defined on this half-plane, we see that the integral
Z
dz
= Log(h(γ(1)) − Log(h(γ(0)) = 0
Γ z
as required.
Remark 3.5.5. Rouche’s theorem can be useful in counting the number of

zeros of a function f – one tries to find an approximation to f whose zeros
are easier to count and then by Rouche’s theorem obtain information about
the zeros of f . Just as for the argument principle above, it also holds for
closed paths which having winding number 1 about their inside.
Example 3.5.6. Suppose that P (z) = z 4 + 5z + 2. Then on the circle
|z| = 2, we have |z|4 = 16 > 5.2 + 2 ≥ |5z + 2|, so that if g(z) = 5z + 2 we see
that P − g = z 4 and P have the same number of roots in B(0, 2). It follows
by Rouche’s theorem that the four roots of P (z) all have a modulus less than
2. On the other hand, if we take |z| = 1, then |5z +2| ≥ 5−2 = 3 > |z 4 | = 1,
hence P (z) and 5z + 2 have the same number of roots in B(0, 1). It follows
P (z) has one root of modulus less than 1, and 3 of modulus between 1 and
2.
Theorem 3.5.7 (Open mapping theorem). Suppose that f : U → C is holo-
morphic and non-constant on a domain U . Then for any open set V ⊂ U
the set f (V ) is also open.
Proof. Suppose that w0 ∈ f (V ), say f (z0 ) = w0 . Then g(z) = f (z) − w0
has a zero at z0 which, since f is non-constant, is isolated. Thus we may
find an r > 0 such that g(z) ̸= 0 on B̄(z0 , r)\{z0 } ⊂ U and in particular
since ∂B(z0 , r) is compact, we have |g(z)| ≥ δ > 0 on ∂B(z0 , r). But then
if |w − w0 | < δ it follows |w − w0 | < |g(z)| on ∂B(z0 , r), hence by Rouche’s
theorem, since g(z) has a zero in B(z0 , r) it follows h(z) = g(z) + (w0 −
w) = f (z) − w does also, that is, f (z) takes the value w in B(z0 , r). Thus
B(w0 , δ) ⊆ f (B(z0 , r)) and hence f (U ) is open as required.
Remark 3.5.8. Note that the proof actually establishes a bit more than the
statement of the theorem: if w0 = f (z0 ) then the multiplicity d of the zero
of the function f (z) − w0 at z0 is called the degree of f at z0 . The proof
shows that locally the function f is d-to-1, counting multiplicities, that is,
there are r, ϵ ∈ R>0 such that for every w ∈ B(w0 , ϵ) the equation f (z) = w
has d solutions counted with multiplicity in the disk B(z0 , r). In particular,
if f ′ (z0 ) ̸= 0 then the degree is 1 and the function is locally 1-to-1.
Theorem 3.5.9 (Inverse function theorem). Suppose that f : U → C is

injective and holomorphic and that f ′ (z) ̸= 0 for all z ∈ U . If g : f (U ) → U
is the inverse of f , then g is holomorphic with g ′ (w) = 1/f ′ (g(w)).
Proof. By the open mapping theorem, the function g is continuous, indeed

if V is open in f (U ) then g −1 (V ) = f (V ) is open by that theorem. To see
that g is holomorphic, fix w0 ∈ f (U ) and let z0 = g(w0 ). Note that since
g and f are continuous, if w → w0 then f (w) → z0 . Writing z = f (w) we
have
g(w) − g(w0 ) z − z0 1
lim = lim = ′
w→w0 w − w0 z→z0 f (z) − f (z0 ) f (z0 )
as required.
Remark 3.5.10. Note that the non-trivial part of the proof of the above
theorem is the fact that g is continuous! In fact the condition that f ′ (z) ̸= 0
follows from the fact that f is bijective – this can be seen using the degree
of f : if f ′ (z0 ) = 0 and f is non-constant, we must have f (z) − f (z0 ) =
(z − z0 )k g(z) where g(z0 ) ̸= 0 and k ≥ 1. Since we can choose a holomorphic
branch of g 1/k near z0 it follows that f (z) is locally k-to-1 near z0 , which
contradicts the injectivity of f . For details see the Appendices. Notice that
this is in contrast with the case of a single real variable, as the example
f (x) = x3 shows. Once again, complex analysis is “nicer” than real analy-
sis!
3.6 Application of the Residue theorem

Let us recall that if γ is a simple positively oriented contour and f is holo-
morphic on and inside of γ except at a finite set S of isolated singularities
none of which lies on γ ∗ itself. Then
Z X
f (z)dz = 2πi Resz0 .
γ z0 ∈S
3.6. APPLICATION OF THE RESIDUE THEOREM 77
In this section, we will discuss how to use this formula to compute various
integrals.
3.6.1 On the computation of residues

A lot will depend on our ability to compute residues. For example, in the
case of poles, we can use the following result.
Lemma 3.6.1. Suppose that f has a pole of order m at z0 , then
1 dm−1
Resz0 (f ) = lim ((z − z0 )m f (z))
z→z0 (m − 1)! dz m−1
Proof. Since f has a pole of order m at z0 we have f (z) = n≥−m cn (z−z0 )n

P
for z sufficiently close to z0 . Thus
(z − z0 )m f (z) = c−m + c−m+1 (z − z0 ) + . . . + c−1 (z − z0 )m−1 + . . .
and the result follows from the formula for the derivatives of a power series.
Remark 3.6.2. The last lemma is perhaps most useful in the case where the
pole is simple since in that case no derivatives need to be computed. In fact,
there is a special case which is worth emphasizing: Suppose that f = g/h
is a ratio of two holomorphic functions defined on a domain U ⊆ C, where
h is non-constant. Then f is meromorphic with poles at the zeros7 of h. In
particular, if h has a simple zero at z0 and g is non-vanishing there, then f
correspondingly has a simple pole at z0 . Since the zero of h is simple at z0 ,
we must have h′ (z0 ) ̸= 0, and hence by the previous result
g(z)(z − z0 ) z − z0
Resz0 (f ) = lim = lim g(z) lim = g(z0 )/h′ (z0 )
z→z0 h(z) z→z0 z→z0 h(z) − h(z0 )
where the last equality holds by standard Algebra of Limits results.
Example 3.6.3. Let f (z) = exp(z)/ sin(z). Since sin has simple zeros at
points zk = πk, the function f has simple poles at zk . By the formula above
(z − zk ) exp(z) exp(zk ) exp(zk )

