Analytic function theory

Basic analysis in one complex variable

Lecture notes

Håkan Samuelsson Kalm

Analytic function theory 3

TABLE OF CONTENT

Preface
1 Möbius transformations and the Riemann sphere

1.1 Circles and lines

1.2 Möbius transformations
1.3 Visualizing Möbius transformations of the Riemann sphere

2 Two-variable real calculus in complex notation

2.1 Differentiability
2.2 Curves and contour integrals

3 Holomorphic functions and Cauchy’s theorem and formula

3.1 Definition and basic properties of holomorphic functions

3.2 Cauchy’s theorem and formula

4 Series and power series

4.1 Series and convergence criteria

4.2 Functions defined by power series
4.3 Holomorphic functions are power series locally

5 Elementary functions via power series

5.1 Exponential functions

5.2 Trigonometric and hyperbolic functions
5.3 Logarithms, arguments, and general powers

6 Liouville’s theorem and the Fundamental theorem of algebra

6.1 Liouville’s theorem

6.2 The Fundamental theorem of algebra

7 Conformal mappings

7.1 Angles between curves

7.2 Constructions of conformal mappings

8 Morera’s theorem and Goursat’s theorem

8.1 Morera’s theorem

8.2 Goursat’s theorem

9 Zeros of holomorphic functions

9.1 Zeros, their orders, and identity theorems

Analytic function theory 4

9.2 Analytic continuation

9.3 Counting the number of zeros

10 Some topology and jazzed-up Cauchy theorems

10.1 Deformation and homotopy

10.2 Winding numbers and homology
10.3 Properties of simply connected open sets

11 Some mapping properties of holomorphic functions

11.1 The Maximum principle and Schwarz’s lemma

11.2 Three basic mapping theorems and the Riemann mapping theorem

12 Singularities and Cauchy’s residue theorem

12.1 Laurent series and classification of singularities

12.2 Residue theorems

13 Calculating real integrals using complex analysis

14 Harmonic functions and the Dirichlet problem

14.1 Harmonic vs. holomorphic functions

14.2 Poisson’s integral formula and some consequences
14.3 The Dirichlet problem and the Riemann mapping theorem

Appendix A
A.1 Complex numbers
A.2 Arithmetics of complex numbers
A.3 Polynomials

Appendix B
B.1 Open and closed subsets of C and of the Riemann sphere
B.2 Some topology and analysis in C and on the Riemann sphere

Bibliography
Analytic function theory 5

Preface
These are the lecture notes of the course “Analytic function theory” (7.5hec), which is given
annually withing the Mathematics Programme at the University of Gothenburg. The course
consists of ca. 21 lectures à 2 × 45 minutes given during 8 weeks. The notes are inspired by the
books [1], [2], and [3] by M. Andersson, L. Hörmander, and H. A. Priestley, respectively.
We base the theory on real calculus, in particular we assume knowledge about differentia-
bility, partial derivatives, line integrals, Green’s formula, and some basic operative knowledge
about differential forms. One difference compared to several text books on complex analysis
(but not [1] and [2]) is that we assume C 1 -smoothness in the definition of a holomorphic func-
tion. The reason is that we want to apply calculus techniques to get to Cauchy’s theorem and
formula in as transparent a way as possible, and then use it as the foundation for the theory.
It is an interesting fact that the assumption of C 1 -smoothness is in fact superfluous, and we
will indeed show this in Chapter 8, but in my opinion not important enough to sacrifice an
efficient development of the theory.
I stress that these are lecture notes; they are intended as a complement to the lectures, not
a substitute. In fact, the information density is rather high and sometimes the style almost
borders that of a mathematical research text. Students in the beginning of their mathematics
studies might therefore find these notes hard to read without the guidance of a teacher. This is
not necessarily a bad thing as it hopefully promotes the important physical meeting between
student and teacher.

Informal introduction
Complex analysis in one variable is about functions depending on two real variables in a certain
way; the dependence of the two real variables is not arbitrary but only “certain combinations”
of the two real variables are allowed. To illustrate, consider a polynomial
X
p(x, y) = ck` xk y ` , ck` ∈ C, (0.1)
0≤k,`≤n

of the two real variables x and y. If we set z := x+iy and z̄ := x−iy, we can write x = (z + z̄)/2
and y = (z − z̄)/2i. Substituting this in (0.1) we get
X
p(z, z̄) = ak` z k z̄ `
0≤k,`≤n

for some new (complex) constants ak` . It might happen that ak` = 0 for ` > 0; this means
that p(z, z̄), so to say, does not depend on z̄. In this case we say that the polynomial p defines
a holomorphic function. In general, a holomorphic function f is a C 1 -function defined in an
∂f
open subset of C such that = 0, which is a way of expressing that f does not depend on z̄.
∂ z̄
Complex analysis is about the study of holomorphic functions. It is hopefully clear that quite
few of the functions and polynomials of two real variables are holomorphic.

Example 0.1. The polynomial p1 (x, y) = x2 + 2ixy − y 2 defines a holomorphic function since
x2 + 2ixy − y 2 = (x + iy)2 = z 2 while the polynomial p2 (x, y) = x2 + y 2 = (x + iy)(x − iy) = z z̄
does not.

The seemingly innocuous condition that a function f in C should be independent of z̄

implies that f has very particular and surprising properties. The purpose of the course is to
reveal the basic such properties.
Analytic function theory 6

1 Möbius transformations and the Riemann sphere

Circles and lines in the plane C ' R2 are intimately connected. The geometrical reason for this
is the possibility to view C as a sphere, the so called Riemann sphere, with one point removed.
Lines and circles in C are then just circles on the Riemann sphere, where the circles going
through the removed point correspond to the lines in C.
A Möbius transformation is an angle-preserving transformation of the Riemann sphere.
Therefore, if C is any circle on the Riemann sphere and f is a Möbius transformation, then
f (C) is again a circle on the sphere. The interpretation of this in the plane C is that any circle
or line in C can be mapped to any other circle or line by a Möbius transformation.

1.1 Circles and lines

A line ` in C can be described in several ways. For instance, if z0 and w0 are two different
points on `, then ` is the set of complex numbers of the form z0 + t(w0 − z0 ) where t is a real
number; if z0 = x0 + iy0 and w0 = u0 + iv0 , then this is just the usual parametrization of the
line in R2 through (x0 , y0 ) and (u0 , v0 ) written in complex notation. Another way to describe
` is first to pick a line `⊥ orthogonal to ` and two points α and β on `⊥ with the same distance
to `. Then ` is the set of points z such that the distances |z − α| and |z − β| are the same, i.e.,

` = {z ∈ C; |z − α| = |z − β|} = {z ∈ C; |z − α|/|z − β| = 1}.

Circles in C may also be described in different ways. Given the center a and the radius r
the circle is conveniently described as the set of z ∈ C such that |z − a| = r, and the circle can
be parametrized by the interval [0, 2π) using the map

t 7→ a + reit = Re a + r cos t + i(Im a + r sin t).

However, a circle can also be described in a, at least at first sight, less intuitive way. Let
α, β ∈ C and λ ∈ [0, ∞) \ {1} be given. One can verify (do it, it’s a good check that you’re
fluent in the computation rules in Appendix A) that the equation
 
z − α
z − β  = λ (1.1)
 

is equivalent to the equation

2
 
z − λ β − α  = λ|α − β| .
 
2
(1.2)
 λ −1  |λ2 − 1|
This means that (1.1) in fact is the equation of a circle with center (λ2 β − α)/(λ2 − 1) and
radius λ|α−β|/|λ2 −1|. Conversely, by Exercise 1.2, any circle may be described by an equation
like (1.1).
We have thus seen that (1.1) is the equation of a line if λ = 1 and the equation of a circle
if λ ∈ [0, ∞) \ {1} (the case λ = 0 is degenerate, the corresponding circle has radius 0, i.e., it is
just a point). This observation is a first indication of the correspondence between circles and
lines.

1.2 Möbius transformations

We will now look at maps that preserve the set of all circles and lines. For any a ∈ C \ {0} and
any b ∈ C the maps
f1,a (z) = az and f2,b (z) = z + b
Analytic function theory 7

map circles to circles and lines to lines since f1,a is a scaling and a rotation about 0 of C and
f2,b is a translation. To be able to change a circle into a line, and vice versa, we introduce
f3 (z) = 1/z. A priori this is not defined for z = 0 but we give it a meaning as follows. Let
b := C ∪ {∞}.
C

This requires some explanations; “∞” is just a (suggestive) notation of an abstract point and
b is the set of all complex numbers together with this abstract point. (We don’t give any
C
geometric interpretation of C
b yet, at this point it is just a set.) We define f3 to be the map
C → C given by
b b

f3 (0) = ∞, f3 (∞) = 0, f3 (z) = 1/z for z ∈ C \ {0}.

b by setting f1,a (∞) = ∞ and f2,b (∞) = ∞.
We also extend the maps f1,a and f2,b to be maps of C
1
In this way we have obtained bijective maps , or automorphisms, f1,a , f2,b , f3 of Cb and their
inverses are of the same type (what are the inverses?). Clearly compositions of such maps are
again automorphisms of C b and we obtain a group of automorphisms where the group operation
is composition. In view of Exercise 1.3, any composition of maps of the types f1,a , f2,b , and f3
is of the form  αz+β
 γz+δ , z 6= −δ/γ, z 6= ∞

f (z) = ∞ , z = −δ/γ (1.3)

α
γ , z = ∞,
where α, β, γ, δ ∈ C and αδ − βγ 6= 0. Conversely, see Exercise 1.4, any map of the form (1.3)
is a composition of maps of the types f1,a , f2,b , and f3 . The automorphisms of C
b of the form
(1.3) are called Möbius transformations.

Theorem 1.1. Möbius transformations preserve the set of circles and lines in C.

Proof. Since any Möbius transformation is a composition of maps of the types f1,a , f2,b , and
f3 , and maps of the types f1,a and f2,b preserve the set of circles and lines, it is sufficient to
check that the map f3 preserves the set of circles and lines.
Let C be a circle or line. We know that it can be described by an equation |z −α|/|z −β| = λ
for some α, β ∈ C and λ > 0. Then

f3 (C) = {w = 1/z; |z − α|/|z − β| = λ}

 
|1/w − α|
= w; =λ .
|1/w − β|

But
|1/w − α| |1 − αw| |α| |w − 1/α|
= =
|1/w − β| |1 − βw| |β| |w − 1/β|
so f3 (C) is described by the equation |w − 1/α|/|w − 1/β| = λ|β|/|α|, i.e., it is a circle or a line.
The cases α = 0 or β = 0 may be checked separately in a similar way.

1.3 Visualizing Möbius transformations of the Riemann sphere

Let Σ be the sphere {(x, y, z) ∈ R3 ; x2 + y 2 + z 2 = 1} and let P be the xy-plane, i.e., P =
{(x, y, z) ∈ R3 ; z = 0}. We denote the north pole (0, 0, 1) by n and the south pole (0, 0, −1)
by s; we will define two stereographic projections, πn : Σ \ {n} → P and πs : Σ \ {s} → P .
1
These are in fact continuous in the sense described in Appendix B.
Analytic function theory 8

Given (x, y, z) ∈ Σ\{n} consider the line in R3 through n and (x, y, z); it can be parametri-
zed by R using the map t 7→ (tx, ty, 1 + t(z − 1)). This line intersects the plane P in a point
that we denote by πn (x, y, z). We can obtain an explicit expression for πn (x, y, z) as follows.
The line intersects P when t is such that 1 + t(z − 1) = 0, that is when t = 1/(1 − z). Putting
this value of t into the parametrization of the line we see that


πn (x, y, z) = x/(1 − z), y/(1 − z), 0 .

Similarly, given (x, y, z) ∈ Σ \ {s} we define πs (x, y, z) to be the point of intersection between
P and the line through s and (x, y, z), and we get


πs (x, y, z) = x/(1 + z), y/(1 + z), 0 .

One can check (Exercise 1.5) that the inverses of πn and πs are given by

x2 + y 2 − 1
 
−1 2x 2y
πn (x, y, 0) = , , and (1.4)
1 + x2 + y 2 1 + x2 + y 2 1 + x2 + y 2

−x2 − y 2 + 1
 
2x 2y
πs−1 (x, y, 0) = , , , (1.5)
1 + x2 + y 2 1 + x2 + y 2 1 + x2 + y 2
respectively.
We can identify the plane P and C in two different ways using the maps id : P → C and
id : P → C defined by id(x, y, 0) = x+iy and id(x, y, 0) = x−iy, respectively. For w = x+iy 6= 0
we now have
|w|2 − 1
 
−1 −1 −1 2x 2y
id ◦ πs ◦ πn ◦ id (w) = id ◦ πs ◦ πn (x, y, 0) = id ◦ πs , ,
1 + |w|2 1 + |w|2 1 + |w|2
 
x y x y w̄
= id , ,0 = −i 2 =
|w|2 |w|2 |w|2 |w| |w|2
1
= .
w

We thus have a geometric interpretation of C b as a sphere, which is called the Riemann

sphere, and an interpretation of the mapping f3 . Moreover, since we already have geometric
interpretations of the mappings f1,a and f2,b , defined above, and any Möbius transformation
is a composition of such maps and f3 -maps, we have a way of thinking about any Möbius
transformation as an automorphism of the Riemann sphere.

Exercises
1.1 Show that Equation (1.1) and Equation (1.2) are equivalent.

1.2 Let a ∈ C and r ≥ 0. Show that there are α, β ∈ C and λ ∈ [0, ∞) \ {1} such that

λ2 β − α λ|α − β|
a= and r = .
λ2 − 1 |λ2 − 1|

1.3 Show, for instance using induction, that any composition of maps of the types f1,a , f2,b ,
and f3 is of the form (1.3).

1.4 Show that a Möbius transformation is a composition of maps of the type f1,a , f2,b , and
f3 .
Analytic function theory 9

1.5 Check that the inverses of πn and πs are given by (1.4) and (1.5), respectively.

1.6 Compute f3 (`) where ` is the line {1 + it; t ∈ R}.

1.7 Find a Möbius transformation that maps the real line to the unit circle.

1.8 Let f (z) = (z − 1)/(z + 1). Find the fixed points of f , i.e., find all points z on C
b such
that f (z) = z. Compute the images of the unit circle and the real axis respectively.

1.9 Sketch the set of points z ∈ C such that

(a) |z + i| = 2, (b) |z| = |z − 1|, (c) |z − eiπ/4 | = |z − ei5π/4 |.

1.10 Sketch the set of points z ∈ C such that

√ √
(a) |z + 1 − i)| > 1, (b) |z + 1| < 2 and |z − 1| < 2, (c) − 3 < Re z < −1,
√
|z − 2| √
(d) 1 < |z − i| < 2, (e) |z|2 − 2Re z > 0 (f) √ = 2.
|z − 1/ 2|

1.11 Find a Möbius transformation that maps 0, 1, and ∞ to 0, 1 + i, and 2, respectively.

What is the image of

(a) R, (b) iR, (c) the unit circle?

Analytic function theory 10

2 Two-variable real calculus in complex notation

We will recall some results and notions from calculus in two real variables and rewrite them in
a notation that is suitable for complex analysis.

2.1 Differentiability
Let f be a (possibly complex-valued) function defined in an open subset Ω ⊂ C ' R2 . Recall
from calculus that f is differentiable at a point z0 = x0 +iy0 ∈ Ω if there are (possibly complex)
numbers A and B such that
p
f (x, y) = f (x0 , y0 ) + A(x − x0 ) + B(y − y0 ) + o( (x − x0 )2 + (y − y0 )2 ). (2.1)

If there are such A and B then the partial derivatives of f at the point (x0 , y0 ) exist and
∂f ∂f
A= (x0 , y0 ), B = (x0 , y0 ). By setting C := (A − iB)/2 and D := (A + iB)/2 we can
∂x ∂y
rewrite (do it!) (2.1) as

f (z) = f (z0 ) + C(z − z0 ) + D(z̄ − z̄0 ) + o(|z − z0 |).

Thus, if we define partial derivatives with respect to z and z̄ by setting

∂f 1  ∂f ∂f 
(z0 ) := (x0 , y0 ) − i (x0 , y0 ) , (2.2)
∂z 2 ∂x ∂y
∂f 1  ∂f ∂f 
(z0 ) := (x0 , y0 ) + i (x0 , y0 ) , (2.3)
∂ z̄ 2 ∂x ∂y
then (2.1) may be written as

∂f ∂f
f (z) = f (z0 ) + (z0 )(z − z0 ) + (z0 )(z̄ − z̄0 ) + o(|z − z0 |). (2.4)
∂z ∂ z̄
Proposition 2.1. Let f : Ω → C be differentiable at z0 ∈ Ω. Then

f (z0 + h) − f (z0 )
lim (2.5)
h→0 h
∂f ∂f
exists if and only if (z0 ) = 0. Moreover, if (2.5) exists then it is equal to (z0 ).
∂ z̄ ∂z
∂f
Proof. Assume first that (z0 ) = 0. Setting z = z0 + h in (2.4) we get
∂ z̄
f (z0 + h) − f (z0 ) ∂f o(|h|)
= (z0 ) + ,
h ∂z h
∂f
and since o(|h|)/h → 0 as h → 0 we see that (2.5) exists and equals (z0 ).
∂z
Conversely, assume that (2.5) exists. Setting z = z0 + h in (2.4) and rewriting we get

∂f h̄ f (z0 + h) − f (z0 ) ∂f o(|h|)

(z0 ) = − (z0 ) + .
∂ z̄ h h ∂z h
By assumption the limit as h → 0 of the right-hand side exists and so the limit of the left-
hand side exists; recall (see, e.g., Appendix B) that this means that for every sequence {hn }
Analytic function theory 11

∂f h̄n
of complex numbers converging to 0, the sequence (z0 ) converges to one fixed number.
∂ z̄ hn
If we let h → 0 along the real axis, e.g., we choose hn = 1/n, then

∂f h̄n ∂f
(z0 ) → (z0 ), n → ∞.
∂ z̄ hn ∂ z̄
On the other hand, if let h → 0 along the imaginary axis, e.g., choosing hn = i/n, then

∂f h̄n ∂f
(z0 ) → − (z0 ), n → ∞.
∂ z̄ hn ∂ z̄
∂f ∂f ∂f
Hence, (z0 ) = − (z0 ) and so (z0 ) = 0.
∂ z̄ ∂ z̄ ∂ z̄
A function f : Ω → C such that the limit (2.5) exists is said to be complex differentiable at
z0 . We notice that if f is complex differentiable at z0 then f is continuous at z0 , Exercise 2.4. We
notice also that the proposition above in this terminology says that a differentiable function is
complex differentiable if and only ∂f /∂ z̄ = 0. We will see in Theorem 3.5 below that a function
f that is complex differentiable, but a priori not necessarily differentiable, in fact still satisfies
∂f /∂ z̄ = 0.

2.2 Curves and contour integrals

Let Ω ⊂ C ' R2 be open. By a parametrized curve in Ω we mean a continuous map γ : I → Ω,
where I ⊂ R is an interval. A parametrized curve γ is said to be simple if either γ is injective
or if I = [a, b], γ(a) = γ(b), and γ is injective on (a, b). A parametrized curve γ is closed if
I = [a, b] is a closed bounded interval and γ(a) = γ(b). A subset C of Ω is called a curve in Ω
if it is the image of a parametrized curve in Ω, i.e., if C = γ(I) for some parametrized curve γ
in Ω. A curve is simple and closed, respectively, if it is the image of some parametrized curve
which is simple and closed, respectively. We will refer to a parametrized curve corresponding
to a curve C as a parametrization of C.
General curves may look very non-curve-like, for instance the so called Peano curve (google
it) fills the 2-dimensional square [0, 1] × [0, 1]. In these notes such curves will not be of in-
terest and we always assume some extra conditions to ensure that our curves are curve-like.
Specifically, we often assume that our parametrized curves are C 1 -smooth, or simply smooth,
meaning that γ is continuously differentiable, or at least that they are piecewise C 1 -smooth.
This means that there are points · · · tk−1 < tk < tk+1 < · · · in I such that the restriction of γ
to the sub-interval (tk , tk+1 ) of I is C 1 -smooth for all k. We say that a subset C of Ω is a C 1 -
smooth (piecewise C 1 -smooth) curve if C = γ(I) for some C 1 -smooth (piecewise C 1 -smooth)
parametrized curve γ.

Example 2.2. Let γ(t) = eit = cos t + i sin t, 0 ≤ t ≤ 2π. Then γ is a smooth closed simple
curve in C. The image of γ is the unit circle {z ∈ C; |z| = 1}, which is smooth, closed, and
simple.

Example 2.3. The set {z = x + iy ∈ C; y = 0, −1 ≤ x ≤ 1} ∪ {z ∈ C; |z| = 1, Im z > 0},

which is the upper part of the unit circle together with the interval [−1, 1], is a simple closed
piecewise smooth curve.

A simple closed curve in C divides C into two connected pieces, one bounded piece, called
the interior of the curve, and one unbounded. This follows in general from the Jordan curve
theorem which is a rather deep result. However, for most curves appearing in these notes this
Analytic function theory 12

will be obvious. An orientation of a curve C is a choice of traveling direction along C. A simple

closed curve C is said to be positively oriented if the traveling direction is such that one has
the interior of C on the left-hand side; it is said to be negatively oriented if the interior is on
the right-hand side. When we say that γ : I → C is a parametrization of an oriented curve C
we always mean that γ(t) runs through C in the direction given by the orientation as t runs
through I from left to right.
Let C ⊂ C ' R2 be a C 1 -smooth simple oriented curve and let P and Q be continuous
functions defined on C. We recall from calculus that the line integral of the differential form2
P dx + Qdy is defined as follows. Let γ : I → R2 be a simple C 1 -smooth parametrization of C
(tacitly assumed compatible with the orientation). We write γ(t) = (x(t), y(t)). Then
Z Z
P γ(t) x0 (t) + Q γ(t) y 0 (t) dt.




P dx + Qdy := (2.6)
C t∈I
This is well defined since we know from calculus that the right-hand side is independent of
the choice of simple parametrization γ of C. More generally, if C is a piecewise smooth curve,
which is also piecewise oriented, then we divide C into pairwise disjoint curve-pieces Cj that
are smooth, simple, and oriented, and we set
Z XZ
P dx + Qdy := P dx + Qdy.
C j Cj

Let us rewrite line integrals in R2 ' C in complex notation. If z = x + iy then z̄ = x − iy

and it is natural to define
dz := dx + idy and dz̄ := dx − idy. (2.7)
It follows that dx = (dz + dz̄)/2 and dy = (dz − dz̄)/2i and a simple computation shows that
we can write a differential form P dx + Qdy as
1 1
P dx + Qdy = (P − iQ)dz + (P + iQ)dz̄.
2 2
Similarly, a differential form f dz + gdz̄ can be expressed in terms of dx and dy as
f dz + gdz̄ = (f + g)dx + i(f − g)dy (2.8)
In particular, if f is a differentiable function we have by (2.2) and (2.3) that
∂f ∂f ∂f ∂f
dz + dz̄ = dx + dy, (2.9)
∂z ∂ z̄ ∂x ∂y
cf. Exercise 2.7 below. In view of (2.8) it is natural to define the contour integral of f dz + gdz̄
along an oriented curve C by
Z Z
f dz + gdz̄ := (f + g)dx + i(f − g)dy,
C C

where the right-hand side is defined by (2.6). If γ(t) = z(t) = x(t) + iy(t), t ∈ I, is a C 1 -smooth
parametrization of C we thus get
Z Z
f (z(t)) + g(z(t)) x0 (t) + i f (z(t)) − g(z(t) )y 0 (t) dt




f dz + gdz̄ =
C
Zt∈I
f (z(t)) x0 (t) + iy 0 (t) + g(z(t)) x0 (t) − iy 0 (t) dt




=
Zt∈I
f (z(t))z 0 (t) + g(z(t))z̄ 0 (t) dt.


=
t∈I
2
We presume some basic operative knowledge about differential forms.
Analytic function theory 13

In particular, Z Z
f dz = f (z(t))z 0 (t)dt.
C t∈I

Example 2.4. Let C = {z ∈ C; |z| = r} with positive orientation (i.e., run through counter-
clockwise)
Z and let f (z) = 1/z; notice that f is defined for z ∈ C. Compute the contour integral
f dz.
C
Solution: Choose γ(t) = reit = r cos t + i sin t, 0 ≤ t ≤ 2π, as simple parametrization of C. We
then have
Z Z Z 2π Z 2π

0 0
f γ(t) γ 0 (t)dt



f dz = f dx + if dy = f γ(t) x (t) + iy (t) dt =
C C 0 0
Z 2π Z 2π
−r sin t + ir cos t
= dt = idt = 2πi.
0 r cos t + ir sin t 0

This computation can be made shorter and more suggestive by noting that if z = reit , then
dz = ireit dt (show this!), and so
2π 2π
ireit dt
Z Z Z
f dz = = idt = 2πi.
C 0 reit 0

We now recall Green’s formula from calculus; it relates the line integral of a differential
form along the boundary of an open set Ω to an integral over Ω. More precisely, we have

Theorem 2.5 (Green’s formula). Let Ω be a bounded open subset of R2 ' C and assume
that its boundary, ∂Ω, is a piecewise C 1 -smooth curve, positively oriented. Let P and Q be
continuously differentiable functions defined in some open set U ⊃ Ω. Then
Z ZZ  
∂Q ∂P
P dx + Qdy = − dxdy.
∂Ω Ω ∂x ∂y

In particular, if f is a continuously differentiable function in U ⊃ Ω it follows that

Z ZZ
∂f
f dz = 2i dxdy. (2.10)
∂Ω Ω ∂ z̄

In fact, by Green’s formula and (2.3) we have

Z Z ZZ  
∂f ∂f
f dz = f dx + if dy = i − dxdy
∂Ω ∂Ω Ω ∂x ∂y
ZZ   ZZ
1 ∂f ∂f ∂f
= 2i +i dxdy = 2i dxdy.
Ω 2 ∂x ∂y Ω ∂ z̄

Remark 2.6. If ∂Ω is a union of piecewise smooth positively oriented curves ∂Ωj , then the
conclusion still holds if we construe the left-hand side as the sum of integrals over the boundary
parts ∂Ωj .

Let γ : I →
R C0 be a smooth simple parametrized curve and assume that I is bounded. Then
the number I |γ (t)|dt is independent of such parametrization and thus only depends on the
curve C := γ(I). We call this number the length of C and denote it by `(C). We define the
length of a piecewise smooth curve by dividing it into smooth simple pieces and sum up the
lengths of these.
Analytic function theory 14

Proposition 2.7. Let C ⊂ C be a piecewise smooth oriented curve and let f be a continuous
function defined on C. Then
Z 
 
 f (z)dz  ≤ sup |f (z)| · `(C).
 
C z∈C

Proof. Possibly after dividing C into suitable pieces we may assume that C is given by a smooth
simple parametrized curve γ : I → C. Then we have
Z  Z  Z Z
 f (z)dz  =  f γ(t) γ 0 (t)dt ≤ f γ(t) γ 0 (t)dt ≤ sup f γ(t)  |γ 0 (t)|dt
  
 
 

   
C I I t∈I I
= sup |f (z)| · `(C).
z∈C

The second inequality is well-known from calculus as is the first, at least if f γ(t) γ 0 (t) is real

for all t. A proof that it holds in general is outlined in Exercise 2.5.

