You are on page 1of 208

Rhodes University

Department of Mathematics
M 3.2
Complex Analysis
Julien Larena
2012
2
Forewords
These lecture notes are intended for a one semester third-year course in
mathematics at Rhodes University.
Mostly, they follow the tracks and spirit of the excellent introductory book
by H.A. Priestley, Introduction to complex analysis, OUP, and do not present
any original result, or any original path to known results. I have tried to
keep the references to real analysis as limited as possible, so that this course
could be studied without much prior knowledge of real analysis. Never-
theless, to cut on some time-consuming proofs, I have omitted to prove
some results as Cauchys convergence criterion for sequences, or Bolzano-
Weierstrass theorem, as these theorems will be proven during the course on
real analysis, and their proofs in complex analysis are very similar, or can
be deduced from the properties in the real case. If a reader is interested in
these proofs, she can refer to W. Rudins book, Principles of Mathematical
Analysis, McGraw-Hill, that treats of properties in general metric spaces.
Also, I did not prove Jordans curve theorem, as it would have been painful,
long, and mostly unnecessary, since we will develop contour integration for
the limited class of non self-intersecting contours, for which the notions of
interior and exterior are quite obvious. I, nevertheless, mentioned the idea
of the general proof, in case some students may be willing to think about
the problem. The class of non intersecting contours made of pieces of arc
circles and line segments is largely sucient for applications in an introduc-
tory course on complex analysis.
i
ii
I shall recommend to read another excellent book: Visual Complex Analysis,
by T. Needham, OUP. It is a wonderful, very graphic, exposition of complex
analysis, with an emphasis on physical and geometrical interpretations of
the notions presented in these notes. I encourage the students who wish to
develop their intuition on complex analysis to read this book.
Students willing to nd more exercises and problems than those we will be
doing in class and tutorials can refer to Complex variables by M. R. Spiegel,
in the collection Schaums outlines, Mc Graw Hill.
Finally, let me emphasize that analysis is a new subject for third year stu-
dents, and a dedicated study will be necessary in order to succeed in un-
derstanding this course. This is mainly due to the introduction of rigorous
proofs. Therefore, students must pay attention to proofs and to methods
used during these proofs, as they are the keys to mastering the techniques
of Complex Analysis.
Cover illustration: Augustin-Louis Cauchy around 1840. Lithography by
Zephirin Belliard after a painting by Jean Roller.
Notations
We use N, Z, R and C to denote the sets of natural numbers, integers, real
numbers and complex numbers, respectively. When we want to exclude 0
from one of these sets, we will simply star the set. For example R

= R\{0}.
In the same way, when we want to indicate that we keep only the positive
(or zero) (resp. negative or zero) real numbers, we will write R
+
(resp. R

).
A lot of notations in complex analysis are directly transferable from their
counterparts in real analysis. When it is the case, we supposed that the
reader could do the translation herself.
Also, we use the standard abbreviations such as i for if and only if, e.g.
for for example and i.e. for that is. The end of a proof is denoted by
the usual symbol: .
iii
iv
Contents
1 The complex plane 1
1.1 The complex plane . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Complex numbers . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Complex Algebra . . . . . . . . . . . . . . . . . . . . . 8
1.1.3 Exercices . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2 Geometry in the complex plane . . . . . . . . . . . . . . . . . 13
1.2.1 Lines, circles, and other subsets . . . . . . . . . . . . . 13
1.2.2 Extended complex plane and Riemann sphere . . . . . 19
1.2.3 Mobius transformations . . . . . . . . . . . . . . . . . 23
1.2.4 Exercices . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.3 A bit of topology in the complex plane . . . . . . . . . . . . . 26
1.3.1 Open and closed sets of the complex plane . . . . . . . 27
1.3.2 Convexity and connectedness . . . . . . . . . . . . . . 32
1.3.3 Limits and continuity . . . . . . . . . . . . . . . . . . 36
1.3.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.4 Curves, paths and contours . . . . . . . . . . . . . . . . . . . 40
1.4.1 Denitions . . . . . . . . . . . . . . . . . . . . . . . . 41
1.4.2 Contours . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.4.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 43
2 Complex functions 45
2.1 Complex series and power series . . . . . . . . . . . . . . . . . 46
v
vi CONTENTS
2.1.1 Complex series . . . . . . . . . . . . . . . . . . . . . . 47
2.1.2 Power series . . . . . . . . . . . . . . . . . . . . . . . . 51
2.1.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.2 Some complex functions . . . . . . . . . . . . . . . . . . . . . 54
2.2.1 The exponential function . . . . . . . . . . . . . . . . 55
2.2.2 Complex trigonometric and hyperbolic functions . . . 58
2.2.3 Roots of unity . . . . . . . . . . . . . . . . . . . . . . 60
2.2.4 The logarithmic function . . . . . . . . . . . . . . . . 61
2.2.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.3 Multifunctions . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.3.1 Example 1: the logarithmic function . . . . . . . . . . 63
2.3.2 Branch points and multibranches . . . . . . . . . . . . 65
2.3.3 Example 2: Fractional powers . . . . . . . . . . . . . . 67
2.3.4 Example 3: An example with two branch points . . . 68
3 Dierentiation 71
3.1 Holomorphic functions . . . . . . . . . . . . . . . . . . . . . . 72
3.1.1 Dierentiation and the Cauchy-Riemann equations . . 72
3.1.2 Holomorphic functions . . . . . . . . . . . . . . . . . . 76
3.1.3 Some useful results . . . . . . . . . . . . . . . . . . . . 78
3.1.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.2 Some holomorphic functions . . . . . . . . . . . . . . . . . . . 81
3.2.1 A result on the dierentiation of power series . . . . . 82
3.2.2 The exponential function . . . . . . . . . . . . . . . . 84
3.2.3 Complex trigonometric and hyperbolic functions . . . 85
3.2.4 The logarithmic function . . . . . . . . . . . . . . . . 85
3.2.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.3 Conformal mapping . . . . . . . . . . . . . . . . . . . . . . . 87
3.3.1 Conformal mapping . . . . . . . . . . . . . . . . . . . 87
3.3.2 Some examples . . . . . . . . . . . . . . . . . . . . . . 89
CONTENTS vii
4 Integration 93
4.1 Integration in the complex plane . . . . . . . . . . . . . . . . 94
4.1.1 Integration along paths . . . . . . . . . . . . . . . . . 94
4.1.2 The fundamental theorem of calculus . . . . . . . . . . 99
4.1.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.2 Cauchys theorem . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.2.1 Historical Cauchys theorem . . . . . . . . . . . . . . . 102
4.2.2 Cauchy-Goursat theorem . . . . . . . . . . . . . . . . 104
4.2.3 Deformation . . . . . . . . . . . . . . . . . . . . . . . 115
4.2.4 The complex logarithm... again . . . . . . . . . . . . . 118
4.2.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.3 Cauchys formul . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.3.1 Cauchys integral formula . . . . . . . . . . . . . . . . 122
4.3.2 Cauchys formul for derivatives . . . . . . . . . . . . 124
4.3.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.4 Power series representation . . . . . . . . . . . . . . . . . . . 129
4.4.1 Integration of series . . . . . . . . . . . . . . . . . . . 129
4.4.2 Taylors theorem . . . . . . . . . . . . . . . . . . . . . 130
4.4.3 Multiplication of power series . . . . . . . . . . . . . . 134
4.4.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.5 Zeros and singularities . . . . . . . . . . . . . . . . . . . . . . 136
4.5.1 Characterizing zeros . . . . . . . . . . . . . . . . . . . 137
4.5.2 Identity and Uniqueness theorems . . . . . . . . . . . 139
4.5.3 Counting zeros . . . . . . . . . . . . . . . . . . . . . . 144
4.5.4 Laurents theorem . . . . . . . . . . . . . . . . . . . . 149
4.5.5 Singularities . . . . . . . . . . . . . . . . . . . . . . . . 154
4.5.6 Meromorphic functions . . . . . . . . . . . . . . . . . 159
4.5.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.6 Cauchys residue theorem . . . . . . . . . . . . . . . . . . . . 161
4.6.1 Residues and Cauchys residue theorem . . . . . . . . 162
viii CONTENTS
4.6.2 Calculation of residues . . . . . . . . . . . . . . . . . . 164
4.6.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 168
5 Applications 171
5.1 Some applications of contour integration . . . . . . . . . . . . 172
5.1.1 Evaluation of real integrals by contour integration . . 172
5.1.2 Some remarks on indented contours . . . . . . . . . . 174
5.1.3 Integral of rational functions . . . . . . . . . . . . . . 177
5.1.4 Integral of other functions with a nite number of poles180
5.1.5 Integrals of functions with an innite number of poles 184
5.1.6 Integrals involving multifunctions . . . . . . . . . . . . 186
5.1.7 Summation of series . . . . . . . . . . . . . . . . . . . 190
5.2 The Fourier transform . . . . . . . . . . . . . . . . . . . . . . 191
5.2.1 Introducing the Fourier transform . . . . . . . . . . . 191
5.2.2 Some applications . . . . . . . . . . . . . . . . . . . . 194
Chapter 1
The complex plane:
Geometry, Topology and
Analysis
1
2 CHAPTER 1. THE COMPLEX PLANE
1.1 The complex plane
1.1.1 Complex numbers
A bit of history
Complex numbers rst appear in the mathematical history during the six-
teenth century, in the work of Girolamo Cardano, Ars Magna (1545), and
shortly after, in LAlgebra, by Rafael Bombelli (1572). In these initial works,
the authors were interested in the solution of cubic algebraic equations of
the form:
x
3
= 3px + 2q .
This problem is equivalent to nding the intersection points of the cubic
curve y = x
3
and the line y = 3px + 2q. It is often said that complex
numbers appeared as necessary entities in relation to nding roots of the
quadratic equations x
2
= mx+c. Indeed, this equation admits a pair of real
solutions as long as m
2
+4c > 0, but no real solution in the case m
2
+4c < 0.
But what is the problem with that? A simple graph will show that there is
no intersection in the plane in that second case. This is clearly not true for
the cubic, as there is always an intersection (cf gure 1.1).
Figure 1.1: Representation of a quadratic and a cubic equations.
1.1. THE COMPLEX PLANE 3
Cardano had shown that this equation could be solved, and the inter-
section was given by:
x =
3
_
q +
_
q
2
p
3
+
3
_
q
_
q
2
p
3
.
Bombelli, inspecting this formula discovered a troubling properties of the
solution: take p = 5 and q = 2, Then, q
2
p
3
< 0, and the solution seems
pathological. But, if one introduces a number i such that i
2
= 1, it takes
the form:
x =
3

2 + 11i +
3

2 11i .
Moreover, the cubic x
3
= 15x + 4 as a solution in the plane, for x = 4.
Bombellis brilliant idea was to link the expression (1.1.1) with the known
real solution x = 4. To do so, he postulated that one could write
3

2 + 11i =
2 + yi and
3

2 11i = 2 yi. In order for this rule to work, he needed to


suppose that the addition of the introduced complex numbers should obey
the intuitive rule: (a + ib) + (c + id) = (a + b) + i(c + d). Then, he tried
to nd whether there was a number y satisfying the properties above. To
do that, he calculated (2 + iy)
3
using the standard rules of algebra of real
numbers:
(a +ib)(c +id) = ac +i
2
bd +i(ad +bc) .
Using i
2
= 1, this yields:
(a +ib)(c +id) = (ac bd) +i(ad +bc) .
That allowed him to show that: (2 i)
3
= 2 11i, leading to x = 4 when
evaluating (1.1.1)! Despite this intriguing properties, complex numbers re-
mained mostly ignored or looked at with contempt until the end of the
eighteenth century
1
. This lack of interest can certainly be linked with a
Platonic prejudice against objects that couldnt be given a solid geometrical
interpretation and hence failed to exists as entities. That is precisely what
1
This is reected in the term imaginary, associated with the multiplier of i in their
expression. Unfortunately, this term has survived until today.
4 CHAPTER 1. THE COMPLEX PLANE
Figure 1.2: The Argand plane with some complex numbers.
an idea pursued independently by Wessel, Argand and Gauss provided at
the end of the eighteenth century. They linked complex numbers to points
(or vectors) in the plane. Taking Bombellis denition of a complex number
z as a complex of two ordinary numbers, a and b, z = a + ib, they re-
marked that it could be identied with a point in the plane with Cartesian
coordinates (a, b). This is illustrated on gure 1.2. Once this identication
has been made, the plane is called the complex plane (or Argand plane, or
Gauss plane), and is usually denoted by C. The power of this identication
lies in the interpretation of the addition of two complex numbers:
The sum of two complex numbers z and w is given by the usual
parallelogram rule of vector addition.
1.1. THE COMPLEX PLANE 5
Figure 1.3: A geometrical interpretation of addition and multiplication of
complex numbers.
The rule for the multiplication is less natural but reads:
The length of the product zw (seen as a vector) is given by the product of
the lengths of z and w (seen as vectors).
The angle of zw with the x axis is given by the sum of the angles of z and
w with the x axis.
Summary
It is time to summarize our denition of complex numbers.
Denition 1. We will denote by C the set of complex numbers (identifying
it without care with the complex plane).
Any complex number z is composed of two real numbers a and b such
that z = a +ib, where i is the complex number satisfying i
2
= 1. To
put it in a formal way: z C, !(a, b) R
2
, z = a +ib.
a is called the real part of z and is noted Re(z). b is called the imagi-
nary part of z and is noted Im(z).
The modulus of z is the positive real number noted |z| representing
the length of the vector associated with z in the complex plane. Note
that if z is real, i.e. if its imaginary part is zero, the associated vector
6 CHAPTER 1. THE COMPLEX PLANE
lies along the real axis; in that case, the modulus of z is simply its
absolute value.
An argument of z is one of the real numbers, noted arg(z), representing
the angle of the vector associated with z in the complex plane with
the x axis. Note that the argument is not unique, because if is an
argument of z, + 2k with k Z is also an argument of z. In the
following, every time we will use arg(z) the statements have to be
understood in the equivalence class modulo 2.
A pure imaginary number is a complex number with a real part equal
to zero.
The x and y axis are sometimes referred to as the real and imaginary
axis, respectively.
Let us note that the Cartesian labelling introduced earlier is very conve-
nient to deal with the addition of complex numbers: Re(z + w) = Re(z) +
Re(w) and Im(z +w) = Im(z) +Im(w). On the contrary, the multiplication
appears uneasy to memorize in the Cartesian notation, and the geometric
interpretation of complex numbers greatly facilitates our task. Instead of
using Cartesian coordinates, lets introduce polar coordinates (r, ). One
automatically has: (x, y) = (r cos , r sin ); cf gure 1.4.
Then, for any two complex numbers z and w, such that |z| = r =
_
x
2
+y
2
and arg(z) = and |w| = r

, arg(w) =

, according to the rule of multipli-


cation given above, it is obvious that: |zw| = rr

and arg(zw) = +

. It is
then convenient, when dealing with multiplications of complex numbers to
use a new notation, called Eulers formula:
Theorem 1. Any complex number z can be written: z = re
i
= r cos() +
ir sin(), where r is the modulus of z and its argument.
1.1. THE COMPLEX PLANE 7
Figure 1.4: Polar representation of complex numbers.
The use of the exponential notation is not due to a coincidence, but it
will be made clearer later, once we have studied the complex exponential
function. Nevertheless, we can try to give here a reason for the notation.
Argument for Eulers formula. Consider the real exponential function: x
R, f(x) = e
x
. The series g(x) =

+
n=0
x
n
n!
converges, and its limit is f(x)
(prove it):
x R, e
x
=
+

n=0
x
n
n!
.
Now, let us boldly replace the real x in this expression by the pure
8 CHAPTER 1. THE COMPLEX PLANE
imaginary number i
2
:
e
i
=
+

n=0
(i)
n
n!
.
It is clear that any even power of i will bring a real term, whereas any odd
power will bring a pure imaginary term. Grouping them like that leads to:
e
i
=
+

n=0
(1)
n

2n
(2n)!
+i
+

p=0
(1)
n

2n+1
(2n + 1)!
,
in which we recognise the series expansion of the real cosinus and sinus:
cos(x) =

+
n=0
(1)
n x
2n
(2n)!
, and sin(x) =

+
p=0
(1)
n x
2n+1
(2n+1)!
. Eulers formula
follows: e
i
= cos() +i sin().
Note that the argument is not uniquely determined: since the cos and
sin functions are 2-periodic, z = re
i
and w = re
i+2k
with k Z are
equal.
Exercise 1. Prove (by induction) that:
R, n N, (cos +i sin )
n
= cos(n) +i sin(n) .
This result is known as de Moivres formula.
1.1.2 Complex Algebra
Algebraic structure: the eld C.
Let us summarize once again the rules of addition and multiplication of two
complex numbers:
Denition 2. For any two complex numbers z = a + ib = re
i
and w =
c +id = Re
i
:
z +w = (a +c) +i(b +d) ,
2
The fact that we can do that is not obvious, and will become clear only after we
have dened the complex exponential function. This is why this development is only an
argument, and by no way a proof!
1.1. THE COMPLEX PLANE 9
zw = rRe
i(+)
.
Note that these operations have very simple properties:
The addition and the multiplication are commutative, i.e.: (z, w)
C
2
, z +w = w +z and zw = wz.
The addition and the multiplication are associative: (z, w, u) C
3
, z+
(w +u) = (z +w) +u and z(wu) = (zw)u.
The multiplication is distributive with respect to the addition: (z, w, u)
C
3
, z(w +u) = zw +zu.
Moreover, the addition as a neutral element, 0, such that z C, z +0 =
z, and the multiplication also has a neutral element, 1, such that z
C, 1.z = z.
Finally, for any z C, there exists a unique element of C, noted z
such that z + (z) = 0, and provided that z = 0, there exists a unique
element of C, noted z
1
or 1/z, such that zz
1
= 1. It is trivial to see
that, for z = x + iy, (z) = x + i(y) = x iy. On the other hand
for z = re
i
, lets note z
1
= Re
i
. Then: zz
1
= 1 rRe
i(+)
=
rRcos(+)+irRsin(+) = 1. Hence: sin(+) = 0 and rRcos() = 1.
This leads to: = + 2k, k Z and R = 1/r. Since the argument is
dened up to 2k, this inverse is unique and one nally has:
|z
1
| =
1
|z|
, (1.1)
arg(z
1
) = arg(z) + 2k, k Z . (1.2)
Theorem 2. The set C equipped with the addition and multiplication dened
above is a eld. The neutral element for the addition is 0; the neutral element
for the multiplication is 1.
Remark 1. Note that i
1
= 1/i = i.
10 CHAPTER 1. THE COMPLEX PLANE
Some more algebra
In the eld C, there exists another operation that is of much interest: it is
the complex conjugation. It corresponds, in the complex plane, to a reexion
in the real axis:
Denition 3. For any z C, the complex conjugate of z is the unique
complex number z such that, given z = x +iy = re
i
, z = x iy = re
i
.
For example, note that

i = i, and: z C, Re(z) = 0 z =
z and Im(z) = 0 z = z.
Proposition 1. The complex conjugation have the following properties, for
any z and w in C:


z = z.
Re(z) =
z+ z
2
.
Im(z) =
i( zz)
2
.
z +w = z + w.
zw = z w.
| z| = |z|.
|z|
2
= z z.
Exercise 2. Prove these properties.
Finally, we can conclude this section by presenting some inequalities of
importance.
Proposition 2. For all z and w in C:
|Re(z)| |z| and |Im(z)| |z|;
The triangle inequality: |z +w| |z| +|w|;
1.1. THE COMPLEX PLANE 11
|z +w| ||z| |w||.
Proof. The rst point is trivial since |z|
2
= Re(z)
2
+ Im(z)
2
.
The second point is an important result:
|z +w|
2
= (z +w)(z +w)
= (z +w)( z + w)
= |z|
2
+|w|
2
+w z +z w
= |z|
2
+|w
2
| + 2Re(z w) cf above.
Hence, using the rst point: |z +w|
2
|z|
2
+|w|
2
+2|z w|. Again, using
the simple properties above: |z w| = |z|| w| = |z||w|, hence the right-hand
side of the inequality is simply the square (|z| +|w|)
2
: |z +w|
2
(|z| +|w|)
2
.
Since |z +w| 0 and |z| +|w| 0, the triangle inequality follows.
Finally, we obtain the last inequality by that for two real numbers x and y,
the inequality |x| y holds i y 0 and y x y. Hence, the third
inequality is satised i |z +w| |z| |w| and |z +w| |w| |z|. But, the
triangle inequality gives:
|z| = |z +w w| |z +w| +| w| = |z +w| +|w|
|w| = |w +z z| |z +w| +| z| = |z +w| +|z|.
The last inequality follows trivially.
It is important to realize that all these inequalities are between real
numbers, constructed from complex numbers. No meaning has been given
to inequalities between complex numbers, simply because it is impossible to
construct an order relation compatible with the structure of the eld C.
What about functions?
Until now, we have dened complex numbers and studied algebraic prop-
erties in the eld C. It is time to introduce complex functions, since this
12 CHAPTER 1. THE COMPLEX PLANE
course is about complex analysis, i.e. the study of complex-valued func-
tions. In real analysis, one denes a function f as a mapping between a
subset S R and R which assigns to each z S a unique f(z) R. The
requirement of uniqueness is crucial in every development of real analysis.
In complex analysis, we will be less restrictive and dene a complex-valued
function f as a mapping between a subset S C and C. Geometrically,
that means that a complex function transform a region of the complex plane
into another region of the complex plane. We dropped the requirement of
uniqueness because we will see that a lot of important functions in complex
analysis are not one-valued and are called multifunctions. We have already
encountered such a multifunction: the argument of a complex number. It is
a complex function if one sees R as a subset of C, and it sends any complex
number z into a subset of S R such that S = R, z = |z|e
i
. Of course,
since complex functions are complex-valued, one can decompose their image
into real and imaginary parts: for f a complex function, there exists u and
v real functions such that f = u + iv, meaning that for all z in the domain
of f, f(z) = u(z) +iv(z), with u(z) = Re(f(z)) and v(z) = Im(f(z)).
1.1.3 Exercices
1. Express the following complex numbers in the Cartesian form x +iy:
(i) (2 + 3i)(1 2i);
(ii) (1 + 3i)(2 + 2i);
(iii)
1+i
12i
;
(iv) 1 2i +
i
2+3i
.
2. Find the polar forms of the following complex numbers, and place
them in the complex plane:
(i)
1
2
+

3
2
i;
(ii)

2
2
(1 +i);
1.2. GEOMETRY IN THE COMPLEX PLANE 13
(iii)

3 +i;
(iv)
1
2
.
3. Find the Cartesian forms of the following complex numbers, and place
them in the complex plane:
(i) 2e
i/4
;
(ii) e
i/6
;
(iii)
1
2
e
i/3
;
(iv) 3e
i/2
.
1.2 Geometry in the complex plane
This section will explore further the link between complex functions and
geometry in the plane. It will present some geometric locii of the complex
plane, and their description in terms of complex numbers. Also, we will
study in details the properties of the Riemann sphere, that will allow us
to treat lines and circle (as well as half-lines and circular arcs) in a unied
way. Finally, we will study a large class of transformations named Mobius
transformations. These transformations nd a lot of applications in complex
analysis, but they also have a particular importance due to their link with
non-Euclidean geometries.
1.2.1 Lines, circles, and other subsets
You have already encountered the introduction of algebraic concepts in ge-
ometry in the past, when you have constructed the Cartesian plane, and
you have seen how ecient it could be. In the Cartesian plane
3
, labelled by
two real coordinates x and y, a line is represented by the algebraic relation
3
The Cartesian plane consists of the introduction of two axis, of coordinates in the
geometric plane, such that any point of the plane is labelled by an ordered set of two
coordinates: M = (x, y), where x and y are real numbers.
14 CHAPTER 1. THE COMPLEX PLANE
ax + by + c = 0, where a, b and c are real numbers; a circle of radius R,
centred on the point O = (a, b), is characterised by (xa)
2
+(y b)
2
= R
2
.
This algebraic structure given to the geometric plane through the in-
troduction of the Cartesian plane can be extended to complex numbers by
making use of the complex plane: lines and circles can then be characterized
by equations involving complex numbers, rather than real numbers.
Equations for line segments and lines
Let us start with a line segment between two points A = (a, b) and B = (c, d)
in the Cartesian plane. Any point M = (x, y) of the segment [A, B] is
characterized by:
x = a +t(c a), t [0, 1]
y = b +l(d b), l [0, 1]
(a c)y = (b d)x + (ad bc) .
The last relation is just the equation for the line containing the line segment.
Putting the rst two relations in the last one, one nds that, necessarily:
l = t. Then, introducing the complex number z = x +iy representing M in
the complex plane:
z = (1 t)(a +ib) +t(c +id) ;
noting that, in the complex plane, A is represented by z
A
= a + ib and B
by z
B
= c +id, on immediately has the equation for a line segment [z
A
, z
B
]
between z
A
and z
B
in the complex plane:
[z
A
, z
B
] = {z = (1 t)z
A
+tz
B
, t [0, 1]} .
This equation can be straightforwardly extended to the whole line through
A and B: (z
A
, z
B
) = {z = (1 t)z
A
+tz
B
, t R}.
Note that a line that is perpendicular to the line joining two points
A = z
A
and B = z
B
and that cuts [A, B] in its centre can be described by:
|z z
A
| = |z z
B
| .
1.2. GEOMETRY IN THE COMPLEX PLANE 15
Exercise 3. Explain why.
Equations for circles and circular arcs
The circle is a very simple geometric object: the circle centred on the point
O and of radius r > 0 is the locus of points at a distance r from M. If
the point O has coordinates (a, b) in the Cartesian plane, such a locus is
therefore characterized by the equation:
(x a)
2
+ (y b)
2
= r
2
.
Exercise 4. Prove it.
This simple form becomes even simpler in the complex plane. Lets note
z = x + iy the complex number associated with a point M on the circle,
and lets all c = a +ib the complex number associated with the centre O of
the circle. Then, z c = (x a) + i(y b), and one immediately sees that
|z c|
2
= (x a)
2
+ (y b)
2
. Hence, the equation for the circle centred on
O and of radius r is:
|z c| = r .
Note that the same argument shows that the disc centred on O and of radius
r is characterized by |z c| r (for the closed disc; see below).
To go further
There exists another useful characterization of circles in the complex plane.
For (a, b) C
2
and R

, = 0, the points associated with z such that:

z a
z b

=
form a circle, known as a circle of Apollonius. We will see that this repre-
sentation is very useful when studying conformal mappings.
16 CHAPTER 1. THE COMPLEX PLANE
Exercise: Show that this locus is actually a circle; conversely, show that
every circle can be describes like that.
We can now turn to the description of circular arcs joining two points A
and B associated respectively to the complex numbers a and b. Let P be an
arbitrary point on this circular arc. Then, a simple geometrical argument
shows that the angle

APB = is constant along the arc. If P is represented


by the complex number z, lets denote arg(z a) = and arg(z b) = .
Then, it is clear from the gure 1.5, that = . Hence:
arg(z a) arg(z b) = [2] ,
or, equivalently, the equation for the arc:
arg
_
z a
z b
_
= [2] .
Some other subsets of the complex plane
We can now describe quickly a few subsets of the complex plane that will
appear in the next chapters. Note that the terms open and closed used
here agree with the denition that we will introduce later, when we examine
the topology of the complex plane.
We have already encountered the notion of disc, i.e. the set of all the
points that are at less than a given distance from one point. Let us
make that notion more precise. The open disc centred on a C and
of radius r R

+
is, by denition:
D(a, r) = {z C, |z a| < r} .
1.2. GEOMETRY IN THE COMPLEX PLANE 17
Figure 1.5: Geometric construction of the equation for a circular arc between
A and B.
18 CHAPTER 1. THE COMPLEX PLANE
The closed disc centred on a C and of radius r R

+
is:
D(a, r) = {z C, |z a| r} .
To put it simply,

D(a, r) consists in the union of D(a, r) and its bound-
ary, the circle characterized by |z a| = r, that we will denote (a, r)
in these notes. We will also need the punctured disc, D

(a, r), centred


on a C and of radius r R

+
:
D

(a, r) = {z C, 0 < |z a| < r} .


It is the disc D(a, r) from which we have removed the centre a.
Another important class of regions is made of the annuli. For (s, r)
R
+
R

+
, they are dened by:
A(a, s, r) = {z C, s < |z a| < r} .
s corresponds to the inner radius of the annulus, and r to its outer
radius.
Note that the case s = 0 corresponds to the punctured disc presented
above.
We will also need to characterize half-planes. The open upper half-
plane is given by:

+
= {z C, Im(z) > 0} ,
and the closed upper half-plane is the union of
+
with its bound-
ary:

+
= {z C, Im(z) 0} .
The other half-planes are dened accordingly (Do it for the lower half-
plane and the two other natural ones).
1.2. GEOMETRY IN THE COMPLEX PLANE 19
Finally, lets introduce sectors, i.e. the region of the complex plane
that is made of the complex numbers with an argument comprised
between two values:
S
,
= {z C

, < arg(z) < } .