Reszk f = lim = = .
z→zk sin(z) sin′ (zk ) cos(zk )
7
Strictly speaking, the poles of f form a subset of the zeros of h, since if g also vanishes
at a point z0 , then f may have a removable singularity at z0 .
If the pole is of the higher order such computations become much more
involved.
Example 3.6.4. Consider f (z) = sin−2 (z) and an isolated singularity at

z0 = 0. Then the Lemma 3.6.1 gives us

d z sin(z) − z cos(z)
Res0 f = lim = lim = 0.
z→0 dz sin(z) z→0 sin2 (z)
Note that even for such a simple function and for a pole of a small
order, the computation of the limit is non-trivial, since we have to argue
that sin(z) − z cos(z) = o(z 2 ). The easiest way of doing this is to consider
the Taylor series of sin and cos.
In many cases it is easier to compute residues or even the entire principal
part by analysing power series without referring to Lemma 3.6.1. A typical
computation is below:
Example 3.6.5. Consider the same function f = 1/ sin2 and z0 = 0. Then

we can write
1 1 1 1
f (z) = = = .
(z − z 3 /6 + . . . )2 z 2 (1 − z 2 /6 + . . . ))2 2 2
z (1 − z /3 + . . . )
Let us denote g(z) = z 2 /3 + . . . . This is an analytic function with g(0) = 0.

This means that if |z| is sufficiently small then |g| < 1/2 so
1 z2
= 1 + g(z) + g 2 (z) + · · · = 1 + + ....
1 − g(z) 3
Combining all of this we get
1 1
f (z) = + + ...
z2 3
Hence the principal part is 1/z 2 and Res0 f = 0.
Example 3.6.6. Consider f (z) = 1/(z 2 sinh(z)3 ). Now sinh(z) = (ez −

e−z )/2 vanishes on πiZ, and these zeros are all simple since dz
d
(sinh(z)) =
n
cosh(z) has cosh(nπi) = (−1) ̸= 0. Thus f (z) has a pole or order 5 at
zero, and poles of order 3 at πin for each n ∈ Z\{0}. Let us calculate the
principal part of f at z = 0 using the same technique. We will write O(z k )
for any holomorphic functions which vanish to order k at 0.
z3 z5 z2 z4
z 2 sinh(z)3 = z 2 (z + + + O(z 7 ))3 = z 5 (1 + + + O(z 6 ))3
3! 5! 3! 5!
3z 2 3z 4 3z 4
= z 5 (1 + + + + O(z 6 ))
3! (3!)2 5!
z 2 13z 4
= z 5 (1 + + + O(z 6 ))
2 120
As before, we introduce
z 2 13z 4 13z 2

5 2 1 3
g(z) = + + O(z ) = z + + O(z )
2 120 2 120
Using the geometric series we get
1
=1 − g(z) + g 2 (z) + . . .
1 + g(z)
2
13z 2 13z 2

2 1 3 4 1
=1 − z + + O(z ) + z + + O(z ) + O(z 6 )
3
2 120 2 120
z2 z 2 17z 4

1 13
=1 − + − z 4 + O(z 5 ) = 1 − + + O(z 5 ).
2 4 120 2 120
Combining all expansions we get
1 1 1 1 1 17
= = 5− 3+ + O(1).
z 2 sinh(z)3 5
z 1 + g(z) z 2z 120z
This gives the entire principal part and in particular Res0 (f ) = 17/120.
There are other variants of the above method which we could have used:
For example, by the binomial theorem for an arbitrary exponent we know
−3 −3 n
that if |z| < 1 then (1 + z) = n≥0 n z = 1 − 3z + 6z 2 + . . .. Arguing
P
as above, it follows that for small enough z we have
z2 z4
sinh(z)−3 = z −3 .(1 + + + O(z 6 ))−3
3! 5!
z2 z4 z 2 z 4 2

−3 6
=z 1 + (−3) + +6 + + O(z )
3! 5! 3! 5!
z2

−3 −3 6 4 6
=z 1− + + z + O(z )
2 5! (3!)2
z 2 17z 4

−3 6
=z 1− + + O(z )
2 120
yielding the same result for the principal part of 1/z 2 sinh(z)3 .
3.6.2 Residue Calculus

As mentioned before, the Residue theorem gives us a very powerful technique
for computing many kinds of integrals. In this section, we give a number of
examples of its application.
Example 3.6.7. Consider the integral

Z 2π
dt
.
0 1 + 3 cos2 (t)
If we let γ be the path t 7→ eit and let z = eit then cos(t) = Re (z) =
1 1
2 (z + z̄) = 2 (z + 1/z). Thus we have
1 1 1 4z 2
= = = ,
1 + 3 cos2 (t) 1 + 3/4(z + 1/z)2 1 + 43 z 2 + 32 + 34 z −2 3 + 10z 2 + 3z 4
Finally, since dz = izdt it follows

2π
−4iz
Z Z
dt
= dz.
0 1 + 3 cos2 (t) γ 3 + 10z 2 + 3z 4
Thus we have turned our real integral into a contour integral, and to evaluate
the contour integral we just need to calculate the residues of the meromor-
−4iz
phic function g(z) = 3+10z 2 +3z 4 at the poles it has inside the unit circle.
Now the poles of g(z) are the zeros of the polynomial p(z) = 3 + 10z 2 + 3z 4 ,
2
√ are at z ∈ {−3, −1/3}. Thus the poles inside the unit circle are at
which
±i/ 3. In particular, since p has degree 4 and has four roots, they must all
be simple zeros, and so g has simple poles at these points. The residue at
a simple pole z0 can be calculated as the limit limz→z0 (z − z0 )g(z), thus we
see (compare with Remark 3.6.2) that
√
−4iz(z − ±i/ 3) √ 1
Resz=±i/√3 (g(z)) = lim√ 2 + 3z 4
= (±4/ 3) √
z→±i/ 3 3 + 10z ′
p (±i/ 3)
√ 1
= (±4/ 3) √ √ = 1/4i.
20(±i/ 3) + 12(±i/ 3)3
It now follows from the Residue theorem that