Exercises
2.1 Let f and g be differentiable functions. Show that
∂
∂f ∂g ∂
∂f ∂g
(a) af + bg = a +b and af + bg = a +b for all complex constants
∂z ∂z ∂z ∂ z̄ ∂ z̄ ∂ z̄
a and b,
∂
∂g ∂f ∂
∂g ∂f
(b) f ·g =f +g and f ·g =f +g ,
∂z ∂z ∂z ∂ z̄ ∂ z̄ ∂ z̄
∂  f  g∂f /∂z − f ∂g/∂z ∂  f  g∂f /∂ z̄ − f ∂g/∂ z̄
(c) = and = if g 6= 0.
∂z g g2 ∂ z̄ g g2
∂f ∂f
2.2 Compute and when
∂z ∂ z̄
(a) f (z) = z, (b) f (z) = z̄, (c) f (z) = az k z̄ ` .

∂ f¯ ∂ f¯
   
∂f ∂f
2.3 Show that = and = .
∂ z̄ ∂z ∂z ∂ z̄
2.4 Show that if f is complex differentiable at z0 , then f is continuous at z0 .

2.5 Let I ⊂ R and let g : I → C be a function. The purpose of this exercise is to outline a
proof of the inequality Z  Z
 
 g(t)dt ≤ |g(t)|dt (2.11)
 
I I
given that we know the corresponding inequality for real-valued functions.
R R
(a) Assume that I g(t)dt is a real number, i.e., that I Im g(t)dt = 0, and show (2.11)
in this case using computation rule (12) in Appendix A.
(b) Show that there is a θ ∈ R such that eiθ I g(t)dt is a real number and conclude
R

(2.11) using part (a).

Z
2.6 Compute the integral f (z)dz, where f (z) = Re z and C is the oriented curve given by
C
the parametrization γ(t) = t + it2 , 0 ≤ t ≤ 1. Notice that f is real-valued; is the integral
a real number?
Analytic function theory 15

2.7 Let γ : I → C be a smooth parametrized curve and let f be a differentiable function in

an open set containing γ(I). Set ψ(t) = f ◦ γ(t) and use the Chain rule of calculus to
∂f ∂f
show that ψ 0 (t) = γ(t) · γ 0 (t) +



γ(t) · γ 0 (t).
∂z ∂ z̄
2.8 Let C be a piecewise smooth oriented curve of finite length in C and let fn , n = 1, 2, . . .,
be continuous functions on C such that fn (z) converges uniformly on C to some function
f (z). Show, e.g., using Proposition 2.7, that
Z Z
lim fn (z)dz = f (z)dz.
n→∞ C C

(
2πi if k = −1,
Z
2.9 Show that (z − a)k dz = ; cf. Example 2.4.
∂D(a,r) 0 if k 6= −1

2.10 Are the following functions differentiable at z = 0?

(a) f (z) = z + z̄, (b) f (z) = |z|2 (c) f (z) = |z|, (d) z̄ 2 .
Z
2.11 Calculate f (z)dz when
C

(a) f (z) = |z|4 and C is the line segment starting at 1 − i and ending at 1 + i,
(b) f (z) = Re z ·Im z and C is the upper half of the unit circle oriented counterclockwise.
Z 
 z+1  R+1
2.12 (a) Show that if R 6= 1 then  dz ≤ · 2πR.
 
 |z|=R z 2 + 1  |R2 − 1|
Z
dz
(b) Show that lim 2
= 0.
R→∞ |z|=R z + 1
Analytic function theory 16

3 Holomorphic functions and Cauchy’s theorem and formula

We define a holomorphic function as a C 1 -smooth function that is complex differentiable and
we show some basic properties. We also show that complex differentiability alone implies that
the Cauchy-Riemann equation(s) are satisfied. We then prove Cauchy’s theorem using Green’s
formula and show how Cauchy’s formula follows from his theorem.

3.1 Definition and basic properties of holomorphic functions

Recall that a function f defined in an open subset Ω of R2 ' C is C 1 -smooth in Ω, f ∈ C 1 (Ω),
if the partial derivatives ∂f /∂x and ∂f /∂y exist at each point z ∈ Ω and depend continuously
on z. From calculus we know that if f is C 1 -smooth in Ω then f is in particular differentiable
at each point of Ω.
Definition 3.1. Let Ω be an open subset of C and let f : Ω → C be a function. Then f is
holomorphic in Ω if f ∈ C 1 (Ω) and f is complex differentiable at each point of Ω, i.e., if the
limit (2.5) exists for each z0 ∈ Ω.
Notice that, by Proposition 2.1, a C 1 -smooth function f is holomorphic if and only if
∂f /∂ z̄ = 0. Moreover, if f is holomorphic then the limit (2.5) is equal to ∂f /∂z at the point
z0 ; we will then usually write f 0 instead ∂f /∂z and refer to f 0 as the derivative of f .
Remark 3.2. The requirement f ∈ C 1 (Ω) is actually superfluous and indeed follows if f is
assumed complex differentiable at each point. This is an interesting fact that we will prove
below (Goursat’s theorem) but, in my opinion, not so important for complex analysis since
checking C 1 -smoothness often is easier than checking that (2.5) exists everywhere.
Let f and g be holomorphic functions in an open set Ω ⊂ C. Then af +bg is holomorphic in
Ω for all constants a and b, f g is holomorphic in Ω, and f /g is holomorphic in Ω if g is non-zero.
Indeed, it is clear that af + bg, f g, and f /g are C 1 -smooth and since ∂f /∂ z̄ = ∂g/∂ z̄ = 0 it
follows that ∂(af + bg)/∂ z̄ = 0, that ∂(f g)/∂ z̄ = 0, and that ∂(f /g)/∂ z̄ = 0, see Exercise 2.1.
Example 3.3. The function f (z) = z is holomorphic in C since it is obviously C 1 , and complex
differentiability is straightforward to check.
P (It is also straightforward to check that ∂f /∂ z̄ = 0,
cf. Exercise 2.2.) Thus, any polynomial nk=0 ak z k is holomorphic in C. Moreover, if p(z) and
q(z) are polynomials, then the rational function p(z)/q(z) is holomorphic in C \ {z; q(z) = 0}.
If f is holomorphic in Ω and g is holomorphic in an open set containing the image f (Ω), then
the composition g ◦ f is holomorphic in Ω. In fact, it is known from calculus that g ◦ f ∈ C 1 (Ω)
and as in one-variable real calculus one can show that g ◦ f is complex differentiable at each
point of Ω. A different argument is suggested in Exercise 3.1 below which also shows that we
have the following version of the Chain rule for holomorphic functions:

(g ◦ f )0 (z) = g 0 f (z) · f 0 (z).

The following result can be seen as a version of the Fundamental theorem of calculus for
holomorphic functions.
Proposition 3.4. Let f be holomorphic in an open setZ Ω and let C be an oriented piecewise
smooth curve in Ω starting at a and ending at b. Then f 0 (z)dz = f (b) − f (a).
C
Z
Notice in particular that the contour integral f 0 (z)dz only depends on the start and end
C
points of C, not on the particular curve connecting these points.
Analytic function theory 17

Proof of Proposition 3.4. Possibly after dividing C into pieces we may assume that we have a
C 1 -smooth parametrization γ : I → C of C. Write γ(t) = x(t) + iy(t) and let t0 and t1 be the
left and right endpoints, respectively, of I. Now, since f is holomorphic we have, in view of
(2.9), that f 0 (z)dz = ∂f /∂x dx + ∂f /∂y dy and so
Z Z Z t1   Z t1
0 ∂f ∂f ∂f 0 ∂f 0 d

f (z)dz = dx + dy = x (t) + y (t) dt = f γ(t) dt
C C ∂x ∂y t ∂x ∂y t0 dt


0
= f γ(t1 ) − f γ(t0 ) = f (b) − f (a).

The third equality follows from the chain rule of calculus (cf. Exercise 2.7) and the
fourth
equality is the Fundamental theorem of calculus applied to the function g(t) = f γ(t) .

One indication that the existence of the limit (2.5) alone has strong consequences is the
next result; it shows that a function f that is complex differentiable at a point, but a priori
not necessarily differentiable, satisfies ∂f /∂ z̄ = 0, cf. Proposition 2.1 above and the comment
following its proof.

Theorem 3.5 (The Cauchy-Riemann equations). Let f be a function that is complex differen-
tiable at a point z0 and write f (z) = u(z) + iv(z). Then the partial derivatives ∂u/∂x, ∂u/∂y,
∂v/∂x, and ∂v/∂y exist at z0 and

∂u ∂v ∂u ∂v
(z0 ) = (z0 ), (z0 ) = − (z0 ). (3.1)
∂x ∂y ∂y ∂x

The equations (3.1) are known as the Cauchy-Riemann equations (at z0 ) and they are
equivalent to the equation ∂f /∂ z̄ = 0. To see this we recall (2.3) and compute:
   
∂f 1 ∂f ∂f 1 ∂u ∂v  ∂u ∂v 
= +i = +i +i +i
∂ z̄ 2 ∂x ∂y 2 ∂x ∂x ∂y ∂y
 
1 ∂u ∂v  ∂u ∂v 
= − +i + .
2 ∂x ∂y ∂y ∂x

Identifying real and imaginary parts we thus see that ∂f /∂ z̄ = 0 at z0 if and only if (3.1)
hold. Moreover, in view of (2.3), ∂f /∂ z̄ = 0 if and only if ∂f /∂x = −i∂f /∂y, which thus is
still another way of writing the Cauchy-Riemann equations. It follows, cf. (2.2), that if the
Cauchy-Riemann equations are satisfied, then ∂f /∂z = ∂f /∂x = −i∂f /∂y.

Proof of Theorem 3.5. We know that the limit (2.5) exists; denote it by f 0 (z0 ). By Lemma 14.10

to the function g(h) := (f (z0 + h) −
(b) applied
f (z0 ))/h, the limits limh→0 Re (f (z0 + h) −
f (z0 ))/h and limh→0 Im (f (z0 + h) − f (z0 ))/h exist and

f (z0 + h) − f (z0 ) f (z0 + h) − f (z0 )

f 0 (z0 ) = lim Re + i lim Im . (3.2)
h→0 h h→0 h


If h is
real, then Re (f (z0 + h) − f (z0 ))/h = (u(z0 + h) − u(z0 ))/h and Im (f (z0 + h) −
f (z0 ))/h = (v(z0 +h)−v(z0 ))/h, and so if we let h → 0 along the real axis, then the right-hand
side of (3.2) equals ∂u/∂x + i∂v/∂x at z0 . Hence, ∂u/∂x and ∂v/∂x exist at z0 and

∂u ∂v
f 0 (z0 ) = (z0 ) + i (z0 ). (3.3)
∂x ∂x


On the other hand, if h = it where t is
real, then Re (f (z0 + h) − f (z0 ))/h = (v(z0 +
it) − v(z0 ))/t and Im (f (z0 + h) − f (z0 ))/h = −(u(z0 + it) − u(z0 ))/t, and so if we let h → 0
Analytic function theory 18

along the imaginary axis, then the right-hand side of (3.2) equals ∂v/∂y − i∂u/∂y at z0 . Thus,
∂v/∂y and ∂u/∂y exist at z0 and
∂v ∂u
f 0 (z0 ) = (z0 ) − i (z0 ). (3.4)
∂y ∂y
Identifying real and imaginary parts in (3.3) and (3.4) the Cauchy-Riemann equations (3.1)
follow.

Proposition 3.6. Let Ω ⊂ C be an open connected set and let f be a holomorphic function in
Ω.
(a) If f 0 = 0 in Ω, then f is constant.
(b) If |f | is constant, then f is constant.
(c) If f (z) ∈ R for all z ∈ Ω, then f is constant.
Parts (b) and (c) say that if the image, f (Ω), is contained in a circle centered at 0 or in the
real axis, then f has to be a constant function. Thus, there are restrictions on what images of
holomorphic functions may look like.

Proof of Proposition 3.6. Part (a) follows from Proposition 3.4. Indeed, any two points a and b
of Ω may be connected
Z by a piecewise smooth curve C since Ω is open and connected. Therefore
f (b) − f (a) = f 0 (z)dz = 0 if f 0 = 0, and so f (a) = f (b) for any two points a and b of Ω. It
C
follows that f is constant. A slightly different argument is suggested in Exercise 3.2.
To show part (b) assume that |f (z)| = c for all z ∈ Ω. If c = 0, then f (z) = 0 for all z ∈ Ω
and we are done. Assume therefore that c 6= 0. We have c2 = |f (z)|2 = f (z)f (z) and so
∂(f f¯) ∂ f¯ ∂f
0= =f =f .
∂ z̄ ∂ z̄ ∂z
∂f
Multiplying by f¯ and using that f f¯ = c2 we then get 0 = c2 . Since c 6= 0 we must have
∂z
∂f /∂z = 0, i.e., f 0 = 0. It thus follows from part (a) that f is constant.
To show part (c) assume that f (z) is real for all z. Then f (z) = f (z) and so ∂f /∂z =
∂ f¯/∂z = (∂f /∂ z̄) = 0, see Exercise 2.3. Hence, f 0 = 0 and so f is constant by part (a).

We conclude this section by saying something about “holomorphicity at ∞”. Let f be a

holomorphic function defined in Ω = {z ∈ C; |z| > r}. If we think of Ω as a subset of the
Riemann sphere, then Ω is a “hat” on it with the north pole removed. The Möbius transfor-
mation g(z) = 1/z interchanges the north and the south poles and transforms Ω to the set
Ω̃ = {w ∈ C; 0 < |w| < 1/r}. Then f transforms to a holomorphic function f˜ defined on Ω̃
by setting f˜(w) := f (1/w). If it happens that f˜, a priori holomorphic in the punctured disc
{0 < w < 1/r}, can be extended to a holomorphic function in the disc {|w| < 1/r}, then we
say that f is holomorphic at ∞.

3.2 Cauchy’s theorem and formula

Without further ado, here is
Theorem 3.7 (Cauchy’s theorem). Let Ω ⊂ C be a bounded open set with piecewise smooth
positively oriented boundary ∂Ω. If f is a function holomorphic in an open set containing Ω,
then Z
f (z) dz = 0.
∂Ω
Analytic function theory 19

Notice that if f happens to be the derivative of a holomorphic function, then Cauchy’s

theorem follows from Proposition 3.4.

Proof of Cauchy’s theorem. From Green’s formula in complex notation, see (2.10), we have
Z ZZ
∂f
f (z) dz = 2i dxdy = 0
∂Ω Ω ∂ z̄

since f is holomorphic.

Remark 3.8. As with Green’s formula, the result also holds if ∂Ω is a union of curves ∂Ωj
provided that they satisfy the hypothesis and that we construe the integral over ∂Ω as the sum
of integrals over the oriented boundary pieces ∂Ωj .

Cauchy’s theorem is fundamental for complex analysis and is the basis of our development of
the theory. The main reason why we assume C 1 -smoothness in the definition of holomorphicity
is that we want to get to Cauchy’s theorem quickly and without ad hoc methods.
Our first application of Cauchy’s theorem is Cauchy’s formula, which provides a way of
recovering a holomorphic function in Ω from its values on ∂Ω. In particular, if two holomorphic
functions agree on ∂Ω, then they agree on Ω.

Theorem 3.9 (Cauchy’s formula). Let Ω ⊂ C and f be as in Cauchy’s theorem. Then

Z
1 f (z) dz
f (a) =
2πi ∂Ω z − a

for each point a ∈ Ω.

To prove Cauchy’s formula we need the following analysis lemma.

Lemma 3.10. Let ϕ be a continuous function defined on the disc {z ∈ C; |z| < r}. Then
Z
lim ϕ(z)dz/z = 2πi · ϕ(0).
%→0 |z|=%

Proof of Lemma 3.10. In view of Example 2.4 we have

Z Z Z
dz dz dz
ϕ(z) = (ϕ(z) − ϕ(0)) + ϕ(0)
|z|=% z |z|=% z |z|=% z
Z
dz
= (ϕ(z) − ϕ(0)) + 2πiϕ(0) =: I% + 2πiϕ(0),
|z|=% z

so if we can show that lim%→0 I% = 0, then we are done. To this end, let  > 0 be given. Since
ϕ is continuous at 0 there is some δ > 0 such that if |z| < δ then |ϕ(z) − ϕ(0)| < . Hence, if
% ≤ δ then, by Proposition 2.7,
Z   
dz   ϕ(z) − ϕ(0) 
 · `(|z| = %) =  2π% = 2π.

|I% | =  (ϕ(z) − ϕ(0))  ≤ sup 
 
 |z|=% z  |z|=% z  %

Since  > 0 is arbitrary it follows that lim%→0 I% = 0.

Analytic function theory 20

Proof of Cauchy’s formula. After a translation we may assume that a = 0. Since now 0 ∈ Ω and
Ω is open there is some % > 0 such that the closure of the disc D(0, %) = {z; |z| < %} is contained
in Ω. Then ∂Ω together with ∂D(0, %) is the boundary of an open set Ω̃ ⊂ Ω. We orient
∂D(0, %) as the boundary of D(0, %) as usual, i.e., ∂D(0, %) is oriented counterclockwise. Letting
−∂D(0, %) denote the curve with opposite orientation we have that −∂D(0, %) is oriented as
a boundary part of Ω̃. By Cauchy’s theorem and Remark 3.8 applied to Ω̃ and the function
f (z)/z we get
Z Z Z Z
f (z)dz f (z)dz f (z)dz f (z)dz
0= + = − .
∂Ω z −∂D(0,%) z ∂Ω z |z|=% z

But this holds for all sufficiently small % > 0, and so, letting % → 0 and using Lemma 3.10, we
get the desired equality.

Exercises
3.1 Let f (z) be holomorphic, let g(w) be holomorphic on the image of f , and let h = g ◦ f
be the composition. Show, using the chain rule of two-variable real calculus and the
∂h ∂h ∂g ∂f
Cauchy-Riemann equations (3.1), that = 0 and that (z0 ) = (f (z0 )) · (z0 ).
∂ z̄ ∂z ∂w ∂z
3.2 Let f be a holomorphic function in an open connected set and assume that f 0 = 0. Show
that ∂f /∂x = ∂f /∂y = 0 and conclude, using a suitable result from calculus, that f is a
constant function.

3.3 Show that if f is holomorphic at ∞ then lim|z|→∞ f (z) exists.

az + b
3.4 Show that f (z) = is holomorphic at ∞ if c 6= 0.
cz + d
Z
3.5 Compute the contour integral f (z)dz when
C

(a) f (z) = z 2 and C is the line segment from 0 to 1 + i,

Pn k
(b) f (z) = k=0 ak z and C is the upper half of the unit circle starting at −1 and
ending at 1.

3.6 Let h be holomorphic in a disc D and set k(z) := h(z).

(a) Show that k satisfies the Cauchy-Riemann equation(s) at a ∈ D if and only if

h0 (a) = 0.
(b) Show that k is holomorphic in D if and only if k (and hence also h) is constant.

3.7 Let f be holomorphic in an open connected set Ω and assume that f (Ω) is contained in
a line or in a circle. Show that f is constant.
z 2 dz
Z
3.8 Use Cauchy’s formula to compute the integral 2
.
∂D(i,1) z + 1

3.9 Let ϕ be a C 1 -smooth function in C and assume that ϕ(z) = 0 if |z| is sufficiently large.
The purpose of this exercise is to outline an argument for the equality
ZZ
1 ∂ϕ dxdy
ϕ(a) = − . (3.5)
π C ∂ z̄ z − a
Analytic function theory 21

(a) Let Ω(a, ) = C \ D(a, ). Show that

ZZ Z
∂ϕ dxdy i ϕ(z) dz
= .
Ω(a,) ∂ z̄ z − a 2 ∂D(a,) z − a

(Hint: (2.10))
ZZ
1 ∂ϕ dxdy
(b) Show (3.5), where the right-hand side is understood as lim − .
→0 π Ω(a,) ∂ z̄ z − a
(Hint: Lemma 3.10)

3.10 Let f be holomorphic in the disc D(a, R). Show that f satisfies the mean value property,
which means that f (a) is the mean value of f on the circle ∂D(a, r) for any r < R, that
Z 2π
1
is f (a) = f (a + reit ) dt for any r < R.
2π 0
(
z 5 /|z|4 if z 6= 0,
3.11 Show that f (z) = satisfies the Cauchy-Riemann equations at z = 0?
0 if z = 0
Is f differentiable at z = 0?

3.12 Let f be holomorphic in an open connected set and assume that Re f is constant. Show
that f is constant.
(Hint: The Cauchy-Riemann equations and Proposition 3.6 (a).)

3.13 The following integrals are 0. Why?

Z Z
dz dz
(a) , (b) .
|z|=1 z + 2 ∂D(i,1) z2 −z−2
Z
dz
3.14 Calculate when (a) C = ∂D(1/2, 1), (b) C = ∂D(1, 1/2), and (c) C =
C z(z 2+ 1)
∂D(2i, 3/2).

3.15 Let C = ∂D(0, 2). Compute

Z 4 Z
z +4 dz
(a) dz, (b) .
C z−i C z2 − (3 + i)z + 3i
Analytic function theory 22

4 Series and power series

Holomorphic functions behave in many ways as polynomials, at least locally. In fact, we will
see in Section 4.3 below that any holomorphic function close to any point is given by a power
series, which loosely speaking is a “polynomial of infinite degree”. The purpose of Sections 4.1
and 4.2 is to recall some theory of power series to have it at hand when needed.

4.1 Series and convergence tests

Let {ak }∞ be sequence of complex numbers. Then we can define a new sequence {sn } by
k=0 P
settingP sn = nk=0 ak . This sequence mayP or may not converge. If it converges we say that the
series ak is convergent and we P define ∞ k=0 ak := limn→∞ sn . If the sequence {sn } is not
convergent we say that the series ak is divergent. P
From basic P properties of limits it follows that if ak is convergent,
P then ak → 0 as
k
P∞ → ∞, and, if bk is another convergent series and c ∈ C, then a k + cbk is convergent and
P ∞ P∞
k=0 (ak + cbk ) = k=0Pak + c k=0 bk . P P
We say that a series ak is
P absolutely convergent if the series |a k | is convergent. If ak
is absolutely convergent, then ak is convergent; this is probably well-known at least if all the
ak are real. An argument showing it in general given the knowledge of it for ak ∈ R is outlined
in Exercise 4.1.
We recall the following convergence tests from calculus.
Direct comparison test: Let {ak } and {bk } be sequences of complex and P real numbers respecti-
P Assume that |ak | ≤ cbk for some constant c and assume that
vely. bk is convergent. Then
ak is absolutely convergent, and in particular convergent.
Ratio test: Let {ak } P
be a sequence and suppose that limk→∞ |ak+1 |/|ak | exists. If the P limit is
< 1, then the series ak is absolutely convergent; if the limit is > 1, then the series ak is
not absolutely convergent (but might still be convergent);
P if the limit is 1, then we do not get
any information about convergence properties of ak .
p
Root test: Let {ak } be a sequence such that limk→∞ k |ak | exists. Then the same conclusions
as in the Ratio test hold.
k ∞
Example 4.1 (Geometric series). P∞Let kz ∈ C and consider the sequence {z }k=0 ; it is called a
k| ≥ 1
geometric sequence. The series k=0 z is called a geometric series. If |z| ≥ 1, then |z
and so z k cannot go to 0 as k → ∞. Hence,
P
zk is divergent if |z| ≥ 1. On the other hand,
(1 − z)(1 + z + z 2 + · · · + z n ) = 1 − z n+1 so, if z 6= 1, we have
n
X 1 − z n+1
zk = .
1−z
k=0

1, then limk→∞ z k = 0, and we see P

If |z| < P that the right-hand side goes to 1/(1 − z) as k → ∞.
Hence, z k is convergent if |z| < 13 and ∞ k
k=0 z = 1/(1 − z).

4.2 Functions defined by power series

P k
A power series is a series of the form ck b , where b and the ck are complex
P numbers. Given
an a ∈ C and a sequence {ck } we get for each z ∈ C a power series ck (z − a)k ; it may
converge for some values of z (it converges at least for z = a) and diverge for other, depending
on the sequence {ck }. IfPit converges for z in some set D ⊂ C, then we can define a function
on D by setting f (z) = ∞ k
k=0 ck (z − a) .
3
One can also use the Ratio or Root test to check this.
Analytic function theory 23

P
Definition 4.2 (Radius of convergence). The radius of convergence of the power series ck (z−
a)k is defined as sup{|z − a|; ck (z − a)k is absolutely convergent}.
P

Let R be the radius of convergence of Pck (z − a)k . Then, by definition, for any r < R
P
there is some w suchPthat r < |w − a| and ck (w − a)k is absolutely convergent. Moreover,
if |w − a| > R, then k
ck (w − a) cannot be absolutely convergent. Notice that R may be ∞;
this
P is the case if and only if there is, for any r > 0, some w ∈ C such that |w P− a| > r and
ck (w − a)k is absolutely convergent. Notice also that R = 0 if and only if ck (z − a)k is
absolutely convergent only for z = a.
Example 4.3. Compute the radius of convergence of k k 2 z k .
P
Solution: We have
|(k + 1)2 z k+1 |
lim = lim (1 + 1/k)2 |z| = |z|.
k→∞ |k 2 z k | k→∞

Ratio test that k k 2 z k is absolutely convergent. On the

P
If |z| < 1 it thus follows from
P the
other hand, if |z| ≥ 1, then k k 2 z k is divergent since k 2 z k does not go to 0 as k → ∞. Hence
the radius of convergence is 1.
We see in this example that the series converges absolutely for all z in the disc determined by
the radius of convergence, and diverges for all z outside of that disc. This is not a coincidence,
in fact we have
Lemma 4.4. Let k ck (z − a)k be a power series with radius of convergence R.
P

(a) If |z − a| < R then k ck (z − a)k is absolutely convergent.

P

(b) If |z − a| > R then k ck (z − a)k is divergent.

P

Proof. After a translation we may assume that a = 0. To prove part (a), P let z be a point such
that |z| < R. By definition of R there is a w such that |z| < |w| < R and ck wk is absolutely
convergent. Then, since |z/w| < 1, we have
 z k
|ck z k | ≤ |ck wk |   ≤ |ck wk |,
 
w
|ck z k | is convergent, i.e., ck z k is
P P
and so it follows by the Direct comparison test that
absolutely convergent.
P To kprove part (b), let z be a point such that P |z| >k
R. By definition of R we know that
|ck z | is divergent but
P wek want to show that ck z is divergent. Assume therefore, to get
a contradiction, that ck z is convergent. Then |ck z k | → 0 as k → ∞ and in particular the
sequence {ck z k }k has to be bounded. There is therefore an M > 0 such that |ck z k | ≤ M for
all k. Let w be any point such that R < |w| < |z|. Then
 w k  w k
|ck wk | = |ck z k |   ≤ M   ,
   
z z
M |w/z|k converges. Thus, by the Direct com-
P
and, since |w/z| P < 1, the geometric series
parison test, k
|ck w | converges. But since |w| > PR thisk
contradicts the definition of R. Our
assumption must therefore be wrong and hence ck z cannot be convergent.