Note that half-planes are also sectors, for = .
To go further
1.2.2 Extended complex plane and Riemann sphere
The Riemann sphere
It is time to see our rst example of a mapping of the complex plane. Con-
sider the function g dened as follows:
C

z 1/z .
Hence, writing z = re
i
, g(z) =
1
r
e
i
. This means that the unit circle
|z| = 1 is mapped into itself. The punctured unit disc D

(0.r) in mapped
into the exterior of the closed unit disc: {z C, |z| > 1}, and the exterior
of the the closed unit disc is mapped into the punctured unit disc D

(0, 1).
Now, it is clear that if one considers a point arbitrarily closed to the origin,
its image will be a point with an arbitrary large modulus, and if one considers
a point such that |z| +, its image will be arbitrarily closed to the origin.
Therefore, even if the mapping is not dened at the origin, it seems that its
behaviour around the origin is very regular, and we would like to extend the
mapping to the origin, so that it can be a mapping of C into itself. In other
20 CHAPTER 1. THE COMPLEX PLANE
words, by taking a limit, we would like to say that the origin is mapped
into a point at innity (|1/z| +), and, conversely, a point at innity
would be mapped into the origin. This can be done very naturally by adding
a single point to the complex plane, in order to make it compact (we will
dene this properly later). The idea of completing the complex plane in
that way is due to Riemann and found a lot of remarkable applications in
geometry. Let us embed the complex plane C into the Euclidean space R
3
by
identifying the complex numbers z = x + iy with the points of coordinates
(x, y, 0) in this space. Let us denote by (u, v, w) the coordinates in R
3
. The
Riemann sphere is dened, in R
3
, as the set:
= {(u, v, w) R
3
, u
2
+v
2
+w
2
= 1} .
It is a sphere that intersect the complex plane on the unit circle |z| = 1.
Let N = (0, 0, 1), the north pole of the Riemann sphere. We are going to
construct the stereographic projection of the Riemann sphere onto the
complex plane. Lets consider a point m of , with coordinates (u, v, w),
such that u
2
+ v
2
+ w
2
= 1. The line (Nm) is generated by the vector

Nm = (u, v, w 1); in other words, any point M on this line is such that
there exists
M
for which

NM =

Nm. If this point M is in the complex
plane, M = (x, y, 0), and one nds:
x = u
y = v
1 = (w 1) .
It is clear that when w = 1, i.e. when m = N, this system is not satised.
In all the other cases, we see that = 1/(1 w), and:
x =
u
1 w
y =
v
1 w
with w
2
= 1 u
2
v
2
1.2. GEOMETRY IN THE COMPLEX PLANE 21
In other words, switching back to complex numbers, we have constructed
a mapping f : \N C from the Riemann sphere without its north pole
onto the complex plane that is dened by:
(u, v, w)
f

u +iv
1 w
.
One can show that this application is bijective, i.e one-to-one, and contin-
uous (Show it!). Let us see what happens when m approaches N. In that
case, w approaches 1 and the image point M in the complex plane is sent to
higher and higher value of its modulus. More precisely, an open neighbour-
hood of N on the Riemann sphere is mapped into the exterior of an open
disc on the complex plane. The smaller the neighbourhood, the farther the
boundary of the exterior is from the origin. Roughly speaking, we would
like to say that every point at innity is an image of N by an extension of
f. Note that it does not matter which direction we consider, the points at
innity are all images of N. Thus, we will dene a new point, which we
will denote , and we will add this point to C and dene the extended
complex plane

C = C {} by constructing an application :

C
such that:
(u, v, w)

_
u+iv
1w
if w = 1
if w = 1 .
If we now come back to the application g(z) = 1/z that we introduced at
the beginning of the subsection, we can extend it into a new application
g :

C

C, that is one-to one on the extended complex plane:
z
g
1/z ,
with the rule:
1

= 0 and
1
0
= .
This means that we are now allowed to divide a non-zero complex number by
zero. More specically, the following algebraic rules apply in the extended
22 CHAPTER 1. THE COMPLEX PLANE
Figure 1.6: Stereographic projection and the Riemann Sphere.
complex plane:
a = a = and a/= 0 , a C
a.= .a = and a/0 = , a C

+= .= =
Remark 2. Note that some operations are not dened, such that ,
0/0 or /.
Why working in the extended complex plane?
The two most important things that are gained by extending the complex
plane are the following:
In the extended complex plane, lines and circles can be unied into a
single class of objects, called the circlines.
1.2. GEOMETRY IN THE COMPLEX PLANE 23
It is now much easier to study in details the behaviour of functions at
innity, since it is only a point in

C.
First of all, consider a circle on that passes through N. Its image by
the stereographic projection is a line on the complex plane. So, in essence,
the point can be viewed as belonging to any line in

C. Now, take a circle
on that is parallel to the complex plane. Then, its image in the complex
plane is clearly a circle centred on 0. It can be shown that any circle that
does not pass through N on projects onto a circle on C, and that every
circle in C can be constructed in that way. Hence, we can regard lines in

C
as circles through . We will then call lines and circles on

C circlines. If
we remember the algebraic parametrizations given previously, we see that
circlines can be described by the equation:

z a
z b

= , > 0 ,
where a line corresponds to = 1.
The other interesting result of this extension lies in the possibility to
treat the innity as a normal point. we will see in the next chapters that
this can allow us to talk about the intersection of curves at innity, or their
behaviour there.
1.2.3 M obius transformations
To nish this section, we are going to introduce a large class of mappings,
called Mobius transformations. These are mappings of

C onto itself that
transform circlines into circlines. These transformations have a wide range
of applications in both algebra and geometry.
Denition 4. A Mobius transformation M is a mapping of

C onto itself
of the form
M(z) =
az +b
cz +d
,where (a, b, c, d) C
4
and ad bc = 0 .
24 CHAPTER 1. THE COMPLEX PLANE
Proposition 3. Mobius transformations are bijective.
Proof. Let us rst remember what bijective means: a function f : D F
is bijective i it is injective (one-to-one) and surjective (onto).
f is injective i ((z
1
, z
2
) D, z
1
= z
2
, f(z
1
) = f(z
2
)).
f is surjective i (w F, z D, w = f(z)).
Let us start with the injectivity of M, and let us prove it by contradiction.
Let (z
1
, z
2
)

C
2
, z
1
= z
2
. Then, let us suppose that M(z
1
) = M(z
2
).
This implies, after a bit of algebra, that (ad bc)z
1
= (ad bc)z
2
. But,
ad bc = 0, hence, z
1
= z
2
, which contradicts our hypothesis. Hence,
z
1
= z
2
M(z
1
) = M(z
2
). The Mobius transformations are thus injective.
To prove the surjectivity, we pick up w

C. Then, we have to nd the
z

C such that w = M(z). This is equivalent to w = (az + b)/(cz + d),
or z = (dw b)/(cw + a). One can then easily check that this z is in

C.
Hence, M is surjective.
Being injective and surjective, M is bijective.
Proposition 4. The inverse of a Mobius transformation, M(z) = (az +
b)/(cz +d) is the Mobius transformation:
M
1
: z
dz b
cw +a
.
Proof. This follows directly from the construction used to proved the sur-
jectivity in the previous proof.
Example 1. Here are some examples of Mobius transformations:
z ze
i
, R: anticlockwise rotation
z z +a , a C: translation
z 1/z: inversion
z Sz , S R

+
: Stretching
1.2. GEOMETRY IN THE COMPLEX PLANE 25
Exercise 5. Show that any Mobius transformations can be decomposed in
a sequence of the previous transformations as follow:
T
1
: z z +
d
c
;
I : z 1/z;
R : z z exp
_
i arg
_
bcad
c
2
__
;
S : z

bcad
c
2

z;
T
2
: z z +
a
c
.
Let us now see the eect of a Mobius transformation of circlines. Let C
be a circline of equation |z|/|z| = . Let f(z) = (az+b)/(cz+d) be a
Mobius transformation. So, if w = f(z), we know that z = (dwb)/(acw).
Hence, we can substitute for z in the equation of the circline C to nd its
image under the Mobius transformation:

w f()
w f()

c +d
c +d

if c +d = 0 and c +d = 0 or,
|w f()| =

a +b
c +d

if c +d = 0 and c +d = 0 or,
|w f()| =

a +b
c +d

if c +d = 0 and c +d = 0 .
(1.3)
Note that c + d and c + d cannot both be zero because ad bc = 0, by
denition. Thus, we see that:
Proposition 5. The image by a Mobius transformation of a circline is a
circline.
26 CHAPTER 1. THE COMPLEX PLANE
1.2.4 Exercices
1. Characterize and represent the sets of the complex plane dened by:
(i) S
1
= {z C, |z 1| < 2};
(ii) S
2
= {z C, |z +i| < 1};
(iii) S
3
= S
1
S
2
;
(iv) S
4
= S
1
S
2
;
(v) S
5
= {z C, arg(z) = /4};
(vi) S
6
= {z C, |z| <

3, arg(z) = /4}.
2. Give an algebraic denition of the following sets of the complex plane
and represent them:
(i) U
1
=
+
\D(2i, 1);
(ii) U
2
=
+
D(3i, 1);
(iii) U
3
: Set of all points with an argument between 0 and 7/6, and
a modulus less than 1;
(iv) U
4
: Set of all points at a distance of 1 from the origin, with an
argument between /3 and /6;
(v) U
5
: Set of all points at a distance more than 1 from the origin,
or at a distance less than 2 from z = i.
1.3 A bit of topology in the complex plane
In the following chapters, in order to proceed with the analysis in the com-
plex plane, and in particular to tackle notions such as convergence, continu-
ity and dierentiability, we will need some elementary notions of topology
that will be presented in this section.
1.3. A BIT OF TOPOLOGY IN THE COMPLEX PLANE 27
Figure 1.7: Denition of an open set S C.
1.3.1 Open and closed sets of the complex plane
Denition 5. A set S C is open if and only if, for any z S, there
exists > 0 (depending on z) such that D(z, ) S.
Roughly speaking, that means that one can always go around any z S
in any direction without leaving S; the distance that is permitted will
vary from one point to another, depending whether z is far or close to the
boundary (the closer z is to the boundary, the smaller the permitted ).
The following properties apply to open sets:
Proposition 6. (i) If S
1
,..., S
n
for n N are open set, then S = S
1

... S
n
is also open.
(ii) If S
j
for j J (where J is some countable index set) are open sets,
then

jJ
S
j
is open.
28 CHAPTER 1. THE COMPLEX PLANE
Proof. (i) Let z S and consider
k
> 0 such that k {1, ..., n}, D(z,
k
)
S
k
. Let = min(
1
, ...
k
). Then > 0 (this is where the fact that the
S
k
are nitely many is crucial), and clearly, k {1, ..., n}, D(z, )
D(z,
k
) S
k
, hence, D(z, ) S.
(ii) This point is trivial: every S
j
is open, and z

jJ
S
j
, j J, z
S
j
; then z

jJ
S
j
, l J, > 0, D(z, ) S
l
}. Since, trivially,
S
l

jJ
S
j
, this ends the proof.
Examples of open sets:
(i) The empty set is open (the condition for it to be open cannot fail);
(ii) C is open;
(iii) a C , D(a, r) is open;
(iv) {z C, |z a| > r} is open;
(v) {z C, s < |z a| < r} is open;
(vi) S
,
is open
Proof. The rst two are trivial.
(iii) Lets start with the disc D(a, r) for a C and r > 0. Let z D(a, r)
and R such that: 0 < < r |z a| (it always exists since, by
denition of D(a, r), it is the locus of the points z such that |za| < r).
Then w D(z, ) i |w z| < . By the triangle inequality, this
implies: |w a| = |w z +z a| |w z| +|z a| +|z a| < r.
Hence, w D(z, ) w D(a, r); in other words, D(z, ) D(a, r),
and D(a, r) is open.
(iv) {z C, |z a| > r} is open by the same kind of argument.
(v) {z C, s < |z a| < r} is open as the intersection of two open sets.
1.3. A BIT OF TOPOLOGY IN THE COMPLEX PLANE 29
(vi) To prove that the sectors S
,
are open, just take z S
,
. Let
1
and
2
be the distances of z to the bounding rays dened by {z
C, arg(z) = } and {z C, arg(z) = }. Then, let = min(
1
,
2
).
We have: > 0, and r, 0 < r < , D(z, r) S
,
by construction.
Let us now consider another class of sets of the complex plane, the closed
sets, and their properties.
Denition 6. Let S C. S is closed i C\S is open.
We see immediately that this new class of sets do not bring any further
subtle denition: to prove that a set in closed is equivalent to prove that
its complementary set in C is open, and one can therefore use the same
criterion as before.
Denition 7. Let S C.
z C is a limit point of S i r R

+
, D

(z, r) S = .
A point of S that is not a limit point of S is called an isolated point
of S.
The union of S and its limit points is called the closure of S, and is
noted

S.
You see that a limit point z of S is such that every open disc centred on
it contains at least one point of S that is not z itself (if z S). Roughly
speaking that means that points of S accumulate around z. These three
denitions are closely related by the following proposition:
Proposition 7. Consider S C.
1. The following propositions are equivalent:
30 CHAPTER 1. THE COMPLEX PLANE
Figure 1.8: w is a limit point of S, but z is not.
(i) S is closed;
(ii) z C, z is a limit point of S z S;
(iii)

S = S.
2. z

S i V S = for every open set V containing z.
3.

S is a closed set.
Proof. We will prove each point successively, in order.
1. First, let us note that D

(z, r) S = D(z, r) S, for z S; that is,


if z S, for it to be a limit point, it is necessary and sucient that
D(z, r) S = . Then:
S is closed C\S is open
z S, r > 0, D(z, r) C\S
z S, r > 0, D

(z, r) S =
No point of C\S is a limit point of S
Any limit point of S is in S
1.3. A BIT OF TOPOLOGY IN THE COMPLEX PLANE 31
This shows that (i) and (ii) are equivalent. (iii) is equivalent to (ii)
because

S is the union of S and its limit points.
2. Let us consider the second proposition. Let z C such that for every
open set V containing z, V S = . Since r > 0, D(z, r) is an open
set containing z, we have, in particular, that D(z, r) S = . So,
either z S, and and this is trivially true, or z S, and this implies
that D

(z, r) S = for every r > 0, which is the denition of a limit


point. Hence, z is in the closure of S: z

S. To show the converse, let
us proceed by contradiction: let z

S and suppose that there exists
an open set V such that z V and V S = . The fact that V is open
guarantees that there exists an r > 0 such that D(z, r) V . Hence,
there exists r > 0, D(z, r) S = , which contradicts the fact that
z

S. This shows the second proposition.
3. To prove the third proposition, it is enough to prove that the closure
of

S is

S itself (by the rst proposition):

S =

S. To prove it by
contradiction, let us suppose that this is false: let z

S such that
z

S. Then, r > 0, D(z, r) S = . On the other hand, since
z

S, there exists an w such that w D(z, r)

S (second proposition
applied to

S instead of S). Hence, w

S, and, D(z, r) being an open
set containing w, the second proposition gives that D(z, r) S = .
That is the desired contradiction.
Here are some examples of closed sets:
All the sectors, when dened with weak inequalities , rather than
strict ones <. They also are the closures of the sectors dened with
strict or mixed inequalities.


D(a, r) for r > 0 is closed because it is the complement of D(a, r){z
C, |za| > r} that is an open set. Of course,

D(a, r) is also the closure
32 CHAPTER 1. THE COMPLEX PLANE
of D(a, r).
Remark 3. Be careful: some sets are neither open nor closed!
Consider for example, for 0 < a < b real numbers, S = {z C, |z| [a, b[}.
One can note that for any r > 0, D(a, r) S; since a S, S cannot be open.
On the other side, for any r > 0, D(b, r) S = , so that D(b, r) C\S;
hence S is not closed either.
Finally, let us introduce the last notions of this subsection: those of
bounded and compact sets.
Denition 8. A set S C is bounded i (M R
+
, z S, |z| M).
Let S C. S compact S bounded and closed.
Examples of compact sets in C are: circles |z a| = r, closed discs

D(a, r), but also line segments [a, b] where (a, b) C


2
.
To come back to the extended complex plane, dening open discs in

C
shouldnt be a problem now: when the centre z C, we use the usual way
described above, and when z = , we simply write, for r > 0:
D(, r) = {z C, |z| > r} {} .
This denition can be made very precise using the stereographic projection:
D(, r) is the image by homeomorphism of the open disc around N on
(for the canonical topology of ). Moreover, one can show that, since is
compact,

C is also compact.
1.3.2 Convexity and connectedness
In this subsection, we will be interested in characterising the shape of
subsets of the complex plane.
Convex and polygonally connected sets
Denition 9. Let S C. S is convex i,
_
(a, b) S
2
, [a, b] S
_
.
1.3. A BIT OF TOPOLOGY IN THE COMPLEX PLANE 33
Figure 1.9: Examples of a convex and a non-convex sets.
This means that, for any two points of S, the line segment joining the
two points is contained in S. For this reason, it is obvious that a set like
the union of two non-intersecting discs is not convex. It the same way, the
complex plane, from which one has removed a line cannot be convex.
Let us look at two examples. C\R and C\[0, +[ are clearly not convex,
but they are nevertheless quite dierent; whereas in the case of C\R, the
two half-planes
+
and

are strictly disconnected, it is not the case for


C\[0, +[: in this case, two points with positive real parts cannot be joined
by a straight line segment, but they can clearly be joined by a nite series of
line segments that avoid [0, +[. This illustrate the fact that convexity is
not sucient to characterize a subset of C. One should introduce a class of
sets for which polygonal routes can be employed to join points in the sets.
Let us rst dene a polygonal route precisely.
Denition 10. Let (z
0
, z
1
, ...z
n1
, z
n
) C
n+1
, for n N

. A polygonal
route from z
0
to z
n
is the set:
[z
0
, z
1
] [z
1
, z
2
] ... [z
n1
, z
n
] .
34 CHAPTER 1. THE COMPLEX PLANE
Figure 1.10: A polygonal route between a and b in S.
This allows us to characterize subsets like C\[0, +[:
Denition 11. A subset S C is polygonally connected i
_
(a, b) S
2
, (z
1
, ..., z
n1
) C
n1
, [a, z
1
] [z
1
, z
2
] ... [z
n1
, b] S
_
This means that for any two points a and b of S, there exists a polygonal
route from a to b that lies completely in S.
It is clear that every convex set in polygonally connected. Any annulus, on
the contrary, is polygonally connected but not convex.
Denition 12. A subset G of C is connected i it cannot be de-
composed into the union of non-empty open sets G
1
and G
2
such that
G
1
G
2
= . In other words, if G is connected and G
1
G and G\G
1
are both open, then, necessarily G
1
= G or G
1
= .
1.3. A BIT OF TOPOLOGY IN THE COMPLEX PLANE 35
A non-empty open connected subset of C is called a region.
Theorem 3. Let G C be a non-empty open set. Then, G is a region i
G is polygonally connected. In particular, any non-empty open convex set is
a region.
Proof. First, suppose that G is a region. Let a G and:
G
1
= {z G, a polygonal route from a to z in G} .
G
1
is then the subset of G that is polygonally connected to a. We will write:
G
2
= G\G
1
. It is clear that G
1
= , because a G
1
. The idea of the
proof is to show that both G
1
and G
2
are open. Since G is a region it is
connected, so this implies that G = G
1
. G is open, so, for any z G, we
can nd a r > 0 such that D(z, r) G. Let w be an element of D(z, r). By
construction [z, w] D(z, r) G. If z G
1
, then there is, by denition a
polygonal route in G from a to z in G
1
, and the addition of [z, w] to this
route gives a new polygonal route from a to w via z, so that, w G
1
. This
means that D(z, r) G
1
, and G
1
is open. On the other hand, if z G
1
(z G
2
), that means that there is no polygonal route from a to z, so clearly
no polygonal route from a to z via w. Hence w G
2
. Then, we have shown
that D(z, r) G
2
, proving that G
2
is open. G
1
and G
2
are thus both open.
It follows, by connectedness of the region G, since G
1
= , that G
1
= G,
and therefore, G is polygonally connected.
Conversely, suppose that G is non-empty, open and polygonally con-
nected. In order to prove the result by contradiction, we will suppose that
G is not a region. In other words, we suppose that there exist two dis-
joint non-empty open sets G
1
and G
2
such that G = G
1
G
2
. Consider
a G
1
and b G
2
. Since G is polygonally connected, we can construct
a polygonal route between a and b in G, P = [z
0
, z
1
] ... [z
n1
, z
n
] with
36 CHAPTER 1. THE COMPLEX PLANE
Figure 1.11: Proof that G is a region i it is polygonally connected.
a = z
0
and b = z
n
. Then, at least one of the line segments [z
k
, z
k+1
] is such
that z
k
G
1
and z
k+1
G
2
. A point of this segment can be described
by z(t) = (1 t)z
k
+ tz
k+1
with t [0, 1]. Since G
1
G
2
= , for each
t [0, 1], either z(t) G
1
, or z(t) G
2
. Moreover, since G
1
and G
2
are
open, if z(t) G
1
(resp. z(t) G
2
), then > 0, z(t + ) G
1
(resp.
> 0, z(t + ) G
2
). Let S = sup{t [0, 1], z(t) G
1
}. It is clear, from
what we just said, that q ]0, 1[ (because G
1
is open, so there is necessarily
an element of the segment close enough to z
k
to still be in G
1
). But we can
iterate this process! Consider z(q). Since G
1
is open, there exists > 0 such
that z(q + ) G
1
, in contradiction with the denition of q. If z(q) G
2
,
since G
2
is open, we can also nd a > 0 such that, for all s satisfying
0 < q < s q, we have z(s) G
2
. This again contradicts the denition
of q. Hence, we see that G has to be a region.
1.3.3 Limits and continuity
It is time to start investigating genuine notions of complex analysis. We will
begin with the concepts of limits and continuity.
1.3. A BIT OF TOPOLOGY IN THE COMPLEX PLANE 37
A few denitions
We rst have to dene what we will call sequences, as well as some properties
of these sequences.
Denition 13. A sequence (z
n
)
nN
is a one-to-one relation between
the natural numbers n N and complex numbers z
n
. In other words,
it is an ordered list of complex numbers. Please note that, in some
cases, we will need to dene a sequence on a subset of N rather than
N itself.
A sequence (z
n
)
nN
is bounded i there exists M R such that,
n N, |z
n
| M.
A sequence (z
n
)
nN
converges with limit a C i
> 0, N N, n N |z
n
a| < .
A sequence (u
k
)
kN
is a subsequence of the sequence (z
n
)
nN
i
there exists natural numbers (n
i
)
iN
with i N, n
i
< n
i+1
, such that
k N, u
k
= z
n
k
.
The convergence of a sequence is simply the fact that, for n big enough,
the members of the sequence accumulate arbitrarily close around a given
complex number a, that is, for this reason called the limit of the sequence.
We will use the following notation to describe a limit:
lim
n+
z
n
= a.
We can now dene limits and continuity for complex-valued functions.
Denition 14. Let f : S C be a function dened on a subset S C.
Let a

S. Then, the limit of f when z tends to a, noted lim
za
f(z)
exists and is equal to w C i
> 0, > 0, (z S, 0 < |z a| < ) |f(z) w| < .
38 CHAPTER 1. THE COMPLEX PLANE
Let a S. f is continuous at a i
> 0, > 0, (z S, |z a| < ) |f(z) f(a)| < .
Note, in the denition of the limit, that |z a| > 0, i.e., the limit is
determined by what happens to the function as it approaches a. f may not
even be dened at this point a, so its value there is of no importance for the
concept of limit. It is dierent for the notion of continuity. Nevertheless,
one sees that a function is continuous i lim
za
f(z) exists and equals f(a).
Obviously, a function is said to be continuous if it is continuous at each
point of its domain.
The operations on limits translate easily from those in the case of real
analysis, and we will use them without further proofs. We simply list them
here for functions (similar results hold for sequences).
Proposition 8. Let f : S C C and g : T C C, Let z
0
S T,
and suppose that:
lim
zz
0
f(z) = A and lim
zz
0
g(z) = B.
Then:
lim
zz
0
(f +g) (z) = A+B
lim
zz
0
(fg) (z) = AB
lim
zz
0
_
f(z)
g(z)
_
=
A
B
if b = 0..
Also, it is intuitive (and easy to prove) that a complex sequence and a
complex function converge i their real and imaginary parts converge in R.
We also list a few important results that we will use later without proofs:
these proofs would be long and time-consuming without bringing anything
decisive to the subject of these lectures. Moreover, they can be easily
adapted from their counterparts in real analysis.
Theorem 4. Any bounded sequence in C has a convergent subsequence.
1.3. A BIT OF TOPOLOGY IN THE COMPLEX PLANE 39
A corollary of this theorem is that any innite compact subset S of C
has a limit point in S. This is known as the Bolzano-Weierstrass theorem.
This leads to the important convergence theorem:
Theorem 5. Cauchy convergence theorem.
(z
n
)
nN
converges i > 0, N N, (m, n) N
2
, m, n N, |z
m
z
n
| < .
Another useful result is:
Theorem 6. Let S C be compact, and f : S C a continuous function.
Then:
f is bounded,i.e.:
M > 0, z S, |f(z)| M
f attains its bounds, i.e.:
(z
1
, z
2
) S
2
, |f(z
1
)| |f(z)| |f(z
2
)| .
Finally, we conclude this section by a last, unproven theorem from real
analysis, that is a direct consequence of the intermediate value theorem:
Theorem 7. Let [a, b] R and f : [a, b] Z a continuous function. Then,
f is constant.
1.3.4 Exercises
1. Determine if the following sets of the complex plane are open, closed
or neither open nor closed:
(i) S
1
= {z C, |z| < 1, 0 < arg(z) < /3};
(ii) S
2
= D(i, 2) D(2, 1);
(iii) S
3
= {z C, Im(z) = 1, 0 < Re(z) < 1}.
2. Find the limits, if they exist, of the following complex sequences:
40 CHAPTER 1. THE COMPLEX PLANE
(i) u
n
= (i)
n
;
(ii) u
n
=
(i)
n
n
;
(iii) u
n
=
1
n
+i;
(iv) u
n
=
n
2
2in+4
in
2
+6n+3i
.
3. Find, if they exist, the following limits of complex functions:
(i) lim
zi
z
2
+z+i
2iz
3
+z+3
;
(ii) lim
z1+

3i
arg(z)
z
;
(iii) lim
zi/2
1
4z
2
+1
.
4. Determine whether or not the following functions are continuous at
the given a C:
(i) f(z) = z
2
+iz + 2, a = i;
(ii) f(z) =
arg(z)
z
2
+1
, a = i;
(iii) f(z) =
z+1
z2
, a = i.
1.4 Curves, paths and contours
This section is an introduction to a key object in complex analysis: contours.
Indeed, in order to develop the theory of integration in the complex plane,
we will need to go beyond the simple description of curves as subset of the
complex plane, as we have done so far. We will need to consider them as
route followed by a moving point; this is exactly what we have done in the
proof of Theorem 3, when we had to consider the points of a segment whose
endpoints were in two dierent open sets. For that, the route followed by
the moving point is described by a real parameter, the complex number
corresponding to the point being a function of this real parameter.
1.4. CURVES, PATHS AND CONTOURS 41
1.4.1 Denitions
Let us begin by dening the concepts of curves and paths.
Denition 15. Let (a, b) R
2
and [a, b] be a closed bounded interval of R.
A curve with parameter interval [a, b] is a continuous function : [a, b]
C. Its initial point is (a), and its nal point is (b). is closed i
(a) = (b).
is said to be simple i (s, t) ]a, b[, s = t, (s) = (t).
For such a curve , we will denote the image of [a, b], i.e. {(t), t [a, b]}
by

. The curve is said to lie in a set S i

S.
Remark 4. [a, b] is a compact set of R. Since

is its image through the


continuous function it is thus also compact (in C). This implies that

is a closed set.
One sees that, in the notion of curve, there is naturally embedded, a
notion of orientation: the image of the curve is described in a particular
direction, from (a) to (b). Of course, for any curve , there is a curve
with the same image but the opposite orientation:
()(t) = (a +b t), for t [a, b].
A line segment between two point z
a
and z
b
is clearly the image in the
complex plane of the curve : [0, 1] C such that t [0, 1], (t) =
(1 t)z
a
+tz
b
.
One sees that it is fairly simple to join curves together to form new
curves: if
1
and
2
are two curves dened respectively on [a
1
, b
1
] and [a
2
, b
2
],
such that
1
(b
1
) =
2
(a
2
), their join can be dened as : [a, b] C, with:
(t) =
_

1
(t) if t [a
1
, b
1
],

2
(t +a
2
b
1
) if t [b
1
, b
1
+b
2
a
2
].
Then, a polygonal route as those used in the previous section is the image
of the join of line segments.
42 CHAPTER 1. THE COMPLEX PLANE
Denition 16. A function f : [a, b] C is said to be dierentiable at
t [a, b] i
lim
h0
g(t +h) g(t)
h
with t +h [a, b], exists.
It is dierentiable i it is dierentiable at any t [a, b]. If it exists, this
limit is noted g

(t) and is called the derivative of g at t.


A curve is smooth if it has a continuous derivative for all value of t
in [a, b].
Denition 17. A path is the join of nitely may smooth curves.
Remark 5. Note that any curve is certainly the join of nitely many smooth
curves, with the requirement that the pieces dont intersect, whereas, in a
path, they can intersect as many times as they want.
1.4.2 Contours
It is clear that circlines are a special case of paths. We have seen that the
line segment between z
a
and z
b
is simply the curve : [0, 1] C such that
t [0, 1], (t) = (1 t)z
a
+ tz
b
, that is also a path. Moreover, a circular
arc centred on a C, of radius r > 0, between two angles and such that
0 2, and described clockwise (resp. anticlockwise) is the image
of the path (resp. ) such that (t) = a +re
it
for t [, ].
A circline path is the join of nitely many paths corresponding to line seg-
ments or circular arcs.
Denition 18. A contour is a simple, closed circline path.
To put it short, the image of a contour is made of nitely many circular
arcs and line segments that do not cross each other. A contour is said to be
positively oriented i, as t increases, its image is described anticlockwise
round any point inside it.
1.4. CURVES, PATHS AND CONTOURS 43
1.4.3 Exercises
1. Give a parametrisation of the following contours:
(i) the square ABCD, with A = 1 i, B = 1 + i, C = 1 + i,
D = 1 +i;
(ii) the arc of circle centered on 0, between A =

3 +i and B = 2i;
(iii) the contour made of the segment [R, R], together with the pos-
itive semi-circle between R and R.
2. Consider the three curves:

1
(t) =

2
2
(1 +i)t for t [0, 1]

2
() = e
i
for
_

4
,
3
4
_

3
(s) =

2
2
(1 +i)s for s [0, 1].
Characterize the image:

2
(
3
)

.
44 CHAPTER 1. THE COMPLEX PLANE
Chapter 2
Complex functions
45
46 CHAPTER 2. COMPLEX FUNCTIONS
We are now entering in the heart of complex analysis. Before studying
the class of functions that are of most interest in complex analysis, i.e. holo-
morphic functions, in the next chapter, we will concentrate in the present
chapter on complex series, and a certain number of complex functions that
are dened through their series expansions. The last such function that we
will present is the complex logarithm, and it will be the occasion to introduce
the concept of multifunction and its phenomenology.
2.1 Complex series and power series
As you remember, many real functions f : R R can be expressed as power
series:
f(x) =
+

n=0
c
n
x
n
,
the c
n
s being real constants, at least on a given interval x ] R, R[, where
R is called the radius of convergence of the series. In real analysis, one
has a few criteria to determine the radius of convergence of a series, but
no clear reason to understand the specic value of this quantity for a given
series. Consider for example the two real functions:
F(x) =
1
1 x
2
and G(x) =
1
1 +x
2
.
By using the geometric series:
1
1 y
=
+

n=0
y
n
for y ] 1, 1[ ,
and making the changes of variable y = x
2
for F and y = x
2
for G, one
nds that:
F(x) =
+

n=0
x
2n
G(x) =
+

n=0
(1)
n
x
2n
.
2.1. COMPLEX SERIES AND POWER SERIES 47
Both these series have a radius of convergence R = 1 (you can apply the
ratio or the root tests). In the case of F, the reason for that can easily be
understood in real analysis: the function diverges at x = 1, so that its
radius of convergence corresponds to the rst singularity in the function;
the divergence in the series expansion is the result of a genuine divergence
in the function itself. But what about G? it is perfectly regular at x = 1,
and yet, its series expansion is only valid in ] 1, 1[. This can easily be
understood if we now go to the complex plane. Consider

G(z) = 1/(1 +z
2
)
to be the extension of G to complex variables. Then, it is clear that

G is
divergent i 1 + z
2
= 0, in other words, at z = i. The distance between
the centre of the expansion, 0, and i is exactly 1. And again, the radius
of convergence of the real series corresponds to the distance to the nearest
singularity, but this time in the complex function that generalizes the real
function to complex variables! This is a general and very powerful result,
and we will see in the following of this course how complex series are much
easier to deal with than real series, and to a certain extend, how a lot of
results from real analysis are much more understandable when we consider
them in the complex plane. In this section, we will properly dene complex
series and complex power series, and explore a few of their properties. We
will see in the next chapter how they are key to complex analysis.
2.1.1 Complex series
Denition 19. Let (c
n
)
nN
be a complex sequence. The complex series

c
n
of generic term c
n
is the sequence of the partial sums of c
n
: s
N
=

N
n=0
c
n
. The series

c
n
is said to converge i the sequence s
N
converges
when N +. The limit is then: s =

+
n=0
c
n
.
As in the case of sequences, and for the same reasons, a complex series
converges i its real and imaginary parts converge in R.
Example 2. Here are a few complex series:
48 CHAPTER 2. COMPLEX FUNCTIONS
(i)

(i)
n
;
(ii)

(i+1)
n
n
2
;
(iii)
_
1
2
_
n
e
in/3
.
Let us list a series of results about convergent series.
Theorem 8. If

c
n
converges, then:
(i) lim
n+
c
n
= 0;
(ii) M > 0, n N, |c
n
| M .
Proof. (i) Use the Cauchy convergence criterion for the partial sums:
n

i=0
c
i
converges
> 0, N N, (m, n) N
2
, m > N, n > N, |
m

i=0
c
i

j=0
c
j
| <
> 0, N N, (m, n) N
2
, m n > N, |
m

k=n
c
k
| < .
In particular, for m = n:
> 0, N N, n N, n > N, |c
n
| < ,
which is exactly the expression of the fact that c
n
converges towards
0.
(ii) The second point is a direct consequence of the rst one: if there exists
an n such that |c
n
| > M, then their is an obvious contradiction with
the rst point.
Proposition 9. Let

a
n
and

b
n
be two convergent complex series.
Then, for any k C,

(a
n
+kb
n
) is a convergent series, and
+

n=0
(a
n
+kb
n
) =
+

n=0
a
n
+k
+

n=0
b
n
.
2.1. COMPLEX SERIES AND POWER SERIES 49
Proof. For ease of notation, let us write A =

+
n=0
a
n
and B =

+
n=0
a
n
.
Then, for k C:
n

i=0
(a
i
+kb
i
) (A+kB) =
_
n

i=0
a
i
A
_
+k
_
n

i=0
b
i
B
_
.
The triangle inequality then gives:

i=0
(a
i
+kb
i
) (A+kB)

i=0
a
i
A

k
_
n

i=0
b
i
B
_

.
Let > 0. Since both

a
n
and

b
n
converge, there exists N > 0 such
that for all n > N:

i=0
a
i
A

<

2

i=0
b
i
B

<

2|k|
.
Hence, for any n > N, |

n
i=0
(a
i
+kb
i
) (A+kB)| < . This shows the
required convergence and limit.
Proposition 10. Let

c
n
be a complex series.
If the (real) series

|c
n
| converges, then

c
n
converges.

c
n
is then said
to be absolutely convergent.
Proof. By the triangle inequality, we have, if m > n:

i=0
c
i

i=0
c
i

k=n
c
k

k=n
|c
k
| .
But,

c
n
converges, so it is a Cauchy sequence in R. Hence:
> 0, N > 0, m > n > N,

i=0
|c
i
|
n

j=0
c
j

k=n
|c
k
|

<
Since n N, |c
n
| > 0, we have |

m
k=n
|c
k
|| =

m
k=n
|c
k
|. So, using the rst
inequality, we have shown that:
> 0, N > 0, m > n > N,

i=0
c
i

i=0
c
i

< .
50 CHAPTER 2. COMPLEX FUNCTIONS
This proves that

n
i=0
c
i
is a Cauchy sequence. Hence it converges.
We can now prove a certain number of criteria that will be very useful
in investigating the convergence of complex series.
Proposition 11. Comparison test.
Let

b
n
be a convergent real series with n N, b
n
0. Let (a
n
)
nN
be a
complex sequence.
If k > 0, n N, |a
n
| kb
n
, then

a
n
is absolutely convergent, and hence
convergent.
The proof is evident and is left to the reader.
Proposition 12. Ratio test (aka: dAlemberts test):
Let

c
n
be a complex series such that:
lim
n+

c
n+1
c
n

= l exists.
Then:
(i)

|c
n
|, and hence

c
n
, converges if l < 1;
(ii)

|c
n
| diverges if l > 1;
(iii) the test is inconclusive if l = 1.
The test is inconclusive if l = 1.
Proof. We consider the complex series

c
n
such that lim
n+

c
n+1
cn

= l
exists, and we note v
n
= |c
n+1
/c
n
|.
(i) Let us suppose that l1. Then for n big enough, all the v
n
s will be
arbitrarily close to l. This means that there is an < 1 such that for
there exists N N for which, if n > N, v
n
< , or, equivalently:
2.1. COMPLEX SERIES AND POWER SERIES 51
|c
n+1
| |c
n
|. In particular:
|c
N+1
| |cN|
|c
N+2
| |C
N+1
|
...
|c
n
| |c
n1
|
This gives: |c
n
| |c
N
|
N

n
, for any n > N. For < 1, the series

converges, so, the comparison test proves that

c
n
is absolutely
convergent, hence convergent.
(ii) A similar argument in the case l > 1 proves that

|c
n
| diverges.
Proposition 13. Root test.
Let (c
n
)
nN
be a complex sequence such that lim
n+
n
_
|c
n
| = l exists.
Then:
(i) if l < 1, then

|c
n
| converges (so does

c
n
);
(ii) if l > 1, then

|c
n
| diverges;
(iii) if l = 1, then the test is inconclusive.
Proof. The proof is analogue to the one for the ratio test: in the case l < 1,
there are < 1, and N N such that for any n > N,
n
_
|c
n
| . So,
|c
n
|
n
. The comparison test ends the proof.
2.1.2 Power series
Now that we have set the general context for complex series, we turn to the
particular case of power series.
Denition 20. A power series is a complex series of the form

c
n
(za)
n
,
where a C, (c
n
)
nN
is a complex sequence, and z C.
52 CHAPTER 2. COMPLEX FUNCTIONS
It is clear that, up to a redenition Z = z a, one could work with
power series of the form

c
n
Z
n
, but the introduction of z a will make
more sense in the future development. It is important to note that power
series are dierent from polynomials: polynomials have only nitely many
terms.
Example 3. Here are a few examples of power series:
(i)

(i)
n
(z 2)
n
;
(ii)

1
n
2
z
n
;
(iii)
_
n
2
+ 2n + 1
_
(z i)
n
.
Note that power series are functions of the complex variable z. Hence,
their convergence will depend on the value of z. This is why we introduce
the notion of radius of convergence.
Denition 21. The radius of convergence, R R{+}, of the power
series

c
n
(z a)
n
is:
R = sup{|z a|,

c
n
(z a)
n
converges}.
We will write R = + if the series converges everywhere (for any value
of z). In other words, R is the radius of the biggest open disk around a on
which

c
n
z
n
converges; outside the closed disk, the series diverges, and it
can converge or diverge on the disk itself. This justies the term radius of
convergence. Actually, we have dened the radius of convergence in terms
of ordinary convergence of the series

c
n
z
n
, but we have a stronger result.
This is the topic of the following lemma.
Lemma 1. Let

c
n
z
n
be a power series with radius of convergence R.
Then:
(i)

c
n
z
n
converges absolutely on D(0, R);
(ii)

c
n
z
n
diverges for any z C\

D(0, R) (i.e. for |z| > R).