Z 2π
dt
2
= 2πi Resz=i/√3 ((g(z)) + Resz=−i/√3 (g(z)) = π.
0 1 + 3 cos (t)
Remark 3.6.8. We can similarly write sin(t) = (z − 1/z)/(2i) and express all
other trigonometric functions in terms of rational functions of z. Thus many
trigonometric integrals can be turned into integrals of rational functions.
All such integrals can be computed quite easily as long as one can find all
singularities.
Often we are interested in integrating along a path which is not closed
or even finite, for example, we might wish to understand the integral of
a function on the positive real axis. The residue theorem can still be a
powerful tool in calculating these integrals, provided we complete the path
to a closed one in such a way that we can control the extra contribution to
the integral along the part of the path we add.
Example 3.6.9. If we have a function f which we wish to integrate over

the whole real line (so we have to treat it as an improper Riemann integral)
then we may consider the contours ΓR given as the concatenation of the
paths γ1 : [−R, R] → C and γ2 : [0, 1] → C where
γ1 (t) = −R + t; γ2 (t) = Reiπt .
(so that ΓR = γ2 ⋆ γR1 traces out the boundary of a half-disk). In many cases
one can show that γ2 f (z)dz tends to 0 as R → ∞, and by calculating the
residues inside the contours ΓR deduce the integral of f on (−∞, ∞). To
see this strategy in action, consider the integral
Z ∞
dx
.
0 1 + x2 + x4
It is easy to check that this integral exists as an improper Riemann integral,

and since the integrand is even, it is equal to
Z R
1 dx
lim dx.
2 R→∞ −R 1 + x2 + x4
If f (z) = 1/(1 + z 2 + z 4 ), then ΓR f (z)dz is equal to 2πi times the sum of

R
the residues inside the path ΓR . The function f (z) = 1/(1 + z 2 + z 4 ) has
poles at z 2 = ±e2πi/3 and hence at {eπi/3 , e2πi/3 , e4πi/3 , e5πi/3 }. They are all
simple poles and of these only {ω, ω 2 } are in the upper-half plane, where
ω = eiπ/3 . Thus by the residue theorem, for all R > 1 we have
Z

f (z)dz = 2πi Resω (f (z)) + Resω2 (f (z)) ,
ΓR
and we may calculate the residues using the limit formula as above (and the
fact that it evaluates to the reciprocal of the derivative of 1+z 2 +z 4 ): Indeed
1 1
since ω 3 = −1 we have Resω (f (z)) = 2ω+4ω 3 = 2ω−4 , while Resω 2 (f (z)) =
1 1
2ω 2 +4ω 6
= 4+2ω 2 . Thus we obtain:
Z
1 1
f (z)dz = 2πi +
ΓR 2ω − 4 2ω 2 + 4

1 1
= πi +
ω − 2 ω2 + 2
√
ω2 + ω

3π π
= πi 2
=− =√ ,
2(ω − ω ) − 5 −3 3
√
(where we used the fact that ω 2 + ω = i 3 and ω − ω 2 = 1). Now clearly
Z Z R Z
dt
f (z)dz = + f (z)dz,
ΓR −R 1 + t2 + t4 γ2
and by the estimation lemma we have

Z
πR
f (z)dz ≤ sup |f (z)|ℓ(γ2 ) ≤ → 0,
γ2 z∈γ2∗ R4 − R2 − 1
as R → ∞, it follows that
Z Z ∞
π dt
√ = lim f (z)dz = .
3 R→∞ ΓR −∞ 1 + t2 + t4
3.6.3 Jordan’s Lemma and applications

The following lemma is a basic fact on convexity. Note that if x, y are vectors
in any vector space then the set {tx + (1 − t)y : t ∈ [0, 1]} describes the line
segment between x and y.
Lemma 3.6.10. Let g : R → R be a twice differentiable function. Then if

[a, b] is an interval on which g ′′ (x) < 0, the function g is concave on [a, b],
that is, for x < y ∈ [a, b] we have
g(tx + (1 − t)y) ≥ tg(x) + (1 − t)g(y), t ∈ [0, 1].
Thus informally speaking, chords between points on the graph of g lie below
the graph itself.
Proof. Given x, y ∈ [a, b] and t ∈ [0, 1] let ξ = tx + (1 − t)y, a point in

the interval between x and y. Now the slope of the chord between (x, g(x))
and (ξ, g(ξ)) is, by the Mean Value Theorem, equal to g ′ (s1 ) where s1 lies
between x and ξ, while the slope of the chord between (ξ, g(ξ)) and (y, g(y))
is equal to g ′ (s2 ) for s2 between ξ and y. If g(ξ) < tg(x) + (1 − t)g(y) it
follows that g ′ (s1 ) < 0 and g ′ (s2 ) > 0. Thus by the mean value theorem for
g ′ (x) applied to the points s1 and s2 it follows there is an s ∈ (s1 , s2 ) with
g ′′ (s) = (g ′ (s2 ) − g ′ (s1 ))/(s2 − s1 ) > 0, contradicting the assumption that
g ′′ (x) is negative on (a, b).
The following lemma is an easy application of this convexity result.
∗
Lemma 3.6.11 (Jordan’s Lemma). Let f be a continuous function on γR
where γR (t) = Reit where t ∈ [0, π]. Then
Z
π
f (z)eiαz dz ≤ MR , MR = max |f (Reit )|.
γR α t∈[0,π]
In particular, suppose that f is holomorphic on H \ S where H = {z ∈

C : Im (z) > 0} is the upper half-plane and S is a finite set of isolated
singularities. Suppose that f (z) → 0 as z → ∞ in H. Then
Z
f (z)eiαz dz → 0
γR
as R → ∞ for all α > 0.

Proof. Applying Lemma 3.6.10 to the function g(t) = sin(t) with x = 0
and y = π/2 we see that sin(t) ≥ π2 t for t ∈ [0, π/2]. Similarly we have
sin(π − t) ≥ 2(π − t)/π for t ∈ [π/2, π]. Thus we have
(
iαz −αR sin(t) e−2αRt/π t ∈ [0, π/2],
|e | ≤ e ≤ −2αR(π−t)/π
e t ∈ [π/2, π].
But then it follows that
Z Z π/2
π π
iαz
f (z)e dz ≤ 2RMR e−2αRt/π dt = MR (1 − e−αR ) < MR .
γR 0 α α
This proves the first claim.
Next, fix some ϵ > 0. Since f → 0 as z → ∞ and S is finite, there is R0
such that MR < ϵ for all R > R0 . This immediately implies that
Z
π
f (z)eiαz dz ≤ ϵ , R > R0 .
γR α
Since ϵ is arbitrary, this proves the second claim.
Remark 3.6.12. If ηR is an arc of a semicircle in the upper half plane, say