From this lemma we thus see that if the power series ck (z −a)k has radius of convergence
P
R then P we get a function f defined in the disc D(a, R) = {z ∈ C; |z − a| < R} by setting
f (z) = ∞ k
k=0 ck (z − a) . We will show in the remainder of this section that functions defined
by power series in fact are C ∞ -smooth and holomorphic. We begin with the preliminary
Analytic function theory 24

ck z k be a power series with radius of convergence R. Then

P
Lemma 4.5. Let
ck kz k−1 has radius of convergence R,
P
(a)
k−`−1 has radius of convergence R for each natural number `.
P
(b) k ck k(k − 1) · · · (k − `)z
Proof. Notice that part (b) follows from repeated use of part (a).
To prove part (a) we first show that P ck kz k−1 has radius of convergence ≤ R. If not,
P
then there is a z such that |z| > R and ck kz k−1 is absolutely convergent. Since k
P |ck z k | =
|z||ck z k−1 | ≤ |z||ck kz k−1 | it thus follows from the Direct comparison test that |ck z | is
convergent, contradicting the definition of R since |z| > R.
ck kz k−1 has radius of convergence ≥ R, let z be any point 6= P
P
To show that 0 such that
|z| < R. From the definition of R there is a w such that |z| < |w| < R and |ck wk | is
convergent. Since |z/w| < 1 we know from one-variable calculus that k|z/w|k → 0 as k → ∞,
and so, in particular, there is a constant M > 0 such that k|z/w|k ≤ M for all k. Hence,
k  z k M
|ck kz k−1 | =   |ck wk | ≤ |ck wk |
|z| w |z|
ck kz k−1 is absolutely convergent.
P
and the Direct comparison test implies that
k be a power series with radius of convergence R > 0 and define
P
Theorem 4.6. Let ck (z −
Pa)∞
f : D(a, R) → C by f (z) = k=0 ck (z − a)k . Then
(a) f is C ∞ -smooth in D(a, R),
(b) f is holomorphic in D(a, R),
P∞
(c) for each natural number `, f (`) (z) = k=` ck k(k − 1) · · · (k − ` + 1)(z − a)k−` .
Notice that part (c) says that the `th derivative of f is given by the term-wise differentiated
power series, which by Lemma 4.5 has radius of convergence R.
Proof of Theorem 4.6. After a translation we may assume that a = 0. We begin with the
technical part of the proof which is to show that for each fixed z ∈ D(0, R) we have
∞
X
f (z + h) − f (z) = h ck kz k−1 + O(|h|2 ). (4.1)
k=1
By the Binomial theorem we have
∞
X
ck (z + h)k − z k


f (z + h) − f (z) =
k=1
∞ k  
X X k k−` `
= ck z h
`
k=1 `=1
∞ k  
X X k k−` `−1
= h ck z h
`
k=1 `=1
∞ k  
  !
X k k−1 X k k−` `−2
= h ck z +h z h
1 `
k=1 `=2
∞ ∞ k  
X
k−1 2
X X k k−` `−2
= h ck kz +h ck z h
`
k=1 k=2 `=2
X∞
= h ck kz k−1 + h2 B(h).
k=1
Analytic function theory 25

To show (4.1) we thus need to show that B(h) is bounded as h → 0, i.e., that there are numbers
C and δ > 0 such that |B(h)| ≤ C when |h| ≤ δ. Take δ < R − |z|. Then, if |h| ≤ δ, we have
∞
 
k  
X X k k−` `−2 
|B(h)| =  ck z h 

 ` 
k=2 `=2
∞ k−2  
X X k
≤ |ck | |z|k−2−m |h|m
m+2
k=2 m=0
∞ k−2  
X X k(k − 1) k−2
= |ck | |z|k−2−m |h|m
(m + 2)(m + 1) m
k=2 m=0
∞ k−2  
X X k−2
≤ |ck |k(k − 1) |z|k−2−m |h|m
m
k=2 m=0
∞
X
k−2
= |ck |k(k − 1) |z| + |h|
k=2
∞
X
k−2
≤ |ck |k(k − 1) |z| + δ ,
k=2
where we in the third step have used basic properties of binomial coefficients and in the fifth
step have used the Binomial theorem backwards. Moreover, in view of Lemma 4.5, the last
series is convergent since |z| + δ < R. Hence, we may take the sum of this series as the constant
C, and so (4.1) follows. We notice in particular that it by (4.1) follows that f is continuous in
D(0, R).
From (4.1) it also readily follows that f is complex differentiable at all points z ∈ D(0, R)
with derivative f 0 (z) given by ∞ k−1 , which has radius of convergence R by Lemma 4.5.
P
k=1 k kz
c
By Theorem 3.5 the partial derivatives ∂f /∂x and ∂f /∂y exist in D(0, R) and f 0 = ∂f /∂x =
−i∂f /∂y, see (3.3) and (3.4). Hence, ∂f /∂x and ∂f /∂y are given by power series converging in
D(0, R) and so, as in the first part of the proof, ∂f /∂x and ∂f /∂y are continuous in D(0, R).
Thus, f is C 1 -smooth and P complexk−1
differentiable in D(0, R), i.e., f is holomorphic in D(0, R),
and, moreover, f 0 (z) = ∞ c
k=1 k kz . But then we can apply
P∞ precisely thek−2 same reasoning to
0 0 00
f to see that f is holomorphic in D(0, R) and f (z) = k=2 ck k(k − 1)z , which again by
Lemma 4.5 has radius of convergence R. Thus f 00 is holomorphic in D(0, R) with derivative
given by a power series converging in D(0, R). Continuing in this way (formally using induction)
Theorem 4.6 follows.

4.3 Holomorphic functions are power series locally

We will see that if f is a holomorphic function in an open set Ω ⊂ C, then for any given a ∈ Ω
and any disc D(a, r) centered at a such that D(a, r) ⊂ Ω, fD(a,r) is given by a power series
ck (z − a)k converging in D(a, r). This is the content of the next result which is a version of
P
Taylor’s theorem for holomorphic functions.
Theorem 4.7 (Taylor’s formula). Let f be P a holomorphic function in an open set containing
D(a, r). Then there is a unique power series ck (z − a)k converging absolutely in D(a, r) such
∞
X
that f (z) = ck (z − a)k for all z ∈ D(a, r). Moreover, the coefficients ck are given by
k=0

f (k) (a)
Z
1 f (w) dw
ck = = . (4.2)
2πi ∂D(a,r) (w − a)k+1 k!
Analytic function theory 26

Remark 4.8. By Cauchy’s theorem, the integration contour ∂D(a, r) in (4.2) may be replaced
by any piecewise smooth simple closed curve enclosing a.

Remark 4.9. If f is holomorphic merely in D(a, r), then the conclusions of Theorem 4.7 still
hold except for that we need to replace the integration contour ∂D(a, r) in (4.2) by a slightly
smaller curve. To see this we simply apply Theorem 4.7 to the smaller disc D(a, r − ) and let
 → 0.

Since holomorphicity and smoothness are local properties and since convergent power series
define holomorphic C ∞ -smooth functions by Theorem 4.6 we get the following corollary of
Theorem 4.7.

Corollary 4.10. If f is holomorphic in an open set Ω ⊂ C, then f ∈ C ∞ (Ω) and f (k) is

holomorphic in Ω for any natural number k.

Proof of Theorem 4.7. P After translation and scaling we may assume that a =P0 and r = 1.
First, notice that if ck z k is absolutely convergent in D(0, 1) and f (z) = ∞ k
k=0 ck z there,
then f (0) = c0 and, by differentiating term-wise ` times,
∞
X 
(`)
f (0) = ck k(k − 1) · · · (k − ` + 1) z k−`  = c` `!.

z=0
k=`

Hence, any absolutely convergent series in D(0, 1) representing f there must have coefficients
ck = f (k) (0)/k!. This proves the uniqueness part.R
Now we define coefficients ck by ck = (2πi)−1 |w|=1 f (w)dw/wk+1 , i.e., so that the left-hand
equality of (4.2) holds, and we set
N
X
AN (z) := f (z) − ck z k .
k=0

To prove the theorem it suffices to show that limN →∞ AN (z) = 0 for each fixed z ∈ D(0, 1)
k has to be convergent for each z ∈ D(0, 1) with sum f (z). Moreover,
P
because then ck z
ck z k then also has to be absolutely convergent in D(0, 1) since if not, then ck z k would
P P
not be convergent in D(0, 1) by Lemma 4.4. From the uniqueness part it will then also follow
that the right-hand equality of (4.2) holds.
To show that limN →∞ AN (z) = 0 weR plug in the definition of the coefficients ck , use
Cauchy’s formula to write f (z) = (2πi)−1 |w|=1 f (w) dw/(w − z), and compute:

Z N Z
1 f (w) dw X k 1 f (w) dw
AN (z) = − z
2πi |w|=1 w−z 2πi |w|=1 wk+1
k=0
N
Z !
1 f (w) X f (w)  z k
= − dw
2πi |w|=1 w−z w w
k=0
N
Z !
1 f (w) 1 X  z k
= − dw
2πi |w|=1 w 1 − z/w w
k=0
1 − (z/w)N +1
Z  
1 f (w) 1
= − dw
2πi |w|=1 w 1 − z/w 1 − z/w
f (w) (z/w)N +1
Z
1
= dw,
2πi |w|=1 w 1 − z/w
Analytic function theory 27

where we in the forth equality have used the formula for a geometric sum, cf. Example 4.1,
noticing that |z/w| < 1 for any z ∈ D(0, 1) and any w with |w| = 1. Now, f is in particular
continuous on the circle {|w| = 1} and so |f (w)| ≤ M if |w| = 1 for some constant M . Moreover,
by the Reverse triangle inequality, |1 − z/w| ≥ 1 − |z/w| = 1 − |z| if |w| = 1, and hence

 f (w) (z/w)N +1  N +1 |z|N +1

 
sup   = sup |f (w)| |z| ≤ M .
|w|=1 w 1 − z/w  |w|=1 |1 − z/w| 1 − |z|

By Proposition 2.7 we thus get

Z 
1  f (w) (z/w)N +1  M |z|N +1
|z|N +1
|AN (z)| = dw ≤ ` |w| = 1 = M ,

2π  |w|=1 w 1 − z/w  2π 1 − |z| 1 − |z|


and since |z| < 1 we see that limN →∞ |AN (z)| = 0.

Exercises
P
4.1 Let ak = xP k + iyk be aPsequence of complex numbers such that |ak | is convergent.
P
Show that k |xk | and k |yk | are convergent. Use Lemma 14.10 to conclude that ak
is convergent.

4.2 Compute the radius of convergence of the power series

X X X X X
(a) z k /k 3 , (b) (−1)k k 5 z k , (c) 8k z 3k , (d) z k /k k , (e) k!z k .
k k k k k

4.3 Find a power series representing f in the disc D where

(a) f (z) = 1/(z + 1) and D = D(0, 1),

(b) f (z) = 1/z and D = D(i, 1),
(c) f (z) = 1/(z 2 − z) and D = D(1/2, 1/2),
(d) f (z) = 1/(z 2 + z + 1) and D = D(0, 1).

4.4 Find a power series representing f in D(0, 1) where

1 1
(a) f (z) = , (b) f (z) = .
(1 − z)2 (1 − z)3

4.5 Cauchy’s estimate. Show that if f is holomorphic in D(0, r) and |f (z)| ≤ M for all
z ∈ D(0, r), then |f (k) (0)| ≤ k!M/rk for all integers k ≥ 0. (Hint: Use Theorem 4.7 and
Proposition 2.7.)
∞
X
4.6 Suppose f (z) = ck z k where the series has infinite radius of convergence. Show that
k=0
for any R > 0
∞
X
|ck |Rk ≤ 2 sup |f (z)|.
k=0 |z|=2R

(Hint: Use (4.2) and Proposition 2.7.)

Analytic function theory 28

5 Elementary functions via power series

We will use power series to define exponential, trigonometric, and hyperbolic functions. We
then define logarithms by inverting the exponential function. As the exponential function turns
out not to be injective this requires some care and involves making choices.

5.1 Exponential functions

P k
The series z /k! converges absolutely in C; this follows for instance from the Ratio test
since |z k+1 /(k + 1)!|/|z k /k!| = |z|/(k + 1) → 0 as k → ∞ for any z ∈ C. Thus, it defines a
holomorphic function in C that we take as definition of the exponential function, i.e.,
∞
z2 z3 X zk
ez := 1 + z + + + ··· = . (5.1)
2! 3! k!
k=0

Notice that we now have two potentially different definitions of eiθ for θ ∈ R but we will soon
see that they coincide. We notice also that (5.1) is an extension to C of the exponential function
on R, i.e., that the right-hand side of (5.1), with
P∞ z = x ∈ R, equals ex as defined in elementary
calculus. Indeed, from calculus we know that k=0 xk /k! is the Taylor series of ex .
The following result extends familiar computation rules of the exponential function from R
to C.
Theorem 5.1. The exponential function ez has the following properties:
∂ez
(a) e0 = 1, (b) = ez , (c) ez+w = ez ew , (d) ez 6= 0 ∀z ∈ C.
∂z
Proof. Part (a) is clear from the definition (5.1). Part (b) follows since
∞ ∞ ∞ ∞ ∞
∂ez ∂ X zk X ∂ zk X kz k−1 X z k−1 X z`
= = = = = = ez ,
∂z ∂z k! ∂z k! k! (k − 1)! `!
k=0 k=0 k=1 k=1 `=0

where we have used Theorem 4.6 (c) to differentiate the series term-wise.
To show part (c), let a ∈ C be an arbitrary complex number and consider the function
f (z) = ez ea−z ; notice that ea−z is a composition of holomorphic functions and thus holomorp-
hic. Differentiating, cf. Exercise 2.1 (b) and Exercise 3.1, and using part (a) we get

f 0 (z) = ez ea−z − ez ea−z = 0.

Hence, by Proposition 3.6, f is constant and so f (z) = f (0) = ea in view of part (a). Thus,
ez ea−z = ea for all a, z ∈ C and choosing a = z + w we get ez ew = ez+w .
Part (d) follows from parts (a) and (c) since for any z ∈ C, ez e−z = e0 = 1, and so ez
cannot be 0.

Another property of the exponential function worth pointing out is the following.
Proposition 5.2. If z = x + iy, then |ez | = ex . In particular, |eiy | = 1.
Proof. In view of Exercise 5.1 we have

|ez |2 = ez ez = ez ez̄ = ez+z̄ = e2x = (ex )2 .

We define the exponential function with base a ∈ (0, ∞) by setting az := ez·log a . It is clear
that the standard computation rules of these exponential functions then extend to C too.
Analytic function theory 29

5.2 Trigonometric and hyperbolic functions

We define the basic trigonometric and hyperbolic functions by the following series.
∞
z2 z4 X z 2k
cos z := 1 − + − ··· = (−1)k (5.2)
2! 4! (2k)!
k=0
∞
z3 z5 X z 2k+1
sin z := z − + − ··· = (−1)k (5.3)
3! 5! (2k + 1)!
k=0
∞
z2 z4 X z 2k
cosh z := 1 + + + ··· = (5.4)
2! 4! (2k)!
k=0
∞
z3 z5 X z 2k+1
sinh z := z + + + ··· = (5.5)
3! 5! (2k + 1)!
k=0
The radius of convergence of these series is ∞, Exercise 5.2, and, as with the exponential
function, we see that these definitions extend the usual trigonometric and hyperbolic functions
from R to holomorphic functions in C. Moreover, by differentiating term-wise (do it!) we get
that
∂ cos z ∂ sin z
= − sin z, = cos z,
∂z ∂z
∂ cosh z ∂ sinh z
= sinh z, = cosh z.
∂z ∂z
From the definitions we see that cos(−z) = cos z, sin(−z) = − sin z, cosh(−z) = cosh z, and
sinh(−z) = − sinh z. We notice also that cos z = cosh(iz), since (iz)2k = (−1)k z 2k , and that
i sin z = sinh(iz) since (iz)2k+1 = i(−1)k z 2k+1 . By summing the series (5.4) and (5.5) with z
replaced by iz we thus obtain
cos z + i sin z = cosh(iz) + sinh(iz) = eiz , (5.6)
cos z − i sin z = cosh(iz) − sinh(iz) = e−iz . (5.7)
In particular, eiθ = cos θ + i sin θ for θ ∈ R. Moreover, by multiplying (5.6) and (5.7) we get
cos2 z + sin2 z = cosh2 (iz) − sinh2 (iz) = 1.
The standard addition formulas for sin and cos extend to hold also for complex variables.
This can be checked, Exercise 5.4, using, e.g., the following identities, which follow easily from
the definitions; see also Exercise 9.1.
1 1 iz
cos z = eiz + e−iz , e − e−iz ,



sin z = (5.8)
2 2i
1 1
cosh z = ez + e−z , sinh z = ez − e−z .



(5.9)
2 2
As we know, | cos x| ≤ 1 and | sin x| ≤ 1 for x ∈ R but this is not the case for a complex
variable. In fact, both cos and sin are unbounded functions on C. For instance, by (5.8),
1  1  1
| cos(x + iy)| = eix−y + e−ix+y  ≥ |eix−y | − |e−ix+y | = e−y − ey ,

2 2 2
which goes to +∞ as y goes to either +∞ or −∞.
Another thing we need to get used to is that the exponential function is periodic in the
imaginary direction since ez+2πik = ez e2πik = ez (cos 2πk + i sin 2πk) = ez for any integer k. In
particular, the exponential function is not injective, which causes problems when we will try
to define logarithms in the next section.
Analytic function theory 30

5.3 Logarithms, arguments, and general powers

As mentioned, the exponential function is not injective on C and therefore it cannot have an
inverse defined on all of C. We encounter a similar problem in calculus when defining arccos
and arcsin, and the way out is to restrict the domains of definition of cos and sin to make them
injective. We will do the same thing with the exponential function.
Consider the horizontal strip Sϕ = {z ∈ C; ϕ < Im z ≤ ϕ + 2π}. Then the restriction of ez
to Sϕ is injective. To see this, assume that ez = ew . Then ez−w = 1 and so, by Exercise 5.3,
z − w = 2πik for some integer k. If both z and w are in Sϕ , then their imaginary parts cannot
differ by a non-zero integer multiple of 2π and hence k = 0, i.e., z = w. Thus, we may invert
the restriction of ez to Sϕ . This amounts to finding w ∈ Sϕ such that ew = z given some z. To
do this we write z on polar form;

z = |z|eiθ = elog |z| eiθ = elog |z|+iθ , (5.10)

where θ is an appropriate angle. Thus we may take w = log |z| + iθ. As above we notice that
there is no unique θ such that (5.10) holds; θ is only determined up to an integer multiple of 2π,
i.e., two different θ’s doing the job may differ by an integer multiple of 2π. This is “the reason”
why the exponential function is not injective. But since we require w to be in the strip Sϕ , θ
must satisfy ϕ < θ ≤ ϕ + 2π and then there is no ambiguity and so θ is uniquely determined.
Given ϕ ∈ R we define the argument function argϕ by setting argϕ z = θ, where ϕ <
θ ≤ ϕ + 2π and z = |z|eiθ . Notice that argϕ is not defined at 0. Notice also that argϕ is not
continuous on C \ {0} because if it were, then the the restriction of argϕ to the unit circle
would be continuous, and then ψ(t) := argϕ (eit ) would be continuous on R. But ψ jumps up
2π at the points t = ϕ + 2πk, k ∈ Z. On the other hand, argϕ restricted to the cut plane
Cϕ := C \ {reiϕ ∈ C; r ≥ 0}, i.e., points on the ray starting at 0 and going through eiϕ are not
allowed, is smooth (Exercise 5.5). We define the logarithm logϕ 4 on Cϕ by

logϕ z = log |z| + i argϕ z.

Then logϕ is smooth and u(z) = log |z| and v(z) = argϕ satisfy the Cauchy-Riemann equations
(3.1) in Cϕ , Exercise 5.6. Hence, logϕ is holomorphic in Cϕ . Knowing this we can compute
the derivative of logϕ using the Chain rule. In fact, for any z ∈ Cϕ we have z = elogϕ z , and
differentiating this identity we get 1 = elogϕ z (logϕ z)0 = z(logϕ z)0 . Hence,

1
(logϕ z)0 = .
z
Different choices of ϕ give different logarithms and one often says that one chooses branch
of the logarithm. The choice ϕ = −π is called the principal branch and we will simply write Log
to denote it. Similarly, we let Arg := arg−π be the principal branch of the argument functions.
Let α ∈ R. Then, given ϕ, we define

z α := eα(logϕ z) .

Notice that the left-hand side depends in general on the choice of ϕ even though it is not visible
from the notation. When dealing with general powers it is necessary to take care to specify
what one means! Notice also that if α is an integer, then there is no ambiguity.
4
This should not be confused with logarithms with respect to other bases than e; we are not concerned with
such logarithms here.
Analytic function theory 31

Example 5.3. Compute all possible values

of (2i)1/2 .
Solution: We shall compute e logϕ (2i) /2 for all possible values of ϕ. The possible values of
argϕ 2i are π/2+2πk, k ∈ Z, so the possible values of logϕ (2i) are logϕ (2i) = log 2+i(π/2+2πk)
for k ∈ Z. Thus, the possible values of (2i)1/2 are

√ iπ/4 iπk
e log 2+i(π/2+2πk) /2 = e(log 2)/2 ei(π/4+πk)
( = 2e e
√ 1 + i iπk 1 + i, if k is even,
= 2 √ e =
2 −1 − i, if k is odd.

We notice in particular that the possible values of (2i)1/2 are the solutions of the equation
z 2 = 2i. In general, if w 6= 0, then the possible values of w1/n are the solutions of the equation
z n = w.
Remark 5.4. We do not define z w for arbitrary complex numbers z and w!

Exercises
= k ck z k , where the series has radius of convergence R. Show that f (z) =
P
5.1 Let
P f (z) k
k c̄k z̄ for all z with |z| < R.

5.2 Show that the series in (5.2), (5.3), (5.4), and (5.4) have infinite radius of convergence.

5.3 Show that

(a) ez = 1 if and only if z = 2πik, k ∈ Z,

(b) cos z = 0 if and only if z = π/2 + kπ, k ∈ Z,
(c) sin z = 0 if and only if z = kπ, k ∈ Z.

5.4 Show that

(a) cos(z + w) = cos z cos w − sin z sin w for all z, w ∈ C,

(b) sin(z + w) = sin z cos w + sin w cos z for all z, w ∈ C.

5.5 Show that argϕ is smooth on the cut plane Cϕ and compute the partial derivatives
∂ argϕ /∂x and ∂ argϕ /∂y.
p
5.6 Let u(x, y) = log x2 + y 2 and v(x, y) = argϕ (x + iy). Show that u and v satisfy the
Cauchy-Riemann equations (3.1) in Cϕ .

5.7 Write ez as a power series k ck (z − 1)k in C.

P

5.8 Write sin2 z as a power series k ck z k in C.

P

2 z
5.9 Find the real and imaginary parts of (a) ez , (b) cos z, (c) ee .

5.10 Solve the following equations

(a) cos2 z = 4, (b) (z 3 − 1) sin(πz) = 0, (c) e2z = −1.

(
(sin z)/z if z 6= 0
5.11 Show that f (z) = is holomorphic in C.
1 if z = 0
(Hint: Use the power series representation of sin z at z = 0.)
Analytic function theory 32

5.12 Calculate argϕ z and logϕ z for z = −1, z = 1 + i, and

(a) ϕ = 0, (b) ϕ = −π/2, (c) ϕ = π/2.

√ i √
2
5.13 (a) Calculate 2 . (b) Calculate i using the principal branch of the logarithm.

5.14 The following integrals are 0. Why?

Z Z
dz sin z
(a) z
, (b) dz.
∂D(2,1) 1 − e |z|=1 cos z

5.15 Calculate
ez
Z Z
Log z
(a) dz, (b) dz.
∂D(2i,3) z2 + π2 |z−i/2|=1/3 (2z − i) sin(iπz)
Analytic function theory 33

6 Liouville’s theorem and the Fundamental theorem of algebra

6.1 Liouville’s theorem
Liouville’s theorem says that a non-constant holomorphic function in C cannot be bounded.
Put in another way, if f is a holomorphic function in C such that its image f (C) is contained
in some disc, then f is constant.
Theorem 6.1 (Liouville’s theorem). Let f be a holomorphic function in C and assume that
there is an M > 0 such that |f (z)| ≤ M for all z ∈ C. Then f is a constant function.
Proof. Let a, b ∈ C be arbitrary points. It suffices to show that f (a) = f (b). Let R >
2 max{|a|, |b|}. Then, for any z with |z| = R we have

|z − a| > R/2 and |z − b| > R/2.

By Cauchy’s formula we have

f (z)(a − b) dz
Z Z Z
1 f (z) dz 1 f (z) dz 1
f (a) − f (b) = − = .
2πi ∂D(0,R) z − a 2πi ∂D(0,R) z − b 2πi ∂D(0,R) (z − a)(z − b)

Hence, by Proposition 2.7 we get

Z 
 
f (a) − f (b) = 1 
 f (z)(a − b) dz 
2π  ∂D(0,R) (z − a)(z − b) 
 

1 |f (z)||a − b|
≤ sup · 2πR
2π |z|=R |z − a||z − b|
M |a − b|R 4M |a − b|
≤ = .
(R/2)(R/2) R
But this holds for all R with R > max{|a|, |b|} and since 4M |a − b|/R → 0 as R → ∞ it follows
that f (a) = f (b).

A slightly different proof, giving a more general Liouville theorem, is outlined in Exerci-
se 6.3.

6.2 The fundamental theorem of algebra

The fundamental theorem of algebra says that any polynomial equation p(z) = 0 has a complex
solution (unless p is a non-zero constant). Let the degree of p be n > 0 and let a1 ∈ C be a
solution of p(z) = 0. By polynomial division it then follows that p(z) = p1 (z)(z − a1 ), where p1
is a polynomial of degree n−1. Applying the Fundamental theorem of algebra to p1 we see that
there is an a2 ∈ C such that p1 (a2 ) = 0 and then, by polynomial division, p1 (z) = p2 (z)(z − a2 )
for some polynomial p2 of degree n − 2. Continuing in this way we finally get a polynomial pn
of degree 0, i.e., pn is a constant c, such that pn−1 (z) = c(z − an ). Hence, the Fundamental
theorem of algebra implies that any polynomial of degree n with complex coefficients can be
factorized as
p(z) = c(z − a1 ) · (z − a2 ) · · · (z − an ),
where c and the aj ’s are complex numbers. The terminology to express this property of the
complex numbers is to say that C is algebraically closed.
Theorem 6.2 (Fundamental theorem of algebra). Let p(z) be a non-constant polynomial with
complex coefficients. Then there is an a ∈ C such that p(a) = 0.
Analytic function theory 34

Proof. Assume, to get a contradiction, that p(z) 6= 0 for all z ∈ C. Then 1/p(z) is a holomorphic
function in C. Since p is a non-constant polynomial, |p(z)| → ∞ as |z| → ∞; an argument for
this fact is outlined in Exercise 6.1. It follows that there is an M > 0 such that 1/|p(z)| ≤ M
for all z ∈ C. In fact, since |p(z)| → ∞ as |z| → ∞ there is an R0 > 0 such that 1/|p(z)| ≤ 1
if |z| > R0 and, moreover, since 1/|p(z)| is a continuous function on the closed and bounded
set D(0, R0 ) there is an M 0 such that 1/|p(z)| ≤ M 0 if z ∈ D(0, R0 ). Hence, we may take
M = max {1, M 0 }. Now, since 1/p(z) is a bounded holomorphic function in C Liouville’s
theorem shows that 1/p(z) is constant, but then p(z) is constant, which is a contradiction. Our
assumption must thus be false and so there is an a ∈ C such that p(a) = 0.

Exercises
Pn k |p(z)|
6.1 Let p(z) = k=0 ck z where cn 6= 0. Show that lim = |cn | and conclude that
|z|→∞ |z|n
|p(z)| ≥ (1/2)|cn ||z|n if |z| is sufficiently large.