2.1. COMPLEX SERIES AND POWER SERIES 53
Proof. (i) Let z D(0, R). Then, by denition |z| < R. By denition
of a supremum, there exists w

D(0, r) such that |z| < |w| R
and

c
n
w
n
converges. Then, |c
n
w
n
| is bounded: there exists M > 0
such that n N, |c
n
w
n
| M. Moreover, |c
n
z
n
| = |c
n
w
n
|.

z
w

z
w

n
. Now, since |z/w| < 1 by construction,

z
w

n
is nothing
but a geometric series and thus, it converges towards 1/(1 q) with
q =

z
w

. Finally, the comparison test implies that

c
n
z
n
converges
absolutely for any z such that |z| < R.
(ii) Now, let us pick z such that |z| > r and suppose for a contradiction
that

c
n
z
n
. Then, there exists M > 0 such that, for any n N,
|c
n
z
n
| < M. Now, the region for which |z| > R is open; son we can pick
up w with R |w| < |z|. Then: |c
n
w
n
| = |c
n
z
n
|

w
z

n
M|

w
z

n
. The
geometric series

w
z

n
converges because |w/z| < 1. Hence, by the
comparison test,

c
n
w
n
converges, which contradicts the denition
of R.
It remains now to give a few criteria to nd the radius of convergence
of power series. These criteria are direct consequences of the ratio and root
tests for series.
Proposition 14. Let

c
n
z
n
be a power series with radius of convergence
R. Then:
R
1
= lim
n+
n
_
|c
n
| = lim
n+

c
n+1
c
n

.
Of course, the last formula only applies if the c
n
s are non-zero.
Proof. Simply applies the ratio and root tests to the power series.
Finally, let us note that we could dene series of functions that are
standard complex series depending on a variable z C, but are not power
series. For example:
(i)

z
2n
1+nz
;
54 CHAPTER 2. COMPLEX FUNCTIONS
(ii)

_
zi
z+2i
_
n
.
The convergence of such series will, again, depend on the value of z, but,
in general, one cannot speak of a radius of convergence anymore, because
the subset of the complex plane where they converge is no longer necessarily
a disk, but can be more complicated. Can you determine the set of z for
which the two examples above converge, using the ratio or the root tests?
2.1.3 Exercises
1. Consider the geometric series:

z
n
, for z C. Prove that the series
converges inside D(0, 1), and that:
z C, |z| < 1,
+

n=0
z
n
=
1
1 +z
.
2. Using the previous result, prove that:
z C, |z| < 1,
+

n=0
(1)
n
z
n
=
1
1 z
.
3. Find the radius of convergence, and the disk of convergence of the
following power series:
(i)

(i)
n
n
(z 2i)
n
;
(ii)

n
2
+3n+2
in
2
+n+1
(z 3)
n
;
(iii)

(1 +i)
n
(z 1)
n
.
2.2 Some complex functions
Now that we know the main properties of power series, we are going to use
them to build some fundamental functions of complex analysis.
2.2. SOME COMPLEX FUNCTIONS 55
2.2.1 The exponential function
The rst, and certainly one of the most important complex function is the
exponential function. There are many way to introduce it, and we will
choose to dene it through its power series expansion.
Proposition 15. Consider the complex power series

z
n
n!
. It has an in-
nite radius of convergence.
Proof. Let us apply the ratio test to the series:

z
n+1
/(n + 1)!
z
n
/n!

=
|z|
n + 1
.
This ratio tends to 0 for any value of z, hence, the series converges with an
innite radius of convergence.
Since this series reduces to the Taylor expansion of the exponential func-
tion when z is restricted to be a real number, we propose the following
denition
Denition 22. We call complex exponential function, and we note
e
z
= exp(z) the limit of the series

z
n
n!
:
z C, e
z
=
+

n=0
z
n
n!
.
Here are the fundamental properties of the exponential:
Proposition 16. (i) e
0
= 1;
(ii) e
z+w
= e
z
e
w
;
(iii) z C, e
z
= 0.
Proof. (i) The result is obvious if one puts z = 0 in the series expansion.
56 CHAPTER 2. COMPLEX FUNCTIONS
(ii) By denition, we have

+
n=0
z
n
n!
= lim
N+

N
n=0
z
n
n!
, and this se-
quence converges. Hence, its product with the same sequence with z
replaced by w also converges, and it does so towards the product of
the two limits. In other words:
e
z
e
w
= lim
N+
N

n=0
N

p=0
z
n
n!
w
p
p!
= lim
N+
_
1 +z +
z
2
2!
+...
z
N
N!
_
[1 +w +
w
2
2!
+... +
w
N
N!
]
= lim
N+
_
1 + (z +w) +
(z +w)
2
2!
+... +
(z +w)
N
N!
_
= e
z+w
The third line was obtained by using Newtons binomial formula: (z +
w)
N
=

N
k=0
n!
(nk)!k!
z
k
w
nk
.
(iii) Using the rst two results, we have e
z
e
z
= 1, hence e
z
= 0.
Note that the coecients in the series expansion of the complex exponen-
tial are real numbers, so, if one restricts z to be a real number, one recovers
the usual real exponential function. This leads to the following property: if
we write: z = x +iy with (x, y) R
2
, we have:
|e
z
| = e
x
.
This necessarily implies that y R, |e
iy
| = 1.
Proof. To prove the rst statement, let us write:
|e
z
|
2
= e
z
e
z
= e
z
e
z
= e
z+ z
= e
2x
= (e
x
)
2
2.2. SOME COMPLEX FUNCTIONS 57
The second line results from the expansion of the exponential and of the
fact that complex conjugation is continuous, implying that one can invert
conjugation and limit. This gives |e
z
| = e
x
, since both sides are positive
real numbers. Finally, since |e
z
| = |e
x+iy
| = |e
x
e
iy
|, we have trivially that
|e
iy
| = 1 (remember that z C, e
z
= 0).
Remark 6. We have not yet made the link between the complex exponential
introduced here and the one we introduced for notational purposes in the
rst chapter, while discussing Eulers formula. This link will come soon,
once we have treated complex trigonometric functions.
Before commenting briey on the geometry of the mapping generated by
the complex exponential, it will be useful to introduce another characteri-
zation of the exponential.
Proposition 17.
z C, e
z
= lim
n+
_
1 +
z
n
_
n
.
Proof. Using the binomial theorem:
_
1 +
z
n
_
n
=
n

p=0
n!
p!(n p)!
_
z
n
_
p
=
n

p=0
n!
p!(n p)!n
p
z
p
Hence, it is enough to prove that
n!
(np)!n
p
tends to 1 when n tends to innity,
because then, we recover the series expansion of the exponential. Indeed,
we have:
n!
(n p)!
= n(n 1)...(n p + 1) n
p
when n + .
So, it is clear that
n!
(np)!n
p
1 when n +.
The geometry of the mapping induced by the exponential can be char-
acterized as follow:
58 CHAPTER 2. COMPLEX FUNCTIONS
A vertical line L = {z C, z = x + iy, x = a R} is mapped into a
circle of centre 0 and of radius e
x
.
An horizontal line l = {z C, z = x + iy, y = b R} is mapped into
a line through 0 making an angle b with the real axis.
2.2.2 Complex trigonometric and hyperbolic functions
As for the exponential, we shall use power series to dene complex trigono-
metric and hyperbolic functions. We will see that there exists a strong dual-
ity between trigonometric and hyperbolic functions of the real variables, in
the sense that cos(z) (resp. sin(z)) reduces to the real cosinus (resp. real si-
nus) on the real axis, and to the real cosh (resp. real sinh) on the imaginary
axis.
Denition 23. Let z C. We dene the trigonometric and hyperbolic
complex functions as:
cos(z) =
+

n=0
(1)
n
z
2n
(2n)!
cosh(z) =
+

n=0
z
2n
(2n)!
sin(z) =
+

n=0
(1)
n
z
2n+1
(2n + 1)!
sinh(z) =
+

n=0
z
2n+1
(2n + 1)!
.
The fact that these series converge can easily be realized by applying the
ratio test to each of them. This will also show that they have an innite
radius of convergence. It is now apparent, by a simple reorganization of the
terms of the trigonometric series, that:
z C, e
iz
= cos(z) +i sin(z) .
2.2. SOME COMPLEX FUNCTIONS 59
This implies, in particular, Eulers formula:
R, e
i
= cos() +i sin() .
Also, we can see that de Moivres formula follows straightforwardly from the
denitions introduced above. Equivalently, one can write the trigonometric
and hyperbolic functions in terms of the exponential. For z C:
cos(z) =
e
iz
+e
iz
2
; sin(z) =
e
iz
e
iz
2i
;
cosh(z) =
e
z
+e
z
2
; sinh(z) =
e
z
e
z
2
.
By a simple comparison of these formulae (or, equivalently, using the
series expansions) one sees that:
z C, cos(iz) = cosh(z) and sin(iz) = i sinh(z) .
Restricting these relations to the real and imaginary axis lead to the prop-
erties cited in the introduction.
Addition properties of complex trigonometric and hyperbolic functions
are exactly identical to their counterparts for real variables. If (z, w) C
2
:
cos(z +w) = cos(z) cos(w) sin(z) sin(w)
sin(z +w) = cos(z) sin(w) + sin(z) cos(w)
cosh(z +w) = cosh(z) cosh(w) + sinh(z) sinh(w)
sinh(z +w) = cosh(z) sinh(w) + sinh(z) cosh(w) .
These properties follow directly from the expressions of the trigonomet-
ric and hyperbolic functions in terms of the exponential. Note that the
similar relations for substractions come from the symmetry properties of
the functions:
cos(z) = cos(z) ; sin(z) = sin(z)
cosh(z) = cosh(z) ; sinh(z) = sinh(z) .
60 CHAPTER 2. COMPLEX FUNCTIONS
2.2.3 Roots of unity
Now that we have dened properly the complex exponential and trigono-
metric functions, we are equipped to study an important set of algebraic
equations For z C and n N

, let us consider the equation z


n
= 1. By
writing z = re
i
, it is immediate that r = 1 (|z
n
| = |z|
n
). Hence, the points
z satisfying the equation are located on the unit circle. Then, we have:
_
e
i
_
n
= 1 e
in
= 1
cos(n) +i sin(n) = 1
cos(n) = 1 and sin(n) = 0
n = 2k, k {0, 1, ..., n 1} . (2.1)
In the last line, note that k is smaller or equal to n1. This comes from the
fact that the argument of a complex number is dened up to a factor of 2:
k n corresponds to values of that lead to z = e
i
equivalent to the ones
covered by k {0, 1, ..., n1}. Hence, the equation z
n
= 1 in C has exactly
n solutions, called the nth roots of unity: z = e
2ki/n
, k {0, 1, ..., n1}.
By construction, they are located on the unit circle, at angles 2ki/n, with
k {0, 1, ..., n 1}: they form a regular n-gon centred at 0 and with one
vertex at z = 1. If n is odd, the only real root is z = 1; if n is even, there
are two real roots: z = 1. The case n = 6 is depicted on gure 2.1.
We will see in this course that any complex polynomial equation of order
n admits exactly n roots (not necessarily distinct); this is the fundamental
theorem of algebra. There exists many ways to prove this theorem, and
we will prove it in the last chapter of these notes. But, at the moment, you
will be able to prove that a polynomial with real coecients, of order two,
has always two roots in the complex plane.
Exercise 6. Let az
2
+ bz + c = 0 for (a, b, c) R
3
and z C, with a = 0.
2.2. SOME COMPLEX FUNCTIONS 61
Figure 2.1: 6th roots of unity forming a regular hexagon.
Prove that this equation has exactly two roots, given by:
z
1,2
=
b
2a

1
2a
_
b
2
4ac , if b
2
4ac 0
z
1,2
=
b
2a

i
2a
_
|b
2
4ac| , if b
2
4ac < 0
(2.2)
2.2.4 The logarithmic function
Before discussing the complex logarithm in details, we need to come back
to our denition for the argument of a complex number. The fact that
the complex exponential is periodic with period 2 implies that the real
number such that for z C

, z = re
i
is not unique. This is the source of
one subtlety of complex analysis that is not present in real analysis: many-
valuedness of complex functions. For any z C

, we will all the argument


62 CHAPTER 2. COMPLEX FUNCTIONS
(instead of an argument) of z the set:
arg z = { R, z = |z|e
i
} .
In real analysis, the logarithm is the inverse of the exponential function,
i.e., for any x R

+
, there exists a unique y R such that e
y
= x. This
y is then denoted y = ln(x). We can apply the same method in complex
analysis: for any z C

, we look for a w C such that e


w
= z. Let us write
w = u +iv with (u, v) R
2
. Then, on has:
|z| = |e
u
e
iv
| = e
u
arg z = {v + 2k, k Z} . (2.3)
Hence, we have:
e
w
= z w = ln |z| +i, with arg z .
We thus dene the logarithm of any z C

via:
ln z = {ln |z| +i, arg z} .
This clearly shows that the logarithm of a complex number z is not uniquely
dened: its real part is unique, equal to ln |z|, but its imaginary part can
be any real number that is an argument of z. This is the rst example of a
multi-valued function: to a given complex number z, the complex logarithm
associates an innity of complex numbers. In the next section, we will see
how to treat such multifunctions.
2.2.5 Exercises
Solve the following equations in C, and represent the solutions in the complex
plane:
(i) 2z
5
+ 1 = 0;
(ii) z
2
+z + 1 = 0;
2.3. MULTIFUNCTIONS 63
(iii) z
4
+ 2z
2
+ 4 = 0;
(ii) sin(2z) = 2;
(iv) z
4
+i = 0;
(v) z
1/3
= 2i;
(vi) ln(z
2
) = 2 +

4
i.
To go further
2.3 Multifunctions
As we have seen with the example of the logarithm, multifunctions appear
every-time we try to invert a complex function (like the exponential) that is
not globally one-to-one. In this section, we are going to develop a method
to construct one-to-one function from multifunctions. We will come back to
these techniques later, once we have introduced the concept of holomorphy.
2.3.1 Example 1: the logarithmic function
Let us start with the logarithm encountered above:
z C

, ln z = {ln |z| +i, arg z} .


Hence, values of that dier from each other by an integer multiple of
2 lead to the same point z = re
i
in the complex plane plane, but give
dierent value for its logarithm ln |z| + i. Nevertheless, if we restrict
[0, 2[ or ] , ], we obtain a single-valued function. We will call such a
restriction a principal-value determination of the argument of z. Usually,
64 CHAPTER 2. COMPLEX FUNCTIONS
Figure 2.2: Cut of the complex plane for the principal determination of the
logarithm.
such a principal-value determination of the argument is noted = Arg(z).
One should note that, despite its name, this prescription does not have
anything particular (despite being the most widely used): any interval of
length 2 that is closed at one end and open at the other one would do the
trick. Let us choose the principal determination by restricting ] , ].
This introduces a branch cut (or cut for short) in the complex plane, i.e.,
the semi-axis of negative real numbers ] , 0] is cut out of the complex
plane and z cannot cross this cut while roaming in the complex plane. This
cut has two edges: the upper edge is identied with the argument = ,
and the lower edge to = . Our principal determination implies that
the lower edge of the cut is excluded from the complex plane, but its upper
edge included. The point 0, that is in the cut is called a branch point.
In the cut plane, we can now dene a family of single-valued functions:
k Z, ] , ], z = re
i
C

, f
k
(z) = ln r +i( + 2k) .
2.3. MULTIFUNCTIONS 65
Clearly, we have that:
z C

, ln z = {f
k
(z), k Z} .
In the cut plane, each f
k
is continuous, by continuity of its real and imag-
inary parts. But on the cut, they are not continuous, and in crossing the
cut from the upper half-plane to the lower half-plane, one has to transfer
from f
k
to f
k+1
: the number k counts the number of time we have accom-
plished a complete (anti-clockwise) tour around the branch point 0 (neg-
ative values indicate clockwise tours). This transfer is continuous because
lim
h0
+ f
k
(ih) = lim
h0
f
k+1
(ih).
2.3.2 Branch points and multibranches
Now that we have seen how to proceed to extract single-valued functions
from the logarithm, we can try and generalize our intuition to more general
multifunctions.
Let us rst prove a simple result:
Proposition 18. There is no restriction which selects a real function (z)
arg z for all z C

, so that : z (z) is a continuous function.


Proof. Let us assume for a contradiction that such a continuous argument
function exists. Then, consider:
v(t) = (e
it
) for t R .
By composition of continuous functions, v : R R is thus continuous.
Moreover, v(t + 2) = v(t), so v is periodic, with a period of 2. Now, v(t)
and t are both arguments of e
it
by construction, so, there exists a function
n : R Z such that:
v(t) t = 2n(t) .
v being continuous, n is also continuous. So, by the theorem on continuous
integer-valued function from chapter 1, it is constant: t R, n(t) = n, and
66 CHAPTER 2. COMPLEX FUNCTIONS
v(t) = t + 2n for any t R. But, v(t) = v(t + 2); this implies that:
t + 2n = t + 2 + 2n, hence the contradiction.
This results means that any principal value of the argument function
of z = re
i
obtained by restricting the domain of has to have a jump
discontinuity somewhere. This is particularly obvious if [0, 2[: then,
when z = e
i
describes a complete circuit anticlockwise, starting from z = 1,
starts at 0, and increases continuously towards 2, where z reaches 1 again.
This result has implications for other multifunctions. For example, the
imaginary part of a complex exponential is an argument function by con-
struction, so there is no continuous logarithm on C

.
Branch points
Consider now a multifunction w(z) that is a non-empty set of C for any z
in the domain of denition of w.
Denition 24. A branch point of w(z) is a point a C such that, for all
r > 0, it is not possible to choose f(z) w(z) such that f is a continuous
function on the circle centred on a and of radius r.
Namely, that means that the denition of w(z) implicitly or explicitly
involves the argument , where z a = |z a|e
i
, in another conguration
than e
i
(i.e. a pure phase). One clearly sees that the motivation for such a
denition comes from the proposition above regarding the argument func-
tion. The fact that 0 is a branch point for the logarithm thus comes from
the fact that the imaginary part of the logarithm is exactly the argument of
the complex number into the logarithm, i.e the angle on a circle centred on
0.
Lets look at another example: ln((z 1)/(z +1)). If one writes z 1 =
|z 1|e
i
and z +1 = |z +1|e
i
, one has ln((z 1)/(z +1)) = ln(|(z 1)/(z +
1)|e
i()
). Hence: ln((z 1)/(z +1)) = {ln |(z 1)/(z +1)| +i( ),
arg(z 1), arg(z + 1)}. So, we see that both the angles around 1
2.3. MULTIFUNCTIONS 67
and 1 appear in the denition of the multifunction: 1 and 1 are branch
points for this multifunction.
Multibranches
As we have seen previously, there is a way to construct continuous selections
from multifunctions. The key to this procedure is to make a change of
variable and replace z by (r, ) around each branch point a, such that z =
a + re
i
. Consider rst a multifunction w(z) with only one branch point
(such as the logarithm). That means that, for z = a:
w(z) = {w(z) = w(r, ), arg(z a)} .
Hence is determined only up to a integer multiple of 2. Therefore, by
restricting ]c, c + 2] with c R, we have the multibranches:
k Z, F
k
(r, ) = w(r, + 2k) .
That are continuous functions of r and . Hence,
w(z) = {F
k
(r, ), k Z, ]c, c + 2]} .
The set (F
k
)
kZ
is called a complete set of multibranches for w(z).
Now, observe what happens when z describes the circle centred on a and of
radius r. Then z = re
it
, with t going from 0 to 2. Consider the kth branch:
F
k
(r, t = 2) = w(r, 2 + 2k) = w(r, 0 + 2(k + 1)) = F
k+1
(r, t = 0). This
means that, when z travels anticlockwise around the branch point, there
is a natural, continuous transfer from F
k
to F
k+1
. This technique only
works with one branch point. The theory with several branch points is
more delicate and will not be treated here, but we will see on a specic
example how to address the problem practically.
2.3.3 Example 2: Fractional powers
We have seen how to treat the logarithm multifunction in order to construct
a complete set of multibranches for it. We will see in this subsection, how
68 CHAPTER 2. COMPLEX FUNCTIONS
to apply this method to another important class of functions: the fractional
powers. Consider n Z

\{1, 1}. Consider the equation w = z


1/n
. Then,
if w = re
i
and z = e
i
, we have: r =
1/n
and =

n
+
2k
n
, for k Z.
Hence, we have a multifunction:
z
1/n
= {|z|
1/n
e
i/n
, z C

, arg z} ,
with a branch point at 0. We choose [0, 2[, so that we cut the plane
along the positive real axis. We dene the branches:
z = re
i
, z = 0, k {0, 1, ..., n 1}, g
k
(z) = r
1/n
e
i(+
2k
n
)
.
Then, we have:
z
1/n
= {g
k
(z), k {0, 1, ..., n 1}} .
This time the complete set of multibranches is nite.
2.3.4 Example 3: An example with two branch points
What happens if the multifunction has two branch points?
The idea in the previous cases, with one branch point, was to cut the
complex plane in a way that prevent the possibility to construct closed
contour paths that include the branch point in their interior. The idea
remains the same when there are more than one branch point. Consider
f(z) =

z
2
+ 1. This can be rewritten as: f(z) =
_
(z i)(z +i), and
introducing z i = re
i
and z +i = e
i
, so that and are the arguments
around i and i respectively, we have: f(z) = f(r, , , ) = re
i(+)/2
.
Hence, if z describes a circle around i (resp. i), varies by 2 and varies
a bit but comes back to its original value (resp. varies by 2 and varies
a bit but comes back to its original value), so, at the end of the loop, f(z)
has a new value: f
new
(z) = f
old
(z), whereas z has returned to the same
value. This proves that i and i are branch points for the multifunction f.
Now, to construct well dened branches, we need two cuts, in order to
2.3. MULTIFUNCTIONS 69
Figure 2.3: Cut of the complex plane for f(z) =

z
2
+ 1.
prevent the two angles and to vary by integer multiples of 2. For
example, we can cut parallel to the real axis, along the negative real parts
for both points, restricting (, ) ] , ]
2
. We could also cut along the
imaginary axis, from i to innity and fromi to . Finally, let us mention
another interesting cut. If we choose a branch cut that is the line segment
[i, i], we clearly prevent any closed path around i or i separately. But,
we allow closed paths that encircle both branch points at the same time.
Actually, it is not a problem: along such a path, both and vary by 2,
and f(z) returns to its initial value, so that these paths do not introduce
the need for any new branch cut. This is a general result: to produce well-
behaved branches of a multifunction, it is enough to introduce cuts that
prevent the existence of closed paths around each isolated branch point.
70 CHAPTER 2. COMPLEX FUNCTIONS
Chapter 3
Complex dierentiation
71
72 CHAPTER 3. DIFFERENTIATION
Dierentiation in the complex plane is central to complex analysis. It
is similar, but not identical to dierentiation in real analysis. It allows one
to introduce a class of complex function called holomorphic functions, and
complex analysis can be viewed as the study of these holomorphic functions.
3.1 Holomorphic functions
3.1.1 Dierentiation and the Cauchy-Riemann equations
We will rst dene dierentiability in the complex plane, and introduce a
necessary condition for a function to be dierentiable: the Cauchy-Riemann
equations.
Denition 25. Let f : S C a complex valued function dened on S C.
Let G S an open subset of S. Then, f is dierentiable at z G i:
lim
h0
f(z +h) f(z)
h
for any h such that z +h G
exists. When this limit exists, it is denoted by f

(z) and it is called the


derivative of f at z.
This denition relies strongly on the existence of the open set G. Indeed,
since G is open, for any z G, there is an r > 0 such that D(z, r) G. In
other words, for any h C such that |h| < r, z+h G. This ensures that, in
the limit written above to dene the derivative, the point z+h can approach
z from any direction as h tends to zero. In other words, the derivative
exists only if the value of the limit does not depend on the way
z + h approaches z. This is similar to what happens in real analysis: a
real function dened on an open interval of R is not dierentiable at x if
the derivatives from the left and from the right of x are not the same. We
can use this idea to illustrate the non-dierentiability of a simple complex
3.1. HOLOMORPHIC FUNCTIONS 73
function. Consider f(z) = Imz dened in C. Let us construct:
f(z +h) f(z)
h
=
Im(z +h) Im(z)
h
=
Im(h)
h
. (3.1)
Choose h R, so that Im(h) = 0, then
f(z+h)f(z)
h
0 when h 0. But,
now, choose h such that Re(h) = 0; then
f(z+h)f(z)
h
i when h 0. So,
in that case, we have found two ways of approaching z that do not give the
same limit; that implies that the function f : z Imz is not dierentiable
in C. Selecting two ways of approaching the point z that do not lead to the
same limit, as we just did, is quite a general method to prove that a function
is not dierentiable.
Theorem 9. Let f : G C where G is an open subset of C. Let f be
dierentiable at z G. Let z = x + iy and f(z) = u(x, y) + iv(x, y). Then
u and v, as real functions, have partial derivatives at (x, y) R
2
. These
partial derivatives satisfy the Cauchy-Riemann equations:
_
u
x
=
v
y
u
y
=
v
x
.
(3.2)
In the following, we will often denote
u
x
= u
x
(and the same for the
derivative with respect to y).
Proof. Since f is dierentiable at z, we have that, for any h C such that
z +h G:
f

(z) = lim
h0
f(z +h) f(z)
h
exists.
Moreover, since we can choose h freely (i.e. we can approach z however we
want in G), we can restrict it to be purely real on the one hand, and purely
imaginary on the other. Thus, we have:
for h R:
f

(z) = lim
h0
_
u(x +h, y) u(x, y)
h
+i
v(x +h, y) v(x, h)
h
_
= u
x
+iv
x
;
74 CHAPTER 3. DIFFERENTIATION
and for h = ik, k R:
f

(z) = lim
h0
_
u(x, y +k) u(x, y)
ik
+
v(x, y +k) v(x, y)
k
_
=
1
i
u
y
+v
y
.
The partial derivatives exist because the limit dening f

(z) exists.
Now, the limit is unique by construction, so:
u
x
+iv
x
=
1
i
u
y
+v
y
,
or v
x
+ iu
x
= u
y
+ iv
y
. So, equating real and imaginary parts, we
recover the Cauchy-Riemann equations.
Note that the Cauchy-Riemann equations are a necessary condition for
a function to be dierentiable. They are not sucient. That means
that the contrapositive of the previous theorem can be used to show that
a function is not dierentiable at a point: if one proves that the Cauchy-
Riemann equations do not hold, then the function is not dierentiable. But,
if they hold, this is not sucient to prove that the function is dierentiable.
To see this, lets go back to our previous example: f(z) = Im(z) for z C.
We have seen that this function is not dierentiable. We can conrm that
by proving that the Cauchy-Riemann equations do not hold. Indeed, we
have u(x, y) = 0 and v(x, y) = y. So, u
x
= 0 = 1 = v
y
.
Now, consider f(z) =
_
|Re(z)Im(z)| for z C. At z = 0, we have, for
h C:
f(0 +h) f(0)
h
=
_
Re(h)Im(h)
h
.
Hence, we see that when we approach 0 with h R, the limit of this ratio
is 0 (because the ratio is identically 0). But, if we approach 0 along the line
making an angle /4 with the real axis, we can write h = t(1+i) with t R,
the ratio is constant and equal to 1/(1 + i) = 0. That proves that f is not
dierentiable at 0. On the other hand, the Cauchy-Riemann equations are
3.1. HOLOMORPHIC FUNCTIONS 75
trivially satised at z = 0. This should emphasize that Cauchy-Riemann
equations should be handled with care.
To go further
Despite this warning, a slight modication of our theorem provides a partial
converse to the previous theorem: it is enough to add the continuity of the
partial derivatives.
Theorem 10. Let f : G C with G an open subset of C. Let z = x+iy C
and f(z) = u(x, y) +iv(x, y). If u and v have continuous rst order partial
derivatives in G that satisfy the Cauchy-Riemann equations at z, then f

(z)
exists.
Proof. Let z G. Let r > 0 such that D(z, r) G. Consider h = p+iq C
such that |h| < r. Then, for (z, h) =
f(z+h)f(z)
h
:
(z, h) =
p
h
_
u(x +p, y +q) u(x, y +q)
p
+i
v(x +p, y +q) v(x, y +q)
p
_
+
q
h
_
u(x, y +q) u(x, y)
q
+i
v(x, y +q) v(x, y)
q
_
Now, since the functions u and v are dierentiable, they are continuous on
G, so one can apply the mean value theorem and nd (, , , ) ]0, 1[
4
such that:
f(z +h) f(z)
h
=
p
q
_
u
x
(x +p, y +q) +i
v
x
(x +p, y +q)
_
+
q
p
_
u
y
(x, y +q) +i
v
y
(x, y +q)
_
.
Now, one can use the continuity of the partial derivatives on G to nd that
for > 0, there exists h > 0 small enough such that:

f(z+h)f(z)
h
g(z)

<
where:
g(z) =
p
q
(u
x
(x, y)+iv
x
(x, y))+
q
h
(u
y
(x, y)+iv
y
(x, y)) = u
x
(x, y)+
1
i
u
y
(x, y) .
76 CHAPTER 3. DIFFERENTIATION
The last equality holds because of the Cauchy-Riemann equations. This
shows that f

(z) exists and is equal to g(z).