ηR (t) = Reit for 0 ≤ t ≤ 2π/3, then the same proof shows that
Z
f (z)eiαz dz → 0 as R → ∞.
ηR
This is sometimes useful when integrating around the boundary of a sector
of the disk (that is a set of the form {reiθ : 0 ≤ r ≤ R, θ ∈ [θ1 , θ2 ]}).
It is also useful to note that if α < 0 then the integral of f (z)eiαz around a
semicircle in the lower half plane tends to zero as the radius of the semicircle
tends to infinity provided |f (z)| → 0 as |z| → ∞ in the lower half plane.
This follows immediately from the above applied to f (−z).
R∞
Example 3.6.13. Consider the integral −∞ sin(x) x dx. This is an improper
RR
integral of an even function, thus it exists if and only if the limit of −R sin(x)
x dx
exists as R → ∞. To compute this consider the integral along the closed
curve ηR given by the concatenation ηR = νR ⋆ γR , where νR : [−R, R] → R
given by νR (t) = t and γR (t) = Reit (where t ∈ [0, π]). Now if we let
iz
f (z) = e z−1 , then f has a removable singularity at z = 0 (as is easily seen
by considering Rthe power series expansion of eiz ) and so is an entire function.
Thus we have ηR f (z)dz = 0 for all R > 0. Thus we have
Z R
eiz
Z Z Z
dz
0= f (z)dz = f (t)dt + dz − .
ηR −R γR z γR z
Now Jordan’s lemma ensures that the second term
R πon the right tends to zero
iReit
as R → ∞, while the third term integrates to 0 Reit dt = iπ. It follows
RR
that −R f (t)dt tends to iπ as R → ∞. and hence taking imaginary parts
R∞
we conclude the improper integral −∞ sin(x)
x dx is equal to π.
iz
Remark 3.6.14. The function f (z) = e z−1 might not have been the first
meromorphic function one could have thought of when presented with the
previous improper integral. A more natural candidate might have been
iz
g(z) = ez . There is an obvious problem with this choice, however, which is
that it has a pole on the contour we wish to integrate around. In the case
where the pole is simple (as it is for eiz /z) there is a standard procedure
for modifying the contour: one indents it by a small circular arc around the
− + ±
pole. Explicitly, we replace the νR with νR ⋆ γϵ ⋆ νR where νR (t) = t and
− +
t ∈ [−R, −ϵ] for νR , and t ∈ [ϵ, R] for νR (and as above γϵ (t) = ϵei(π−t) for
t ∈ [0, π]). Since sin(x)
x is bounded at x = 0 the sum
Z −ϵ Z R Z R
sin(x) sin(x) sin(x)
dx + dx → dx,
−R x ϵ x −R x
as ϵ → 0, while the integral along γϵ can be computed explicitly: by the

iz
Taylor expansion of eiz we see that Resz=0 ez = R1, so that eiz − 1/z is
bounded near 0. It follows that as ϵ → 0 we have γϵ (eiz /z − 1/z)dz → 0.
On the other hand
Z π
−ϵiei(π−t)
Z
dz
=ϵ dt = −iπ,
γϵ z 0 ϵei(π−t)
so that we see
eiz
Z
dz → −iπ, ϵ → 0.
γϵ z
− +
Combining all of this we conclude that if Γϵ = νR ⋆ γ ϵ ⋆ νR ⋆ γR then
−ϵ R
eix eiz eix eiz
Z Z Z Z Z
0= f (z)dz = dx + dz + dx + dz.
Γϵ −R x γϵ z ϵ x γR z
R
eiz eiz
Z Z Z
sin(x)
= 2i + + dz
ϵ x γϵ z γR z
R
eiz
Z Z
sin(x)
→ 2i dx − iπ + dz.
0 x γR z
as ϵ → 0. Then letting R → ∞, it follows from Jordans Lemma that the

third term tends to zero so we see that
Z ∞ Z ∞
sin(x) sin(x)
dx = 2 dx = π
−∞ x 0 x
as required.
We record a general version of the calculation we made for the contribu-
tion of the indentation to a contour in the following Lemma.
Lemma 3.6.15. Let f : U → C be a meromorphic function with a simple
pole at a ∈ U and let γϵ : [α, β] → C be the path γϵ (t) = a + ϵeit , then
Z
lim f (z)dz = Resa (f )(β − α)i.
ϵ→0 γϵ
Proof. Since f has a simple pole at a, we may write

c
f (z) = + g(z)
z−a
where g(z) is holomorphic near z and c = Resa (f ) (indeed c/(z − a) is
just the principal part of f at a). But now as g is holomorphic at a, it is
continuous at a, and so bounded. Let M, r > 0 be such that |g(z)| < M for
all z ∈ B(a, r). Then if 0 < ϵ < r we have
Z
g(z)dz ≤ ℓ(γϵ )M = (β − α)ϵM,
γϵ
which clearly tends to zero as ϵ → 0. On the other hand, we have

Z Z β Z β
c c it
dz = it
iϵe dt = (ic)dt = ic(β − α).
γϵ z − a α ϵe α
R R R
Since γϵ f (z)dz = γϵ c/(z − a)dz + γϵ g(z)dz the result follows.
3.6.4 Summation of infinite series

Residue calculus can also be a useful tool in calculating infinite sums, as we
now show. For this, we use the function f (z) = cot(πz). Note that since
sin(πz) vanishes precisely at the integers, f (z) is meromorphic with poles at
each integer n ∈ Z. Moreover, since f is periodic with period 1, in order to
understand the poles of f it suffices to calculate the principal part of f at
z = 0. We can use the method of the previous section to do this:
3 5
We have sin(z) = z − z3! + z5! + O(z 7 ), so that sin(z) vanishes with
multiplicity 1 at z = 0 and we may write sin(z) = z(1 − zh(z)) where
h(z) = z/3! − z 3 /5! + O(z 5 ) is holomorphic at z = 0. Then
1 1 1 X 1
= (1 − zh(z))−1 = 1 + z n h(z)n = + h(z) + O(z 2 ).
sin(z) z z z
n≥1
Multiplying by cos(z) we see that the principal part of cot(z) is the same
as that of z1 cos(z) which, using the Taylor expansion of cos(z), is clearly z1
again. By periodicity, it follows that cot(πz) has a simple pole with residue
1/π at each integer n ∈ Z.
We can also use this strategy to find further terms of the Laurent series of
cot(z): Since our h(z) actually vanishes at z = 0, the terms h(z)n z n vanish
to order 2n. It follows that we obtain all the terms of the Laurent series of
cot(z) at 0Pup to order 3, say, just by considering the first two terms of the
series 1 + n≥1 z n h(z)n , that is, 1 + zh(z). Since cos(z) = 1 − z 2 /2! + z 4 /4!,
it follows that cot(z) has a Laurent series
z2 1 z z3
+ O(z 4 )). + ( − + O(z 5 ))