6.2 Show that if f is holomorphic in C and |f (z)| ≥ M > 0 for all z ∈ C, then f is constant.

6.3 Generalized Liouville theorem. The purpose of this exercise is to show that if f is a
holomorphic function in C and |f (z)| ≤ C(1 + |z|)n for some constants C and n and all
z ∈ C, then f is a polynomial of degree ≤ n. This means roughly speaking that if |f (z)|
does not grow faster than a polynomial as |z| → ∞, then f is a polynomial.

(a) Use Exercise 4.5 to show that |f (k) (0)| ≤ k!C(1+r)n /rk for all r > 0 and all integers
k ≥ 0.
(b) Use Theorem 4.7 to conclude that f is a polynomial of degree ≤ n.

6.4 Let f be holomorphic in C. Show that f is constant if

(a) ef (z) is bounded,

(b) Re f or Im f is bounded. (Hint: Part (a) and Proposition 5.2.)

6.5 Let f be holomorphic in C and assume that f is periodic in both the x- and the y-
direction. Show that f is constant.

6.6 Let f be holomorphic in C and assume that f (z) → 0 as |z| → ∞. Show that f = 0.
Analytic function theory 35

7 Conformal mappings
Concrete problems often become computationally easier if the underlying geometry is simple.
For instance, certain physical problems amount to solving a differential equation in an open
set Ω ⊂ C given relevant data specified on the boundary ∂Ω. If Ω is geometrically simple then
the computations naturally become simpler; in particular the unit disc, the upper half-plane,
and annuli have simple geometries with lots of symmetries that facilitate computations. It is
therefore desirable to be able to transform problems with complicated underlying geometry
into problems involving, e.g., the unit disc.
Conformal mappings are mappings that preserve angles between curves. Loosely speaking
this means that they preserve the local geometry up to rotation and scaling. Consequently,
problems of geometric nature transform well under conformal mappings, and so it is desirable
to be able to map complicated sets to simpler ones by conformal mappings.

7.1 Angles between curves

Let C1 and C2 be two smooth oriented curves intersecting at the point a. We define the angle
between C1 and C2 at a as follows. Let wj 6= 0 be a tangent to Cj at a pointing in the direction
given by the orientation; we identify tangent vectors in R2 with complex numbers in the usual
way. Then the angle between C1 and C2 at a is defined by
w2
ν(C1 , C2 ) = Arg .
w1
In terms of parametrizations this means that if γj : Ij → C is a parametrization of Cj such that
0 ∈ Ij , γj (0) = a, and |γj0 (0)| = 1, then ν(C1 , C2 ) is the unique number in (−π, π] such that
eiν(C1 ,C2 ) γ10 (0) = γ20 (0).
Notice that the angle is a number in the interval (−π, π]. Notice also that if ν(C1 , C2 ) 6= π,
then ν(C1 , C2 ) = −ν(C2 , C1 ); if ν(C1 , C2 ) = π, then ν(C1 , C2 ) = ν(C2 , C1 ).
Example 7.1. Let C1 be the unit circle oriented counterclockwise and let C2 be the imaginary
axis oriented upwards. The angle between C1 and C2 at −i is π/2 and the angle at i is −π/2.
Holomorphic functions with non-zero derivative preserve angles; this is the content of
Theorem 7.2. Let C1 and C2 be two smooth oriented curves intersecting at a and let f be
a holomorphic function in an open set containing a such that f 0 (a) 6= 0. Then f (C1 ) and
f (C1 ) are smooth oriented curves in an open set containing f (a) intersecting at f (a), and
ν(C1 , C2 ) = ν(f (C1 ), f (C2 )).
Proof. Let γ1 and γ2 be parametrizations as above and set Γj (t) = f (γj (t)), j = 1, 2. Then Γj
is a smooth parametrization of f (Cj ) and Γj (0) = f (γj (0)) = f (a). Moreover, by the Chain
rule, Γ0j (0) = f 0 (γj (0)) · γj0 (0) = f 0 (a) · γj0 (0), see Exercise 2.7. Hence,
Γ02 (0) f 0 (a) · γ20 (0) γ20 (0)
ν(f (C1 ), f (C2 )) = Arg = Arg = Arg = ν(C1 , C2 ).
Γ01 (0) f 0 (a) · γ10 (0) γ10 (0)

Definition 7.3 (Conformal mapping). A conformal mapping of an open set Ω ⊂ C is a

holomorphic function f defined on Ω such that f 0 (z) 6= 0 for all z ∈ Ω.
If f is a conformal mapping of Ω then Ω and f (Ω) have the same local geometry up to
rotation and scaling but the global shapes of Ω and f (Ω) may be quite different. In the next
section we will see lots of examples of conformal mappings and how they can change the global
shapes.
Analytic function theory 36

7.2 Constructions of conformal mappings

We illustrate how one can construct conformal mappings with various properties by composing
Möbius transformations, exponential functions, logarithms and powers. Notice that, by the
Chain rule, the composition of conformal mappings is again conformal.
Möbius transformations are conformal mappings. Recall that if f (z) = (az + b)/(cz + d) is
a Möbius transformation then we require that ad − bc 6= 0. Hence,
a(cz + d) − (az + b)c ad − bc
f 0 (z) = 2
= 6= 0,
(cz + d) (cz + d)2
at least for z 6= −d/c; if c = 0 then f is conformal for all z ∈ C. With appropriate definitions a
Möbius transformation is in fact a conformal mapping of the Riemann sphere, but we will not
go further into conformal mappings in this generality.
Example 7.4. Let f (z) = (z − i)/(z + i). Compute the images of the upper half-plane, the
lower half-plane, the real axis, the imaginary axis, and the line {z; Im z = 1}.
Solution: A point z is in the upper half-plane if and only if its distance to i is less than its
distance to −i, i.e., |z − i| < |z + i|. Similarly, z is in the lower half-plane if and only if |z − i| >
|z + i| and z is on the real axis if and only if |z − i| = |z + i|. Hence, |f (z)| = |z − i|/|z + i| < 1
if z is in the upper half-plane, |f (z)| = 1 if z is real, and |f (z)| > 1 if z is in the lower half-
plane. Since Möbius transformations are bijective maps of the Riemann sphere it follows that
the image of the upper half-plane is the unit disc D(0, 1), the image of the lower half-plane is
C \ D(0, 1) (together with ∞), and the image of the real axis is the unit circle ∂D(0, 1).
It remains to compute the images of the imaginary axis and the line {z; Im z = 1}. The
imaginary axis goes through −i, 0, and ∞ so its image must be a circle or line going through
f (−1) = ∞, f (0) = −1, and f (∞) = 1. Hence, the image of the imaginary axis is the real
axis. Notice that the image of the imaginary axis intersects the image of the real axis at a right
angle. To find the image of the line {z; Im z = 1} we notice that it must be a circle since −i,
the point which is mapped to ∞, is not on the line. Moreover, this circle must contain f (i) = 0
and f (∞) = 1 and, in addition, it must intersect the image of the imaginary axis, i.e., the real
axis, at a right angle since f is conformal. It follows that the image of {z; Im z = 1} is the
circle centered at 1/2 with radius 1/2. (Draw the picture!)
The exponential function is conformal in C since (ez )0 = ez 6= 0 by Theorem 5.1. Let S
be the vertical strip {z; a < Re z < b}. The image of S under the exponential function is the
annulus {w; ea < |w| < eb } since

exp(S) = {w = ez ; a < Re z < b} = {w = ex eiy ; a < x < b} = {w = reiy ; ea < r < eb }.

Similarly we see that the image of a vertical line is a circle centered at 0, and the image of a
horizontal line is a ray from 0. Notice in particular that the image of a vertical line intersects
the image of a horizontal line at a right angle.
Logarithms are conformal where they are defined since (logϕ z)0 = 1/z for z ∈ Cϕ ; see
Chapter 5 for the notation.
Example 7.5. Let Ω = {z; θ1 < Arg z < θ2 }, where −π ≤ θ1 < θ2 ≤ π. Compute the image
of Ω under the principal branch of the logarithm.
Solution: We have

Log (Ω) = {w = Log z; θ1 < Arg z < θ2 } = {w = log |z| + iArg z; θ1 < Arg z < θ2 },

and so the image is the horizontal strip {w; θ1 < Im w < θ2 }.

Analytic function theory 37

Power functions f (z) = z k for k ∈ Z are conformal in C\{0} and power functions f (z) = z α
for α ∈ R\Z are conformal where they are defined, see Exercise 7.3. To see what power functions
may do to sets consider for instance f (z) = z 1/2 defined by the principal branch of the logarithm
in the cut plane C−π . Then

f (C−π ) = {w = e(Log z)/2 ; z ∈ C−π } = {w = e(log |z|+iArg z)/2 ; −π < Arg z < π}
p p
= {w = |z|ei(Arg z)/2 ; −π < Arg z < π} = {w = |z|eiθ ; −π/2 < θ < π/2},

which is the right half-plane. In a similar way we see that the image of the wedge {z =
reiθ ; −π/4 < θ < π/4} under the map f (z) = z 2 is the right half-plane too. Power functions
can thus be used to open up wedges or make them sharper.
We now illustrate how one can construct conformal mappings with certain properties by
composing mappings of the above type. Consider a lens-shaped set Ω bounded by two circular
arcs C1 and C2 starting at 0 and ending at 1; for definiteness say that C1 is the lower curve
and let θ be the angle between C1 and C2 . Our task is to find a conformal map of Ω onto the
unit disc. Viewed on the Riemann sphere there is no essential difference between Ω and the
set between two rays going out from the same point with angle θ. More precisely, if we choose
a Möbius transformation f1 that maps f1 (1) = ∞ and f1 (0) = 0, then f1 (C1 ) and f1 (C2 ) are
rays starting at 0 and f1 (Ω) is the wedge between f1 (C1 ) and f1 (C2 ) with angle θ. The choice
f1 (z) = z/(z − 1) does the job and we set Ω1 := f1 (Ω).
We may open up the the wedge Ω1 to be a half-plane by using a suitable power. We first
make a rotation so that the ray f1 (C1 ) becomes the positive real axis; if the angle between the
positive real axis and f1 (C1 ) is ν then f2 (z) := e−iν z accomplishes this. Thus, Ω2 := f2 (Ω1 ) is
the wedge with angle θ between the positive real axis and the ray f2 (f1 (C1 )). To open up Ω2
to a wedge with angle π we let f3 (z) := z π/θ where we use, e.g., the principal branch of the
logarithm to define z π/θ . Then one checks (do it!) that Ω3 := f3 (Ω2 ) is the upper half-plane.
Finally we let f4 (z) := (z − i)/(z + i), which maps the upper half-plane to the unit disc,
see Example 7.4. The composition

π/θ
e−iν z/(z − 1) −i
f4 ◦ f3 ◦ f2 ◦ f1 (z) =
π/θ
e−iν z/(z − 1) +i
thus is a conformal mapping of Ω onto the unit disc.

Exercises
7.1 Compute the images of the real and imaginary axes and
(a) the lower half-plane under the map f (z) = (z + i)/(z − i),
(b) the right half-plane under the map f (z) = (z − 1)/(z + 1),
(c) the left half-plane under the map f (z) = (z + 1)/(z − 1).
7.2 Compute the image under the exponential function of the box

(a) {z = x + iy; a < x < b, 0 ≤ y < 2π}, (b) {z = x + iy; a < x < b, θ1 < y < θ2 },

and of the strip

(c) {z = x + iy; −π/4 < y < π/2}, (d) {z = x + iy; log 2 < x < log 3}.

7.3 Show that (z α )0 = αz α−1 if z α and z α−1 are defined using the same branch of the
logarithm.
Analytic function theory 38

7.4 Find a conformal mapping of the slit disc D(0, 1) \ (−∞, 0] onto D(0, 1).

7.5 Find a conformal mapping of the piece of cake {z; 0 < Arg z < π/2, |z| < 1} onto the
upper half-plane.

7.6 (a) Calculate the image of the first quarter (i.e., {z; Re z > 0, Im z > 0}) under the map
f (z) = z 3 . (b) Calculate the image of {z; π/3 < Arg z < π/2} under the map f (z) = z 4 .

7.7 Let f (z) = −iz 2 and a ∈ R+ . Sketch the image of

(a) the vertical line {z; Re z = a},

(b) the set {z; 0 < Re z < 1}.

7.8 Find and sketch f (Ω) when

(a) f (z) = z/(z − 2) and Ω = {z; |z − 1| > 1},

(b) f (z) = (z − 1/2)/(z − 2) and Ω = D(0, 1),
(c) f (z) = iz/(z − 1) and Ω = {z; |z − 1/2| > 1/2, Im z 6= 0},
(d) f (z) = 1/z and Ω is the square with corners at ±1 ± i,
(e) f (z) = z/(z − i) and Ω = {z; Im z > 0, } \ {z; Re z = 0, 0 ≤ Im z ≤ 1},
(f) f (z) = (z − 1)/(z + 1) and Ω = {z; −1 < Re z < 1, |z| > 1},
(g) f (z) = (z − i)/(z + i) and Ω = {z; |z| > 1, |z + i| < 2}.

7.9 Find f (Ω) when

 2 2
z +1
(a) f (z) = and Ω = {z; |z| < 1, 0 < Arg z < π/2},
z2 − 1
i − e2z
(b) f (z) = and Ω = {z; 0 < Im z < π/2},
i + e2z
(c) f (z) = e1/(z−1) and Ω = {z; |z| < 1, |z − 1/2| > 1/2}.

(Hint: Write f as a composition of simpler maps.)

7.10 Find a conformal map

(a) from {z; −1 < Re z < 1} onto D(0, 1),

(b) from {z; |z − 1| < 1, Im z > 0} onto D(0, 1),
√ √
(c) from {z; |z − i| < 2, |z + i| < 2} onto D(0, 1),
(d) from {z; |z| < 1, 0 < Arg z < π/2} onto the upper half-plane,
√
(e) from {z; |z − (i − 1)/ 2| > 1, Im z > Re z} onto D(0, 1),
(f) from {z; |z − 1| > 1, |z + 1| > 1} onto D(0, 1).
Analytic function theory 39

8 Morera’s theorem and Goursat’s theorem

We will see that our definition of holomorphicity is equivalent to the “classical” one, i.e., that
complex differentiability in an open set implies holomorphicity in our sense. Morera’s theo-
rem says that a continuous function whose integral around the boundary of any triangle is 0
must be holomorphic in our sense. Goursat’s theorem shows that a function which is complex
differentiable in an open set satisfies the hypothesis of Morera’s theorem.

8.1 Morera’s theorem

By a triangle in an open set Ω we mean a bounded open subset ∆ ⊂ Ω such that ∆ ⊂ Ω and
the boundary ∂∆ of ∆ consists of three line segments.

Theorem 8.1 (Morera’s Z theorem). Let Ω ⊂ C be an open set and let f be a continuous
function in Ω such that f (z) dz = 0 for all triangles ∆ in Ω. Then f is holomorphic in Ω.
∂∆

Proof. Holomorphicity is a local property so it suffices to show that, given any point z0 ∈ Ω, f is
holomorphic in some disc D(z0 , r), r > 0. Let z0 ∈ Ω and choose r > 0 such that D(z0 , r) ⊂ Ω;
this is possible since Ω is open. We define a function F in D(z0 , r) by setting
Z
F (z) = f (z) dz,
[z0 ,z]

where [z0 , z] is the line segment starting at z0 and ending at z. Fix w ∈ D(z0 , r) and let h
be a complex number with |h| so small that the triangle with corners at z0 , w, and w + h is
contained in D(z0 , r). Since the integral of f around the boundary of any triangle in Ω is 0 we
have Z Z Z
0= f (z) dz + f (z) dz + f (z) dz.
[z0 ,w] [w,w+h] [w+h,z0 ]

Hence we get
Z Z Z
F (w + h) − F (w) = f (z) dz − f (z) dz = f (z) dz
[z0 ,w+h] [z0 ,w] [w,w+h]
Z 1 1 Z


= f (w + th) hdt = h f (w) + f (w + th) − f (w) dt
t=0 0
Z 1


= h · f (w) + h f (w + th) − f (w) dt,
0

where we in the third step have parametrized [w, w + h] by [0, 1] using t 7→ w + th. Since f is
continuous, |f (w + th) − f (w)| → 0 as h → 0 and so we see that

F (w + h) − F (w) = h · f (w) + o(|h|). (8.1)

It follows that F is complex differentiable in D(z0 , r) with F 0 (w) = f (w). By Theorem 3.5 the
partial derivatives ∂F/∂x and ∂F/∂y exist in D(z0 , r) and F satisfies the Cauchy-Riemann
equations there. In view of the paragraph following Theorem 3.5 this means that ∂F/∂x =
−i∂F/∂y. But letting h → 0 with h ∈ R in (8.1) we see that ∂F/∂x = f . Since f is continuous,
∂F/∂x and ∂F/∂y are too and so F ∈ C 1 (D(z0 , r)). Thus, F is holomorphic in D(z0 , r) and
so, by Corollary 4.10, f = F 0 is holomorphic in D(z0 , r).
Analytic function theory 40

8.2 Goursat’s theorem

Theorem 8.2 (Goursat’s
Z theorem). Let f be a function which is complex differentiable in an
open set Ω ⊂ C. Then f (z) dz = 0 for any triangle ∆ ⊂ Ω.
∂∆

Proof. WeR notice first that f is continuous in Ω, see Exercise 2.4. Let ∆0 ⊂ Ωjbe a triangle and
set I := ∂∆0 f (z) dz; we will show that I = 0. Divide ∆0 into four triangles ∆0 , j = 1, . . . , 4, by
inscribing a triangle in ∆0 with corners on the midpoints of the edges of ∆0 . Then, Exercise 8.1,
Z 4 Z
X
f (z) dz = f (z) dz, (8.2)
∂∆0 j=1 ∂∆j0

and so there must be at least one j such that | ∂∆j f (z) dz| ≥ |I|/4. Let ∆1 be one of the ∆j0 ’s
R
0
with this property. Notice that the circumference of ∆1 , `(∂∆1 ), is half that of ∆0 . Repeating
this process of dividing a triangle into four smaller triangles we get a decreasing sequence of
triangles, ∆0 ⊃ ∆1 ⊃ ∆2 ⊃ · · · , such that
Z 
f (z) dz  ≥ |I|/4k and `(∂∆k ) = `(∂∆0 )/2k .
 


∂∆k

Let a ∈ ∩k ∆k ; it is a fact of topology that a decreasing sequence of compact subsets

has non-empty intersection. Since f is complex differentiable at a there is an A (the limit of
(f (a + h) − f (a))/h as h → 0) such that f (a + h) − f (a) = Ah + o(|h|). Hence, for any given
 > 0 there is a δ > 0 such that if |z − a| < δ then
 
f (z) − f (a) − A(z − a) ≤ |z − a|.

Take k soR large that ∆k ⊂ D(a, δ); this is possible since the ∆j ’s shrink to a. By Cauchy’s
theorem ∂∆k (f (a) + A(z − a)) dz = 0 and so we get
Z  Z 
  


 f (z) dz  = 
 f (z) − f (a) − A(z − a) dz 
∂∆k ∂∆k
≤ sup |f (z) − f (a) − A(z − a)| · `(∂∆k )
z∈∂∆k
≤ sup |z − a|`(∂∆k )
z∈∂∆k

≤  · `(∂∆k )2 =  · `(∂∆0 )2 /4k .

Z
Thus, for k large enough, |I|/4k ≤ f (z) dz ≤  · `(∂∆0 )2 /4k so |I| ≤  · `(∂∆0 )2 . But  > 0
∂∆k
is arbitrary so |I| = 0, i.e., I = 0.

Exercises
8.1 Show that (8.2) holds.

8.2 Let Ω ⊂ C be open and let {fn }n be a sequence of holomorphic functions in Ω. Assume
that fn (z) converges uniformly on each compact subset of Ω to some function f (z).

(a) Show that f is holomorphic in Ω.

(b) Show that lim fn(k) (a) = f (k) (a) for each a ∈ Ω and k = 1, 2, . . ..
n→∞
Analytic function theory 41

(Hint: (a) Use Exercise 2.8 to show that f satisfies the hypothesis of Morera’s theorem.
(k)
(b) Apply Cauchy’s estimate, Exercise 4.5, to f (k) − fn ).
∞
X
8.3 (a) Show that the series 1/nz converges if Re z > 1.
n=1
∞
X N
X
(b) Let f (z) = 1/nz and fN (z) = 1/nz . Show that if δ > 0 then
n=1 n=1

sup |f (z) − fN (z)| → 0, N → ∞.

Re z≥1+δ

Conclude, using Exercise 8.2, that f is holomorphic in {z; Re z > 1}. The function
f is known as the Riemann zeta function.
X
(Hint: Compare to the series 1/n1+δ .)
n
Analytic function theory 42

9 Zeros of holomorphic functions

A polynomial of degree at most n cannot have more than n zeros unless it is identically zero.
As we will see, something similar holds for holomorphic functions; a holomorphic function
cannot have “too many” zeros unless it is identically zero. The reason is, loosely speaking, that
a holomorphic function locally is given by a convergent power series, which can be thought of
as a polynomial of infinite degree. This means that holomorphic functions, just as polynomials,
are rigid objects; two holomorphic functions which agree at sufficiently many points have to
agree everywhere. This is in stark contrast to smooth or continuous functions which are much
more plastic.

9.1 Zeros, their orders, and rigidity theorems

Let Ω ⊂ C and let f be a holomorphic function in Ω. We let Z(f ) := {a ∈ Ω; f (a) = 0} be the
zero set of f . We say that a ∈ Z(f ) has multiplicity m, or that a is a zero of order m, if

f (a) = f 0 (a) = · · · = f (m−1) (a) = 0, f (m) (a) 6= 0.

Example 9.1. Let f (z) = z 2 − 1. Then Z(f ) = {1, −1} and both 1 and −1 are zeros of order
1 since f (±1) = 0 and f 0 (±1) = ±2 6= 0.

Example 9.2. The function cos z − 1 has a zero of order 2 at z = 0 since cos 0 − 1 = 0,
(cos z − 1)0 (0) = − sin 0 = 0, and (cos z − 1)00 (0) = − cos 0 = −1 6= 0.

A holomorphic function cannot have a zero of infinite order unless the function is identically
zero. More precisely, we have

Theorem 9.3 (Identity theorem I). Let Ω ⊂ C be a connected open set and let f be holomorphic
in Ω. If there is an a ∈ Ω such that f (k) (a) = 0 for all integers k ≥ 0, then f is identically 0
in Ω.

Proof. Consider the set S of points a ∈ Ω such that f (k) (a) = 0 for all integers k ≥ 0. Then S
is a closed subset of Ω since all f (k) are continuous. On the other hand, by Theorem 4.7,
∞
X f (k) (a)
f (z) = (z − a)k (9.1)
k!
k=0

in some disc D(a, r) and so, if a ∈ S, then f = 0 in some open disc centered at a. This disc
then is a subset of S and so S open. Hence, both S and Ω \ S are open subsets of Ω, and since
Ω is connected either S or Ω \ S has to be empty, see Appendix B. Thus, if there is an a ∈ S,
then S = Ω and so f is 0 in Ω.

Example 9.4. Let Ω be an open connected set, C a curve in Ω, and f a holomorphic function
in Ω. Show that if f = 0 on C, then f = 0 in Ω.
Solution: Let z0 ∈ C. Since f 0 (z0 ) is given by the limit (2.5) it follows by letting h → 0 along
C that f 0 (z0 ) = 0. But z0 ∈ C is arbitrary so f 0 = 0 on C. Repeating the argument it follows
that f (k) = 0 on C for all k ≥ 0. By the Identity theorem I, f = 0 in Ω.

Proposition 9.5. Let f be holomorphic in a disc D(a, r). Then the following are equivalent.

(a) f has a zero of order m at a.

(b) There is a holomorphic function g in D(a, r) such that g(a) 6= 0 and f (z) = (z − a)m g(z).
Analytic function theory 43

(c) The limit limz→a f (z)/(z − a)m exists and is non-zero.

Proof. Let ` be the least integer such that f (`) (a) 6= 0. Then f has a zero of order m at a if
and only if ` = m. Moreover, by Theorem 4.7, see also (9.1), we have
!
` f (`) (a) f (`+1) (a) f (`+2) (a) 2
f (z) = (z − a) · + (z − a) + (z − a) + · · · . (9.2)
`! (` + 1)! (` + 2)!

Thus, if f has a zero of order m at a then f (z) = (z − a)m g(z) and g(a) = f (m) (a)/m! 6= 0.
Hence, (a) ⇒ (b).
To see that (b) ⇒ (c) we just notice that if f (z) = (z − a)m g(z) with g(a) 6= 0, then
limz→a f (z)/(z − a)m = g(a) 6= 0.
It remains to see that (c) ⇒ (a). Assume that limz→a f (z)/(z − a)m exists and is non-zero,
and let ` be the number defined above; we’ll show that ` = m. In view of (9.2), if ` < m
then |f (z)|/|z − a|m → ∞ as z → a and so limz→a f (z)/(z − a)m cannot exist; if ` > m then
limz→a f (z)/(z − a)m = 0. Hence, ` = m and we are done.

Example 9.6. Compute the multiplicity of the zero at z = 0 of sin12 z.

Solution: sin z has a zero of order 1 at z = 0 since sin 0 = 0 and (sin z)0 (0) = cos 0 = 1 6= 0. By
Proposition 9.5, sin z = zg(z) where g is holomorphic and g(0) 6= 0. Hence, sin12 z = z 12 g(z)12 .
Since g(0)12 6= 0, Proposition 9.5 shows that sin12 z has a zero of multiplicity 12 at z = 0.
The next result, which can be seen as a jazzed-up version of the Identity theorem I, says
that a non-trivial holomorphic function must have isolated zeros. That a function in some open
set has an isolated zero at a means that for some r > 0 the only zero of f in the disc D(a, r)
is a.
Theorem 9.7 (Identity theorem II). Let f be a holomorphic function in an open connected
set Ω ⊂ C. Then either f is identically 0 in Ω or f has only isolated zeros in Ω.
Proof. Let E be the set of limit points of Z(f ) in Ω (see Appendix B); it is an exercise (do it!)
to show that E ⊂ Z(f ). The set E thus is the set of non-isolated zeros of f in Ω. We will show
that either f is identically 0 or that E is empty, in which case f only has isolated zeros.
Assume that E is non-empty and let a ∈ E. If f is not identically zero, then a must be a
zero of f with some finite multiplicity m by the Identity theorem I. Thus, by Proposition 9.5,
f (z) = (z − a)m g(z) in some open disc centered at a where g is holomorphic and g(a) 6= 0.
Since g in particular is continuous, g 6= 0 in some open set containing a and so there is a disc
D(a, r), r > 0, such that g 6= 0 in D(a, r). But then a is an isolated zero of f . This contradicts
that a ∈ E and so E must be empty.

Corollary 9.8. Let Ω ⊂ C be open and connected and let f and g be holomorphic functions
in Ω. If there is a set E ⊂ Ω that has a limit point in Ω and f (z) = g(z) for z ∈ E, then f = g
in Ω.
Proof. The function F (z) := f (z) − g(z) is holomorphic in Ω and Z(F ) contains E. Thus, F
has non-isolated zeros in Ω and must therefore be identically zero in Ω by the Identity theorem
II.

Example 9.9. Show that sin2 z + cos2 z = 1 in C.