In principle, this theorem can be used to test the dierentiability of a
function. But it is not very practical, and we will shortly see much more
powerful results to achieve this goal.
3.1.2 Holomorphic functions
Until now, we have used the decomposition of complex numbers and com-
plex functions into their real and imaginary parts to talk about complex
dierentiation. It is time to forget about all this and to deal directly with
the complex variable. Again, the notion of open sets will be central to the
developments presented here.
Denition 26. A complex function f that is dierentiable at any point of
an open set G included in its domain of denition is said to be holomorphic
in G.
This means that,for any z G, irrespective of the way h tends to zero,
lim
h0
(f(z+h)f(z))/h exists. We will denote by H(G) the set of functions
holomorphic in a given open set G.
Denition 27. A function f is said to be holomorphic at a point a C
is there exists r > 0 such that f is dened and holomorphic in D(a, r).
It is important to realize that being holomorphic at a point a is a stronger
condition than being dierentiable at a: in order to be holomorphic at a, f
has to be dierentiable at a and at every point of a disk centred on a.
We can now list a few properties of holomorphic functions. The proofs
are left to the reader, as they are identical to their counterparts in real
analysis. Let G be an open subset of G. The following properties can be
3.1. HOLOMORPHIC FUNCTIONS 77
derived by proving the appropriate dierentiability conditions at each point
z G.
Let f and g be holomorphic in G and let C. Then, f, f +g and fg
are holomorphic in G and the following rules for dierentiation apply,
for all z G:
(f)

(z) = f

(z)
(f +g)

(z) = f

(z) +g

(z)
(fg)

(z) = f

(z)g(z) +f(z)g

(z) .
Let f be holomorphic in G and g be holomorphic in an open set con-
taining f(G). Then, g f is holomorphic in G and, for all z G:
(g f)

(z) = (g

f)(z)f

(z) = g

(f(z))f

(z) .
This property is often referred to as the chain rule.
Let f be holomorphic in G such that z G, f(z) = 0. Then, 1/f is
holomorphic in G and, for any z G:
_
1
f
_

(z) =
f

(z)
(f(z))
2
.
These rules can now be used to test the holomorphy of complicated functions
knowing the holomorphy of simple functions. It will be much easier than
using the Cauchy-Riemann equations.
For example, the function f(z) = z is trivially dierentiable for any
z C, as are any constant functions. This implies that any polynomial:
P(z) =
N

n=0
c
n
z
n
where N N and n {0, 1, .., N}, c
n
C, is holomorphic in C. Remember
that a polynomial is the sum of nitely many terms. Power series, that are
78 CHAPTER 3. DIFFERENTIATION
the sums of innitely many terms, will be treated separately. In the same
way, any rational function P(z)/Q(z) where P(z) and Q(z) are polynomials
is holomorphic in any open set in which Q(z) is never zero. For example,
1/(1 +z
2
) is holomorphic in C\{i, i}.
To go further
Finally, let us see what happens in the extended complex plane C. In C,
holomorphy is stated as above, but what happens at ? We have already
seen a map z 1/z that interchanges with a point of C, namely 0. Let f
be a function dened on a set {z C, |z| > r} for some r > 0. By dening

f
such that

f(z) = f(1/z), we have f() =

f(0). Hence, any property of

f at
0, such as continuity, limit, holomorphy, can be transferred to f at innity.
Consider, for instance, f(z) = z
2
. Then,

f(w) = 1/w
2
, that is not holomor-
phic at w = 0 (Check it); so we can say that f is not holomorphic at innity.
Conversely, consider f(z) = 1/(1 + z
2
) for |z| > 1. Then, f() = 0, and

f(w) = w
2
/(1 +w
2
) for |w| < 1:

f is holomorphic at 0, so f is holomorphic
at .
3.1.3 Some useful results
Now, we will prove a certain number of results on holomorphic functions
that will be useful in the rest of this course. The rst one states that an
holomorphic function is necessarily continuous. This is analogous to the real
case, where dierentiability implies continuity (but in complex analysis, we
need holomorphy, not mere dierentiability).
Proposition 19. Let f : S C with S C. Let G be an open subset of S.
If f is holomorphic in G, then it is continuous on G.
Moreover, if F G is a compact subset of G, then f is bounded on F.
3.1. HOLOMORPHIC FUNCTIONS 79
Proof. Suppose f is holomorphic on G. Then, for any z G and for any
h C such that z +h G:
f

(z) = lim
h0
f(z +h) f(z)
h
.
So, we can dene, for h = 0, the function:
(h) =
f(z +h) f(z)
h
f

(z) .
So, by denition, (h) 0 when h 0. Hence, we have:
f(z +h) f(z) = h(f

(z) +(h)) .
This implies that:
lim
h0
|f(z +h) f(z)| = 0 ,
which is exactly the requirement for f to be continuous on G.
Now, remember that we have stated, in the rst chapter that a continuous
function on a compact subset of C is bounded on this domain. So, since f
is continuous on G, it is continuous on F G, and since F is compact, f is
bounded on F.
Proposition 20. Let f : S C C be holomorphic on a region G S.
Then, f is constant on G if any of the following condition is true:
(i) z G, f

(z) = 0;
(ii) |f| is constant in G;
(iii) z G, Im(f(z)) = 0.
Proof. Remember that a region is a non-empty open connected subset of C.
Let us suppose that G = D(0, 1). For any z = x +iy D(0, 1), let us write
f(z) = u(x, y)+iv(x, y). Since f is holomorphic on G, the Cauchy-Riemann
equations hold, and:
f

(z) = u
x
+iv
x
= v
y
iu
y
.
80 CHAPTER 3. DIFFERENTIATION
(i) Suppose z G, f

(z) = 0. Then, u
x
= v
x
= u
y
= v
y
= 0 identically
on G. Let p = a+ib and q = c+id be two arbitrary point of G. Then,
construct r = c + ib and s = a + id. It is obvious that |r|
2
+ |s|
2
=
|p|
2
+ |q|
2
< 2, so, at least one of r or s is in D(0, 1). Let us suppose
r D(0, 1), without loss of generality. The functions x u(x, b) and
y u(c, y) are real functions with vanishing derivatives, so, by virtue
of the mean value theorem, they are constant. Hence:
u(a, b) = u(c, b) and u(c, b) = u(c, d) .
So, u(a, b) = u(c, d). With the same argument, we can prove that:
v(a, b) = v(c, b) and v(c, b) = v(c, d) ,
so that v(a, b) = v(c, d). This shows that f(p) = f(q) for any p and q
in G.
(ii) Now, let us suppose that z D(0, 1), |f(z)| = c, with c R
+
con-
stant. Then: u
2
+v
2
= c
2
, so that:
uu
x
+vv
x
= 0 and uu
y
+vv
y
= 0 .
By using the Cauchy-Riemann equations, this leads to uu
x
vu
y
=
uu
y
+ vu
x
= 0, so: (u
2
+ v
2
)u
x
= 0. If u
2
+ v
2
= 0, then f = 0
on D(0, 1), so f is constant. If u
2
+ v
2
= 0, then, u
x
= 0 on D(0, 1).
Similarly, u
y
= v
x
= v
y
= 0 on D(0, 1), so that f is constant on D(0, 1)
(according to the rst point).
(iii) Finally, take f real valued on D(0, 1). Then, v = 0, so that v
x
=
v
y
= 0, which implies, through the Cauchy-Riemann equations, that
u
x
= u
y
= 0. Hence, f is constant on D(0, 1) (according to the rst
point).
The proof presented here generalizes to any region of C by replacing the
simple route (p, r, q) of the rst point to an arbitrary polygonal route con-
sisting of horizontal and vertical line segments. It is more messy, but does
not involve anything new.
3.2. SOME HOLOMORPHIC FUNCTIONS 81
3.1.4 Exercises
1. Which of these functions are dierentiable at the given point a C:
(i) f(z) = z|z|, a = 0;
(ii) f(z) = |z|
2
, a = 0;
(iii) f(z) = arg(h), a = 0 (where arg is restricted to [0, 2[);
(iv) f(z) = z, for any a C.
2. Give the domain of holomorphy of the following function, and calculate
its derivative in this domain:
f(z) =
z
z
2
+ 1
.
3. Let f : G C C be holomorphic in the open set G. For z = x+iy
C, dene:
f
z
=
1
2
_
f
x
+i
f
y
_
f
z
=
1
2
_
f
x
i
f
y
_
,
where f is regarded as a function of (x, y), f(z) = f(z(x, y)), on the
right-hand side.
Check that the partial derivatives of f(z(x, y)) with respect to x and
y exist, and show that:
f
z
= 0 and
f
z
= f

(z).
Conversely, prove that a dierentiable function f that satises
f
z
= 0
is holomorphic in G.
3.2 Some holomorphic functions
We will now come back to the functions we have dened previously, and look
at their holomorphy. Before that, we have to prove a very powerful result
82 CHAPTER 3. DIFFERENTIATION
on the dierentiation of power series: any function that can be written as a
power series is holomorphic in the disc of convergence of the power series.
Later in the course, we will prove another very powerful result: the fact that
the contrapositive is also true, namely that any holomorphic function can be
written as a power series. These two results together mean that holomorphic
functions and power series are one same notion!
3.2.1 A result on the dierentiation of power series
Let us see what the derivative of a power series could be.
Lemma 2. Let (c
n
)
nN
. The power series

c
n
z
n
and

nc
n
z
n1
have the
same radius of convergence.
Proof. Let us suppose rst that

c
n
z
n
converges for |z| < R, R being its
radius of convergence (remember that this implies that

|c
n
z
n
| converges
for |z| < R). Let R
+
such that |z| < < R. Assume that z = 0. Then:
|nc
n
z
n1
| =
n
|z|
_
|z|

_
n
|c
n

n
| .
Since |z|/ < 1, the ratio test tells us that

n(|z|/)
n
converges. That
implies that the general term of this series tends to 0 when n tends to
innity, so it is bounded:
M > 0, n N, n(|z|/)
n
M .
So:
|nc
n
z
n1
|
M
|z|
|c
n

n
|
The real series

|c
n
|
n
converges because || < R by construction. Hence
the comparison test gives that

nc
n
z
n1
converges absolutely, hence con-
verges for any z such that |z| < R. The fact that it does not converge for
|z| > R follows from a similar argument, from the fact that

c
n
z
n
diverges
for |z| > R. Conversely, if

|nc
n
z
n1
| converges for |z| < R, then:
n N

, |c
n
z
n
| |z||nc
n
z
n1
| ,
3.2. SOME HOLOMORPHIC FUNCTIONS 83
so

|c
n
z
n
| converges by the comparison test, and thus,

c
n
z
n
converges.
Theorem 11. Let f(z) =

+
n=0
c
n
z
n
with a radius of convergence R > 0.
Then, f H(D(0, R)), and:
z D(0, R), f

(z) =
+

n=1
nc
n
z
n1
.
Proof. The previous lemma allows to dene, for all z D(0, R), a function
g such that:
g(z) =
+

n=1
nc
n
z
n1
.
For z D(0, R) and h C such that z +h D(0, R):
f(z +h) f(z)
h
g(z) =
+

n=1
_
(z +h)
n
z
n
h
nz
n1
_
c
n
.
We will need the binomial expansion:
(z +h)
n
=
n

k=0
C
k
n
z
nk
h
k
with C
k
n
=
n!
k!(n k)!
.
Then:
(z +h)
n
z
n
h
nz
n1
=
nhz
n1
+... +C
k
n
h
k
z
nk
+... +h
n
h
nz
n1
= h
_
C
2
n
z
n2
+... +C
k
n
h
k2
z
nk
+... +h
n2
_
= h
n

k=2
C
k
n
h
k2
z
n2
= h
n2

i=0
n!
(n (i + 2))!(i + 2)!
h
i
z
ni2
84 CHAPTER 3. DIFFERENTIATION
by the change i = k 2. So, using the triangle inequality recursively:
+

n=0

(z +h)
n
z
n
h
nz
n1

|c
n
| |h|
+

n=0
n2

i=0
|c
n
|
n!
(n 2 i)!(i + 2)!
|h|
i
|z|
n2i
|h|
+

n=0
n(n 1)
n2

i=0
|c
n
|
(n 2)!|h|
i
|z|
n2i
(n 2 i)!(i + 2)!
|h|
+

n=0
n(n 1)|c
n
|(|z| +|h|)
n2
.
Let > 0 such that |z| < < R and |z|+|h| < (always possible since h can
be as small as desired). Then, clearly, |h|

+
n=0
n(n 1)|c
n
|(|z| +|h|)
n2
<
|h|

+
n=0
n(n 1)|c
n
|
n2
. Moreover, by applying the lemma to

nc
n

n1
and

n(n1)c
n

n2
, we know that

+
n=2
n(n1)
n2
converges to a nite
value independent of h, so f

(z) exists and is exactly equal to g(z).


Example 4. Consider the power series

(i)
n
2
n
(zi)
n
. By applying the ratio
or the root test, we see that it denes a function f(z) =

+
n=0
(i)
n
2
n
(z i)
n
on its disk of convergence D(i, 2). Then, f H (D(i, 2)), and:
z D(i, 2), f

(z) =
+

n=1
ni
n
2
n
(z i)
n1
=
+

n=0
(n + 1)i
n+1
2
n+1
(z i)
n
.
3.2.2 The exponential function
Since we have dened the exponential via its power series expansion, with
an innite radius of convergence:
z C, e
z
=
+

n=0
z
n
n!
,
the theorems we have proved in the previous subsection show that:
the exponential is holomorphic on the entire complex plane C;
3.2. SOME HOLOMORPHIC FUNCTIONS 85
the derivative of the exponential (e
z
)

=
d
dz
e
z
is given, for any z C,
by:
d
dz
e
z
=
+

n=1
n
z
n1
n!
=
+

n=1
z
n1
(n 1)!
=
+

k=0
z
k
k!
= e
z
. (3.3)
3.2.3 Complex trigonometric and hyperbolic functions
For the trigonometric and hyperbolic functions, the same procedure as in
the case of the exponential tells us that they are all holomorphic on the
entire complex plane, with, for all z C:
d
dz
cos z = sin z ,
d
dz
sin z = cos z ,
d
dz
cosh z = sinh z ,
d
dz
sinh z = cosh z .
3.2.4 The logarithmic function
For the logarithm, things are a bit more subtle, since we have seen that the
function itself is not well dened on the entire complex plane. Rather, one
has to introduce a cut and select a particular determination of the logarithm.
By putting a cut along the negative real-axis, we have created an innite
sequence of branches f
k
for the logarithm, that constitute a complete set of
multibranches:
ln z = {f
k
(r, ), k Z}
with
k Z, z = re
i
, r > 0, =] , ], f
k
(z) = f
k
(r, ) = ln r +i( + 2k) .
Then, we have the result:
86 CHAPTER 3. DIFFERENTIATION
Proposition 21. For all k Z, f
k
is holomorphic in C

= C\] , 0],
with:
z C

, f

k
(z) =
1
z
.
Proof. Let z C

, and let h C such that z + h C

. Denote =
f
k
(z+h)f
k
(z). Then, the continuity of f
k
on C

implies that lim


h0
= 0.
Now, using the fact that e
f
k
(z)
= z for any z C

, we have:
h = e
f
k
(z+h)
e
f
k
(z)
= e
f
k
(z)
(e

1) = z(e

1) ,
so that:
f
k
(z +h) f
k
(z)
h
=
1
z

1
. (3.4)
Consider g() =

e

1
, for = 0. Then:
1
g()
=
1

_
+

n=0

n
n!
1
_
=
1

n=1

n
n!
=
+

k=0

k
(k + 1)!
.
Hence:
0

1
g()
1

k=1

k
(k + 1)!

k=1
||
k
(k + 1)!

||
2
when 0 .
So, when tends to zero, the dominant term in this sum is || and it tends
to zero, so, we have: 1/g() 1 when tends to zero; or equivalently,
g() 1. This means that, when h tends to zero,
f
k
(z+h)f
k
(z)
h
tends to
1/z. Hence f
k
is holomorphic with f

k
(z) = 1/z on C

.
3.3. CONFORMAL MAPPING 87
3.2.5 Exercises
Find the domain of holomorphy of the following functions, and calculate
their derivatives in this domain:
(i) f(z) = exp
_
1 +iz
3
_
;
(ii) f(z) =
sin(iz+3)
z
2
1
;
(iii) f(z) = tan(z)cos(z);
(iv) f(z) = tanh(z)cosh(z);
(v) f(z) = cosh (2 sin(z) +i).
To go further
3.3 Conformal mapping
In this section, we will briey study mappings between regions of the com-
plex plane that preserve angles. In particular, we will see that any holomor-
phic function whose derivative is non-zero denes such a mapping.
3.3.1 Conformal mapping
Consider a path with parameter interval [0, 1], for convenience. Then,
there is a well-dened tangent to at = (0): + t

(0), for t 0,
provided

(0) = 0. This tangent makes an angle arg

(0) with the real


axis.
Now, let
1
and
2
be two paths, both with parameter interval [0, 1], with a
common starting point
1
(0) =
2
(0) = . Assume that

1
(0) and

2
(0) are
88 CHAPTER 3. DIFFERENTIATION
both non-zero, so that each path has a well-dened tangent at . The angle
between
1
and
2
at is then simply the angle between their tangents at
that point: arg

1
(0) arg

2
(0).
Theorem 12. Conformality theorem.
Let f : C C be holomorphic in an open set G, and let
1
and
2
be paths,
with parameter interval [0, 1], in G meeting at =
1
(0) =
2
(0). Suppose
that f

() = 0. Then, f preserves angles between paths in G meeting at .


This means that two paths meeting at with an angles are transformed by
f into two paths meeting at f() with an angle .
Proof. Let = arg

1
(0) arg

2
(0) be the angle between
1
and
2
at .
The paths
1
and
2
are mapped by f to paths f
1
and f
2
, respectively
(note that they are, indeed, paths, because f is holomorphic). These two
paths meet at f(), with an angle = arg (f
1
)

(0) arg (f
2
)

(0).
We have:
(f
1
)

(0)
(f
2
)

(0)
=
f

()

1
(0)
f

()

2
(0)
=

1
(0)

2
(0)
.
Hence, taking the argument of both side, and remembering that arg(z/w) =
arg(z) arg(w): = , which is the result that needed to be proven.
Denition 28. A complex-valued function f is conformal in an open
set G C (or

C), if f H(G) and z G, f

(z) = 0. It is said to be
conformal at a point C if it is conformal in a disc D(, r) for some
r > 0.
Hence, the conformality theorem shows that a conformal mapping pre-
serves both the magnitude and sense of angles between paths.
The conformality theorem admits a partial converse:
Theorem 13. Let G be an open set of C. let f : C C be a function such
that its partial derivatives f
x
and f
y
exist and are continuous in G. If f is
conformal in G, then, f is holomorphic in G.
3.3. CONFORMAL MAPPING 89
Proof. Consider a path in G. Let = f . Then, by a simple manipu-
lation, one can write:

(t) =
1
2
(f
x
if
y
)

(t) +
1
2
(f
x
+if
y
)

(t),
where the partial derivatives are evaluated at (t). Now, up to a change of
origin and units, we can consider that the square[1, 1] [i, i] is in G, and
consider two paths:

1
(t) = it for t [0, 1], i.e. the line segment [0, i];

2
(t) = t for t [0, 1], i.e. the line segment [0, 1].
They intersect at 0, with an angle /2: arg (

1
(0)/

2
(0)) = /2. If f pre-
serves the angle and sense in G we must therefore have: arg (

1
(0)/

2
(0)) =
/2. This implies that arg (f
x
/f
y
) = /2, hence: f
y
= if
x
, at 0. Writing
f(z) = u(x, y) +iv(x, y), this gives the Cauchy-Riemann equations.
Note that the argument would be valid for any perpendicular curve at any
point of G, provided we use the correct reparametrization of the paths, so,
we can say that f has continuous partial derivatives that satisfy the Cauchy-
Riemann equations everywhere in G. This implies that f is holomorphic in
G.
3.3.2 Some examples
We start by proving that Mobius transformations are conformal.
Theorem 14. Let f : z (az + b)/(cz + d) with ad bc = 0 be a general
Mobius transformation. f is conformal in C\{d/c} for c = 0.
Proof. It is enough to prove that Mobius transformation are holomorphic
with f

(z) = 0 on C\{d/c}. This is easily done by realizing that f is the


ratio of two holomorphic functions. Moreover, we have:
z C\{d/c}, f

(z) =
ad bc
(cz +d)
2
= 0.
90 CHAPTER 3. DIFFERENTIATION
When we studied Mobius transformation, we considered their behaviour
in

C rather that C, We would therefore like to extend our notion of confor-
mality to

C. If f maps C to , we will build g : z 1/f(z), and say
that f is conformal at if g is conformal at . We will also say that f is
conformal at if

f such that

f(z) = f(1/z) is conformal at z = 0. Let
c = 0. Then:

f() =
b +a
d +c
.
From the previous theorem, this is conformal at 0, so f is conformal at
innity.
Now, consider the behaviour at z = d/c. This point is mapped into by
f. Let = 1/w where w = 1/f(z). Then:
=
cz +d
az +b
,
and this has a non-zero derivative at z = d/c, so f is conformal at z =
d/c. Geometrically, it means that f maps a pair of circles tangent at
z = d/c into a pair of parallel lines.
Finally, in the case c = 0, f() = . By considering = 1/w as a function
of = 1/z, one can show immediately that the derivative at z = 0 is non-
zero.
Hence, we can conclude that Mobius transformations are conformal at every
point of

C.
Hence, we can now list a lot of standard conformal mappings that are
Mobius transformations:
z
zi
z+i
maps the open upper half-plane onto the unit disc;
z
z+i
zi
maps the open lower half-plane onto the unit disc;
z
z1
z+1
maps the open right half-plane onto the unit disc;
z
z+1
z1
maps the open left half-plane onto the unit disc.
Other examples of conformal maps are:
3.3. CONFORMAL MAPPING 91
The exponential map. It is conformal in C because it is holomorphic
in C, with z C,
_
e
iz
_

= e
iz
= 0.
z z
n
for n N\{0, 1} is holomorphic in C

. At z = 0, angles are
magnied by a factor of n, so the map is not conformal.
Any holomorphic branch of the logarithm. The conformality also fol-
lows from the holomorphy and the non-zero value of the derivative.
Any holomorphic branch of a general power z z

, for > 0.
In the last two cases, it is important to choose the cut so that the region we
wish to map is not aected by the introduction of the cut.
Then, conformal mappings can be constructed at will by any standard
operations on functions applied to the few we have listed above, as long as
the resulting function remains holomorphic with a non-zero derivative in the
region one wishes to transform.
92 CHAPTER 3. DIFFERENTIATION
Chapter 4
Complex integration
93
94 CHAPTER 4. INTEGRATION
The content of this chapter is the core of complex analysis. In partic-
ular, we will prove Cauchys theorem, Cauchys formulae and the residue
theorem. We will see that all this machinery of complex integration al-
lows us to prove very important theorems that are encountered everywhere
in mathematics, as, for example, the fundamental theorem of calculus, the
fundamental theorem of algebra, Liouvilles theorem, but also methods to
integrate real functions, to deal with Laplace and Fourier transforms etc.
4.1 Integration in the complex plane
To start, let us remember that we have dened a path in the rst chap-
ter of this course, as the join of nitely many smooth (i.e. dierentiable)
curves: It is a function : [a, b] R C that is piecewise continuous and
dierentiable. A contour was just a closed path made of bits of circlines
(circular arcs and line segments). These paths and contours will be essential
in the theory of complex integration. Let us recall what piecewise continuity
means. A function h : [a, b] R C is piecewise continuous on [a, b] i
there exists real numbers (t
i
)
i{0,1,...,n}
such that a = t
0
< t
1
< ... < t
n
= b
and continuous functions h
k
: [t
k
, t
k+1
] C such that h(t) = h
k
(t) for
t ]t
k
, t
k+1
[. Note that h need not be dened at the points t
k
. That means
that h is continuous everywhere on [a, b] except possibly for a nite number
of discontinuities . A real-valued function h that is piecewise continuous is
integrable with:
_
b
a
h(t)dt =
n1

n=0
_
t
k+1
t
k
h
k
(t)dt .
This comes from the fact that continuous real functions are integrable.
4.1.1 Integration along paths
What is the meaning of an object like:
_
b
a
f(z)dz ,
4.1. INTEGRATION IN THE COMPLEX PLANE 95
when a and b are complex numbers, and f : S C C a complex function?
In real analysis, Riemanns construction of the integral:
_
b
a
f(x)dx for [a, b] R and f a real function,
relies on the partition of [a, b] into smaller interval [x
i
, x
i+1
] such that i
{0, 1, ..., n}, x
i
= x
0
+ i
xnx
0
n
, with x
0
= a and x
n
= b. Then, the integral
is dened as the limit, when n tends to innity, i.e. when the size of the
subintervals tends to zero, of the sum:
n1

i=0
f(x
i
)(x
i+1
x
i
) .
Then, it can be shown that the limit does not depend on the way the inter-
val [a, b] is cut into subintervals. One could have the idea to generalize that
to a complex integral, and to calculate the integral of a function between
two complex numbers by a succession of small increments that start at a
and connect it to b. But, in the complex plane, there exists innitely many
curves that join a point a to a b. Which one should one choose?
Actually, we will dene complex integration by using a particular path be-
tween a and b. Then, we will show, later that, under certain conditions,
the result of the integration does not depend on the path. Be-
fore treating the general case of complex function, let us see the case of a
complex-valued functions dened on an interval of R.
Denition 29. Let f : [a, b] R C. Let x [a, b], f(x) = Re(f(x)) +
iIm(f(x)), where Re(f) and Im(f) are real functions. We say that f is
integrable i Re(f) and Im(f) are both integrable. If this is the case,
then, we dene:
_
b
a
f(x)dx =
_
b
a
Re(f(x))dx +i
_
b
a
Im(f(x))dx .
The easiest example of such an integral is:
_
2
0
e
ix
dx =
_
2
0
cos(x)dx +i
_
2
0
sin(x)dx = [sin(x)]
2
0
i[cos(x)]
2
0
= 0.
96 CHAPTER 4. INTEGRATION
We are now equipped to dene properly the integral of a complex function
along a path.
Denition 30. Let be a path such that : [a, b] R C. By denition,
there exists (t
i
)
i{0,1,...,n}
with a = t
0
< t
1
< ... < t
n
= b such that is
continuously dierentiable on every [t
k
, t
k+1
]. Let f :

C be continuous.
We dene the integral of f along , or round , if is closed, by:
_

f(z)dz =
_
b
a
f((t))

(t)dt .
Please, note that the integral on the right-hand side is well dened,
because (f )

is piecewise continuous, and hence integrable.


We can justify the previous formula by noting that we have: z


t [a, b], z = (t). Hence, by deriving with respect to t:
dz
dt
=

(t) .
This tells us that when we move from t to t+dt, the point on the curve moves
from z to z +dz with dz =

(t)dt. The following result, a direct application


of the denition, will be used extensively in the rest of this chapter.
Proposition 22. Let a C and r > 0. Let (a, r) be the circle centred on
a of radius r ((a, r) is then a contour). Then:
_
(a,r)
(z a)
n
dz =
_
0 if n = 1
2i if n = 1 .
4.1. INTEGRATION IN THE COMPLEX PLANE 97
Proof. We have, by denition: (a, r)(t) = a +re
it
with t [0, 2]. So:
_
(a,r)
(z a)
n
dz =
_
2
0
_
re
it
_
n
re
it
dt
= ir
n+1
_
2
0
e
i(n+1)t
dt
= ir
n+1
__
2
0
cos ((n + 1)t) dt +i
_
2
0
sin ((n + 1)t) dt
_
=
_

_
ir
n+1
_
_
sin((n+1)t)
n+1
_
2
0
i
_
cos((n+1)t)
n+1
_
2
0
_
for n = 1
i[t]
2
0
for n = 1 .
=
_
0 for n = 1
2i for n = 1 .
(4.1)
The rest of this subsection will be a list of some technical results.
Proposition 23. Let : [a, b] C be a path, and f :

C be continuous.
Then:
(i)
_

f(z)dz =
_

f(z)dz;
(ii) If ]a, b[ and we note
1
and
2
the restriction of to [a, ] and [, b]
respectively:
_

f(z)dz =
_

1
f(z)dz +
_

2
f(z)dz.
(iii) Let : [ a,

b] C be a path such that = , where : [ a,

b] [a, b]
has a positive continuous derivative. Then:
_

f(z)dz =
_

f(z)dz.
This states that the integral along a path is only dependent on the path,
and not on the parametrization of the path.
98 CHAPTER 4. INTEGRATION
Proof. The rst two points are direct consequences of the denition of
_

f(z)dz. To prove the third point, we can assume that and are smooth:
if they are not, we can use the second point and sum integrals on smooth
restrictions of the path. For t [ a,

b], we have:

(t) =

((t))

(t).
Now, we have:
_

f(z)dz =
_

b
a
f( (t))

(t)dt
=
_

b
a
f(((t)))

((t))

(t)dt
Writing s = (t), we have ds =

(t)dt, and since

> 0, the boundary of


the integral stay in the same order. Finally: t = a s = a and t =

b
s = b. Hence:
_

f(z)dz =
_
b
a
f((s))

(s)ds =
_

f(z)dz .
The following proposition can be proved by using the second property
of the proposition above, as well as the third one, applied to the particular
case of a reparametrization by translation of the parameter interval.
Proposition 24. Let : [a, b] C be a path. Suppose that is the join of
paths (
i
)
i{1,...,n}
. Let f :

C be continuous. Then:
_

f(z)dz =
n

k=1
_

k
f(z)dz .
Let us now apply what we have learned to an example
Example 5. Let be the contour whose image is formed by the join of

1
= [R, R] and

2
, the upper semi-circle centred on 0 of radius R and
4.1. INTEGRATION IN THE COMPLEX PLANE 99
described counter-clockwise (cf gure 4.2.3). We will denote
2
= (0, R).
Consider f(z) = z
2
. We have:

1
(s) = (1 t)(R) +tR for t [0, 1] ,
and:

2
(s) = Re
is
for s [0, ] .
So, we have:
_

f(z)dz =
_
1
0
((2t 1)R)
2
2Rdt +
_

0
R
2
e
2is
iRe
is
ds
=
_
2R
3
_
4
3
t
3
2t
2
+t
__
1
0
+
_
R
3
3
e
3is
_

0
= 0 .
Note that the integral along [R, R] could have been recovered directly,
since z [R, R] Im(z) = 0, so that it is just a real integral.
4.1.2 The fundamental theorem of calculus
In real analysis, a usually simple way to evaluate integral, is to recognize
the integrand as the continuous derivative of a known function, and then
to apply the fundamental theorem of calculus. It turns out that a similar
procedure is possible for complex integral, thanks to the complex version of
the fundamental theorem of calculus:
Theorem 15. Let : [a, b] C be a path, and let F be a complex function
that is dened on an open set containing

. If F

(z) exists and is continuous


at each point of

, then:
_

(z)dz = F((b)) F((a)) .