cot(z) = (1 −
2! z 3! 5!
1 z
= − + O(z 3 )
z 3
The fact that f (z) has simple poles at each integer will allow us to sum
infinite series with the help of the following:
Lemma 3.6.16. Let f (z) = cot(πz) and let ΓN denotes the square path
with vertices (N + 1/2)(±1 ± i). There is a constant C independent of N
such that |f (z)| ≤ C for all z ∈ Γ∗N .
Proof. We need to consider the horizontal and vertical sides of the square
separately. Note that cot(πz) = (eiπz + e−iπz )/(eiπz − e−iπz ). Thus on the
horizontal sides of ΓN where z = x ± (N + 1/2)i and −(N + 1/2) ≤ x ≤
(N + 1/2) we have
eiπ(x±(N +1/2)i) + e−iπ(x±(N +1/2)i)

| cot(πz)| =
eiπ(x±(N +1/2)i − e−iπ(x±(N +1/2)i)
eπ(N +1/2) + e−π(N +1/2)
≤
eπ(N +1/2) − e−π(N +1/2)
= coth(π(N + 1/2)).
Now since coth(x) is a decreasing function for x ≥ 0 it follows that on the
horizontal sides of ΓN we have | cot(πz)| ≤ coth(3π/2).
On the vertical sides we have z = ±(N + 1/2) + iy, where −N − 1/2 ≤
y ≤ N + 1/2. Observing that cot(z + N π) = cot(z) for any integer N and
that cot(z + π/2) = − tan(z), we find that if z = ±(N + 1/2) + iy for any
y ∈ R then
| cot(πz)| = | − tan(iy)| = | − tanh(y)| ≤ 1.
Thus we may set C = max{1, coth(3π/2)}.
We now show how this can be used to sum an infinite series:

Example 3.6.17. Let g(z) = cot(πz)/z 2 . By our discussion of the poles of
cot(πz) above it follows that g(z) has simple poles with residues πn1 2 at each
non-zero integer n and residue −π/3 at z = 0.
Consider now the integral of g(z) around the paths ΓN : By Lemma
3.6.16 we know |g(z)| ≤ C/|z|2 for z ∈ Γ∗N , and for all N ≥ 1. Thus by the
estimation lemma we see that
Z
g(z)dz ≤ C(4N + 2)/(N + 1/2)2 → 0,
ΓN
as N → ∞. But by the residue theorem we know that

Z X 1
g(z)dz = −π/3 + .
ΓN πn2
n̸=0,
−N ≤n≤N
It therefore follows that

∞
X 1
= π 2 /6
n2
n=1
Remark 3.6.18. Notice that the contours ΓN and the function cot(πz) clearly
allow us to sum other infinite series in a similarPway – for example if we
wished to calculate the sum of the infinite series n≥1 n21+1 then we would
consider the integrals of g(z) = cot(πz)/(1 + z 2 ) over the contours ΓN .
Remark 3.6.19. (Non-examinable – for interest only! ): Note that taking
g(z) = (1/z 2k ) cot(πz) P
for any positive integer k, the above strategy gives a
method for computing ∞ n=1 1/n
2k (check that you see why we need to take
even powers of n). The analysis for the case k = 1 goes through in general,
we just need to compute more and more of the Laurent series of cot(πz) the
larger we take k to be.
One can show that ζ(s) = ∞ s
P
n=1 1/n converges to a holomorphic func-
tion of s for any s ∈ C with Re (s) > 1 (as usual, we define ns = exp(s log(n))
where log is the ordinary real logarithm). As s → 1 it can be checked that
ζ(s) → ∞, however it can be shown that ζ(s) extends to a meromorphic
function on all of C\{1}. The identity theorem shows that this extension is
unique if it exists8 . (This uniqueness is known as the principle of “analytic
continuation”.) The location of the zeros of the ζ-function is the famous Rie-
mann hypothesis: apart from the “trivial zeros” at negative even integers,
they are conjectured to all lie on the line Re (z) = 1/2. Its values at special
points however are also of interest: Euler was the first to calculate ζ(2k)
for positive integers k, but the values ζ(2k + 1) (for k a positive integer)
remain mysterious – it was only shown in 1978 by Roger Apéry that ζ(3) is
irrational for example. Our analysis above is sufficient to determine ζ(2k)
once one succeeds in computing explicitly the Laurent series for cot(πz) or
equivalently the Taylor series of z cot(πz) = iz + 2iz/(e2iz − 1).
3.6.5 Keyhole contours

There are many ingenious paths which can be used to calculate integrals via
residue theory. One common contour is known (for obvious reasons) as a
keyhole contour. It is constructed from two circular paths of radius ϵ and R,
where we let R become arbitrarily large, and ϵ arbitrarily small, and we join
the two circles by line segments with a narrow neck in between. Explicitly,
if 0 < ϵ < R is given, pick a δ > 0 small, and set η+ (t) = t + iδ, η− (t) =
8
It is this uniqueness and the fact that onePcan readily compute that ζ(−1) = −1/12
that results in the rather outrageous formula ∞ n=1 n = −1/12.
Figure 3.3: A keyhole contour.
(R−t)−iδ, where in each case t runs over the closed intervals with endpoints
such that the endpoints of η± lie on the circles of radius ϵ and R about the
origin. Let γR be the positively oriented path on the circle of radius R
joining the endpoints of η+ and η− on that circle (thus traversing the “long”
arc of the circle between the two points) and similarly let γϵ the path on the
circle of radius ϵ which is negatively oriented and joins the endpoints of γ±
on the circle of radius ϵ. Then we set ΓR,ϵ = η+ ⋆ γR ⋆ η− ⋆ γϵ (see Figure
3.3). The keyhole contour can sometimes be useful to evaluate real integrals
where the integrand is multi-valued as a function on the complex plane since
it avoids a straight branch cut. We can see it in the next example:
R ∞ x1/2 1/2 /(1 +
Example 3.6.20. Consider the integral 0 1+x 2 dx. Let f (z) = z
z 2 ), where we use the branch of the square root function which is continuous
on C\R>0 , that is, if z = reit with t ∈ [0, 2π) then z 1/2 = r1/2 eit/2 .
We use the keyhole contour ΓR,ϵ . On the circle of radius R, we have
|f (z)| ≤ R1/2 /(R2 − 1), so by the estimation lemma, this contribution to the
integral of f over ΓR,ϵ tends to zero as R → ∞. Similarly, |f (z)| is bounded
by ϵ1/2 /(1 − ϵ2 ) on the circle of radius ϵ, thus again by the estimation lemma
this contribution to the integral of f over ΓR,ϵ tends to zero as ϵ → 0. Finally,
the discontinuity of our branch of z 1/2 on R>0 ensures that the contributions
of the two line segments of the contour do not cancel but rather both tend
R∞ x1/2
to 0 1+x2
dx
as δ and ϵ tend to zero.
R ∞ x1/2 R
To compute 0 1+x 2 dx we evaluate the integral ΓR,ϵ f (z)dz using the
residue theorem: The function f (z) clearly has simple poles at z = ±i, and
their residues are 21 e−πi/4 and 21 e5πi/4 respectively. It follows that
√
Z
1 −πi/4 1 5πi/4
f (z)dz = 2πi e + e = π 2.
ΓR,ϵ 2 2
R∞ x1/2
√
Taking the limit as R → ∞ and ϵ → 0 we see that 2 0 1+x2
dx = π 2, so
that Z ∞ 1/2
x dx π
2
=√ .
0 1+x 2
Appendices
91
93
All appendices are non-examinable. They are here for the sake of com-
pleteness and do not form part of the syllabus.
94
Appendix A
On the homotopy and