Solution: We have already showed this in Section 5.2 but here is an argument based on Corol-
lary 9.8. Let f (z) = sin2 z + cos2 z and g(z) = 1, which are holomorphic in C = Ω. We know
that f (z) = g(z) for z ∈ R. But any point on R is a limit point of R and R ⊂ C so Corollary 9.8
gives that f (z) = g(z) for all z ∈ C.
Analytic function theory 44

9.2 Analytic continuation

Let f be a holomorphic function in an open set Ω ⊂ C. Suppose that f˜ is a holomorphic
function in some open set Ω̃ and that f = f˜ in Ω ∩ Ω̃. Then we can extend f to a holomorphic
function in Ω ∪ Ω̃ by setting f (z) := f˜(z) for z ∈ Ω̃. So far this has nothing to do with f and
f˜ being holomorphic but the point is that if Ω ∪ Ω̃ is connected, then f can be extended to
a holomorphic function in Ω ∪ Ω̃ in only one way. In fact, since Ω is open any point of Ω is a
limit point of Ω and so, if Ω ∪ Ω̃ is connected, any holomorphic extension of f to Ω ∪ Ω̃ must
agree with f˜ in Ω̃ by Corollary 9.8.
If a holomorphic function f a priori defined in some open set Ω0 can be extended to a
holomorphic function in some bigger open set Ω1 we say that f can be analytically continued
to Ω1 ; if Ω1 is connected the analytic continuation of f is unique.

Example 9.10. The power series ∞ z k has radius of convergence 1 and thus defines a
P
k=0 P
∞ k
holomorphic function f in D(0, 1). But k=0 z = 1/(1 − z) in D(0, 1) and 1/(1 − z) is
holomorphic in C \ {1} so f can be analytically continued to C \ {1}.

Holomorphic functions cannot always be analytically continued but there is a, at least

theoretically, natural way to find the analytic continuation if it exists. Suppose that f is
holomorphic in some open set Ω and let a ∈ Ω. Let r be the distance from a to the boundary
∂Ω. Then, by Theorem 4.7, f is represented by its Taylor series in D(a, r). Now if this series
happens to converge in a bigger disc then f can be analytically continued beyond Ω. On the
other hand, if f can be analytically continued across the boundary points of Ω closest to a,
then the Taylor series will converge in a larger disc by Theorem 4.7.
Possibly unexpected things might happen in connection with analytic continuation. As an
illustration, consider the restriction of Log z to the set {z; −π/2 < Arg z < π}. Obviously it
can be extended across the negative imaginary axis to the third quarter as the principal branch
of the logarithm. But it can also be extended across the negative real axis to the third quarter
by defining it to be log−π/2 z for z in the left half-plane. These two extensions do not agree
in the third quarter but differ by 2πi. How come this doesn’t contradict Corollary 9.8 or the
uniqueness of analytic continuation?

9.3 Counting the number of zeros

Let Ω ⊂ C be a bounded open set and assume that the boundary ∂Ω is a finite union of
piecewise smooth simple closed curves. Then
(
1 if 0 ∈ Ω
Z
1 dz
= . (9.3)
2πi ∂Ω z 0 if 0 ∈
/Ω

This follows from Cauchy’s formula (applied to the function 1) if 0 ∈ Ω and from Cauchy’s
theorem if 0 ∈/ Ω since 1/z then is holomorphic in some open set containing Ω.
One way to think of (9.3) is that the left-hand side counts the number of zeros of the
function f (z) = z in Ω. Another interpretation is that the left-hand side computes the total
number of times the (image under f of the) boundary ∂Ω winds around the origin; this is the
same as the total change in argument, divided by 2π, that a point traveling along (the image
under f of) ∂Ω experiences.
The following result is a generalization of (9.3).

Theorem 9.11 (Argument principle). Let Ω ⊂ C be a bounded open set whose boundary ∂Ω
consists of a finite number of piecewise smooth simple closed curves. If f is holomorphic in an
Analytic function theory 45

open set containing Ω and f (z) 6= 0 for z ∈ ∂Ω, then

f 0 (z) dz
Z
1
(9.4)
2πi ∂Ω f (z)

equals the number of zeros of f in Ω counted with multiplicity.

Proof. First notice that the set of zeros of f in Ω must be finite because if not, then, since Ω
is bounded, the zero set of f in Ω would have a limit point in Ω by the Bolzano-Weierstrass
theorem (google it!) and so f would be identically zero in at least one connected component
of Ω. But this is impossible since f (z) 6= 0 for z ∈ ∂Ω. Let a1 , . . . , an be the distinct zeros of f
in Ω and let their multiplicities by m1 , . . . , mn , respectively.
If r > 0 is sufficiently small then the discs D(a1 , r), . . . , D(an , r) are pairwise disjoint and
contained in Ω. Then f 0 (z)/f (z) is holomorphic in an open set containing Ω \ ∪k D(ak , r) and
so Cauchy’s theorem shows that
n
f 0 (z) dz f 0 (z) dz
Z Z
1 1 X
= .
2πi ∂Ω f (z) 2πi ∂D(ak ,r) f (z)
k=1

To compute the right-hand side we note that by Proposition 9.5, f (z) = (z − ak )mk · gk (z) in
D(ak , r) where gk is holomorphic and gk (ak ) 6= 0. Shrinking r if necessary we may assume that
gk 6= 0 in D(ak , r). Thus

f 0 (z) mk (z − ak )mk −1 gk (z) + (z − ak )mk gk0 (z) 1 gk0 (z)

= = mk +
f (z) (z − ak )mk gk (z) z − ak gk (z)

in D(ak , r) \ {ak } and so we get

f 0 (z) dz gk0 (z) dz

Z Z Z
1 mk dz 1
= + = mk ,
2πi ∂D(ak ,r) f (z) 2πi ∂D(ak ,r) z − ak 2πi ∂D(ak ,r) gk (z)

where the last equality follows from Cauchy’s formula (applied to the function 1) and Cauchy’s
theorem (applied to g 0 /g) noticing that g 0 /g is holomorphic in an open set containing D(ak , r).
Hence, (9.4) equals m1 + · · · + mn and the theorem follows.

We’ll say some words hopefully explaining the name “Argument principle”; we begin with

Example 9.12. Let Ω be the unit disc and f (z) = z n . Then ∂Ω is the unit circle and γ(t) = eit ,
0 ≤ t ≤ 2π, is a simple parametrization of it. The composition f ◦ γ(t) is a parametrization of
f (∂Ω) which again is the unit circle. But as f ◦ γ(t) = eint , 0 ≤ t ≤ 2π, this parametrization
is not simple; f ◦ γ(t) runs around the unit circle n times as t runs through [0, 2π]. Thus the
total change of argument of f ◦ γ(t) is 2πn. Hence, the number of zeros of f in Ω (= n) is 1/2π
times the total change in argument of f (z) as z traverses a curve enclosing the zeros of f .

With this example in mind we consider a more general Ω and f . Assume for simplicity
that ∂Ω consists of one simple smooth closed curve and let γ : [0, 1] → C be a simple smooth
parametrization of it. Then f ◦ γ is a parametrization of the curve f (∂Ω) but, as we have seen,
the parametrization need not be simple; f ◦γ(t) might run through f (∂Ω) several times as t runs
through [0, 1]. Let m be the number of times f ◦γ(t) runs around f (∂Ω) as t runs through [0, 1].
Notice that 2πm is the total change in argument that f ◦ γ(t) experiences as t runs through
[0, 1], cf. the paragraph preceding Theorem 9.11. We claim that m = m1 + · · · + mn , where mj
Analytic function theory 46

are the multiplicities of the zeros of f in Ω. To see this, notice first that since w(t) = f ◦ γ(t)
is an “m-fold” parametrization of f (∂Ω) we have
1
(f ◦ γ)0 (t) dt
Z Z
1 m dw
= = m.
2πi t=0 f ◦ γ(t) 2πi f (∂Ω) w

On the other hand, by the Chain rule (see Exercise 2.7), we have (f ◦ γ)0 (t) = f 0 (γ(t)) · γ 0 (t),
and so
(f ◦ γ)0 (t) dt
Z 1 0
f (γ(t)) · γ 0 (t) dt f 0 (z) dz
Z 1 Z
1 1 1
= = .
2πi t=0 f ◦ γ(t) 2πi t=0 f (γ(t)) 2πi ∂Ω f (z)
The claim thus follows from Theorem 9.11.
The next result, Rouche’s theorem, which is a consequence of the Argument principle, is
often useful to compute/estimate the number of zeros in concrete examples.

Theorem 9.13 (Rouche’s theorem). Let Ω ⊂ C be a bounded open set whose boundary ∂Ω
consists of a finite number of piecewise smooth simple closed curves. Let f and g be holomorphic
in an open set containing Ω and assume that |f (z)| > |g(z)| for z ∈ ∂Ω. Then f and f + g
have the same number of zeros in Ω, multiplicities taken into account.

Proof. Consider the function

f 0 (z) + tg 0 (z)
Z
1
ϕ(t) := dz
2πi ∂Ω f (z) + tg(z)

defined for t ∈ [0, 1]. Notice that |f (z) + tg(z)| ≥ |f (z)| − t|g(z)| ≥ |f (z)| − |g(z)| > 0 for
all (z, t) ∈ ∂Ω × [0, 1]. It follows that there is some c > 0 such that |f (z) + tg(z)| ≥ c for all
(z, t) ∈ ∂Ω×[0, 1] (why?). By the Argument principle, ϕ(t) is the number of zeros of f (z)+tg(z)
in Ω, counting multiplicities. In particular, ϕ(0) is the number of zeros of f in Ω and ϕ(1) is
the number of zeros of f + g in Ω.
We claim that ϕ is continuous. Given the claim we have that ϕ is a continuous function on
[0, 1] taking integer values. But then ϕ must be constant by the Intermediate value theorem
and so ϕ(0) = ϕ(1). It remains to show the claim. To do this we make a straightforward
computation to get
Z  0
f (z) + tg 0 (z) f 0 (z) + t0 g 0 (z)

1
ϕ(t) − ϕ(t0 ) = − dz
2πi ∂Ω f (z) + tg(z) f (z) + t0 g(z)
t − t0 g 0 (z)f (z) − f 0 (z)g(z)
Z
=

dz.
2πi ∂Ω f (z) + tg(z) f (z) + t0 g(z)

Since g 0 (z)f (z) − f 0 (z)g(z) is continuous there is a C > 0 such that |g 0 (z)f (z) − f 0 (z)g(z)| ≤ C
for all z ∈ ∂Ω and so the absolute value of the integrand is ≤ C/c2 for all (z, t) ∈ ∂Ω × [0, 1].
By Proposition 2.7 we thus see that ϕ(t) − ϕ(t0 ) = O(|t − t0 |) and the claim follows.

Example 9.14. Find the number of zeros of z 5 − 3z + 1 in the unit disc.

Solution: Set f (z) = −3z + 1 and g(z) = z 5 . Clearly, the only zero of f is z = 1/3 which is in
D(0, 1). For z on the boundary of D(0, 1), i.e., |z| = 1, we have

|f (z)| = | − 3z + 1| ≥ | − 3z| − |1| = 2 > 1 = |z|5 = |g(z)|.

By Rouche’s theorem, f (z) and f (z) + g(z) = z 5 − 3z + 1 have the same number of zeros in
D(0, 1) and so z 5 − 3z + 1 has one zero in D(0, 1).
Analytic function theory 47

Exercises
9.1 Show, using Identity theorem I or II, that

(a) sin(2z) = 2 sin z cos z, (b) cos(2z) = cos2 z − sin2 z,

for all z ∈ C.

9.2 Find the zeros (cf. Exercise 5.10) and their multiplicities of
2
(a) (z 2 − 1) sin(πz), (b) ez − 1, (c) e2z + 1, (d) cosh100 z, (e) cos2 z − 4.

9.3 Let {zn } ⊂ D(0, 1) be a sequence of points 6= 0 such that zn → 0 as n → ∞. Is there a

holomorphic function f in D(0, 1) such that f (zn ) = (−1)n zn for all n?

9.4 Is there a holomorphic function in Ω such that f (0) = 1 and f (1 − 1/2n) = 0 for
n = 1, 2, . . . if (a) Ω = C, (b) Ω is the unit disc?

9.5 Find all holomorphic functions f in C satisfying f (1/n) = n2 f (1/n)3 for n = 1, 2, . . ..

9.6 Find the number of zeros of

(a) 5z 6 + z 4 − z + 2 in the unit disc,

(b) z 4 + z 2 + 3z + 5.5 in the annulus {z; 1 < |z| < 2},
(c) z 4 + z 2 + 5z in the annulus {z; 1 < |z| < 2},
(d) cos(πz) − 100z 100 in the unit disc,
(e) z 5 − 5z + sin z in the annulus {z; 1 < |z| < 2}.
N
X zk
9.7 Let R be a given positive number. Show that 6= 0 for all z ∈ D(0, R) if N is
k!
k=0
sufficiently large.
Analytic function theory 48

10 Some topology and jazzed-up Cauchy theorems

One important application of Cauchy’s theorem is that when computing contour integrals of
holomorphic functions, the integration contour may often be changed into a simpler one without
affecting the integral. We saw one striking application of this in the proof of the Argument
principle, Theorem 9.11. In this section we’ll look more closely at how integration contours can
be changed with out affecting “Cauchy integrals”. We’ll also get a glimpse of the connection
between topological properties of an open set and analytic properties of holomorphic functions
living in that set.

10.1 Deformation and homotopy

Let Ω ⊂ C be open and let γ0 and γ1 be parametrized closed curves in Ω. A closed curve
can be parametrized by the interval [0, 1] and in this chapter we will tacitly assume that all
parametrized closed curves are parametrized by [0, 1]. The curves γ0 and γ1 are said to be
homotopic in Ω if there is a continuous map H : [0, 1] × [0, 1] → Ω such that
H(s, 0) = γ0 (s), H(s, 1) = γ1 (s), H(0, t) = H(1, t) ∀t ∈ [0, 1].
The last condition ensures that s 7→ H(s, t) is a closed curve. One should think of the curves
s 7→ H(s, t) as interpolating continuously between γ0 and γ1 in Ω as t goes from 0 to 1, i.e.,
that γ0 is continuously deformed inside Ω to γ1 as t goes from 0 to 1. We will call the map H
a homotopy between γ0 and γ1 .
The notion of two curves being homotopic turns out to be an equivalence relation on the
set of parametrized closed curves in Ω and thus gives a partition of this set into equivalence
classes, called homotopy classes. A connected open set such that all curves are homotopic, i.e.,
the number of homotopy classes is 1, is called simply connected. Intuitively one can think of a
simply connect open set as having “no holes”.
If a is a point in an open set Ω then the constant map γ(s) = a for all s ∈ [0, 1] is a
parametrized closed curve in Ω. A parametrized closed curve in Ω is said to be null-homotopic in
Ω if it is homotopic to a point in Ω, viewed as such a trivial curve. Notice that an open connected
set is simply connected if and only if every parametrized closed curve is null-homotopic.
Example 10.1. Let Ω be an open convex set. Then Ω is simply connected. In fact, let γ be
a closed parametrized curve in Ω and let a ∈ Ω. Since Ω is convex, the line segment between
a and γ(s) is contained in Ω for all s ∈ [0, 1], and so H(s, t) := γ(s) · (1 − t) + ta ∈ Ω for all
(s, t) ∈ [0, 1]×[0, 1]. Since H clearly depends continuously on (s, t), H(s, 0) = γ(s), H(s, 1) = a,
and H(0, t) = γ(0)(1 − t) + ta = γ(1)(1 − t) + ta = H(1, t) it follows that H : [0, 1] × [0, 1] → Ω
is a homotopy between γ and the trivial curve a.
If f is holomorphic in Ω and γ is a parametrized smooth closed curve in Ω we set
Z Z 1
f γ(t) γ 0 (t)dt;


f (z) dz :=
γ t=0

if γ is merely piecewise smooth we divide γ into smooth pieces and add up the integrals. We
call such integrals Cauchy integrals; the content of the following result is that Cauchy integrals
only depend on the homotopy class of the curve.
Theorem 10.2 (Cauchy’s theorem, homotopy version). Let Ω ⊂ C be an open set and let γ0 och
γ1 be homotopic parametrized closed piecewise smooth curves in Ω. Then, for any holomorphic
function f in Ω, Z Z
f (z) dz = f (z) dz.
γ0 γ1
Analytic function theory 49

R
In particular, if Ω is simply connected, then γ f (z) dz = 0 for any holomorphic function f in
Ω and any parametrized closed piecewise smooth curve γ in Ω.
R
Sketch of proof. The last statement follows from the first since γ f (z) dz = 0 for any function
if γ is constant.
To prove the first we take a homotopy H between γ0 and γ1 . We divide [0, 1] into n pieces
of equal length. This gives us a grid of (n +
1) · (n + 1) points (i/n, j/n), i, j = 0, . . . , n, in
[0, 1] × [0, 1]. Consider the images H i/n, j/n in Ω of these points. Let `ij , for
i = 0, 1, . . . , n
and j = 0, 1, . . .
, n − 1, be the oriented line segment starting at H i/n, j/n and ending at
H i/n, (j + 1)/n ; notice that `0j = `nj . Let Γij ,
for i = 0, 1, . . . , n − 1 and j = 1, .
. . , n − 1, be
the oriented line segment starting at H i/n, j/n and ending at H (i + 1)/n, j/n ; we let Γi0
be the curve-piece γ0 ([i/n, (i + 1)/n]) and Γin be the curve-piece γ1 ([i/n, (i + 1)/n]). Since H
is continuous and Ω is open `ij and Γij are contained in Ω if n is large enough. In fact, since
H([0, 1] × [0, 1]) is a compact subset of Ω (why?) it has a positive distance to ∂Ω and so we
may choose n so large that all `ij and Γij are contained in Ω for all i and j.
We get oriented closed curves Pij , for i = 0, . . . , n − 1 and j = 0, . . . , n − 1 by joining Γij ,
`i+1,j , −Γi,j+1 , and −`ij in this order. It is an exercise (do it!) to use Cauchy’s theorem and
check that Z
f (z) dz = 0, ∀i, j = 0, . . . , n − 1.
Pij

We also get oriented closed curves Γj , for j = 0, 1, . . . , n, by joining Γ0j , Γ1j , . . . , Γn−1,j in
this order. Notice that Γ0 and Γn are (the images of) γ0 and γ1 respectively. The theorem will
follow if we can show that Z Z
f (z) dz − f (z) dz = 0, (10.1)
Γj+1 Γj

for j = 0, . . . , n − 1. To show (10.1) we add and subtract integrals along the lines `ij :
Z Z n−1
XZ Z Z Z
f (z) dz − f (z) dz = f (z) dz − f (z) dz + f (z) dz − f (z) dz
Γj+1 Γj i=0 Γi,j+1 Γij `ij `i+1,j
n−1
XZ Z
− f (z) dz − f (z) dz
i=0 `ij `i+1,j
Z
= − f (z) dz − 0 = 0,
Pij

Pn−1 R
where the second last equality follows since `0j = `nj , which makes the sum i=0 `ij f (z) dz −
R
`i+1,j f (z) dz telescoping.

10.2 Winding numbers and homology

In the previous section we
Z saw that Cauchy integrals only depend on the homotopy class of
the curve. In particular, f (z)dz = 0 if γ is null-homotopic in the set where f is holomorphic.
γ
But homotopy is not the whole story. If Ω is C with two distinct points removed then one can
find a curve
Z γ in Ω (google “null homologous” and “picture”) that is not null-homotopic in Ω,
but still f (z)dz = 0 for any holomorphic function f i Ω. The “reason” for this is that the
γ
total number of times that γ winds around any of the points outside of Ω is 0. This is loosely
speaking the content of the homology version of Cauchy’s theorem.
Analytic function theory 50

P
Let Ω ⊂ C be an open set. A cycle in Ω is a formal finite sum Γ = k nk Ck where nk ∈ Z
and Ck is a piecewise smooth oriented closed curve in Ω. The support, |Γ|, of Γ is the union of
the Ck ’s. If f is a continuous function on |Γ| then we set
Z X Z
f (z) dz := nk f (z) dz.
Γ k Ck

P P
We let `(Γ) := k |nk |`(Ck ) be the length of Γ. One should think of the cycle Γ = k nk Ck as
the curves Ck run through nk times; if nk is negative this means that Ck should be run through
|nk | times in the direction opposite to the one given by the orientation.

Example 10.3. Let Ω be an open set with piecewise smooth boundary ∂Ω. Then ∂Ω, oriented
as a boundary, (i.e., Ω is on the left-hand side) is a cycle in C. If f is holomorphic in an open
set containing Ω, then f (∂Ω) has a natural structure as a cycle. For instance, if Ω = D(0, 1)
and f (z) = z n then f (∂D(0, 1)) = n∂D(0, 1).

Definition 10.4. If Γ is a cycle in C and a ∈ C \ |Γ| we define the winding number, or index,
denoted IndΓ (a), of Γ with respect to a as
Z
1 dz
IndΓ (a) = . (10.2)
2πi Γ z − a

The interpretation of IndΓ (a) is as the total number of times that Γ winds around a. Suppose
that Γ = ∂Ω for some open set Ω with piecewise smooth boundary. Then by Theorem 9.11
applied to f (z) = z − a it follows that the right-hand side of (10.2) equals the number of zeros
of f in Ω. This number is, in view of the discussion after Example 9.12, the number of times
f (∂Ω) winds around 0. Since f (z) = z − a is a translation, this number in turn is the same as
the number of times ∂Ω winds around a. Hence, the interpretation of IndΓ (a) is justified in the
case Γ = ∂Ω. For a general cycle Γ the justification is contained in the proof of the following
result.

Theorem 10.5. Let Γ be a cycle in C.

(a) For each a ∈ C \ |Γ|, IndΓ (a) is an integer.

(b) The function a 7→ IndΓ (a), defined in C \ |Γ|, is constant on each connected component
of C \ |Γ|. Moreover, IndΓ (a) = 0 if |a| is sufficiently large.

Proof. To prove (a) we may assume that Γ is aR piecewise smooth oriented closed curve (why?)
and that a = 0; we should then show that Γ dz/z is 2πi times an integer. Choose points
w1 , w2 , . . . , wN on Γ and discs D(wj , r), j = 1, . . . , N − 1, such that (i) w1 = wN , (ii) Γ ⊂
∪j D(wj , r), (iii) 0 ∈/ ∪j D(wj , r), (iv) wj+1 ∈ D(wj , r) for j = 1, . . . , N −1, and (v) the traveling
direction, given by the orientation, of the piece of Γ between wj and wj+1 , inside D(wj , r) is
from wj to wj+1 .
Since 0 ∈ / D(wj , r) we may choose a branch logϕj of the logarithm that is holomorphic in
D(wj , r); recall that (logϕj z)0 = 1/z where it is defined. Let Γj be the part of Γ between wj
Analytic function theory 51

and wj+1 inside D(wj , r). In view of Proposition 3.4 we get

Z N −1 Z N −1 Z N −1
dz X dz X
0
X
= = (logϕj z) dz = logϕj wj+1 − logϕj wj
Γ z Γj z Γj
j=1 j=1 j=1
N
X −1
= log |wj+1 | + i argϕj wj+1 − log |wj | − i argϕj wj
j=1
N
X −1
= i argϕj wj+1 − argϕj wj
j=1
N
X
= i argϕj−1 wj − argϕj wj .
j=2

But argϕj−1 wj − argϕj wj is 2π times some integer and so part (a) of the theorem follows.
To show part (b) we notice that IndΓ (a) depends continuously on a in C \ |Γ| by Ex-
ercise 10.1. Thus, IndΓ (a) is a continuous integer-valued function in C \ |Γ|. It follows that
IndΓ (a) is constant on each connected component of C \ |Γ| (why?). If |a| > 2 supz∈|Γ| |z|, then
|a|/2 − |z| > 0 for all z ∈ |Γ|, and so |z − a| ≥ |a| − |z| = |a|/2 + |a|/2 − |z| > |a|/2. Hence,
 Z 
IndΓ (a) =  1 dz  1 `(Γ)
  
≤
Γ z−a 2π |a|/2
 2πi 

by Proposition 2.7. For |a| sufficiently large, `(Γ)/π|a| < 1. Thus, since IndΓ (a) is an integer,
IndΓ (a) = 0 if |a| is sufficiently large.

A cycle Γ in an open set Ω is said to be null-homologous in Ω if IndΓ (a) = 0 for all a ∈ C\Ω;
two cycles Γ1 and Γ2 are said to be homologous in Ω if Γ1 − Γ2 is null-homologous. The notion
of two cycles being homologous in Ω is an equivalence relation on the set of cycles in Ω and
the corresponding equivalence classes are called homology classes.
The content of the homology version of Cauchy’s theorem below is that Cauchy integrals
only depend on the homology class. Notice that if C is a null-homotopic piecewise smooth
oriented closed curve in an open set Ω, then C is null-homologous in Ω. In fact, 1/(z − a) is
holomorphic in C \ {a} and C is null-homotopic in C \ {a} so IndC (a) = 0 by the homotopy
version of Cauchy’s theorem.

Remark 10.6. Our definition of homology is not the standard one. However it is equivalent
to the standard definition, see M. Andersson’s book “Topics in complex analysis”.

Theorem 10.7 (Cauchy’s theorem, homology version). Let Ω ⊂ C be an open set and let Γ
be a null-homologous cycle in Ω. Then, for any holomorphic function f in Ω,
Z
f (z) dz = 0.
Γ

Sketch of proof. 5 Let ϕ be a C 1 -smooth function in C such that ϕ(z) = 0 if |z| is suffici-
ently large and notice that IndΓ (z) is a bounded function defined almost everywhere in C by
5
This proof sketch can be made rigorous by using some distribution theory and/or integration theory.
Analytic function theory 52

Theorem 10.5. In view of Exercise 3.9 we get

ZZ ZZ Z
∂ϕ 1 dw ∂ϕ
2i IndΓ (z) dxdy = 2i dxdy
C ∂ z̄ C 2πi Γ w − z ∂ z̄
Z ZZ
1 ∂ϕ dxdy
= − dw
Γ π C ∂ z̄ z − w
Z
= ϕ(w) dw.
Γ

Let f be a holomorphic function in Ω and let χ be a smooth function in C such that

χ(z) = 1 if |z| ≤ R and χ(z) = 0 if |z| ≥ R + 1, where R is to be chosen suitably. Replacing ϕ
in the preceding computation by f χ we get
ZZ Z
∂χ
2i IndΓ (z) f (z) dxdy = f (w)χ(w) dw. (10.3)
C ∂ z̄ Γ

By Theorem 10.5, IndΓ (z) = 0 if |z| > R0 for R0 large enough, and so the left-hand side of
(10.3) is 0 if R > R0 . ROn the other hand, if R is so big that |Γ| ⊂ D(0, R), then the right-hand
side of (10.3) equals Γ f (w) dw. The theorem thus follows.

10.3 Properties of simply connected open sets

We’ll see that in simply connected open sets the theory of holomorphic functions is particularly
well behaved.
Theorem 10.8. Let Ω ⊂ C be an open connected set. If Ω is simply connected then
(a) Indγ (a) = 0 for any parametrized piecewise smooth closed curve γ in Ω and any a ∈
/ Ω,
Z
(b) f (z)dz = 0 for any cycle Γ in Ω and any holomorphic function f in Ω,
Γ

(c) each holomorphic function f in Ω has a holomorphic primitive in Ω, i.e., there is a

holomorphic F in Ω such that F 0 = f ,

(d) each holomorphic function f in Ω such that f (z) 6= 0 for all z ∈ Ω has a holomorphic
logarithm in Ω, i.e., there is a holomorphic g in Ω such that f (z) = eg(z) and g 0 (z) =
f 0 (z)/f (z) for all z ∈ Ω.
We will prove that if Ω is simply connected then property (a) holds. Then we show that
(a) ⇒ (b) ⇒ (c) ⇒ (d) without assuming that Ω is simply connected. It is also true that (d)
implies that Ω is simply connected, see, e.g., Andersson. Hence, any one of the properties (a)
– (d) is in fact equivalent to that Ω is simply connected.

Proof. If Ω is simply connected then (a) follows as in the paragraph preceding Remark 10.6
since any closed curve in Ω is null-homotopic.
Assume that (a) holds. Then every parametrized piecewise smooth closed curve in Ω is
null-homologous and it follows that any cycle in Ω is null-homologous. Hence, (b) follows from
the homology version of Cauchy’s theorem.
R let f be a holomorphic function in Ω. Fix some point a ∈ Ω and
Assume that (b) holds and
set, for any z ∈ Ω, F (z) = γ(a,z) f (z)dz, where γ(a, z) is a piecewise smooth curve starting at
a and ending at z; since (b) holds, F (z) is independent of the choice of γ(a, z) (why?). As in
the proof of Morera’s theorem we then see that F is holomorphic and that F 0 = f . Hence, (c)
holds.
Analytic function theory 53

Assume that (c) holds and let f be a holomorphic function in Ω such that f (z) 6= 0 for all
z ∈ Ω. Then f 0 /f is holomorphic in Ω and there is a holomorphic function g̃ in Ω such that
g̃ 0 = f 0 /f . Hence,
 g̃ 0
e eg̃ g̃ 0 f − eg̃ f 0 eg̃ f 0 − eg̃ f 0
= = = 0,
f f2 f2
and so eg̃ /f = C for some (necessarily non-zero) constant C by Proposition 3.6. Choosing c
such that C = ec we get eg̃−c = f and we may thus take g := g̃ − c to see that (d) holds.

Exercises
10.1 Let Γ be a cycle in C, let a ∈ C \ |Γ|, and let d be the distance between a and |Γ|. Show,
  `(Γ)
e.g., using Proposition 2.7, that IndΓ (a + h) − IndΓ (a) ≤ |h| 2 if |h| ≤ d/2.
πd
Analytic function theory 54

11 Some mapping properties of holomorphic functions

11.1 The Maximum principle and Schwarz’s lemma
The modulus of a holomorphic function cannot have a local maximum unless the function is
constant. Another way to say this is that the modulus of a holomorphic function is maximal
on the boundary. This is the content of the Maximum principle, a.k.a. the Maximum modulus
principle. We begin with a local version.

Proposition 11.1 (Maximum principle, local version). If f is holomorphic in a disc D(a, R)

and |f (z)| ≤ |f (a)| for all z ∈ D(a, R), then f is constant.

Proof. By expressing the integral in Cauchy’s formula using the natural parametrization of the
circle ∂D(a, r) we get
Z 2π
1
f (a) = f (a + reit )dt, for r < R,
2π 0

see Exercise 3.10. Hence, we get

2π 2π
Z Z 
it
 
0 = 2π|f (a)| − 2π|f (a)| = |f (a)| dt −  f (a + re )dt (11.1)
0 0
Z 2π Z 2π
≥ |f (a)|dt − |f (a + reit )|dt
0 0
Z 2π
= |f (a)| − |f (a + reit )| dt ≥ 0,
0

where the last step follows since |f (a)| ≥ |f (a + reit )| for all r < R and all t by assumption.
Thus the inequalities
R 2π in this computation must be equalities. But since |f (a)|−|f (a+reit )| ≥ 0,
we cannot have 0 |f (a)| − |f (a + re )| dt = 0 unless |f (a)| − |f (a + reit )| = 0. Hence, |f | is
it

constant in D(a, R) and so f is constant in D(a, R) by Proposition 3.6.

Theorem 11.2 (Maximum principle, global version). Let Ω ⊂ C be an open bounded set. Let
f be holomorphic in Ω and assume that f extends to a continuous function on Ω. Then |f |
attains its maximum on ∂Ω, i.e., sup |f (z)| = sup |f (z)|.
z∈Ω z∈∂Ω

Proof. Since |f | is a continuous function on the compact set Ω there is an a ∈ Ω such that
|f (a)| = supz∈Ω |f (z)|. If a ∈ ∂Ω = Ω \ Ω we are done, so assume that a ∈ Ω. Since Ω is open
there is a disc D(a, R) ⊂ Ω and so, by the local version of the Maximum principle, f is constant
in D(a, R). But then, by Corollary 9.8, f is constant in the connected component Ω0 of Ω that
contains a. Thus, f (z) = f (a) for any boundary point of Ω0 ; such points are also boundary
points of Ω and so there are boundary points of Ω where |f | attains its maximum.

One application of the Maximum principle is Schwarz’s lemma, which in particular will
enable us to determine all bijective holomorphic maps from the unit disc to itself.

Theorem 11.3 (Schwarz’s lemma). Let f be holomorphic in the disc D(0, R) and assume that
f (0) = 0 and that |f (z)| ≤ M for all z ∈ D(0, R). Then |f (z)| ≤ M |z|/R for all z ∈ D(0, R)
and |f 0 (0)| ≤ M/R. Moreover, if |f (z)| = M |z|/R for some z ∈ D(0, R) \ {0} or if |f 0 (0)| =
M
M/R, then f (z) = z eiθ for some constant θ ∈ R.
R
Analytic function theory 55

Proof. We may assume that M = R = 1; otherwise consider the function f˜(z) = f (zR)/M .
Since f (0) = 0 we can write f (z) = zg(z) for some g holomorphic in D(0, 1), cf. Propo-
sition 9.5. Notice that f 0 (0) = g(0). For any r < 1, g is continuous on D(0, r) and |g(z)| =
|f (z)|/|z| ≤ 1/r if |z| = r. Thus, by the Maximum principle, |g(z)| ≤ 1/r in D(0, r). Letting
r → 1− it follows that |g(z)| ≤ 1 in D(0, 1), and so |f (z)| ≤ |z| in D(0, 1).
Assume that |f (z)| = |z| for some z ∈ D(0, 1) \ {0} or that |f 0 (0)| = 1. By the first part of
the proof, |g| then has a maximum = 1 in D(0, 1) and so by the local version of the Maximum
principle g is a constant, which has to have modulus 1. Thus, g(z) = eiθ for some θ ∈ R, and
the last part of the theorem follows.

Notice that the hypothesis of Schwarz’s lemma is that f is a holomorphic function from
the disc D(0, R) to the disc D(0, M ) such that f (0) = 0. If f (0) = a 6= 0 we can still get some
information from Schwarz’s lemma. Set α = a/M ∈ D(0, 1) and let
z−α
φα (z) = . (11.2)
ᾱz − 1
One can check that
φα is a bijective map from D(0, 1) to itself (Exercise 11.3). The function
h(z) := φα f (z)/M then is a holomorphic map from D(0, R)
to D(0, 1) such that h(0) = 0.
Schwarz’s lemma thus gives |h(z)| ≤ |z|/R, i.e., |φα f (z)/M | ≤ |z|/R, which at least says
something about f .
Let us also show how Schwarz’s lemma can be used to prove that any bijective holomorphic
map from the unit disc to itself, i.e., a holomorphic automorphism of D(0, 1), is of the form
φα for some α ∈ D(0, 1), up to a multiplicative constant of modulus 1. Let f be a holomorphic
automorphism of D(0, 1). Suppose first that f (0) = 0. Then Schwarz’s lemma gives |f (z)| ≤ |z|
for all z ∈ D(0, 1). On the other hand, Schwarz’s lemma applied to f −1 gives |f −1 (z)| ≤ |z| for
all z ∈ D(0, 1). Hence, |z| = |f −1 (f (z))| ≤ |f (z)| ≤ |z| for all z ∈ D(0, 1) and so the inequalities
must be equalities. From Schwarz’s lemma again it follows that f (z) = eiθ z = ei(π+θ) φ0 (z) for
some θ ∈ R. If f (0) = a 6= 0, then φa ◦ f is an automorphism of D(0, 1) that maps 0 to 0. From
what we have just seen, φa ◦ f (z) = eiθ z for some θ ∈ R and so, since φ−1 a = φa (Exercise 11.3),

f (z) = φa (eiθ z) = eiθ φe−iθ a (z);

the last equality is a simple exercise.

11.2 Three basic mapping theorems and the Riemann mapping theorem
Recall from Proposition 3.6 and Exercise 3.7 that if the image of a holomorphic function is
contained in a circle or a line then the function must be constant. This suggests that non-
constant holomorphic functions cannot have “thin” images. In fact, the next result says that
the image of a non-constant holomorphic function necessarily is open.

Theorem 11.4. Let f be holomorphic in an open connected set Ω ⊂ C. Then either f is

constant or f (Ω) is open.

Proof. Assume that f is not constant. Let a ∈ Ω be an arbitrary point; we will show that there
is a disc centered at f (a) contained in f (Ω).
We may assume that f (a) = 0 (possibly after replacing f by f − f (a)). Since f is not
constant, a is an isolated zero of f by the Identity theorem II and so we may choose a disc
D(a, r) ⊂ Ω such that f (z) 6= 0 for z ∈ ∂D(a, r). Let  := inf z∈∂D(a,r) |f (z)|; since |f | is
continuous and non-zero on the compact set ∂D(a, r), we have  > 0. Consider the disc D(0, )
and let w ∈ D(0, ). Let g be the constant function g(z) = −w. For z ∈ ∂D(a, r) we have
Analytic function theory 56

|f (z)| ≥  > |w| = |g(z)| and so by Rouche’s theorem f and f + g have the same number
of zeros in D(a, r). Thus, since f has a zero in D(a, r), there is a b ∈ D(a, r) such that
f (b) + g(b) = 0. Hence, f (b) = w and it follows that D(0, ) ⊂ f (D(a, r)) ⊂ f (Ω).

One can formulate Theorem 11.4 in terms of solvability of equations as follows: Let f : Ω →
C be holomorphic and non-constant in the connected open set Ω. If there is a solution to
f (z) = a in Ω, then there is a solution in Ω to f (z) = a0 if |a − a0 | is sufficiently small.
With some variation of the proof of Theorem 11.4 we also get

Theorem 11.5. If f is an injective holomorphic function defined in some open set Ω ⊂ C,

then f is conformal in Ω, i.e., f 0 (z) 6= 0 for all z ∈ Ω.

Proof. Assume, to get a contradiction, that there is a point a ∈ Ω such that f 0 (a) = 0. We
may assume that f (a) = 0 (possibly after replacing f by f − f (a)). Since f is injective, f
cannot be identically zero in any disc centered at a and so, by the Identity theorem I, a is a
zero of some finite order m; notice that then a is a zero of f 0 of order m − 1 ≥ 1. Thus, by
Proposition 9.5, there is a disc D(a, r) ⊂ Ω such that a is the only zero of f and f 0 in D(a, r).
Set  := inf z∈∂D(a,r) |f (z)|. As in the previous proof,  > 0 and, for any w ∈ D(0, ), f (z) and
f (z) − w have the same number of zeros in D(a, r). Take w0 ∈ D(0, ) \ {0}. Since f has a zero
of order m ≥ 2 at a, f (z) − w0 must have at least two zeros in D(a, r) taking multiplicity into
account. Now, f (z) − w0 cannot have two distinct zeros since f is injective and so f (z) − w0
must have a zero of multiplicity at least two at some b ∈ D(a, r); notice that b ∈ D \ {a} since
f (a) − w0 = w0 6= 0. But then f 0 (b) = 0, which contradicts that a is the only zero of f 0 in
D(a, r).

Theorem 11.6 (Inverse function theorem). Let f be an injective holomorphic function defined
in some open set Ω ⊂ C. Then f has a holomorphic inverse f −1 : f (Ω) → Ω.

Proof. Since f is injective it is set-theoretic nonsense that there is a unique inverse f −1 : f (Ω) →
Ω; we need to show that f −1 is holomorphic.
Let w0 ∈ f (Ω) and let z0 = f −1 (w0 ). Since f is injective f 0 (z0 ) 6= 0 by Theorem 11.5 and so
the differential of f at z0 is non-zero, cf. (2.9). It thus follows from the Inverse function theorem
of calculus that f −1 is C 1 -smooth close to w0 . To show that f −1 is complex differentiable at w0
note that if wj is a sequence converging to w0 then zj := f −1 (wj ) converges to z0 by continuity.
Hence,

f −1 (w) − f −1 (w0 ) f −1 (f (z)) − f −1 (f (z0 )) z − z0 1

lim = lim = lim = 0 ,
w→w0 w − w0 z→z0 f (z) − f (z0 ) z→z0 f (z) − f (z0 ) f (z0 )

where we compute the limit by considering sequences {wj } such that wj 6= w0 ; this ensures
that f (zj ) − f (z0 ) 6= 0 since f is injective.

If f is an injective holomorphic function defined in some open Ω, f −1 thus is a holomorphic

function in f (Ω), which is open by Theorem 11.4, at least if Ω is connected. Moreover, by
Theorem 11.5, both f and f −1 are conformal. We say that two open sets are conformally
equivalent if there is a bijective holomorphic map between them. In Chapter 7 we indicated
the importance of finding conformal mappings from complicated open sets to simpler ones and
we also looked at some techniques to construct such maps. Given a quite general open set it
is very hard to construct an explicit conformal mapping onto some substantially simpler one,
but, in fact, theoretically it is usually possible. The first and main result in this direction is
the Riemann mapping theorem.
Analytic function theory 57

Theorem 11.7 (Riemann mapping theorem). Let Ω ⊂ C, Ω 6= C, be an open simply connected

set. Then Ω is conformally equivalent to the unit disc.

A proof of this is beyond the scope of these notes but can be found in, e.g., M. Andersson’s
book “Topics in complex analysis”. However, in Chapter 14 below we will make it physically
reasonable that the Riemann mapping theorem holds.
Notice that C cannot be conformally equivalent to the unit disc because if it where, then
there would be a bijective holomorphic map f : C → D(0, 1) and this is not possible by Liou-
ville’s theorem.
In view of the Riemann mapping theorem the hypothesis in Schwarz’s lemma becomes less
restrictive than it might look. To be more specific, let f be a holomorphic function in an open
simply connected set Ω 6= C and assume that f (Ω) 6= C. By the Riemann mapping theorem
there are bijective holomorphic maps ϕ : D(0, 1) → Ω and ψ : f (Ω) → D(0, 1) and so ψ ◦ f ◦ ϕ
is a map from the unit disc to itself. By composing further with suitable φα ’s, see (11.2), we
may assume that 0 is mapped to 0. Then the hypothesis of Schwarz’s lemma is satisfied and
one might get some information about f . One illustration of this is outlined in Exercise 11.2.

Exercises
11.1 Let f be a holomorphic function in C such that f (z) ∈ R for all z ∈ ∂D(0, 1).

(a) Show that |e±if (z) | = 1 for z with |z| = 1.

(b) Use the Maximum principle to show that f (z) ∈ R for all z ∈ D(0, 1) and conclude
that f is a constant function.

11.2 Let Π+ be the upper

 half-plane,
 let a ∈ Π+ , and let f : Π+ → Π+ be a holomorphic
 f (z) − f (a)   z − a 
map. Show that 

≤
   for all z ∈ Π+ . (Hint: Consider the composition
 f (z) − f (a)   z − ā 
φ2 ◦ f ◦ φ−1
1 where φ1 (z) = (z − a)/(z − ā) and φ2 (z) = (z − f (a))/(z − f (a)).)

11.3 (a) Show that |z − α| < |ᾱz − 1| if and only if |z|2 (1 − |α|2 ) < 1 − |α|2 , and conclude that
|φα (z)| < 1 if z, α ∈ D(0, 1).
(b) Show that w = φα (z) if and only if z = φα (w).
Analytic function theory 58

12 Singularities and Cauchy’s residue theorem

By singularities we here mean isolated points where a holomorphic function is not defined. We
will see that a holomorphic function is represented by a Laurent series in a neighborhood of a
singularity; a Laurent series is a power series that also contains negative powers. Laurent series
expansions allow us to classify the singularities of holomorphic functions.
Cauchy’s residue theorem is a generalization of Cauchy’s theorem/formula that applies to
holomorphic functions with singularities.

12.1 Laurent series and classification of singularities

Let A be the annulus {z; r < |z| < R} and let f be a holomorphic function in an open set
containing A. For any z ∈ A we have by Cauchy’s formula that
Z Z Z
1 f (w) dw 1 f (w) dw 1 f (w) dw
f (z) = = − (12.1)
2πi ∂A w − z 2πi |w|=R w − z 2πi |w|=r w − z
Z Z
1 f (w) dw 1 f (w) dw
= + .
2πi |w|=R w 1 − z/w 2πi |w|=r z 1 − w/z

and |w/z| < 1 if |w| = r we have 1/(1 − z/w) = ∞ k

P
Since |z/w| < 1 if |w| = R P k=0 (z/w) if
∞ k
|w| = R and 1/(1 − w/z) = k=0 (w/z) if |w| = r. As in the proof of Theorem 4.7 one shows
Z ∞ Z
f (w) dw X f (w)  z k
= dw and
|w|=R w 1 − z/w |w|=R w w
k=0
Z ∞ Z
f (w) dw X f (w)  w k
= dw,
|w|=r z 1 − w/z |w|=r z z
k=0

and so, by (12.1),

∞ Z ∞ Z
1 X k f (w) 1 X −k−1
f (z) = z dw + z f (w)wk dw.
2πi |w|=R wk+1 2πi |w|=r
k=0 k=0

Both of these series converge if r < |z| < R and so it follows from Lemma 4.4 that the first
series is absolutely convergent if |z| < R and the second one is absolutely convergent if |z| > r.
Setting ck := 1/(2πi) C f (w)/wk+1Pdw, where C is any curve such that C − ∂D(0, R) is null-
R

homologous in A, we have f (z) = ∞ k

−∞ ck z and the series converges absolutely in A. This is
the Laurent series of f (centered at 0).
Theorem 12.1 (Laurent’s theorem). Let A = {z ∈ C; r < |z − a| < R} P and let f be a
holomorphic function in an open set containing A. Then there is a unique series ∞−∞ ck (z−a)
k

converging absolutely in A such that

∞
X
f (z) = ck (z − a)k
k=−∞

for z ∈ A. The coefficients are given by

Z
1 f (w) dw
ck = , (12.2)
2πi C (w − a)k+1

where C is any curve such that C − ∂D(a, %), r < % < R, is null-homologous in A.
Analytic function theory 59

Proof. We may assume that P a = 0 kand we have already showed that if ck is defined by (12.2)
(with a = 0) then f (z) = ∞ −∞ ck z for z ∈ A and the series is absolutely convergent in A. It
remainsPto show uniqueness.
Let ∞ k
P∞ k
−∞ dk z be absolutely convergent in A and assume that f (z) = −∞ dk z for z ∈ A.
To see that dk = ck for all k we notice that if r < % < R, then
∞
wk
Z Z
f (w) dw X
2πic` = `+1
= d k `+1
dw = 2πid` ; (12.3)
|w|=% w k=−∞ |w|=% w

the last equality follows from Exercise 2.9 and an argument showing that it is allowed to
interchange the order of integration and summation in the second equality is outlined in Ex-
ercise 12.1.

To find a Laurent series explicitly in a concrete example it is often easier to use some ad
hoc method instead of trying to compute the coefficients by (12.2). If one somehow can cook
up a Laurent-type series representing a given function then, by uniqueness, it has to be the
Laurent series.
1
Example 12.2. Find the Laurent series of f (z) = centered at 1.
z(z − 1)
Solution: One way to solve this problem is first to split f into partial fractions,
1 1 1
= − , (12.4)
z(z − 1) z−1 z
and then use the formula for a geometric series to write
∞
1 1 X
= = (−1)k (z − 1)k ,
z 1 + (z − 1)
k=0

which converges in D(1, 1). Hence, the Laurent series of f in D(1, 1) \ {1} is
∞
1 X
− (−1)k (z − 1)k .
z−1
k=0

One could also use the formula for a geometric series directly to get
∞ ∞
1 1 X X
= (−1)k (z − 1)k = (−1)k+1 (z − 1)k ,
z(z − 1) z−1
k=0 k=−1

which is the same as we got before.

The part of the Laurent series containing negative powers is called the principal part. In
the example above the principal part is 1/(z − 1).
P∞Let f be holomorphic in the annulus A = {z; r < |z − a| < R}. Then its Laurent series
c (z − a) k converges to f in {z; r +  < |z − a| < R − } for any  > 0 by Laurent’s
−∞ k
theorem, and so it converges to f in A. If the principal part of the Laurent series vanishes, i.e.,
if ck = 0 for k < 0, then f extends to a holomorphic function in the disc D(a, R). In fact, if
the Laurent series converges in {z; r < |z − a| < R} and has no negative powers of z − a then
the series converges in D(a, R) by Lemma 4.4 and thus gives the holomorphic extension.
Let r = 0, so that f has a singularity at a. If ck = 0 for k < 0, then, as we have just seen, f
extends across a, and we say that f has a removable singularity at a. If there is an m > 0 such
that c−m 6= 0 and ck = 0 for k < −m, then we say that f has a pole of order m at a. In the
example above the function has a pole of order 1 at 1. If ck 6= 0 for infinitely many negative
k’s, then we say that f has an essential singularity at a.
Analytic function theory 60

Proposition 12.3. Let f be holomorphic in the punctured disc D(a, R) \ {a} and assume that
there is a constant M > 0 such that |f (z)| ≤ M for all z ∈ D(a, R) \ {a}. Then f has a
removable singularity at a.
Proof. Let ck , k ∈ Z, be the coefficients in the Laurent series of f centered at a and let
0 < % < R. By Laurent’s theorem,
   
 1 Z f (z) dz  1  f (z) 
|ck | = 

≤ sup   2π% ≤ M .
 2πi |z−a|=% (z − a) k+1  2π |z−a|=% |z − a|k+1  %k

If k < 0, then M/%k → 0 as % → 0 and so ck = 0 if k < 0. Hence, f has a removable singularity

at a.

A function f holomorphic in an open set Ω except for poles at isolated points aj ∈ Ω is

said to be meromorphic in Ω. Alternatively, f is meromorphic in Ω if for each a ∈ Ω either f or
1/f is holomorphic in D(a, r) for some r > 0. This follows in particular from the next result.
Theorem 12.4. Let f be holomorphic in a puntured disc D(a, R) \ {a}. Then f has a pole of
order m at a if and only if 1/f is holomorphic in some disc D(a, r) and has a zero of order m
at a.
Proof. Assume first that f has a pole of order m at a. Then the Laurent series looks like
∞
X ∞
X
ck (z − a)k = (z − a)−m ck−m (z − a)k ,
k=−m k=0

and converges in D(a, R)\{a}. It then follows from Lemma 4.4 that ∞ k
P
k=0 ck−m (z −a) =: g(z)
converges in D(a, R) and thus g is holomorphic in D(a, R). Moreover, g(a) = c−m 6= 0 and so
1/g is holomorphic in some disc D(a, r) (why?). Hence, 1/f (z) = (z − a)m /g(z) is holomorphic
in D(a, r) and has a zero of order m at a by Proposition 9.5.
Showing the converse statement essentially amounts to do the above reasoning backwards.
Assume that 1/f is holomorphic in some disc centered at a and that it has a zero of order m
there. Then, by Proposition 9.5, 1/f (z) = (z − a)m g̃(z) where g̃ P
is holomorphic and g̃(a) 6= 0.
Hence, g(z) := 1/g̃(z) is holomorphic in a disc centered at a. Let ∞ k
k=0 dk (z −a) be the Taylor
series of g and notice that d0 = g(a) 6= 0. We get
∞ ∞
g(z) 1 X
k
X
f (z) = = dk (z − a) = dm+k (z − a)k ,
(z − a)m (z − a)m
k=0 k=−m

which by uniqueness has to be the Laurent series of f . Thus f has a pole of order m at a.

In view of Proposition 9.5 we now also get two more characterizations of f having a pole
of order m at a namely 1) that f (z) = g(z)/(z − a)m for some holomorphic g with g(a) 6= 0
and 2) that limz→a (z − a)m f (z) exists and is non-zero.
We conclude this section by saying a few words about singularities at ∞. If f is holomorphic
in {z; |z| > R} for some R > 0 we say that f has a singularity at ∞. Set f˜(w) = f (1/w);
then f˜ is holomorphic in the punctured disc D(0, 1/R) \ {0}. We say that f has a removable
singularity at ∞, a pole of order m at ∞, and an essential singularity at ∞ if f˜ has a removable
singularity, a pole of order m, and an essential singularity, respectively, at 0.
ExampleP 12.5. If f is a polynomial of degree
P n, then f has Pa pole of order n at ∞. In fact, if
f (z) = nk=0 ck z k then f (1/w) = (1/wn ) nk=0 ck wn−k and nk=0 ck wn−k is holomorphic and
non-zero at w = 0.
Analytic function theory 61

By the Fundamental theorem of algebra, a polynomial thus has the same number of zeros
and poles on the Riemann sphere. Something more general is in fact true; any meromorphic
function on the Riemann sphere has the same number of zeros and poles. We will not prove
this statement in these notes.

12.2 Residue theorems

Let f be holomorphic in a punctured disc D(a, R) \ {a} and let ∞ k
P
−∞ ck (z − a) be the Laurent
series. The coefficient c−1 is of particular importance; it is called the residue of f at a and
denoted Res(f ; a). Its importance stems from the fact that
Z
f (z) dz = 2πiRes (f ; a). (12.5)
∂D(a,%)

To see this we compute as in the uniqueness part of the proof of Laurent’s theorem, cf. (12.3);
Z Z ∞
X ∞
X Z
k
f (z) dz = ck (z − a) dz = ck (z − a)k dz = 2πic−1 .
∂D(a,%) ∂D(a,%) k=−∞ k=−∞ ∂D(a,%)

Theorem 12.6 (Cauchy’s residue theorem). Let Ω ⊂ C be an open set and let f be holomorphic
in Ω \ {a1 , . . . , aN }. Let ω ⊂ Ω be an open subset such that aj ∈ ω for all j, ω ⊂ Ω, and ∂ω is
a finite number of piecewise smooth closed curves. Then
Z N
X
f (z) dz = 2πi Res(f ; aj ).
∂ω j=1

Proof. Choose r > 0 such that D(aj , r) ⊂ ω for all j. Then f is holomorphic in an open set
containing ω \ ∪j D(aj , r) and so it follows from Cauchy’s theorem that
Z N Z
X
f (z) dz = f (z) dz.
∂ω j=1 ∂D(aj ,r)

PN
But by (12.5) the right-hand side equals 2πi j=1 Res (f ; aj ) and we are done.
Z
dz
Example 12.7. Compute the integral .
∂D(0,2) z(z − 1)
Solution: From (12.4) we read off that 1/(z(z − a)) has poles of order 1 at z = 0 and z = 1,
which are in D(0, 2), and that the residues at these points are −1 and 1, respectively (why?).
By the residue theorem the integral thus equals 2πi(−1 + 1) = 0.

Theorem 12.6 is sufficient to solve most basic problems but we also give a jazzed-up version.

Theorem 12.8. Let Ω ⊂ C be an open set and let f be holomorphic in Ω \ {a1 , . . . , aN }. If Γ

is a null-homologous cycle in Ω such that aj ∈
/ |Γ|, then
Z N
X
f (z) dz = 2πi IndΓ (aj )Res(f ; aj ).
Γ j=1

Proof. Choose r > 0 such that the discs D(aj , r), j = 1, . . . , N , are pairwise disjoint and
P
D(aj , r) ⊂ Ω and D(aj , r) ∩ |Γ| = ∅ for all j. Consider the cycle γ = j IndΓ (aj )∂D(aj , r).
Analytic function theory 62

Since D(aj , r) ⊂ Ω the winding number Indγ (z) is 0 for all z ∈ / Ω. Hence, the cycle Γ̃ := Γ − γ
is null-homologous in Ω. Since also IndΓ̃ (aj ) = IndΓ (aj ) − Indγ (aj ) = 0 by the choice of γ, the
cycle Γ̃ is in fact null-homologous in Ω\{a1 , . . . , aN }. Since f is holomorphic
R R Ω\{a1 , . . . , aN }
in
it follows from the homology version of Cauchy’s theorem that Γ f (z) dz = γ f (z) dz, and so
Z N
X Z N
X
f (z) dz = IndΓ (aj ) f (z) dz = 2πi IndΓ (aj )Res(f ; aj ).
Γ j=1 ∂D(aj ,r) j=1

In order for the Residue theorem to be useful we need ways of computing residues efficiently;
we list a few here.
Assume first that f has a pole of order ≤ m at a. ThenPg(z) = (z − a)m f (z) is holomorphic
by the comment following the proof of P Theorem 12.4; let ∞ k
k=0 ck (z − a) be the Taylor series
∞ k−m
of g. The Laurent series of f thus is k=0 ck (z − a) and we read off that Res (f ; a) =
cm−1 = g (m−1) (a)/(m − 1)!. We get the following computation rules.
g (m−1) (a)
Residue computation rule 1a: If f has a pole of order ≤ m at a then Res (f ; a) =
(m − 1)!
where g(z) = (z − a)m f (z).
Residue computation rule 1b: If f is of the form f (z) = (z − a)−m ∞ k
P
k=0 ck (z − a) , then
Res (f ; a) = cm−1 .
Residue computation rule 1c: If f has a pole of order 1 at a then Res (f ; a) = lim (z − a)f (z).
z→a
Assume now that f is of the form f (z) = g(z)/h(z), where h has a zero of order 1 at a. Then
h(z) = (z − a)h̃(z) where h̃(a) 6= 0 in view of Proposition 9.5; a straightforward computation
(do it!) shows that h̃(a) = h0 (a). Hence, f (z) = (g(z)/h̃(z))/(z − a) and it follows by Residue
computation 1c that Res (f ; a) = g(a)/h̃(a) = g(a)/h0 (a). We thus have
Residue computation rule 2: If f is of the form f (z) = g(z)/h(z), where h has a zero of
order 1 at a, then Res (f ; a) = g(a)/h0 (a).

Exercises
12.1 Let ∞ k
P
−∞ dk z be absolutely convergent in A = {z; r < |z| < R}, let r < % < R, and
choose u and v such that r < u < % < v < R.
(a) Show that there is a constant C such that |dk |v k ≤ C and |dk |uk ≤ C for all k ∈ Z.
(b) Show that  ∞ 
 X   % N +1 1
sup  dk wk  ≤ C ,
 
|w|=% k=N +1  v 1 − %/v
∞
   N +1
 X 1  u 1
sup  d−k k  ≤ C .

|w|=%  w  % 1 − u/%
k=N +1

(c) Show that

Z ∞ N Z
X dw X dw
dk wk `+1
− dk wk → 0 as N → ∞,
|w|=% k=0 w |w|=% w`+1
k=0
Z ∞ N Z
X d−k dw X d−k dw
k `+1
− k `+1
→ 0 as N → ∞,
|w|=% w w |w|=% w w
k=1 k=1
Analytic function theory 63

and conclude that the second equality in (12.3) is valid.

12.2 Let f be holomorphic in C, let p be a polynomial of degree d with d distinct zeros

a1 , . . . , ad , and choose R > 0 such that D(0, R) contains all the zeros of p. The purpose
of this exercise is to show that for any z ∈ D(0, R)
Z d
X p(z) f (aj )
p(z) f (w)
f (z) = dw + . (12.6)
2πi ∂D(0,R) (w − z)p(w) z − aj p0 (aj )
k=1

This is an example of a division formula; it provides an explicit decomposition, analo-

gous to the decomposition of integers given by the division algorithm, of the function
f as f (z) = p(z)q(z) + r(z), where q and r are holomorphic. The function f thus is
holomorphically divisible by p with quotient q if the “reminder” r is 0, which we see is
the case if f (aj ) = 0 for all j.

(a) Fix z and set Hz (w) = (p(z) − p(w))/(z − w). Show that Hz (w) has a removable
singularity at w = z and thus defines a holomorphic function in C.
(b) Use the Residue theorem to show that
Z d
X p(z) f (aj )
1 Hz (w)f (w)
dw = .
2πi ∂D(0,R) p(w) z − aj p0 (aj )
k=1

(c) Use Cauchy’s formula to show that

Z Z
1 Hz (w)f (w) p(z) f (w)
dw = f (z) − dw,
2πi ∂D(0,R) p(w) 2πi ∂D(0,R) (w − z)p(w)

and conclude that (12.6) holds.

12.3 Let f and g be polynomials such that deg f < deg g =: d and assume that the zeros {aj }
of g are simple. Show that there are constants cj , j = 1, . . . , d, such that
d
f (z) X cj
= .
g(z) z − aj
j=1

Notice that this shows that the standard Ansatz for making a partial fractions decom-
position will work for denominators with simple zeros. What are the constants cj ?
Hint: Exercise 12.2.

12.4 Show that if Z|z| = 1, then Im z = (z −1/z)/2i, and use this and Cauchy’s residue theorem
Im z
to calculate dz.
|z|=2 − 1
z
Z
dz
12.5 Let a, b ∈ C and |a| < |b|. Calculate when (i) 0 < r < |a|, (ii)
|z|=r (z − a)(z − b)
|a| < r < |b|, and (iii) r > |b|.

12.6 Find the Laurent series of f in A when

(a) f (z) = 1/(z + 1) and A = {z; |z| > 1},

(b) f (z) = 1/z and A = {z; |z − i| > 1},
Analytic function theory 64

(c) f (z) = 1/(z(z − 1)) and A = {z; 1 < |z − 2| < 2},

(d) f (z) = 1/(z 2 + z + 1) and A = {z; |z| > 2}.

12.7 Find the Laurent series of f in A when

(a) f (z) = 1/(z − 1)2 and A = {z; |z| > 1},

(b) f (z) = 1/(z 2 − 1)2 and A = {z; 0 < |z − 1| < 2}.

12.8 Find the principal part of the Laurent expansion centered at 0 of

1 1 z
(a) , (b) , (c) , (d) sin(1/z).
ez − 1 cos z − 1 sinh2 z
Are the singularities removable, pole, or essential?

12.9 Let f : D(a, r) \ {a} → C be holomorphic and assume that there is b ∈ C such that
|f (z) − b| ≥  > 0 for all z ∈ D(a, r) \ {a}.
1
(a) Show that g(z) := is holomorphic in D(a, r) and g(z) 6= 0 for z ∈ D(a, r) \
f (z) − b
{a}.
(b) Show that the singularity of f at a is at worst a pole (i.e., not an essential singula-
rity).

R) \ {a} → C is holomorphic with an essential

Use (a) and (b) to prove that if h : D(a,
singularity at a, then h D(a, r) \ {a} is dense in C for any r < R, i.e., for any b ∈ C
there is a sequence aj → a such that h(aj ) → b.

12.10 Find and classify all singularities of the following functions and calculate the correspon-
ding residues.

z−1 1 ez − 1 eiz
(a) , (b) , (c) , (d) ,
(z − 2)(z − 3) z(2 − z) z2 z2 + z + 1

cos z 1 cos(πz) 1
(e) , (f) , (g) , (h) .
(z − 1)3 (z 2 − i)2 sin(πz) z 2 sin z
Z Z
dz dz
12.11 Compute (a) and (b) .
|z|=2 (z − 1)(z + 1)2
3
|z|=2 cos z
Analytic function theory 65

13 Calculating real integrals using complex analysis

We consider a few examples illustrating some techniques to compute certain real integrals by
using complex analysis methods, in particular the Residue theorem.
Z 2π
dt
Example 13.1. Compute the integral .
0 1 + 8 cos2 t
The idea is to rewrite the integral as a contour integral and then use the Residue theorem.
A natural first attempt is to try to rewrite the integral as an integral over ∂D(0, 1): Notice
that z = eit , 0 ≤ t < 2π, is a parametrization of ∂D(0, 1) and that then dz = ieit dt = izdt and
cos t = (eit + e−it )/2 = (z + 1/z)/2. Hence,
Z 2π Z Z
dt dz/iz dz/iz
2
= 2
= 2 2
0 1 + 8 cos t ∂D(0,1) 1 + 8((z + 1/z)/2) ∂D(0,1) 1 + 2(z + 2 + 1/z )
Z
zdz
= −i 4 + 5z 2 + 2
. (13.1)
∂D(0,1) 2z

The denominator is a degree 2 expression in z 2 and may be factorized as √2z 4 + 5z 2 √ +2 =

(2z 2 + 1)(z 2 + 2).
√ It follows that the denominator has zeros of order 1 at ±i/ 2 and ±i 2. Of
2 2
these only ±i/ 2 are in D(0, 1) and the residue of z/((2z + 1)(z + 2)) =: g(z)/(2z + 1) at2

these points are

g(z)  1  1
√ = √ = ,

 2
4z z=±i/ 2 4(z + 2) z=±i/ 2 6


by Residue computation 2. Hence, by (13.1) and the Residue theorem,

Z 2π
dt
= −i2πi(1/6 + 1/6) = 2π/3.
0 1 + 8 cos2 t
Z ∞
dx
Example 13.2. Compute the integral .
0 1 + x4
This is an integral over the interval [0, ∞) which does not bound any open set. To use the
Residue
R ∞ theorem we needRto close up the curve in an appropriate way. A first thing to notice is
dx R dx
that 0 1+x 4 = lim R→∞ 0 1+x4 , and the interval [0, R] is at least finite. We also notice that
the integrand is an even function so that
Z R
1 R dx
Z
dx
4
= .
0 1+x 2 −R 1 + x4
Denote the interval [−R, R] by γR and let ΓR be the upper half of the circle {z; |z| = R}
oriented counterclockwise. Then γR R+ ΓR is the boundary of the upper half of the disc D(0, R)
and we should be able to compute γR +ΓR dz/(1 + z 4 ) by the Residue theorem. On the other
hand, Z Z R Z
dz dx dz
4
= 4
+ 4
, (13.2)
γR +ΓR 1 + z −R 1 + x ΓR 1 + z
and so, if we also can compute the second integral on the right-hand side, we will get information
about the integral of interest.
We begin by computing the left-hand side of (13.2). Since 1+z 4 = (z −eiπ/4 )(z −ei3π/4 )(z −
e i5π/4 )(z − ei7π/4 ) we see that, in the upper half of the disc D(0, R), f (z) := 1/(1 + z 4 ) has
poles of order 1 at eiπ/4 and ei3π/4 (at least if R > 1). The residues at these points can be
computed using, e.g., Residue computation 2:
1  1+i
Res(f ; eiπ/4 ) = 3  iπ/4 = · · · = − √ ,
4z z=e 4 2
Analytic function theory 66

1  1−i
Res(f ; ei3π/4 ) = 3  i3π/4 = · · · = √ .
4z z=e 4 2
Hence, for any R > 1 we have
Z
dz π
4
= 2πi(Res(f ; eiπ/4 ) + Res(f ; ei3π/4 )) = · · · = √ .
γR +ΓR 1 + z 2

The second integral on the right-hand side of (13.2) actually goes to 0 as R → ∞. To see
this we use Proposition 2.7 and the Reverse triangle inequality:
Z 
 dz  1 1 πR

4
≤ sup 4
· `(ΓR ) ≤ sup 4
· πR = 4 ,
ΓR 1 + z z∈ΓR |1 + z | z∈ΓR |z| − 1 R −1
 

which goes to 0 as R → ∞.Z

∞ Z R Z ∞
dx dx π dx π
From (13.2) we now get 4
= lim 4
= √ and we conclude that 4
= √ .
−∞ 1 + x R→∞ −R 1 + x 2 0 1+x 2 2
Z ∞
cos x dx
Example 13.3. Compute the integral 2
.
−∞ x − 2x + 2
We first present the solution and then we make some comments that hopefully explain
why we do as we do. Let γR be the interval [−R, R], let ΓR be the upper half of the circle
eiz
{z; |z| = R}, and let f (z) = 2 . Since z 2 − 2z + 2 = (z − a)(z − ā), where a = 1 + i,
z − 2z + 2
the function f is holomorphic in C \ {a, ā} and the residue at a is

eiz  eia e−1+i

Res(f ; a) = = =
z − ā z=a a − ā 2i

√
by Residue computation 1c. Hence, for any R > 2 (so that a is in the upper half of the disc
D(0, R)),
eiz dz e−1+i
Z
2
= 2πi = πe−1+i (13.3)
γR +ΓR z − 2z + 2 2i
by the Residue theorem. On the other hand,
Z R
eiz dz eix dx eiz dz
Z Z
2
= 2
+ 2
, (13.4)
γR +ΓR z − 2z + 2 −R x − 2x + 2 ΓR z − 2z + 2

and we claim that the second integral on the right-hand side goes to 0 as R → ∞. The claim
follows since
eiz dz  |eiz | eRe iz
Z 


2
≤ sup 2
· `(Γ R ) ≤ sup 2
· πR
ΓR z − 2z + 2 z∈ΓR |z − 2z + 2| z∈ΓR |z| − 2|z| − 2
 
e−Im z πR
≤ sup 2
· πR ≤ 2 → 0 as R → ∞,
z∈ΓR R − 2R − 2 R − 2R − 2

where the second inequality follows from Proposition 5.2 (and the Reverse triangle inequality),
and the fourth inequality follows since Im z ≥ 0 on ΓR so that e−Im z ≤ e0 = 1 for all z ∈ ΓR .
In view of (13.3) and (13.4) we thus have
∞ R
eix dx eix dx
Z Z
= lim = πe−1+i . (13.5)
−∞ x2 − 2x + 2 R→∞ −R x2 − 2x + 2
Analytic function theory 67

However, eix = cos x + i sin x, so the integral we are interested in computing is the real part of
the left-hand side of (13.5). Hence, the sought integral equals Re(πe−1+i ) = Re(πe−1 (cos 1 +
i sin 1)) = πe−1 cos 1.
Our comments concern the choice of f and integration contour. One might wonder why
we don’t choose f as cos z/(z 2 − 2z + z); this is a holomorphic function in C \ {a, ā} and on
the real axis it is the integrand in the integral we want to compute. The step where we use
the Residue theorem works just as well with this choice of f but the step where we show that
the integral over ΓR goes to 0 will not work out in a nice way. The reason is that we cannot
estimate | cos z| for z ∈ ΓR in a sufficiently good way; recall that cos grows exponentially on
the imaginary axis.
One might also wonder why we choose the upper half of the circle {|z| = R} to get a closed
integration contour and not the lower half. This is just a matter of taste, we can choose ΓR as
the lower part but then we need to choose f as e−iz /(z 2 − 2z + z) to make the step where the
integral over ΓR goes to 0 work out.
Other ways of closing up the integration contour might also work but will probably lead to
more elaborate computations.
Z ∞
sin2 x dx
Example 13.4. Compute the integral .
0 x2
It is often a good idea first to check that the integral in question actually makes sense.
In this case there is no problem; the singularity at 0 is removable (why?) and the integrand
decays as 1/x2 as x → ±∞ so the integral converges nicely. To compute it we begin by
1 − ei2z
noticing that sin2 x = (1 − cos(2x))/2 = Re(1 − ei2x )/2 and set f (z) = . However, f
2z 2
does not have a removable singularity at 0 since the numerator now has a zero only of order
1 at 0. Thus, f has a pole of order 1 at 0 and so we cannot choose [−R, R] as part of an
integration contour as we have done before, we need to take a detour to avoid the origin. Let
γ1 (δ, R) be the interval [−R, −δ], let γ2 (δ, R) be the interval [δ, R], let ΓR be the upper half
of the circle {z; |z| = R} oriented counterclockwise, and let Γδ be the upper half of the circle
{z; |z| = δ} oriented clockwise. Then γ1 (δ, R) + Γδ + γ2 (δ, R) + ΓR is the boundary of the set
ΩδR := {z ∈ C; Im z > 0} ∩ D(0, R) \ D(0, δ) and f is holomorphic in an open set containing
δ
ΩR . Therefore, by Cauchy’s theorem, we have
Z Z Z
f (z) dz + f (z) dz + f (z) dz = 0. (13.6)
γ1 (δ,R)+γ2 (δ,R) Γδ ΓR

The first integral should be related to the integral we want to compute and, indeed, we have
Z −δ Z R
1 − ei2x 1 − ei2x
Z
f (z) dz = 2
dx + dx
γ1 (δ,R)+γ2 (δ,R) −R 2x δ 2x2
1 − e−i2x
Z R Z R
1 − ei2x
= dx + dx
δ 2x2 δ 2x2
2 − (ei2x + e−i2x )
Z R Z R
2 − 2 cos(2x)
= 2
dx = dx
δ 2x δ 2x2
Z R
2 sin2 x
= dx,
δ x2
which converges to 2 times the sought integral as δ → 0 and R → ∞.
The third integral in (13.6) goes to 0 as R → ∞ since
1 − ei2z  |1 − ei2z | 1 + e−2Im z
Z 
 π

2
dz ≤ sup 2
· πR ≤ sup 2
· πR ≤ .

ΓR 2z 
z∈ΓR 2|z| z∈ΓR 2R R
Analytic function theory 68

We also need to compute the second integral in (13.6). We do this similarly to the proof of
Lemma 3.10. Notice that ei2z = 1 + 2iz + O(|z|2 ) by Taylor’s formula. Hence,
1 − ei2z −2iz O(|z|2 )
Z Z Z
dz = dz + dz
Γδ 2z 2 Γδ 2z
2
Γδ 2z 2
Z Z Z 0 Z
dz
= −i + O(1)dz = −i idt + O(1)dz
Γδ z Γδ t=π Γδ
Z
= −π + O(1)dz.
Γδ

The last integral here goes to 0 as δ → 0 since the integrand is bounded and the length of Γδ
goes to 0. Thus the second integral in (13.6) goes to −π as δ → 0. Putting all this together
Z ∞
sin2 x
and letting δ → 0 and R → ∞ in (13.6) we get 2 dx − π = 0. The sought integral
0 x2
thus equals π/2.
Z R
x−1
Example 13.5. Compute the limit lim cos x dx.
R→∞ −R x2 + 1
Notice that the integrand decays as 1/|x| as x → ±∞ so it isn’t integrable on R in the usual
z − 1 iz
sense. Part of the exercise is to see that the given limit exists anyway. Let f (z) = 2 e
z +1
and let γR and ΓR be as in Examples 13.2 and 13.3. The function f is holomorphic in C \ {±i}
and the residue at i can be seen, e.g., by Residue computation 2, to be Res(f ; i) = e−1 (1 + i)/2.
As in several of the examples above the Residue theorem gives
Z R
x − 1 ix z − 1 iz
Z
2
e dx + 2+1
e dz = 2πiRes(f ; i) = πe−1 (−1 + i).
−R x + 1 ΓR z
We claim that the second integral on the left-hand side goes to 0 as R → ∞. Given the claim
it follows that the sought limit equals Re(πe−1 (−1 + i)) = −πe−1 .
It remains to show the claim. In this case the method to estimate the integral by Proposi-
tion 2.7 and using that |eiz | ≤ 1 in the upper half-plane as we have done before does not work.
In fact, this method (do the computations) shows that the modulus of the second integral is
bounded by πR(R + 1)/(R2 + 1) which does not go to 0 as R → ∞. The point is that we need
to improve the estimate |eiz | ≤ 1 for z = Reit and 0 ≤ t ≤ π. First we make a preliminary
computation:
Z π
Reit − 1 iReit
Z  
 z − 1 iz   it 


2
e dz  =  
2 i2t
e iRe dt
ΓR z + 1 0 R e +1

Z π
R2 + R π/2 −R sin t
Z
R+1 −R sin t
≤ 2
Re dt = 2 2 e dt.
0 R −1 R −1 0
To estimate the last integral we use
π 2
Jordan’s inequality: If 0 ≤ t ≤
, then t ≤ sin t ≤ t.
2 π
In view of this we have − sin t ≤ −2t/π for 0 ≤ t ≤ π/2 and so
" #π/2
R2 + R
Z π/2
R2 + R
Z π/2
R 2+R e−2Rt/π
e−R sin t dt ≤ e−2Rt/π dt = 2 −π
R2 − 1 0 R2 − 1 0 R −1 2R
0
π R2 + R −R
= − (e − 1),
2R R2 − 1
which goes to 0 as R → ∞. This proves the claim.
Analytic function theory 69

Exercises
Z
dz
13.1 Compute the integral .
∂D(0,8) 1 + ez
2π
sin2 t
Z
13.2 Compute the integral dt.
0 5 + 4 cos t
Z ∞
cos(2x)
13.3 Compute the integral dx.
0 x2 + 4
∞
e−ix
Z
13.4 Compute the integral dx.
−∞ x2 + x + 1
Z ∞
cos(ax)
13.5 Compute the integral dx for all a > 0.
−∞ x2 + a2
Z ∞
dx
13.6 Compute the integral for all a, b ∈ R such that 0 < a < b.
0 (x2 + a2 )(x2 + b2 )
Z ∞
sin x
13.7 Compute the integral dx.
0 x(x2 + 1)
Z R
sin x
13.8 Compute the limit lim dx.
R→∞ −R x
Z ∞
dx
13.9 Compute the integral for all integers n ≥ 2.
0 1 + xn
(Hint: Integrate around a suitable sector (piece of cake) of D(0, R) of angle 2π/n.)
Z ∞
dx
13.10 Compute the integral 2
.
−∞ x − x + 1

π cos(πz)
13.11 Let f (z) = and let, for N = 1, 2, 3, . . ., SN be the square with corners at
z 2 sin(πz)
±(N + 1/2) ± i(N + 1/2).
N
π2 
Z  X 1
(a) Show that f (z)dz = 2πi 2 − .
∂SN n2 3
n=1
cos(πz)
(b) Show that there is a constant C such that sup | | ≤ C for all N .
z∈SN sin(πz)
∞
π2
Z X 1
(c) Show that lim f (z)dz = 0 and conclude that = .
N →∞ ∂SN n2 6
n=1
Analytic function theory 70

14 Harmonic functions and the Dirichlet problem

One reason why holomorphic functions are important is that they are closely related to harmo-
nic functions which in turn are fundamental in physics. If Ω is an open set in the plane and we
place charges on its boundary, then the resulting electric potential is a harmonic function in Ω.
If we instead place heat sources on the boundary and wait until the temperature distribution
T becomes stationary, then T is a harmonic function in Ω. These are certainly not the only
examples where harmonic functions turn up in physics.
The Dirichlet problem is a certain boundary value problem; given a continuous function on
∂Ω one seeks a continuous function on Ω that agrees with the given one on the boundary and
that is harmonic in Ω. In view of the preceding examples it is at least physically reasonable to
believe that the Dirichlet problem has a unique solution.

14.1 Harmonic vs. holomorphic functions

Let Ω ⊂ C be an open set. A C 2 -smooth function u in Ω is harmonic in Ω if it satisfies the
partial differential equation
∂2u ∂2u
+ 2 =0
∂x2 ∂y
∂2 ∂2
in Ω. The differential operator + is called the Laplace operator and is often denoted
∂x2 ∂y 2
by ∆. Notice that if ϕ is a real-valued function, then ∆ϕ is also real-valued. It follows that the
real and the imaginary parts of a harmonic function are harmonic as well. In fact, if f = u + iv
then ∆f = ∆u + i∆v and hence, ∆f = 0 if and only if ∆u = ∆v = 0.
A first indication that harmonic functions and holomorphic functions are related is given
by the following result. Recalling the definition of ∂/∂z and ∂/∂ z̄ (see Chapter 2), the proof
is a straightforward computation (Exercise 14.1).
∂2ϕ
Proposition 14.1. If ϕ is a C 2 -smooth function then ∆ϕ = 4 .
∂z∂ z̄
It follows that holomorphic functions are harmonic. Thus, the real and the imaginary parts
of a holomorphic function are harmonic. A partial converse of this statement is also true; if u
is harmonic and real-valued, then there is, at least locally, a holomorphic function f such that
u is the real part of f . This is the content of

Theorem 14.2. Let Ω ⊂ C be an open simply connected set. If u is a real-valued harmonic

function in Ω then there is a real-valued harmonic function v in Ω such that f = u + iv is
holomorphic in Ω.

The function v is called a harmonic conjugate of u. Harmonic conjugates are not unique
but if v1 and v2 both are harmonic conjugates of u then v1 = v2 +c for some constant c. In fact,
both u + iv1 and u + iv2 are holomorphic so v1 − v2 = −i(u + iv1 − (u + iv2 )) is holomorphic
and real-valued. Thus v1 − v2 is constant by Proposition 3.6.

Example 14.3. If u(x, y) = x2 − y 2 , then v(x, y) = 2xy is a harmonic conjugate of u since

u + iv = x2 − y 2 + i2xy = (x + iy)2 = z 2 , which is holomorphic.

Complex-valued harmonic functions also have harmonic conjugates locally, albeit complex-
valued. Such harmonic conjugates are only unique up to holomorphic functions, see Exerci-
se 14.4.
Analytic function theory 71

∂u ∂ ∂u
Proof of Theorem 14.2. The function is holomorphic since = ∆u/4 = 0 (see Sec-
∂z ∂ z̄ ∂z
tion 3.1) Therefore, since Ω is simply connected, there is a holomorphic function F in Ω such
∂u ∂(F − u)
that F 0 = by Theorem 10.8. This means that = 0. In view of Exercise 2.3 we get
∂z ∂z
 
∂(F − u) ∂(F − u) ∂(F − u)
= = =0
∂ z̄ ∂ z̄ ∂z
since u = u. Thus F − u is holomorphic and so also F + F − u is holomorphic. But F + F − u is
real-valued so it must be constant by Proposition 3.6, i.e., F +F −u = C for some real constant
C. Setting f = 2F − C we get a holomorphic function in Ω such that Re f = 2Re F − C =
F + F − C = u. We may thus take v = Im f .

It is necessary that Ω is simply connected for Theorem 14.2 to be true, see Exercise 14.6.
Actually, if Ω is not simply connected then there always is a harmonic function lacking harmonic
conjugate in Ω but we will not prove this.

14.2 Poisson’s integral formula and some consequences

Poisson’s integral formula follows from Cauchy’s formula and expresses a harmonic function u
in the unit disc in terms of the values of u on the boundary. Many properties of holomorphic
functions stem from Cauchy’s formula and, similarly, we will see that Poisson’s integral formula
implies fundamental properties of harmonic functions.
Theorem 14.4 (Poisson’s integral formula). If u is continuous on D(0, 1) and harmonic in
D(0, 1), then, for any 0 ≤ r < 1 and t ∈ R,
Z 2π
1 1 − r2
u(reit ) = u(eiθ )dθ. (14.1)
2π 0 1 − 2r cos(t − θ) + r2
Proof. We may assume that u is real-valued. Otherwise we consider the real and imaginary
parts of u separately.
Assume first that u is harmonic in an open set containing D(0, 1) and let v be a harmonic
conjugate of u so that f := u + iv is holomorphic in an open set containing D(0, 1). Fix a point
w = reit ∈ D(0, 1) and set g(z) = (1 − r2 )/(1 − w̄z). Then g is holomorphic in an open set
containing D(0, 1) and g(w) = 1. We apply Cauchy’s formula to f (z)g(z):
Z
1 f (z)g(z) dz
f (reit ) = f (w) = f (w)g(w) =
2πi |z|=1 z−w
1 − r2 dz
Z
1
= f (z) .
2πi |z|=1 1 − w̄z z − w
Since |z| = 1 we have z z̄ = |z|2 = 1 and so z − w = (1 − wz̄)z. We thus get
1 − r2
Z
it 1 1 dz
f (re ) = f (z)
2πi |z|=1 1 − w̄z 1 − wz̄ z
1 − r2
Z
1 dz
= f (z) .
2πi |z|=1 1 − 2Re(wz̄) + |w|2 z

We parametrize the curve {|z| = 1} by z = eiθ , 0 ≤ θ ≤ 2π. Then dz/z = ieiθ dθ/eiθ = idθ and
Re(wz̄) = Re(reit e−iθ ) = r cos(t − θ) so we get
Z 2π
1 1 − r2
f (reit ) = f (z) dθ.
2π 0 1 − 2r cos(t − θ) + r2
Analytic function theory 72

Taking the real part we obtain (14.1).

If u is merely continuous on D(0, 1) and harmonic in D(0, 1) we set u% (z) := u(%z), where
0 ≤ % < 1. Then u% is harmonic in an open set containing D(0, 1) and so (14.1) holds with
u replaced by u% . But u is uniformly continuous on D(0, 1) so for any given  > 0 there is a
δ > 0 such that |u(z) − u% (z)| = |u(z) − u(%z)| ≤  if |z − %z| ≤ δ; then (14.1) follows by letting
% → 1.

An alternative proof of Poisson’s formula (in case f is holomorphic) is outlined in Exerci-

se 14.7.

Corollary 14.5 (Mean value property of harmonic functions). If u is harmonic in an open

Z 2π
1
set containing D(0, R) then u(0) = u(%eiθ )dθ if 0 ≤ % ≤ R.
2π 0
Proof. Let u% (z) := u(%z), where 0 ≤ % ≤ R. Then u% is harmonic in an open set containing
D(0, 1). Poisson’s integral formula thus gives
Z 2π Z 2π
1 iθ 1
u(0) = u% (0) = u% (e ) dθ = u(%eiθ ) dθ.
2π 0 2π 0

Theorem 14.6 (Maximum principle for harmonic functions). Let Ω ⊂ C be an open bounded
connected set. If u is a continuous function on Ω that is harmonic in Ω, then |u| attains its
maximum on the boundary, i.e., sup |u(z)| = sup |u(z)|.
z∈Ω z∈∂Ω

Proof. The proof relies on the Mean value property and is essentially the same as the proof of
the Maximum principle in Chapter 11; we recall the main points.
Let M = supz∈Ω |u(z)|. Since Ω is bounded and |u| is continuous on Ω, M is finite and
there is an a ∈ Ω such that M = |u(a)|. If a ∈ ∂Ω we are done so assume that a ∈ Ω. Let
A = {z ∈ Ω; |u(z)| = M }; since a ∈ A, A is non-empty. Moreover, since |u| is continuous
it follows that A is closed (why?). If A in addition is open it follows that A = Ω since Ω is
connected (why?). But then |u| is a constant function so it attains its maximum (=only value)
on the boundary.
It remains to show that A is open. Let b ∈ A and choose R > 0 such that D(b, R) ⊂ Ω.
Using the Mean value property of harmonic functions and computing as in (11.1) (with f
replaced by u, a replaced by b, and r replaced by %) we get
Z 2π
1
|u(b)| − |u(b + %eiθ )| dθ = 0
2π 0

for all % ≤ R. But since b is a maximum point, the integrand must be non-negative, and a non-
negative function cannot have integral 0 unless the function is 0. Hence, |u(b)|−|u(b+%eiθ )| = 0
for all θ ∈ [0, 2π] and all % ≤ R, i.e., |u| is constant in D(b, R). Thus, D(b, R) ⊂ A, which shows
that A is open.

14.3 The Dirichlet problem and the Riemann mapping theorem revisited
By the Dirichlet problem we will mean the following: Let Ω ⊂ C be an open bounded and
connected set and let υ be a continuous function on ∂Ω. Find a continuous function u on Ω
that is harmonic in Ω and agrees with υ on ∂Ω.
Analytic function theory 73

If the Dirichlet problem is solvable then the solution is unique; this follows from the Maxi-
mum principle. In fact, if u1 and u2 are continuous on Ω, harmonic in Ω, and u1 = u2 on ∂Ω,
then u1 − u2 is continuous on Ω, harmonic in Ω, and 0 on ∂Ω. Hence, |u1 − u2 | = 0 on ∂Ω and
so, by the Maximum principle, |u1 − u2 | ≤ 0 in Ω and we thus get u1 = u2 in Ω.
For Ω = D(0, 1) the Dirichlet problem is solvable. Let υ be a continuous function on ∂Ω.
Notice that if u is a solution of the corresponding Dirichlet problem then, by Poisson’s formula,
the values of u in D(0, 1) are given by
Z 2π
it 1 1 − r2
u(re ) = υ(eiθ )dθ. (14.2)
2π 0 1 − 2r cos(t − θ) + r2
The integral on the right-hand side only involves the given function υ so if we want to find
a solution of the Dirichlet problem it is natural to define a tentative solution as the integral
on the right-hand side. This will at least give us a harmonic function u in D(0, 1). In fact, if
z = reit , then (Exercise 14.2)

1 − r2
 iθ 
e +z
= Re iθ , (14.3)
1 − 2r cos(t − θ) + r2 e −z

so the integrand in (14.2), as a function of reit ∈ D(0, 1), is a harmonic function. It is also true
that u defined by (14.2) in D(0, 1) has a continuous extension to D(0, 1) that agrees with υ on
∂D(0, 1), but we will not show this.
Let now Ω be an open bounded and simply connected set. By the Riemann mapping
theorem there is holomorphic bijective map ϕ : Ω → D(0, 1). If ∂Ω is sufficiently nice it is not
unreasonable to believe that ϕ can be extended to a continuous bijective map Ω → D(0, 1). In
this case we can solve the Dirichlet problem in Ω as follows. Let ν be a continuous function
on ∂Ω. Then υ := ν ◦ ϕ−1 ∂D(0,1) is a continuous function on ∂D(0, 1). Since we can solve the
Dirichlet problem in D(0, 1) there is a continuous function u on D(0, 1) that is harmonic in
D(0, 1) and agrees with υ on ∂D(0, 1). Then v := u◦ϕ is a continuous function on Ω that agrees
with ν on ∂Ω (why?) and it is harmonic in Ω since ∆(u ◦ ϕ) = |ϕ0 |2 ∆u = 0, see Exercise 14.3.
The Riemann mapping theorem thus indicates that the Dirichlet problem is solvable in
sufficiently nice simply connected open sets. On the other hand, if the Dirichlet problem is
solvable in simply connected open sets, we can almost prove the Riemann mapping theorem.
Since the Dirichlet problem should be solvable for physical reasons, see the introduction to this
chapter, we at least get a physical justification of the Riemann mapping theorem.
We end these notes by a “hand-waving” proof of the Riemann mapping theorem given that
the Dirichlet problem can be solved. Let Ω be an open bounded and simply connected set and
pick a point a ∈ Ω. Then − log |z − a| defines a continuous function on ∂Ω. If the Dirichlet
problem can be solved in Ω we find a continuous function u on Ω that agrees with − log |z − a|
on ∂Ω and is harmonic in Ω. Since Ω is simply connected there is a harmonic conjugate v of u
by Theorem 14.2. Then ϕ(z) := (z − a)eu(z)+iv(z) is holomorphic in Ω and has precisely 1 zero,
multiplicity counted, since eu(z)+iv(z) 6= 0. For z ∈ ∂Ω we have

|ϕ(z)| = |z − a|eRe f (z) = |z − a|eu(z) = |z − a|e− log |z−a| = 1

and it follows from the Maximum principle that ϕ maps Ω to D(0, 1). We will show that ϕ is
bijective under the additional assumption that ϕ can be extended to a holomorphic function
in some open set containing Ω and that ∂Ω is piecewise smooth. In this case ϕ maps ∂Ω to
∂D(0, 1) and it follows that
ϕ0 (z) dz
Z
1
ψ(w) =
2πi ∂Ω ϕ(z) − w
Analytic function theory 74

is a continuous function in D(0, 1). Moreover, by the Argument principle, ψ(w) is the number
of points z ∈ Ω such that ϕ(z) = w. We want to show that ψ(w) = 1 for all w ∈ D(0, 1). But
ψ is continuous and integer-valued so it must be constant, and this constant has to be 1 since
ψ(0) = 1.

Exercises
14.1 Show Proposition 14.1.

14.2 Show (14.3), where z = reit .

14.3 Let u(w) be C 2 -smooth and let ϕ(z) be holomorphic. Show, using the Chain rule, that
∂ 2 (u ◦ ϕ) ∂2u
= |ϕ0 |2 .
∂z∂ z̄ ∂w∂ w̄
14.4 Let f be a possibly complex-valued harmonic function in a simply connected open set
Ω. Show that there is a harmonic function g in Ω such that f + ig is holomorphic in Ω.
Show also that if g1 and g2 are harmonic functions such that f + ig1 and f + ig2 are
holomorphic, then g1 − g2 is holomorphic.

14.5 Let Ω be a bounded open set with piecewise smooth boundary and assume that the
Dirichlet problem is solvable in Ω. Set νa (z) = 1/(z − a) for z ∈ ∂Ω and a ∈ Ω and let
1 ∂v̄
va (z) be the solution of the corresponding Dirichlet problem. Show that Ka (z) :=
π ∂z
is holomorphic in Ω. Show also that if g is holomorphic in an open set containing Ω then
Z
g(a) = g(z)Ka (z) dxdy.
Ω

(Hint: Cauchy’s formula and (2.10).)

14.6 (a) Show that log |z| is harmonic in C \ {0}.

(b) Find a harmonic conjugate of log |z| in C \ (−∞, 0].
(c) Show that log |z| does not have a harmonic conjugate in C \ {0}.

14.7 Let f be holomorphic in an open set containing D(0, 1).

(a) Use Cauchy’s formula and theorem to show that if w ∈ D(0, 1) then
Z Z
f (z) f (z)
2πif (w) = dz − dz. (14.4)
∂D(0,1) z − w ∂D(0,1) z − 1/w̄

(b) Rewrite the right-hand side of (14.4) and show

2π
1 − r2
Z
it 1
f (re ) = f (eiθ )dθ.
2π 0 1 − 2r cos(t − θ) + r2
Analytic function theory 75

Appendix A
Complex numbers
A complex number z is an ordered pair (x, y) of real numbers x, y ∈ R. We usually write
z = x + iy, where the i in front of y here means that y is the second element of the ordered
pair. We identify the set C of complex numbers with R2 and we picture the complex number
z = x + iy as the point (x, y) in R2 . The x-axis is called the real axis and the y-axis is called the
imaginary axis. The standard projections on these axes are denoted Re and Im respectively, so
that Re (x + iy) = x and Im (x + iy) = y. For a complex number z, Re z is called the real part
of z and Im z is called the imaginary part of z. (The imaginary part of a complex number is
thus a real number!) The set of real numbers is then naturally identified with the real axis by
identifying x ∈ R with (x, 0). The conjugate, z̄, of a complex numberp z = x + iy is z̄ := x − iy.
The absolute value, or modulus, of z = x + iy is defined to be |z| := x2 + y 2 and is thus the
Euclidean distance between 0 and (x, y) in R2 .
A complex number can also be represented using polar coordinates on R2 . If z = x + iy is
a complex number, let r = |z| and let θ be the angle between the positive real axis and the ray
from 0 through (x, y). Then (x, y) = (r cos θ, r sin θ), and the polar representation of z is

z = r cos θ + ir sin θ = reiθ . (14.5)

At this point eiθ is just a notation but in Chapter 5 we define the exponential of any complex
number (in a natural way extending the exponential of real numbers) and show that (14.5)
holds.

Arithmetics of complex numbers

Addition and multiplication of two complex numbers z = x + iy and w = u + iv are defined as
follows:

z + w = (x + u) + i(y + v),
z · w = (xu − yv) + i(xv + yu).

Addition is easy to visualize: Thinking of z and w as vectors in R2 starting at 0 and ending

at (x, y) and (u, v) respectively, the sum z + w corresponds to the vector sum of these two
vectors. To visualize multiplication, it is convenient to use polar representation. Let z = reiθ
and w = ρeiϕ . Using the definition of complex multiplication and the addition formulas for sin
and cos we get

z · w = (r cos θ + ir sin θ = reiθ ) · (ρ cos ϕ + iρ sin ϕ) (14.6)

= rρ ((cos θ cos ϕ − sin θ sin ϕ) + i(sin θ cos ϕ + cos θ sin ϕ))
= rρ (cos(θ + ϕ) + i sin(θ + ϕ)) .

Thus, the modulus of the product z · w is the product of the moduli |z| and |w|, and the
angle corresponding to the product z · w is the sum of the angles corresponding to z and
w. In particular, if |w| = 1, then you get z · w by rotating (the vector corresponding to) z
counterclockwise by the angle corresponding to w.
The computation in (14.6) and induction (do the details!) shows that

(cos θ + i sin θ)n = cos(nθ) + i sin(nθ)

for any positive integer n. This equality is called de Moivre’s formula. In our new notation it
can alternatively be written (eiθ )n = einθ , which give some justification for the notation.
Analytic function theory 76

The first eleven of the following computation rules are easily checked (do it!) from the
definitions of complex addition and multiplication.

(1) z + w = w + z

(2) (z + w) + ζ = z + (w + ζ)

(3) z · w = w · z

(4) (z · w) · ζ = z · (w · ζ)

(5) z · (w + ζ) = z · w + z · ζ

(6) z̄¯ = z

(7) z + z̄ = 2Re z, z − z̄ = 2iIm z

(8) z + w = z̄ + w̄

(9) z · w = z̄ · w̄

(10) |z̄| = |z|

(11) z · z̄ = |z|2

(12) |Re z| ≤ |z|, |Im z| ≤ |z|

(13) |z + w| ≤ |z| + |w|

 
(14) |z + w| ≥ |z| − |w|.

To show (12), recall that |z|2 = (Re z)2 + (Im z)2 and notice that the terms are non-negative.
Thus, (Re z)2 = |z|2 − (Im z)2 ≤ |z|2 and so |Re z| ≤ |z|. We get |Im z| ≤ |z| in a similar way.
To show (13), which is known as the Triangle inequality, we use (some of) the previous
computation rules:

|z + w|2 = (z + w)(z + w) = (z + w)(z̄ + w̄)

= |z|2 + z w̄ + z̄w + |w|2 = |z|2 + z w̄ + z w̄ + |w|2
= |z|2 + 2Re (z w̄) + |w|2 ≤ |z|2 + 2|z| · |w| + |w|2
= (|z| + |w|)2 .

The Reverse triangle inequality, (14), follows from the Triangle inequality in the same way as
in real calculus.
Please note that inequalities can only occur between real numbers, it has no meaning to
write “z ≤ w” if any of z and w is complex!
Division of complex numbers is defined as follows. First, if z = x + iy and t ∈ R \ {0},
then we define z/t = (1/t) · z = (x/t) + i(y/t). Second, if z ∈ C and w ∈ C \ {0}, then z/w is
defined to be the unique complex number q satisfying w · q = z. To compute q we multiply by
w̄ and get |w|2 · q = z · w̄; dividing by the non-zero real number |w|2 we obtain q = z · w̄/|w|2 .
In particular, w−1 := 1/w = w̄/|w|2 ; it is straightforward to check that de Moivre’s formula
holds also for negative integers.
Analytic function theory 77

Polynomials
A polynomial of the complex variable z is a function C → C of the form
n
X
p(z) = ak z k ,
k=0

where the coefficients ak are complex numbers. The integer n is called the degree of the po-
lynomial. Notice that, setting z = x + iy, p becomes a polynomial of the two real variables x
and y (with complex coefficients). However, not every polynomial of two real variables can be
obtained in this way, cf. the Informal introduction above.
The solutions of a polynomial equation p(z) = 0 are called the roots of the polynomials.
The Fundamental theorem of algebra, which is proved (by analytic techniques!) in Chapter 6,
states that every polynomial has a root in C. By the standard algorithm of polynomial division
it then follows that a polynomial of degree n has precisely n roots in C taking multiplicity into
account, see Chapters 6 and 9 above.

Example 14.7 (Roots of unity). Find the roots of the polynomial z n − 1, n ≥ 0; these are
called the nth roots of unity.
Solution: We use polar representation and write z = reiθ . By de Moivre’s formula we are
looking for r ≥ 0 and θ ∈ R such that rn einθ = 1. From the identity sin2 θ + cos2 θ = 1 and
the computation rule (11) it follows that |einθ | = 1. Thus, 1 = |1| = |rn einθ | = rn so r = 1; it
remains to find θ ∈ R such that einθ = 1, i.e., such that cos(nθ) + i sin(nθ) = 1. Taking the
real and imaginary parts of this equation we get
(
cos(nθ) = 1
sin(nθ) = 0.

From calculus we know that the solutions of these equations are nθ = 2πk, k ∈ Z, so θ = 2πk/n,
k ∈ Z. For k = 0, 1, . . . , n − 1, ei2πk/n are different complex numbers of modulus 1 lying on the
vertices of the regular n-gon centered at 0 with one vertex at (1, 0). Now, any k ∈ Z can be
written k = k0 + ` · n for some integer ` and some k0 ∈ {0, 1, . . . , n − 1} and so

ei2πk/n = ei2π(k0 +`·n)/n = ei2πk0 /n+i2π` = ei2πk0 /n · ei2π` = ei2πk0 /n ,

where the third equality follows from the addition formulas for sin and cos, cf. (14.6), and the
forth since ei2π` = cos(2π`) + i sin(2π`) = 1. Hence we do not get any new solutions.
In conclusion, the roots of z n −1 are the n complex numbers 1, ei2π/n , ei2π2/n , . . . , ei2π(n−1)/n .

Exercises
A.1 Write the following complex numbers in polar form.
√ √
(a) i + 1, (b) i3 , (c) (1 − i)/ 2, (d) i − 3, (e) (1 + i)(−1 + i),
√ √
(f) 1/(1 + i), (g) (1 − i)/( 3 + 1), (h) (2 3 + 2i)3 .

A.2 Calculate  n
n 1+i
(a) i , (b)
1−i

A.3 Calculate the conjugate of z = (2 + i)n + (2 − i)n . Show that z ∈ R.

Analytic function theory 78

A.4 Find the roots of the following polynomials and illustrate geometrically.

(a) z 3 − 1, (b) z 4 + 1, (c) z 5 − 1.

√
A.5 Let z = x + iy, (x, y ∈ R). Show that |z| ≤ |x| + |y| ≤ 2|z|.
(Hint for the second inequality: Rewrite the inequality 0 ≤ (|x| − |y|)2 .)
Analytic function theory 79

Appendix B
Open and closed subsets of C and of the Riemann sphere
A subset S ⊂ C is open in C if for each a ∈ S there is some r > 0 such that the disc centered
at a with radius r is contained in S, i.e., {z ∈ C; |z − a| < r} ⊂ S.

Example 14.8. Let S = {z ∈ C; |z| < 1}. Then S is open. In fact, if a ∈ S, then |a| < 1 and
so there is some r > 0 such that |a| + r < 1. Hence, if z satisfies |z − a| < r, then it follows
from the Triangle inequality that |z| = |z − a + a| ≤ |z − a| + |a| < r + |a| < 1, i.e., the disc
centered at a with radius r is contained in S.

It is straightforward to check from the definition that the following holds.

• The empty set, ∅, and C are open,

• if S1 , . . . , Sn are open subsets of C then S1 ∩ · · · ∩ Sn is open,

• if {Si }i∈I is an arbitrary family of open subsets then ∪i∈I Si is open.

Notice also that any open subset of C is a union of discs since if S is open, then for each
a ∈ S there is an ra > 0 such that D(a, ra ) := {z ∈ C; |z − a| < ra } ⊂ S. It follows that,
S = ∪a∈S D(a, ra ).
b = C ∪ {∞} be the extended complex plane, or the Riemann sphere. A subset S ⊂ C
Let C b
is open in C if S is a union of discs D(a, r) ⊂ C and subsets of the form {∞} ∪ {z ∈ C; |z| > r}.
b
If we identify Cb with the unit sphere in R3 as in Chapter 1, then it is not hard to believe (and
it is indeed true) that the open subsets of C b precisely correspond to the intersection between
the unit sphere and the open subsets of R3 .
A subset S of C (respectively C) b if C \ S (resp. C
b is closed in C (resp. C) b \ S) is open.
Let S be a subset of C or C. A point a ∈ C is a limit point of S if S ∩{z ∈ C; 0 < |z −a| < r}
b
is non-empty for all r > 0. The point ∞ is a limit point of S if S ∩{z ∈ C; |z| > r} is non-empty
for all r > 0.
The closure, S, of a set S is the union of S and all its limit points.

Proposition 14.9. Let S ⊂ C. Then (1) a ∈ S if and only if S ∩ U is non-empty for every
open set U containing a, (2) S is closed, and (3) S is closed if and only if S = S.

The boundary of an open subset S ⊂ C is defined, and denoted, by ∂S := S \ S.

Some topology and analysis in C and on the Riemann sphere

Specifying a topology on some given set X amounts to defining what subsets of X are to be
called open, and this family of subsets should have the three properties listed after Example 14.8
above, with C in the first property replaced by X. Having a topology on X is the minimum
requirement to be able to speak about convergence, continuity, and basic geometric features of
X. The families of open subsets of C and C b defined above are topologies on C resp. C. b
Let S ⊂ C; S is said to be compact if it is closed and bounded, which means that there is
some r > 0 such that S ⊂ {z ∈ C; |z| < r}. S is convex if for any two points a, b ∈ S, the line
segment connection a and b is contained in S. If S is open then S is said to be connected if S
cannot be written as the union of two disjoint non-empty open sets S1 , S2 ⊂ S; S is said to be
path-connected if for any two points a, b ∈ S there is a continuous map γ : [0, 1] → C such that
γ(0) = a, γ(1) = b, and γ([0, 1]) ⊂ S. It can be shown that an open subset of C is connected
if and only if it is path-connected.
Analytic function theory 80

A sequence of complex numbers (points on C) b is an enumeration, with repetitions allowed,

b we write, e.g., {z1 , z2 , . . .}, {zn }∞ , or simply {zn }.
of a set of complex numbers (points on C); n=1
We say that a sequence {zn } converges to a ∈ C if for every  > 0 there is some N (that may
depend on ) such that if n ≥ N then |zn − a| < . This is the same thing as saying that for
every open set U containing a, zn is in U if n is large enough. Phrased in this way it also makes
sense if a = ∞ ∈ C.
b We write limn→∞ zn = a to denote that the sequence {zn } converges to a.
Let f be a function defined on an open subset S of C or C b with values in either C or C. b
Then f is continuous at a point a ∈ S if for any sequence {zn } of points in S converging to
a the sequence {f (zn )} converges to f (a). This is the same as saying that for any open set U
containing f (a) there is an open set V ⊂ S containing a such that f (V ) ⊂ U . If f is continuous
at every point of S then we say that f is continuous (on S). In general, we say that f (z)
has a limit as z → a, or that limz→a f (z) exists, if {f (zn )} converges to the same fixed point
for any sequence {zn } (inside S) converging to a; limz→a f (z) then denotes this point. Notice
that in this terminology, f is continuous at a if and only if limz→a f (z) exists; necessarily then
limz→a f (z) = f (a) since the “constant” sequence zn = a converges to a.
The following lemma says in particular that convergence of sequences of complex numbers
is the same as convergence of the corresponding sequences of points in R2 , and that a function
f : C → C is continuous if and only if it is continuous considered as a function R2 → R2 .

Lemma 14.10. (a) A sequence {zn } of complex numbers converges if and only if the sequences
{Re zn } and {Im zn } of real numbers converge. Moreover, if {zn } converges to a, then {Re zn }
converges to Re a, {Im zn } converges to Im a, and {|zn |} converges to |a|.
(b) Let f (z) = u(z) + iv(z), where u(z) = Re f (z) and v(z) = Im f (z), be a map from an
open set S ⊂ C to C. Then limz→a f (z) exists if and only if both limz→a u(z) and limz→a v(z) ex-
ist. Moreover, if limz→a f (z) = A, then limz→a u(z) = Re A, limz→a v(z) = Im A, limz→a |f (z)| =
|A|, and limz→a f (z) = Ā.
(c) Let f (z) = u(z) + iv(z) be as in (b). Then f is continuous at a ∈ S if and only if both
u and v are continuous at a. Moreover, if f is continuous then |f | and f¯ are continuous.

Proof. The proofs of (a), (b), and (c) are similar; we prove (c) and leave (a) and (b) as exercises.
Assume that f is continuous at a and let  > 0 be given. From the definition of continuity
above it follows (how?) that there is a δ > 0 such that if |z − a| < δ then |f (z) − f (a)| < .
Hence, by computation rule (12) in Appendix A, if |z − a| < δ we have

|u(z) − u(a)| = |Re(f (z) − f (a))| ≤ |f (z) − f (a)| < .

Similarly, |v(z) − v(a)| < . Thus u and v are continuous at a.

Conversely, assume that u and v are continuous and let  > 0 be given. By assumption
it then follows that there is a δ > 0 such that if |z − a| < δ then |u(z) − u(a)| < /2 and
|v(z) − v(a)| < /2. Thus, if |z − a| < δ we get, using the Triangle inequality, that

|f (z) − f (a)| = |u(z) − u(a) + i(v(z) − v(a))| (14.7)

≤ |u(z) − u(a)| + |v(z) − v(a)| < /2 + /2 = .

Hence, f is continuous at a.
For the last statement of (c), assume again that f is continuous at a. We then know that
u and v are continuous and it follows in the usual way that u2 + v 2 is continuous, and non-
negative. Since the square-root√function is continuous [0, ∞) → R it follows in the standard
way that that the composition u2 + v 2 = |f | is continuous. Finally, f¯ = u−iv and computing
as in (14.7) shows that f¯ is continuous at a.
Analytic function theory 81

Exercises
B.1 Show that the following are open subsets of C

(a) {z ∈ C; Re z > 0}, (b) C \ R,

and that the following are closed

(c) {z ∈ C; |z| ≥ 1}, (d) {z ∈ C; |z − 1| ≤ |z + i|}.

B.2 Is {z ∈ C; Re z < 0} open, closed, or neither?

Analytic function theory 82

Bibliography
[1] Andersson, M. Topics in complex analysis. Universitext. Springer-Verlag, New York,
1997. viii+157 pp.

[2] Hörmander, L. An introduction to complex analysis in several variables. Third edi-

tion. North-Holland Mathematical Library, 7. North-Holland Publishing Co., Amsterdam,
1990. xii+254 pp.

[3] Priestley, H. A. Introduction to complex analysis. Revised second edition. Oxford Uni-
versity Press, Oxford, 2003. xiv+328 pp.