It is clear that if is closed, this integral is zero.
100 CHAPTER 4. INTEGRATION
Figure 4.1: Contour for
_

z
2
dz, in the example
Proof. Let us rst assume that is smooth. Then, F is dierentiable on
[a, b] and (F )

(t) = F

((t))

(t). Then:
_

(z)dz =
_
b
a
F

((t))

(t)dt
=
_
b
a
(F )

(t)dt
=
_
b
a
(Re (F ))

(t)dt +i
_
b
a
(Im(F ))

(t)dt
= [Re (F ) (t)]
b
a
+i [Im(F ) (t)]
b
a
= F((b)) F((a)) .
Note that we applied the real version of the fundamental theorem of calculus
to the real and imaginary parts separately.
If is not smooth, we choose a = t
0
< t
1
< ... < t
n
= b such that
4.1. INTEGRATION IN THE COMPLEX PLANE 101
the restrictions
k
s of to the [t
k
, t
k+1
] are smooth, and, by applying the
previous result to each smooth portion, we have:
_

(z)dz =
n1

k=0
_

k
F

((t))

(t)dt
=
n1

k=0
(F ((t
k+1
)) F ((t
k
)))
= F((b)) F((a)) .
Let us cite a result that is very useful when the fundamental theorem of
calculus cannot be used (and even sometimes when it can).
Proposition 25. Let : [a, b] C be a path and f :

C be a continuous
function. Then:

f(z)dz

_
b
a

f((t))

(t)

dt .
Proof. This proposition is a direct consequence of the denition of the com-
plex integral, and of the well-known result of real analysis:

_
b
a
f(x)dx

_
b
a
|f(x)| dx .
In particular, we have the corollary:
Corollary 1. If M > 0, z

, |f(z)| M, then:

f(z)dz

M length() ,
where:
length() =
_
b
a
|

(t)|dt .
Note that the function length dened above give exactly what we would
expect the length of a path or a contour to be. To convince yourself of that,
you can try for a line segment or a circle, for example.
102 CHAPTER 4. INTEGRATION
4.1.3 Exercises
Calculate the following integrals:
(i)
_

(z
2
+z + 2)dz, with = [1 +i, 0];
(ii)
_

ze
iz
2
dz, with = (0, 2);
(iii)
_

cos(z)dz, with = [0, 1] [1, 1 +i];


(iv)
_

z sinh(z
2
)dz, with = (0, 1), the counter-clockwise, positive semi-
circle centred on 0;
(v)
_

1
(zi)
2
dz, with = (i, 2);
(vi)
_

1
zi
dz, with = (i, 2).
4.2 Cauchys theorem
In this section, we are going to prove an extremely important result, called
Cauchys theorem. There are actually two Cauchys theorem: the historical
one, rst proven by Cauchy, and the Cauchy-Goursat theorem. The rst
one makes an extra assumption, compared to the second one, namely that
the rst derivative of the integrand, f, in continuous. We will rst prove
the historical theorem, and then, the Cauchy-Goursats theorem.
4.2.1 Historical Cauchys theorem
Theorem 16. Historical Cauchys theorem.
Let be a contour (simple closed path),

C. Let f be a complex function


that is holomorphic on

and in the interior of

. Suppose that f

is
continous on

and in the interior of

. Then:
_

f(z)dz = 0.
4.2. CAUCHYS THEOREM 103
Proof. By writing : t [a, b] R z(t) ]C, we have:
_

f(z)dz =
_
b
a
f (z(t)) z

(t)dt.
Now, let us write that f(z) = u(x, y) + iv(x, y), with z(t) = x(t) + iy(t).
Then:
_

f(z)dz =
_
b
a
_
ux

vy

_
dt +i
_
b
a
_
vx

+uy

_
dt.
By noting that

dt = d, we thus have, by going back to integrals in the


real Cartesian plane:
_

f(z)dz =
_

udx
_

vdy +i
_

vdx +i
_

udy.
It is time to remember a result from calculus: Greens theorem. It states
that, if two real-valued functions P(x, y) and Q(x, y), dened on the Carte-
sian plane, together with their rst-order partial derivatives are continuous
on

and in the inside of (noted I()), then:


_

Pdx +
_

Qdy =
_

I()
_
Q
x

P
y
_
dxdy.
Here, f and f

are continous on

I(), by hypothesis, so u, v and their


partial derivatives are also continous on

I(). We can then apply Greens


theorem, to get:
_

f(z)dz =
_

I()
(v
x
u
y
) dxdy +i
_

I()
(u
x
v
y
) dxdy.
But, Cauchy-Riemann equations applied to f ensure that the two integrands
are identically zero. The result thus follows.
To go further
104 CHAPTER 4. INTEGRATION
4.2.2 Cauchy-Goursat theorem
Here we will relax the hypothesis of continuity of f

. It is actually extremely
useful, since it will allow us to prove, later, that f

(and all the other higher-


order derivatives) is actually also holomorphic. In order to prove this version
of Cauchys theorem, we will need the fundamental theorem of calculus that
we proved in the previous section. But, once we will have Cauchys theorem,
its power will completely supersede the one of the fundamental theorem of
calculus, and we may not need the latter any more. We will restrict our
proof of Cauchys theorem to a version that is sucient for any application
that we have in mind in an introduction to complex analysis.
Properties of contours
In the rst chapter, we have dened a contour as a simple closed path whose
image is the join of a nite number of line segments and circular arcs. We
now have to go deeper in the properties of these objects.
Lemma 3. Let be a path and G C be an open subset of C such that

G. Then:
m R

+
, z

, D(z, m) G .
Proof. Since

G and G is open, that means that any z

is also in
G, so:
z

, M
z
> 0, D(z, M
z
) G .
Take m = min{M
z
, z

}. Then, clearly, the lemma follows because


D(z, m) D(z, M
z
) for all z

.
This lemma will be useful in proving the covering theorem:
Theorem 17. Let G C be an open set, and : [a, b] C be a path such
that

G. Then, there exists a real constant m > 0 and a sequence of


open disks (D
k
)
k{0,1,...,N}
such that:
k {0, 1, ..., N}, D
k
= D((t
k
), m) with a = t
0
< t
1
< ... < t
N
= b;
4.2. CAUCHYS THEOREM 105
k {0, 1, ..., N}, D
k
D(k + 1) = ;
k {0, 1, ..., N}, ([t
k
, t
k+1
]) D
k
;

N
k=0
D
k
G.
Proof. By using the previous lemma, we can choose m > 0 such that z

, D(z, m) G. We have to prove that

can be covered by a nite set


of such disks, each overlapping the next. Suppose, for a start, that is
smooth. Then, by applying the real mean value theorem, we have:
(s, t) [a, b]
2
, c [a, b], (Re) (s) (Re) (t) = (s t) (Re)

(c) ,
and the same for the imaginary part of . Since they are continuous on a
closed interval, [a, b], (Re)

and (Im)

are bounded. Hence:


(s, t) [a, b]
2
, > 0, |s t| < |(s) (t)| < m .
This states the uniform continuity of . This result remains valid if is not
smooth, since we can apply the same argument to the smooth pieces that
constitute . Now, we can choose a = t
0
< t
1
< ... < t
N
= b such that
k {0, 1, ..., N}, |t
k+1
t
k
| < . If we choose D
k
= D((t
k
), m), then, the
theorem is valid.
We can now state Jordan curve theorem for a contour:
Theorem 18. Let be a contour. Then, the complement, in the complex
plane, of

is of the form I() O(), where I() and O() are disjoint,
connected sets. I() is the inside of

and is bounded, whereas O() is the


outside of

and is unbounded.
Proof. We will only give the outline of the proof. Let a

. Let l be a
ray with endpoint a. Let N(a, l) be the number of times l cuts

. One can
convince himself that whether N(a, l) is odd or even does not depend on the
direction of l, but only on the position of a in C\

. Let I() be the set of


points for which N(a, l) is odd, and O() the set of points for which N(a, l)
106 CHAPTER 4. INTEGRATION
is even. The fact that I() and O() are open follows from the fact that

is closed (since it is compact, as image under a continuous function, of a


compact interval of R). Finally, to prove the connectedness of I() (the idea
is the same for O()), it is sucient to prove that any two points a and b in
I() can be joined by a path in I() made of circlines. The idea is to join a
and b to two points a

and b

in I() that are close to

, and follow

at
a xed small distance from it, staying in I() in order to join a

to b

.
The previous theorem allowed us to characterize the inside and the out-
side of a contour. What about the notion of boundary?
Theorem 19. The boundary of a set S is dened to be: S = S C\S.
Note that when S is open, C\S is closed, so C\S = C\S. Hence, if S is
open, S = S\S. For a contour , both

I() and

O() are closed,


and

is the boundary of I() and O().


Finally, we need a last result that will be crucial in our proof of Cauchys
theorem: it is a way to break up an integral along a polygonal contour into
a sum of integrals along triangles.
Theorem 20. Let be a polygonal contour in C. Let z
1
, z
2
, ..., z
n
, for n > 3
be the vertices of

. Then, it is possible to insert n3 line segments [z


j
, z
k
]
that subdivide I() into n2 triangles. Each of the inserted segments ]z
j
, z
k
[
lies in I().
Proof. Again, we will only present an outline of the proof. If I() is convex,
then, the segments [z
1
, z
k
] for k {3, ..., n 1} triangulate it. Otherwise,
one of the interior angle at some vertex, say z
1
, is greater than . Let l
be a ray emanating from z
1
such that, for r > 0 suciently small along
l, D(z
1
, r) I() = (this means that l points into I()). Moving along
this ray l from z
1
, there is a rst point of intersection of l with

; call it
w
l
= z
1
. For one ray l at least, the point w
l
is a vertex of the polygon. Let
z
k
be such a vertex. The segment [z
1
, z
k
] can then be used to create two new
4.2. CAUCHYS THEOREM 107
polygonal contours, each of whose images in C has fewer than n vertices (by
partitioning in two contours). The argument is the repeated until only
triangles remain.
Cauchy-Goursat theorem
Cauchys theorem states that, under suitable conditions on the function f,
the closed path and the set G on which f is holomorphic, we have:
_

f(z)dz = 0.
Our derivation of the fundamental integral
_
(0,1)
z
1
dz = 2i, shows that
Cauchys theorem fails for a function that is not holomorphic at every point
inside. Indeed, z 1/z is holomorphic everywhere in (0, 1), except at 0.
We say that f is holomorphic inside and on a contour i f H(G) for
some open set G such that

I() G.
The fundamental theorem of calculus implies that
_

(z)dz = 0 if is
a closed path in an open set G on which F is dened. Hence, it is tempting
to approach Cauchys theorem by trying to nd conditions under which
f H(G) has an antiderivative F (such that F

= f). Actually, we will see


that provided G is convex, this is so if
_

f(z)dz = 0 for all triangles in G.


Therefore, we shall prove Cauchys theorem for triangles rst.
Our general strategy to prove Cauchys theorem is summarized as follow:
1. Proof of Cauchys theorem for triangles.
2. Proof of the indenite integral theorem.
3. Proof of the antiderivative theorem (implied by Cauchys theorem for
triangles together with the indenite integral theorem).
4. Proof of Cauchys theorem for convex regions (implied by the an-
tiderivative theorem together with the fundamental theorem of cal-
culus).
108 CHAPTER 4. INTEGRATION
5. Proof of Cauchys theorem for a contour.
Theorem 21. Suppose that f : C C is holomorphic on an open set G
which contains a triangle and I(). Then:
_

f(z)dz = 0.
Proof. Let = [u, v, w], i.e. the triangle formed by joining [u, v], [v, w]
and [w, u]. Let u

, v

and w

be the midpoints of [v, w], [w, u] and [u, v],


respectively. Consider the triangles
0
= [u

, v

, w

],
1
= [u, w

, v

],
2
=
[v, u

, w

], and
3
= [w, v

, u

]. Then:
I =
_

f(z)dz =
_

f(z)dz =
3

k=0
_

k
f(z)dz.
The triangle inequality reads:
|I|
4

k=0

k
f(z)dz

.
Consider the k for which

k
f(z)dz

is maximum, and relabel it 1, then:

1
f(z)dz

1
4
|I| .
Note that, by Thales theorem, length(
1
) =
1
2
length(). Repeat the ar-
gument with
1
instead of . By induction, we can thus generate a sequence
of triangles
0
,
1
,
2
, ... such that:

0
= ;
n N,
n+1

n
, where
n
is the closed triangular area with
boundary

n
;
n N, length(
n
) = 2
n
L, with L = length();
4.2. CAUCHYS THEOREM 109
Figure 4.2: Subdivision of a triangle for the proof of Cauchys theorem for
a triangle.
n N, 4
n
|I|

_
n
f(z)dz

.
The set

+
n=0

n
contains a point Z common to all the triangles
n
. This
seems obvious, but to prove it, select a point z
n

n
for all n 0. Since all
the points belong to
0
, a bounded set, the sequence (z
n
)
nN
is bounded.
Therefore, it has a convergent subsequence. Let us call its limit Z. For each
n 0, Z is a limit point of the subset (z
k
)
kn
of
n
, so it belongs to
n
.
Now, let > 0. f is dierentiable at Z since it is holomorphic in the triangle.
So, there is a r > 0 such that:
z D(Z, r),

f(z) f(Z) (z Z)f

(Z)

< |z Z|.
Let N N such that
N
D(Z, r). Then, for all z
N
,
|z Z| 2
N
L.
110 CHAPTER 4. INTEGRATION
Moreover:
_

N
_
f(Z) + (z Z)f

(Z)
_
dz = 0,
according to the fundamental theorem of calculus for a closed contour ap-
plied to the function F(z) = f(Z)z + (z
2
/2 Zz)f

(Z). Now:

N
f(z)dz

N
_
f(z) f(Z) (z Z)f

(Z)
_
dz

,
and, since the integrand on the right-hand side is bounded:

N
f(z)dz

2
N
L length() = 2
2N
L
2
.
Since we have: 4
N
|I|

N
f(z)dz

, we have: |I| L
2
for an arbitrary
small , so I = 0.
Theorem 22. Indenite integral Theorem.
Let f : C C be a continuous function on a convex region G C such
that
_

f(z)dz = 0 for any triangle G. Let a G. Then, the function


F dened by:
z G, F(z) =
_
[a,z]
f(w)dw,
is holomorphic in G with F

= f.
Proof. Let z G, and D(z, r) G for r > 0, such that |h| < r implies
z +h G. For |h| < r, the line segments [a, z], [z, z +h] and [a, z +h] are all
in G since G is convex. By hypothesis,
_
[a,z,z+h]
f(z)dz = 0 =
_
[a,z]
f(w)dw+
_
[z,z+h]
f(w)dw +
_
[z+h,a]
f(w)dw. Hence:
F(z +h) F(z) =
_
[a,z+h]
f(w)dw
_
[a,z]
f(w)dw
=
_
[z+h,a]
f(w)dw
_
[z,a]
f(w)dw
=
_
[z,z+h]
f(w)dw.
We can choose the parametrization such that
_
[z,z+h]
dw = h.
4.2. CAUCHYS THEOREM 111
Figure 4.3: Proof of the indenite integral theorem.
Hence:

F(z +h) F(z)


h
f(z)

=
1
|h|

_
[z,z+h]
(f(w) f(z)) dw

1
|h|
|h| sup
w[z,z+h]
|f(z) f(w)|. (4.2)
The right-hand side clearly tends to 0 as h tends to zero. Hence, F is
holomorphic in G with F

= f.
This leads to the rst version of the antiderivative theorem:
Theorem 23. Let G be a convex region and let f H(G). Then, there
exists F H(G) such that F

= f.
Proof. This results simply from the two previous theorems. Indeed, since
f H(G), we have
_

f = 0 for any triangle G. So, by the indenite


integral theorem, there exists a function F such that F H(G) and F

=
f.
We can now prove Cauchys theorem for a convex region:
112 CHAPTER 4. INTEGRATION
Theorem 24. Cauchys theorem for a convex region.
Let G be a convex region and f H(G). Then, for every closed path such
that

G:
_

f(z)dz = 0.
Proof. The result follows from using the antiderivative theorem and applying
the fundamental theorem of calculus. Indeed, the antiderivative theorem
tells us that there exists F H(G) such that F

= f. Hence,
_

f(z)dz =
_

(z)dz = 0, the last equality holding because of the fundamental theorem


of calculus applied along a closed path.
Very often, the region in which we wish to apply Cauchys theorem is
not convex. In that case, we use the following form of the theorem:
Theorem 25. Cauchys theorem for a contour (Cauchy-Goursat
theorem).
Let f : C C be holomorphic inside and on a contour . Then:
_

f(z)dz = 0.
Proof. First, suppose that is a polygon. By triangulating , we can write:
_

f(z)dz =
N

k=1
_

k
f(z)dz,
where each
k
is a triangle (note that, as in the proof of Cauchys theorem
for triangles, the integrals along the inserted line segments cancel). Then,
by applying Cauchys theorem for triangles, the integrals of f along each
k
are zero. So
_

f(z)dz = 0.
Now, let be any contour. Let G be an open set containing

I().
By denition, f is holomorphic on

I(). We will approximate by a


polygonal contour. To do this, we use the covering theorem and introduce
overlapping discs D
k
= D((t
k
), m) for k {0, ..., N}, with t
0
< t
1
< ... <
t
N
and (t
0
) = (t
N
), that satisfy the conditions of the covering theorem.
4.2. CAUCHYS THEOREM 113
By increasing the number of discs if necessary, we can assume that each

k
, i.e. each restriction of to [t
k
, t
k+1
] is a line segment or a circular arc
(remember that a contour, in this course is the join of nitely many line
segments and circular arcs). Moreover, the line segments
k
= [(t
k
),
t
k+1
]
for k {0, ..., N1} join to form a polygonal contour such that

I( ) is
contained in

N
k=0
D
k
I(), and so, is also contained in G. By application
of the rst part of the proof, we have:
_

f(z)dz = 0.
Figure 4.4: Proof of Cauchys theorem for a convex region.
Moreover, the join of
k
and
k
is a closed path in D
k
for every k. D
k
being convex, we therefore have, by applying Cauchys theorem for a convex
114 CHAPTER 4. INTEGRATION
region:
k {0, ..., N 1},
_

k
f(z)dz
_

k
f(z)dz = 0.
Hence:
_

f(z)dz =
N1

k=0
_

k
f(z)dz =
N1

k=0
_

k
f(z)dz =
_

f(z)dz = 0.
Example 6. In the following two examples, I =
_
(0,1)
f(z)dz is zero:
(i) f(z) = e
z
2
. The function is holomorphic on C (as a composition of
holomorphic functions), so it is holomorphic on D(0, 1). Cauchys the-
orem for a contour then gives the result. Note that without Cauchys
theorem, there is no way to prove this (in particular with the funda-
mental theorem of calculus).
(ii) f(z) =
e
iz
2
4+z
2
. The zeros of the denominator do not lie in D(0, 1), so f
is holomorphic on D(0, 1), and Cauchys theorem applies.
When the function is not holomorphic inside and on the contour, one has
to rely on the denition of the path integral and use the parametrization of
the path. For example, consider f(z) = (Im(z))
2
and the contour (0, 1).
It is not holomorphic anywhere (the Cauchy-Riemann equations are not
veried anywhere except at z = 0). Then, one has to write (t) = e
it
for
t [0, 2]. By de Moivres formula: f(gamma(t)) = sin(2t) = 2 cos(t) sin(t),
and:
I =
_
(0,1)
f(z)dz =
_
2
0
f((t))ie
it
dt = 0.
4.2. CAUCHYS THEOREM 115
4.2.3 Deformation
Now that we have proven Cauchys theorem for closed contour, we would
like to be able to replace some complicated contours by simpler ones, when
performing calculation. This is the topic of this subsection. For example,
when evaluating integrals, we would like to replace a closed contour by a
circle centred on a I(). These results will be important later.
Theorem 26. Deformation theorem.
(i) Let be a positively oriented contour and a I() such that D(a, r)
I() for a given r > 0. Let f : C C be holomorphic inside and on
, except possibly at a. Then:
_

f(z)dz =
_
(a,r)
f(z)dz.
(ii) Let and be two positively oriented contours such that

lies inside

, i.e.

I( ) I(). Let f : C C be holomorphic inside and


on , except, possibly, at a I ( ). Then:
_

f(z)dz =
_

f(z)dz.
(iii) Let
1
and
2
be two circline paths with commons initial and nal
points. Let =
1
(
2
), and suppose that is simple. Let f : C C
be holomorphic inside and on . Then:
_

1
f(z)dz =
_

2
f(z)dz.
Proof. We prove each point successively.
(i) Let c be the initial point of . Let > 0 such that f H(D(c, )).
Then, I() D(c, ) = . Take d I() D(c, ). Since, according
116 CHAPTER 4. INTEGRATION
Figure 4.5: Proof of the deformation theorem.
to Jordans curve theorem, I() is (polygonally) connected, there is a
polygonal path
1
in I() joining d to a (and this path is simple). Let
the parameter interval of
1
be [, ]. there is a point b = (T) on

1
such that t [0, T[, |
1
(t)a| > r (this means that b is the rst point
at which

1
meets the circle |z a| = r). Then
2
= [c, d]
1
joins
c to b. Now, let be the join: =
2
((a, r)) (
2
). is
not a contour because we trace
2
twice, in both directions. However,
the proof of Cauchys theorem for contours can be generalized to such
paths (exercise). So:
_

f(z)dz = 0 =
_

f(z)dz
_
(a,r)
f(z)dz.
(ii) This point can be proven by choosing some disk D(a, r) I( ) and
by applying the rst point twice to and , to obtain:
_

f(z)dz =
_
(a,r)
f(z)dz =
_

f(z)dz.
4.2. CAUCHYS THEOREM 117
(iii) The last point is a direct result of Cauchys theorem, together with
the decomposition of the integral along a join.
This theorem can give us a generalization of the fundamental integral
_
(a,r)
(z a)
n
dz for n Z.
Proposition 26. Let be a positively oriented contour, and let a C such
that a

. Then:
_

(z a)
n
dz =
_

_
0 if a O()
0 if a I() and n = 1
2i if a I() and n = 1
Proof. For n = 1, the results come from the fundamental theorem of cal-
culus. For n = 1, there is no antiderivative for (z a)
1
, so we cannot
use the fundamental theorem of calculus. However, Cauchys theorem is
applicable when a O(), because then f is holomorphic inside and on
the contour. When a I(), we can use the deformation theorem and the
known result for (a, r) to prove the statement.
Example 7. Consider f(z) = 2/(4z
2
1) and I =
_
(0,1)
f(z)dz. The
function f is holomorphic everywhere except at the two points z = 1/2,
where its denominator cancels. Let us separate these poles and write f(z) =
1/(2z 1) 1/(2z + 1). Then, we can apply the deformation theorem to
each part and write:
I =
_
(1/2,1/4)
1
2(z 1/2)
dz
_
(1/2,1/4)
1
2(z + 1/2)
dz =
1
2
2i
1
2
2i = 0.
118 CHAPTER 4. INTEGRATION
To go further
4.2.4 The complex logarithm... again
When we introduced the complex logarithm, we dened it as the innite set
of solutions to the equation e
w
= z, for z = 0. In order to get a well-dened
logarithm, we have seen how to introduce a cut in the complex plane, by
restricting the argument of z, that denes branches of the logarithm. Later,
we have seen that these branches are holomorphic. We can now examine
this logarithm in more details.
In real analysis, the logarithm is given by ln x =
_
x
1
1
u
du, for x ]0, +[.
We would like to see if there is an analogous relation for the complex loga-
rithm. Let z = |z|e
i
= 0, with ] , ]. This means that we are working
in the plane with a cut along ] , 0], so that the argument of z is uniquely
determined. Let:
F
0
(z) =
_
(z)
1
w
dw,
where (z) is the join of
1
(z) = [1, |z|] and
2
dened by:

2
(z) =
_
|z|e
it
, for t [0, ] if Imz 0
|z|e
i(t)
, for t [, 0] if Imz < 0.
Note that the path does not cross the cut. Suppose that Imz 0.
Then:
F
0
(z) =
_
|z|
0
1
u
du +
_

0
1
|z|e
it
|z|ie
it
dt = ln |z| +i.
Similarly, one can show that F
0
(z) = ln |z| + i for Imz < 0 (Exercise). So,
we have a valid integral formula for the complex logarithm, once we have
4.2. CAUCHYS THEOREM 119
Figure 4.6: Contour for the derivation of the logarithm using an indenite
integral.
restricted ourselves to a branch of the logarithm. In order to obtain this
representation, we used a specic path in the complex plane with a cut.
What would happen for a dierent path in the cut plane, or even in C

?
Let us remember the fundamental integral:
_

1
w
dw =
_
0 if 0 O()
2i if 0 I().
For k Z, let us dene:

k
(z) =
_
_
_
_

k
i=1
(0, 1)
_
(z) if k 0
_

k
i=1
(0, 1)
_
(z) if k < 0.
This means that we turn |k| times around 0 on the circle (0, 1), in the
positive sense if k 0, and in the negative sense if k < 0. Then, by the
120 CHAPTER 4. INTEGRATION
same method as above:
z C

,
_

k
(z)
1
w
dw = F
0
(z) + 2ki.
Note that, if we were replacing (0, 1) by a contour that does not encircle
0, then the factor 2ki would disappear. Actually, this can be generalized:
if (z) is any circline path from 1 to z in C

, then
_
(z)
1
w
dw takes its values
in ln z, as (z) varies.
Theorem 27. Let G be a convex region not containing 0. Then, there exists
a function f = ln
G
H(G) such that z G, e
f(z)
= z, and:
(a, z) G
2
, f(z) f(a) =
_

1
w
dw,
where is any path in G with endpoints a and z. The function f is uniquely
determined up to the addition of an integer multiple of 2i. Hence:
z G, ln
G
z = ln |z| +i(z),
where (z) arg z and z (z) is a continuous function in G.
Proof. By the antiderivative theorem, there exists a function f H(G) such
that: z G, f

(z) = 1/z. Then:


d
dz
_
ze
f(z)
_
= e
f(z)
zf

(z)e
f(z)
= 0.
This implies that ze
f(z)
is constant in G; hence: z = Ce
f(z)
, with C C

.
By a redenition of f, we can always impose C = 1. The integral formula
for f then follows from the indenite integral theorem and the antiderivative
theorem.
Suppose now that we have two functions f and g, holomorphic in G such
that: z G, e
f(z)
= e
g(z)
. Then, f g has zero derivative on G and
is therefore constant: z G, (f g)(z) = K C

. Thus, e
K
= 1, or
equivalently, K = 2ki for k Z.
The last part comes from the construction of the logarithm, and from the
fact that the imaginary part of an holomorphic function is continuous.
4.3. CAUCHYS FORMUL 121
4.2.5 Exercises
Determine whether the following integrals are zero:
(i)
_
(0,1)
e
z
+z
2
z
3
dz;
(ii)
_
(i,1)
sin(z)
(zi/2)
dz;
(iii)
_
(2i,1)
sin(z)
(zi/2)
dz;
(iv)
_

tan(z)dz, with = [0, 1] [1, 1 + i] [1 + i, + i] [ + i, i]


[ i, i] [i, 0].
4.3 Cauchys formul
Now that we have proven Cauchys theorem, we can derive a lot of very
important and powerful results, that all derive from Cauchy formul:
Liouvilles theorem: A function which is holomorphic in C cannot
be bounded, unless it is constant.
Innite dierentiability: Any holomorphic function (i.e. that is
dierentiable once), is actually automatically innitely dierentiable.
Taylors theorem: Any holomorphic function is locally representable
by a power series. This is the contrapositive of the result we have
proven, i.e. that any function that can be written as a power series is
holomorphic on its disc of convergence.
Identity theorem: This is a corollary of Taylors theorem. It states
that if f is holomorphic in a region G and is zero in an open disc in
G, then f is identically zero in G.
122 CHAPTER 4. INTEGRATION
One should be aware that these results are extremely strong and do not have
any analogue in real analysis.
4.3.1 Cauchys integral formula
Cauchys integral formula gives the value of a complex function at a point
a C in terms of a boundary value integral evaluated on a contour encircling
the point a. In order to prove the formula, we will need the deformation
theorem, to replace the given, arbitrary contour, by a small circle around a.
In the following, unless explicitly stated, contours will be positively oriented.
Theorem 28. Cauchys integral formula.
Let f be holomorphic inside and on a positively oriented contour . Then,
if a is inside :
f(a) =
1
2i
_

f(w)
w a
dw.
Proof. Since a I() and I() is open, there exists R > 0 such that
D(a, R) I(). By the deformation theorem, for any r < R, we have:
_

f(w)
w a
dw =
_
(a,r)
f(w)
w a
dw.
Moreover, since f(a) is a constant, we have:
_
(a,r)
f(a)
w a
dw = f(a)
_
(a,r)
1
w a
dw = 2if(a).
Hence:

1
2i
_

f(w)
w a
dw f(a)

1
2i
_
(a,r)
f(w) f(a)
w a
dw

1
2i
_
2
0
f(a +re
i
) f(a)
re
i
ire
i
d

1
2
2 sup
[0,2]
|f(a +re
i
) f(a)|.
The last line is obtained by using the comparison theorem. Finally, since f
is continuous (because holomorphic) at a, the supremum tends to 0 when r
tends to 0. Since the left-hand side is independent of r, it should be zero.
4.3. CAUCHYS FORMUL 123
Example 8. We can apply this formula immediately to a few cases.

_
(3,5)
cos z
z
dz = 2i[cos z]
z=0
= 2i.

_
(i,1)
z
2
z
2
+1
dz =
_
(i,1)
z
2
/(z+i)
zi
dz = 2i[z
2
/(z +i)]
z=i
= .
I =
_
(0,2)
e
iz/2
z
2
1
dz has an integrand with two non-holomorphic points,
1 and 1 that are both in the contour, so, strictly speaking, we cannot
apply Cauchys formula. But, we can decompose it in partial fractions:
I =
1
2
_
(0,2)
e
iz/2
z 1
dz
1
2
_
(0,2)
e
iz/2
z + 1
dz.
And, now, we can apply Cauchys formula to each integral, to obtain:
I = [
1
2
e
iz/2
]
z=1
[
1
2
e
iz/2
]
z=1
= i. The use of partial fraction is
ecient, but could be a bit laborious. We will see later a much more
powerful method to deal with integrals with several non-holomorphic
points.
Theorem 29. Liouvilles theorem.
Let f : C C be holomorphic and bounded in C. Then f is constant.
Proof. Suppose that w C, |f(w)| M. Let (a, b) C
2
and dene
R 2 max{|a|, |b|}, so that |w a| R/2 and |w b| R/2 whenever
|w| = R (this results from |w + z| ||w| |z||). Then, apply Cauchys
formula with = (0, R):
f(a) f(b) =
1
2i
_

f(w)
_
1
w a

1
w b
_
dw
=
a b
2i
_

f(w)
(w a)(w b)
dw.
So, by bounding the integrand:
|f(a) f(b)|
1
2
2RM
|a b|
(R/2)
2
=
4M|a b|
R
.
The right-hand side of the inequality can be made arbitrarily small by taking
R arbitrarily large. So, for any (a, b) C
2
, f(a) = f(b).
124 CHAPTER 4. INTEGRATION
Liouvilles theorem can now be applied to give a remarkable proof of the
fundamental theorem of algebra.
Theorem 30. Fundamental theorem of algebra.
Let p(z) be a non-constant polynomial with complex coecients. Then, there
exists C such that p() = 0. This implies that a complex polynomial
of degree n 1 has n roots (taking into account the multiplicity,
so, not necessarily distinct) in C.
Proof. Let us suppose, for a contradiction, that: z C, p(z) = 0. Since
|p(z)| tends to + when |z| tends to +, there is an R > 0 such that:
|z| > R |1/p(z)| < 1. Moreover, on the compact set D(0, R), 1/p(z)is
continuous (because holomorphic), so it is bounded. Since R can be made
has big as necessary, 1/p(z) is bounded on C. But it is also holomorphic, so
Liouvilles theorem implies that it must be constant. This contradicts the
hypothesis of the theorem. This implies that p(z) must have one root, .
Then, one can factorize : p(z) = (z)p
1
(z), and apply the same argument
to p
1
(z), etc. By induction, one then constructs exactly n roots for p(z) (the
induction stops when one arrives at a constant polynomial, i.e. after exactly
n iterations).
4.3.2 Cauchys formul for derivatives
We have seen that there is a simple relation between the value of an holo-
morphic function at a point and a simple integral along a contour encircling
that point. We are going to show something evening more stunning: there
is such simple relations for any derivative of an holomorphic function. This
tells us that an holomorphic function, for which we have required deriv-
ability, is actually innitely dierentiable! This is very dierent from what
happens in real analysis, where the dierentiability of a function at rst
order does not guarantee the innite dierentiability.
Theorem 31. Cauchys formula for the rst derivative.
Let f : C C be holomorphic inside and in a positively oriented contour .
4.3. CAUCHYS FORMUL 125
Let a I(). Then:
f

(a) =
1
2i
_

f(w)
(w a)
2
dw.
Proof. As in the the proof of Cauchys integral formula, we use the deforma-
tion theorem to replace integrations along by integrations along a circle
of, say (a, 2r) where r > 0 is chosen so that (a, 2r) I() (which is al-
ways possible since I() is open). Then, for h C with |h| < 2r, Cauchys
integral formula applied twice gives:
f(a +h) f(a)
h
=
1
2hi
_
(a,2r)
f(w)
_
1
w a h

1
w a
_
dw
=
1
2i
_
(a,2r)
f(w)
(w a h)(w a)
dw.
Hence:
f(a +h) f(a)
h

1
2i
_
(a,2r)
f(w)
(w a)
2
dw
=
h
2i
_
(a,2r)
f(w)
(w a h)(w a)
2
dw. .
We now have to prove that the right-hand side tends to zero as h tends to
zero. Let us choose h such that |h| < r. Then: w (a, 2r)

, |wa h|
|w a| |h| > r. Moreover, since f is holomorphic inside and on , it is
continuous on (a, 2r)

, and this set is compact. So, there exists a constant


M > 0 such that w (a, 2r)

, |f(w)| M. So, since |w a| = 2r on


(a, 2r)

, we have:

f(a +h) f(a)


h

1
2i
_
(a,2r)
f(w)
(w a)
2
dw

|h|M
2 4r
3
4r =
|h|M
2r
2
,
and the right-hand side tends to zero when h tends to zero.
126 CHAPTER 4. INTEGRATION
But, we are not going to stop there:
Theorem 32. Let f : C C be holomorphic in an open set G C. Then:
f

H(G);
f has derivatives of all orders in G.
Proof. Let a G, and r > 0 such that

D(a, 2r) G. For h C with
|h| < r, Cauchys formula for the rst derivative gives:
f

(a +h) f

(a)
h
=
1
2i
_
(a,2r)
f(w)
_
1
(w a h)
2

1
(w a)
2
_
dw.
By applying the same estimation argument as in the proof of Cauchys
formula for the rst derivative, the integral on the right-hand side can be
shown to converge towards:
2
_
(a,2r)
f(w)
(w a)
3
dw.
This guarantees that f

(a) exists for all a G, so that f

H(G). By
induction, we see that f
(n)
exists and is holomorphic on G for all n N.
This theorem allows one to formulate a partial converse to Cauchys
theorem.
Theorem 33. Moreras theorem.
Let f : C C be continuous on an open set G C, with
_

f(w)dw = 0 for
all triangles G. Then f H(G).
Proof. Let a G and r > 0 such that D(a, r) G. Since D(a, r) is a
convex region, we can use the indenite integral theorem to get a function
F H(D(a, r)) such that F

= f. Then, the previous theorem implies that


f H(D(a, r)). Since a is arbitrary, we have f H(G).
Finally, we can give a formula for any derivative of a holomorphic func-
tion:
4.3. CAUCHYS FORMUL 127
Theorem 34. Cauchys formula for derivatives.
Let f be holomorphic inside and on a positively oriented contour . Let
a I(). Then f
(n)
(a) exists for any n N, and:
a I(), f
(n)
(a) =
n!
2i
_

f(w)
(w a)
n+1
dw.
Figure 4.7: Proof of Cauchys formula for derivatives.
Proof. We will prove the result by induction. It is obviously true for n = 0,
since it give Cauchys integral formula. Assume that the result is true at
order k. By the deformation theorem, we may assume that = (a, 2r)
with r > 0 suitably chosen. Take |h| < r. Since the result is true at order
128 CHAPTER 4. INTEGRATION
k, we have:
f
(k)
(a +h) f
(k)
(a) =
k!
2i
_

f(w)
_
1
(w a h)
k+1

1
(w a)
k+1
_
dw
=
(k + 1)!
2i
_

f(w)
_
[a,a+h]
(w )
k2
ddw.
The last line is obtained by using the fundamental theorem of calculus. We
now have to show that (h) =
f
(k)
(a+h)f
(k)
(a)
h

(k+1)!
2i
_

f(w)
(wa)
k+2
dw tends
to 0 as h tends to 0. We have:
(h) =
(k + 1)!
2ih
_

f(w)
_
[a,a+h]
_
1
(w )
k+2

1
(w a)
k+2
_
ddw
=
(k + 2)!
2ih
_

f(w)
_
[a,]
1
(w )
k+3
dddw.
Again, the last line comes from the fundamental theorem of calculus. Since
f is holomorphic, it is continuous on the compact set

, so it is bounded
by a constant M > 0. For [a, ] and [a, a + h], we have |w | r
for all w

. Also, | a| r as long as |wa| = 2r, and | a| |h| by


construction. Hence:
|(h)|
(k + 2)!
2|h|
M|h|
2
r
k+3
4r |h|.
This shows that |(h)| tends to zero when h tends to zero.
4.3.3 Exercises
Calculate the following integrals:
(i)
_
(0,2)
sin(z)
(zi)
dz;
(ii)
_
(i,1)
cosh(z
2
)
(zi/2)
2
dz;
(iii)
_

e
z
cos(z)
(z1i)
3
dz, with [0, 2] (1, 2);
(iv)
_
(0,1)
z2
(z+2)z
2
dz.
4.4. POWER SERIES REPRESENTATION 129
4.4 Power series representation
We proved earlier that a convergent power series is holomorphic in its disc of
convergence. In this section, we explore further the link between holomorphy
and expansion in power series.
4.4.1 Integration of series
The results of this subsection are mostly technical, and they will be used in
the subsequent manipulations of series and integrals. They could be replaced
by the more sophisticated notion of uniform convergence, but this will not
be needed for the results we want to prove in the following.
Theorem 35. Let be a path. Let U, u
0
, u
1
,... be continuous functions on

, and assume that z

, U(z) =

+
n=0
u
n
(z). Assume that there exist
constants M
k
> 0 for k N such that

M
k
converges and k N, z

, |u
k
(z)| M
k
. Then:
+

k=0
_

u
k
(z)dz =
_

k=0
u
k
(z)dz =
_

U(z)dz.
Proof. For N N, let U
N
=

N
k=0
u
k
(n). Both U
N
and U are continuous,
and hence integrable, on

. We have:

U(z)dz
N

k=0
_

u
k
(z)dz

(U(z) U
N
(z)) dz

sup
z

{|U(z) U
N
(z)|} length()
sup
z

k=N+1
|u
k
(z)| length()

k=N+1
M
k
length().
Since

M
k
converges, M
k
tends to 0 when k tends to innity. So, the
right-hand side tends to zero.
130 CHAPTER 4. INTEGRATION
Theorem 36. Coecients in a power series.
Let f(z) =

+
k=0
c
k
z
k
with a radius of convergence R > 0. Then:
r ]0, R[, n N, c
n
=
1
2i
_
(0,r)
f(z)
z
n+1
dz.
Proof. Provided we can interchange the sum and the integral, we have, for
n N and r [0, R[:
_
(0,r)
f(z)
z
n+1
=
_
(0,r)
_
=

k=0
c
k
z
k
_
z
n1
dz
=
+

k=0
c
k
_
(0,1)
z
kn1
dz
= 2ic
n
.
Hence, we only have to prove that we can interchange integration and sum.
To do that, we need to apply the previous theorem to the case = (0, r)
and u
k
(z) = c
k
z
kn1
. Then, U = z
n1
f(z). U is continuous because f is
continuous inside the disc on convergence. Moreover, on (0, r) D(0, R),
we have |u
k
(z)| = M
k
= |c
k
|r
kn1
, and

M
k
converges (since the series

u
k
(z) converges absolutely inside its disc on convergence). Therefore, all
the hypothesis of the previous theorem apply, and we can safely interchange
integral and sum.
4.4.2 Taylors theorem
We can now use Cauchys integral formula to prove that any function that
is holomorphic in a disc D(0, R) has a power series expansion with radius
of convergence R.
Theorem 37. Taylors theorem Let f H(D(a, R)) for R > 0. Then,
there exists a unique set of constants (c
n
)
nN
such that:
z D(a, R), f(z) =
+

n=0
c
n
(z a)
n
.
4.4. POWER SERIES REPRESENTATION 131
The constants c
n
are given by:
n N, c
n
=
1
2i
_

f(w)
(w a)
n+1
dw =
f
(n)
n!
,
where is any positively oriented contour included in D(a, R) and enclosing
a.
Proof. Let z D(a, r) and r > 0 such that |z a| < r < R. Consider
= (a, r). If this is not the case, one can use the deformation theorem to
recover the circle. By Cauchys integral formula:
f(z) =
1
2i
_

f(w)
w z
dw.
Since w

, |z a| < |w a| = r, we have:
1
w z
=
1
w a
(1 (z a)/(w a))
1
=
1
w a
+

k=0
_
z a
w a
_
k
.
Hence:
f(z) =
1
2i
_

k=0
(z a)
k
(w a)
k+1
f(w)dw.
On

that is compact, f is bounded (because it is continuous), so there


exists an M > 0 such that:

(z a)
k
f(w)
(w a)
k+1

M
k
=
M
r
|z a|
k
r
k
.
Since |z a|/r < 1, the series

M
k
converges. Hence, once again we can
legitimately commute sum and integral, to obtain:
f(z) =
+

n=0
_
1
2i
_

f(w)
(w a)
n+1
dw
_
(z a)
n
.
Cauchys formula for derivatives gives then the desired answer. The unique-
ness of the decomposition is a result of the theorem on coecients of power
series.
132 CHAPTER 4. INTEGRATION
Example 9. Consider f(z) = z
5
sin(2z). Using the series expansion for the
sinus, sin(2z) =

+
n=0
(1)
n
(2z)
2n+1
(2n+1)!
we obtain:
f(z) =
+

n=0
(1)
n
2
2n+1
z
2(n+3)
(2n + 1)!
By uniqueness of the series expansion, this is the Taylor series. Then we
have the derivatives at all orders of the functions f. Note that it would have
been dicult to obtain these derivatives by a direct calculation..
Let see what happens in the case of a holomorphic branch of the loga-
rithm. Let us cut the plane along ] , 0] so that ] , ]. We let f be
the holomorphic branch of the logarithm in C

= C\] , 0], given by:


z = |z|e
i
= 0, ] , ], f(z) = ln |z| +i.
Since f H(D(1, 1)), f must have a Taylor expansion in the disc D(1, 1):
f(z) =

+
n=0
c
n
(z 1)
n
. First, note that:
1
z
=
1
1 + (z 1)
=
+

n=0
(1)
n
(z 1)
n
for |z 1| < 1.
Then:
1
z
=
d
dz
ln z =
+

n=1
nc
n
(z 1)
n1
for |z 1| < 1,
thanks to the theorem on the dierentiation of power series. By uniqueness
of the series expansion, we have:
c
n
=
(1)
n1
n
for n 1.
The value of c
0
depends on the branch that is considered. Here, we have
f(1) = 0, so, necessarily, c
0
= 0. Therefore:
z D(1, 1), f(z) =
+

n=1
(1)
n1
n
(z 1)
n
.
4.4. POWER SERIES REPRESENTATION 133
Consider now f H(D(0, R)) for some R > 0. Then, for n N, we have
the coecients of the Taylor expansion of f, for 0 < r < R:
c
n
=
1
2i
_
(0,r)
f(w)
w
n+1
dw,
and:
|c
n
| =

1
2i
_
(0,r)
f(w)
w
n+1
dw

1
2
sup
|z|=r
{|f(z)z
n1
|} length((0, r))

1
2
M(r)r
n1
2r
r
n
M(r),
where M(r) = sup{|f(z)|, |z| = r}. This leads to the following interesting
result:
Theorem 38. Let f be holomorphic in C, with Taylor expansion f(z) =

+
n=0
c
n
z
n
valid for all z C. Suppose that there exist two positive constants
M and K and k N

such that:
z C, |z| K, |f(z)| M|z|
k
.
Then, f is a polynomial of degree at most k.
Proof. Let r K, and note that M(r) Mr
k
. Then, using the estimation
above, we obtain:
n N, |c
n
| Mr
kn
.
Since r can be chosen arbitrary large (f is holomorphic on C), then we must
have c
n
= 0 for all n > k. This means that the Taylor expansion is truncated
at least at order k, so f has to be polynomial of degree smaller or equal to
k.
Remark 7. We can sum up the various ideas we have developed about power
series and complex functions.
134 CHAPTER 4. INTEGRATION
We showed that every power series around a C with a radius of
convergence R > 0 denes a holomorphic function in D(a, r).
We showed here that every function that is holomorphic in an open
set G is locally (in G) representable by a power series. We will say
that it is analytic in G.
We showed that any holomorphic function with a non vanishing deriva-
tive is conformal.
We showed that any regular (that is, with continuous partial deriva-
tives) conformal mapping is holomorphic.
So, keeping in mind the restrictions that make the results precise, we could
state that:
ANALYTIC = HOLOMORPHIC = CONFORMAL.
4.4.3 Multiplication of power series
We will use Taylors theorem to prove an important result about the multi-
plication of power series.
Theorem 39. Suppose that:
f(z) =
+

n=0
a
n
z
n
and g(z) =
+

n=0
b
n
z
n
are complex series with radii of convergence R
1
and R
2
, respectively. Let
h(z) =

+
n=0
c
n
z
n
, with n N, c
n
=

n
k=0
a
k
b
nk
. Then,

+
n=0
c
n
z
n
has
a radius of convergence at least R = min{R
1
, R
2
}, and z D(0, R), h(z) =
f(z)g(z).
Proof. In D(0, R), both f and g are holomorphic, and we have a
n
= f
(n)
(0)/n!
and b
n
= g
(n)
(0)/n!. The product fg is also holomorphic in D(0, R), as
4.4. POWER SERIES REPRESENTATION 135
product of holomorphic functions, and it is represented by the Taylor series:
f(z)g(z) =
+

n=0
c
n
z
n
,
with:
n!c
n
= (fg)
(n)
(0) =
n

k=0
n!
k!(n k)!
f
(k)
(0)g
(nk)
(0) = n!
n

k=0
a
k
b
nk
.
We used Leibniz rule to derive n times a product. The rule can easily be
checked by induction.
Example 10. The nth Hermite function H
n
is dened, for t R by:
n N, H
n
(t) = (1)
n
e
1
2
t
2
_
d
dt
_
n
e
t
2
.
We have:
(x, t) R
2
,
+

n=0
H
n
(t)
x
n
n!
= e

1
2
t
2
+2xtx
2
.
So that we have a generating function whose derivatives give the Hermite
functions: H
n
(t) =
_
d
n
dx
n
e

1
2
t
2
+2xtx
2
_
|x=0
.
To prove this, note that e

1
2
t
2
+2xtx
2
= e
t
2
/2
e
(xt)
2
. By composition of
holomorphic functions, we have that e
z
2
is holomorphic in C. Hence, it has
a Taylor expansion for any a C:
e
z
2
=
+

n=0
_
d
n
dx
n
e
z
2
_
|z=a
(z a)
n
n!
.
By choosing a = t and z = x t, the required formula follows.
4.4.4 Exercises
1. Find the Taylor expansions of the following function, at the given point
a C:
(i) f(z) =
sin(z)
z
, at a = 0;
136 CHAPTER 4. INTEGRATION
(ii) f(z) = (z i)
2
cos(z i), at a = i;
(iii) f(z) = z
3
sinh(z), at a = 0.
2. Give the value of all the successive derivatives of the following functions
at a = 0:
(i) f(z) = z
5
cosh(z);
(ii) f(z) = (z
3
+ 2z
2
+iz + 1) sin(z);
(iii) f(z) = sin(z) sinh(z).
4.5 Zeros and singularities
This chapter is concerned with two important sets of points in the domain
of an holomorphic function.
The rst set is the set of the zeros of the function. If f is holomorphic in an
open set G C, this set is Z(f) = {z G, f(z) = 0}. It is important for
two reasons. The rst one has to do with the second set: if f is zero at a
point a C, then 1/f has a singularity at that point. Integrating a function
along a contour inside which it has a singularity will, in general, give a non-
zero result (remember Cauchys formul), so locating and characterizing the
zeros are important for integration. The second reason is that we will prove
a very strong result, as a consequence of Taylors theorem: if a function is
holomorphic in a region of C, then it cannot be zero in this region except at
isolated points, unless it is identically zero in the region.
The second set we mentioned in the set of points at which an otherwise
holomorphic function fails to be holomorphic, i.e. the set of singularities
of the function. We have highlighted previously its link with the set of
zeros of the denominator of the function. It turns out that the behaviour of
the function at its singularities can be studied in details if one replaces the
Taylor expansion by the Laurent expansion of the function at the singular
point.
4.5. ZEROS AND SINGULARITIES 137
4.5.1 Characterizing zeros
Denition 31. Let f : S C C be holomorphic at a S. Then the
point a is said to be a zero of f i f(a) = 0. a is said to be a zero of order
m N i:
0 = f(a) = f

(a) = ... = f
(m1)
(a) and f
(m)
(a) = 0.
Zeros of order 1 and 2 are usually called simple and double, respectively.
Note that, by convention, a zero is of order 0 if f is holomorphic at a and
f(a) = 0, i.e. if a is not a zero of f.
Remember that f is holomorphic at a i f H(D(a, r)) for some r > 0.
Example 11. f(z) = (z a)
m
has a zero of order m at a.
f(z) = sin(z) has innitely many zeros at z = k, for k Z. At these
points, (sin z)

= cos z is non-zero, so, the zeros of sin z are all simple.


This is also true of all the zeros of cos z, sinh z and cosh z.
The next theorem proves to be very useful in characterizing zeros.
Theorem 40. Characterization theorem for zeros of order m.
Let f H(D(a, r)). Suppose that its Taylor expansion in D(a, r) is f(z) =

+
n=0
c
n
(z a)
n
. Then, the following propositions are equivalent:
(i) m N, 0 = f(a) = f

(a) = ... = f
(m1)
(a) and f
(m)
= 0;
(ii) z D(a, r), f(z) =

+
n=m
c
n
(z a)
n
with c
m
= 0;
(iii) g H(D(a, r)), g(a) = 0, z D(a, r), f(z) = (z a)
m
g(z);
(iv) C C

, lim
za
(z a)
m
f(z) = C.
Proof. The rst two points are equivalent by virtue of Taylors theorem.
Now, assume (ii) and dene in D(a, r):
g(z) =
+

n=m
c
n
(z a)
nm
=
+

k=0
c
m+k
(z a)
k
.
138 CHAPTER 4. INTEGRATION
The power series

+
n=m
c
n
(z a)
n
is convergent (sub-series of a convergent
series), so g H(D(a, r)). Moreover, g(a) = c
m
= 0. This means that (iii)
holds. Conversely, if (iii) holds, then (ii) follows immediately from the fact
that g is holomorphic, so that it admits a Taylor expansion.
The continuity of g makes (iv) follow from (iii) immediately.
To prove the converse, suppose (iv). Then, by denition, for > 0 arbitrary,
there exists > 0 such that:
|w a| <

(w a)
m
f(w) C

< .
Take, < min{, r}. Then:
|w a| =

m
f(w)

|C|

m
f(w) C

<

m
f(w)

|C|
|f(w)| (|C| +)
m
. (4.3)
Remember the estimate for the coecient c
n
in the Taylor expansion given
previously:
|c
n
|
n
sup{|f(z)|, |z| = } for any 0 < < r.
So, we have:
|c
n
| (|C| +)
mn
.
If, m < n, then
mn
can be made arbitrarily small by taking suciently
small, so c
n
must be zero (it is independent of ). We thus have f(z) =

+
n=m
c
n
(z a)
n
. Moreover, the inequalities above show that |c
m
C| < ,
so c
m
= C = 0. Thus we have (ii), and since (ii) implies (iii), (iv) implies
(iii).
Proposition 27. Consider two functions f and g that are holomorphic at
a C and for which a is a zero of order m 0 and n 0 respectively.
Then, fg is holomorphic at a, and a is zero of fg of order m+n.
4.5. ZEROS AND SINGULARITIES 139
Proof. The fact that a is a zero of fg is evident. fg is holomorphic at a as
product of two holomorphic functions at a. Now, consider:
lim
za
(z a)
m+n
(fg)(z) = lim
za
[(z a)
m
f(z)(z a)
n
g(z)]
= lim
za
(z a)
m
f(z) lim
za
(z a)
n
g(z).
(4.4)
Since both f and g are holomorphic at a with a zero of order m and n
respectively, there are two non-zero constants C and D such that lim
za
(z
a)
m
f(z) = C, and lim
za
(z a)
n
g(z) = D, according to point (4) of the
previous theorem. Hence, fg has a zero of order m+n at a.
Example 12. The previous proposition makes it easy to determine the
order of the zeros of a function that can be decomposed into a product of
other functions. For example, f(z) = z
2
sin
4
(z) has a zero of order 2+4 = 6
at z = 0 and zeros of order 0 + 4 = 4 at z = k for k Z

.
4.5.2 Identity and Uniqueness theorems
In the previous subsection, we concentrated on the behaviour of holomorphic
functions around one of their zeros. We will now deal with the set Z(f) itself
for a function f dened on a region G of the complex plane. Remember that
a region is a non-empty open connected set of the complex plane.
We rst need to summarize what we know about limit points. The denition
was given in the rst chapter: if S C is a set, a C is a limit point of S
i:
> 0, D

(a, ) S = .
Proposition 28. Let G be an open set such that S G. In each of the
following cases, we construct limit point of S.
(i) If S = {z
n
, n N}, with (z
n
)
nN
a converging sequence such that
lim
n+
z
n
= z G, then z is a limit point of S.
140 CHAPTER 4. INTEGRATION
(ii) If S = [a, b] with a = b, then every point of S is a limit point of S.
(iii) If S = D(a, r) with a C and r > 0, then, D(a, r) G is the set of all
the limit points of S. Note that it contains S itself.
Proof. (i) S = {z
n
, n N}. z is the limit of the sequence (z
n
)
nN
, so, for
any > 0, there is N N such that, if n N, then, |z
n
z| < ;
in other words, z
n
D

(z, ). Hence, since z


n
S by construction,
S D

(z, ) = for any > 0. This proves that z is a limit point of S.


(ii) S = [a, b] with a = b. Then, take c [a, b]. It is obvious that, for any
> 0, D

(c, ) intersect [a, b].


(iii) S = D(a, r) with a C and r > 0. Let z D(a, r). Then, either
z S, or z (a, r). If z S, since S is open, we clearly have
D

(z, ) S = for any > 0. If z (a, r) this is also the case. So,
D(a, r) L where Lis the set of limit points of S. Now, let us take
z L. Then, if z D(a, r), it should be outside the circle (a, r),
so |z a| > r. But, in that case, one can nd an such that D(z, )
does not contain any element of S: just take < = |z a| r. So
L D(a, r); hence L = D(a, r).
Theorem 41. Identity theorem for a disc.
Let f H(D(a, r)) such that f(a) = 0. Then, either:
(i) f is identically zero in D(a, r), or:
(ii) the zero of f at a is isolated, i.e., there exists > 0 such that D

(a, )
Z(f) = .
Consequently, if a is a limit point of Z(f), then f 0 in D(a, r).
Proof. In D(a, r), we can apply Taylors theorem to f and write:
z D(a, r), f(z) =
+

n=0
c
n
(z a)
2
.
4.5. ZEROS AND SINGULARITIES 141
Two things can happen. Either n N, c
n
= 0, in which case, (i) holds. Or,
there exists a smallest m such that c
m
= 0, and we can write:
f(z) = (z a)
m
g(z) with g(z) =
+

k=0
c
k+m
(z a)
k
.
The series dening g has a radius of convergence at least equal to r (it
is a sub-series of the one dening f). Hence, g is continuous on D(a, r)
(any power series is continuous into its disc of convergence). But, since
g(a) = c
m
= 0, the continuity implies that there exists > 0 such that
g(z) = 0 for any z D

(a, ). Thus, we also have: z D(a, ), f(z) = 0,


hence a is an isolated zero of f.
Remark 8. Consider the function:
f(z) =
_
1 if z D(2, 1)
0 if z D(2, 1).
This function is not identically zero in G = D(2, 1) D(2, 1), even
though every point of D(2, 1) is a limit point of Z(f).
This should emphasize that the previous theorem cannot be generalized to
any open set G of C. In the present case, the problem is that G is not
connected. We will see in the next theorem that it is the only condition to
be added for the identity theorem to be true in general.
Theorem 42. Identity Theorem.
Let G be a region. Let f H(G). Assume that Z(f), the set of zeros of f
has a limit point in G. Then, f is identically zero in G.
In particular, if f 0 on some open disc D(a, r) G for r > 0, then f is
identically zero in G.
Proof. Let L be the set of the limit points of Z(f) in G. The idea is to prove
that L Z(f), and that L and G\L are both open. Since G is a region, it is
connected, so it cannot be expressed as the union of two disjoint non-empty
sets. Since L is not empty by hypotheses, this means that G\L = , i.e.,
142 CHAPTER 4. INTEGRATION
G = L. Since L Z(f), that means that G Z(f): f 0 on G.
To prove that L Z(f), let a L. Then, for any > 0, D

(a, ) contains
at least a point of Z(f). Choose the sequence 1/n, then, any disc D(a, 1/n)
contains a point, say a
n
such that a
n
Z(f), i.e. n N, f(a
n
) = 0. Now,
when n tends to innity, D(a, 1/n) tends to {a}, so, a
n
tends to a, and, by
continuity of f, f(a
n
) tends to f(a). So, f(a) = 0,i.e. a Z(f). Hence,
L Z(f).
To prove that L and G\L are both open, consider a L. Then, the identity
theorem for a disc implies that f 0 in D(a, r) for some r > 0 such that
D(a, r) G (so that f is holomorphic on the disc). But then, D(a, r) L,
so L is open. Now, take a G\L. Since a is not a limit point of Z(f), there
exists D

(a, r) for some r > 0 such that z D

(a, r), f(z) = 0: no point of


D(a, r) is a limit point of Z(f), hence D(a, r) G\L. This shows that G\L
is open.
The identity theorem leads to a useful corollary:
Corollary 2. Uniqueness theorem.
Let G C be a region. Let f H(G) and g H(G) such that z
S, f(z) = g(z), where S has a limit point in G. Then, f = g in G.
Example 13. We will present some applications of the previous theorems
Let f H(D(0, 1)) such that z ]0, 1[, f(z) = 0. Every point of ]0, 1[
is a limit point of Z(f), and ]0, 1[ D(0, 1), so f 0 on D(0, 1), by
the identity theorem.
Let f H(C) such that n N

, f(1/n) = sin(1/n). Then, 0 is a


limit point of S = {1/n, n N

} and f(z) = sin(z) on S. By the


uniqueness theorem, f(z) = sin(z) in C.
Be careful that the limit point of Z(f) always have to be in the domain
of holomorphy of the function. To illustrate this, consider f H(C

)
such that n N

, f(1/(n)) = sin(1/(n)). It does not follow that


4.5. ZEROS AND SINGULARITIES 143
f(z) = sin(1/z) in C

. Indeed, f 0 also satises the conditions. The


problem here is that the limit point of Z(g) with g(z) = f(z)sin(1/z)
is 0, that does not lie in C

, the region of holomorphy of g.


We can also use the uniqueness theorem to prove some contradiction.
For example, we want to prove that there is no f H(D(0, 1)) such
that f(x) = |x|
3
for x ] 1, 1[. Suppose, for a contradiction that such
a f exists. D(0, 1) is a region and 0 is a limit point in D(0, 1) for both
[0, 1[ and ] 1, 0]. Then, f(z) = z
3
on the segment [0, 1[ in D(0, 1), so
that the uniqueness theorem gives f(z) = z
3
on the whole of D(0, 1).
In the same way, f(z) = z
3
on ] 1, 0] in D(0, 1), so that f(z) = z
3
on D(0, 1). This is the contradiction we were looking for.
Using the uniqueness theorem, we can extend some functional relations
on bigger domains than the ones on which they are originally dened. For
example, note that the identity cos
2
z + sin
2
z = 1 for z C can be derived
directly from the denitions of the cosine and sine functions (try), but we
can also notice that cos
2
z + sin
2
z 1 is holomorphic in C and identically
zero on the real axis, whose points are obviously all limit points of R in C,
so that the identity has to hold on the domain of holomorphy C. Another
example is the binomial expansion: (1 + z)
n
=

+
k=0
n!
k!(nk)!
z
k
for n Z

.
It is true for z R such that |z| < 1. Moreover, the series converges in
D(0, 1) so it is holomorphic on D(0, 1), and (1 +z)
n
is holomorphic for any
z C\{1}. Hence, the identity theorem gives the equality in D(0, 1).
To go further
As a nal application of the uniqueness theorem, we shall mention a tech-
nique called analytic continuation. It consists in extending a function
f H(G), where G C is a region, to a function g H(G

), where G

C
is a region such that G G

. Moreover, if such a g exists, it is unique.


144 CHAPTER 4. INTEGRATION
We can illustrate the idea of analytic continuation on a simple example. Let
z D(0, 1), f(z) =
+

n=0
z
n
and z C\{1}, g(z) =
1
1 z
.
We have f H(D(0, 1)) and g H(C\{1}). Clearly, D(0, 1) C\{1}, and
f = g on D(0, 1). g is said to be the analytic continuation of f on C\{1}.
Usually, it is hard to say whether a given holomorphic function can be
extended, but we can try to nd a natural way to do it. Let a C.. Let
f(z) =

+
n=0
c
n
(z a)
n
, dened on D(a, R) for some R > 0. The Taylor
expansion of f around b D

(a, R) is unique and is given by:


z D(b, R |b a|),

f(z) =
+

n=0
f
(n)
(b)
n!
(z b)
n
.
We have restricted the expansion to D(b, R |b a|) since it is the largest
open disk centred on b and contained in D(a, R), where the function is
holomorphic. But, it can happen that the series dening

f actually converges
on a disc D larger than D(b, R |b a|). If this is the case, we can extend
f to g H(D(a, r) D) by dening:
g(z) =
_
f(z) if z D(a, R)

f(z) if z D.
The process can then be repeated by nding another disc overlapping
with D, etc, so that we nd a chain of overlapping discs.
4.5.3 Counting zeros
In this subsection, we will present a result that links the number of zeros of
a given function f to integrals of functionals of f along a contour encircling
the zeros. Then, we will see a consequence of this link, Rouches theorem,
4.5. ZEROS AND SINGULARITIES 145
that allows to compare the number of zeros of two comparable functions.
We will use this result to give another proof of the fundamental theorem of
algebra.
Theorem 43. Let f : S C C be holomorphic inside and on a positively
oriented contour , with

S. Suppose that f is non-zero on and have


N N zeros inside . Then:
1
2i
_

(z)
f(z)
dz = N.
Note that a zero of order m N is counted m times.
Proof. The function f

/f is holomorphic inside and on , except where f is


zero inside . Let us call the zeros of f inside (a
n
)
n{1,...,N}
and suppose
that they are of orders (M
i
)
i{1,...,N}
, respectively. Then, we know that
there exist r
k
> 0 for k {1, ..., N}, such that D(a
k
, r
k
) for k {1, ..., N}
are disjoint open discs contained in the inside of and on which we can nd
functions g
k
that are holomorphic and non-zero such that:
k {1, ..., N}, z D(a
k
, r
k
), f(z) = (z a
k
)
m
k
g
k
(z).
Then, we have:
k {1, ..., N}, z D

(a
k
, r
k
),
f

(z)
f(z)
=
g

k
(z)
g
k
(z)
+
m
k
z a
k
.
Then, dene:
F(z) =
_
_
_
f

(z)
f(z)

N
i=1
m
i
za
i
for z

N
i=1
D(a
i
, r
i
)
g

k
(z)
g
k
(z)

1iN,i=k
m
i
za
i
for z D(a
k
, r
k
) and k {1, ..., N}.
F is holomorphic on and inside , so that Cauchys theorem gives:
_

F(z)dz =
0, i.e.:
_

(z)
f(z)
=
N

k=1
m
k
_

1
z a
k
dz = m
k
2iN.
146 CHAPTER 4. INTEGRATION
Theorem 44. Rouches theorem.
Let f and g be holomorphic inside and on a contour , such that z

, |f(z)| > |g(z)|. Then, f and f +g have the same number of zeros inside
(a zero is counted as many times as its order).
Proof. Let t [0, 1]. z

, |f(z)| > |g(z)| |f(z)+tg(z)| > ||f(z)| |g(z)|| =


|f(z)| |g(z)| > 0, hence, z

, f(z) + tg(z) = 0. Assume, without loss


of generality, that is positively oriented. Dene:
t [0, 1], (t) =
1
2i
_

(f

+tg

)(z)
(f +tg)(z)
dz.
The preceding theorem states that (t) is the number of zeros of f + tg
inside . Hence, is integer-valued, and , if it is continuous, it must be
constant. In this case, (0) = (1), i.e. the number of zeros of f is the same
as the number of zeros of f +g. Let us then prove that is continuous.
Let (t, s) [0, 1]
2
. We have:
(t) (s) =
t s
2i
_

(g

f fg

)(z)
(f +tg)(z)(f +sg)(z)
dz.
Moreover, g and g

f f

g are continuous (because holomorphic) on

that is
a compact set, so they are bounded. We can then take the biggest of the two
bounds, call it M R, and we have: z

, |g(z)| M, |(g

f f

g)(z)|
M. Also, since |f + tg| is strictly positive on

, we can nd m > 0 such


that, z

, |(f +tg)(z)| m. Then:


|(f +sg)(z)| |(f +tg)(z)| |s t||g(z)|
1
2
m if |s t|
m
2M
.
(Note that |st| = |t s| t +s, so that |st|g(z) t|g(z)| s|g(z)| >
|f(z)| s|g(z)| > 0). This means that, for |s t| suciently small,
|(t) (s)|
M|t s|
m
length().
That implies that is continuous.
4.5. ZEROS AND SINGULARITIES 147
As an application of Rouches theorem, consider f(z) = 2 + z
2
and
g(z) = e
z
, and take to be the contour made by joining the semi-circle
between R and R and the line segment [R, R], with R >

3. Note that
f has only one zero in the upper half-plane: i

2. On the line segment, we


have:
|f(z)| 2 > 1 = |g(z)|,
and on the semi-circle:
|f(z)| R
2
2 > 1 |g(z)|.
Then, Rouche theorem tells us that f(z) +g(z) = 2 +z
2
e
iz
has the same
number of zeros in {z C, Im(z) > 0, |z| < R}, R >

3 as f(z) = 2 + z
2
,
i.e., just one zero. Now, R can be made arbitrary large, so f +g has exactly
one zero in the upper half-plane.
We can nally prove a series of very important results. First, we can
start by a new proof of the fundamental theorem of algebra.
Theorem 45. Fundamental theorem of algebra.
A complex polynomial of degree n 1 has n roots (taking into account the
multiplicity, so, not necessarily distinct) in C.
Proof. Consider g(z) = a
1
z
n1
+... +a
n
for n N

and k {1, ..., n}, a


k

C. Consider also f(z) = a
0
z
n
for a
0
C

. f has exactly one zero of


order n in C: 0. Moreover, f and g are both holomorphic on C. Now,
consider z C such that |z| > M = max
_
1,
1
|a
0
|

n
i=1
|a
i
|
_
, so that |f(z)| >

n
k=1
|a
k
z
nk
| |g(z)| for any z {z C, |z| > M}. Hence, by choosing
R > M, we can use Rouches theorem applied to (0, R) and conclude that
a
0
z
n
and (f + g)(z) = a
0
z
n
+ a
1
z
n1
+ ... + a
n
have the same number of
zeros in D(0, R), i.e. exactly n. By choosing R arbitrarily large, we see that
this is true in C.
The next three results are generically known as the maximum modulus
theorem.
148 CHAPTER 4. INTEGRATION
Theorem 46. Local maximum modulus theorem Let f H(D(a, R))
for a C and R > 0. Suppose that z D(a, R), |f(z)| |f(a)|. Then f is
constant.
Proof. Let 0 < r < R. By Cauchys integral formula:
f(a) =
1
2i
_
a,r
f(z)
z a
dz
=
1
2i
_
2
0
f(a +re
i
)rie
i
re
i
d
=
1
2
_
2
0
f(a +re
i
)d.
Thus, we have:
|f(a)|
1
2
_
2
0
|f(a +re
i
)|d |f(a)|,
the last inequality coming from the hypothesis that |f(z)| |f(a)| for all
z D(a, R). Therefore:
_
2
0
_
|f(a)| |f(a +re
i
)|
_
= 0.
The integrand is continuous and non-negative, so it must be zero identically.
Since this is true for any 0 < r < R, it follows that |f| is constant. This, in
turn, implies that f is constant.
Theorem 47. Maximum modulus theorem.
Let G C be a bounded region. Let f H(G) be continuous and non-
constant on G. Then, |f| attains its maximum on the boundary G = G\G.
Proof. Since G is bounded and closed, it is compact, so, on G, the continu-
ous function |f| is bounded and attains its supremum M at some point of
G.
Assume that M is not attained on G. Then, a G, |f(a)| = M. G
being open, there is r > 0 such that D(a, r) G. Then, the local maxi-
mum theorem implies that f is constant on D(a, R). The identity theorem
4.5. ZEROS AND SINGULARITIES 149
then implies that f is constant in G. By continuity, f is constant in G, in
contradiction with the hypothesis.
Finally, we have the following corollary of the maximum modulus theo-
rem, usually useful in applying the theorem:
Lemma 4. Schwarz lemma.
Let f H(D(0, R)) for R > 0. Suppose that f(0) = 0 and that M >
0, z D(0, R), |f(z)| M. Then:
z D(0, R), |f(z)|
M
R
|z|.
If equality occurs for some z with |z| < R, then there exists R such that
z D(0, R), f(z) = Mze
i
/R.
Proof. Since f(0) = 0, there is g H(D(0, R)) such that z D(0, R), zg(z) =
f(z). On |z| = r < R, this gives:
|g(z)| |f(z)|r M/r.
Applying the maximum modulus theorem to g in G = D(0, r), we get
|g(z)| M/r for |z| r. Now, if r tends to R, we get |g(z)| M/R
for |z| < R. For any z = 0, we then have the required bound on f. In z = 0,
f(0) = 0, so the inequality is also satised.
For the last point, consider the function f(z)/z for z = 0. We have shown
that |f(z)/z| M/R for |z| < R, and we now suppose that it attains its
maximum at |z

| < R, i.e. in the bounded region D(0, R). According to


the proof of the maximum modulus theorem, that means that f(z)/z is con-
stant in D(0, R), with a modulus: |f(z)/z| = M/R, which means that there
is R such that f(z) = Mz
e
i
R
, because a factor e
i
does not alter the
modulus.
4.5.4 Laurents theorem
Now that we have studied the zeros of holomorphic functions in detail, we
see that they actually carry a lot more properties than we suspected. In
150 CHAPTER 4. INTEGRATION
the remainder of this section, we will study another set of important points:
the ones at which a function fails to be holomorphic. We will see in the
next section that they are also very important in the study of properties of
holomorphic functions.
First, we have to introduce a new kind of series expansions that constitute
a good alternative to Taylors expansion for functions holomorphic except
at isolated singular points.
Let us rst remember what we know about binomial expansions. For
z C such that |z| < 1, we can expand 1/(1 z) in positive powers of z:
z D(0, 1),
1
1 z
=
+

n=0
z
n
.
Outside D(0, 1), the series on the right-hand side diverges, but we know
that: z C\D(0, 1) 1/z D(0, 1), so that we can write:
z C\D(0, 1),
1
1 z
=
1
z
1
1 (1/z)
=
+

n=0
z
n1
=
1

k=
z
k
,
where we just wrote k = n1. Using the same trick, we can write (az)
1
as a series in positive power of z on D(0, a) and as a series in negative powers
on C\D(0, a). For instance, note that in the annulus {z C, 1 < |z| < 3},
we have:
4
(1 z)(z + 3)
=
1
1 z
+
1
z + 3
=
1

n=
z
n
+
+

n=0
(1)
n
3
n1
z
n
.
This naturally leads us towards the study of double-ended series. By
denition:
Denition 32. Double-ended series.
A double ended series

+
n=
a
n
converges to s = s
1
+ s
2
C i

+
n=0
a
n
converges to s
1
C and

+
n=1
a
n
converges to s
2
C.
Of course, if f is holomorphic on D

(a, r) for r > 0, but has a problem


at a, then, the standard power series expansion

+
n=0
c
n
(z a)
n
cannot
4.5. ZEROS AND SINGULARITIES 151
represented correctly f at a, since the series is well-dened and behaves
regularly at z = a. Nevertheless, the example of the binomial expansion
leads us to think that double-ended power series could be the expansion we
are looking for. For f H(D

(a, r)), we want to show that f(z) can be


expanded as a Laurent series:
z D

(a, r), f(z) =


+

n=
c
n
(z a)
n
.
Actually, we will prove that a function that is holomorphic in an annulus
has an expansion like that.
Theorem 48. Laurents theorem.
For (R, S) [0, +], R < S, let A = {z C, R < |z a| < S} be an
annulus. Let f H(A). Then:
z A, f(z) =
+

n=
c
n
(z a)
n
,
where:
r ]R, S[, n Z, c
n
=
1
2i
_
(a,r)
f(w)
(w a)
n+1
dw.
Proof. By a simple translation of the origin in the complex plane, if neces-
sary, we can let a = 0. Let z A, and choose P R and Q R such that
R < P < |z| < Q < S.
Let and be as shown on Fig. 4.8. Then:
f(z) =
1
2i
_

f(w)
w z
dw,
and:
0 =
1
2i
_

f(w)
w z
dw.
152 CHAPTER 4. INTEGRATION
h
Figure 4.8: Paths and for the proof of Laurents theorem.
Adding these two numbers, and noting that the integrals along the line
segments cancel, we have:
f(z) =
1
2i
_
(0,Q)
f(w)
w z
dw
1
2i
_
(0,P)
f(w)
w z
dw
=
1
2i
_
(0,Q)
+

n=0
z
n
w
n+1
f(w)dw
1
2i
_
(0,P)
+

m=0

w
m
z
m+1
f(w)dw.
The last line was obtained by applying the appropriate binomial expansion:
w (0, Q)

, |z/w| < 1 and w (0, P)

, |w/z| < 1. The functions


involved are all continuous on the paths, so we can interchange the sums
and the integrals, and we nd:
f(z) =
+

n=0
_
1
2i
_
(0,Q)
f(w)
w
n+1
dw
_
z
n
+
+

m=0
_
1
2i
_
(0,P)
f(w)w
m
dw
_
z
m1
.
We can re-index the second sum by posing n = m1. Finally, we invoke
the deformation theorem to replace (0, Q) and (0, P) by (0, R).
4.5. ZEROS AND SINGULARITIES 153
Proposition 29. The Laurent expansion of a function f H(A) where A
is an annulus is unique.
Proof. Again, suppose that a = 0. Suppose that:
z A, f(z) =
+

n=
d
n
z
n
.
Choose r such that R < r < S. Then:
2ic
n
=
_
(0,r)
f(w)w
n1
dw
=
_

0,r
+

k=
d
k
w
kn1
dw
=
_
(0,r)
+

k=0
d
k
w
kn1
dw +
_
(0,r)
+

m=1
d
m
w
mn1
dw.
If we interchange summations and integrations:
2ic
n
=
+

k=
d
k
_
(0,r)
w
kn1
= 2id
n
.
The uniqueness of Taylor and Laurent expansion allows us to relate them
when the rst one exits. Suppose that f H(D(a, r)). Then, it has a Tay-
lor expansion there, and a Laurent expansion in D

(a, r), The uniqueness


of Laurent coecients forces these expansions to coincide on D

(a, r), i.e.


n Z

, c
n
= 0.
As we did for Taylor coecients, we can estimate Laurent coecients.
Suppose that f is holomorphic in A = {z C, R < |z| < S}. Let f have the
Laurent expansion f(z) =

+
n=0
c
n
z
n
in A, with c
n
=
_
(0,r)
f(w)w
n1
dw
for r ]R, S[. Then, we have:
n Z, |c
n
| r
n
sup{|f(z)|, |z| = r}.
Two cases are worth mentioning:
154 CHAPTER 4. INTEGRATION
Let f be holomorphic for |z| > R and suppose that z C, |z| >
R, |f(z)| M, with M R
+
. Then, |c
n
| Mr
n
for all r > R.
This forces c
n
= 0 for all n > 0, so that the Laurent expansion of f is

0
n=
c
n
z
n
.
Suppose that f is holomorphic in D

(0, S) for S > 0. We can take r


arbitrarily small and use the estimate above to nd c
n
= 0 for all n < 0,
so that f(z) =

+
n=0
c
n
z
n
for 0 < |z| < S. In addition, if we dene
f(0) = c
0
, the expansion is valid on D(0, S), and f is holomorphic in
D(0, S). This will be useful when we study removable singularities.
4.5.5 Singularities
Denition 33. Let f : C C.
a C is called a regular point of f i f is holomorphic at a (f H(D(a, r))
for some r).
A point a C is a singularity of f if a is a limit point of regular points
which is not itself regular.
If a C is a singularity of f and f is holomorphic in some punctured disc
D

(a, r) for some r > 0, then a is an isolated singularity. If this is not


the case, a is a non-isolated or essential singularity.
Let f : C C. Suppose that f has an isolated singularity at a. Then,
f is holomorphic in an annulus {z C, 0 < |z a| < r} for r > 0. Thus, it
has a Laurent expansion in this annulus:
f(z) =
+

n=
c
n
(z a)
n
=
+

n=0
c
n
(z a)
n
+
1

n=
c
n
(z a)
n
.
The rst series on the right-hand side is holomorphic in D(a, r) and does not
carry any singular behaviour. All the singularities are in the second sum,
that is called the principal part of the Laurent expansion. Isolated singu-
larities can then be classied according to the behaviour of the coecients
c
n
for n < 0. The point a is said to be:
4.5. ZEROS AND SINGULARITIES 155
a removable singularity if c
n
= 0 for all n < 0;
a pole of order m 1 if c
m
= 0 and c
n
= 0 for all n < m;
an isolated essential singularity if there is no m Z

such that
c
n
= 0 for all n < m.
The following examples will help us see these notions:
Example 14. 1/(z 1)
2
is its own Laurent expansion around z = 1,
where it has a double pole.
Consider the function cosec(z) = 1/ sin(z). We know that sin is holo-
morphic and has a Taylor expansion:
sin(z) = z
z
3
3!
+
z
5
5!
... = z
_
1
z
2
3!
+h(z)
_
.
All the terms after the rst two have been amalgamated in a function
h(z) that is holomorphic (because it is a power series). Near z = 0,
the dominant term in h is the one proportional to z
4
and: there is
K > 0 such that h(z) K|z
4
|. Then:
cosec(z) =
1
z
_
1
z
2
3!
+O(z
4
)
_
1
=
1
z
_
1 +
z
2
3!
+O(z
4
)
_
for |z| suciently small.
This means that the principal part of the Laurent expansion of cosec
about 0 is 1/z, so that cosec has a simple pole at 0.
(1 cos(z))z
2
is holomorphic everywhere except at z = 0, where it
is not dened. The Laurent expansion around 0 is given by:
(1 cos(z))z
2
=
1
2!

z
2
4!
+
z
4
6!
... =
+

n=0
(1)
n
z
(2n1)
(2n)!
.
Hence, the singularity at z = 0 is removable.
For z C

:
sin(1/z) =
+

n=0
(1)
n
1
(2n + 1)!z
2(n+1)
.
So, sin(1/z) has an isolated essential singularity at 0.
156 CHAPTER 4. INTEGRATION
Classifying singularities by explicitly computing the Laurent coecients
of a function is long and dicult, even for simple functions. Fortunately,
there exists more powerful methods. The general idea is that, if an holo-
morphic function f has an isolated zero at a C, then 1/f has an isolated
singularity at a C. The proof of the following proposition is similar to the
one about zeros of holomorphic functions, so it is left to the reader as an
exercise.
Proposition 30. Characterization of poles of order m.
Let f H(D

(a, r)) for a C and r > 0. Then, f has a pole of order


m N

at a i:
lim
za
(z a)
m
f(z) = D C

.
Theorem 49. Let f be holomorphic in an open disc D(a, r) for a C and
r > 0. Then, f has a zero of order m N at a i 1/f has a pole of order
m N at a.
Proof. Suppose that 1/f has a pole at a. Then f is holomorphic in a punc-
tured disc centred on a. Therefore, the zeros of f cannot be non-isolated
(identity theorem), so we can apply the theorem on the characterization of
zeros. Conversely, a zero a of f of order m is isolated, so 1/f is holomorphic
in some punctured disc D

(a, r). The result then follows from the theorem


on the characterization of zeros and the previous proposition, with some
algebra of limits.
The following proposition is a corollary of the theorems characterizing
zeros and poles:
Corollary 3. Suppose that f has a pole of order m at a C.
Let g H(D(a, r)) for some r > 0. Then, at a, fg has:
a pole of order m if g(a) = 0;
a pole of order mn if g has a zero of order n < m at a;
4.5. ZEROS AND SINGULARITIES 157
a removable singularity if g has a zero of order n m at a.
Suppose that g has a pole of order n at a. Then, fg has a pole of order
n +m at a.
Example 15. With these results in hand, we can simple determine the
order of the poles of some functions.
z sin(z) has isolated zeros at z = k for k Z. They are all of order
1, except the one at z = 0, that is of order 2. Therefore, 1/(z sin(z))
has a simple pole at every z = k with k Z

, and a double pole at


z = 0.
Consider:
f(z) =
(z 1) cos(z)
(z + 2)(2z 1)(z
2
+ 1)
3
sin
2
z
.
The denominator has a simple zero at 1/2, two triple zeros at i, a
triple zero at 2 (two come from the sinus), and double zeros at each
integer k = 2. The numerator has a simple zero at 1 and at each
(2k + 1)/2 for each integer k. Then, we see that f has triple poles at
2 and i, a double pole at k Z\{2, 1}, a simple pole at 1 and a
removable singularity at 1/2.
The last thing we have to understand is the behaviour of functions in
the neighbourhood of their singularities.
We can start with removable singularities. Let f H(D

(a, r)) such


that f has a removable singularity at a C. The Laurent expansion of f in
D

(a, r) is f(z) =

+
n=0
c
n
(z a)
n
. Then, lim
za
f(z) = c
0
, by continuity
of the series at a. By dening f(a) = c
0
, we can construct an extended f:
z D(a, r), f(z) =
+

n=0
c
n
(z a)
n
.
This means that f is now holomorphic in D(a, r). In that sense, a remov-
able singularity is nothing in particular: a simple redenition of f at z = a
158 CHAPTER 4. INTEGRATION
suppresses the singularity.
The next case is the one of a non-removable isolated singularity
(pole or essential singularity). Let f have an isolated singularity at z = a
with the Laurent expansion on D

(a, r):
f(z) =
+

n=
c
n
(z a)
n
.
Two cases can occur:
a is a pole. Let m N be the order of the pole. Then, we have
lim
za
(z a)
m
f(z) = D = 0. So, a simple manipulation of limits of
continuous functions shows that lim
za
|f(z)| = +.
a is an essential singularity. Then, we have:
Theorem 50. Casorati-Weierstrass theorem.
Let f H(D

(a, r)) for a C and some r > 0. Suppose that a is an


isolated essential singularity. Let w C. Then, there exists a sequence
(a
n
)
nN
with limit a, so that lim
n+
f(a
n
) = w.
Proof. Outline of proof.
Take f H(D

(a, r)) and w C. Suppose that > 0, z


D

(a, r), |f(z) w| . Then, by considering z 1/(f w), show


that f cannot have an essential singularity at a. The theorem is the
negation of this proposition.
Actually, Picard showed a much more powerful result: in any D

(a, )
for > 0, f assumes every complex value except, possibly, one. In the
case of e
1/z
, the essential singularity is at 0, and the exceptional value
is 0.
Finally, we should mention briey non-isolated singularities. The Lau-
rent expansion of a function f at a point a exists only if f is holomorphic in
4.5. ZEROS AND SINGULARITIES 159
a punctured disc D

(a, r) with r > 0. When a is a limit point of singularities


of f, this is not possible. Here are two examples of that case.
f(z) = cosec(1/z) has singularities at z = 0, where the function itself
is undened, and at z = 1/k with k Z

, where sin(1/z) = 0. At
z = 1/k, we have simple poles. But now, by choosing k arbitrarily
large, we can nd one of these poles arbitrarily close to 0, so 0 is a limit
point of the set of poles of f. Hence, there is no punctured disc around
0 on which f is holomorphic. 0 is thus a non-isolated singularity.
f(z) = z
3
(1 +e
1/z
)
1
has singularities at 0 and at 1/((2k +1)i) for
k Z where 1 + e
1/z
has simple zeros. The point 0 is again a limit
point of the singularities at 1/((2k +1)i), so that 0 is a non-isolated
singularity.
To go further
4.5.6 Meromorphic functions
The last topic in this section is the study of singularities in the extended
complex plane

C.
We saw previously that one can use the inversion map z 1/z to analyse
the behaviour of functions at or near . Let f : {z C, |z| > r} C where
r R

+
. We dene

f by:
w D

(0, 1/r),

f(w) = f(1/w),
and

f(0) = f(). Then, we can transfer notions about

f at 0 to the corre-
sponding notions about f at . In particular, we can talk of singularities,
poles etc at innity.
160 CHAPTER 4. INTEGRATION
Consider f(z) = z
3
. Then,

f(w) = w
3
has a triple pole at 0. Hence,
we will say that f has a triple pole at .
In the same way, z
2
sin(z) has a removable singularity at .
tan(z) has a non-isolated singularity at . Indeed, is a limit point
of the set of poles of tan(z): {(2k + 1)/2, k Z}.
Denition 34. Let G

C be open. A complex-valued function which
is holomorphic in G except, possibly for nitely many poles, is said to be
meromorphic in G.
The following theorem gives properties of meromorphic functions.
Theorem 51. f holomorphic in

C f constant.
f meromorphic in

C f is a rational function.
Proof. The rst point comes from the fact that f is bounded on

C. Indeed, f
is continuous and

C is compact. In details,

f is continuous on the compact

D(0, 1) C, thus it is bounded. Hence, f is bounded on {z C, |z|


1} {}. This implies that f is bounded. Finally, Liouvilles theorem
guarantees that f is constant.
For the second point, assume that f has poles of order m
k
at a
k
C for
k {1, ..., N}, and a pole of order m at . Then:
lim
za
N

k=1
(z a)
m
k
z
m
f(z) = D = 0.
Hence:
g(z) =
N

k=1
(z a)
m
k
z
m
f(z)
has at worst removable singularities. One can then remove them to make g
holomorphic in

C. The rst point forces g to be constant.
4.6. CAUCHYS RESIDUE THEOREM 161
4.5.7 Exercises
1. Find the zeros of the following functions and give their orders:
(i) f(z) = z
3
sin(z);
(ii) f(z) = (1 e
z
2
) sinh(z);
(iii) f(z) = (z
2
+z + 1) sin(iz).
2. Find the singularities of the following functions and characterize them:
(i) f(z) =
z
2
+4
(zi)
3
(z2i)
4
(ii) f(z) =
1e
z
sin(z)
;
(iii) f(z) =
tan(z)
z
3
+1
;
(iv) f(z) =
sinh(z)
tan(z)
;
(v) f(z) =
1
z
4
+iz
2
+1
.
3. Find the Laurent expansion of the following functions around the given
point a C, and characterize a:
(i) f(z) =
1cos(z)
z
3
, at a = 0;
(ii) f(z) = sin
_
1
zi
_
, at a = i;
(iii) f(z) =
sinh(z)
z
4
, at a = 0;
(iv) f(z) = (z
4
+z
3
+iz) sin
_
1
z
_
, at a = 0.
4.6 Cauchys residue theorem
In this section, we will extend the number of techniques available to evaluate
integrals in the complex plane, by concentrating on the relationship between
integrals around contours and singularities inside these contours.
162 CHAPTER 4. INTEGRATION
4.6.1 Residues and Cauchys residue theorem
Lemma 5. Let f be holomorphic inside and on a positively oriented contour
, except at a point a I(), where it has a pole of order m. The unique
Laurent expansion of f around a is:
f(z) =
+

n=m
c
n
(z a)
n
.
Then:
_

f(z)dz = 2ic
1
.
Proof. This lemma is a direct consequence of Laurents theorem and the
Deformation Theorem.
Denition 35. Let f H(D

(a, r)) for a C and r > 0. Suppose that


f has a pole at a. The residue of f at a is the unique coecient c
1
of
(z a)
1
in the Laurent expansion of f around a. It is denoted res{f(z), a}.
Residues are important through the following theorem.
Theorem 52. Cauchys residue theorem.
Let f be holomorphic inside and on a positively oriented contour , except
at a nite number of poles (a
k
)
k{1,...,N}
inside . Then:
_

f(z)dz = 2i
N

k=1
res{f(z), a
k
}.
Proof. Let us denote, for k {1, ..., N}, f
k
the principal part of the Laurent
expansion of f around a
k
. Then the function g dened by:
g = f
N

k=1
f
k
has only removable singularities at a
1
,..., a
N
. Once we have removed them
by redening appropriately g at the points a
1
,..., a
N
, g is holomorphic ev-
erywhere inside and on the contour . We can thus apply Cauchys theorem
4.6. CAUCHYS RESIDUE THEOREM 163
to g:
_

g(z)dz = 0. On the other hand:


_

g(z)dz =
_

f(z)dz
N

k=1
_

f
k
(z)dz = 2i
N

k=1
res{f(z), a
k
},
by using the previous lemma for each principal part.
Example 16. Cauchys residue theorem can be used to nd functions when
we know about their poles. Suppose that f is holomorphic in C except for
simple poles at the cubic roots of unity: 1, = e
2i/3
and
2
, where it
has residues 1, = 0 and 1/, respectively. Suppose that there exists a
constant K > 0 such that z C, |z| 2, |z
2
f(z)| K. By Cauchys
residue theorem, for R 2, we have:
2i(1 + +
1
) =
_
(0,R)
f(z)dz.
Hence:
|2i(1 + +
1
)|
_
2
0
|f(Re
i
)|Rd 2
K
R
.
Since R can be arbitrarily large, We have necessarily: 1 + +
1
= 0.
This equation has two solutions = and =
2
. We can then dene a
function g that has only removable singularities by subtracting the principal
parts of the Laurent expansion of f around the three poles:
g(z) = f(z)
1
z 1


z


1
z
2
.
After removing the removable singularities, g is holomorphic in C. The
bound K on |z
2
f(z)| implies that f tends to zero when |z| tends to innity,
and the same is true of g. This implies that g is bounded in C, so, by
Liouvilles theorem, it is constant, and the constant must be 0 (because of the
limit at innity). By using the values of found by solving 1++
1
= 0,
we get two possible functions satisfying the conditions:
f(z) = 3/(z
3
1) or f(z) = 3z/(z
3
1).
164 CHAPTER 4. INTEGRATION
4.6.2 Calculation of residues
Deriving residues by a direct calculation of Laurent expansions can be some-
times very boring. In this subsection, we provide formulthat allows one to
nd residues more easily.
Classication of poles
We saw previously how to characterize poles by their relationship with zeros.
We saw that the function:
f(z) =
g(z)
h(z)
has a pole of order m N at a if there exists r > 0 such that:
(g, h) H(D(a, r))
2
,
g(a) = 0,
h has a zero of order m at a.
The last condition is satised i one of the following equivalent conditions
is satised:
h(a) = h

(a) = ... = h
(m1)
(a) = 0 and h
(m)
(a) = 0,
OR: h(z) = (z a)
m
k(z) where k H(D(a, r)) with k(a) = 0.
If f has a pole of order m at a, we call the pole simple if m = 1 and
multiple otherwise. It is said to be overt if f(z) is expressed in the form
g(z)/(z a)
m
with g H(D(a, r)) for r > 0 and g(a) = 0, and covert
otherwise. Of course, whether a pole is overt or covert depends on the way
f is written, so that overt an covert poles can be changed into each others,
even though this is not usually recommended in a residue calculation.
Example 17. 1/((z i)(z + i)) has two overt simple poles: one at i
and one at i.
1/(z
2
+ 1) has covert simple poles at i.
4.6. CAUCHYS RESIDUE THEOREM 165
tan
2
(z) has covert double poles at (2k + 1)/2 for k Z.
Residue at a simple pole.
Let f H(D

(a, r)). Assume that f has a simple pole at a. Then, one has:
z D

(a, r)f(z) =
+

n=1
c
n
(z a)
n
.
Multiplying by (z a):
z D

(a, r), (z a)f(z) =


+

n=1
c
n
(z a)
n+1
= c
1
+
+

n=0
c
n
(z a)
n+1
.
The second part of the right-hand side tends to zero as z tends to a, so we
have:
lim
za
(z a)f(z) = res{f(z), a}.
Now, we can study both types of simple poles:
If a is an overt simple pole, then, there exists g H(D(a, r)) for
some r > 0 such that f(z) = g(z)/(z a) with g(a) = 0. Then, we
have:
res{f(z), a} = g(a).
If a is a covert simple pole, then, we can write f(z) = h(z)/k(z)
where (h, k) H(D(a, r))
2
with h(a) = 0, k(a) = 0 and k

(a) = 0.
Hence:
res{f(z), a} = lim
za
(z a)
h(z)
k(z)
= h(a) lim
za
z a
k(z) k(a)
=
h(a)
k

(a)
.
Then, for a covert simple pole, we have:
res{f(z), a} =
h(a)
k

(a)
if f(z) =
h(z)
k(z)
.
166 CHAPTER 4. INTEGRATION
Residue at a multiple pole.
Suppose that f as a pole of order m > 1 at a C. Then, again, two things
can occur.
f has an overt multiple pole. Then, f(z) = g(z)/(z a)
m
with
g H(D(a, r)) and g(a) = 0. In that case, by applying Cauchys
formula:
g
(m1)
(a) =
(m1)!
2i
_
(a,r/2)
g(z)
(z a)
m
dz
=
(m1)!
2i
_
(a,r/2)
f(z)dz
= res{f(z), a}.
Hence, we have:
res{f(z), a} =
1
(m1)!
g
(m1)
(a).
f has a covert multiple pole. Then, there is no simple formula. To
nd the residue, you have to either convert the pole to an overt pole
or compute the Laurent coecient c
1
from scratch by expanding the
function in powers of (z a) for (z a) small.
Example 18. Consider f(z) = 1/((2 z)(z
2
+ 4)), f has 3 simple
poles: one at 2 and two at 2i. The pole at 2 is overt:
res{f(z), 2} = res
_
(z
2
+ 4)
1
z 2
, 2
_
=
1
8
.
The poles at 2i are covert. f(z) = (2z)
1
/(z
2
+4), and (z
2
+4)

=
2z, hence:
res{f(z), 2i} =
_
(2 z)
1
2z
_
z=2i
=
1 i
16
.
4.6. CAUCHYS RESIDUE THEOREM 167
Consider f(z) = 1/(1 + z
4
). f has covert simple poles at z
k
=
e
(2k+1)i/4
for k {0, 1, 2, 3}. Hence:
res{f(z), z
k
} =
_
1
(1 +z
4
)

_
z=z
k
=
_
1
4z
3
_
z=z
k
=
1
4z
3
k
=
1
4z
k
=
1
4
e
(2k+1)i/4
, (4.5)
since z
4
k
= 1 z
3
k
= z.
Consider f(z) = e
iz
/z
4
. It has an overt pole of order 4 at 0. Then:
res{f(z), 0} =
1
3!
_
d
3
dz
3
e
iz
_
z=0
=
i
6
.
One could also have used the Laurent expansion:
e
iz
z
4
=
1
z
4
+
i
z
3

1
2!z
2

i
3!z
+... for |z| > 0.
Then, res{f(z), 0} = c
1
= i/6.
If f(z) = z
3
/ sin(z), then f has a covert pole of order 4 at 0, and we
have:
f(z) =
1
z
3

1
z
_
1 +
z
2
3!

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

7
60
+O(z
2
).
So, res{f(z), 0} = 0.
Integrals around the unit circle.
We would like to use Cauchys residue theorem to prove that:
_
2
0
1
8 cos
2
()
d =
2
3
.
The idea is to rewrite the integral as an integral on the unit circle: (0, 1) =
{z C, z = () = e
i
, [0, 2]}. Then:
dz =

()d = izd,
168 CHAPTER 4. INTEGRATION
and:
cos() = (e
i
+e
i
)/2 = (z +z
1
)/2.
Hence:
_
2
0
1
8 cos
2
()
d =
_
(0,1)
1
1 + 2(z
2
+ 2 +z
2
)
dz
iz
=
_
(0,1)
z
z
4
+ 5z
2
+ 2
dz
=
_
(0,1)
z
(2z
2
+ 1)(z
2
+ 2)
dz.
The integrand has simple poles at i/

2 and i

2. The poles at i

2
are outside (0, 1), so they do not contribute to the integral. So:
_
2
0
1
8 cos
2
()
d = 2i
_
res
_
z
(2z
2
+ 1)(z
2
+ 2)
,
i

2
_
+ res
_
z
(2z
2
+ 1)(z
2
+ 2)
,
i

2
__
=
_
2i
4(z
2
+ 2)
_
z=i/

2
+
_
2i
4(z
2
+ 2)
_
z=i/

2
=
2
3
.
4.6.3 Exercises
1. Calculate the residues of the following functions at their poles:
(i) f(z) =
e
z+1
(zi)
;
(ii) f(z) =
cos(z)
(z3)
2
;
(iii) f(z) =
sin(z)
z
3
;
(iv) f(z) =
z
2
+z+1
sin(z)
;
(v) f(z) = tan(z);
(vi) f(z) = tanh(z);
(vii) f(z) =
1
z
4
cos(z)
.
2. Calculate the following integrals:
4.6. CAUCHYS RESIDUE THEOREM 169
(i)
_
(0,1)
z
2
+2
4z
2
+1
dz;
(ii)
_
(0,1)
sin(z)
8z
4
+1
dz;
(iii)
_
(i,1)
cos(z)
sin(zi)
dz;
(iv)
_
(0,4)
z
2
+z+1
cosh(z)
dz;
(v)
_
(0,2)
z+1
z
2
+2z+i
dz.
170 CHAPTER 4. INTEGRATION
Chapter 5
Applications of integration in
the complex plane
171
172 CHAPTER 5. APPLICATIONS
5.1 Some applications of contour integration
In this section, we are going to apply everything we know about contour
integration to dierent context. This will make us realize how powerful
complex integration is, as a technique applied to various parts of mathemat-
ics, besides the interest it has on its own.
5.1.1 Evaluation of real integrals by contour integration
First, we would like to see how to apply contour integration to the evaluation
of real integrals. Let us start with an illustrative example. Consider:
I =
_
+
0
1
1 +x
4
dx.
Such an integral, considered in R is extremely dicult to solve. Let us rst
note that the integrand is an even function, so that, for R R

+
:
_
R
0
1
1 +x
4
dx =
1
2
_
R
R
1
1 +x
4
dx.
The integral I has to be understood as: I = lim
R+
_
R
0
1
1+x
4
dx. This is
known as the principal value of the integral.
Let be the positively oriented semi-circular contour obtained by joining
the upper arc of (0, R), (0, R), with the line segment [R, R]. We have:
_

1
1 +z
4
dz =
_
R
R
1
1 +x
4
dx +
_

0
Re
i
1 +R
4
e
4i
d.
By taking the limit R +, the rst term on the right-hand side is
just 2I. Consider the second term. We have:

_

0
Re
i
1 +R
4
e
4i
d


_

0

Re
i
1 +R
4
e
4i

R
|R
4
1|
.
5.1. SOME APPLICATIONS OF CONTOUR INTEGRATION 173
Figure 5.1: Contour for the evaluation of I =
_
+
0
1
1+x
4
dx.
Now, lim
R+
R
|R
4
1|
= 0, so that
_

0
Re
i
1+R
4
e
4i
d tends to zero when R tends
to +.
Thus, we only have to compute the value of
_

1
1+z
4
dz by using Cauchys
residue theorem. The integrand is holomorphic inside and on except for
covert simple poles at e
(2k+1)i/4
for k {0, 1, 2, 3}. The only of these poles
that are inside are z
1
= e
i/4
=
1+i

2
and z
2
= e
3i/4
=
1+i

2
. So, we have:
res{1/(1 +z
4
), z
k
} =
_
1
(1 +z
4
)

_
z=z
k
=
1
4
z
k
for k {1, 2},
because, for k {1, 2}, z
4
k
= 1 z
3
k
= z
k
. Hence, by Cauchys residue
theorem:
_

1
1 +z
2
dz =
2i
4
_
e
i/4
+e
3i/4
_
=
i
2
_
1 +i

2
+
1 +i

2
_
=

2
.
174 CHAPTER 5. APPLICATIONS
Thus, we can conclude that:
_
+
0
1
1 +x
4
dx =

2

2
.
In the previous section, we saw how to calculate a real integral
_
2
0
1
8 cos
2
()
d =
2/3 by reducing it to a complex integral around a contour, to which we ap-
plied Cauchys residue theorem. Here we saw a more complicated example.
The key elements in these evaluations are:
Relate the real integral I that is to be calculated, or an approximation
of it (by estimation as before), to some contour integral
_

f(z)dz.
Apply Cauchys residue theorem to
_

f(z)dz. That requires that f


has at most nitely many poles inside and none on itself.
The contour has to be chosen so that the integral of f along each
portion of it either contributes to I or can be handled by estimation
(or any other way that allows to calculate them).
These guidelines are really to be taken for what they are, i.e. advices.
The best here is to practice through a lot of examples. We will see some of
them in the following subsections.
5.1.2 Some remarks on indented contours
Consider a complex-valued function f and a contour . As emphasized ear-
lier, when we want to integrate f around it is vital, to apply the theorems
on contour integration that f does not have any pole on itself, or that,
if f is a multifunction, it does not have a branch point on . Nevertheless,
this can happen, and in that case, we would like to be able to avoid the
point that create troubles by adding a small circular arc of radius to ,
so that the new path avoids the point. Then, we would like to control the
limit of the integral along this new path when tends to zero. This can be
made precise by the following theorem.
5.1. SOME APPLICATIONS OF CONTOUR INTEGRATION 175
Theorem 53. Indentation lemma for a simple pole. Let f H(D

(a, r))
have a simple pole at a with a residue b. We call indentation around a
a circular arc

() = a +e
i
for [
1
,
2
] [0, 2] and 0 < < r. Then,
we have:
lim
0
_

f(z)dz = ib(
2

1
).
We see that in the case of a circle, we recover Cauchys residue theorem.
Proof. We saw already that, a being a simple pole with residue b, we have:
b = lim
za
(z a)f(z).
Let > 0, then, by denition of the limit, there exists > 0 such that:
0 < |z a| < |(z a)f(z)| < .
Consider ]0, min{r, }[. If z =

(), then

() = ie
i
= i(z a), so:

f(z)dz ib(
2

1
)

_

2

1
_
f(

())

() ib
_
d

_

2

1
i ((z() a)f(z()) b)

< (
2

1
).
This is exactly the statement that ib(
2

1
) is the limit of
_

f(z)dz when
tends to 0.
Remark 9. The previous theorem is only valid for simple poles. It does not
apply to multiple poles. Indeed, for a multiple pole, |(z a)f(z)| is not
controllable when z approaches a: it blows up too fast.
We would like also like to be able to apply Cauchys theorem or the
residue theorem to a function that is a holomorphic branch of a multifunc-
tion. In order to do so, we must specify a contour that lies into the cut
plane. To illustrate this, let us consider a branch of the logarithm in the
plane cut along [0, +[:
f(z) = ln(r) +i, for z = re
i
= 0 and [0, 2[.
176 CHAPTER 5. APPLICATIONS
Figure 5.2: A keyhole contour to avoid a branch cut.
We cannot integrate f around (0, R) because of the cut. But, consider
the keyhole contour in Fig. 5.1.2 In the cut plane, we can take the horizontal
line to be arbitrarily close to the cut. Call their distance to the cut. Then,
f is holomorphic inside and on . Moreover if we parametrize

by (t)
with t [a, b] R:
_

f(z)dz =
_
b
a
f((t))

(t)dt,
with

(t)dt = (x

+iy

)dt if we write z = x +iy

. Hence:
_

f(z)dz =
_

Re(f(x, y))dx
_

Im(f(x, y))dy+i
_

Im(f(x, y))dx+i
_

Re(f(x, y))dy.
Hence, noting that, along [A, B] there is no variation of the imaginary part
5.1. SOME APPLICATIONS OF CONTOUR INTEGRATION 177
of z, we have:
_
[A,B]
f(z)dz =
_
R

Re(f(x +i))dx +i [Im(f(x +i))]


|[A,B]
_
R

dx.
On [A, B], at rst order in , we have: = arg(z) arcsin(/) /.
Thus:
[Im(f(x +i))]
|[A,B]
_
R

dx
R

.
Thus, lim
0
+ [Im(f(x +i))]
|[A,B]
_
R

dx = 0. This implies that:


lim
0
+
_
[A,B]
f(z)dz =
_
R

lim
0
+
ln(
_
x
2
+
2
)dx =
_
R

ln tdt.
In the same way:
lim
0
+
_
[C,D]
f(z)dz =
_

R
(ln(t) + 2i) dt.
Thus, when we integrate in the cut plane, we can always integrate along the
edge, using the edge-values integrand, on both sides of the edge.
What happens at the branch point? Well, the question is dicult. Note
that if a is a branch point of f, then f is not holomorphic in any punctured
disc centred on a. Therefore, a cannot be an isolated singularity of f, and
in particular, it cannot be a pole of f, so that the indentation method does
not apply. Usually, one has to rely on estimation techniques.
For example, for the logarithm with the previous key-hole contour, one
has, at the limit 0:
_

f(z)dz =
_
(0,R)
f(z)dz + lim
0
_
[C,D]
f(z)dz
_
(0,)
f(z)dz + lim
0
_
[A,B]
f(z)dz
= 2iR +
_
1


1
R
+ 2i( R)
_
2i +
_
1
R

1

_
= 0.
5.1.3 Integral of rational functions
We are just going to illustrate the general method by considering examples.
178 CHAPTER 5. APPLICATIONS
Example 19. Consider I =
_
+
0
1
(x
2
+1)
2
(x
2
+4)
dx. To evaluate this integral
directly is virtually impossible. We then consider the contour in the com-
plex plane made of the joint of the line segment [R, R] and the positively
oriented semi-circle, (0, R), joining R to R, for R R

+
such that R > 2.
Then, we integrate f(z) = 1/((z
2
+1)
2
(z
2
+4)) around . The choice R > 2
ensures that two poles of f are inside , i (double pole) and 2i (simple pole)
and none on , so that f is holomorphic inside and on , except at i I()
and 2i I(). Then, by Cauchys residue theorem:
_
R
R
f(z)dz +
_
(0,R)
f(z)dz = 2i (res{f(z), i} + res{f(z), 2i}) .
We see that i is an overt double pole: f(z) =
1/((z+i)
2
(z
2
+4))
(zi)
, and 2i a covert
simple pole, so:
res{f(z), i} =
_
d
dz
1
(z +i)
2
(z
2
+ 4)
_
z=i
=
_
2z(z +i) 2(z
2
+ 4)
(z +i)
3
(z
2
+ 4)
2
_
z=i
=
i
36
res{f(z), 2i} =
_
1/(z
2
+ 1)
(z
2
+ 4)

_
z=2i
=
i
18
.
Moreover:

_
(0,R)
f(z)dz

_

0
1
(R
2
1)
2
(R
2
4)
Rd = O(1/R
5
),
and:
_
R
R
f(x)dx = 2
_
R
0
1
(x
2
+ 1)
2
(x
2
+ 4)
dx.
So, by taking the limit R +, we see that the integral along the semi-
circle tends to 0, and:
I =

6
.
Remark 10. Note that, once again, the parity of the integrand was crucial
in the derivation of the result.
Example 20. Consider now:
I =
_
+
0
1
1 +x
10
.
5.1. SOME APPLICATIONS OF CONTOUR INTEGRATION 179
The function f(z) = 1/(1 +z
10
) is holomorphic everywhere in C except for
simple covert poles at the points z
k
= e
i(2k+1)i/10
for k {0, 1, 2, ..., 9}.
We could use a semicircular contour, but that would include many poles.
Rather, let us consider the contour shown on Fig. 26. The only pole inside
the contour is = e
i/10
. Its residue is given by:
res{f(z), } =
_
1
(1 +z
10
)

_
z=
=
1
10
9
=

10
.
Figure 5.3: Contour for the evaluation of I =
_
+
0
1
1+x
10
.
On the line segment between 0 and Re
i/5
: z = te
i/5
with t [0, R], so
that dz/dt = e
i/5
, and:
1 +z
10
= 1 +t
10
e
2i
= 1 +t
10
.
180 CHAPTER 5. APPLICATIONS
So, by Cauchys residue theorem:
_
R
0
1
1 +x
10
dx +
_
/5
0
Rie
i
1 +R
10
e
10i
d +
_
0
R
e
i/5
1 +t
10
dt =
2ie
i/10
10
.
The integral on the circular arc is O(1/R
9
), so it vanishes at the limit R
+. Hence:
_
1 e
i/5
_
_
+
0
1
1 +x
10
dx =
i
5
e
i/10
.
Finally, noting that:
1 e
i/5
= e
i/10
(e
i/10
e
i/10
) = 2ie
i/10
sin(/10),
we have:
_
+
0
1
1 +x
10
dx =

10
1
sin(/10)
=

10
cosec(/10).
In the previous example, note how the integral along the slanting line
gives a multiple of the integral along the real axis. When this happens, and
two subpaths yield integrals that are multiple of one another, we say that
we have integral reinforcement.
5.1.4 Integral of other functions with a nite number of poles
In this subsection, we would like to consider integrals of the form:
_
I
(x) sin(mx)dx or
_
I
(x) cos(mx)dx or
_
I
(x)e
imx
dx,
where I is [0, +[, ]0, +[ or R, m 0, and (z) = p(z)/q(z) is a rational
function with deg q > 1 + deg p.
Example 21. Our rst example is:
J =
_
+

cos(x)
x
2
+x + 1
dx.
We consider the complex function:
f(z) =
e
iz
z
2
+z + 1
.
5.1. SOME APPLICATIONS OF CONTOUR INTEGRATION 181
It is holomorphic everywhere except for simple poles at the roots of z
2
+z +
1 = 0, i.e. at e
2i/3
and e
4i/3
. Then, consider again the contour mad of
the join of [R, R] and the positively oriented semi-circle (0, R) for R > 1.
Then, only e
2i/3
is inside the contour. By Cauchys residue theorem, we
thus have:
_
R
R
f(x)dx +
_
(0,R)
f(z)dz = 2ires{f(z), e
2i/3
} = 2i
_
e
iz
2z + 1
_
z=e
2i/3
,
i.e.:
_
R
R
f(x)dx +
_
(0,R)
f(z)dz ==
2

3
e
i(1/2+i

3/2)
.
As usual, the integral along the semi-circle can be evaluated:

_
(0,R)
f(z)dz

_

0
R
|R
2
R 1|
d = O(1/R),
and vanishes at the limit R +. Hence, taking the limit and equating
real and imaginary parts:
_
+

cos(x)
x
2
+x + 1
=
2

3
e

3/2
cos(1/2).
A few comments are in order:
The integrand is not an even function, so we cannot use the usual trick
to evaluate the integral on [0, +[.
We couldnt have chosen f(z) = cos(z)/(z
2
+z +1), because then, the
integral along the semi-circle would not have converged towards 0 as
R became arbitrarily large. Indeed: | cos(Re
i
)|
2
= cosh
2
(Rsin())
sin
2
(Rcos()), that grows like cosh
2
R when approaches /2, so we
couldnt have found a supremum.
Let us see another example.
Example 22. Try and evaluate I =
_
+
0
sin
2
x
x
2
.
Because of the problem we mentioned previously with the cosinus, we cannot
182 CHAPTER 5. APPLICATIONS
choose f(z) = sin
2
(z)/z
2
: we would have problems nding a supremum
that tends to zero along a circular arc. Instead, we would like to nd a
function whose real part is sin
2
x/x
2
when z = x R. Remembering that
2 sin
2
x = 1 cos(2x), we are led to consider f(z) = (1 e
2iz
)/z
2
. The
function is holomorphic everywhere except at z = 0, where it has a pole.
The Laurent expansion of the function is given by:
f(z) =
1
z
2
_
1
+

n=0
(2iz)
n
n!
_
=
+

k=1
(2i)
k+2
(k + 2)!
z
k
.
So, the pole is simple with residue 2i. Hence, we cannot use the usual
contour, because the line segment [R, R] encounters the pole. We then use
an indented contour that goes around the pole at z = 0 (cf Fig. 26).
Figure 5.4: Contour for the evaluation of I =
_
+
0
sin
2
x
x
2
.
5.1. SOME APPLICATIONS OF CONTOUR INTEGRATION 183
Then, f is holomorphic inside and on the contour, so that, by Cauchys
theorem:
_

R
f(x)dx
_
(0,)
f(z)dz +
_
R

f(z)dz +
_
(0,R)
f(z)dz = 0.
The rst and third integrals combine to give:
_
R

1 e
2ix
x
2
dx +
_
R

1 e
2ix
x
2
dx = 2
_
R

1 cos(2x)
x
2
dx =
_
R

4 sin
2
x
x
2
.
Because the pole is simple, we can apply the indentation lemma, and we
have:
lim
0
_
(0,)
f(z)dz = i( 0)res{f(z), 0} = 2.
Finally:

_
(0,R)
f(z)dz

_
2
0
1 +e
2Rsin()
R
2
Rd = O(1/R).
So, letting 0 and R +, we get:
_
+
0
sin
2
x
x
2
=

2
.
Finally, let us present a last example, in which we highlight the impor-
tance of the choice of the contour.
Example 23. Consider:
I =
_
+

e
2ix
1 +x
4
dx.
The obvious choice of function is to take f(z) = e
2iz
/(1 + z
4
). Then,
f is holomorphic everywhere except for simple poles at z
k
= e
(2k+1)i
for
k {0, 1, 2, 3}. If we write z = Re
i
, we see that |e
2iz
| = e
2Rsin()
, that
is not bounded when R tends to innity and ]0, [, so that we cannot
use the usual semi-circle (0, R) in the upper half-plane. Nevertheless, it is
bounded as long as [, 0], so we can draw the semi-circle from R to
R in the lower half-plane. Lets call it (0, R). If we join it with the line
184 CHAPTER 5. APPLICATIONS
segment [R, R], we get a contour that encircles two poles of f, at e
3i/4
and e
i/4
. The residues at that poles are given by:
res{f(z), e
i/4
} =
_
e
2iz
4z
3
_
z=e
i/4
=

2e

2
e
i

2
8
(1 i)
res{f(z), e
3i/4
} =
_
e
2iz
4z
3
_
z=e
3i/4
=

2e

2
e
i

2
8
(1 +i).
Then, Cauchys residue theorem gives:

_
R
R
e
2ix
1 +x
4
dx +
_
0

e
2iRe
i
1 +R
4
e
4i
d =

2
e

2
_
cos(

2) + sin(

2)
_
.
So that, bounding the integral along the circular arc by O(1/R
3
) and taking
the limit R +, we get:
_
+

e
2ix
1 +x
4
=

2
2
e

2
_
cos(

2) + sin(

2)
_
.
5.1.5 Integrals of functions with an innite number of poles
In this subsection, we would like to exemplify a method to evaluate integrals
of the form:
_
+

(x) cos(mx)dx or
_
+

(x) sin(mx)dx,
where (z) is a function with an innite number of regularly spaced poles;
for example, (z) = 1/ cos(z), (z) = 1/ sin(z) or (z) = 1/(1 e
z
).
Example 24. For a ] 1, 1[, consider the integral:
I =
_
+

e
ax
cosh(x)
dx.
The function f(z) = e
az
/ cosh(z) has simple poles at the zeros of cosh(x),
i.e. at z
k
=
2k+1
2
i, for k Z. If we were to consider a semi-circular
contour, at the limit of the radius tending to innity, we would have to
sum an innite number of residues. So, we consider a rectangular contour
which size is limited along the imaginary axis (height ), but can increase
arbitrarily along the real axis (cf Fig. 24).
5.1. SOME APPLICATIONS OF CONTOUR INTEGRATION 185
Figure 5.5: Contour for the evaluation of I =
_
+

e
ax
cosh(x)
dx.
Then, f is holomorphic inside and on this contour, except at z = i/2,
where there is a simple pole. We have:
res{f(z), i/2} =
_
e
az
sinh(z)
_
z=i/2
= ie
ai/2
.
So that:
_
R
R
e
ax
cosh(x)
dx +
_

0
e
a(R+iy)
cosh(R +iy)
idy +
_
R
R
e
ai
e
ax
cosh(x +i)
dx
+
_
0

e
a(R+iy)
cosh(R +iy)
dy = 2e
ai/2
.
186 CHAPTER 5. APPLICATIONS
The integrals along the vertical lines can be evaluated as follows:

_

0
e
a(R+iy)
cosh(R +iy)
idy


_

0

2e
a(R+iy)
e
(R+iy)
+e
(R+iy)

dy

_

0
2e
aR
|e
R
e
R
|
dy 2e
(a1)R
when R +
Since a < 1, that proves that the integral tends to zero as R becomes
arbitrarily big. Similarly:

_

0
e
a(R+iy)
cosh(R +iy)
idy

_

0
2e
aR
|e
R
e
R
|
dy 2e
(1+a)R
as R +,
and, since a > 1, the integral also tends to zero. Finally, note that the
two remaining integrals reinforce to give the required real integral, because
cosh(x +i) = cosh(x) and e
a(x+i)
= e
ai
e
ax
. Hence:
_
+

e
ax
cosh(x)
dx =

cos(a/2)
for a ] 1, 1[.
5.1.6 Integrals involving multifunctions
In this subsection, we consider integrals of the form:
_
+
0
(x) ln(x)dx and
_
+
0
(x)x
a1
dx for a > 0,
where (z) is meromorphic. We then have to work in the cut plane, with
a selected holomorphic branch of the multifunction. The branch point at 0
is avoided thanks to an indentation. First, let us see an example with the
logarithm.
Example 25. Consider:
I =
_
+
0
ln x
1 +x
2
dx.
We cut the plane along ] , 0], and select the holomorphic branch:
ln(z) = ln |z| +i for arg(z)] , ].
5.1. SOME APPLICATIONS OF CONTOUR INTEGRATION 187
Then, f(z) = ln(z)/(1+z
2
) is holomorphic in the cut plane except for simple
poles at i. Let be the contour of Fig. 25, with R > 1. On the top side
of the cut, = , so ln(z) = ln(x) + i with z = x > 0. Cauchys residue
theorem then gives:
_
R

ln(x)
1 +x
2
dx +
_
(0,R)
f(z)dz +
_

R
ln(x) +i
1 +x
2
(dx)

_
(0,)
f(z)dz = 2ires{f(z), i}.
Figure 5.6: Contour for the evaluation of I =
_
+
0
ln x
1+x
2
dx.
Moreover:
res{f(z), i} =
_
ln(z)
2z
_
z=i
=
ln(i)
2i
=
i/2
2i
=

4
.
188 CHAPTER 5. APPLICATIONS
Also:

_
(0,R)
f(z)dz


_

0

ln(R) +i
1 +R
2
e
2i

_

0
(ln(R) +)R
R
2
1
R
1
ln(R) when R +.
And, similarly:

_
(0,)
f(z)dz

_
(| ln()| +)
1
2
d ln() when 0.
Hence, by taking the limits R + and 0, we get:
2
_
+
0
ln(x)
1 +x
2
dx +i
_
+
0
1
1 +x
2
dx =

2
2
i,
so that, by equating real and imaginary parts, we get:
_
+
0
ln(x)
1 +x
2
dx = 0.
Note that the imaginary part gives us an integral we knew how to calculate
already (the integral is just arctan(x)).
Let us conclude with an example involving a power law.
Example 26. Consider:
I =
_
+
0

x
1 +x
3
dx.
The function 1/(1 + z
3
) has simple poles at 1, e
i/3
and e
i/3
. The pole
at 1 leads us to control the square root by cutting the plane along [0, +[
(to avoid cutting on a pole of the integrand). We then take the holomorphic
branch:

z =
_
|z|e
i/2
with [0, 2[, and we use the contour shown n
Fig. ??. On the top side of the cut, z = x > 0 and

z =

x, while on the
bottom side, z = |z|e
i2
and

z =

x with x > 0. The integrals along


5.1. SOME APPLICATIONS OF CONTOUR INTEGRATION 189
the two sides of the cut are in opposite directions, so that they reinforce.
The residues at the three poles inside are:
res{f(z), 1} =
_
z
3z
2
_
z=1
=
i
3
res{f(z), e
i/3
} =
i
3
res{f(z), e
i/3
} =
i
3
.
Figure 5.7: Contour for the evaluation of I =
_
+
0

x
1+x
3
dx.
So, by evaluating the integrals along the circular arcs and taking the
appropriate limits, we nd:
_
+
0

z
1 +z
3
=

3
.
190 CHAPTER 5. APPLICATIONS
5.1.7 Summation of series
We have encountered convergence tests, that tell us whether or not a given
series converges. But they do not give us the values of the sums for conver-
gent series. Now, if an innite sum can be recognized as a sum of residues
of a meromorphic function, then by using contour integration, we may be
able to evaluate it. Let us start with an example.
Example 27. Consider the known series:
+

n=0
1
n
2
=

2
6
.
How can we recover this result using contour integration?
The function f(z) = /(z
2
tan(z)) is holomorphic everywhere, except for
a triple pole at 0 and simple covert poles at n Z

. The residues at the


poles are:
res{f(z), 0} =

2
3
n Z

, res{f(z), n} =
1
n
2
.
Consider the square contour

N
with vertices at (N + 1/2)(1 i). f is
holomorphic inside and on
N
except for poles at 0, 1, 2,..., N. Then,
Cauchys residue theorem leads to:
_

N
f(z)dz = 2i
_
2
N

n=1
1
n
2


2
3
_
.
Now:

N
f(z)dz

sup
z

N
|f(z)| length(
N
)
sup
z

1
tan(z)

4(2N + 1)
(N + 1/2)
2
.
First, note that on the square

N
, there exists a constant C R

+
such that

1
tan(z)

C. The right-hand side of this inequality behaves like 1/N when


5.2. THE FOURIER TRANSFORM 191
N is big, so it tends to zero when N tends to innity. Thus,
_

N
f(z)dz
tends to 0 when N tends to innity. This proves the result on the series.
It turns out that the method presented on the example above applies to
any series

+
n=1
(n), where the function satises the following properties:
n N

, (n) = (n);
is a rational function;
(z) |z|
2
for large |z|.
If this is the case, then we integrate f(z) = (z)/ tan(z) along the square

N
. The term / tan(z) creates simple poles of f at each n Z at which
is holomorphic and non-zero, of residue (n). The bounds on and
|1/ tan(z)| then ensures that
_

N
f(z)dz 0 when N +.
In the same way, we can evaluate series of the form:

+
n=1
(1)
n
(n),
with satisfying the previously listed conditions, by integrating f(z) =
(z)/ sin(z) around the square

N
as before. The 1/ sin(z) term creates
simple covert poles at n Z, of residue (1)(n), provided is holomorphic
and non-zero there. Then, using Cauchys residue theorem and bounding
|1/ sin(z)| on the square

N
, we can deduce the value of the series.
5.2 The Fourier transform
The Fourier transform is a very important theoretical concept, as well as a
central practical tool in solving many mathematical problems that arise in
the modelisation of phenomena. Here, we concentrate on Fourier transforms
coming from probability theory, as well as some examples of how to use teh
Fourier transform to solve dierential equations.
5.2.1 Introducing the Fourier transform
Let J denote [0, +[ or R. We will denote by I(J) the set of complex-valued
functions dened on J which are piecewise continuous on any bounded sub-
192 CHAPTER 5. APPLICATIONS
interval of J and such that each of f and |f| has a weel-dened integral on
J.
Denition 36. Let f : R C be a real or complex valued function such
that f I(R). The Fourier transform of f is dened by:
s R, (Ff) (s) =

f(s) =
_
+

f(x)e
isx
dx.
Note that this denition is not unique, and some variants can be found
throughout the litterature: e
isx
can be used instead of e
isx
, and sometimes,
a normalisation factor

2 may be included.
Proposition 31. Provided all the transforms exist, we have:
(a, b) C
2
, F[af +bg] = a

f +b g;
a R

+
, F[af(x/a)] = a

f(as);
a R, F
_
e
ixa
f(x)

=

f(s +a).
Proof. The proof is left to the reader.
The following results are important to be able to apply the Fourier trans-
form to nd solutions to dierential equations. Their proofs are simple
calculations that can be justied in integration theory.
Proposition 32. Let f I(R). Suppose that f satises the following con-
ditions:
n N, (f, f

, f

, ..., f
(n)
) I(R), and f
(n)
continuous.
Then:
F
_
f
(n)
(x)
_
= (is)
n

f(s).
Proof. The result is obtained by repeated integration by parts. The con-
ditions are here to ensure that these integrations by parts can be done n
times. They are strong enough to lead to f
(k)
(x)e
isx
0 as |x| +,
5.2. THE FOURIER TRANSFORM 193
for k {0, 1, ..., n 1}, so that we can get rid of the fully integrated terms.
These decay conditions are usually met in the problems we want to apply
the Fourier transform to.
Proposition 33. For f I(R) such that the derivatives of f exist and are
continuous up to order at least n N:
F[x
n
f(x)] = i
n

f
(n)
(s).
Proof. The conditions on f are sucient to justify dierentiation under the
integral sign n times, so that the result is obtained by dierentiation n times
of

f.
We see that the Fourier transform changes derivatives into simple prod-
uct. That will be key in the methods used to solve dierential equations
later.
Proposition 34. Inversion theorem.
Let f I(R) such that f and f

are piecewise continuous on R. Then:


1
2
_
f(x
+
) +f(x

))
_
=
1
2
_
+

f(s)e
isx
ds.
If f is continuous, then we have:
f(x) =
1
2
_
+

f(s)e
isx
ds.
The proof of this result rely a lot on subtleties of integration theorey
that are beyond teh scope of these notes.
Proposition 35. Convolution for the Fourier transform.
Let (f, g) I(R). Then:

f(s) g(s) =

h(s),
where h is the convolution:
h(x) =
_
+

f(y)g(x y)dy = (f g)(x).


Hence, the Fourier transform of a convolution product is teh product of
the Fourier transforms:

f g =

f g.
194 CHAPTER 5. APPLICATIONS
5.2.2 Some applications
Application in probability theory
Here, we consider some fundamental probability distributions on R and com-
pute their characteristic functions. These functions are very useful in prob-
ability theory since they encode information about the moments of the as-
sociated distributions. It turns out that for a given probability density f,
the characteristic fubction is simply the Fourier transform of f.
Example 28. Cauchy distribution.
It is given by the density function f(x) =
1
(1+x
2
)
. The Fourier transform is
then:

f(s) =
_
+

e
isx
(1 +x
2
)
dx.
The inegrand has two simple poles i and i. If we write z = Re
i
, we see
that |e
isz
| = e
Rs sin()
. So, if s 0, we can use a semi-circular contour in
the upper-half plane, and if s > 0, we have to use a semi-circular contour in
the lower half-plane, so that |e
isz
| along the semi-circular arc is always a
negative exponential, and can be bounded from above by 1. We then arrive
at:

f(s) =
_
_
_
2ires
_
e
isz
(1+z
2
),i
_
= e
s
if s 0
2ires
_
e
isz
(1+z
2
),i
_
= e
s
if s > 0.
Hence, the characteristic function is given by:
s R,

f(s) = e
|s|
.
Example 29. Normal distribution.
In this case:
f(x) =
1

2
e

1
2
x
2
.
First:

2

f(s) =
_
+

e
x
2
/2
e
isx
dx = e
s
2
/2
_
+

e
(x+is)
2
/2
dx.
5.2. THE FOURIER TRANSFORM 195
Then, we integrate e
z
2
/2
around the rectangle with vertices R, R, R+is
and R + is. By Cauchys theorem, since there is no pole in the complex
plane:
_
R
R
e
x
2
/2
dx+
_
s
0
e
(R+iy)
2
/2
idy+
_
R
R
e
(x+is)
2
/2
dx+
_
0
s
e
(R+iy)
2
/2
idy = 0.
Moreover:

_
s
0
e
(R+iy)
2
/2
idy

e
R
2
/2
_
|s|
0
e
y
2
/2
dy 0 when R +.
Similarly,
_
s
0
e
(R+iy)
2
/2
idy tends to zero when R tends to +. Hence,
letting R +:
_
+

e
x
2
/2
dx =
_
+

e
(x+is)
2
/2
dx.
It is well-known, and it can be shown using contour integration (Show it),
that the left-hand side is

2. So:

f(s) = e

1
2
s
2
.
Example 30. Gamma distribution.
For > 0 and t > 0, the Gamma distribution is given by:
f(x) =
1
(t)

t
x
t1
e
x

[0,+[
(x),
where
I
is the characteristic function of the interval I, and:
(t) =
_
+
0
x
t1
e
x
dx
is the gamma function. For an arbitrary t > 0, we have to deal with a
multifunction, so we have to select a convenient branch cut. We will work
in the plane cut along ] , 0], and take the branch:
z
t1
= |z
t1
|e
i(t1)
, for z = |z|e
i
, ] , ].
We take > 0, and we choose a wedge contour formed of the join of: the
line segment [, R], the circular arc centred on 0, of radius R and aperture
196 CHAPTER 5. APPLICATIONS
] /2, /2], the line segment [Re
i
, ], and the clockwise circular arc
centred on 0, of radius and aperture . Let g(z) = z
t1
e
z
. Then, we
have:

g(z)dz


_
||
0

(Re
i
)
t1
e
Re
i
Rie
i

_
||
0
R
t
e
Rcos()
d
||R
t
e
Rcos
0 when R +.
Also:

g(z)dz

||
t
e
cos
0 when 0.
Finally, on the line segment [Re
i
, ], we can write z = (+is)u with u > 0,
= cos and s = sin . Then, we have:
_
[Re
1
,]
g(z)dz = ( +is)
t
_
R

u
t1
e
(+is)u
du
_
[,R]
g(z)dz =
_
R

x
t1
e
x
dx.
Applying Cauchys theorem and taking the limits R + and 0,
we get:
( +is)
t
_
+
0
u
t1
e
(+is)u
du +
_
+
0
x
t1
e
x
dx = 0.
The last term is just (t), and we have:

f(s) =

t
(t)
_
+
0
u
t1
e
(+is)u
du,
so that:

f(s) =
_

+is
_
t
.
Application to dierential equations
We will conclude this subsection by two examples that illustrate how to use
the Fourier transform in solving dierential equations.
5.2. THE FOURIER TRANSFORM 197
Example 31. An ordinary dierential equation Consider f : R R
such that f, f

and f

all belong to I(R). Suppose that f, f

and f

all
tend to zero as |x| +. Consider the ordinary dierential equation:
x R, f

(x) f(x) = e
x
2
.
If we apply the Fourier transform to both sides of the equation, we obtain:
s
2

f(s)

f(s) = F
_
e
x
2
_
.
Hence:

f(s) = F
_
e
x
2
_
1
1+s
2
. If we call g(s) = 1/(1 + s
2
), we have (by
using the same method as for the Cauchy distribution, but applied to the
inverse Fourier transform): g(x) = e
|x|
. So, we can write:

f(s) = F
_
e
x
2
_
g(s),
and using the convolution product:
f(x) =
_
+

e
y
2
g(x y)dy =
_
+

e
|yx|y
2
dy.
Example 32. A partial dierential equation: Laplace equation in
a half-plane.
Let u(x, y) be dened and continuous on {(x, y) R
2
, y 0}. Suppose that
it satises:
u
xx
+u
yy
= 0;
x R, u(x, 0) = f(x) where f is integrable on R.
Such a formulation is called a boundary value problem. We shall solve
it for a suitable behavior of u for large values of r =
_
x
2
+y
2
, i.e. at
innity.
For a xed value of y, we perform the Fourier transform of u on the variable
x:
u(s, y) =
_
+

u(x, y)e
isx
dx.
198 CHAPTER 5. APPLICATIONS
The partial dierential equation then becomes an ordinary dierential equa-
tion:
d
2
u
dy
2
= s
2
u.
The boundary condition becomes u(s, 0) =

f(s). The general solution of
the ordinary dierential equation is given by:
u(s, y) = Ae
sy
+Be
sy
,
where A and B are real constants. If we want a solution that decays at
innity, certainly u(s, y) 0 when y +, so that A = 0 if s > 0 and
B = 0 if s < 0, so, we have, taking into account the bounday condition:
u(s, y) =

f(s)e
|s|y
.
Using the convolution, we nally get:
u(x, y) =
y

_
+

f(t)
(x t)
2
+y
2
dt.
This solution is called the Poisson integral for the half-plane.

You might also like