homology versions of
Cauchy’s theorem
In this appendix we give proofs of the homotopy and homology versions of

Cauchy’s theorem which are stated in the body of the notes. These proofs
are non-examinable but are included for the sake of completeness.
We will need the following theorem that shows that certain functions
defined by integrals are holomorphic.
Theorem A.0.1. Let U be an open subset of C and suppose that F : U ×[a, b]

is a function satisfying
1. The function z 7→ F (z, s) is holomorphic in z for each s ∈ [a, b].
2. F is continuous on U × [a, b]
Then the function f : U → C defined by

Z b
f (z) = F (z, s)ds
a
is holomorphic.
Proof. Changing variables we may assume that [a, b] = [0, 1] (explicitly, one
replaces s by (s − a)/(b − a)). By Theorem 3.2.12 it is enough to show that
we may find a sequence of holomorphic functions fn (z) which converge of
f (z) uniformly on compact subsets of U . To find such a sequence, recall
from Prelims Analysis that the Riemann integral of a continuous function is
95
96APPENDIX A. ON THE HOMOTOPY AND HOMOLOGY VERSIONS OF CAUCHY’S THEO
equal to the limit of its Riemann sums as the mesh of the partition used for
the sum tends to zero. Using the partition xi = i/n for 0 ≤ i ≤ n evaluating
at the right-most end-point of each interval, we see that
n
1X
fn (z) = F (z, i/n),
n
i=1
R1
is a Riemann sum for the integral 0 F (z, s)ds, hence as n → ∞ we have
fn (z) → f (z) for each z ∈ U , i.e. the sequence (fn ) converges pointwise to
f on all of U . To complete the proof of the theorem it thus suffices to check
that fn → f as n → ∞ uniformly on compact subsets of U . But if K ⊆ U is
compact, then since F is clearly continuous on the compact set K × [0, 1], it
is uniformly continuous there, hence, given any ϵ > 0, there is a δ > 0 such
that |F (z, s) − F (z, t)| < ϵ for all z ∈ B̄(a, ρ) and s, t ∈ [0, 1] with |s − t| < δ.
But then if n > δ −1 we have for all z ∈ K
Z 1 n
1X
|f (z) − fn (z)| = F (z, s)dz − F (z, i/n)
0 n
i=1
X n Z i/n

= F (z, s) − F (z, i/n) ds
i=1 (i−1)/n
n Z i/n
X
≤ |F (z, s) − F (z, i/n)|ds
i=1 (i−1)/n
Xn
< ϵ/n = ϵ.
i=1
Thus fn (z) tends to f (z) uniformly on K as required.
Theorem A.0.2. Let U be a domain in C and a, b ∈ U . Suppose that γ

and η are paths from a to b which are homotopic in U and f : U → C is a
holomorphic function. Then
Z Z
f (z)dz = f (z)dz.
γ η
Proof. The key to the proof of this theorem is to show that the integrals of
f along two paths from a to b which “stay close to each other” are equal.
We show this by covering both paths by finitely many open disks and using
the existence of a primitive for f in each of the disks.
More precisely, suppose that h : [0, 1] × [0, 1] is a homotopy between γ
and η. Let us write K = h([0, 1] × [0, 1]) be the image of the map h, a
97
Figure A.1: Dissecting the homotopy
compact subset of U . Since K is sequentially compact there is an ϵ > 0 such

that B(z, ϵ) ⊆ U for all z ∈ K (Lemma 8.2.3 of the metric spaces part of
the course).
Next, we use the fact that, since [0, 1] × [0, 1] is compact, h is uniformly
continuous. Thus we may find a δ > 0 such that |h(t1 , s1 ) − h(t2 , w2 )| < ϵ
whenever ∥(t1 , s1 ) − (t2 , s2 )∥ < δ. Now pick N ∈ N such that 1/N < δ and
dissect the square [0, 1] × [0, 1] into N 2 small squares of side length 1/N .
For convenience, we will write ti = i/N for i ∈ {0, 1, . . . , N }
For each k ∈ {1, 2, . . . , N − 1}, let νk be the piecewise linear path which
connects the point h(tj , k/N ) to h(tj+1 , k/N ) for each j ∈ {0, 1, . . . , N ).
Explicitly, for t ∈ [tj , tj+1 ], we set
νk (t) = h(tj , k/N )(1 − N t − j) + h(tj+1 , k/N )(N t − j)
We claim that
Z Z Z Z Z
f (z)dz = f (z)dz = f (z)dz = . . . = f (z)dz = f (z)dz
γ ν1 ν2 νN −1 η
R
which will prove the theorem. In fact, we will only show that γ f (z)dz =
R
ν1 f (z)dz, since the other cases are almost identical.
We may assume the numbering of our squares Si is such that S1 , . . . , SN
list the bottom row of our N 2 squares from left to right. Let mi be the
98APPENDIX A. ON THE HOMOTOPY AND HOMOLOGY VERSIONS OF CAUCHY’S THEO
centre of the square Si and let pi = h(mi ). Then h(Si ) ⊆ B(pi , ϵ) so that
γ([ti , ti+1 ]) ⊆ B(pi , ϵ) and ν1 ([ti , ti+1 ]) ⊆ B(pi , ϵ) (since B(pi , ϵ) is convex
and by assumption contains ν1 (ti ) and ν1 (ti+1 )). Since B(pi , ϵ) is convex, f
has primitive Fi on each B(pi , ϵ). Moreover, as primitives of f on a domain
are unique up to a constant, it follows that Fi and Fi+1 differ by a constant
on B(pi , ϵ) ∩ B(pi+1 , ϵ), where they are both defined. In particular, since
γ(ti ), ν1 (ti ) ∈ B(pi , ϵ) ∩ B(pi+1 , ϵ), (1 ≤ i ≤ N − 1), we have
Fi (γ(ti )) − Fi+1 (γ(ti )) = Fi (ν1 (ti )) − Fi+1 (ν1 (ti )). (A.0.1)
Now by the Fundamental Theorem we have
Z
f (z)dz = Fi (γ(ti+1 )) − Fi (γ1 (ti )),
γ|[ti ,ti+1 ]
Z
f (z)dz = Fi (ν1 (ti+1 )) − Fi (ν1 (ti ))
ν1|[ti ,ti+1 ]
Combining we find that:

Z N
X −1 Z
f (z)dz = f (z)dz
γ i=0 γ|[ti ,ti+1 ]
N
X −1

= Fi+1 (γ(ti+1 )) − Fi+1 (γ(ti ))
i=0
N
X −1

= FN (γ(tN )) − F1 (γ(0)) + Fi (γ(ti )) − Fi+1 (γ(ti ))
i=1
N
X −1

= FN (b) − F0 (a) + (Fi (ν1 (ti+1 )) − Fi+1 (ν1 (ti+1 )
i=0
N
X −1

= (Fi+1 (ν1 (ti+1 )) − Fi+1 (ν1 (ti ))
i=0
N
X −1 Z Z
= f (z)dz = f (z)dz
i=0 ν1 |[ti ,ti+1 ] ν1
where in the fourth equality we used Equation (A.0.1).
Remark A.0.3. The use of the piecewise linear paths νk might seem unnat-
ural – it might seem simpler to use the paths given by the homotopy, that
is the paths γk (t) = h(t, k/N ). The reason we did not do this is because
99
we only assume that h is continuous, so we do not know that the path γk is

piecewise C 1 which we need in order to be able to integrate along it.
The proof of the homology form of Cauchy’s theorem uses Liouville’s

theorem, which we proved using Cauchy’s theorem for a disc.
Theorem A.0.4. Let f : U → C be a holomorphic function and let γ : [0, 1] →

U be a closed path whose inside lies entirely in U , that is I(γ, z) = 0 for all
/ U . Then we have, for all z ∈ U \γ ∗ ,
z∈
Z Z
f (ζ)
f (ζ)dζ = 0; dζ = 2πiI(γ, z)f (z), ∀z ∈ U \γ ∗ .
γ γ ζ −z
Moreover, if U is simply connected and γ : [a, b] → U is any closed path,

then I(γ, z) = 0 for any z ∈
/ U , so the above identities hold for all closed
paths in such U .
Proof. We first prove the general form of the integral formula. Note that
using the integral formula for the winding number and rearranging, we wish
to show that
f (ζ) − f (z)
Z
F (z) = dζ = 0
γ ζ −z
for all z ∈ U \γ ∗ . Now if g(ζ, z) = (f (ζ) − f (z))/(ζ − z), then since f

is complex differentiable, g extends to a continuous function on U × U if
we set g(z, z) = f ′ (z). Thus the function F is in fact defined for all z ∈ U .
Moreover, if we fix ζ then, by standard properties of differentiable functions,
g(ζ, z) is clearly complex differentiable as a function of z everywhere except
at z = ζ. But since it extends to a continuous function at ζ, it is bounded
near ζ, hence by Riemann’s removable singularity theorem, z 7→ g(ζ, z) is in
fact holomorphic on all of U . It follows by Theorem A.0.1 that
Z 1
F (z) = g(γ(t), z)γ ′ (t)dt
0
is a holomorphic function of z.
Now let ins(γ) = {z ∈ C : I(γ, z) ̸= 0} be the inside of γ, so by assump-
tion we have ins(γ) ⊂ U , and let V = C\(γ ∗ ∪ ins(γ)) be the complement
of γ ∗ and its inside. If z ∈ U ∩ V , that is, z ∈ U but not inside γ or on γ ∗ ,
100APPENDIX A. ON THE HOMOTOPY AND HOMOLOGY VERSIONS OF CAUCHY’S THE
then
Z Z
f (ζ)dζ dζ
F (z) = − f (z)
γ ζ −z γ ζ −z
Z
f (ζ)dζ
= − f (z)I(γ, z)
γ ζ −z
Z
f (ζ)dζ
= = G(z)
γ ζ −z
since I(γ, z) = 0. Now G(z) is an integral which only involves the values of
f on γ ∗ hence it is defined for all z ∈
/ γ ∗ , and by Theorem A.0.1, G(z) is
holomorphic. In particular, G defines a holomorphic function on V , which
agrees with F on all of U ∩V , and thus gives an extension of F to a holomor-
phic function on all of C. (Note that by the above, F and G will in general
not agree on the inside of γ.) Indeed if we set H(z) = F (z) for all z ∈ U and
H(z) = G(z) for all z ∈ V then H is a well-defined holomorphic function on
all of C. We claim that |H| → 0 as |z| → ∞, so that by Liouville’s theorem,
H(z) = 0, and so F (z) = 0 as required. But since ins(γ) is bounded, there
is an R > 0 such that V ⊇ C\B(0, R), and so H(z) = G(z) for |z| > R. But
then setting M = supζ∈γ ∗ |f (ζ)| we see
Z
f (ζ)dζ ℓ(γ).M
|H(z)| = ≤ .
γ ζ −z |z| − R
which clearly tends to zero as |z| → ∞, hence |H(z)| → 0 as |z| → ∞ as

required.
For the second formula, simply apply the integral formula to g(z) =
(z − w)f (z) for any w ∈ / γ ∗ . Finally, to see that if U is simply connected the
inside of γ always lies in U , note
R that if w ∈
/ U then 1/(z −w) is holomorphic
dz
on all of U , and so I(γ, w) = γ z−w = 0 by the homotopy form of Cauchy’s
theorem.
Remark A.0.5. It is often easier to check that a domain is simply connected

than it is to compute the interior of a path. Note that the above proof uses
Liouville’s theorem, whose proof depends on Cauchy’s Integral Formula for
a circular path, which was a consequence of Cauchy’s theorem for a triangle,
but apart from the final part of the proof on simply connected regions, we
did not use the more sophisticated homotopy form of Cauchy’s theorem. We
have thus established the winding number and homotopy forms of Cauchy’s
theorem essentially independently of each other.
Appendix B
Remark on the Inverse

Function Theorem
In this appendix we supply1 the details for the claim made in the remark
after the proof of the holomorphic version of the inverse function theorem.
There is an enhancement of the Inverse Function Theorem in the holo-
morphic setting, which shows that the condition f ′ (z) ̸= 0 is automatic (in
contrast to the case of real differentiable functions, where it is essential as
one sees by considering the example of the function f (x) = x3 on the real
line). Indeed suppose that f : U → C is a holomorphic function on an open
subset U ⊂ C, and that we have z0 ∈ U such that f ′ (z0 ) = 0.
Claim: In this case, f is at least 2 to 1 near z0 , and hence is not injective.
Proof of Claim: If we let w0 = f (z0 ) and g(z) = f (z) − w0 , it follows g has a
zero at z0 , and thus it is either identically zero on the connected component
of U containing z0 (in which case it is very far from being injective!) or
we may write g(z) = (z − z0 )k h(z) where h(z) is holomorphic on U and
h(z0 ) ̸= 0. Our assumption that f ′ (z0 ) = 0 implies that k, the multiplicity
of the zero of g at z0 is at least 2.
Now since h(z0 ) ̸= 0, we have ϵ = |h(z0 )| > 0 and hence by the continuity
of h at z0 we may find a δ > 0 such that h(B(z0 , δ)) ⊆ B(h(z0 ), ϵ). But
then by taking a cut along the ray {−t.h(z0 ) : t ∈ R>0 } we can define
a holomorphic branch of z 7→ z 1/k on the whole of B(h(z0 ), ϵ). Now let
ϕ : B(z0 , δ) → C be the holomorphic function given by ϕ(z) = (z−z0 ).h(z)1/k
(where by our choice of δ this is well-defined) so that ϕ′ (z0 ) = h(z0 )1/k ̸= 0.
Then clearly f (z) = w0 + ϕ(z)k on B(z0 , δ). Since ϕ(z) is holomorphic,the
1
For interest, not examination!
101
102APPENDIX B. REMARK ON THE INVERSE FUNCTION THEOREM
open mapping theorem ensures that ϕ(B(z0 , δ)) is an open set, which since it
contains 0 = ϕ(z0 ), contains B(0, r) for some r > 0. But then since z 7→ z k
is k-to-1 as a map from B(0, r)\{0} → B(0, rk )\{0} it follows that f takes
every value in B(w0 , rk )\{w0 } at least k times.

A2 Complex Analysis 2024

Uploaded by

Copyright:

Available Formats

A2 Complex Analysis 2024

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A2 Complex Analysis 2024

Uploaded by

Copyright:

Available Formats

A2.

September 30, 2024

2 Paths and Integration 27

3 Cauchy’s Formula and its applications 55

3.4 Isolated singularities . . . . . . . . . . . . . . . . . . . . . . . 64

A On the homotopy and homology versions of Cauchy’s theo-

B Remark on the Inverse Function Theorem 101

The aim of this part of the course is to study functions f : C → C, asking

Cauchy’s Integral formulae. Taylor expansion. Liouville’s Theorem.

0.2 Further reading

0.3 Basic notations

which obviously can be written as

where | · | is the modulus of a complex number.

Lemma 0.3.1. Let z, w ∈ C. Then

1. |z|2 = z z̄, where z̄ is the complex conjugate of z;

2. If z = reiθ , where r ∈ [0, ∞) and θ ∈ R, then |z| = r;

Proof. (1) If z = a + ib then z z̄ = (a + ib)(a − ib) = a2 + b2 .

(3) One can calculate directly, writing z = a + ib and w = c + id. Alter-

Definition 0.3.2. We say that U ⊂ C is open if for every z ∈ U there is

In complex analysis, it is often convenient to work with connected open

Definition 0.3.3. A connected open subset D ⊆ C of the complex plane

Now we come to a crucial part in the course – the discussion of what it

lim f (x + iy), lim lim f (x + iy) and lim lim f (x + iy)

lim lim f (x + iy) = lim lim f (x + iy) = 0

1.1 Complex differentiability

1. (Sums, products) If f, g are differentiable at a then f + g and f g

2. (Quotients) If f, g are differentiable at a and g(a) ̸= 0 then f /g is

3. (Chain rule) If U and V are open subsets of C and f : V → U and

(g ◦ f )′ (a) = g ′ (f (a))f ′ (a).

Lemma 1.1.3. Let a ∈ C, let U be a neighbourhood of a and let f : U → C.

f (z) = f (a) + f ′ (a)(z − a) + ε(z)(z − a),

It is an easy exercise to check that this definition is indeed equivalent to

Definition 1.1.4 (Holomorphic function). Let U ⊆ C be an open set (for

Sometimes one says that f is holomorphic at a point a; this means that

1.2 Cauchy-Riemann equations

and f ′ (z0 ) = ∂x u(z0 ) + i∂x v(z0 ).

Combining two different expressions for f ′ (z0 ) we get

There is an alternative way of proving Theorem 1.2.2 which also explains

Alternative proof of Theorem 1.2.2. First, we show that the function f is

This immediately gives the Cauchy-Riemann equations and that f ′ (z0 ) =

f (z) = f (z0 ) + L(z − z0 ) + R(z)

f (z) = f (z0 ) + (∂x u(x0 , y0 ) + i∂x v(x0 , y0 ))(z − z0 ) + R(z).

The assumption that function is continuously differentiable is important.

Example 1.2.8. Let f (z) = xy/(x2 + y 2 ) for z ̸= 0 and define f (0) = 0.

Remark 1.2.9. Complex conjugation is a very natural and useful transfor-

Lemma 1.2.10. Suppose f : C → C is holomorphic and that f ′ is identically

1.2.1 Wirtinger derivatives

Definition 1.2.11. Let f : U → C be a function with components (u, v),

Wirtinger (partial) derivatives by

Proof. Straightforward calculation.

Remark 1.2.13 (Non-examinable). Wirtinger derivatives are very natural.

1.3 Harmonic functions

Definition 1.3.1. Suppose that u : R2 → R is a function on some open

Definition 1.3.2. Suppose that u : R2 → R is a function on some open

Theorem 1.3.3. Let U ⊆ C be open, and suppose that f : U → C is

Proof. From the Cauchy-Riemann equations,

∂xx u = ∂xy v (= ∂x ∂y v), ∂yy u = −∂yx v.

and the result follows.

Let us make some further comments on this result: