Complex Analysis

Milivoje Lukić
Emory University
Email address: [Link]@[Link]

Abstract. These lecture notes were written for the complex analysis
courses at Rice University and Emory University. Do not distribute
beyond this course.
These notes will be expanded and revised throughout the semester.

Copyright Milivoje Lukić 2025.

 Contents

Chapter 1. Complex numbers 5

1. Historical motivation: solving quadratic and cubic equations 5
2. The field of complex numbers 8
3. Complex conjugation, absolute value, and metric 10
4. Convexity and the affine structure 12
5. Inversion 14
6. Exercises 15

Chapter 2. Differential calculus and power series 21

1. Holomorphic functions 21
2. Power series 24
3. Exponential and trigonometric functions 28
4. Exercises 35

Chapter 3. Contour integration 39

1. Fundamental theorem of calculus for C 1 curves 39
2. Rectifiable and piecewise C 1 curves 41
3. Contour integrals on piecewise C 1 curves 43
4. Antiderivatives from contour integrals 46
5. Reparametrization and arclength 47
6. Bonus: a gallery of curves 51
7. Bonus: Lebesgue integral representation of rectifiable curves 54
8. Exercises 61

Chapter 4. The Cauchy integral theorem 63

1. Connected components; path connectedness 63
2. Antiderivatives and closed contours 66
3. Triangles and the Goursat argument 67
4. Cauchy integral theorem on star-shaped domains 70
5. Cauchy integral theorem on simply connected domains 70
6. Branches of the logarithm and roots 73
7. Exercises 77

Chapter 5. Cauchy integral formula and analytic functions 79

1. Cauchy integral formula for the disk 79
2. Cauchy power series theorem 82
3. Applications of the Cauchy power series theorem 84
 CONTENTS 3

4. Bonus: Abel’s theorem 88

5. Zeros of analytic functions and the identity theorem 90
6. Morera’s theorem and the reflection principle 92
7. Exercises 95

Chapter 6. Limits of analytic functions 99

1. Limits of analytic functions and Montel’s theorem 99
2. Bonus: metrizability of uniform convergence on compacts 101
3. Series, parametric integrals, and products of analytic functions 102
4. Bonus: Weierstrass product theorem 106
5. Bonus: the Gamma function 109
6. Exercises 113

Chapter 7. The argument principle and consequences 115

1. Winding numbers 115
2. The argument principle and Rouché’s theorem 119
3. Hurwitz’s theorem 121
4. Inverse function theorem 122
5. Bonus: implicit function theorem 124
6. Open mapping, the maximum principle, and the Schwarz lemma 125
7. A first look at conformal bijections 127
8. The Riemann mapping theorem 131
9. Exercises 133

Chapter 8. Meromorphic functions and the residue theorem 137

1. Cauchy integral formula for the annulus 137
2. Laurent series 139
3. Classification of isolated singularities 141
4. Meromorphic functions 144
5. Bonus: partial fraction decompositions 146
6. Chains of contours and the global Cauchy integral formula 149
7. Residue theorem and argument principle 152
8. Applications of contour integration 154
9. Exercises 160

Chapter 9. Riemann surfaces 165

1. The Riemann sphere and Möbius transformations 165
2. Riemann surfaces 168
3. Meromorphic maps on the Riemann sphere 170
4. Exercises 171

Appendix A. Some prerequisites 173

1. Upper and lower limits (lim sup and lim inf) 173
2. Connected sets 177

Appendix B. Class schedule Fall 2025 181

4 CONTENTS

Appendix C. Some superfluous notes 183

1. Rectifiability and improper integrals 183
 CHAPTER 1

Complex numbers

In this chapter, we introduce complex numbers, their algebraic structure,

and the connection with Euclidean geometry.

1. Historical motivation: solving quadratic and cubic equations

It’s been known since antiquity that the quadratic equation
ax2 + bx + c = 0 (1.1)
can be solved by using square roots. In particular:
Theorem 1.1. If a ̸= 0 and b2 − 4ac ≥ 0, the quadratic equation (1.1)
has real solutions √
−b ± b2 − 4ac
.
2a
Conversely, coefficients can be determined from zeros: by Viete’s formu-
las, if the quadratic equation has real solutions x1 , x2 , then
b c
− = x1 + x2 , = x1 x2 .
a a
2
√ point, the case b − 4ac < 0 can be used to motivate the object
At this
i = −1, that is, an imaginary constant i declared to have the algebraic
property i2 = −1. Then the quadratic equation has “complex solutions”
of the form x + yi, where x, y ∈ R. For instance, we can declare that the
quadratic equation x2 − 2x + 5 = 0 has complex solutions x = 1 ± 2i.
However, it is not yet clear whether this is a useful construct, or merely a
way to pretend that an equation has solutions.
Solving the cubic equation
ax3 + bx2 + cx + d = 0
was a greater challenge; its general solution was found in the 16th century
by del Ferro and later Tartaglia and first published by Cardano. By dividing
by a we can reduce to the case a = 1. By a linear substitution x̃ = x + b/3
we can reduce to the case b = 0. Tartaglia’s approach starts by comparing
the equation
x3 = px + q (1.2)
with the identity
(s + t)3 = 3st(s + t) + (s3 + t3 ).
5
6 1. COMPLEX NUMBERS

Therefore, x = s + t will be a solution of the cubic equation if we can find

s, t such that
3st = p, s3 + t3 = q.
By Viete’s formulas, this is equivalent to asking s3 , t3 to be the two solutions
of the quadratic equation
 p 3
y 2 − qy + = 0.
3
This leads to r 
3 3 q q 2  p 3
s ,t = ± −
2 2 3
and finally
s r  s r 
3 q q 2 p 3 3 q q 2  p 3
 
x= + − + − − . (1.3)
2 2 3 2 2 3
This is known as Tartaglia’s formula. Recall that t 7→ t3 is a bijection from
R to R, and the cubic root is its inverse; in particular, we can take cubic
roots of negative numbers.
The values s3 , t3 are real if q 2 /4 − p3 /27 ≥ 0, so we have proved:
Theorem 1.2. If p, q ∈ R and q 2 /4−p3 /27 ≥ 0, the cubic equation (1.2)
has a real solution (1.3).
Tartaglia’s formula is not completely straightforward to use:
Example 1.3. Find a real solution of x3 = −6x + 20.
Solution. Using Tartaglia’s formula,
√ √ √ √
q q q q
3 3 3 3
x = 10 + 108 + 10 − 108 = 10 + 6 3 + 10 − 6 3.
With some algebraic effort, we may guess that
√ √
(1 ± 3)3 = 10 ± 6 3
so a real solution is
√ √
x = (1 + 3) + (1 − 3) = 2. □
The quantity
q2 p3
D= −
4 27
is called the discriminant of the cubic polynomial, since it determines the
number of real solutions (see Exercise 2 and Figure 1.1).
Theorem 1.2 finds a real zero if D ≥ 0 and Exercise 3 gives two complex
conjugate solutions. In the case D < 0, the cubic equation still has a real
zero, but the interpretation of Tartaglia’s formula requires complex numbers.
Let us see this in the following example, which dates back to Cardano.
Example 1.4. Find a real solution of x3 = 15x + 4.
 1. HISTORICAL MOTIVATION: SOLVING QUADRATIC AND CUBIC EQUATIONS 7

Figure 1.1. Examples of cubic polynomials with positive,

zero, and negative discriminant, respectively

Solution. Tartaglia’s formula says

√ √
x = 3 2 + 11i + 3 2 − 11i.
To make sense of these cubic roots, we have to make some assumptions
about how to compute with complex numbers. If we assume that algebraic
rules for complex numbers are like those for real numbers, then using the
binomial formula
(a + bi)3 = a3 + 3a2 bi + 3ab2 i2 + b3 i3
= a3 + 3a2 bi − 3ab2 − b3 i
and we can somehow guess that
(2 ± i)3 = 2 ± 11i.
√
3
If we then interpret 2 ± 11i = 2 ± i, we would find
x = (2 + i) + (2 − i) = 4.
Although we have not justified these calculations, the final answer x = 4 is
indeed a real solution of the cubic equation. □

Note that a calculation using complex numbers led to a real solution.

This promotes the idea that complex numbers can be a useful abstract tool,
but it leads to new algebraic questions. For instance, cubic root is not
uniquely determined on complex numbers: we will eventually see that any
nonzero complex number has three cubic roots. If we could choose each
cubic root in Tartaglia’s formula in three ways, this would result in nine
solutions of the cubic equation, which is impossible. Exercise 4 addresses this
issue rigorously and reduces the problem of cubic equations with negative
discriminant to the problem of finding third roots of complex numbers.
Exercise 5 shows how to solve quartic equations in terms of roots (rad-
icals). Solvability in terms of roots leads to interesting algebraic questions,
which are solved by Galois theory and not part of this course. On the
other hand, existence of real/complex solutions is an analysis fact (ironi-
cally, called the fundamental theorem of algebra): every polynomial with
complex coefficients of degree at least 1 has at least one complex zero. We
will prove this in this course, but it will take some time to get there.
8 1. COMPLEX NUMBERS

2. The field of complex numbers

Definition 1.5. An algebra A over the field R is a vector space over R
equipped with a bilinear product A × A → A (bilinear means: linear in each
of the two variables).
The reader already knows at least one algebra:
Example 1.6. The vector space of n × n real matrices, denoted Rn×n ,
equipped with matrix multiplication, is an algebra over R.
Complex numbers are often informally introduced as numbers of the
form x + yi, where x, y ∈ R, with a prescribed property ii = −1. This can
be formalized into an axiomatic way of introducing complex numbers:
Definition 1.7. The algebra of complex numbers C is an algebra over
the field R with a distinguished basis 1, i such that
11 = 1, 1i = i, i1 = i, ii = −1. (1.4)
The elements of C are called complex numbers.
Since 1, i is a basis, part of this definition is that every complex number
can be uniquely written in the form
z = a1 + bi, a, b ∈ R,
and conversely, that for every a, b ∈ R, this gives a complex number. The
definition requires that we distinguish between the scalar (real number) 1
and the vector (complex number) 1, but our first observation will be that
we can identify 1 with 1 and identify R with a subset of C.
Lemma 1.8. The map x 7→ x1 embeds the field R into C, in the following
sense:
(a) (x1 + x2 )1 = x1 1 + x2 1 for all x1 , x2 ∈ R
(b) (x1 x2 )1 = (x1 1)(x2 1) for all x1 , x2 ∈ R
(c) 01 = 0
(d) 11 = 1
(e) (x1)z = xz for all x ∈ R, z ∈ C
Proof. The proof consists of direct calculations, using properties of an
algebra and 11 = 1 and 1i = i. □
Due to this, it is standard to identify R with the subset R1 of C, and
the real constant 1 with the complex number 1. Accordingly, from now on
we will write complex numbers in the form
z = a + bi, a, b ∈ R.
The coefficients a, b ∈ R are called the real and imaginary part of z, respec-
tively, and we write Re z = a, Im z = b.
Lemma 1.8 addresses another formal issue. Algebraically speaking, there
are three different multiplication operations here: multiplication in R, scalar
 2. THE FIELD OF COMPLEX NUMBERS 9

multiplication (real times complex number), and the bilinear product (com-
plex times complex number). By Lemma 1.8, the first two are special cases
of the bilinear product. We simply call this complex multiplication from
now on.
In general, multiplication in an algebra may not be commutative or even
associative. However, the algebra of complex numbers has good properties:
Proposition 1.9. C is a field with the additive neutral element 0 and
the multiplicative neutral element 1, i.e.
(a) (z1 + z2 ) + z3 = z1 + (z2 + z3 )
(b) z+0=0+z =z
(c) for every z there exists −z such that z + (−z) = (−z) + z = 0
(d) z1 + z2 = z2 + z1
(e) (z1 z2 )z3 = z1 (z2 z3 )
(f ) z1 = 1z = z
(g) for every z ̸= 0 there exists z −1 such that zz −1 = z −1 z = 1
(h) z1 z2 = z2 z1
(i) z1 (z2 + z3 ) = z1 z2 + z1 z3
Proof. (a), (b), (c), (d) follow from the vector space structure.
(e) For arbitrary zj = aj +bj i, direct calculations of (z1 z2 )z3 and z1 (z2 z3 )
imply that they are equal, which implies associativity.
(f) This follows from 11 = 1 and 1i = i1 = i.
(g) For a, b ∈ R not both zero, it is directly verified that the multiplica-
tive inverse of z = a + bi is the complex number
a b
z −1 = − 2 i.
a2 +b2 a + b2
(h) By direct calculations using 1i = i1, we obtain z1 z2 = z2 z1 .
(i) follows from bilinearity of the product. □

Constructions of the algebra of complex numbers. The algebra

of complex numbers can be constructed in several ways. The most common
construction is based on R2 as a vector space over R:
Proposition 1.10. The vector space R2 with the bilinear product
(a, b)(c, d) = (ac − bd, ad + bc)
forms an algebra of complex numbers with distinguished basis
1 = (1, 0), i = (0, 1).
Proof. It is elementary to see that this product is bilinear and that it
obeys (1.4). Moreover, (1, 0), (0, 1) form a basis of R2 . □

Another construction is based on a 2-dimensional subset of the vector

space R2×2 of 2 × 2 matrices with real entries.
10 1. COMPLEX NUMBERS

Proposition 1.11. The algebra R2×2 of 2 × 2 real matrices has a sub-

algebra   
a −b
A= | a, b ∈ R .
b a
This is an algebra of complex numbers with distinguished basis
   
1 0 0 −1
1= , i= .
0 1 1 0
The verification of this model is left to Exercise 6. This model will
naturally appear when we discuss Cauchy–Riemann relations.
Uniqueness of the algebra/field of complex numbers. Although
there are different ways of constructing complex numbers, these different
constructions are equivalent: they produce models which behave in the same
way. This is shown by a bijection which preserves the structure:
Proposition 1.12. If C1 , C2 are algebras of complex numbers, there
exists a bijection from C1 to C2 which preserves addition, multiplication,
and maps the distinguished basis of C1 to that of C2 .
Proof. The map f (a11 + bi1 ) = a12 + bi2 has these properties. □

3. Complex conjugation, absolute value, and metric

Definition 1.13. The complex conjugate of z = a + bi with a, b ∈ R is
z = a − bi
Lemma 1.14. Complex conjugation is a field automorphism, i.e. it is a
bijection and for all z1 , z2 ∈ C,
z 1 + z 2 = z1 + z2 , z 1 z 2 = z1 z2 , 1 = 1.
Conversely, any field automorphism of C that preserves R is identity or
complex conjugation (Exercise 7).
Computationally, the complex conjugate is related to the inverse: note
that
zz = (a + bi)(a − bi) = a2 + b2
so for any z = a + bi ̸= 0,
z a − bi
z −1 = 2 2
= 2 .
a +b a + b2
More generally, the standard trick to simplify a fraction wz is to expand the
zw
fraction by w, because the form ww has a real denominator which can be
distributed into the real and imaginary parts. For example:
2−i (2 − i)(1 − i) 1 − 3i 1 3
= = = − i
1+i (1 + i)(1 − i) 2 2 2
It is also common to express Re z, Im z in terms of z, z as follows,
z+z z−z
Re z = , Im z = .
2 2i
 3. COMPLEX CONJUGATION, ABSOLUTE VALUE, AND METRIC 11

Definition 1.15. For z = a+bi, we define its modulus or absolute value

p
|z| = a2 + b2 .
Lemma 1.16. For all z ∈ C,
|z|2 = z z̄.
Proof. Both sides of this equality are equal to a2 + b2 . □
Corollary 1.17. For all z, w ∈ C, |zw| = |z||w|.
Note that |z| gives the usual Euclidean length of the vector (a, b). In
particular, it is well known that Euclidean distance obeys the triangle in-
equality, but we can also give a direct proof:
Lemma 1.18. For all z1 , z2 ∈ C,
|z1 + z2 | ≤ |z1 | + |z2 |.
Equality holds if and only if z1 z2 ≥ 0.
Proof. First, note that
p
|w| = (Re w)2 + (Im w)2 ≥ |Re w| ≥ Re w
so Re w ≤ |w|, and equality holds if and only if w ≥ 0.
Now, we will prove the square of the triangle inequality. Starting with
the left-hand side,
|z1 + z2 |2 = (z1 + z2 )(z1 + z2 ) = z1 z̄1 + z1 z̄2 + z2 z̄1 + z2 z̄2 .
Since
z1 z̄2 + z2 z̄1 = 2 Re(z1 z̄2 ) ≤ 2|z1 z̄2 | = 2|z1 ||z2 |,
this gives
|z1 + z2 |2 ≤ |z1 |2 + 2|z1 ||z2 | + |z2 |2 = (|z1 | + |z2 |)2 .
Taking the square roots of both sides shows the triangle inequality. □
Corollary 1.19. C is a metric space with the metric d(z, w) = |z − w|.
This metric coincides with the Euclidean metric on R2 , so we often
talk about the complex plane, which is a visualization of complex numbers
as points in a Euclidean plane, with the identification of complex number
z = x + yi and the point (x, y) ∈ R2 . We carry over some geometric
terminology and talk about lines, circles, and disks in the complex plane.
Balls of radius r are precisely Euclidean disks of radius r, and we will denote
Dr (z0 ) = {z ∈ C | |z − z0 | < r}.
See also Exercises 13, 14.
Exercises 15, 17 illustrate how complex numbers can be used to study
the geometry of the Euclidean plane.
As on any metric space, we can talk about open and closed sets, limits
of sequences and functions, series, and continuity.
The following continuity statements hold similarly as in real analysis:
12 1. COMPLEX NUMBERS

Theorem 1.20. The following functions are continuous:

(a) complex addition C × C → C, (z, w) 7→ z + w
(b) complex multiplication C × C → C, (z, w) 7→ zw
(c) multiplicative inverse C \ {0} → C \ {0}, z 7→ 1/z
(d) complex conjugation C → C, z 7→ z̄
(e) absolute value C → C, z 7→ |z|
These are used in the usual way to establish continuity
Pn of other “simple”
functions. For instance, any polynomial p(z) = j=0 aj z j is a continuous
function.

4. Convexity and the affine structure

The vector space structure of C provides some further linear notions.

Convexity. By definition, C is a vector space of dimension 2 over the

field of scalars R, with basis {1, i}. The vector space structure over R gives
a notion of convexity:
Definition 1.21. A set S ⊂ C is convex if for any z1 , z2 ∈ S, t ∈ [0, 1],
it also holds that (1 − t)z1 + tz2 ∈ S.
Trivially, C itself is convex. In addition:
Example 1.22. The upper half-plane
C+ = {z ∈ C | Im z > 0}
is a convex set.
Proof. If z1 , z2 ∈ C+ , then Im z1 > 0 and Im z2 > 0, so for any t ∈
[0, 1],
Im((1 − t)z1 + tz2 ) = (1 − t) Im z1 + t Im z2 > 0
so (1 − t)z1 + tz2 ∈ C+ . □

Similarly, any half-plane in C is convex.

Example 1.23. The unit disk
D = {z ∈ C | |z| < 1}
is a convex set.
Proof. If z, w ∈ D, then for any t ∈ (0, 1),
|(1 − t)z + tw| ≤ |(1 − t)z| + |tw| = (1 − t)|z| + t|w| < (1 − t) + t = 1
so (1 − t)z + tw ∈ D. □

Similarly, any disk Dr (z0 ) is a convex set.

 4. CONVEXITY AND THE AFFINE STRUCTURE 13

Definition 1.24. A convex combination of z1 , . . . , zn ∈ C is any linear

combination of the form
Xn
cj zj
j=1
Pn
with cj ≥ 0 for all j and j=1 cj = 1. The convex hull of A is the set of all
convex combinations of elements of A,
 
X n n
X 
cvh A = cj zj | n ∈ N, z1 , . . . , zn ∈ A, c1 , . . . , cn ≥ 0, cj = 1 .
 
j=1 j=1

Note that convex combinations, like linear combinations, are only of

finitely many elements at a time; even if the set A is infinite, there are no
series in the definition of cvh A.
Lemma 1.25. The convex hull of A is the smallest convex set containing
A in the sense that:
(a) cvh A is convex and A ⊂ cvh A;
(b) if S is convex and A ⊂ S, then cvh A ⊂ S.
Proof. (a) With n = 1, we see that 1z1 is a convex combination of
z1 ∈ A, so A ⊂ cvh A.
Any two elements w1 , w2 ∈ cvh A can be written in the form
n
X m
X
w1 = cj zj , w2 = c̃j z̃j
j=1 j=1

so, for t ∈ [0, 1],

n+m
(
X (1 − t)cj j≤n
(1 − t)w1 + tw2 = dj zj , dj =
j=1
tc̃j−n j>n
Pn+m
with the notation zj = z̃j−n for j > n. Since dj ≥ 0 and j=1 dj = 1, this
shows that cvh A is convex.
(b) By induction in n using
Xn n−1
X  cj 
cj zj = (1 − cn ) zj + cn zn ,
1 − cn
j=1 j=1

if a convex set contains z1 , . . . , zn , then it contains all their convex combi-

nations, so if S is convex and A ⊂ S, then cvh A ⊂ S. □
In other words, the convex hull of A ⊂ C is the intersection of all convex
sets containing A:
\
cvh A = S.
S⊂C
A⊂S
S is convex
14 1. COMPLEX NUMBERS

The affine structure and affine maps on C. An affine map on C is

a map of the form
z 7→ αz + β
for some α, β ∈ C.
An affine map is invertible if and only if α ̸= 0.
Example 1.26. For any z0 ∈ C and r > 0, the affine map
z − z0
z 7→
r
maps the disk Dr (z0 ) bijectively to the unit disk D = D1 (0). In later chap-
ters, we will often formulate results on D, with the realization that these
results can be moved to an arbitrary disk by composing with an affine map.
Example 1.27. The affine map z 7→ iz maps the right half-plane {z |
Re z > 0} to the upper half-plane, denoted
C+ = {z ∈ C | Im z > 0}.
Proof. This follows from
iz − (iz) iz + iz z+z
Im(iz) = = = = Re z. □
2i 2i 2
Similarly, affine maps can be used to map any half-plane to C+ .
Affine maps preserve linear combinations, convexity, ratios
(z2 − z0 )/(z1 − z0 ),
and orthogonality.

5. Inversion
Inversion is the map z 7→ 1/z (note: in Euclidean geometry, inversion
usually refers to a different map on the Euclidean plane, corresponding to
z 7→ 1/z). Inversion is a bijection from C \ {0} to itself. It has interesting
geometric properties:
Example 1.28. Inversion maps D \ {0} bijectively to C \ D.
Proof. For any z ∈ C \ {0},
1 1
=
z |z|
1
so 0 < |z| < 1 if and only if z > 1. □
Example 1.29. Inversion maps the disk |z − 12 | < 1
2 bijectively to the
half-plane Re z > 1.
Proof. For any z ∈ C \ {0},
1
− |z − 12 |2
 
1 1 1 1 z + z − 2zz 4
Re − 1 = + −1= =
z 2 z z 2zz |z|2
It follows that |z − 12 | < 1
2 if and only if Re z1 > 1. □
 6. EXERCISES 15

Composing inversion with affine maps, we can construct bijections be-

tween arbitrary disks and half-planes. One notable example is:
Example 1.30. The Cayley transformation
1+z
γ(z) = i
1−z
maps D bijectively to C+ .
Proof. f (z) = −z+1
2 maps D bijectively to D1/2 (1/2).
g(z) = 1/z maps D1/2 (1/2) bijectively to the half-plane Re z > 1.
h(z) = i(z − 1) maps the half-plane Re z > 1 bijectively to C+ .
A calculation shows γ = h ◦ g ◦ f . □
The Cayley transformation and its inverse
w−i
γ −1 (w) =
w+i
are often useful. By continuity of γ and γ −1 , the Cayley transformation
maps ∂D \ {1} bijectively to R.
It would have been more elegant to view inversion and the Cayley trans-
form as bijections C ∪ {∞} → C ∪ {∞}; we will make this point of view
rigorous when we consider the Riemann sphere and Möbius transformations.
More generally, inversion maps any disk, half-plane, or exterior of a disk,
to another set of the same form (Exercise 21). What looks like a list of cases
will be simplified soon by viewing all these sets as “generalized disks” on
the Riemann sphere.
Inversion does not preserve ratios, but it preserves the cross-ratios of
four points,
z2 − z 0 z3 − z0
(z0 , z1 ; z2 , z3 ) = :
z2 − z 1 z3 − z1
6. Exercises
√ √
1. (a) Verify that (1 ± 3)3 = 10 ± 6 3.
(b) Verify that (2 ± i)3 = 2 ± 11i.
2. Let p, q ∈ R and denote D = q 2 /4 − p3 /27. Prove that the polynomial
x3 − px − q = 0 has:
(a) one real root and two complex conjugate roots if D > 0
(b) all real roots, including a real root of multiplicity at least 2 if D = 0
(c) all real simple roots if D < 0
Hint: find the critical points (zeros of the derivative) of the cubic
polynomial and use the intermediate value theorem.
3. (a) Show that the cubic equation z 3 − 1 = 0 has the real solution z = 1
and two complex conjugate solutions ω, ω where
√
1 3
ω =− +i .
2 2
Hint: factor z − 1 to reduce to a quadratic equation.
16 1. COMPLEX NUMBERS

(b) Show that ωω = 1.

(c) Assume p, q ∈ R and q 2 /4 − p3 /27 > 0. Show that the cubic equation
(1.2), in addition to the real solution (1.3), two distinct non-real
solutions
s r  s r 
3 q q 2  p 3 3 q q 2  p 3
z=ω + − +ω − −
2 2 3 2 2 3
and
s r  s r 
3 q q 2  p 3 3 q q 2  p 3
z=ω + − +ω − − .
2 2 3 2 2 3
4. Assume p, q ∈ R and q 2 /4 − p3 /27 < 0, and let w be a complex number
such that r 
3 q p 3  q 2
w = +i − . (1.5)
2 3 2
(a) Prove that x = w + w is a real solution of the cubic equation (1.2).
(b) Prove that different values of w satisfying (1.5) give different values
of x.
(Thus, once we know that there are three different values of w obeying
(1.5), we will know that this procedure gives all three solutions of the
cubic equation)
5. This exercise explains a method for solving quartic equations in the re-
duced form
x4 + αx2 + βx + γ = 0 (1.6)
where α, β, γ ∈ R.
(a) Show that the factorization
x4 + αx2 + βx + γ = (x2 − sx + r)(x2 + sx + t) (1.7)
holds if and only if
r + t − s2 = α, (r − t)s = β, rt = γ (1.8)
(b) Show that (1.8) implies
s6 + 2αs4 + (α2 − 4γ)s2 − β 2 = 0.
Hint: use the identity (r + t)2 − (r − t)2 = 4rt.
(c) Show that the equation
σ 3 + 2ασ 2 + (α2 − 4γ)σ − β 2 = 0 (1.9)
has at least one nonnegative solution.
(d) For any strictly positive solution σ of (1.9), show that
√ α + s2 + βs−1 α + s2 − βs−1
s= σ, r= , t=
2 2
gives a factorization (1.7).
 6. EXERCISES 17

(e) If σ = 0 is a solution of (1.9), show that

p p
α + α2 − 4γ α − α2 − 4γ
s = 0, r = , t=
2 2
gives a factorization (1.7).
This gives a method for solving the quartic equation (1.6) by solving
the cubic equation (1.9), finding a factorization (1.7), and solving some
quadratic equations.
6. Denote   
a −b
A= | a, b ∈ R .
b a
Prove that A is a subset of 2 × 2 real matrices closed under addition and
multiplication, that it is a field, and that it is an isomorphic field to C.
7. If f : C → C is a field automorphism that preserves R, prove that f = id
or f is complex conjugation.
8. Compute
3+i
2−i
(your answer should be in the form a + bi with a, b ∈ R).
9. In this problem, (a) is a well-known fact using real numbers; (b) is a
related identity derived using complex numbers.
(a) For n ∈ N, prove that
n  
X n
(−1)j = 0.
j
j=0

Hint: start with the binomial sum in the form nj=0 nj z j = (1 + z)n
P

evaluated at a suitable z.
(b) For m ∈ N, find the value of the sum
m  
X 2m
(−1)k .
2k
k=0

You should find the value in an explicit form which depends on the
value of m modulo 4; for instance, it should be obvious from your
answer for which m the sum is positive or negative. Pn n j n


Hint: start again with the binomial sum j=0 j z = (1 + z) ,
evaluated at z = i, and take the real part. To try to find a pattern
in the powers of 1 + i, start by computing (1 + i)2 .
10. Let z1 , z2 ∈ C. Prove that |z1 + z2 | < |z1 | + |z2 | if and only if z1 /z2 ∈ /
(−∞, 0].
11. Let z1 , z2 ∈ C. Prove that |z1 − z2 | = |z1 | + |z2 | if and only if 0 is a convex
combination of z1 and z2 .
12. Prove the parallelogram identity: for all z1 , z2 ∈ C,
|z1 + z2 |2 + |z1 − z2 |2 = 2|z1 |2 + 2|z2 |2 .
 18 1. COMPLEX NUMBERS

13. Prove that any line in C can be represented in the form

Re(αz) = C
for some α ∈ C \ {0} and C ∈ R, and conversely, that every equation of
this form describes a line.
Hint: recall that in Euclidean geometry lines are described by equa-
tions Ax + By = C, with (A, B) ̸= (0, 0).
14. Prove that any open half-plane in C can be represented in the form
Re(αz) > C
for some α ∈ C \ {0} and C ∈ R, and conversely, that every inequality of
this form describes an open half-plane.
Hint: recall that half-planes in R2 are described by equations Ax +
By > C, with (A, B) ̸= (0, 0).
15. (Circles of Apollonius) Prove that for any ρ > 1, the set of points z such
that |z|/|z − 1| = ρ is a circle. Find the center and radius of that circle.
Hint: square the equation |z| = ρ|z − 1|, group terms and complete
the square to
obtain
axformula of the form |z − c|2 = r2 .
x0 x1 2


16. (a) Points y0 , y1 , y2 ∈ R are said to form a triangle if they are not
2

collinear. Prove that z0 , z1 , z2 ∈ C form a triangle if and only if they

are distinct and (z2 − z0 )/(z1 − z0 ) ∈ C \ R
(b) The triangle is said to be positively oriented if
x1 − x0 x2 − x0
> 0.
y1 − y0 y2 − y0
Prove that this is equivalent to
z2 − z0
Im > 0.
z1 − z0
(c) If the triangle [z0 , z1 , z2 ] is positively oriented, prove that [z1 , z2 , z0 ]
is positively oriented.
17. (a) For any triangle in C, prove that there exists an affine map z 7→ cz+β,
c > 0 which maps it to a triangle [z0 , z1 , z2 ] with |z0 | = |z1 | = |z2 | = 1.
(b) For a triangle [z0 , z1 , z2 ] with |z0 | = |z1 | = |z2 | = 1, prove that all
three altitudes of the triangle pass through the point
h = z0 + z1 + z 2
which is called the orthocenter of the triangle.
Hint: vectors corresponding to complex numbers u, v are mutually
orthogonal if and only if Re(uv) = 0, since Re(uv) corresponds to the
Euclidean dot product.
(c) Prove that in any triangle, the orthocenter, circumcenter, and cen-
troid lie on the same line, called the Euler line.
18. Let a, b, c, d be points on the unit circle. Assume a ̸= b and c ̸= d.
 6. EXERCISES 19

(a) Prove that lines ab and cd are parallel if and only if ab = cd. (no-
tational note: here ab denotes both the line passing through points
a, b, and the product of complex numbers a and b)
(b) If ab ̸= cd, prove that the intersection of lines ab and cd is the point
ab(c + d) − cd(a + b)
.
ab − cd
19. (Pascal’s theorem on the circle) Let z1 , z2 , z3 , w1 , w2 , w3 ∈ ∂D. Use cyclic
notation zj+3 = zj , wj+3 = wj . Denote by uj the intersection of lines
zj+1 wj+2 and zj+2 wj+1 . If u1 , u2 , u3 exist, prove that they lie on one line.
20. (a) Consider a sequence P of complexPnumbers such that Re zj ≥ 0 for
all j. If the series ∞ z
j=1 j and ∞
z
j=1 j
2 are convergent, prove that
P∞ 2 is convergent.
j=1 |z j |
(b) PProve that there P∞exists a sequence zj of complex
P∞ numbers such that
∞ 2 2
j=1 zj and j=1 zj are convergent, but j=1 |zj | is divergent.
21. Prove some of the following statements. Inversion z 7→ 1/z maps the
following sets bijectively:
(a) Dr (z0 ) with |z0 | > r to a disk DR (w)
(b) Dr (z0 ) with |z0 | = r to a half-plane Re(αw) = C with C > 0
(c) Dr (z0 ) \ {0} with |z0 | < r to an exterior of a disk C \ DR (w)
(d) half-plane Re(αz) = C with C > 0 to a disk
(e) half-plane Re(αz) = C with C = 0 to a half-plane
(f) half-plane Re(αz) = C with C < 0 to an exterior of a disk
22. Prove that inversion preserves cross-ratios.
 CHAPTER 2

Differential calculus and power series

In this chapter, we begin to study complex analysis. The origins of

complex analysis are in special functions, including de Moivre’s formula
(cos t + i sin t)n = cos(nt) + i sin(nt)
and Euler’s formula
es+it = es (cos t + i sin t).
These early observations motivated the study of functions of a complex vari-
able, but in this course we will take a more modern approach to exponential
and trigonometric functions: we will construct them from scratch as func-
tions of a complex variable.

1. Holomorphic functions
An open subset Ω ⊂ C will also be called a domain. We now define the
holomorphic property (informally speaking, holomorphic = complex differ-
entiable):
Definition 2.1. Assume f : Ω → C. For any z0 ∈ Ω, we say f is
holomorphic at z0 if the limit
f (z) − f (z0 )
f ′ (z0 ) = lim (2.1)
z→z0 z − z0
exists. This value is called the derivative of f at z0 .
If f is holomorphic at every z0 ∈ Ω, we say that f is holomorphic.
The basic properties of the complex derivative are the same as in the
real-variable case; we will skip many of the proofs, since they are identical
to the real case.
Theorem 2.2. f is holomorphic at z0 if and only if there exists a value
f ′ (z
0 ) ∈ C such that

f (z) = f (z0 ) + f ′ (z0 )(z − z0 ) + o(|z − z0 |), z → z0 . (2.2)

Proof. (2.1) is equivalent to (2.2) by algebraic rearrangements. □
Theorem 2.3. If f is holomorphic at z0 , then f is continuous at z0 .
Proof. As z → z0 ,
f (z) − f (z0 )
f (z) − f (z0 ) = (z − z0 ) → f ′ (z0 ) × 0 = 0. □
z − z0
21
22 2. DIFFERENTIAL CALCULUS AND POWER SERIES

Theorem 2.4 (Sum and product rules). If f, g : Ω → C are holomorphic

at the point z0 ∈ Ω, then
(a) (f + g)′ (z0 ) = f ′ (z0 ) + g ′ (z0 )
(b) (f g)′ (z0 ) = f ′ (z0 )g(z0 ) + f (z0 )g ′ (z0 )
Example 2.5. (1) If f (z) = c then f ′ (z) = 0.
(2) If f (z) = z then f ′ (z) = 1.
(3) For any nP ∈ N, if f (z) = z n then fP′ (z) = nz n−1 .

(4) If f (z) = k=0 ck z then f (z) = nk=1 kck z k−1 .

n k ′

(5) If f (z) = 1/z, then f ′ (z) = −1/z 2 , because

f (z) − f (z0 ) 1 1
=− → − 2, z → z0 .
z − z0 zz0 z0
Theorem 2.6 (Chain rule). If f : Ω → C is holomorphic at z0 , g :
Ω̃ → C holomorphic at the point f (z0 ), and f (Ω) ⊂ Ω̃. Then the function
h = g ◦ f is holomorphic at z0 and
h′ (z0 ) = g ′ (f (z0 ))f ′ (z0 ).
Proof. Define u : Ω → C by
(
f (z)−f (z0 )
z−z0 − f ′ (z0 ) z ̸= z0
u(z) =
0 z = z0
Since f is holomorphic at z0 , note that u is continuous at z0 , and write
f (z) − f (z0 ) = (z − z0 )(f ′ (z0 ) + u(z)). (2.3)
This is mostly formal, but note that (2.3) holds for all z ∈ Ω, including
z = z0 .
Similarly, at w0 = f (z0 ),
g(w) − g(w0 ) = (w − w0 )(g ′ (w0 ) + v(w))
where v is continuous at w0 and v(w0 ) = 0. Then
h(z) − h(z0 ) = g(f (z)) − g(f (z0 ))
= (f (z) − f (z0 ))(g ′ (f (z0 )) + v(f (z)))
= (z − z0 )(f ′ (z0 ) + u(z))(g ′ (f (z0 )) + v(f (z)))
Since z → z0 implies u(z) → 0, f (z) → f (z0 ), and then v(f (z)) → v(f (z0 )),
this implies
h(z) − h(z0 )
= (f ′ (z0 ) + u(z))(g ′ (f (z0 )) + v(f (z))) → f ′ (z0 )g ′ (f (z0 ))
z − z0
as z → z0 . □
Corollary 2.7 (Quotient rule). If f, g are holomorphic at z0 ∈ Ω and
′ ′
g(z0 ) ̸= 0, then f /g is holomorphic at z0 and (f /g)′ (z0 ) = g(z0 )f (zg(z
0 )−g (z0 )f (z0 )
0 )2
 1. HOLOMORPHIC FUNCTIONS 23

Cauchy–Riemann equations. Denote u = Re f , v = Im f . Let us

view f also as a vector-valued map F : Ω → R2 with
F (x, y) = (Re f (x + iy), Im f (x + iy)) = (u(x + iy), v(x + iy)).
In multivariable calculus, we say that a function F : Ω → R2 is differentiable
at a point (x0 , y0 ) ∈ Ω if
 
x − x0
F (x, y) = F (x0 , y0 ) + A + o(|(x, y) − (x0 , y0 )|), (x, y) → (x0 , y0 ).
y − y0
(2.4)
2×2
for some 2 × 2 matrix A ∈ R . The value of the matrix A is called the
derivative and denoted (DF )(x0 , y0 ).
To compare (2.2) and (2.4), note that the constant terms are the same:
F (x0 , y0 ) = f (x0 + iy0 ) by definition. Moreover, the error terms are the
same, since the metric on C is precisely the one on R2 . Thus, it remains to
compare the linear terms, using this lemma:
Lemma 2.8. The map L(z) = (a + ib)z, viewed as a map L : R2 → R2 ,
is a linear transformation which is represented in matrix form as
    
Re L(z) a −b Re z
= .
Im L(z) b a Im z
Proof. This is just a calculation, reading off matrix coefficients from
L(z) = (a + bi)(Re z + i Im z) = (a Re z − b Im z) + i(a Im z + b Re z). □
Theorem 2.9 (Cauchy–Riemann equations). f is holomorphic at z0 if
and only if F is differentiable at (x0 , y0 ) and the partial derivatives obey
∂u ∂v ∂v ∂u
= , =− .
∂x ∂y ∂x ∂y
In this case,
∂u ∂v ∂v ∂u
Re f ′ = = , Im f ′ = =− .
∂x ∂y ∂x ∂y
Proof. Comparing the holomorphicity statement (2.2) with the (real)
differentiability statement (2.4), we see that the constant terms and the
error terms are the same. By the previous lemma, the linear term in (2.2)
is of the form   
a −b x − x0
(a + ib)(z − z0 ) =
b a y − y0
where a = Re f ′ (z0 ), b = Im f ′ (z0 ). Thus, f is holomorphic with derivative
f ′ if and only if f is differentiable with matrix derivative
Re f ′ − Im f ′
   
∂x u ∂ y u
= □
∂x v ∂y v Im f ′ Re f ′
Example 2.10. The function f (z) = z̄ is not holomorphic.
24 2. DIFFERENTIAL CALCULUS AND POWER SERIES

Proof. This function corresponds to F (x, y) = (x, −y). As a function

F : R2 → R2 , this is differentiable, but
 
1 0
DF =
0 −1
so the Cauchy–Riemann equations do not hold. □

Geometrically speaking, complex conjugation is a reflection which is

incompatible with the Cauchy–Riemann equations. As a result, expressions
involving z̄ are usually not holomorphic (for instance, a polynomial in z, z̄
is only holomorphic if it can be expressed as a polynomial in z only). An
important family of exceptions to this ”rule” is obtained when there are two
complex conjugations:
Example 2.11. If the function f : Ω → C is holomorphic, then the
function f♯ (z) = f (z) is holomorphic on Ω♯ = {z | z ∈ Ω}. Moreover,
f♯′ (z) = f ′ (z).

Proof. We first note that complex conjugation is a continuous function

on C, since |z − w| = |z − w|. Next, from the definition of derivative,
 
f♯ (z) − f♯ (z0 ) f (z) − f (z 0 ) f (z) − f (z 0 )
= =
z − z0 z − z0 z − z0
Since complex conjugation is continuous,
 
f♯ (z) − f♯ (z0 ) f (z) − f (z 0 )
lim = lim = f ′ (z 0 ). □
z→z0 z − z0 z→z0 z − z0

An alternative proof would use Cauchy–Riemann equations and the cal-

culation      
1 0 a −b 1 0 a b
= .
0 −1 b a 0 −1 −b a

2. Power series
A power series in z centered at z0 is a formal sum
∞
X
an (z − z0 )n
n=0

where z0 ∈ C, an ∈ C for all n, and z is a complex variable.

Radius of convergence. In analysis, it is crucial to understand where

the series converges. In real analysis, a power series has an interval of
convergence; in complex analysis, by following the same arguments, we will
obtain a disk of convergence.
 2. POWER SERIES 25

Theorem 2.12 (Cauchy radius formula). Let

1
R= .
lim supn→∞ |an |1/n
Note R ∈ [0, ∞]; in particular, R is allowed to be 0 or ∞. The power series
converges absolutely for |z − z0 | < R and diverges for |z − z0 | > R. It defines
a continuous function on DR (z0 ).
Moreover, for any radius 0 < r < R, the power series converges ab-
solutely and uniformly in Dr (z0 ) = {z ∈ C | |z − z0 | ≤ r} in the sense
that
X∞
sup |an (z − z0 )n | < ∞.
n=0 |z|≤r

This is sometimes called the ”root test” for series convergence, and is
related to the ”ratio test”. The reader should remember the Weierstrass
criterion, which we recall without proof:
Theorem 2.13 (Weierstrass). Let X be a metric space. Consider func-
tions fn : X P → C and bn ∈ R such that P |fn (x)| ≤ bn for all n ∈ N and
x ∈ X. If ∞ b
n=1 n < ∞, then the series ∞
n=1 fn (x) converges on X ab-
solutely and uniformly. In particular, if fn are continuous functions, the
series is also continuous.
Proof of Theorem 2.12. Without loss of generality we can assume
z0 = 0 throughout the proof.
We begin with the region of convergence. For R = 0 there is nothing to
prove, so assume R > 0 and fix 0 < r < R. Let us also fix some ρ such that
r < ρ < R. Then lim supn→∞ |an |1/n = R1 < ρ1 , so ∃N such that
1
|an | ≤ , ∀n ≥ N.
ρn
Thus, if we denote
bn = sup |an z n |,
|z|≤r
we have  n
r
bn ≤ , ∀n ≥ N.
ρ
Since 0 < r/ρ < 1, this implies that ∞
P
n=0 bn is convergent. Thus, by the
Weierstrass criterion, the power series converges absolutely and uniformly
on Dr (0).
On the other hand, if |z| > R, then
1 1
lim sup|an |1/n = >
n→∞ R |z|
so there exists a subsequence (nj )∞
j=1 such that |anj |
1/nj ≥ 1
|z| for all j and
therefore
|anj z nj | > 1.
26 2. DIFFERENTIAL CALCULUS AND POWER SERIES

In particular, among the terms of the power series, infinitely many of them
are larger than 1 in magnitude. Since the terms of the power series do not
converge to 0, by the divergence test, the series is divergent. □
Inside the disk of convergence DR (0), a power series defines a function
∞
X
f (z) = an z n
n=0

By uniform convergence, f is continuous on Dr (0) for any r < R, so it is

continuous on DR (0).
P∞ n
Example 2.14. The power series n=0 z has radius of convergence
R = 1. On D1 (0), this power series describes the function
∞
X 1
zn = .
1−z
n=0
Solution. Immediately from the definition, R = 1/ lim supn→∞ 1 = 1.
The value of the series for z ∈ D1 (0) follows from the famous finite sum
N −1
X 1 − zN
zn = (2.5)
1−z
n=0
(proved, e.g., by induction in N ), by letting N → ∞. □
Term-by-term differentiation. Our next goal is to prove that in the
disk of convergence, the power series can also be differentiated term-by-term.
For a polynomial, this would be trivial, but exchanging differentiation with
an infinite series requires a careful consideration of the error terms. We start
with some preliminaries.
Recall the binomial theorem: for n ∈ N, z, w ∈ C,
n    
n
X n j n−j n n!
(z + w) = w z , = .
j j j!(n − j)!
j=0
This is a well known identity proved, e.g., by induction in n.
Lemma 2.15. limn→∞ n1/n = 1.
Proof. Define δn = n1/n − 1. Since n > 1, δn > 0. For n ≥ 2, writing
a binomial sum with all terms positive and estimating the sum by a single
term,
n    
n
X n j n 2 n(n − 1) 2
n = (1 + δn ) = δn ≥ δn = δn .
j 2 2
j=0
2
This implies δn2 ≤ n−1 , and therefore,
r
2
0 ≤ δn ≤ .
n−1
By the squeeze theorem, limn→∞ δn = 0, so limn→∞ n1/n = 1. □
 2. POWER SERIES 27
P∞ n
P∞ n−1
Corollary 2.16. The power series n=0 an z and n=0 nan z have
the same radius of convergence.
Proof. Since
lim sup|nan |1/n = lim n1/n lim sup|an |1/n = lim sup|an |1/n
n→∞ n→∞ n→∞ n→∞
P∞ P∞
zn zn
the power series n=0 an and n=0 nan have the same radius of con-
vergence. Dividing the power series by z does not affect convergence, so it
doesn’t affect the radius, which concludes the proof. □
Lemma 2.17. For n ∈ N and z, w ∈ C, we have
n(n − 1) 2
|(z + w)n − z n − nwz n−1 | ≤ |w| (|z| + |w|)n−2 .
2
Proof. For j ≥ 2, use
     
n n(n − 1) n − 2 n(n − 1) n − 2
= ≤
j j(j − 1) j − 2 2 j−2
Using the binomial theorem, estimate
n  
X n j n−j
(z + w)n − z n − nwz n−1 = w z
j
j=2
n  
X n(n − 1) n − 2
≤ |w|j |z|n−j
2 j−2
j=2
n  
n(n − 1) 2 X n − 2
= |w| |w|j−2 |z|n−j
2 j−2
j=2
n(n − 1) 2
= |w| (|z| + |w|)n−2 . □
2
Theorem 2.18. Assume that
∞
X
f (z) = an (z − z0 )n
n=0
has radius of convergence R > 0. In DR (z0 ), the function f is holomorphic
and its derivative is given by a power series,
∞
X
′
f (z) = nan (z − z0 )n−1 .
n=1

Proof. We assume without loss of generality that z0 = 0.

As proved above, the candidate power series also has radius of conver-
gence R, so it defines a function
∞
X
g(z) = nan z n−1
n=1
 28 2. DIFFERENTIAL CALCULUS AND POWER SERIES

on DR (0).
Fix z ∈ DR (0) and r such that |z| + r < R. Let z ∈ DR (0). For |w| < r,
by the previous lemma,
n(n − 1) 2 n(n − 1) 2
|(z+w)n −z n −nwz n−1 | ≤ |w| (|z|+|w|)n−2 ≤ |w| (|z|+r)n−2 .
2 2
Multiplying by |an |, summing in n, and using a triangle inequality,
∞
X n(n − 1)
|f (z + w) − f (z) − wg(z)| ≤ |w|2 |an |(|z| + r)n−2 .
2
n=0
P∞ n(n−1)
The series n=0 2 an xn−2 has radius of convergence R, by a double
application of Corollary 2.16. Thus, we can define a finite constant C by
∞
X n(n − 1)
C= |an |(|z| + r)n−2 < ∞.
2
n=0
We have proved
|f (z + w) − f (z) − wg(z)| ≤ C|w|2
for all small enough w. Thus,
f (z + w) − f (z) − wg(z) = o(|w|), w → 0,
which proves that f is holomorphic at z and f ′ (z) = g(z). □

3. Exponential and trigonometric functions

This can also be
proved by an appli- Lemma 2.19. If f : Dr (z0 ) → C is holomorphic and f ′ = 0, then f is
cation of the fun- constant.
damental theorem Proof. Without loss of generality let z0 = 0. Fix w ∈ Dr (0) and
of calculus which consider the function g : D → C given by
we will cover soon
g(z) = f (wz).
By the chain rule, g ′ (z)
= f ′ (wz)w
= 0 for all z.
Restricting to real parameters, we have g : [−1, 1] → C with zero de-
rivative, so g is constant on [−1, 1] by the mean value theorem from real
analysis. In particular, g(1) = g(0), so f (w) = f (0). □
Theorem 2.20. The power series
∞
X zn
exp(z) =
n!
n=0
has radius of convergence R = ∞ and defines the function exp : C → C with
the properties:
(a) exp(0) = 1
(b) exp′ (z) = exp(z)
(c) exp(z) ̸= 0 for all z ∈ C
(d) exp(z + w) = exp(z) exp(w)
 3. EXPONENTIAL AND TRIGONOMETRIC FUNCTIONS 29

(e) exp(z) = exp(z)

Proof. Among the numbers 1, . . . , n, at least n2 of them are of size at
least n2 . Thus, n! ≥ (n/2)n/2 , so
 1/n  1/n
1 1/2
 
1 1
lim sup ≤ lim sup ≤ lim sup =0
n→∞ n! n→∞ n! n→∞ n/2
so R = ∞ and this defines a function exp : C → C.
(a) follows from the point evaluation at z = 0.
(b) follows from term-by-term differentiation of power series, since
∞ ∞ ∞
X z n−1 X z n−1 X zk
exp′ (z) = n = = = exp(z).
n! (n − 1)! k!
n=0 n=1 k=0
Now fix w ∈ C and consider the holomorphic function
g(z) = exp(z + w) exp(−z).
Differentiating by the product rule gives
g ′ (z) = exp(z + w) exp(−z) − exp(z + w) exp(−z) = 0
so g is constant. Since g(0) = exp(w) exp(0) = exp(w), we conclude
exp(z + w) exp(−z) = exp(w), ∀z, w ∈ C.
Plugging in w = 0 shows exp(z) exp(−z) = 1 which proves (c) and allows us
to rewrite the previous identity as exp(z + w) = exp(z) exp(w), proving (d).
(e) Since exp is defined by a Taylor series with real coefficients,
∞ ∞
X zn X zn
exp(z) = = = exp(z). (2.6)
n! n!
n=0 n=0
□
In particular, z = z implies exp(z) = exp(z). In other words, z ∈ R
implies exp(z) ∈ R. Let us analyze the behavior of exp on R:
Theorem 2.21. For all x ∈ R, exp(x) > 0, and exp : R → (0, ∞) is a
strictly increasing bijection.
Proof. For x ≥ 0, the power series for exp(x) consists of positive terms
so obviously exp(x) > 0. For x < 0, it follows from exp(−x) exp(x) = 1 that
exp(x) = 1/ exp(−x) > 0.
Since it has a positive derivative on R, exp is strictly increasing on R.
For x ≥ 0, exp(x) ≥ 1 + x implies
lim exp(x) = +∞.
x→+∞

Now exp(−x) exp(x) = 1 implies

1
lim exp(x) = lim = 0.
x→−∞ x→−∞ exp(−x)
30 2. DIFFERENTIAL CALCULUS AND POWER SERIES

A function with these properties is a bijection by the intermediate value

theorem from real analysis. □
It is standard to define e = exp(1) and to write
ez = exp(z),
and this matches with other definitions of the exponential function.
Since ∞ n /n! = e−1 is an alternating series, its value obeys pre-
P
n=0 (−1)
cise inequalities in terms of partial sums Sk = kn=0 (−1)n /n! as
P

S2k+1 < e−1 < S2k , ∀k ∈ N.

Already from S3 < e−1 < S2 we conclude
1 1 1 1
1−1+ − < <1−1+ ,
2 6 e 2
and therefore
2 < e < 3.
Using larger values of k or other techniques, one concludes
e = 2.71828 . . .
The same series can be used to prove:
Theorem 2.22. The number e is irrational.
Proof. Assume e is rational. Then (2n − 1)!e−1 ∈ Z for some n ∈ N.
By the alternating series above, we have
1
0 < e−1 − S2n−1 <
(2n)!
so
1
0 < (2n − 1)!e−1 − (2n − 1)!S2n−1 < < 1.
2n
We note that (2n−1)!S2n−1 ∈ Z, and that a difference of two integers cannot
be in the interval (0, 1), leading to a contradiction. □
We now turn to trigonometric functions.
Definition 2.23. We define functions cos, sin : C → C by
eiz + e−iz eiz − e−iz
cos(z) = , sin(z) = , (2.7)
2 2i
From the power series representation for exp(z), we obtain
∞ ∞
X (−1)k 2k X (−1)k 2k+1
cos z = z , sin z = z . (2.8)
(2k)! (2k + 1)!
k=0 k=0
From these power series we read off that cos 0 = 1 and sin 0 = 0 and
sin′ = cos, cos′ = − sin .
Further properties of trigonometric functions, such as addition formulas for
sin, cos, can be derived from ez+w = ez ew (Exercise 5).
 3. EXPONENTIAL AND TRIGONOMETRIC FUNCTIONS 31

Directly from the definitions (2.7),

cos z + i sin z = eiz .
Although this holds for all z ∈ C, it is usually used for real z as part of
Euler’s formula,
ex+iy = ex eiy = ex (cos y + i sin y).
Lemma 2.24. For all t ∈ R, |eit | = 1 and
Re eit = cos t, Im eit = sin t. (2.9)
Proof. Using (2.6), we obtain
|eit |2 = eit eit = eit e−it = e0 = 1.
Likewise, the power series (2.8) have real coefficients, so cos t, sin t ∈ R for
t ∈ R. From this and Euler’s formula, (2.9) follows. □
Theorem 2.25. There exists a number π ∈ (0, 4) with the properties:
(a) the map t 7→ eit is a bijection from [0, 2π) to {z ∈ C | |z| = 1}
(b) the map t 7→ eit is a bijection from [0, π/2) to {z ∈ C | |z| = 1, Re z >
0, Im z ≥ 0}
(c) eiπ/2 = i
Proof. Using the power series representation for cos 2 and grouping
together terms k = 2j − 1 and k = 2j,
∞ 
24j−2 24j
X 
cos 2 = 1 + − +
(4j − 2)! (4j)!
j=1

For any j ∈ N,
24j 24j−2
<
(4j)! (4j − 2)!
so using that to estimate terms with j ≥ 2 gives
1 
24j−2 24j 22 24

X 1
cos 2 < 1 + − + =1− + =− .
(4j − 2)! (4j)! 2! 4! 3
j=1

Since cos 0 = 1, by continuity and the intermediate value theorem, cos has a
zero in (0, 2). Thus, the set {t ∈ [0, ∞) | cos t = 0} is nonempty; moreover,
it is closed since cos is continuous. Thus, we can define
π = 2 min{t ∈ [0, ∞) | cos t = 0}.
Since cos 0 > 0, it follows that π > 0 and cos t > 0 on t ∈ (0, π/2). By
cos 2 < 0 we know π/2 < 2.
Now sin′ = cos implies that sin is strictly increasing on [0, π/2]. More-
over, sin 0 = 0, so sin is strictly positive on (0, π/2]. By sin2 t + cos2 t = 1 it
follows that sin2 π2 = 1 so sin π2 = 1. In particular,
π π
eiπ/2 = cos + i sin = 0 + i = i.
2 2
32 2. DIFFERENTIAL CALCULUS AND POWER SERIES

Note that the arc

A = {z ∈ C | |z| = 1, Re z > 0, Im z ≥ 0}
is parametrized by Im z = y ∈ [0, 1) as
p
A = { 1 − y 2 + iy | y ∈ [0, 1)}.
p
Since sin : [0, π2 ) → [0, 1) is a bijection and cos t = 1 − sin2 t on this
it π
interval, it follows that the map t 7→ e is a bijection [0, 2 ) → A.
By separating cases based on the signs of Re z, Im z, we observe that
for every z with |z| = 1, exactly one of the four numbers z, iz, −z, −iz is
an element of A. Thus, by applying the previous result to z, iz, −z, −iz, it
follows that t 7→ eit is a bijection of [0, 2π) to {z ∈ C | |z| = 1}. □
Remark 2.26. If we have at our disposal some integral calculus, we can
provide a different proof of existence of a positive zero of cos.
Assume L is such that cos has no zeros on [0, L]. Then, as above, sin
and cos are strictly positive on (0, L). The curve γ(t) = eit obeys |γ ′ (t)| = 1
and Z LZ Z L L
L= |γ ′ (t)| dt ≤ |Re γ ′ (t)| dt + |Im γ ′ (t)| dt.
0 0 0
Since 0 ≤ Re γ ≤ 1 and 0 ≤ Im γ ≤ 1 and their derivatives have constant
sign, we obtain by the fundamental theorem of calculus that
Z L Z L
|Re γ ′ (t)| dt = − Re γ ′ (t) dt = Re γ(0) − Re γ(L) ≤ 1,
0 0
Z L Z L
|Im γ ′ (t)| dt = Im γ ′ (t) dt = Im γ(1) − Im γ(L) ≤ 1,
0 0
so L ≤ 2. Thus, cos has a zero on [0, 2]. Now everything proceeds as before.
It is clear that exp : C → C is neither onto nor 1-1. The amount to
which these properties “fail” is described precisely here:
Theorem 2.27. The function exp takes all nonzero values in C. More-
over, for z1 , z2 ∈ C, ez1 = ez2 ⇐⇒ z1 − z2 ∈ 2πiZ.
Proof. Let w ̸= 0. Let us solve the equation ez = w. Denote z = x+iy.
By Euler’s formula, since ex > 0 and |eiy | = 1,
|ez | = |ex eiy | = ex
so ex = |w|. Since |w| > 0 and x 7→ ex is a bijection from R to (0, ∞), this
determines x ∈ R uniquely.
Moreover,
w ez
= x = eiy . (2.10)
|w| e
w
Since |w| is a point on the unit circle, by the previous theorem, there is a
unique y ∈ [0, 2π) which obeys (2.10). Moreover, by e2πi = 1, y + 2πk has
the same property for k ∈ Z and there are no other solutions. □
 3. EXPONENTIAL AND TRIGONOMETRIC FUNCTIONS 33

We sometimes denote
(
w/|w| ̸ 0
w=
sgnw =
0 w=0

Polar form of complex numbers.

Definition 2.28. The polar form of w ∈ C \ {0} is a representation
w = reiθ
where r > 0, θ ∈ R. Here r = |w| is uniquely determined, and θ is uniquely
determined up to +2πZ. Any such θ is called an argument of w.
If z1 = r1 eiϕ1 and z2 = r2 eiϕ2 , then z1 z2 = r1 r2 ei(ϕ1 +ϕ2 ) . This provides
a geometric interpretation of complex multiplication: if w ̸= 0 has the polar
form w = reiθ , the map
z 7→ wz
corresponds to a scaling of the complex plane by a factor of r > 0 and a
rotation of the complex plane by angle θ.
Some values of the exponential function are useful, and can now be com-
puted without appealing to geometric intuition or trigonometry, for instance:
√
1 i 3
Example 2.29. eiπ/3 = 2 + 2

Proof. Denote z = eiπ/3 . Then z 3 = eiπ = −1. The equation z 3 +1 = 0

factorizes as (z√+ 1)(z 2 − z + 1) = 0. Solving the quadratic equation, we find
solutions 12 ± i 2 3 . Since 0 < π3 < π2 , the number z is in the first quadrant,
i.e., Re z > 0 and Im z > 0. Among all solutions of the √ cubic equation
z 3 + 1 = 0, the only solution in the first quadrant is 12 + i 2 3 . □

Example 2.30. eiπ/4 = √1 + √i

2 2

Proof. Denote z = x + iy = eiπ/4 . Since z 2 = eiπ/2 = i, we obtain in

2 2 2 2
√ x − y = 0 and 2xy = i. This shows x = y and
Cartesian coordinates
then x = y√= ±1/ 2. Since 0 ≤ π/4 < π/2, we know that x, y ≥ 0, so
x = y = 1/ 2. □
Alternative proof. Denote z = eiπ/4 . Then z 4 = eiπ = −1. The
equation z 4 + 1 = 0 factorizes as
√ √
z 4 + 1 = (z 2 + 1)2 − 2z 2 = (z 2 − 2z + 1)(z 2 + 2z + 1)
so by solving the quadratic equations we find four solutions ± √12 ± √i .
2
Since 0 < π4 < π2 , the number eiπ/4 is in the first quadrant and we conclude
eiπ/4 = √12 + √i2 . □

Taking the complex conjugate of these values and multiplying by powers

of eπi/2 = i, we obtain all the values in Figure 2.1. See also Exercise 6.
34 2. DIFFERENTIAL CALCULUS AND POWER SERIES

n-th roots of complex numbers.

Theorem 2.31. For any n ∈ N and w ∈ C \ {0}, the equation
zn = w
has n distinct roots. Denoting by θ an argument of w, the solutions are of
the form
z = |w|1/n ei(θ+2kπ)/n , k = 0, 1, . . . , n − 1.
Proof. Writing z = reiϕ , we see that
rn einϕ = ρeiθ
which implies
r = ρ1/n , nϕ ∈ θ + 2πZ.
This gives all solutions in the form
z = ρ1/n ei(θ+2kπ)/n
′
with k ∈ Z. Since ei(θ+2kπ)/n = ei(θ+2k π)/n if and only if k − k ′ is divisible
by n, there are n distinct solutions, obtained for k = 0, 1, . . . , n − 1. □
Definition 2.32. For n ∈ N, the n-th roots of unity are all the solutions
of ω n = 1.
The n-th roots of unity can be listed as
2kπi
ωk = e n , k = 0, 1, . . . , n − 1.
It is sometimes convenient to write ω = ω1 and ωk = ω k .

√ eiπ/2 =i √
e2iπ/3 =− 12 + i 2 3 eiπ/3 = 12 + i 2 3

e3iπ/4 =− √1 + √i eiπ/4 = √1 + √i
2 2 2 2
√ √
e5iπ/6 =− 2 + 2i
3
eiπ/6 = 2
3
+ 2i

eπi =−1 e0i =1

√ √
3
e7iπ/6 =− 2
− 2i e11iπ/6 = 2
3
− 2i

e5iπ/4 =− √1 − √i e7iπ/4 = √1 − √i
2 2 2 2

√ √
e4iπ/3 =− 12 − i 2 3 e5iπ/3 = 12 − i 2 3
e3iπ/2 =−i

Figure 2.1. Some known values of eit

 4. EXERCISES 35

Just as the map z 7→ eiϕ z encodes rotation by angle ϕ around the origin,
the map
z − z0 7→ eiϕ (z − z0 )
encodes rotation by angle ϕ around z0 .

4. Exercises
1. Prove that the function f (z) = z z̄ is not holomorphic, except at z = 0.
Comment: the fact that this function is holomorphic at one point is
not significant; in complex analysis, we always study functions holomor-
phic on some open region.
2. This exercise uses the view of a function f of a complex variable z = x+iy
as a function of two real variables x, y. Assume that f is differentiable in
the sense of real variable multivariable calculus, denote its partial deriva-
tives by ∂x f , ∂y f , and define
1 1
∂z f =(∂x − i∂y ) f, ∂z f = (∂x + i∂y ) f
2 2
(a) Prove that f is holomorphic if and only if ∂z f = 0.
(b) If f is holomorphic, prove that ∂z f = f ′ .
(c) Prove that ∂z obeys the product rule
∂z (f g) = (∂z f )g + f (∂z g)
(d) For k, l = 0, 1, 2, . . . , prove that ∂z (z k z l ) = lz k z l−1 .
3. If f : Ω → C is holomorphic, prove that the function
g(z) = f (1/z)
is holomorphic on {z ∈ C | 1/z ∈ Ω}.
4. (a) Find a power series (centered around 0) which converges to
1
(1 − z)3
on some disk centered at 0, and determine the radius of convergence
of this power series.
(b) Compute the value of the series
∞
X n(n − 1)
.
2n
n=2

5. Verify directly using (2.7) the addition formulas for sin and cos:
(a) sin(z + w) = sin z cos w + cos z sin w
(b) cos(z + w) = cos z cos w − sin z sin w
6. Prove without
√ appeal to √ geometric intuition or trigonometry that:
√ √
(a) eiπ/8 = 2+ 2
2
+ i 2−
2
2

Hint: find real and imaginary parts of z = eiπ/8 by using z 2 = eiπ/4

and |z|2 = 1.
 36 2. DIFFERENTIAL CALCULUS AND POWER SERIES

√ √ √
(b) ei2π/5 = 5−1+i 4 10+2 5
Hint: prove that z = ei2π/5 satisfies z 2 + z + 1 + z −1 + z −2 = 0. Use
the substitution z + z −1 = 2x and express z 2 + z −2 in terms of x to
find a quadratic
√ equation for x.
7. (a) Compute (− 3 + i) . 100
√
(b) Find the smallest n ∈ N = {1, 2, 3, . . . } such that (− 3+i)n ∈ (0, ∞).
8. Find all solutions of the equation z 6 + z 3 + 1 = 0 in polar form.
Hint: expand by z 3 − 1 and use the roots of unity.
9. Find all solutions of the equation z 3 = 3z − 1 as cosines of rational
multiples of π.
Hint: use Tartaglia’s formula.
10. As functions on C, prove that sin z = 0 if and only if z ∈ πZ, and that
cos z = 0 if and only if z ∈ π2 + πZ.
11. From the exponential function, we can also define the hyperbolic cosine
and hyperbolic sine as functions cosh, sinh : C → C given by
ez + e−z ez − e−z
cosh z = , sinh z = .
2 2
For any z = x + iy, prove that
|cos(x + iy)|2 = cos2 x + sinh2 y.
12. From the exponential function, we can also define the hyperbolic cosine
and hyperbolic sine as functions cosh, sinh : C → C given by
ez + e−z ez − e−z
cosh z = , sinh z = .
2 2
For any z = x + iy, prove that
|sinh(x + iy)|2 = sinh2 x + sin2 y.
13. For z ∈ C \ πZ, define
sinh z ez − e−z
tanh z = = z .
cosh z e + e−z
For z, w, z + w ∈ C \ πZ, prove that
tanh z + tanh w
tanh(z + w) = .
1 + tanh z tanh w
14. Compute the sum
n
X
cos kθ,
k=0
giving your final answer in terms of trigonometric functions.
Hint: write it as the real part of a geometric sum. This should give
??
you a result of a form such as Re 1−eiθ . Manipulate the fraction to obtain
a real denominator and then rewrite everything in terms of trigonometric
functions.
1 1 1
15. Let ABCDEF G be a regular 7-gon. Prove that AC + AD = AB .
 4. EXERCISES 37

16. In this problem, we consider the differential equation

y ′′ (z) = zy(z).
(a) Assuming that there is a solution of the form y(z) = ∞ n
P
n=0 an z , by a
formal differentiation, find a relation between the terms of the series
in the form of a recurrence relation.
(b) Prove that this recurrence relation has a unique solution for any
choice of a0 = α, a1 = β.
(c) Prove that for any α, β, the resulting power series has infinite ra-
dius of convergence. Conclude that this power series really defines a
solution of the differential equation.
Pn
17. If zj ∈ C for j = 1, . . . , n and j=1 |zj | = 1, prove that there exists
S ⊂ {1, . . . , n} such that
X 1
zj ≥ .
6
j∈S

Hint: divide C into three sectors Pwith internal angles 2π/3 each.
n
18. If zj ∈ C for j = 1, . . . , n and j=1 |zj | = 1, prove that there exists
S ⊂ {1, . . . , n} such that
X 1
zj ≥ .
π
j∈S
Pn
Hint: consider the sum j=1 (Re(e−iϕ zj ))+ as a function of ϕ ∈ R and
compute its average over ϕ ∈ [0, 2π). Use ϕ which maximizes the sum.
 CHAPTER 3

Contour integration

A curve in Ω is a continuous function γ : [a, b] → C.

Curves appear in complex analysis both for topological and analytical
reasons. Curves are used to probe the topology of a domain; for this pur-
pose, arbitrary (continuous) curves are used. For analysis, integration along
curves requires, at the very least, that the curve has finite length; such curves
are called rectifiable. A special case of rectifiable curves are piecewise C 1
curves; this special case can be studied using the Riemann integral, and it
is sufficient for the development of complex analysis. We will begin by de-
scribing integration on piecewise C 1 curves. We will also discuss integration
on rectifiable curves as a bonus topic.

1. Fundamental theorem of calculus for C 1 curves

Let us now focus on the following class of curves:
Definition 3.1. The curve γ : [a, b] → C is a C 1 curve if it is differen-
tiable at every point x ∈ [a, b] (in particular, it has a right derivative at a
and a left derivative at b) and its derivative γ ′ is continuous on [a, b].
A counterexample
p is a circular arc, parametrized by imaginary part. The
2
map γ(y) = 1 − y + iy maps [0, 1] bijectively to the arc {z ∈ C | |z| =
1, Re z ≥ 0, Im z ≥ 0}. It is continuously differentiable on [0, 1), with
y
γ ′ (y) = − p + i,
1 − y2
p
but it is not a C 1 curve since limy↑1 |γ ′ (y)| = limy↑1 1/ 1 − y 2 = ∞.
We note that γ is C 1 if and only if Re γ and Im γ are C 1 , and in this
case,
γ ′ = (Re γ)′ + i(Im γ)′ .
This will allow us to transfer to curves some facts from single variable cal-
culus, such as the fundamental theorem of calculus:
Lemma 3.2 (Fundamental theorem of calculus for complex-valued func-
tions). If γ : [a, b] → C is a C 1 curve, then
Z b
γ(b) − γ(a) = γ ′ (x) dx.
a
39
40 3. CONTOUR INTEGRATION

Proof. If γ is real-valued, this is the fundamental theorem of calculus.

In general, applying the fundamental theorem of calculus to Re γ and Im γ
and using γ = Re γ + i Im γ completes the proof. □
Corollary 3.3. If f : DR (z0 ) → C is holomorphic and f ′ = 0, then f
is constant.
Proof. Fix z ∈ DR (z0 ) and observe the curve
γ(t) = f (z0 + t(z1 − z0 )), t ∈ [0, 1].
By the chain rule, γ is differentiable and
γ ′ (t) = f ′ (z0 + t(z1 − z0 ))(z1 − z0 ) = 0.
Thus, γ is a C 1 curve and, by the fundamental theorem of calculus, γ(1) =
γ(0). In other words, f (z) = f (z0 ). □
We also note a basic upper bound for Riemann integrals:
Lemma 3.4 (Upper bound for a Riemann integral). If h : [a, b] → C is
Riemann-integrable, then
Z b Z b
h(x) dx ≤ |h(x)| dx.
a a
Rb
Proof. Denote C = a h(x) dx. If C = 0, set arbitrary ω ∈ ∂D;
otherwise, use ω = C/|C| ∈ ∂D. Then |ω| = 1 and Cω ≥ 0, thus,
|C| = |Cω| = Re(Cω). In other words,
Z b  Z b 
h(x) dx = Re ω h(x) dx .
a a
Using this, we estimate
 Z b  Z b Z b
Re ω h(x) dx = Re (ωh(x)) dx ≤ |h(x)| dx. □
a a a

This has an immediate geometric application to C 1 curves:

Corollary 3.5. For any C 1 curve γ : [a, b] → C,
Z b
|γ(b) − γ(a)| ≤ |γ ′ (x)| dx.
a
As an illustration, we prove:
p √
Example 3.6. The constant π satisfies π ≥ 6 2 − 3 = 3.1058 . . . .
Proof. The arc of the unit circle γ(t) = eit , t ∈ [0, π/6], is C 1 and
satisfies |γ ′ (t)| = |ieit | = 1, so
Z π/6 Z π/6
iπ/6 i0 it π
|e −e |= ie dt ≤ ieit dt = .
0 0 6
 2. RECTIFIABLE AND PIECEWISE C 1 CURVES 41
√
Using the computed value eiπ/6 = 3/2 + i/2 we obtain
√
√
q
π 3 i
≥ + − 1 = 2 − 3.
6 2 2
p √
Thus, π ≥ 6 2 − 3 = 3.1058 . . . . Note that even without the square root
calculations, since the imaginary part of eiπ/6 − ei0 is 1/2, we would obtain
π/6 > 1/2 so π > 3. □
We have previously proved π < 4. We will later mention more precise
methods for computing π.
Finally, we warn the reader that there is no mean value theorem for
curves:
Example 3.7. The curve γ(t) = eit obeys γ(0) = γ(2π), but γ ′ (t) ̸= 0
for all t.

2. Rectifiable and piecewise C 1 curves

In this section, we study the length of curves, and prove that piecewise
C 1 curves have finite length. Along the way, we review some facts about
Riemann integration of complex-valued functions.
Definition 3.8. A partition of the interval [a, b] is a a finite sequence
t = (tj )nj=0 with a = t0 < t1 < · · · < tn = b. We denote n = n(t) and denote
by P[a,b] the set of all partitions of [a, b].
For any partition t, by connecting consecutive points in the sequence
γ(t0 ), . . . , γ(tn ), we obtain a polygonal curve which approximates γ. The
length of a curve is defined as the supremum of lengths of these approxi-
mants:
Definition 3.9. The length of the curve γ : [a, b] → C is
n(t)
X
ℓ(γ) = sup |γ(tj ) − γ(tj−1 )|.
t∈P[a,b] j=1

The curve is said to be rectifiable if ℓ(γ) < ∞.

Lemma 3.10. For any curve γ : [a, b] → C and any x ∈ (a, b),
ℓ(γ|[a,x] ) + ℓ(γ|[x,b] ) = ℓ(γ).
In particular, γ is rectifiable if and only if both γ|[a,x] and γ|[x,b] are rectifi-
able.
Proof. Combining arbitrary partitions of [a, x] and [x, b] gives any par-
tition of [a, b] containing x. Thus,
n(t)
X
ℓ(γ|[a,x] ) + ℓ(γ|[x,b] ) = sup |γ(tj ) − γ(tj−1 )|.
t∈P[a,b] j=1
∃j tj =x
42 3. CONTOUR INTEGRATION

By the triangle inequality, the sum over the partition t can only increase
when we add the point x to the partition, so the right-hand side is equal to
ℓ(γ). □
Theorem 3.11. Any C 1 curve γ : [a, b] → C is rectifiable, with length
Z b
ℓ(γ) = |γ ′ (x)|dx.
a
Proof. For any partition t and j, by the fundamental theorem of cal-
culus and upper bound for Riemann integrals,
Z tj
γ(tj ) − γ(tj−1 ) = γ ′ (x) dx
tj−1
so Z tj
|γ(tj ) − γ(tj−1 )| ≤ |γ ′ (x)| dx.
tj−1
Summing over j and taking the supremum over t shows that γ is rectifiable
and Z b
ℓ(γ) ≤ |γ ′ (x)| dx. (3.1)
a
To prove the opposite inequality, fix ϵ > 0. By uniform continuity of γ ′ on
[a, b], there exists δ > 0 such that |v − u| ≤ δ implies |γ ′ (v) − γ ′ (u)| ≤ ϵ.
Thus, if u < v and v − u ≤ δ, by the fundamental theorem of calculus,
Z v Z v
′ ′
γ (x) dx − (v − u)γ (u) = (γ ′ (x) − γ ′ (u)) dx ≤ (v − u)ϵ
u u
so by the triangle inequality
Z v
γ ′ (x) dx ≥ (v − u)|γ ′ (u)| − (v − u)ϵ.
u

Applying this to the equispaced partition tj,n = a + (b − a) nj for n ∈ N large

enough that (b − a)/n < δ, and summing in j = 1, . . . , n, we obtain
n
X n
X
ℓ(γ) ≥ |γ(tj,n ) − γ(tj−1,n )| ≥ (tj,n − tj−1,n )|γ ′ (tj−1,n )| − (b − a)ϵ.
j=1 j=1
Rb ′ (x)| dx,
The right-hand side contains a Riemann sum for the integral a |γ
so in the limit n → ∞, we obtain
Z b
ℓ(γ) ≥ |γ ′ (x)| dx − (b − a)ϵ.
a
Since ϵ > 0 is arbitrary, we conclude
Z b
ℓ(γ) ≥ |γ ′ (x)| dx,
a
which together with (3.1) concludes the proof. □
 3. CONTOUR INTEGRALS ON PIECEWISE C 1 CURVES 43

Since it is useful to work with polygonal contours, we generalize this to

the following class.
Definition 3.12. The curve γ : [a, b] → C is a piecewise C 1 curve
if there exists a partition s of the interval [a, b] such that the restriction
γ|[sj−1 ,sj ] is C 1 for each j.
Theorem 3.13. Any piecewise C 1 curve γ : [a, b] → C is rectifiable, with
length
Z b
ℓ(γ) = |γ ′ (x)|dx.
a
Proof. By Lemma 3.10 and additivity of integrals, this follows from
the special case when γ is C 1 . □
We mention that lack of differentiability at one point can be enough to
destroy rectifiability: the curve
(
t + it sin(π/t) t ∈ (0, 1]
γ(t) =
0 t=0
is continuously differentiable on (0, 1] and continuous on [0, 1], but it is not
rectifiable (see Exercise 1).
Let us revisit pthe example of an arc, parametrized by imaginary part.
The map γ(y) = 1 − y 2 + iy maps p [0, 1] bijectively to the arc {z ∈ C |
|z| = 1, Re z ≥ 0, Im z ≥ 0}. Since 1 − y 2 is monotone decreasing and y is
monotone increasing, for any partition of [0, 1], |γ(tj )−γ(t
P j−1 )| ≤ Im(γ(tj )−
γ(tj−1 ))−Re(γ(tj )−γ(tj−1 )) and summing in j gives nj=1 |γ(tj )−γ(tj−1 )| ≤
Im(γ(1) − γ(0)) − Re(γ(1) − γ(0)) = 2. Thus, this curve is rectifiable, with
ℓ(γ) ≤ 2. This can be used as a more foundational example to help √ study π
and the exponential function. Notice also ℓ(γ) ≥ |γ(1) − γ(0)| = 2.

3. Contour integrals on piecewise C 1 curves

Our next goal is to define integration along a curve γ of a function f
which is continuous on Ran γ, denoted
I
f (z) dz.
γ
This contour integral plays a central role in complex analysis. Here dz is
intuitively thought of as an infinitesimal displacement along the curve; it is
an infinitesimal complex quantity. Complex multiplication of f (z) and dz
will be built into this definition.
Occasionally, we will want to use integrals over the length element |dz|,
which will be denoted I
f (z) |dz|.
γ
To integrate on piecewise C1 curves, we can use the Riemann integral.
44 3. CONTOUR INTEGRATION

Definition 3.14. Let γ be a piecewise C 1 curve, and f a continuous

function on Ran γ. Then we define
I Z b
f (z) dz = f (γ(t))γ ′ (t) dt
γ a
I Z b
f (z) |dz| = f (γ(t))|γ ′ (t)| dt
γ a

To see that this is well-defined, note that γ ′ exists except at the points
tj and f (γ(t))γ ′ (t) has left and right limits at the points tj and is continuous
elsewhere. For this reason f (γ(t))γ ′ (t) is Riemann-integrable on the intervals
[tj−1 , tj ] and then on [0, 1].
Example 3.15. By the fundamental theorem of calculus for curves and
the length formula,
I
1 dz = γ(b) − γ(a),
γ
I
1 |dz| = ℓ(γ).
γ

Example 3.16. For the positively oriented unit circle γ(t) = eit , t ∈
[0, 2π],
I Z 2π ′ Z 2π
dz γ (t)
= dt = i dt = 2πi.
γ z 0 γ(t) 0

Algebraic properties of contour integrals. As a direct consequence

of linearity of Riemann/Lebesgue integration:
Lemma 3.17. Contour integrals are linear functionals, i.e.,
I I I
(αf + βg)(z) dz = α f (z)dz + β g(z)dz
γ γ γ
for any α, β ∈ C and continuous functions f, g : Ω → C.
Lemma 3.18 (Contour additivity). If γ : [a, b] → C is piecewise C 1 ,
x ∈ (a, b), and γ1 = γ|[a,x] , γ2 = γ|[x,b] , then
I I I
f (z) dz = f (z) dz + f (z) dz.
γ γ2 γ1

Substitution rule for contour integrals. Composing γ with a suit-

able map f gives a new curve f ◦ γ, and there is a substitution rule relating
contour integrals over the two curves:
Theorem 3.19 (Substitution rule for contour integrals). Let f : Ω → C
be holomorphic, with f ′ continuous. Let γ : [a, b] → Ω be a piecewise C 1
curve in Ω. Let h be a continuous function on the range of f ◦ γ. Then f ◦ γ
is a piecewise C 1 curve and
(f ◦ γ)′ (t) = f ′ (γ(t))γ ′ (t) (3.2)
 3. CONTOUR INTEGRALS ON PIECEWISE C 1 CURVES 45

for all but finitely many values of t. Moreover, if h is continuous on the

range of f ◦ γ, then
I I
h(w) dw = h(f (z))f ′ (z) dz (3.3)
f ◦γ γ
I I
h(w) |dw| = h(f (z))|f ′ (z)| |dz|. (3.4)
f ◦γ γ

Proof. Using the chain rule, we obtain (3.2). Then we notice that both
sides of (3.3) are equal to the Riemann integral
Z
h(f (γ(t)))f ′ (γ(t))γ ′ (t) dt

and both sides of (3.4) are equal to

Z
h(f (γ(t)))|f ′ (γ(t))γ ′ (t)| dt. □

Combining this with the fundamental theorem of calculus for curves, we

obtain a fundamental theorem of calculus for contour integrals:
Theorem 3.20 (Fundamental theorem of calculus for contour integrals).
Let γ : [a, b] → Ω be a piecewise C 1 curve in Ω. Let f be holomorphic in Ω
and f ′ continuous.
I
f (γ(b)) − f (γ(a)) = f ′ (z) dz
γ
I
ℓ(f ◦ γ) = |f ′ (z)| |dz|.
γ

Proof. We obtain the identities by applying Example 3.15 to the curve

f ◦ γ and combining this with the formulas (3.3), (3.4) applied to h = 1. □
Example 3.21. For the positively oriented unit circle γ(t) = eit , t ∈
[0, 2π], and any n ∈ Z \ {−1},
I I  n+1 ′
n z
z dz = dz = 0.
γ γ n+1
Inequalities for contour integrals. Next, we turn to estimates (up-
per bounds) for contour integrals. We denote for a set S and a function f
on S,
∥f ∥S = sup|f (z)|.
z∈S
If S is compact and f is continuous, the supremum is guaranteed to be finite.
When needed, we will also use the notation γ for the set of values of γ;
for instance, ∥f ∥γ will be the supremum of |f (z)| over z = γ(t), t ∈ [a, b].
Lemma 3.22. Contour integrals obey the inequalities
I I
f (z)dz ≤ |f (z)| |dz| ≤ ℓ(γ)∥f ∥γ .
γ γ
46 3. CONTOUR INTEGRATION

Proof. In terms of Riemann integrals, this is simply

Z Z Z
′
f (γ(t))γ (t) dt ≤ f (γ(t))γ (t) dt ≤ ∥f ∥γ |γ ′ (t)| dt.
′
□

Corollary 3.23. If functions fn converge to f uniformly on Ran γ,

then I I
lim fn (z) dz = f (z) dz.
n→∞ γ γ
Proof. Uniform convergence on γ means that ∥fn −f ∥γ → 0 as n → ∞.
Thus, the claim follows from the estimate
I I
fn (z) dz − f (z) dz ≤ ℓ(γ)∥fn − f ∥γ . □
γ γ

4. Antiderivatives from contour integrals

We consider the line segment [z1 z2 ] as the C 1 curve parametrized by
t ∈ [0, 1] by z1 + t(z2 − z1 ).
Lemma 3.24. Assume f : Dr (z1 ) → C is continuous and for every z2 ∈
Dr (z1 ), I
g(z2 ) = g(z1 ) + f (ζ) dζ.
[z1 z2 ]
Then g is holomorphic at z1 and g ′ (z1 ) = f (z1 ).
Proof. For any ϵ > 0, there exists δ > 0 such that |z − z1 | < δ implies
|f (z) − f (z1 )| < ϵ. Then for |z2 − z1 | < δ we have
I
|g(z2 ) − g(z1 ) − (z − z1 )f (z1 )| = (f (ζ) − f (z1 )) dζ ≤ ϵℓ([z1 z2 ])
[z1 z2 ]
so
|g(z2 ) − g(z1 ) − (z − z1 )f (z1 )| ≤ ϵℓ([z1 z2 ])
so
g(z2 ) − g(z1 )
− f (z1 ) ≤ ϵ.
z2 − z1
Since ϵ > 0 is arbitrary, it follows that g is holomorphic at z1 and g ′ (z1 ) =
f (z1 ). □
Corollary 3.25 (Term-by-term integration of power series). Assume
the function f is defined on DR (z0 ) by a power series
X∞
f (z) = an (z − z0 )n
n=0
with radius of convergence R > 0. Then the function
∞
X an
g(z) = (z − z0 )n+1 (3.5)
n+1
n=0
is holomorphic on DR (z0 ) and g′ = f.
 5. REPARAMETRIZATION AND ARCLENGTH 47

Proof. Without loss of generality let z0 = 0. Recall that f is holomor-

phic. Moreover, the power series (3.5) has the same radius of convergence
R so it defines a holomorphic function g on DR (z0 ).
For any n ∈ N0 , using (z n+1 )′ = (n + 1)z n and the fundamental theorem
of calculus, I
an
an ζ n dζ = z2n+1 − z1n+1 .


[z1 z2 ] n+1
Summing in n from 0 to N and taking N → ∞, we conclude
I
f (ζ) dζ = g(z2 ) − g(z1 ).
[z1 z2 ]
Note that for the left-hand side, we used uniform convergence of the power
series on the line segment [z1 z2 ]. Thus, by Lemma 3.24, g is holomorphic at
the arbitrary point z1 ∈ Dr (z0 ) and g ′ (z1 ) = f (z1 ). □

5. Reparametrization and arclength

In this section, we consider the effect of precomposing a curve with a
real-valued function; this changes the parameter, i.e., reparametrizes the
curve. The geometric intuition is to think of two curves as equivalent if
they can be reparametrized into the same form, and we will address some
geometric complications.
We begin with preliminary results about the length of parts of rectifiable
curves:
Theorem 3.26. For any rectifiable curve γ : [a, b] → C,
(a) For any x0 ∈ [a, b),
lim ℓ(γ|[x0 ,x] ) = 0
x↓x0
(b) For any x0 ∈ (a, b],
lim ℓ(γ|[x,x0 ] ) = 0.
x↑x0
(c) The function
α(x) = ℓ(γ|[a,x] )
is continuous in x ∈ [a, b].
Proof. (a) By additivity of length, the curve γ|[x0 ,b] is rectifiable, so it
suffices to prove the case x0 = a.
Fix ϵ > 0. By continuity of γ at a, there exists δ > 0 such that |γ(x) −
γ(a)| < ϵ for x ∈ (a, a + δ). By the definition of ℓ(γ), there exists a partition
a = t0 < t1 < · · · < tn = b such that
Xn
|γ(tj ) − γ(tj−1 )| > ℓ(γ) − ϵ.
j=1

Adding additional points to the partition will preserve this inequality, so we

can assume without loss of generality that the partition contains a point in
(a, a + δ). Thus, t1 ∈ (a, a + δ).
48 3. CONTOUR INTEGRATION

Separating the term γ(t1 )−γ(t0 ) and using |γ(t1 )−γ(a)| < ϵ, we conclude
n
X
|γ(tj ) − γ(tj−1 )| > ℓ(γ) − 2ϵ.
j=2

Viewing t1 < · · · < tn = b as a partition of γ|[t1 ,b] , we now notice that

ℓ(γ|[t1 ,b] ) > ℓ(γ) − 2ϵ
and therefore by additivity of length, ℓ(γ|[a,t1 ] ) < 2ϵ. Thus, for every x ∈
(a, t1 ], ℓ(γ|[a,x] ) < 2ϵ. Since ϵ > 0 is arbitrary, this implies limx↓a ℓ(γ|[a,x] ) =
0.
(b) is proved analogously to (a).
(c) Fix x0 ∈ [a, b). By (a), we have
lim ℓ(γ|[x0 ,x] ) = 0
x↓x0

and adding ℓ(γ|[a,x0 ] ) = α(x0 ) to both sides gives

lim α(x) = α(x0 ).
x↓x0

Thus, α is right-continuous. Analogously, α is left-continuous. □

It is geometrically motivated to consider curves parametrized by length:
Definition 3.27. A curve γ : [a, b] → C is said to be arclength-parametrized
if a = 0 and for every x ∈ [a, b], ℓ(γ|[a,x] ) = x.
Arclength reparametrization of a rectifiable curve γ is expressed as a
decomposition of the function γ:
Theorem 3.28 (Arclength reparametrization). For any rectifiable curve
γ : [a, b] → C, there exists a unique triple (γ1 , α, L), where L ∈ [0, ∞),
γ1 : [0, L] → C is an arclength-parametrized curve, and α : [a, b] → [0, L] an
increasing surjection, such that
γ = γ1 ◦ α.
Proof. Uniqueness: We will first prove uniqueness of such a triple
(γ1 , α, L).
We say that a finite sequence a = t0 ≤ · · · ≤ tn = b is a partition with
repetitions of [a, b]. In the definition of length as a supremum over partitions,
note that we can allow partitions with repetitions, since repetitions don’t
affect the sum.
Next, note that an increasing surjection α : [a, b] → [0, L] maps parti-
tions with repetitions of [0, L] bijectively to partitions with repetitions of
[a, b]. Thus, we conclude
ℓ(γ) = ℓ(γ1 ).
Since γ1 is arclength parametrized, we conclude L = ℓ(γ). Likewise, for any
x,
ℓ(γ|[a,x] ) = ℓ(γ1 |[0,α(x)] ) = α(x),
 5. REPARAMETRIZATION AND ARCLENGTH 49

which determines α uniquely. Since α is surjective, γ1 ◦ α = γ determines

γ1 uniquely by general principles: for any s ∈ [0, L], there exists x with
α(x) = s, and γ(x) = γ1 (α(x)) = γ1 (s) determines the value γ1 (s) uniquely.
Existence: Define α : [a, b] → R by α(x) = ℓ(γ|[a,x] ); this function is
increasing and continuous, so it is a surjection to the interval [α(a), α(b)] =
[0, ℓ(γ)]. For any x < y,
|γ(y) − γ(x)| ≤ ℓ(γ|[x,y] ) = α(y) − α(x). (3.6)
In particular, α(x) = α(y) implies γ(x) = γ(y). Thus, by general principles,
there exists γ1 such that γ1 ◦ α = γ. As in the proof of uniqueness, for any
x,
ℓ(γ1 |[0,α(x)] ) = ℓ(γ|[a,x] ),
but now we use this to conclude ℓ(γ1 |[0,α(x)] ) = α(x) for any x and therefore
ℓ(γ1 |[0,s] ) = s for any s. This proves that γ1 is arclength-parametrized. □
Definition 3.29. The curve γ1 from the previous theorem is called the
arclength parametrization, or natural parametrization, of γ.
Two rectifiable curves are said to be equivalent if they have the same ar-
clength parametrization. This is an equivalence relation, and its equivalence
classes are called rectifiable contours.
Remark 3.30. Our level of generality allows curves which are locally
constant. For instance, we allow the rectifiable curves
( (
0 0≤t≤1 t 0≤t≤1
γ1 (t) = , γ2 (t) =
t−1 1<t≤2 1 1<t≤2
and consider them to be mutually equivalent, since both have the arclength
parametrization
γ1 (t) = t, 0 ≤ t ≤ 1.
In our definition, it doesn’t matter that there is no bijection β : [0, 2] → [0, 2]
such that γ̃ = γ ◦ β.
We often describe contours geometrically, for instance:
Example 3.31. The “unit circle oriented counterclockwise” is the closed
contour with the arclength parametrization given by γ(t) = eit , t ∈ [0, 2π].
Example 3.32. The “line segment from z0 to z1 ” is the contour corre-
sponding to the curve γ(x) = z0 + x(z1 − z0 ), 0 ≤ x ≤ 1. Its arclength repre-
sentation is γ1 (t) = z0 + teiϕ , t ∈ [0, L], using the polar form z1 − z0 = Leiϕ .
Arclength reparametrization of (piecewise) C 1 curves. An issue
with piecewise C 1 curves is that even if a curve γ is piecewise C 1 , its ar-
clength parametrization γ1 may not be (see Exercise 9). Thus, to stay within
this class, we must assume that γ1 is piecewise C 1 .
These issues are commonly avoided by assuming that γ ′ ̸= 0 (see Exer-
cise 10), further restricting the class of curves. We prefer not to rule out
50 3. CONTOUR INTEGRATION

zero derivative; in certain topological discussions, constant curves appear

naturally, so it seems formally necessary to allow those.
Theorem 3.33 (Reparametrization of contour integrals). Let γ1 : [0, L] →
C be a piecewise C 1 curve and α : [a, b] → [0, L] a piecewise C 1 increasing
surjection. Then γ = γ1 ◦ α is also a piecewise C 1 curve and for any con-
tinuous function h on the range of γ,
I I
h(z) dz = h(z) dz (3.7)
γ γ
I I1
h(z) |dz| = h(z) |dz|. (3.8)
γ γ1

Proof. For every x ∈ [a, b], the preimage α−1 ({x}) is a point or a closed
interval. Finitely many intervals can arise from values of x such that γ1′ is
discontinuous at x.
The function γ is continuous, as a composition of continuous maps. The
function γ is C 1 on any interval [u, v] such that: α is C 1 on [u, v] and γ1
is C 1 on [α(u), α(v)]. The function γ is also C 1 on any interval α−1 ({x}),
with x such that γ1′ is discontinuous there; inside such an interval, γ ′ = 0.
This makes [0, L] the union of finitely many closed intervals on which γ is
C 1 . Thus, γ is piecewise C 1 .
For any y, s such that α(y) = s, we get
Z y Z s
γ ′ (x) dx = γ(y) − γ(a) = γ1 (s) − γ1 (0) = γ1′ (t) dt.
a 0
Thus, for a characteristic function of an interval H = χ[0,s] ,
Z b Z L
′
H(α(x))γ (x) dx = H(t)γ1′ (t) dt.
a 0
Taking linear combinations, we conclude the same identity for piecewise
constant functions, and then by uniform approximation, for any continuous
H : [0, L] → C. Applying this to H = h ◦ γ1 gives (3.7); (3.8) is proved
analogously. □
Since integrals over the curve γ only depend on its equivalence class, we
are justified in calling them contour integrals.
We note that a much more natural level of generality can be achieved by
introducing contour integration over arbitrary rectifiable curves; since that
approach uses Lebesgue integration, we consider it as a bonus topic in this
course.
Concatenation of contours. For curves γj : [aj , bj ] → C, j = 1, 2 with
γ1 (b1 ) = γ2 (a2 ), their concatenation is the curve γ2 ∗γ1 : [a1 , b1 +b2 −a2 ] → C
given by
(
γ1 (t) a1 ≤ t ≤ b1
(γ2 ∗ γ1 )(t) =
γ2 (t + b1 − a2 ) b1 < t ≤ b1 + b2 − a2
 6. BONUS: A GALLERY OF CURVES 51

This operation commutes with arclength reparametrization, so it can also

be viewed as an operation on contours. Contour integrals are additive with
respect to concatenation:
I I I
h(z) dz = h(z) dz + h(z) dz,
γ2 ∗γ1 γ2 γ1
I I I
h(z) |dz| = h(z) |dz| + h(z) |dz|.
γ2 ∗γ1 γ2 γ1

Orientation reversal. For a curve γ : [a, b] → C, there is the orientation-

reversed curve γ ← : [a, b] → C,
γ ← (t) = γ(a + b − t).
This operation commutes with arclength reparametrization, so it can also
be viewed as an operation on contours. Orientation reversal affects the sign
of the contour integral:
I I
h(z) dz = − h(z) dz,
γ γ←
I I
h(z) |dz| = h(z) |dz|.
γ γ←

6. Bonus: a gallery of curves

In this bonus section, we consider a few examples of curves with in-
teresting or counter-intuitive properties. We always parametrize curves by
t ∈ [0, 1].
Example 3.34 (Cantor staircase). We define functions βn : [0, 1] → [0, 1]
inductively as follows: β0 (x) = x and

1
 2 βn (3x)
 0 ≤ x ≤ 13
βn+1 (x) = 2 1 1 2
3 <x< 3
1 1
 2
2 + 2 βn (3x − 2) 3 ≤x≤1
The functions converge uniformly to a continuous increasing function β :
[0, 1] → [0, 1], called the Cantor function. The ”Cantor staircase” γ(t) =
t + iβ(t) is a rectifiable curve and ℓ(γ) = 2.
Proof. Using the recursion to express βn+1 and βn in terms of βn and
βn−1 , respectively, shows that ∥βn+1 − βn ∥∞ = 12 ∥βn − βn−1 ∥∞ . By the
Weierstrass criterion, it follows that βn converge uniformly to a continuous
function β. Since all βn are increasing, β is increasing: for x < y,
β(x) = lim βn (x) ≤ lim βn (y) = β(y).
n→∞ n→∞

For any partition t = (tj )nj=0 , by the triangle inequality, we conclude that
|γ(tj ) − γ(tj−1 )| ≤ (tj − tj−1 ) + (β(tj ) − β(tj−1 )),
52 3. CONTOUR INTEGRATION

so summing in j gives the upper bound tn − t0 + β(tn ) − β(t0 ) = 2. Thus,

γ is rectifiable and ℓ(γ) ≤ 2.
The recursion implies that βn (0) = 0 and βn (1) = 1 for all n, so β(0) = 0
and β(1) = 1. Taking the limit in the recursion gives

1
 2 β(3x)
 0 ≤ x ≤ 13
β(x) = 12 1
3 <x< 3
2
1 1
 2
2 + 2 β(3x − 2) 3 ≤x≤1
so further values of β are computed by induction in n: for any a1 , . . . , an ∈
{0, 1},
   
n n n n
X 2aj  X aj 1 X 2aj  1 X aj
β  = , β  + = n+ .
3j 2j 3n 3j 2 2j
j=1 j=1 j=1 j=1

2n+1 −1
Using these 2n+1 points for the partition (tj )j=0 , we observe that for each
odd interval, β(t2j+1 ) − β(t2j ) = 2−n and therefore
2n−1
X−1
|γ(t2j+1 ) − γ(t2j )| ≥ 2n × 2−n = 1.
j=0

For even intervals (t2j , t2j−1 ), we estimate their total length as sum of
n −1
2X n −1
2X n
X 2j−1 1
|γ(t2j ) − γ(t2j−1 )| = |β(t2j ) − β(t2j−1 )| = =1− .
3j 3n
j=1 j=1 j=1
1
It follows that ℓ(γ) ≥ 2 − 3n for any n ∈ N and therefore ℓ(γ) ≥ 2. □

i 1+i

Figure 3.1. The graph of the Cantor function is a rectifiable

curve with length 2.
 6. BONUS: A GALLERY OF CURVES 53

Example 3.35 (Koch curve). Define γn : [0, 1] → C recursively by

γ0 (t) = t
and 1
 3 γn (4t)
 t ∈ [0, 1/4]
 1 + ω γ (4t − 1)

t ∈ (1/4, 1/2]
n
γn+1 (t) = 31 ω3 ω
3 + 3 + 3 γn (4t − 2) t ∈ (1/2, 3/4]



2 1
3 + 3 γn (4t − 3) t ∈ (3/4, 1]
Then γn converge uniformly to a curve γ : [0, 1] → C which has no self-
intersections and is not rectifiable.
Proof. A direct calculation gives
1
∥γ1 − γ0 ∥ = √
2 3
and it follows from the recursion that
1
∥γn+1 − γn ∥ = ∥γn − γn−1 ∥.
3
For this reason, γn has a uniform limit as n → ∞ to a continuous map
γ : [0, 1] → C, and that curve obeys the self-similar property
1

 3 γ(4t) t ∈ [0, 1/4]
 1 + ω γ(4t − 1)

t ∈ (1/4, 1/2]
γ(t) = 31 ω3 ω
+ + 3 γ(4t − 2) t ∈ (1/2, 3/4]
 32 13



3 + 3 γ(4t − 3) t ∈ (3/4, 1]
From the obvious inequality ℓ(γ) ≥ |γ(1) − γ(0)| = 1, we know the curve
has positive length. From the self-similar property and additivity of length,
we know
4
X 4
ℓ(γ) = ℓ(γ|[(j−1)/4,j/4] ) = ℓ(γ)
3
j=1

which implies ℓ(γ) = ∞.

By induction, it is straightforward to verify that all iterates of the Koch
curve are contained in the same triangle
1 1
γn ([0, 1]) ⊂ T = {z ∈ C | Im z ≥ 0, Im z ≤ √ Re z, Im z ≤ √ (1 − Re z)}.
3 3
Thus, γ([0, 1]) ⊂ T .
By comparing the four images of T in the affine maps used, we observe
that γ(x) ̸= γ(0) for all x ≥ 1/4.
By induction in n, we observe that γ(x) ̸= γ(0) for all x ≥ 1/4n .
Thus, γ(x) ̸= γ(0) for all x ̸= 0. Likewise, γ(x) ̸= γ(1) for all x ̸= 1, so
γ((0, 1)) ⊂ T \ {0, 1}.
54 3. CONTOUR INTEGRATION

Since the four images of T \ {0, 1} are mutually disjoint, we conclude by

induction that if ⌊4n x⌋ =
̸ ⌊4n y⌋ for some n ∈ N, then γ(x) ̸= γ(y). Thus, γ
is injective. □
From the Koch curve, it is common to construct the Koch snowflake,
which is a closed curve defined by

ωγ(3t)
 t ∈ [0, 1/3]
σ(t) = ω + (1 − ω)γ(3t − 1) t ∈ (1/3, 2/3]

1 − γ(3t − 2) t ∈ (2/3, 1]


Example 3.36 (The Lebesgue space-filling curve). Use the “middle

thirds” Cantor set  
∞
X 2aj 
C= | a j ∈ {0, 1}
 3j 
j=1
and map the point
∞
X 2aj
x= , aj ∈ {0, 1}
3j
j=1
to
∞ ∞
X a2j−1 X a2j
γ(x) = +i .
2j 2j
j=1 j=1
This is a continuous surjective map γ : C → [0, 1]2 . Linear interpolation on
the gaps (a, b) ⊂ [0, 1] \ C makes it a continuous surjective map γ : [0, 1] →
[0, 1]2 .
The Lebesgue space-filling curve maps γ(0) = 0 and γ(1) = 1 + i, but
it can be concatenated with the line segment [1 + i, 0] to produce a closed
curve with range [0, 1]2 .

7. Bonus: Lebesgue integral representation of rectifiable curves

In this section, we derive an integral representation for an arbitrary
rectifiable curve, in the language of measure theory. This is the modern
level of generality; it produces more natural results and allows us to integrate
over arbitrary rectifiable curves. Since complex analysis can be developed
in a satisfactory way with a more restrictive class of curves, this section
can be skipped. The prerequisite for this section is Lebesgue integration in
one dimension, in particular, the Lebesgue–Stieltjes measures on R and the
Lebesgue differentiation theorem.
Definition 3.37. For a Borel measure µ on R, we use the notation
Z y Z
f dµ = sgn(y − x) f dµ.
x (min{x,y},max{x,y}]

Note that, in the case when µ is Lebesgue measure, this corresponds to

the standard notation for Riemann integrals.
 7. BONUS: LEBESGUE INTEGRAL REPRESENTATION OF RECTIFIABLE CURVES55

Theorem 3.38. For every rectifiable curve γ : [a, b] → C, there is

a unique pair (µ, s) where µ is a finite Borel measure on (a, b] and s ∈
L∞ ((a, b], dµ) such that |s| = 1 µ-a.e. and
Z x
γ(x) = γ(a) + s dµ, ∀x ∈ [a, b]. (3.9)
a
Moreover, µ and s have the following properties: µ is a continuous measure,
µ((x, y]) = ℓ(γ|[x,y] ), ∀x < y,
and for µ-a.e. x,
γ(x + ϵ) − γ(x − ϵ)
s(x) = lim . (3.10)
ϵ↓0 ℓ(γ|[x−ϵ,x+ϵ] )
Proof. We begin by proving existence and additional properties. The
function 
0
 x≤a
α(x) = ℓ(γ|[a,x] ) a < x ≤ b

ℓ(γ) x>b


is a continuous increasing function of x. There exists a unique Borel measure

µ on R such that
µ((x1 , x2 ]) = α(x2 ) − α(x1 )
for all x1 < x2 . Note that
|γ(x2 ) − γ(x1 )| ≤ ℓ(γ|[x1 ,x2 ] ) ≤ α(x2 ) − α(x1 ).
We now denote
γ0 = α, γ1 = α, γ2 = α − Re γ, γ3 = α − Im γ.
Since for x < y,
Re γ(y) − Re γ(x) ≤ |γ(y) − γ(x)| ≤ α(y) − α(x),
the function γ2 is increasing. Analogously, γ3 is increasing. In this way we
obtain a representation
X3
γ= i k γk
k=0
where γk are increasing functions on [a, b] which obey
|γk (y) − γk (x)| ≤ µ((x, y]).
These are distribution functions for some measures µk . By the above,
µk ((x, y]) ≤ µ((x, y]), so µk (B) ≤ µ(B) for all Borel sets B. Thus, µk
are continuous with respect to µ, so by Radon–Nikodym, they are abso-
lutely continuous: there exist Borel functions fk ≥ 0 such that dµk = fk dµ.
In summary, Z
γk (x) = γk (a) + fk χ(x,y] dµ
56 3. CONTOUR INTEGRATION

with 0 ≤ fk ≤ 1. Taking linear combinations, we obtain a representation

Z x X3
γ(x) = γ(a) + s(t) dµ(t), s(t) = ik fk (t),
a k=0

with s ∈ L1 ((a, b], dµ). By the Lebesgue differentiation theorem, for µ-a.e.
x,
γ(x + ϵ) − γ(x − ϵ)
s(x) = lim (3.11)
ϵ↓0 µ((x − ϵ, x + ϵ])
which proves |s| ≤ 1 µ-a.e..
On the other hand, for every u < v,
Z v
|γ(v) − γ(u)| ≤ |s| dµ
u
so applying this to an arbitrary partition and taking the supremum gives
Z b Z b
µ((a, b]) = ℓ(γ) ≤ |s| dµ ≤ dµ.
a a
To have equality, we conclude that |s| = 1 µ-a.e..
It remains to prove uniqueness. Assume that for another pair (ν, t),
where ν is a finite measure and |t| = 1 ν-a.e., we have
Z x
γ(x) = γ(a) + t dν, ∀x ∈ [a, b].
a
Then Z y
|γ(y) − γ(x)| ≤ |t| dν
x
so applying this to partitions implies
Z y
ℓ(γ|(x,y] ) ≤ |t| dν.
x
Therefore µ((x, y]) ≤ ν((x, y]) so µ(B) ≤ ν(B) for all Borel sets B. Thus, µ
is continuous with respect to ν, so it is absolutely continuous, so dµ = f dν.
In the chain of inequalities
Z x+ϵ
|γ(x + ϵ) − γ(x − ϵ)| ≤ µ((x − ϵ, x + ϵ]) ≤ dν
x−ϵ

dividing by ν((x − ϵ, x + ϵ]) and letting ϵ → 0 gives, by the Lebesgue differ-

entiation theorem, for ν-a.e. x,
|t(x)| = f (x) = |t(x)|.
Thus, f = 1 ν-a.e., so µ = ν. Now by
Z x+ϵ Z x+ϵ
t dµ = γ(x + ϵ) − γ(x − ϵ) = s dµ
x−ϵ x−ϵ

dividing by µ((x − ϵ, x + ϵ]) and letting ϵ → 0 gives t = s µ-a.e.. □

 7. BONUS: LEBESGUE INTEGRAL REPRESENTATION OF RECTIFIABLE CURVES57

We will call (3.9) the integral representation of the rectifiable curve γ.

Informally speaking, the infinitesimal version of this representation is
dγ = s dµ,
which will motivate the definition of contour integrals over rectifiable curves.
Theorem 3.39. If γ : [a, b] → Ω is rectifiable and f : Ω → C is holomor-
phic with f ′ continuous, then f ◦γ is rectifiable. If the integral representation
of γ is denoted (3.9), then f ◦ γ has the integral representation
Z x
f (γ(x)) = f (γ(a)) + sgnf ′ (γ(t))s(t) dν(t), dν = |f ′ ◦ γ| dµ (3.12)
a
and length
Z b
ℓ(f ◦ γ) = |f ′ ◦ γ| dµ.
a

Proof. In the proof we will repeatedly use the following consequence

of the fundamental theorem of calculus: for a line segment [z, w] ∈ Ω,
Z 1
f (z) − f (w)
= f ′ (z + t(w − z)) dt.
z−w 0

This follows from the fundamental theorem of calculus applied to the C 1

function h(t) = f (z + t(w − z)), t ∈ [0, 1].
Since Ran γ is a compact subset of Ω, it is at a positive distance from
Ωc . Thus, there exists ϵ > 0 such that
Kϵ = {z ∈ C | dist(z, Ran γ) ≤ ϵ}
is a compact subset of Ω. Since γ is uniformly continuous, there exists δ > 0
such that |x − y| < δ implies |γ(x) − γ(y)| < ϵ.
Denote
C = max|f ′ (z)|.
z∈Kϵ
For any z, w ∈ Ran γ with |z − w| ≤ ϵ, the line segment with endpoints z, w
lies in Kϵ , so
|f (z) − f (w)| ≤ C|z − w|.
By applying this estimate to an arbitrary fine enough partition, we conclude
that f ◦ γ is rectifiable with
ℓ(f ◦ γ) ≤ Cℓ(γ).
Since Z y
|f (γ(y)) − f (γ(x))| ≤ C|γ(y) − γ(x)| ≤ C |s| dµ,
x
for the arclength measure of f ◦ γ we conclude
Z y
ℓ(f ◦ γ|[x,y] ) ≤ C |s| dµ
x
58 3. CONTOUR INTEGRATION

so this arclength measure is continuous with respect to the measure |s| dµ.
Thus, by Radon–Nikodym, it is absolutely continuous with respect to |s| dµ,
so it has an integral representation
Z x
f (γ(x)) = f (γ(a)) + h(t)|s(t)| dµ(t)
a
for some function h ∈ L∞ ((a, b], dµ).
For µ-a.e. x, we have
γ(x + ϵ) − γ(x − ϵ)
lim = s(x) ∈ ∂D
ϵ→0 ν((x − ϵ, x + ϵ))

and in particular γ(x + ϵ) ̸= γ(x − ϵ) for all small enough ϵ > 0.

For any such x, by the fundamental theorem of calculus from real anal-
ysis,
Z 1
f (γ(x + ϵ)) − f (γ(x − ϵ))
= f ′ ((1 − t)γ(x − ϵ) + tγ(x + ϵ)) dt
γ(x + ϵ) − γ(x − ϵ) 0
which converges to f ′ (γ(x)) as ϵ → 0.
By the Lebesgue differentiation theorem, we obtain for µ-a.e. t
f ′ (γ(t))s(t) = h(t)|s(t)|
and then (3.12). □
Once we rewrite this in the language of contour integrals, it will be
precisely the fundamental theorem of calculus for contour integrals.
Contour integral on rectifiable curves using the Lebesgue inte-
gral. Every rectifiable curve γ : [a, b] → C has an integral representation
Z x
γ(x) = γ(a) + s(t) dµ(t) (3.13)
a
where µ is a finite continuous measure on (a, b] and |s| = 1 µ-a.e.. This
representation allows us to think of an infinitesimal curve element formally
as dγ = s dµ, which motivates the following definition:
Definition 3.40. For the rectifiable curve γ with representation (3.13),
the contour integral of a continuous function f : Ran γ → C is defined by
I Z b
f (z)dz = (f ◦ γ)s dµ (3.14)
γ a
I Z b
f (z)|dz| = (f ◦ γ)|s| dµ (3.15)
γ a
Example 3.41. Directly from the definitions,
I Z b
1 dz = s dµ = γ(b) − γ(a),
γ a
I Z b
1 |dz| = |s| dµ = ℓ(γ).
γ a
 7. BONUS: LEBESGUE INTEGRAL REPRESENTATION OF RECTIFIABLE CURVES59

By Theorem 3.39 we have:

Theorem 3.42 (Fundamental theorem of calculus for contour integrals).
Let γ : [a, b] → Ω be a rectifiable curve in Ω. Let f be holomorphic in Ω and
f ′ continuous. Then
I
f (γ(b)) − f (γ(a)) = f ′ (z) dz
γ
I
ℓ(f ◦ γ) = |f ′ (z)| |dz|.
γ

More generally, contour integrals over the curve f ◦ γ can be expressed

in terms of contour integrals over the curve γ:
Theorem 3.43 (Substitution rule for contour integrals). Let γ : [a, b] →
Ω be a rectifiable curve in Ω. Let f be holomorphic in Ω and f ′ continuous.
Let h be a continuous function on the range of f ◦ γ. Then
I I
h(w) dw = h(f (z))f ′ (z) dz
f ◦γ γ
I I
h(w) |dw| = h(f (z))|f ′ (z)| |dz|.
f ◦γ γ

Properties of contour integrals. As a direct consequence of linearity

of Lebesgue integration:
Lemma 3.44. Contour integrals are linear functionals, i.e.,
I I I
(αf + βg)(z) dz = α f (z)dz + β g(z)dz
γ γ γ

for any α, β ∈ C and continuous functions f, g : Ω → C.

Lemma 3.45 (Contour additivity). If γ : [a, b] → C is rectifiable, x ∈
(a, b), and γ1 = γ|[a,x] , γ2 = γ|[x,b] , then
I I I
f (z) dz = f (z) dz + f (z) dz.
γ γ2 γ1

Lemma 3.46. Contour integrals over rectifiable curves obey the inequal-
ities I I
f (z)dz ≤ |f (z)| |dz| ≤ ℓ(γ)∥f ∥γ .
γ γ

Proof. In terms of the integral representation of the rectifiable curve

γ, this reduces to standard inequalities for Lebesgue integrals,
Z Z Z
(f ◦ γ)s dµ ≤ |(f ◦ γ)s| dµ ≤ ∥f ∥γ |s| dµ. □

By Theorem 3.39 we have:

60 3. CONTOUR INTEGRATION

Theorem 3.47 (Fundamental theorem of calculus for contour integrals).

Let γ : [a, b] → Ω be a rectifiable curve in Ω. Let f be holomorphic in Ω and
f ′ continuous. Then
I
f (γ(b)) − f (γ(a)) = f ′ (z) dz
γ
I
ℓ(f ◦ γ) = |f ′ (z)| |dz|.
γ

More generally, contour integrals over the curve f ◦ γ can be expressed

in terms of contour integrals over the curve γ:
Theorem 3.48 (Substitution rule for contour integrals). Let γ : [a, b] →
Ω be a rectifiable curve in Ω. Let f be holomorphic in Ω and f ′ continuous.
Let h be a continuous function on the range of f ◦ γ. Then
I I
h(w) dw = h(f (z))f ′ (z) dz
f ◦γ γ
I I
h(w) |dw| = h(f (z))|f ′ (z)| |dz|.
f ◦γ γ

Theorem 3.49. If γ is a rectifiable curve and γ1 its arclength parametriza-

tion, then for any continuous h on Ran γ,
I I
h(z) dz = h(z) dz, (3.16)
γ γ1
I I
h(z) |dz| = h(z) |dz|. (3.17)
γ γ1

Proof. The pushforward of the measure µ by α : [a, b] → [0, L] is

defined as the measure ν(B) = µ(α−1 (B)). Since µ((x, y]) = α(y) − α(x),
we conclude
ν((u, v)]) = µ(α−1 ((u, v])) = µ(( max x, max y]) = v − u
x:α(x)=u y:α(y)=v

i.e., ν is Lebesgue measure. We will repeatedly Ruse the basic property of

pushforward measures: for any g ∈ L1 ([0, L], dν), g dν = (g ◦ α) dµ. This
R
already implies (3.17) since
I Z Z I
h |dz| = h(γ1 (α(t))) dµ(t) = h(γ1 (u)) du = h |dz|.
γ γ1
The arclength-parametrized curve γ1 has Lebesgue measure in its inte-
gral representation:
Z u Z maxx:α(x)=u x
γ1 (u) = s1 (τ ) dτ = s1 (α(t)) dµ(t).
0 0
Since µ gives zero measure to intervals of constancy of α, for every x,
Z x
γ1 (α(x)) = s1 (α(t)) dµ(t).
0
 8. EXERCISES 61

Comparing with the representation of γ(x), we conclude s1 ◦ α = s in

L∞ ([a, b], dµ). Thus, by the property of the pushforward measure,
I Z Z I
h dz = h(γ1 (α(t)))s(t) dµ(t) = h(γ1 (u))s1 (u) du = h dz. □
γ γ1

Further properties for concatenation of curves and orientation reversal

follow as before.

8. Exercises
1. The curve γ : [0, 1] → C is defined by
(
t + it sin(π/t) t ∈ (0, 1]
γ(t) =
0 t=0
(a) Prove that γ is continuous.
Hint: explain why the only concern is at t = 0 and prove continuity
at t = 0 by some custom argument.
(b) Prove that γ is not rectifiable.
Hint: use some of the local maxima and minima of sin to con-
struct partitions t = (tj )nj=0 of [0, 1] with arbitrarily large length
Pn−1
j=0 |γ(tj+1 ) − γ(tj )|.
2. (a) For any z1 , . . . , zk , z1′ , . . . , zk′ ∈ D, prove that
|z1 · · · zk − z1′ · · · zk′ | ≤ k max |zj − zj′ |.
j∈{1,...,k}

Hint: telescoping sum.

(b) For any z1 , . . . , zk ∈ D, prove that
z1k + · · · + zkk
− z1 · · · zk ≤ (k − 1)2 max |zi − zj |2 .
k i,j∈{1,...,k}

z k +···+z k
Hint: denoting P (z1 , . . . , zk ) = 1 k k − z1 · · · zk , compute the par-
tial derivative ∂zk P and then estimate |P (z1 , . . . , zk−1 , zk )−P (z1 , . . . , zk−1 , z1 )|.
3. If γ : [a, b] → C \ {0} is a rectifiable curve and n ∈ Z \ {−1}, prove that
γ(b)n+1 − γ(a)n+1
I
z n dz = .
γ n+1
Hint: use the fundamental theorem of calculus for contour integrals.
4. Denote Ω = C \ (−∞, 0] and assume f : Ω → C is holomorphic on Ω,
with f ′ continuous.
(a) If |f ′ (z)| ≤ 1/|z| for all z, prove that
|f (i) − f (−i)| ≤ π.
(b) If |f ′ (z)| ≤ 1 for all z, prove that
|f (i) − f (−i)| ≤ 2.
 62 3. CONTOUR INTEGRATION

5. Let f : D \ [0, 1) → C be holomorphic, with f ′ continuous. If for some

κ > −1,
|f ′ (z)| = O(|z|κ ), z → 0,
prove that:
(a) there exist C, ρ > 0 such that for all z1 , z2 ∈ Dρ (0) \ [0, 1),
|f (z1 ) − f (z2 )| ≤ C(max{|z1 |, |z2 |})κ+1 .
Hint: integrate f ′ along a contour consisting of a line segment and a
circular arc.
(b) The limit
lim f (z)
z→0
D\[0,1)
exists in C.
Hint: prove that for any sequence zn ∈ D \ [0, 1) with zn → 0, the
sequence f (zn ) is Cauchy. Then the statement follows by general
metric space arguments.
6. Let Ω ⊂ C be open and convex. Assume that f : Ω → C is holomorphic,
f ′ continuous, and Re f ′ > 0 on Ω. Prove that f is injective.
Hint: for z0 ̸= z1 , express the real part of the ratio (f (z1 )−f (z0 ))/(z1 −
z0 ) as a Riemann
P integral.
7. Assume that ∞
P∞ n
n=2 n|an | ≤ |a1 |. Prove that the function f (z) = n=0 an z
is well-defined and holomorphic on D. If f is not identically zero, prove
that f is injective on D.
8. Let γ : [a, b] → C be a rectifiable curve. Prove that for every ϵ > 0 there
exists δ > 0 such that for all x, y ∈ [a, b] with 0 < y − x < δ,
ℓ(γ|[x,y] ) < ϵ.
9. Prove that the curve γ : [0, 1] → C given by γ(0) = 0 and
eiπ/n 1 1
γ(t) = sin2 (πn(n + 1)t), <t≤ , n∈N
n3 n+1 n
is C 1 , but that its arclength parametrization is not piecewise C 1 .
10. If the curve γ is piecewise C 1 and γ ′ ̸= 0 (including one-sided derivatives
at points where the derivative doesn’t exist), prove that the arclength
parametrization of γ is piecewise C 1 .
11. Let γ : [a, b] → C be a rectifiable curve. Prove that
ℓ(γ) = |γ(b) − γ(a)|
if and only if γ is (equivalent to) the line segment [γ(a), γ(b)].
 CHAPTER 4

The Cauchy integral theorem

So far, the differential and integral calculus we developed has been very
similar to real variable calculus. In Chapter 2 we developed differentiation,
and in Chapter 3 we developed contour integration, which takes the role of
the definite integral. There is something missing: a notion of antiderivative
(indefinite integral). This is the subject of this chapter. We will see that
this is closely related to integrals over closed contours and to the Cauchy
integral theorem, which will be the first major theorem of this course.
We will also see that properties of the domain begin to play an important
role. We start this chapter by examining connectedness, and how it pertains
to open subsets of the complex plane.

1. Connected components; path connectedness

We begin by recalling the notion of connected subsets of a metric space
(see Appendix 2).
Definition 4.1. Let X be a metric space.
Two subsets A, B ⊂ X are separated if A ∩ B and A ∩ B are empty.
A subset E ⊂ X is connected if, for any separated sets A, B ⊂ X,
E ⊂ A ∪ B implies E ⊂ A or E ⊂ B.
Since A ∩ B = ∅, an equivalent characterization is: a subset E ⊂ X is
connected if, for any separated sets A, B ⊂ X, E ⊂ A ∪ B implies E ∩ A = ∅
or E ∩ B = ∅.
Connected components.
Definition 4.2. The connected component of a point x ∈ X is the
union of all connected subsets of X that contain x. We denote it Γx .
Since {x} is connected, Γx is well-defined and x ∈ Γx .
By the following lemma, a connected component is connected:
Lemma 4.3. Consider a nonempty family F ⊂ P(X) such that for every
E ∈ F, E is connected and x ∈ E. Then ∪E∈F E is connected.
In particular, the connected component of any point is connected.
Proof. Assume that ∪E∈F E ⊂ A ∪ B. Then x ∈ A or x ∈ B. Without
loss of generality, assume x ∈ A. Since every E ∈ F is connected, E ⊂ A∪B
and x ∈ E ∩ A implies E ⊂ A. Thus, ∪E∈F E ⊂ A. □
63
64 4. THE CAUCHY INTEGRAL THEOREM

Theorem 4.4. The binary relation y ∈ Γx is an equivalence relation. In

particular, connected components form a partition of the space X.
Proof. We have already seen that x ∈ Γx , so the relation is reflexive.
Directly from the definition, y ∈ Γx if and only if there exists E con-
nected such that x, y ∈ E. Thus, y ∈ Γx if and only if x ∈ Γy .
If x ∈ Γy and y ∈ Γz , then x, z ∈ Γy . By the previous observation, using
Γy as the connected set, we have x ∈ Γz . □
In particular, a space X is connected if and only if it consists of only
one connected component.
It is now easy to describe connected components in some simple geomet-
ric situations:
Lemma 4.5. Any convex subset of C is connected. In particular, any
half-plane is connected.
Proof. For any x, y in a convex set S, the line segment with endpoints
x, y is connected as a continuous image of [0, 1]. This line segment is con-
tained in S. Thus, S is connected. □
Example 4.6. The connected components of C \ R are the half-planes
C± = {z ∈ C | ± Im z > 0}.
Proof. C± are connected. Thus, there exists a connected component
which contains C+ and a connected component which contains C− .
Since Im is continuous, C+ ⊂ C \ C− and C− ⊂ C \ C+ . Thus, C+ and
C− are separated, so every connected set E ⊂ C \ R is contained in one or
them. Thus, they are the maximal connected subsets of C \ R. □
Example 4.7. The connected components of C \ {z | |z| = 1} are D and
C \ D.
Proof. The set A = D is connected because it is convex.
B = C \ D is connected because it is the continuous image of {z | Re z >
0} in the exponential function.
Thus, there is a connected component containing A and a connected
component containing B.
By continuity of absolute value, A ⊂ D and B ⊂ C \ D, so A, B are
separated. Thus, any connected set is contained in A or in B.
Thus, A and B are the two connected components. □
In general, if we find a partition of a space into connected closed sets,
then that is also the partition into connected components.
Path-connectedness. In an open subset of C, the following notions of
connectedness and path-connectedness are equivalent:
Theorem 4.8. If Ω is an open subset of C, for any two points z0 , z1 ∈ Ω,
the following are equivalent:
 1. CONNECTED COMPONENTS; PATH CONNECTEDNESS 65

(a) z0 , z1 are in the same connected component of Ω

(b) There exists a continuous map γ : [0, 1] → Ω with γ(0) = z0 , γ(1) = z1
(c) There exists a piecewise C 1 map γ : [0, 1] → Ω with γ(0) = z0 , γ(1) = z1
(d) There exists a piecewise linear map γ : [0, 1] → Ω with γ(0) = z0 ,
γ(1) = z1
Proof. (d) =⇒ (c) =⇒ (b) is obvious.
(b) =⇒ (a): The set γ([0, 1]) is connected as the continuous image of
the connected set [0, 1]. Since γ([0, 1]) contains z0 , z1 , these two points are
in the same connected component of Ω.
(a) =⇒ (d): Denote by E the connected component of z0 . Fix z0 ∈ Ω.
Denote
A = {z1 ∈ Ω | ∃ piecewise linear γ in Ω : γ(0) = z0 , γ(1) = z1 }
We will prove that A is open and closed in Ω.
For any z1 ∈ Ω, there exists δ > 0 such that Dδ (z1 ) ⊂ Ω. Fix z2 ∈ Dδ (z1 ).
If z1 ∈ A, concatenating a curve from z0 to z1 with the line segment from
z1 to z2 gives a curve from z0 to z2 . Thus, A is open.
Similarly, if z2 ∈ A, then z1 ∈ A. By contraposition, if z1 ∈ Ω \ A, then
z2 ∈ Ω \ A. Thus, A is closed in Ω.
Thus, A is open and closed in Ω; in other words, A and Ω \ A are
separated. Since E is connected, this means that E ⊂ A or E ⊂ Ω \ A.
From z0 ∈ A ∩ E, it follows that E ⊂ A. □
Holomorphic functions with zero derivative.
Lemma 4.9. If g is holomorphic on an open set Ω, and g ′ = 0 on Ω,
then g is constant on each connected component of Ω.
Proof. For any z0 , z1 in the same connected component of Ω, there
exists a rectifiable curve γ in Ω connecting z0 to z1 . By the fundamental
theorem of calculus,
I
g(z1 ) − g(z0 ) = g ′ (z) dz = 0. □
γ

Lemma 4.10. A function f on an open set Ω ⊂ C is continuous/holomorphic

on Ω if and only if it is continuous/holomorphic on each of the connected
components of Ω.
Proof. For any z ∈ Ω, there exists ϵ > 0 such that Dϵ (z) ⊂ Ω.
Then Dϵ (z) is contained in the connected component of Ω, so continu-
ity/holomorphicity at z only depends on the values of f in the connected
component of z. □
Due to this, from now on, we will consider functions f on an open
connected set Ω ⊂ C.
Definition 4.11. A domain is an open, connected, nonempty subset of
C.
66 4. THE CAUCHY INTEGRAL THEOREM

2. Antiderivatives and closed contours

Recall that a domain Ω is always assumed to be open and connected.
A curve γ : [a, b] → Ω is said to be closed if γ(b) = γ(a). Applying the
fundamental theorem of calculus to closed curves gives an immediate, but
very important corollary:
Corollary 4.12. Let γ be a closed piecewise C 1 curve in Ω. Let g be
holomorphic in Ω and g ′ continuous. Then
I
g ′ (z)dz = 0.
γ

Proof. By the fundamental theorem of calculus, this is equal to g(γ(b))−

g(γ(a)). Since γ(a) = γ(b), this is equal to zero. □
Corollary 4.13. For any polynomial p(z) and any closed piecewise C 1
curve γ in Ω, I
p(z)dz = 0.
γ
Pn j
Proof. Any polynomial p(z) = j=0 aj z is the derivative of some
Pn aj j+1
polynomial g(z) = j=0 j+1 z , so
I I
p(z)dz = g ′ (z)dz = 0. □
γ γ

This also tells us that it isn’t always possible to find a holomorphic

function with a prescribed derivative:
Corollary 4.14. There is no holomorphic function g : C \ {0} → C
such that g ′ (z) = 1/z.
Proof. If such aHfunction g existed, by the fundamental theorem of cal-
culus we would have |z|=1 g ′ (z) dz = 0. Since we have previously computed
the value of this integral as 2πi, this would be a contradiction. □
The following theorem gives a general criterion for the existence of an
antiderivative:
Theorem 4.15. Let f : Ω → C be continuous. The following are equiv-
alent:
(a) f has an antiderivative on Ω.
(b) For every closed piecewise C 1 curve γ on Ω,
I
f (z) dz = 0.
γ

(c) For any piecewise C1 curves γ0 , γ1 with the same starting point z0 and
the same endpoint z1 ,
I I
f (z) dz = f (z) dz.
γ0 γ1
 3. TRIANGLES AND THE GOURSAT ARGUMENT 67

Proof. (a) =⇒ (b): If f = g ′ for some holomorphic g : Ω → C, then

by the fundamental theorem of calculus,
I I
f (z) dz = g ′ (z) dz = 0.
γ γ

(b) =⇒ (c): If γ0 , γ1 have the same starting points and endpoints, then
γ1← ∗ γ0 is a closed curve so
I I I
f (z) dz − f (z) dz = f (z) dz = 0.
γ0 γ1 γ1← ∗γ0

(c) =⇒ (a): We fix z0 in Ω and, for each z ∈ Ω, define

I
g(z) = f (ζ) dζ
γ
where γ is some curve with starting point z0 and endpoint z; by connected-
ness, such a curve exists, and by assumption (c), the value of the integral
doesn’t depend on the choice of curve. We will prove that g ′ = f .
Fix z1 ∈ Ω and r > 0 such that Dr (z1 ) ⊂ Ω. If γ is a rectifiable curve
with starting point z0 and endpoint z1 , then for z ∈ Dr (z1 ), the curve [z1 z]∗γ
has starting point z0 and endpoint z, so
I I I
g(z) = f (ζ) dζ + f (ζ) dζ = g(z1 ) + f (ζ) dζ.
γ [z1 z] [z1 z]

Thus, by Lemma 3.24, g is holomorphic at z1 and g ′ (z 1) = f (z1 ). □

We will now work towards a geometric/topological description of do-
mains on which every holomorphic function has an antiderivative.

3. Triangles and the Goursat argument

In this section, we will prove the vanishing of contour integrals over
suitable triangles, following an elegant proof of Goursat.

Triangles. A triangle is a contour γ composed of three line segments

[z0 z1 z2 z0 ] = [z2 z0 ] ∗ [z1 z2 ] ∗ [z0 z1 ].
A geometer usually assumes that vertices z0 , z1 , z2 are not collinear, but we
allow arbitrary z0 , z1 , z2 ∈ C.
We have previously defined the convex hull of a set. We define the convex
hull of a curve as the convex hull of its image,
cvh γ = cvh Ran γ,
and prove a geometrically obvious observation about triangles:
Lemma 4.16. For any z0 , z1 , z2 ∈ C, the convex hull of the triangle
γ = [z0 z1 z2 z0 ] is the compact set
cvh γ = {c0 z0 + c1 z1 + c2 z2 | c0 , c1 , c2 ≥ 0, c0 + c1 + c2 = 1}.
68 4. THE CAUCHY INTEGRAL THEOREM

Proof. Recall that

cvh{z0 , z1 , z2 } = {c0 z0 + c1 z1 + c2 z2 | c0 , c1 , c2 ≥ 0, c0 + c1 + c2 = 1}.
With c2 = 0, we see that this contains all convex combinations of z0 , z1 ,
i.e., the line segment [z0 z1 ]. Analogously it contains [z1 z2 ] and [z2 z0 ], so
Ran γ ⊂ cvh{z0 , z1 , z2 }. Now from
{z0 , z1 , z2 } ⊂ Ran γ ⊂ cvh{z0 , z1 , z2 }
it follows that cvh{z0 , z1 , z2 } ⊂ cvh γ ⊂ cvh{z0 , z1 , z2 } and therefore cvh γ =
cvh{z0 , z1 , z2 }.
P2 3
The map (c0 , c1 , c2 ) 7→ j=0 cj zj is a continuous map from R to C.
Note that {(c0 , c1 , c2 ) ∈ R3 | c0 , c1 , c2 ≥ 0, c0 + c1 + c2 = 1} is a compact
subset of R3 , and cvh γ is the continuous image of that compact. □
Theorem 4.17 (Goursat). If f is holomorphic on a domain Ω and γ a
triangle with cvh(γ) ⊂ Ω, then
I
f (z)dz = 0.
γ

The proof needs a lemma:

Lemma 4.18. Under the assumptions of Theorem 4.17, there exists a
triangle γ̃ with cvh γ̃ ⊂ cvh γ, diam cvh γ̃ = 21 diam cvh γ, ℓ(γ̃) = 21 ℓ(γ), and
I I
1
f (z)dz ≥ f (z)dz .
γ̃ 4 γ
z +z
Proof. For a triangle γ = [z0 z1 z2 z0 ], consider points wj = j+1 2 j+2
and consider the four triangles γ0 = [w0 w1 w2 w0 ], γ1 = [z0 w2 w1 z0 ], γ2 =
[z1 w0 w2 z1 ], γ3 = [z2 w1 w0 z2 ]. Each of them has half the length and half the
diameter of the original triangle, since they can be obtained by affine maps
with scaling 21 from γ. For instance, the affine map z 7→ z0 + 21 (z − z0 ) maps
γ to γ1 .
By contour arithmetic, noting that the integrals over line segments
[wj wj+1 ] and [wj+1 wj ] cancel out, we conclude that
I X 3 I
f (z)dz = f (z)dz.
γ j=0 γj
H
Thus, at least one of the integrals γj f (z)dz must have absolute value at
least 41 γ f (z)dz .
H
□

Proof of Theorem 4.17. Using the previous lemma as the inductive

step, we construct a sequence of triangles γn such that
I I
1
f (z)dz ≥ n f (z)dz
γn 4 γ
 3. TRIANGLES AND THE GOURSAT ARGUMENT 69

and cvh(γn+1 ) ⊂ cvh(γn ),

diam cvh(γ) ℓ(γ)
diam cvh(γn ) = n
, ℓ(γn ) = n
2 2
Since the sets cvh(γn ) are compact and nested, they have a nonempty inter-
section; since their diameter goes to zero, the intersection is a single point
which we denote w.
For every ϵ > 0, there exists δ > 0 such that 0 < |z − w| < δ implies
f (z) − f (w)
− f ′ (w) < ϵ.
z−w
Multiplying by |z − w| and verifying the case z = w, we conclude that for
all z ∈ Dδ (w),
|f (z) − f (w) − f ′ (w)(z − w)| ≤ ϵ|z − w|.
For all n large enough, diam cvh(γn ) < δ. For such n, for z on the curve γn ,
we have
|f (z) − f (w) − f ′ (w)(z − w)| ≤ ϵ diam cvh(γn )
so by integrating over γn we conclude
I
f (z) − f (w) − f ′ (w)(z − w) dz ≤ ϵℓ(γn ) diam cvh(γn ).


γn

Since integrals of polynomials over closed curves are zero (Corollary 4.13),
I
f (w) + f ′ (w)(z − w) dz = 0


γn
so the previous inequality simplifies to
I
f (z) dz ≤ ϵℓ(γn ) diam cvh(γn ).
γn

Combining this with the known (in)equalities for γn we conclude

I
1 ℓ(γ) diam cvh(γ)
n
f (z) dz ≤ ϵ n .
4 γ 2 2n
Now the exponentials cancel and we obtain an inequality independent of n,
I
f (z) dz ≤ ϵℓ(γ) diam cvh(γ).
γ
H
Since this holds for all ϵ > 0, we conclude γ f (z) dz = 0. □
It is notable that this theorem requires only holomorphicity and doesn’t
require continuity of the derivative. This is because the proof only uses holo-
morphicity at one point. This can be compared to the mean value theorem
from real analysis, and it is equally fundamental. Goursat’s theorem is the
first glimpse that a differential property of f implies an integral property of
f.
70 4. THE CAUCHY INTEGRAL THEOREM

4. Cauchy integral theorem on star-shaped domains

Definition 4.19. A domain Ω is said to be star-shaped if it has the
following property: there exists z∗ ∈ Ω such that for all z ∈ Ω, the line
segment [z∗ z] is contained in Ω. In this case z∗ is said to be a hub for Ω.
Obviously all convex sets are star-shaped. The set C \ (−∞, 0] is also
star-shaped, with hub z∗ = 1.
In a star-shaped domain, there is a natural way to construct an anti-
derivative; we state that construction in its full generality since we will need
it one more time in this text.
Lemma 4.20. If f is continuous on a star-shaped domain Ω and
I
f (z)dz = 0
γ
for all triangles γ with cvh(γ) ⊂ Ω, then f has an antiderivative on Ω.
Proof. Define for z ∈ Ω
I
g(z) = f (ζ) dζ.
[z∗ z]
Fix z1 ∈ Ω and δ > 0 such that Dδ (z1 ) ⊂ Ω. For z2 ∈ Dδ (z1 ), [z1 z2 ] ⊂ Ω.
For any w on the line segment [z1 z2 ], since z∗ is a hub of Ω, [z∗ w] ⊂ Ω.
Thus, cvh[z∗ z1 z2 ] ⊂ Ω. By the Goursat theorem,
I
g(z2 ) − g(z1 ) = f (ζ) dζ
[z1 z2 ]
so by Lemma 3.24, g is holomorphic at z1 and g ′ (z1 ) = f (z1 ). □
Combining this with Goursat’s theorem and Theorem 4.15, we obtain
the Cauchy integral theorem for star-shaped domains:
Theorem 4.21 (Cauchy Integral Theorem for star-shaped domains). On
a star-shaped domain, every holomorphic function has an antiderivative.
In particular, if f is holomorphic on some star-shaped domain Ω, then
for every closed piecewise C 1 curve γ in Ω,
I
f (z)dz = 0.
γ

5. Cauchy integral theorem on simply connected domains

In this discussion, it will be convenient to assume all curves are parametrized
by t ∈ [0, 1]. Any curve γ : [a, b] → C can be reparametrized this way by
the linear reparametrization γ̃(t) = γ(a + t(b − a)).
Definition 4.22. For closed curves γ0 , γ1 : [0, 1] → Ω, a loop homotopy
in Ω is a continuous F : [0, 1]2 → Ω with the properties
(a) the initial curve is γ0 : F (t, 0) = γ0 (t) for all t
(b) the final curve is γ1 : F (t, 1) = γ1 (t) for all t
 5. CAUCHY INTEGRAL THEOREM ON SIMPLY CONNECTED DOMAINS 71

γ0 γ1

Figure 4.1. Left: a loop homotopy between two closed

curves is a continuous family of closed curves. Center: a
grid of points used in the proof of the Cauchy integral theo-
rem. Right: replacing the intermediate curves by polygonal
curves to ensure their rectifiability

(c) each curve along the homotopy is closed: F (0, s) = F (1, s) for all s
If such a homotopy exists, γ0 , γ1 are said to be loop homotopic in Ω.

Loop homotopy is an equivalence relation (Exercise 4) and curves in

the same contour are loop homotopic (Exercise 5); thus, the binary relation
of being loop homotopic can also be viewed as a binary relation between
contours.

Theorem 4.23. If rectifiable closed curves γ0 , γ1 are loop homotopic in

a domain Ω, then for any f holomorphic on Ω,
I I
f (z) dz = f (z) dz. (4.1)
γ0 γ1

Proof. Denote a loop homotopy by F : [0, 1]2 → Ω. Since Ran F is

compact, we denote
r = dist(Ran F, C \ Ω) > 0.
By uniform continuity of F , there exists n ∈ N such that ∥(t1 , s1 )−(t2 , s2 )∥ <
2/n implies |F (t1 , s1 )−F (t2 , s2 )| < r. In particular, the image of each square
[k/n, (k + 1)/n] × [l/n, (l + 1)/n] is contained in a disk Dr (F (k/n, l/n)) ⊂ Ω.
The boundary of each square is mapped by F to a closed curve which
may not be rectifiable. Thus, instead of using the map F , we keep only the
grid of points zk,l = F (k/n, l/n) and connect these vertices by line segments,
except those coming from γ0 , γ1 . In particular, we consider curves

γ0 |[k/n,(k+1)/n] l = 0

γk,l = [zk,l zk+1,l ] 1≤l ≤n−1

γ1 |[k/n,(k+1)/n] l = n

72 4. THE CAUCHY INTEGRAL THEOREM

←
Each curve [zk,l+1 zk,l ] ∗ γk,l+1 ∗ [zk+1,l zk+1,l+1 ] ∗ γk,l is closed, rectifiable, and
lies inside the disk Dr (zk,l ) ⊂ Ω, so by Theorem 4.21,
I I I I
f (z) dz + f (z) dz = f (z) dz + f (z) dz.
[zk+1,l zk+1,l+1 ] γk,l [zk,l zk,l+1 ] γk,l+1
Summing in k = 0, 1, . . . , n − 1 and canceling the line segments which occur
on both sides, including the line segment [z0,l z0,l+1 ] = [zn,l zn,l+1 ], we obtain
n−1
XI n−1
XI
f (z) dz = f (z) dz.
k=0 γk,l k=0 γk,l+1

Moreover, since z0,l = z0,l+1 = zn,l = zn,l+1 = z∗ , this simplifies to

n−1
XI n−1
XI
f (z) dz = f (z) dz.
k=0 γk,l k=0 γk,l+1

Iterating this in l = 0, 1, . . . , n − 1, we obtain

n−1
XI n−1
XI
f (z) dz = f (z) dz
k=0 γk,0 k=0 γk,n

which is precisely (4.1). □

Definition 4.24. The domain Ω is (topologically) simply connected if
every closed curve is loop-homotopic in Ω to a constant curve.
Sometimes simple connectedness is defined in terms of basepoint-preserving
homotopy; this is equivalent to our definition (Exercise 6).
Example 4.25. Any star-shaped domain is simply connected, with the
homotopy F (t, s) = (1 − s)γ(t) + sz∗ where z∗ is a hub and γ an arbitrary
closed curve.
Example 4.26. C \ {0} is not simply connected.
Proof. The function 1/z is holomorphic on C \ {0}, and we have pre-
viously computed I
1
dz = 2πi.
|z|=1 z
Thus, the unit circle γ(t) = eit , t ∈ [0, 2π] is not loop homotopic to a
constant curve. Thus, C \ {0} is not simply connected. □
Corollary 4.27 (Cauchy integral theorem on simply connected do-
mains). If f is holomorphic on a simply connected domain Ω and γ a closed
rectifiable curve on Ω, then
I
f (z)dz = 0.
γ

Proof. This follows from Theorem 4.23, since contour integrals over a
constant curve are zero. □
 6. BRANCHES OF THE LOGARITHM AND ROOTS 73

On a simply connected domain, it will sometimes be convenient to use

notation motivated by Riemann integrals:
Definition 4.28. For a holomorphic function f on a simply connected
domain Ω, we will denote by
Z z
f (ζ) dζ
z0

the value of γ f (ζ) dζ for some (and therefore every) piecewise C 1 curve γ
H

with starting point z0 and endpoint z.

Some texts make the following definition:
Definition 4.29. A domain Ω is called holomorphically simply con-
nected (hsc) if for every closed piecewise C 1 curve γ in Ω and every holo-
morphic function f on Ω, I
f (z)dz = 0.
γ

We have proved that simply connected domains are hsc. Later in this
course, we will prove as a consequence of the Riemann mapping theorem
that the opposite is also true: every hsc domain is simply connected. Even
later, we will prove that a domain is hsc if and only if it has no ”holes”, where
a hole is defined as a bounded connected component of its complement. For
instance, the domain C \ {0} has a hole {0}.

6. Branches of the logarithm and roots

For any function g : Ω → C, the function f = eg is nonzero. If g is
continuous or holomorphic, then so is f . Going in the opposite direction is
harder: if f is nonzero, then for any z, there exist values of g(z) such that
eg(z) = f (z), but they are not unique and there is no guarantee that we
can choose these values in a way that makes g continuous or holomorphic.
However, on a simply connected domain, this can be achieved:
Theorem 4.30. Let Ω be simply connected and f holomorphic on Ω
with f ′ holomorphic. Suppose f (z) ̸= 0 for all z ∈ Ω. Then there exists g
holomorphic on Ω such that eg = f . Any other continuous function G such
that eG = f is of the form G(z) = g(z) + 2πik for some k ∈ Z independent
of z.
Proof. Pick z0 ∈ Ω. Since f (z0 ) ̸= 0, there exists g0 ∈ C such that
eg0 = f (z0 ) (Theorem 2.27). It follows from our assumptions that f ′ /f is
holomorphic, so there exists g such that g ′ = f ′ /f on Ω and g(z0 ) = g0 . In
other words, we take
Z z ′
f (w)
g(z) = g(z0 ) + dw.
z0 f (w)
74 4. THE CAUCHY INTEGRAL THEOREM

Consider the function h = f e−g . This is holomorphic and a direct calculation

gives
h′ = f ′ e−g − f e−g g ′ = f g ′ e−g − f g ′ e−g = 0
so h is constant. Moreover, h(z0 ) = f (z0 )e−g(z0 ) = 1, so h is identically 1
and f = eg .
For any other function G with the same properties, eg = eG implies
e G−g = 1 which implies G(z) − g(z) ∈ 2πiZ. Since G − g is a continuous
function which takes values only in the discrete set 2πiZ, it must be a
constant function equal to 2πik for some k ∈ Z. □
Definition 4.31. For f : Ω → C holomorphic, a “branch of log f ” is a
holomorphic function g : Ω → C such that eg = f .
In particular, in the special case f (z) = z, a “branch of log on Ω” is a
holomorphic function g : Ω → C such that eg(z) = z for all z ∈ Ω.
The special case f (z) = z of the previous theorem gives:
Corollary 4.32. On any simply connected domain Ω such that 0 ∈ / Ω,
there is a holomorphic function log : Ω → C such that elog w = w for all
w ∈ Ω.
Different choices of this function are called different branches of the
logarithm. On Ω, any two branches differ simply by an additive term in
2πiZ.
Most commonly used are cut planes Ωϕ = C \ {reiϕ | r ≥ 0}. These are
simply connected because they are star-shaped (the point −eiϕ is a hub).
Example 4.33 (Branches of logarithm on cut planes). For every ϕ ∈ R,
there is a branch of logarithm on Ωϕ such that
ϕ < Im log < ϕ + 2π.
Moreover, for any r > 0,
lim log z = log r + i(ϕ + 2π) (4.2)
z→reiϕ
Im(z/eiϕ )>0

and
lim log z = log r + iϕ. (4.3)
z→reiϕ
Im(z/eiϕ )<0

Proof. If Im log z ∈ ϕ + 2πZ, this implies

z = eRe z+i Im z = eRe z eiϕ ∈
/ Ωϕ .
By contraposition, on Ωϕ , Im log z ∈
/ ϕ + 2πZ.
Since Ωϕ is simply connected and
ei(ϕ+π) = −eiϕ ,
there exists a branch of logarithm on Ωϕ with log(−eiϕ ) = i(ϕ + π). By
continuity, we conclude ϕ < Im log < ϕ + 2π on Ωϕ .
 6. BRANCHES OF THE LOGARITHM AND ROOTS 75

The branches for Ωϕ and Ωϕ+π coincide on the half-plane Im(z/eiϕ ) > 0.
Using the second of those branches, we conclude (4.2). Similarly, comparing
Ωϕ and Ωϕ−π , we conclude (4.3). □
Example 4.34. The branch of logarithm on C \ (−∞, 0] with log 1 = 0
has a power series representation around z0 = 1 given by
∞
X (−1)n−1
log z = (z − 1)n , ∀z ∈ D1 (1). (4.4)
n
n=1

Proof. The radius of convergence is unaffected by the factor 1/n so the

Cauchy radius formula gives radius of convergence R = 1. Term by term
differentiation of
∞
X (−1)n−1
f (z) = (z − 1)n
n
n=1
gives
∞
′
X 1 1
f (z) = (−1)n−1 (z − 1)n−1 = =
1+z−1 z
n=1
so f (z) − log z has zero derivative on D1 (1). By a consequence of the fun-
damental theorem of calculus, this shows that f (z) − log z is a constant
function.
Since f (1) = 0 = log 1, that constant is zero, so log z = f (z) on D1 (1).
□
Corollary 4.35. For any t ∈ (−1, 1),
∞
X (−1)k t2k+1
Im log(1 + it) = .
2k + 1
k=0
In particular, the value of π is given by an alternating series,
∞ √
X
k 2 3
π= (−1) .
(2k + 1)3k
k=0

Proof. Plugging in z = 1 + it into (4.4) and taking imaginary parts

gives
∞
X (−1)k t2k+1
Im log(1 + it) = .
2k + 1
k=0
Since √ !
i 2 3 i 2
1+ √ = √ + = √ eiπ/6 ,
3 3 2 2 3
√
we read off Im log(1 + i/ 3) = π/6 and conclude
∞
π X (−1)k
= √ . □
6 (2k + 1)3k 3
k=0
76 4. THE CAUCHY INTEGRAL THEOREM

There exist other

√ alternating series representations for π with rational
terms (without 3, in particular) and with faster convergence (see Exer-
cises).
The branch of the logarithm on C \ (−∞, 0] with −π < Im log < π is
sometimes called the principal value of the logarithm. However, this is only
a matter of convention; in this course, we will avoid such conventions and
whenever we use a logarithm, we will always specify the branch.
It is common to denote
arg w = Im log w.
This is called the argument of w and it is the polar angle we saw before.
Since |ez | = eRe z we have Re log w = log|w|, so log can also be written in
the form
log w = log|w| + i arg w.
For instance, the branch on Ωπ mentioned before is described by saying
arg 1 = 0 or log 1 = 0 or −π < arg < π.
Regardless of the branch, we have to be careful about further algebraic
properties we may be used to. For instance, for the branch of logarithm
mentioned above, using a third root of unity ω = e2πi/3 ,
2πi
log ω = log e2πi/3 =
3
but
2πi
log(ω 2 ) = log(ω −1 ) = log e−2πi/3 = −
3
so in particular
log(ω 2 ) ̸= 2 log ω.
Thus, the property log(w1 w2 ) = log w1 + log w2 cannot hold everywhere.
Instead, we only know in general that
log(w1 w2 ) ∈ log w1 + log w2 + 2πiZ.
We can use the following link to draw some pictures with logarithms:
[Link]
Similarly to the above discussion, we obtain holomorphic n-th roots:
Corollary 4.36. Let Ω be simply connected and f holomorphic on Ω
with f ′ holomorphic. Suppose f (z) ̸= 0 for all z ∈ Ω. Then for any n ∈ N
there exists holomorphic hn on Ω such that (hn )n = f . Any other holomor-
phic function Hn such that (Hn )n = f obeys Hn (z) = ωhn (z) for some n-th
root of unity ω independent of z.
Proof. There exists holomorphic g such that eg = f . For any n, the
function hn = eg/n has the correct properties. Moreover, if Hn has the same
properties, then (hn /Hn )n = 1 on Ω. Thus, hn /Hn only takes values which
are n-th roots of unity. By continuity of hn /Hn , it follows that hn /Hn is a
constant function equal to an n-th root of unity. □
 7. EXERCISES 77

For instance, on the star-shaped domain C \ (−∞, 0], there is a holomor-

phic n-th root function hn (z) = z 1/n with hn (1) = 1.
Example 4.37. Fix a branch of log on a simply connected domain Ω ⊂
C \ {0}. For all α ∈ R, the functions

z α = eα log z (4.5)
are holomorphic on Ω and obey:
(a) z α z β = z α+β for all α, β ∈ R
(b) z 0 = 1, z 1 = z, and more generally, for any n ∈ Z, z n matches the
standard notation for n-th powers
(c) (z α )′ = αz α−1
Proof. (a) Follows from eu ev = eu+v .
(b) We directly obtain z 1 = elog z = z and z 0 = e0 = 1. By (a), for n ∈ N,
inductively, z n = z n−1 z so this must match the old definition. Similarly use
z −n z n = z 0 = 1 to conclude definitions match for negative integers.
(c) Differentiating (4.5) gives
 ′ α
α ′
(z ) = e α log z
= eα log z (α log z)′ = z α = αz α−1 .
z
Note that the last step uses z α /z = z α−1 by (a) and (b). □

7. Exercises
1. Assume that A1 , B1 are separated sets and that A2 , B2 are separated sets.
Prove that A1 ∩ A2 and B1 ∪ B2 are separated sets.
2. If E ⊂ C is connected and E ⊂ F ⊂ E, prove that F is connected.
3. (a) Prove that there is no continuous bijection γ : [0, 1] → D or γ :
[0, 1] → Ĉ.
Hint: prove that γ −1 would be continuous and consider connectedness
of [0, 1] \ {1/2} and its image.
(b) Prove that there is no continuous bijection γ : ∂D → D or γ : ∂D → Ĉ.
4. Denote γ0 ∼ γ1 if and only if γ0 , γ1 are loop homotopic in Ω. Prove that
∼ is an equivalence relation.
5. Prove that if two curves are in the same contour, they are loop homotopic.
6. For a closed curve γ : [0, 1] → Ω, the point γ(0) = γ(1) is sometimes
called the basepoint. A loop homotopy F : [0, 1]2 → Ω in which the
basepoint is preserved along the homotopy, i.e. F (0, s) is independent of
s, is called a basepoint-preserving homotopy of closed curves. If γ0 , γ1
have the same basepoint and are loop homotopic, prove that they have a
basepoint-preserving homotopy.
7. Assume that z1 , z2 , z3 are not colinear and let γ = [z1 z2 z3 ] and w ∈
cvh(γ) \ γ. Prove that any line through w intersects γ at precisely two
points w1 , w2 , so that w is between w1 , w2 .
 78 4. THE CAUCHY INTEGRAL THEOREM

∈ C can uniquely
Hint: one approach is toPprove the fact that every z P
be written in the form z = 3j=1 cj zj where cj ∈ R and 3j=1 cj = 1, and
use this to pull the problem to the plane {c ∈ R3 | c1 + c2 + c3 = 1}.
8. Consider the branch of logarithm on C \ (−∞, 0] with log 1 = 0.
(a) Find the power series representation for log z around an arbitrary
point z0 ∈ C \ (−∞, 0]. What is the radius of convergence of that
power series?
(b) Can the disk of convergence from (b) intersect the half-line (−∞, 0],
and if so, what do you think happens beyond that half-line (what do
the values of the power series represent)?
9. Use the identity   
i i 5 5
1+ 1+ = + i
2 3 6 6
to prove that
∞
π X (−1)k
 
1 1
= + .
4 2k + 1 22k+1 32k+1
k=0

Note: the identity is often expressed as arctan 12 + arctan 13 = 1.

10. Prove that
∞
π X (−1)k
 
4 1
= − .
4 2k + 1 52k+1 2392k+1
k=0
11. Prove that
√ there is no holomorphic√square root on C \ {0}, i.e. no holo-
2
morphic z on C \ {0} such that ( z) = z for all z ∈ Cp\ {0}.
12. Let a < b. Prove that there exists a branch of square root (z − a)(z − b)
on C \ [a, b]. √
13. Prove that there is no branch of square root √z on D.
14. Prove that there is no branch of square root sin pz on D.
15. Prove that there exists a branch of square root sin(z 2 ) on D.
 CHAPTER 5

Cauchy integral formula and analytic functions

We will begin this chapter by proving the Cauchy integral formula for the
disk. This is a remarkable formula with no analog in real analysis; it allows
us to express values of a holomorphic function inside a disk by an integral
along the boundary of the disk. We then study the numerous consequences
of this integral representation.

1. Cauchy integral formula for the disk

As a warm-up, consider the following. Let Ω be a domain which contains
Dr (z0 ). Since Dr (z0 ) is compact, there is a positive distance δ from Dr (z0 )
to C \ Ω. In particular, DR (z0 ) ⊂ Ω for R = r + δ > r. Thus, for any f
holomorphic on Ω, by applying the Cauchy Integral Theorem on the star-
shaped domain DR (z0 ),
I
f (ζ) dζ = 0.
|ζ−z0 |=r

Note that it doesn’t matter whether the domain Ω is star-shaped or simply

connected, since we can pass to a subdomain with these properties.
Compare the above application of the Cauchy integral theorem with the
following statement:
Theorem 5.1 (Cauchy Integral Formula on a disk). Let Ω be a domain
which contains Dr (z0 ). Let z ∈ Dr (z0 ). For any f holomorphic on Ω,
I
f (ζ)
dζ = 2πif (z).
|ζ−z0 |=r ζ − z

This statement is not incompatible with the Cauchy integral theorem,

because the ratio fζ−z
(ζ)
is only holomorphic on Ω\{z}. One technicality of the
proof is precisely in working with this punctured domain Ω \ {z}. We may
pass to a punctured disk DR (z0 ) \ {z}, but this is also not simply connected.
The first part of the proof is geometric and consists of moving the contour
integral to a smaller circle centered at z. We separate this as a lemma and
provide two proofs:
Lemma 5.2. In the setting of Theorem 5.1, for every 0 < ρ ≤ r −|z −z0 |,
I I
f (ζ) f (ζ)
dζ = dζ. (5.1)
|ζ−z0 |=r ζ − z |ζ−z|=ρ ζ − z
79
80 5. CAUCHY INTEGRAL FORMULA AND ANALYTIC FUNCTIONS

σ+ v+
τ+
u+
z

z0 u− τ−

v− σ−

Figure 5.1. Contours used in the first proof of the lemma

First proof of lemma, using only star-shaped domains. Denote

ϕ = arg(z − z0 ) for some choice of argument; if z = z0 , fix any ϕ ∈ R.
On the circle |ζ − z| = ρ, observe the points
u± = z ± ρeiϕ
and on the circle |ζ − z0 | = r, observe the points
v± = z0 ± reiϕ .
Consider the semicircle contours
σ± (t) = z0 ± rei(ϕ+t) , t ∈ [0, π]
and
τ± (t) = z ± ρei(ϕ+π−t) , t ∈ [0, π]
and the closed contours
γ± = τ± ∗ [v∓ u∓ ] ∗ σ± ∗ [u± v± ].
The curve γ± lies in the domain
Ω± = DR (z0 ) \ {z ∓ tieiϕ | t ∈ [0, ∞)}
which is star-shaped with hub z + ρeiϕ . Since the function ζ 7→ f (ζ)/(ζ − z)
is holomorphic in Ω± ,
I
f (ζ)
dζ = 0.
γ± ζ −z
Summing the two integrals, the contributions from the line segments cancel
and we obtain (5.1). □
 1. CAUCHY INTEGRAL FORMULA FOR THE DISK 81

Second proof of lemma, using loop homotopy. By linearly inter-

polating the center and the radius of the circle (see Figure 4.1), we produce
the loop homotopy

F (t, s) = z0 + s(z − z0 ) + [r + s(ρ − r)]e2πit .

It seems geometrically obvious that this is a loop homotopy in DR (z0 ) \ {z},

and we justify that as follows. Since F (t, s) is a convex combination of points
F (t, 0) = z0 + re2πit and F (t, 1) = z + ρe2πit ,

|F (t, s) − z0 | ≤ max{|F (t, 0) − z0 |, |F (t, 1) − z1 |} ≤ r

since |z − z0 | + ρ ≤ r. Thus, F (t, s) ∈ Dr (z0 ). We also have to check that

F does not take the value z. If it did, z would be in the line segment with
endpoints F (t, 0) and F (t, 1), or equivalently, (F (t, 1) − z)(F (t, 0) − z) ≤ 0.
This simplifies to
ρe−2πit (z0 + re2πit − z) ≤ 0
and since ρ > 0, it is equivalent to

r + e−2πit (z0 − z) ≤ 0.

This is impossible, since |z − z0 | < r.

Since the function ζ 7→ f (ζ)/(ζ − z) is holomorphic in DR (z0 ) \ {z}, by
Theorem 4.23, (5.1) follows. □

Remark 5.3. The previous proof can be improved as follows. Since

F (t, s) intersects

Proof of the Cauchy integral formula on a disk. We compute

I I
f (z) 1
dζ = f (z) dζ = 2πif (z),
|ζ−z|=ρ ζ − z |ζ−z|=ρ ζ − z

where the final step is an easy calculation with γ(t) = z + ρe2πit , t ∈ [0, 1].
Thus,

f (ζ) − f (z) f (ζ) − f (z)

I
dζ ≤ 2πρ sup
|ζ−z|=ρ ζ −z ζ:|ζ−z|=ρ ζ −z
= 2π sup |f (ζ) − f (z)|
ζ:|ζ−z|=ρ

and this goes to 0 as ρ ↓ 0. Combining these statements with (5.1) concludes

the proof. □

This formula is of tremendous importance in complex analysis. It will

allow us to describe local properties of holomorphic functions.
82 5. CAUCHY INTEGRAL FORMULA AND ANALYTIC FUNCTIONS

2. Cauchy power series theorem

The Cauchy integral formula represents the values of a holomorphic func-
tion f inside the disk Dr (z0 ) in terms of values on its boundary:
I
1 f (ζ)
f (z) = dζ.
2πi |ζ−z0 |=r ζ − z
This indicates that holomorphicity is a surprisingly rigid notion. It can also
be used to study the z-dependence more closely. On the right-hand side,
the integral kernel fζ−z
(ζ)
is not only holomorphic in z, it can be expanded as
a power series around z0 , since
∞
X (z − z0 )n
1 1 1
= = , z ∈ Dr (z0 ).
ζ −z ζ − z0 1 − z−z
ζ−z0
0 (ζ − z0 )n+1
n=0
Plugging this into the Cauchy integral formula and exchanging the integral
and the series seems to give a power series expansion for f on Dr (z0 ). Its
convergence and the exchange of integral and series has to be justified, which
leads to the following lemma:
Lemma 5.4. If f is holomorphic in Ω and Dr (z0 ) ⊂ Ω, then with the
coefficients I
1 f (ζ)
an = dζ (5.2)
2πi |ζ−z0 |=r (ζ − z0 )n+1
we have
∞
X
f (z) = an (z − z0 )n (5.3)
n=0
for all z ∈ Dr (z0 ); in particular, this power series has radius of convergence
at least r.
Proof. For z ∈ Dr (z0 ), compare f (z) to a partial sum:
N −1 N −1
!
1
I
f (ζ) f (ζ)(z − z )n
0
X X
f (z) − an (z − z0 )n = − dζ
2πi |ζ−z0 |=r ζ − z (ζ − z0 )n+1
n=0 n=0
Computing the geometric sum gives
N −1  N
z − z0
I
X
n 1 f (ζ)
f (z) − an (z − z0 ) = dζ.
2πi |ζ−z0 |=r ζ −z ζ − z0
n=0
By standard integral inequalities this gives
N −1
X 1 |z − z0 |N f (ζ)
f (z) − an (z − z0 )n ≤ 2πr sup .
2π rN ζ:|ζ−z0 |=r ζ −z
n=0
Since the function f (ζ)/(ζ − z) is continuous on ∂Dr (z0 ), the supremum is
finite and independent of N . Since |z − z0 | < r, as N → ∞, we obtain (5.4).
Thus, the function is given by a convergent power series on Dr (z0 ). □
 2. CAUCHY POWER SERIES THEOREM 83

Definition 5.5. A function f : Ω → C is analytic if at every point

z0 ∈ Ω, it has a power series representation centered at z0 with a nonzero
radius of convergence.
The above lemma implies:
Theorem 5.6 (Fundamental theorem of complex analysis). A function
f : Ω → C is holomorphic if and only if it is analytic.
Proof. We have already seen that power series are holomorphic, so
analytic functions are holomorphic. Conversely, any holomorphic function
is analytic by Lemma 5.4. □
Corollary 5.7. The derivative of a holomorphic function is holomor-
phic.
Proof. If f is holomorphic, it is analytic, so f ′ is analytic, so f ′ is
holomorphic. □
Remark 5.8. This corollary immediately improves the wording of some
of the earlier theorems. For instance:
(a) in the fundamental theorem of calculus, we assumed that f is holomor-
phic and f ′ continuous. Now we know that it suffices to assume f is
holomorphic.
(b) to prove existence of a branch of log f , we assumed that f was holomor-
phic with f ′ holomorphic. Now we know that it suffices to assume f is
holomorphic.
Continuing this principle, we can differentiate n times and express an in
terms of the n-th derivative of f . We can also optimize the radius:
Theorem 5.9 (Cauchy power series theorem). If f is holomorphic in Ω
and DR (z0 ) ⊂ Ω, then f has a convergent power series representation
∞
X f (n) (z0 )
f (z) = (z − z0 )n (5.4)
n!
n=0
for all z ∈ DR (z0 ). In particular, the radius of convergence of this series is
at least R.
Proof. For any r < R, we have a convergent power series representation
∞
X
f (z) = an (z − z0 )n , z ∈ Dr (z0 ).
n=0
Term by term differentiation, applied k times, implies that
∞
X
f (k) (z0 ) = n(n − 1) . . . (n − k + 1)an (z − z0 )n−k , z ∈ Dr (z0 ).
n=k

Evaluating at z = z0 , we conclude f (k) (z0 ) = k!ak for any k. In particular,

the coefficients ak do not depend on r < R, and we conclude that (5.4) holds
on ∪r<R Dr (z0 ) = DR (z0 ). □
84 5. CAUCHY INTEGRAL FORMULA AND ANALYTIC FUNCTIONS

3. Applications of the Cauchy power series theorem

Examples. An important aspect of the Cauchy power series theorem is
that the radius of convergence is at least R if DR (z0 ) ⊂ Ω. This often allows
us to read off the radius of convergence with very little explicit calculation,
as in the following examples.
Recall that the power series representation centered at z0 = 0 is often
called the Maclaurin series.
Example 5.10. The Maclaurin series of the rational function
1
f (z) = 2
z +1
has radius of convergence R = 1.
Proof. As a function on C, if we factorize z 2 + 1 = (z + i)(z − i), we
see that f is holomorphic on Ω = C \ {i, −i}. Thus, D ⊂ Ω, so f has a
Maclaurin series
∞
X
f (z) = cn z n
n=0
around 0 with radius R ≥ 1.
If this power series had radius of convergence R > 1, the power series
would define an analytic function on DR (0), and in particular, the limit
∞
X ∞
X
lim f (z) = lim cn (iy)n = cn in
z=iy z=iy
y↑1 y↑1 n=0 n=0

would be convergent. However,

1
lim |f (z)| = =∞
z=iy limz=iy |z 2 + 1|
y↑1 y↑1

which gives a contradiction. Thus, the radius is R = 1. □

Second proof. Of course, in this case, we can compute the power

series directly as
∞
1 X
f (z) = = (−z 2 )n
1 + z2
n=0
and use the root test to determine R = 1. □

We caution the reader that some special consideration is still needed to

ensure that the radius of convergence is not larger than that suggested by
the geometry of the domain:
Example 5.11. For a branch of log on Ω = C\(−∞, 0], the Taylor series
centered at z0 has radius of convergence R = |z0 |, even though if Re z0 < 0,
D|z0 | (z0 ) ̸⊂ Ω.
 3. APPLICATIONS OF THE CAUCHY POWER SERIES THEOREM 85

Figure 5.2. For a Taylor series for a branch of log, the disk
of convergence of a Taylor series is not determined by the
choice of branch, and may extend beyond the domain of the
branch

Proof. Since
∞  ∞
z − z0 n X (−1)n

1 1 1 1 X n
= = − = n+1 (z − z0 )
z z0 1 + z−z
z
0 z 0 z 0 z0
0 n=0 n=0

has radius of convergence R = |z0 |, its antiderivative log has the same radius
of convergence. □

We can understand this more geometrically if we restate what the Cauchy

power series theorem gives us. Recall that we use notation g|D to denote
the restriction of a function g to the domain D.
Definition 5.12. Let Ω1 , Ω2 be domains with Ω1 ⊂ Ω2 . If f is analytic
on Ω1 , g is analytic on Ω2 , and g|Ω1 = f , then g is said to be an analytic
continuation of f to Ω2 .
Newly added
Lemma 5.13. Let f be holomorphic on Ω and z0 ∈ Ω. Fix some r > 0 lemma to clarify
with Dr (z0 ) ⊂ Ω. The radius of convergence of the power series at z0 is the this point
largest R such that the restriction of f to the disk Dr (z0 ) has an analytic
continuation to DR (z0 ). We understand the case R = ∞ by the convention
D∞ (z0 ) = C.
Proof. Denote by R the radius of convergence. The power series gives
an analytic continuation to DR (z0 ). Conversely, if f has an analytic contin-
uation to some Dρ (z0 ), then by the Cauchy power series theorem, the radius
of convergence is at least ρ, so ρ ≤ R. □

In the example with log, the restriction of our branch of log has a holo-
morphic extension to another branch of log, on a different cut plane.
The Cauchy power series theorem is also a powerful calculus tool. In
real analysis, obtaining a power series representation of a function entails
comparing a function to its Taylor polynomials by a Taylor remainder es-
timate. For a function whose Taylor coefficients don’t decay exponentially,
this can be effortful, even for a simple function like (1 + x)α . The Cauchy
power series theorem takes care of the remainder estimates automatically,
86 5. CAUCHY INTEGRAL FORMULA AND ANALYTIC FUNCTIONS

and we just need to obtain all the coefficients by repeatedly differentiating

and read off (a lower bound for) the radius of convergence from the domain:
Example 5.14 (Binomial series). Fix the branch of log on C \ (−∞, 0]
with log 1 = 0 and recall the holomorphic functions
z α = eα log z
for α ∈ R.
(a) On (0, ∞), these functions match the real analysis definition of xα .
(b) The function z α has a power series representation around z0 = 1 given
by
∞  
α
X α
z = (z − 1)n , z ∈ D1 (1)
n
n=0
where the binomial coefficients are defined for α ∈ R, n ∈ N0 by
 
α α(α − 1) . . . (α − n + 1)
= (5.5)
n n!
(c) The power series (5.5) has radius of convergence R = ∞ if α ∈ N0 and
R = 1 if α ∈ R \ N0 .
Proof. (a) The real analysis definition obeys xα = eα log x and our log
matches the log from real analysis on (0, ∞).
(b) Differentiating z α gives
α
(z α )′ = (eα log z )′ = eα log z = αz α−1 .
z
Iterating this n times gives
(z α )(n) = α(α − 1) . . . (α − n + 1)z α−n .
Evaluating at z = 1 gives coefficients of the power series expansion. Holo-
morphicity on C \ (−∞, 0] shows that the radius of convergence is R ≥ 1.
(c) If α ∈ N0 , the function f (z) = z n is analytic on C, so R = ∞.
Otherwise, let us assume R > 1; then the power series would define an
analytic function
∞  
X α
f (z) = (z − 1)n
n
n=0
on the larger disk DR (1). Then f (z) = z α on D1 (1) so for k ∈ N0 ,
f (k) (z) = α(α − 1) . . . (α − k + 1)z α−k .
In particular, for k > α,
lim|f (k) (x)| = ∞
x↓0

which would contradict analyticity of f and its derivatives on DR (1). □

 3. APPLICATIONS OF THE CAUCHY POWER SERIES THEOREM 87

Cauchy estimates and consequences. Recall the Cauchy formulas

I
1 f (z)
an = dz
2πi |z−z0 |=r (z − z0 )n+1

Applying the basic integral estimate to this gives:

Theorem 5.15 (Cauchy estimates). If f is holomorphic on a domain

containing Dr (z0 ) and ∞ n
P
n=0 an (z − z0 ) is its power series expansion at z0 ,
then
Z 2π
−n dθ
|an | ≤ r |f (z0 + reiθ )| ≤ r−n max |f (ζ)|.
0 2π ζ:|ζ−z0 |=r

This is notable because it gives estimates for f and its derivatives in

terms of estimates on f elsewhere. For instance:

Example 5.16. If f is analytic on D, say f (z) = ∞ n

P
n=0 an z and |f (z)| <
1 everywhere, then |an | ≤ 1 for each n. For each n, this inequality is optimal.

Proof. The Cauchy estimate for f in Dr (0) gives |an | ≤ r−n . We can
conclude this for any r < 1, so letting r ↑ 1 optimizes the bound in r and
gives |an | ≤ 1.
The inequalities are optimal because for any n, the function f (z) = z n
gives a case of equality. □

Definition 5.17. Holomorphic functions on Ω = C are called entire

functions.

Theorem 5.18 (Liouville). Any bounded entire function is constant.

Proof. By the Cauchy power series theorem, an entire function must

have a Maclaurin series with infinite radius of convergence. For any n ≥ 1,
taking r → ∞ in the Cauchy estimates gives an = 0. Thus, f (z) = a0 for
all z ∈ C. □

Corollary 5.19 (Fundamental theorem of algebra). Every polynomial

of degree at least 1 with complex coefficients has a zero.

Proof. Let p(z) = nk=0 ak z k be a polynomial of degree n, i.e., n ≥ 1

P
and an ̸= 0. Then
p(z)
lim = an
z→∞ z n

and limz→∞ |z n | = ∞, so limz→∞ |p(z)| = ∞.

Assume that p has no zeros in C. Then f = 1/p is an entire function
and limz→∞ |f (z)| = 0. Thus, f is bounded, so constant, so f is identically
zero. This is a contradiction. □
88 5. CAUCHY INTEGRAL FORMULA AND ANALYTIC FUNCTIONS

A computational comment.
5.20 (Product of power series). If two power series ∞ n
P
P∞Theorem n=0 an z ,
n
n=0 bn z have radii of convergence at least R, then the power series with
coefficients
Xn
cn = ak bn−k ,
k=0
also has radius of convergence at least R and
∞ ∞
! ∞ !
X X X
cn z n = an z n bn z n , z ∈ DR (0).
n=0 n=0 n=0

Proof. The two power series define analytic functions f, g : DR (0) → C,

so their product is also analytic on DR (0). From
n  
X n (k)
(f g)(n) (z) = f (z)g (n−k) (z),
k
k=0
the formula for the Taylor coefficients of f g follows. □

4. Bonus: Abel’s theorem

At the boundary of the disk of convergence, the power series may be
convergent or divergent. Abel’s theorem relates convergence at a boundary
point to the behavior of the function inside the disk.
Theorem 5.21 (Abel). Let f (x) = ∞ k
P
k=0 ak x have radius of conver-
gence 1. Assume that the series
∞
X
ak
k=0

is convergent. Then f has a left limit (radial limit) at 1,

∞
X
lim f (x) = ak .
x↑1
k=0

Proof. Denote Sn = nk=0 ak with S−1 = 0. Then the sequence Sn is

P
convergent. Thus, for x ∈ (0, 1), by telescoping,
n
X
f (x) = lim (Sk − Sk−1 )xk
n→∞
k=0
n−1
!
X
= lim Sn xn + Sk (xk − xk+1 )
n→∞
k=0
∞
X
= Sk xk (1 − x)
k=0
 4. BONUS: ABEL’S THEOREM 89

(in the last step, we use convergence of Sn to conclude decay of Sn xn ).

Subtracting S and using
∞
X
xk (1 − x) = 1
k=0
we write
∞
X
f (x) − S = (Sk − S)xk (1 − x).
k=0
For any ϵ > 0, there exists K such that for all k ≥ K, |Sk − S| < ϵ, so
K−1
X ∞
X
|f (x) − S| ≤ |Sk − S|xk (1 − x) + |Sk − S|xk (1 − x)
k=0 k=K
K−1
X X∞
≤ |Sk − S|xk (1 − x) + ϵxk (1 − x)
k=0 k=K
and therefore
K−1
X
|f (x) − S| ≤ |Sk − S|xk (1 − x) + ϵ.
k=0
Also, for any k,
lim|Sk − S|xk (1 − x) = 0,
x↑1
so
K−1
X
lim |Sk − S|xk (1 − x) = 0.
x↑1
k=0
Thus, there exists x0 such that for all x ∈ (x0 , 1),
|f (x) − S| ≤ ϵ + ϵ = 2ϵ. □
The converse is false: the function
∞
1 X
f (z) = = (−1)k z k
1+z
k=0
P∞
has a finite limit at 1 but the series is k=0 (−1)k which is divergent.
Abel’s theorem gives some interesting identities:
Corollary 5.22.
∞
X (−1)k−1
= ln 2.
k
k=1

Proof. This is an alternating series, so it is convergent. By Abel’s

theorem, it is equal to
∞ ∞
X (−1)k−1 X (−1)k−1
= lim rk = lim ln(1 + r) = ln 2. □
k r↑1 k r↑1
k=1 k=1
90 5. CAUCHY INTEGRAL FORMULA AND ANALYTIC FUNCTIONS

5. Zeros of analytic functions and the identity theorem

In this section, we study general properties of zeros of analytic func-
tions. We begin with a factorization result, reminiscent of factorization of
polynomials:
Lemma 5.23 (Factorization lemma). If f : Ω → C is analytic and z0 ∈
Ω, k ∈ N such that
f (z0 ) = · · · = f (k−1) (z0 ) = 0, (5.6)
then the function
f (z)
(
(z−z0 )k
z ∈ Ω \ {z0 }
g(z) = f (k) (z0 )
k! z = z0
is analytic on Ω and
f (z) = (z − z0 )k g(z), ∀z ∈ Ω.
Proof. The function g is clearly holomorphic on Ω \ {z0 } as the ratio
of holomorphic functions. For r > 0 such that Dr (z0 ) ⊂ Ω, the function f
has a power series representation
∞
X f (n) (z0 )
f (z) = (z − z0 )n
n!
n=0
on z ∈ Dr (z0 ). Due to (5.6), the terms with n < k vanish, and dividing by
(z − z0 )k gives
∞
X f (n) (z0 )
g(z) = (z − z0 )n−k
n!
n=k
for z ∈ Dr (z0 ) \ {z0 }. Due to our choice of g(z0 ), equality holds also for z =
z0 . Thus, g has a power series representation on Dr (z0 ), so g is holomorphic
at z0 . □
Recall a subset of a metric space is called discrete if it has no accumu-
lation points. We can now prove our main result:
Theorem 5.24. Let f : Ω → C be analytic, and not identically zero.
Then the set of zeros of f is discrete in Ω, and for each zero z0 , there exists
m ∈ N such that
f (z0 ) = f ′ (z0 ) = · · · = f (m−1) (z0 ) = 0, f (m) (z0 ) ̸= 0. (5.7)
Proof. Denote
A = {z ∈ Ω | ∀n ∈ N0 f (n) (z) = 0}.
Since each f (n) is continuous, the set A is an intersection of closed sets, and
therefore closed.
However, if z0 ∈ A and r > 0 such that Dr (z0 ) ⊂ Ω, by the Cauchy
power series theorem, f is represented in Dr (z0 ) by the trivial power series,
which implies Dr (z0 ) ⊂ A. Thus, A is open.
 5. ZEROS OF ANALYTIC FUNCTIONS AND THE IDENTITY THEOREM 91

Since Ω is connected, it follows that if A ̸= Ω, then A = ∅.

For any zero z0 of f , since z0 ∈
/ A, there exists a smallest m ∈ N such
(m)
that f (z0 ) ̸= 0. Then (5.7) holds. Moreover, by the factorization lemma,

f (z) = (z − z0 )m g(z)

with an analytic function g : Ω → C and g(z0 ) = f (m) (z0 )/m! ̸= 0. By

continuity of g, there exists δ > 0 such that g is nonzero on Dδ (z0 ), so f is
nonzero on Dδ (z0 ) \ {z0 }. Thus, zeros of f are isolated.
If the set of zeros of f had an accumulation point z0 ∈ Ω, then by
continuity f (z0 ) = 0, so z0 would be a non-isolated zero of f , leading to a
contradiction. □

Definition 5.25. We say that z0 is a zero of multiplicity (or order) m

if (5.7) holds.

Remark 5.26. From the factorization lemma, further conclusions for

the multiplicity of zeros follow as in the case of polynomials. For instance, if
f, g have a zero at z0 of multiplicities m, n, respectively, then f g has a zero at
z0 of multiplicity m + n. This gives us better insight into some of the topics
considered before. For instance: if f has a zero at z0 of odd multiplicity m,
it has no branch of square root in a neighborhood of z0 , because that branch
would have to have a zero of multiplicity m/2 ∈ / N.

The set of zeros is discrete in Ω, but it can have accumulation points

outside of Ω. For instance:

Example 5.27. f (z) = sin(π/z) is analytic on C \ {0}, not identically

zero, and has zeros at the points 1/n, n ∈ Z \ {0}. Those zeros accumulate
at 0.

Statements about zeros, applied to a difference of two analytic functions,

give a criterion for their equality:

Theorem 5.28 (Identity theorem for analytic functions). Let f, g : Ω →

C be analytic. Assume there is a convergent sequence zn → z∞ without
repetitions with zn , z∞ ∈ Ω with f (zn ) = g(zn ) for all n ∈ N. Then f = g.

Proof. The analytic function f − g has zeros at zn , which accumulate

at z∞ ∈ Ω, so f − g is identically zero on Ω. □

Example 5.29. If f : C → C is analytic and f (1/n) ∈ R for all n ∈ N,

then
f (z) = f (z), ∀z ∈ C.

Proof. The function f♯ (z) = f (z) is also analytic on C, by a conse-

quence of the Cauchy–Riemann relations. Since f♯ (1/n) = f (1/n) for n ∈ N,
f♯ (z) = f (z) for z ∈ C by the identity theorem. □
92 5. CAUCHY INTEGRAL FORMULA AND ANALYTIC FUNCTIONS

Analytic continuation. The identity theorem can be used to under-

stand how analytic functions can be extended to larger domains:
Corollary 5.30. If Ω1 , Ω2 are domains with Ω1 ⊂ Ω2 , any analytic
function on Ω1 has at most one analytic continuation to Ω2 .
One way to produce an analytic continuation is to produce an analytic
function with overlapping domains and values:
Corollary 5.31. Let fj : Ωj → C be analytic functions, j = 1, 2. If
f1 = f2 on Ω1 ∩ Ω2 , then the function
(
f1 (z) z ∈ Ω1
f (z) =
f2 (z) z ∈ Ω2 \ Ω1
is analytic on Ω1 ∪ Ω2 .
Proof. This follows from locality of analyticity. □
In the above, as always, Ωj were domains; in particular, they were open.
Example 5.32. The function
∞
n
X
f (z) = z2
n=0
is well-defined on D and does not have an analytic continuation to any larger
domain.
Proof. If it had an analytic continuation to some larger domain Ω, this
domain would contain an arc of the unit circle. That arc would contain some
eiθ such that 2m θ ∈ 2πZ for some m ∈ N. Then for r ∈ (0, 1),
m−1 ∞
n nθ n
X X
f (reiθ ) = r2 ei2 + r2 .
n=0 n=m
Since
∞ M
2n n
X X
lim inf r ≥ lim inf r2 = M − m
r↑1 r↑1
n=m n=m
P∞ 2n
P∞ 2n
for any M ∈ N, this implies lim inf r↑1 = ∞ so lim supr↑1
n=m r n=m r =
∞ so
lim|f (reiθ )| = ∞
r↑1
which leads to a contradiction. □

6. Morera’s theorem and the reflection principle

Morera’s theorem is another counterintuitive result of complex analysis:
it allows us to check whether a function is holomorphic by integrating it.
This is valuable, since integrals are usually easier to control than derivatives.
 6. MORERA’S THEOREM AND THE REFLECTION PRINCIPLE 93

Dr (z) Ω
I
R
Ω♯

Figure 5.3. The proof of Lemma 5.34 and the reflection

principle uses triangles which may intersect both Ω and Ω♯

Theorem 5.33 (Morera’s theorem). Let Hf : Ω → C be continuous. If

for every z0 ∈ Ω there exists r > 0, such that γ f (z)dz = 0 for all triangles
γ in Dr (z0 ), then f is holomorphic on Ω.
Proof. Fix z0 ∈ Ω. By Lemma 4.20, there exists holomorphic g :
Dr (z0 ) → C such that g ′ = f . Thus, as the derivative of a holomorphic
function, f is holomorphic on Dr (z0 ) and in particular at z0 . □

Morera’s theorem can be used to glue together two analytic functions

along a shared boundary on which they are continuous:
Lemma 5.34. Let Ω ⊂ C+ be a domain and I ⊂ R an open subset of R
such that for every x ∈ I, there exists r > 0 such that
Dr (x) ∩ C+ ⊂ Ω.
Denote Ω♯ = {z̄ | z ∈ Ω}. Assume that f1 is analytic on Ω, continuous on
Ω ∪ I, that f2 is analytic on Ω♯ , continuous on Ω♯ ∪ I, and that
f1 (x) = f2 (x), ∀x ∈ I.
Then the function
(
f1 (z) z ∈Ω∪I
g(z) =
f2 (z) z̄ ∈ Ω
is analytic on Π = Ω ∪ I ∪ {z̄ | z ∈ Ω}.
Proof. Denote Ω♯ = {z | z ∈ Ω}.
The function g is continuous on Π because f1 , f2 are continuous and
f1 = f2 on I. At z ∈ Ω, g is holomorphic because f1 is. At z ∈ Ω♯ , g is
holomorphic because f2 is. We will use Morera’s theorem to prove that g is
holomorphic at points z ∈ I.
Fix z ∈ I and r > 0 such that Dr (z) ∩ C+ ⊂ Ω. For any closed rectifiable
curve γ in C+ ∩ Dr (z), by the Cauchy integral theorem,
I
g(ζ) dζ = 0.
γ
94 5. CAUCHY INTEGRAL FORMULA AND ANALYTIC FUNCTIONS

If, instead, γ is a closed rectifiable curve in (C+ ∪ I) ∩ Dr (z), let ϵ0 =

dist(Ran γ, Dr (z)c ). Then, since g is uniformly continuous on the compact
set
{z + iϵ | z ∈ Ran γ, δ ∈ [0, δ0 ]} ⊂ Dr (z),
we have
I I
g(ζ) dζ − g(ζ + iδ) dζ ≤ ℓ(γ)∥g(·) − g(· + iδ)∥γ → 0, δ ↓ 0.
γ γ

Since γ + iδ is a closed rectifiable curve in C+ ∩ Dr (z), this implies

I I I
g(ζ) dζ = lim g(ζ + iδ) dζ = lim g(ζ) dζ = 0.
γ δ↓0 γ δ↓0 γ+iδ

Analogously,
H if γ is a closed rectifiable curve in (C− ∪ I) ∩ Dr (z), then
γ g(ζ) dζ = 0.
For any triangle γ in Dr (z), if it is not contained in C+ ∪ I or C− ∪ I, its
contour integral is the sum of two contour integrals over polygonal curves
γ± in (C± ∪ I) ∩ Dr (z). Thus,
I I I
g(ζ) dζ = g(ζ) dζ + g(ζ) dζ = 0.
γ γ+ γ−

By Morera’s theorem, g is holomorphic at z. □

Although it is of general interest, Lemma 5.34 is often not stated in
this form, but rather in the special case known as the reflection principle.
In Example 5.29, we saw how symmetry can arise for an analytic function.
In the reflection principle, we use symmetry to extend the domain of an
analytic function:
Theorem 5.35 (Reflection principle). Let Ω ⊂ C+ be a domain and
I ⊂ R an open subset of R such that for every x ∈ I, there exists r > 0 such
that
Dr (x) ∩ C+ ⊂ Ω.
Assume that f is analytic on Ω, continuous on Ω ∪ I, and
f (x) ∈ R, ∀x ∈ I.
Then the function (
f (z) z ∈Ω∪I
g(z) =
f (z̄) z̄ ∈ Ω
is analytic on Π = Ω ∪ I ∪ {z̄ | z ∈ Ω}.
Proof. Denote f♯ : Ω♯ ∪ I → C, f♯ (z) = f (z). By Example 5.29, f♯ is
analytic on Ω♯ . For z ∈ I, z = z and f (z) ∈ R, so
f♯ (z) = f (z) = f (z) = f (z).
Applying Lemma 5.34 to f, f♯ concludes the proof. □
 7. EXERCISES 95

This is the standard formulation, in which the real line has a special
role. However, the reflection principle can be more general: it can be modi-
fied to reflect with respect to a circular arc or a more general analytic arc.
This is done by precomposing or postcomposing by maps which change the
geometry of the boundary, or modifying the proof to work with different
boundaries.

7. Exercises
1. Let f be holomorphic in an open set containing Dr (z0 ). Prove that for
n ∈ N and z ∈ Dr (z0 ),
I
(n) n! f (ζ)
f (z) = dζ.
2πi |ζ−z0 |=r (ζ − z)n+1
Note that the case n = 0 would correspond to the Cauchy integral for-
mula.
2. (a) Find the Taylor polynomial of degree 6 centered at 0 for tan z.
(b) Find the radius of convergence of the Maclaurin series for tan z.
3. For the Fibonacci sequence F0 = 0, F1 = 1, Fn = Fn−1 + Fn−2 , compute
the power series
∞
X
f (z) = Fn z n
n=0
and find its radius
P of convergence. Find a partial fraction decomposition
in the form a
j j /(1 − cj z) and by using power series expansions for
functions of the form 1/(1 − cz), compute the power series representation
of f around 0 and obtain an explicit formula for Fn .
4. For the branch of arcsin on D such that arcsin 0 = 0, prove that
∞
X (2n)! z 2n+1
arcsin z = .
22n (n!)2 2n + 1
n=0
Then use sin(π/6) = 1/2 to derive a series representation for π.
5. Assume that f is analytic on D = D1 (0) and
1
|f (z)| < , ∀z ∈ D.
1 − |z|
What is the best bound you can get from Cauchy estimates for |an |? How
does the bound behave as n → ∞?
6. If f : C → C is an entire function and d ∈ N0 such that
|f (z)| = o(|z|d+1 ), z→∞
prove that f is a polynomial of degree at most d.
7. An entire function f is said to have finite order if for some α ∈ (0, ∞),
there exist C, D ∈ (0, ∞) such that
α
|f (z)| ≤ CeD|z| , ∀z ∈ C.
The infimum of all such α is then called the order of f and denoted ρ(f ).
 96 5. CAUCHY INTEGRAL FORMULA AND ANALYTIC FUNCTIONS

If f has finite order, prove that f ′ has finite order and that ρ(f ′ ) ≤
ρ(f ).
Hint: use Cauchy estimates for f ′ (z) over a circle of radius 1 centered
at arbitrary z.
8. Prove that
∞
X (−1)j π
= .
2j + 1 4
j=0

Hint: compute limr↑1 log(1 + ir).   P∞  1 

9. Find the values of the sums ∞ 1 1 2 1
P
k=0 3k+1 + 3k+2 − 3k+3 and k=0 3k+1 − 3k+2 .
10. Under the assumptions of Abel’s theorem, for any ϕ ∈ (0, π/2), prove the
nontangential limit
∞
X
lim f (z) = ak .
z→1
Im(e−iϕ (z−1))≥0 k=0
Im(eiϕ (z−1))≤0

11. If f, g are analytic functions on Ω and at every zero of g, f also has a

zero, of same or larger multiplicity. Prove that there exists h analytic on
Ω such that f = gh.
12. If f, g are analytic functions on Ω and f g = 0, prove that f = 0 or g = 0.
13. Find all functions f analytic on a neighborhood of 0 such that f (1/n2 ) =
n2 −1
n4
for all large enough n ∈ N.
14. Find all functions f analytic on a neighborhood of 0 such that f (1/n2 ) =
n3 −1
n4
for all large enough n ∈ N.
15. Assume that h : C → C is 1-periodic, entire, and
lim sup |h(x + iy)| = 0.
|y|→∞ x∈[0,1]

Prove that h = 0.
Hint: prove that h is bounded; or prove that there exists analytic
f : C \ {0} → C such that
h(z) = f (e2πiz )
and study the behavior of f at 0 and ∞.
16. Let f : C+ → C be analytic and for all z ∈ ∂D ∩ C+ , f (z) ∈ R. If f has
no zeros in D ∩ C+ , prove that f has no zeros.
17. Suppose that f : Dr (0) → C obeys f (2z) = 2f (z)f ′ (z) whenever |z| <
r/2. Prove that f has an analytic continuation to C.
18. Let I ⊂ R be an open interval and f : C+ ∪ I → C a continuous function.
Assume f is analytic on C+ and f (I) ⊂ ∂D. Prove that f has an analytic
continuation to some domain Π with C+ ∪ I ⊂ Π given by
(
f (z) z ∈ C+ ∪ I
g(z) =
1/f (z) z ∈ (Π \ (C+ ∪ I))
 7. EXERCISES 97

Note that f (z) = z−i

z+i satisfies the assumptions but doesn’t have an ana-
lytic continuation to C.
19. Fix r ∈ (0, 1). Find all functions f which are holomorphic on A = {z ∈
C | r < |z| < 1}, continuous on A, and equal to zero on ∂D.
20. Let f : D → C be continuous. Assume f is analytic on D and f (∂D) ⊂ R.
Prove that f is constant.
Hint: reflect across the unit circle to construct an analytic continua-
tion of f to C and prove that it is bounded.
 CHAPTER 6

Limits of analytic functions

To further develop complex analysis, we often need to approximate one

analytic function by others, or construct a new analytic function as a limit
of others. A power series is representation of a function as a limit of poly-
nomials of larger and larger degree, but we will need more general limiting
processes. This is the subject of this chapter.

1. Limits of analytic functions and Montel’s theorem

In the study of power series, it was important that series converge uni-
formly on compact subsets of the disk of convergence. This is the mode
of convergence we will now consider for a sequence of analytic functions on
some domain Ω.
Limits of analytic functions are analytic.
Theorem 6.1. If fn : Ω → C are analytic and fn → f uniformly on
compact subsets K ⊂ Ω, then
(a) f is analytic on Ω.
(b) fn′ → f ′ uniformly on compact subsets K ⊂ Ω.
Proof. (a) Uniform convergence implies that the limit f is continuous
on any compact K ⊂ Ω. Thus, f is continuous on Ω.
For any triangle γ in Ω, Ran γ is compact, so exchanging uniform limits
with integrals gives
I I
f (ζ) dζ = lim fn (ζ) dζ = 0.
γ n→∞ γ

Thus, by Morera’s theorem, f is holomorphic on Ω.

(b) For any compact K, there exists r > 0 such that the compact
K ′ = {z ∈ C | dist(z, K) ≤ r}
is also contained in Ω. For z ∈ K, use the formula
fn (ζ) − f (ζ)
I
1
(fn − f )′ (z) = dζ.
2πi |ζ−z|=r (ζ − z)2
The standard integral estimate gives
2πr fn (ζ) − f (ζ)
|fn′ (z) − f ′ (z)| ≤ sup
2π ζ∈K ′ r2
which implies uniform convergence of fn′ to f ′ on K. □
99
100 6. LIMITS OF ANALYTIC FUNCTIONS

Montel’s theorem. Every bounded sequence in C has a convergent

subsequence; this is a manifestation of the fact that bounded closed subsets
of C are compact. We will now prove a result of this flavor for sequences of
analytic functions.
Denote by A(Ω) the set of analytic functions on Ω.
Theorem 6.2 (Montel’s theorem, 1st form). Let (fn )∞
n=1 be a sequence
in A(Ω). If, for every compact K ⊂ Ω,
sup∥fn ∥K < ∞,
n∈N
then there is a subsequence (fnk )∞
k=1 which converges uniformly on compacts
to some g ∈ A(Ω).
The proof requires some prerequisites about uniform convergence on
compacts, which we will recall. For a set K, we denote by ∥·∥K the uniform
norm,
∥f ∥K = sup |f (z)|.
z∈K
We allow the case K = ∅ with ∥·∥∅ = 0 in what follows. In particular, if K is
compact, the space of complex-valued continuous functions on K, denoted
C(K), has a metric given by d(f, g) = ∥f − g∥K . This is a complete metric
space, i.e., it is a Banach space with norm ∥·∥K .
Definition 6.3. Let Ω ⊂ C be a domain (open and connected set). A
compact exhaustion of Ω is a sequence (Km )∞
m=1 of compact sets such that
Km ⊂ int Km+1
for all m ∈ N and [
Km = Ω.
m∈N

Example 6.4. For any domain Ω, the sets

Km = {z ∈ Ω | |z| ≤ m, dist(z, C \ Ω) ≥ 1/m}
are a compact exhaustion of Ω.
Proof. For any z ∈ Km , Drm (z) ⊂ Km+1 where rm = (m(m + 1))−1 .
□
Proof of Theorem 6.2. For each m, as in the previous proof, we have
sup∥fn ∥Km < ∞, sup∥fn′ ∥Km < ∞.
n n
In particular, for z, w ∈ Km with |z − w| ≤ rm , the line segment [zw] is in
Km+1 so
|fn (z) − fn (w)| ≤ |z − w|∥fn′ ∥Km+1 .
It follows that (fn )∞
n=1 are equicontinuous in C(Km ). By the Arzelà–Ascoli
theorem, there is a convergent subesequence in C(Km ). By a diagonalization
argument, there is a subsequence (fnk )∞ k=1 which is convergent in C(Km ) for
 2. BONUS: METRIZABILITY OF UNIFORM CONVERGENCE ON COMPACTS 101

all m. Denote by gm the limit in C(Km ). Then gm+1 |Km = gm so, taking
the union of domains, we obtain g : Ω → C such that fnk → g uniformly on
each Km . By Morera’s theorem, g ∈ A(Ω).
For any compact K ⊂ Ω, there exists m such that K ⊂ Km , so fnk → g
uniformly on K. □

2. Bonus: metrizability of uniform convergence on compacts

Our intuition about convergence is largely based on the notion of con-
vergence in a metric space. However, when we studied uniform convergence
on compacts in A(Ω), we didn’t define it through a metric. It is useful to
know that there is such a metric:
Theorem 6.5. Let Ω be a domain and (Km )∞m=1 a compact exhaustion
of Ω. Denote
∞
X
d(f, g) = min{2−m , ∥f − g∥Km }.
m=1
Then d is a metric on A(Ω) and, in this metric:
(a) fn → f in A(Ω) if and only if for every compact K ⊂ Ω, fn converge to
f uniformly on K
(b) A(Ω) is a complete metric space
(c) (Montel’s theorem, 2nd form) A set S ⊂ A(Ω) is compact if and only if
it is closed and, for every compact K ⊂ Ω,
sup∥f ∥K < ∞.
f ∈S

Proof. Denote dm (f, g) = min{2−m , ∥f − g∥Km }. For any m, we prove

dm (f, h) ≤ dm (f, g) + dm (g, h)
by separating cases:
(i) if ∥f − g∥Km ≥ 2−m or ∥g − h∥Km ≥ 2−m , then
dm (f, g) + dm (g, h) ≥ 2−m ≥ dm (f, h);
(ii) if ∥f − g∥Km < 2−m and ∥g − h∥Km < 2−m , then
dm (f, g) + dm (g, h) = ∥f − g∥Km + ∥g − h∥Km ≥ ∥f − h∥Km ≥ dm (f, h).
Summing in m, it follows that d obeys the triangle inequality. Obviously d
is symmetric, and d(f, g) = 0 implies ∥f − g∥Km = 0, so f = g on Ω.
(a)⇐: By using different trivial estimates for m ≤ M and m > M , we
have
M
X X∞ M
X
d(fn , f ) ≤ ∥f − g∥Km + 2−m ≤ ∥f − g∥Km + 2−M .
m=1 m=M +1 m=1

Since ∥fn − f ∥Km → 0 for each m, we conclude

lim sup d(fn , f ) ≤ 2−M .
n→∞
102 6. LIMITS OF ANALYTIC FUNCTIONS

Since M ∈ N is arbitrary, this implies d(fn , f ) → 0 as n → ∞.

(a)⇒: For any compact K ⊂ Ω, if we pick m such that m ≥ maxz∈K |z|
and
m ≥ max 1/dist(z, C \ Ω),
z∈K
then K ⊂ Km . If d(fn , f ) → 0, then min{2−m , ∥fn − f ∥Km } → 0 and
therefore ∥fn − f ∥Km → 0, which implies ∥fn − f ∥K → 0 as n → ∞.
(b) Let (fn )∞ ∞
n=1 be a Cauchy sequence in A(Ω). Then (fn )n=1 is a Cauchy
sequence in C(Km ), so it converges to some gm ∈ C(Km ) uniformly on Km .
Then gm+1 |Km = gm so, taking the union of domains, we obtain g : Ω → C
such that fn → g uniformly on each Km . By Fubini and Morera, g ∈ A(Ω).
(c) ⇒: Compact sets are always closed. Since convergence in A(Ω)
implies convergence in C(K), any compact set must be bounded in C(K).
(c) ⇐: By Theorem 6.2, any sequence in S has a convergent subsequence.
Since S is closed, the subsequential limit lies in S. This means that S is
sequentially compact, so it is compact. □
Example 6.6. For any domain Ω, the set {f ∈ A(Ω) | ∥f ∥Ω ≤ 1} is
compact.
Compactness is often useful for proving convergence; recall that for a
sequence in a compact metric space, if every convergent subsequence has the
same limit f , then the entire sequence converges to f (compare Exercise 1).

3. Series, parametric integrals, and products of analytic

functions
In this section, we consider several important ways to obtain a new
analytic function as some kind of limit of analytic functions.

Series of analytic functions.

Theorem 6.7 (Weierstrass test). If fn : Ω → C are analytic and, for
every compact K ⊂ Ω,
∞
X
lim sup |fn (z)| = 0
N →∞ z∈K
n=N
then
∞
X
g(z) = fn (z)
n=1
defines an analytic function g : Ω → C and
∞
X
g ′ (z) = fn′ (z).
n=1

This is a P
consequence of the previous result applied to the sequence of
partial sums N n=1 fn .
3. SERIES, PARAMETRIC INTEGRALS, AND PRODUCTS OF ANALYTIC FUNCTIONS
103

Analytic functions from parametric integrals (“Fubini and Mor-

era”). Often analytic functions are given by integral representations. In
those cases, well-posedness and analyticity come from analyticity of the in-
tegral kernel, by the Fubini and Morera theorems.
We state here a version for contour integrals, which will be used when
we study the local behavior of analytic functions:
Theorem 6.8. Let σ be a piecewise C 1 curve in C. Assume that F (z, w)
is jointly continuous on Ω × Ran σ and, for every w ∈ Ran σ, F (z, w) is
analytic in z. Then the parametric integral
I
f (z) = F (z, w) dw
σ
is analytic in z ∈ Ω, and
I
′
f (z) = (∂z F )(z, w) dw
σ
Proof. Since F (z, w) is jointly continuous on Ω × Ran σ, it is uniformly
continuous on K × Ran σ for any compact K ⊂ Ω. By uniform continuity
of the integrand, the integral is continuous in z.
By the same joint continuity, Fubini’s theorem can be applied to ex-
change integrals: for any triangle γ in Ω,
I I I
1
f (z) dz = F (z, w) dw dz
γ 2πi γ σ
I I
1
= F (z, w) dz dw
2πi σ γ
I
1
= 0 dw
2πi σ
=0
(in the second to last step, we used analyticity of F (z, w) for fixed w ∈
Ran σ). By Morera’s theorem, f is holomorphic.
By the Cauchy integral formula, the derivative
I
1 F (ζ, w)
∂z F (z, w) = dζ
2πi |ζ−z|=r (ζ − z)2
is jointly continuous on Ω×Ran σ. Likewise, by the Cauchy integral formula,
for r > 0 small enough, using continuity of ∂z F on Ω × Ran σ, we can apply
Fubini’s theorem to conclude
I
′ 1 f (ζ)
f (z) = dζ
2πi |ζ−z|=r (ζ − z)2
I I
1 F (ζ, w)
= dw dζ
2πi |ζ−z|=r σ (ζ − z)2
I I
1 F (ζ, w)
= dζ dw
2πi σ |ζ−z|=r (ζ − z)2
104 6. LIMITS OF ANALYTIC FUNCTIONS
I
= ∂z F (z, w) dw. □
σ

This can be generalized if instead of contour integration we assume the

integration in w is over some measure space; instead of compactness of the
parameter space we then use dominated convergence to establish continuity
of f and Fubini’s theorem to establish its zero integral over triangles. For
an application of this principle to the Fourier transform, see Exercise 3.

Infinite products of analytic functions.

Theorem 6.9. If fn : Ω → C are analytic, not identically zero, and, for
every compact K ⊂ Ω,
X∞
∥1 − fn ∥K < ∞, (6.1)
n=1
then:
(a)
∞
Y
g(z) = fn (z)
n=1
defines an analytic function g : Ω → C
(b) g is not identically zero; any zero of g is a zero of some fn , and its
multiplicity is the sum of multiplicities from the fn ’s.
(c) for every compact K ⊂ Ω \ {z | g(z) = 0},
∞
X fn′
<∞ (6.2)
fn K
n=1
and uniformly in K,
∞
g ′ (z) X fn′ (z)
= .
g(z) fn (z)
n=1
w
The motivation for the criterion Pnfor small w, 1 + w ≈ e ,
Qn (6.1) is that,
so if if fn ≈ 1 for large n, then j=N fj ≈ exp( j=N (fj − 1)). To prove it,
we need a precise inequality:
Lemma 6.10. For t ≥ 0, 1 + t ≤ et .
Proof. The function f (t) = et − t − 1 obeys f (0) = f ′ (0) = 0 and
f ′′ (t)
= et > 0 for t > 0, so f ′ is strictly increasing, so f ′ (t) > f ′ (0) = 0.
Then f is strictly increasing so f (t) > f (0) = 0 for t > 0. □
Proof of Theorem 6.9. (a) Fix a compact K ⊂ Ω and fix N ∈ N
such that for all n ≥ N , ∥1 − fn ∥K ≤ 21 . Denote
n
Y
hn = fj .
j=N
3. SERIES, PARAMETRIC INTEGRALS, AND PRODUCTS OF ANALYTIC FUNCTIONS
105

Then, using |1 − w| ≤ 1 + |w| ≤ e|w| , we obtain

 
Yn n
X
∥hn ∥K ≤ (1 + |1 − fj |) ≤ exp  ∥1 − fj ∥K 
j=N j=N

which implies that hn is a bounded sequence in C(K). Moreover, by a

telescoping sum, for m ≥ n ≥ N ,
m
X m
X
hm − hn = (hj − hj−1 ) = hj (fj − 1)
j=n+1 j=n+1

implies that
m
X
∥hm − hn ∥K ≤ sup ∥hj ∥K ∥fj − 1∥K
j≥N j=n+1
and therefore (hn )∞
n=N is a Cauchy sequence in C(K). It follows that hn
converges in C(K) as n → ∞ to some h ∈ C(K). Thus,
n
Y N
Y −1 N
Y −1
fj = hn fj → h fj , n → ∞, (6.3)
j=1 j=1 j=1

and since this product converges uniformly on compacts, the limit is an

analytic function on Ω.
(b) If t = |w| ≤ 21 , then
(1 − t)(1 + 2t) = 1 + t − 2t2 = 1 + t(1 − 2t) ≥ 1
so 1 − |w| ≥ (1 + 2|w|)−1 . Combining this with 1 + 2|w| ≤ e2|w| , we conclude
1
|1 − w| ≥ 1 − |w| ≥ ≥ e−2|w| ,
1 + 2|w|
so in the setting of the proof of (a),
 
n
X
∥h−1
n ∥K ≤ exp 2
 ∥1 − fj ∥K  .
j=N

It follows that for all z ∈ K,

 
n
X
|hn (z)| ≥ exp −2 ∥1 − fj ∥K  ,
j=N

and taking n → ∞, that

 
∞
X
|h(z)| ≥ exp −2 ∥1 − fj ∥K  .
j=N

In particular, h is nonvanishing on K, so zeros of the limit (6.3) are precisely

those of the individual factors fj with j < N , counted with the sum of
multiplicities.
106 6. LIMITS OF ANALYTIC FUNCTIONS

(c) For any compact K, by a Cauchy estimate,

∞
X
lim ∥fn′ ∥K = 0.
N →∞
n=N
1
For any compact K, with N chosen as before and n ≥ N , ∥1 − fn ∥K ≤ 2
implies that fn has no zeros on K and ∥fn−1 ∥K ≤ 2, so
fn′
≤ 2∥fn′ ∥K
fn K

which proves (6.2).

Each product gn = nj=1 fj obeys
Q

n
X fj′
gn′
= .
gn fj
j=1

Let us take the limit as n → ∞ of both sides in Ω\{z | g(z) = 0}. Since gn →
g uniformly on compacts, gn′ → g ′ uniformly on compacts, so gn′ /gn → g ′ /g
uniformly on compacts. Meanwhile, by the Weierstrass test, (6.2) implies
fj′
that ∞
P
j=1 fj is convergent. □

4. Bonus: Weierstrass product theorem

To construct an entire function with zeros at prescribed points z1 , . . . , zn ,
an obvious choice is the polynomial
n
Y
p(z) = (z − zj ).
j=1

Zeros of multiplicity greater than 1 can be obtained by allowing repetition

in the sequence. If we wish to prescribe infinitely many zeros, factors of
the form z − zj won’t work: the product ∞
Q
j=1 (z − zj ) cannot converge
on a nonempty open set, so it does not define an analytic function. More
promising are factors of the form 1−z/zj : by Theorem 6.9, if j |zj |−1 < ∞,
P
the product
∞  
Y z
1−
zj
j=1

is convergent and has zeros precisely at the points zj . This requires zj ̸= 0,

but it is easy to handle zeros at zero separately, by multiplying the product
by a power of z.
Note that this used the assumption j |zj |−1 < ∞. We wish to remove
P
that assumption and accommodate an arbitrary sequence zj → ∞. To do
so, we use factors which are further renormalized.
 4. BONUS: WEIERSTRASS PRODUCT THEOREM 107

Definition 6.11. The Weierstrass product factors En : C → C are

 
n j
X w 
En (w) = (1 − w) exp 
j
j=1

for n ∈ N, and E0 (w) = 1 − w.

Note that these factors have a simple zero at w = 1 and no other zeros.
We will use them to construct factors for a product with prescribed zeros.
Convergence of the product will be ensured by the following basic estimate:
Lemma 6.12. For w ∈ D,
|1 − En (w)| ≤ |w|n+1 .
Proof. The key is to prove that 1 − En (w) has a Maclaurin series
with positive coefficients and a zero at w = 0 of multiplicity n + 1. A
straightforward computation gives
 
n j
X w
−En′ (w) = wn exp  
j
j=1

which expands to a Taylor series with R = ∞ with positive coefficients,

∞
X
−En′ (w) = bn,j wn+j .
j=0

Integrating gives
∞
X bn,j
1 − En (w) = wn+j+1 .
n+j+1
j=0

Now, for w ∈ D,
∞ ∞
n+1
X bn,j X bn,j
|1 − En (w)| ≤ |w| |w|n ≤ |w|n+1
n+j+1 n+j+1
j=0 j=0
so
|1 − En (w)| ≤ |w|n+1 (1 − En (1))
which concludes the proof. □
Theorem 6.13 (Weierstrass product theorem). For any sequence zn ∈
C \ {0} with zn → ∞ as n → ∞, for each R < ∞,
∞  
X z
sup 1 − En <∞
z:|z|≤R zn
n=0
so
∞  
Y z
f (z) = En
zn
n=0
108 6. LIMITS OF ANALYTIC FUNCTIONS

defines an entire function with zeros precisely at the zn ’s with multiplicity

card{j : zj = zn }.
Of course, we can include a zero at zero of multiplicity m by multiplying
by z m .
Proof. For any R, there exists N such that n ≥ N implies |zn | ≥ 2R
and therefore
z n+1
 
z 1
1 − En ≤ ≤ n+1
zn zn 2
and so on. □
To generalize this to an arbitrary domain Ω ⊂ C, it is natural to allow
sequences zn with accumulation points on ∂Ω.
Lemma 6.14. If the sequence (zn )∞
n=1 is discrete in a domain Ω, then
−1

lim min |zn | , dist(zn , ∂Ω) = 0.
n→∞

Proof. If this were false, there would exist a subsequence (znk )∞

k=1 with
sup|znk | < ∞ and inf dist(znk , ∂Ω) > 0. By the first condition, this subse-
quence would have a convergent subsubsequence with limit z∗ ; by the second,
the limit z∗ would be an element of Ω, leading to a contradiction. □
Theorem 6.15 (Weierstrass product theorem on arbitrary domains).
For any sequence (zn )∞n=1 in a domain Ω with no accumulation points in Ω,
there exists an analytic function f on Ω with zeros precisely at the zn , with
multiplicity card{j : zj = zn }.
Proof. We define factors fn ∈ A(Ω) such that fn has a simple zero at
zn and no other zeros. To separate cases, denote
 
1
A = n ∈ N | dist(zn , ∂Ω) > .
|zn |
For n ∈ A we define  
z
fn (z) = En .
zn
For n ∈ N \ A, we pick wn ∈ ∂Ω such that
|zn − wn | = dist(zn , ∂Ω)
and define  
z n − wn
fn (z) = En .
z − wn
Now fix a compact K ⊂ Ω; there exists m > 0 such that z ∈ K implies
|z| ≤ m and D1/m (z) ⊂ Ω. By the lemma, there exists N such that for all
n ≥ N , |zn | ≥ 2m or dist(zn , ∂Ω) ≤ 1/(2m). If n ≥ N and n ∈ A, then
|zn | ≥ 2m so for all z ∈ K,
z 1
≤
zn 2
 5. BONUS: THE GAMMA FUNCTION 109

and therefore
1
|fn (z)| ≤ . (6.4)
2n+1
If n ≥ N and n ∈
/ A, then dist(zn , ∂Ω) ≤ 1/(2m) so
zn − w 1/(2m) 1
≤ =
z−w 1/m 2
so
Q∞ (6.4) holds again. In summary, (6.4) holds for all n ≥ N and all z ∈ K, so
n=1 fn (z) converges to a nontrivial analytic function with the prescribed
zeros. □

5. Bonus: the Gamma function

The Gamma function is an important special function which can be
defined in several ways. We will use Weierstrass’ definition of the Gamma
function as an infinite product,
∞
1 Y
= ze γz
(1 + z/n)e−z/n (6.5)
Γ(z)
n=1
where γ is the Euler–Mascheroni constant,
 
n−1
X1
γ = lim  − log n . (6.6)
n→∞ j
j=1

Lemma 6.16. The Euler–Mascheroni constant (6.6) is well-defined and

0 < γ < 1.
Proof. Denote
n−1 n
X 1 X 1
an = − log n, bn = − log n.
j j
j=1 j=1

Strict concavity of log on (0, ∞) implies log(1+t) < t for t ∈ (−1, 0)∪(0, ∞),
so  
1 n+1 1 1
an+1 − an = − log = − log 1 + >0
n n n n
and
 
1 n+1 1 1
bn+1 − bn = − log = + log 1 − < 0.
n+1 n n+1 n+1
Thus, (an )∞ ∞
n=1 is strictly increasing and (bn )n=1 strictly decreasing, but bn −
1
an = n → 0 as n → ∞. In particular, the sequences converge to the same
limit γ. Evaluating a1 = 0 and b1 = 1 gives 0 < γ < 1. □
More precise estimates give γ = 0.57 . . . .
special case of
Lemma 6.17. For w ∈ D, |1 − (1 − w)ew | ≤ |w|2 . the Weierstrass
estimates
110 6. LIMITS OF ANALYTIC FUNCTIONS

P∞ Denote
Proof. f (w) = 1−(1−w)ew . A direct calculation gives f ′ (w) =
wew = n=0 w n+1 /n! so, together with f (0) = 0, we conclude
∞
X wn+2
f (w) = .
(n + 2)n!
n=0
Thus, for w ∈ D,
∞
X 1
|f (w)| ≤ |w|2 = |w|2 f (1) = |w|2 . □
(n + 2)n!
n=0
We can now prove that the function Γ is well-defined:
Lemma 6.18. The right-hand side of (6.5) is a convergent product whose
limit is an entire function with simple zeros at 0, −1, −2, . . . and no other
zeros. Thus, Γ is an analytic function without zeros on C \ {−n | n ∈ N0 }.
Proof. Uniform convergence on compact subsets of C follows from the
estimate
|z|2
|(1 + z/n)e−z/n | ≤ 2 , ∀z ∈ Dn (0)
P∞ −2 n
since n=1 n < ∞. The other properties follow, since (1 + z/n)e−z/n has
a simple zero at z = −n and no other zeros. □
We now prove that Γ(z) can be described by a limit of functions (this
is sometimes used as an alternative definition) and use this to prove some
properties.
Proposition 6.19. (a) With the notation nz = ez log n with the branch
log n > 0,
(n − 1)!nz
Γ(z) = lim
n→∞ z(z + 1) . . . (z + n − 1)
for all z ∈ C \ {−n | n ∈ N0 }.
(b) Γ(z + 1) = zΓ(z)
(c) Γ(1) = 1
(d) Γ(n) = (n − 1)! for all n ∈ N.
(e) |Γ(z)| ≤ |Γ(Re z)|.
(f ) Γ is bounded on the strip 1 ≤ Re z ≤ 2.
Proof. (a) Denote
(n − 1)!nz
fn (z) = .
z(z + 1) . . . (z + n − 1)
Then
 
 z z
fn (z)−1 = z 1 + ... 1 + e−z log n
1 n−1
 !
n−1
Y n−1
X1
=z (1 + z/j)e−z/j exp z − log n 
j
j=1 j=1
 5. BONUS: THE GAMMA FUNCTION 111

so
lim fn (z) = Γ(z).
n→∞
(b) This follows from the relation
 z
n
fn (z + 1) = zfn+1 (z)
n+1
by letting n → ∞.
(c) From the product definition and (b),
1 1
= lim = 1.
Γ(1) z→0 zΓ(z)
(d) This follows by induction using (b).
(e) This follows from |z + j| ≥ |Re z + j| and |nz | = |nRe z |.
(f) Since Γ is analytic on Re z > 0, it is bounded on the compact [1, 2].
Thus, by (e), it is bounded on 1 ≤ Re z ≤ 2. □
We now prove a criterion which allows us to recognize a function as the
Gamma function:
Theorem 6.20 (Wielandt). Let f be analytic on the right half-plane
Re z > 0. If f obeys
f (z + 1) = zf (z) (6.7)
and f is bounded on {z ∈ C | Re z ∈ [1, 2]}, then
f (z) = f (1)Γ(z).
Proof. Note that if f obeys the assumptions of the theorem, then so
does f − f (1)Γ. Thus, it suffices to prove the special case f (1) = 0; applying
that to f − f (1)Γ gives the general case.
With the assumption f (1) = 0, we define for Re z > −1
(
f (z+1)
z z ̸= 0
f1 (z) = ′
f (1) z=0
and this is holomorphic on {z | Re z > −1} by Lemma 5.23. Moreover, the
condition f (z + 1) = zf (z) implies f (z) = f1 (z) for Re z > 0. Thus, f1 is
an analytic continuation of f to Re z > −1.
Now this procedure can be iterated: we can define
(
fn−1 (z+1)
z Re z > −n, z ̸= 0
fn (z) = ′
fn−1 (1) z=0
and obtain an analytic continuation of fn−1 on Re z > −n. Combining all
these, we conclude that f has an analytic continuation to C which obeys
f (z + 1) = zf (z) for all z ∈ C.
In addition, f (z) = z −1 f (z + 1) is bounded on the strip Π = {z | Re z ∈
[0, 1]}: it is bounded on Π \ D because there |f (z)| ≤ |f (z + 1)| and f is
bounded on 1 ≤ Re z ≤ 2; moreover, it is bounded on D by continuity.
112 6. LIMITS OF ANALYTIC FUNCTIONS

Then the entire function

u(z) = f (z)f (1 − z)

is bounded on 0 ≤ Re z ≤ 1. Moreover,

u(z + 1) = f (z + 1)f (−z) = zf (z)f (−z) = −f (z)f (1 − z) = −u(z)

which implies that u is bounded on C. Thus, u is constant. Then u(1) = 0

implies u = 0, so f = 0 or f (1 − z) = 0 for all z (otherwise the set of zeros
of u would be discrete, as the union of two discrete sets). □

Corollary 6.21. For Re z > 0,

Z ∞
Γ(z) = e−t tz−1 dt.
0

This uses the convention tz−1 = e(z−1) log t with log t ∈ R for t ∈ (0, ∞).

Proof. For n ∈ N, the functions

Z n
fn (z) = e−t tz−1 dt
1/n

are analytic on C by Theorem 6.8. Note that

|e−t tz−1 | = e−t tRe z−1 .

For ϵ, C ∈ (0, ∞), note convergence of the improper integrals

Z 1 Z ∞
e−t tϵ−1 dt, e−t tC−1 dt
0 1

and use tRe z−1 ≤ tϵ−1 for t ∈ (0, 1] and tRe z−1 ≤ tC−1 for t ∈ [1, ∞) to
conclude that fn converge uniformly on the strips 0 < ϵ ≤ Re z ≤ C < ∞.
Their limit is the improper integral
Z ∞
f (z) = e−t tz−1 dt.
0

In particular, fn converge to f uniformly on compact subsets of {z ∈ C |

Re z > 0}, so f is analytic there. The same estimates above show bounded-
ness of f on the strip 1 ≤ Re z ≤ 2.
By integration by parts,
Z n Z n
−t z−1 −t z n
zfn (z) = e zt dt = e t |1/n + e−t tz dt = e−t tz |n1/n + fn (z + 1)
1/n 1/n
R∞
so as n → ∞, we conclude zf (z) = f (z+1). Moreover, f (1) = 0 e−t dt = 1.
By Wielandt’s theorem, f = Γ. □
 6. EXERCISES 113

6. Exercises
1. (a) If f : C+ → D is an analytic function and limy↓0 f (iy) exists, prove
that for any α ∈ (0, π/2), the nontangential limit
lim f (z)
z→0
α≤arg z≤π−α

also exists.
Hint: prove that if a sequence of the form fn (z) = f (z/rn ), where
rn → ∞ as n → ∞, has a limit in A(C+ ), that limit is the constant
function equal to limy↓0 f (iy). Use compactness. Recall that for a
sequence in a compact metric space, if every convergent subsequence
has the same limit f , then the entire sequence converges to f .
(b) Prove the same claim for a map f : C+ → C+ , interpreting the limits
as limits in Ĉ.
2. Let w ∈ C([0, 1]). Prove that the function
Z 1
w(ζ)
f (z) = dζ
0 ζ −z
is analytic in C \ [0, 1]. This is called the Cauchy transform of w.
3. For an integrable function f on R, its Fourier transform is given by
Z
1
ˆ
f (k) = √ f (x)e−ikx dx.
2π
If f is compactly supported, prove that fˆ is an analytic function on C.
Note: informally, more decay of f (x) as x → ∞ implies more smooth-
ness of fˆ. This result is an example of that.
4. If f is integrable on (0, ∞), prove that the function
Z ∞
g(z) = f (t)e−zt dt
0
is well-defined and holomorphic on Re z > 0. This function is called the
Laplace transform of f .
5. Let f be an analytic function on a simply connected domain Ω. Assume
that all zeros of f are of even multiplicity. Prove that there exists g
analytic on Ω such that g 2 = f .
Hint: prove that there is a factorization f = h2 f∗ with f∗ nonzero.
6. For m = 2, 3, . . . and mz ∈
/ {0, −1, −2, . . . }, prove that
   
(m−1)/2 mz−1/2 1 m−1
Γ(mz)(2π) =m Γ(z)Γ z + ...Γ z + .
m m
Hint: use Wielandt’s theorem.
7. (a) For z ∈
/ {0, −1, −2, . . . }, prove that
∞ 
Γ′ (z)

1 X 1 1
= −γ − − −
Γ(z) z z+n n
n=1
114 6. LIMITS OF ANALYTIC FUNCTIONS

and
∞
Γ′ (z)
 
d 1 X 1
= 2+ .
dz Γ(z) z (z + n)2
n=1
(b) (Log-Convexity) For x, y ∈ (0, ∞) and 0 ≤ θ ≤ 1, prove that
Γ(θx + (1 − θ)y) ≤ Γ(x)θ Γ(y)1−θ .
 CHAPTER 7

The argument principle and consequences

We have already seen that an integral representation, such as that in the

Cauchy integral formula, has important consequences. The central theme of
this chapter is that the number of zeros of an analytic function, and their
location, also has an integral representation; this is known as the argument
principle.

1. Winding numbers
As we try to make a full circle around the origin, the logarithm changes
by 2πi, and the argument changes by 2π. This can be used to make rigorous
the natural topological notion of winding number, which counts how many
times a curve γ goes around the origin, or more generally, around a given
point w ∈ / Ran γ. Since this is a topological discussion, it is natural to start
with a general continuous curve (not necessarily closed and not necessarily
rectifiable).
Definition 7.1. Let γ : [a, b] → C \ {0} be continuous. A continuous
choice of argument for γ is a continuous function ϕ : [a, b] → R such that

γ(t) = |γ(t)|eiϕ(t) ∀t ∈ [a, b].

The motivation is that locally, ϕ(t) behaves like the argument of γ(t),
but that ϕ exists even when we cannot choose a single branch of argument
on Ran γ. For instance, the curve γ(t) = e2πit , t ∈ [0, 1] has a continuous
choice of argument ϕ(t) = 2πt. Note also that ϕ(1) − ϕ(0) = 2π, which
matches our intuitive idea that the circle winds around the origin exactly
once.
Topologically speaking, the following theorem says that R is a universal
cover for ∂D, with the covering map ϕ 7→ eiϕ .
Theorem 7.2. Any continuous curve γ : [a, b] → C \ {0} has a con-
tinuous choice of argument ϕ. Moreover, all other continuous choices of
argument are of the form ϕ + 2πn, where n ∈ Z is a constant.
We break up the proof into several lemmas.

Lemma 7.3 (Uniqueness up to +2πZ). If ϕ, ϕ̃ are continuous choices of

argument for the same curve γ, then ϕ = ϕ̃ + 2πn for some n ∈ Z.
115
116 7. THE ARGUMENT PRINCIPLE AND CONSEQUENCES

Proof. If ϕ and ϕ̃ have the correct properties, then

eiϕ(t) = eiϕ̃(t) ∀t ∈ [a, b]
so ϕ(t) − ϕ̃(t) ∈ 2πZ for all t ∈ [a, b]. By continuity, this implies that ϕ − ϕ̃
is constant. □
Lemma 7.4. If γ : [a, b] → Ω for some simply connected domain Ω ⊂
C \ {0}, then for any branch of arg = Im log on Ω, the function
ϕ(t) = arg γ(t)
is a continuous branch of argument for γ.
Proof. This is immediate from the known properties of log. □
Lemma 7.5. If there exist continuous branches of argument for γ|[a,c]
and γ|[c,b] , then there exists a continuous branch of argument for γ|[a,b] .
Proof. Let ϕ1 : [a, c] → R and ϕ2 : [c, b] → R such that γ(t) =
|γ(t)|eiϕj (t) whenever the right-hand side is defined. In particular,
|γ(c)|eiϕ1 (c) = γ(c) = |γ(c)|eiϕ2 (c)
implies that ϕ1 (c) − ϕ2 (c) = 2πn for some n ∈ Z. Then the function
(
ϕ1 (t) t ∈ [a, c]
ϕ(t) =
ϕ2 (t) + 2πn t ∈ (c, b]
is a continuous choice of argument for γ|[a,b] . □
We can now prove the theorem:
Proof of Theorem 7.2. Uniqueness was already proved. Denote r =
mint |γ(t)| > 0. By uniform continuity, there exists δ > 0 such that
|t − s| < δ =⇒ |γ(t) − γ(s)| < r. (7.1)
Pick k ∈ N such that (b − a)/k < δ and denote tj = a + j(b − a)/k. Then
t +t
for each j, the curve γ|[tj−1 ,tj ] is contained in the disk Dr (γ( j−12 j )) which
does not contain 0, so there is a continuous choice of argument for γ|[tj−1 ,tj ] .
By Lemma 7.5, there is a continuous choice of argument for γ. □
Definition 7.6. For a closed continuous curve γ : [a, b] → C \ {0} with
a continuous choice of argument ϕ : [a, b] → R, its winding number around
the origin is
ϕ(b) − ϕ(a)
n(γ) := .
2π
Note that this is well-defined because the difference ϕ(b) − ϕ(a) is in-
dependent of the continuous argument. Moreover, if γ is a closed curve,
then ϕ(b) − ϕ(a) ∈ 2πZ, since γ(b) = γ(a). Thus, the winding number is an
integer.
 1. WINDING NUMBERS 117

Corollary 7.7. If γ : [a, b] → Ω is a closed curve in some simply

connected domain Ω ⊂ C \ {0}, then n(γ) = 0.
Proof. In this case Lemma 7.4 can be applied globally. For any choice
of arg on Ω, we have ϕ(t) = arg γ(t) for all t ∈ [a, b] so ϕ(a) = ϕ(b). □

Continuity of winding numbers. The following lemma says, infor-

mally speaking, that if two curves are never in antipodal positions (oppo-
site directions from the origin), then they have the same winding numbers.
Later, we will combine this with the argument principle to obtain Rouché’s
theorem.
Lemma 7.8 (Rouché’s theorem for winding numbers). If γ1 , γ2 : [a, b] →
C are two closed continuous curves on C and
|γ1 (t) − γ2 (t)| < |γ1 (t)| + |γ2 (t)|, ∀t, (7.2)
then 0 ∈
/ Ran γ1 , 0 ∈
/ Ran γ2 , and n(γ1 ) = n(γ2 ).
Proof. The assumption (7.2) is equivalent to saying that γ1 , γ2 are
nonzero and
γ1 (t)
∈
/ C \ (−∞, 0], ∀t.
γ2 (t)
In particular, the curve γ1 /γ2 has winding number 0, since it is a closed
curve contained in the simply connected domain C \ (−∞, 0].
If ϕ1 , ϕ2 are continuous choices of arguments for γ1 , γ2 , then ϕ1 − ϕ2 is
a continuous choice of argument for γ1 /γ2 , so
n(γ1 ) − n(γ2 ) = n(γ1 /γ2 ) = 0. □
Theorem 7.9. In C\{0} and in D\{0}, closed curves are loop homotopic
if and only if they have the same winding number.
Proof. Consider closed curves γ0 , γ1 : [0, 1] → C \ {0}.
Let F : [0, 1] × [0, 1] → C \ {0} be a loop homotopy between them
and denote γs (t) = F (t, s). By compactness, there exists ϵ > 0 such that
Dϵ (0) ∩ Ran F = ∅, and since F is uniformly continuous, there exists δ > 0
such that
∥(t1 , s1 ) − (t2 , s2 )∥ < δ =⇒ |F (t1 , s1 ) − F (t2 , s2 )| < ϵ.
Thus, if |s1 − s2 | < δ, Lemma 7.8 applies to give n(γ(s1 )) = n(γ(s2 )). Thus,
the winding number is locally constant. By connectedness of [0, 1], this
implies that the set of values of n(γ(s)) is a connected subset of Z, so the
winding number is independent of s.
Conversely, if n(γ0 ) = n(γ1 ), consider their lifts
γj = eσj , σj = ln|γj | + iϕj
where ϕj denotes a continuous choice of argument for γj .
118 7. THE ARGUMENT PRINCIPLE AND CONSEQUENCES

Then σ0 , σ1 are curves in C and σj (1) − σj (0) = 2πin(γj ). In particular,

since n(γ0 ) = n(γ1 ),
F (t, s) = eσ0 (t)+s(σ1 (t)−σ0 (t))
is a loop homotopy between γ0 , γ1 (equality of winding numbers ensures that
F (1, s) = F (0, s) for all s).
The same proof holds in D \ {0}: then the lifts σj are curves in the
left half-plane {w ∈ C | Re w < 0}, so their convex combinations σ0 (t) +
s(σ1 (t) − σ0 (t)) lie in the same half-plane. □
Winding number around a point w ∈ C. More generally, we can
define the winding number around some other point w ∈ C:
Definition 7.10. For a closed continuous curve γ : [a, b] → C and any
w ∈ C \ Ran γ, the winding number of γ around w is the integer
n(γ, w) = n(γ − w).
Theorem 7.11. The winding number of a closed curve γ is constant
on each connected component of C \ Ran γ. It is zero on the unbounded
connected component of C \ Ran γ.
Proof. If z0 , z1 are in the same connected component of C\Ran γ, then
there is a curve σ in C \ Ran γ with starting point z0 and endpoint z1 . Then
F (t, s) = γ(t) − σ(s) is a loop homotopy in C \ {0} between γ − z0 and γ − z1 ,
so n(γ, z0 ) = n(γ, z1 ).
The unbounded connected component contains C \ DR (0), where R =
maxz∈Ran γ |z|, and it suffices to evaluate the winding number at one such
point, say w = −R − 1. At that point, Re(γ − w) > 0, so n(γ, w) =
n(γ − w) = 0. □
Example 7.12. If γ denotes the counterclockwise circular contour γ(t) =
z0 + re2πit , then (
1 |w − z0 | < r
n(γ, w) =
0 |w − z0 | > r
Proof. We have previously proved that C \ ∂Dr (z0 ) has the connected
components Dr (z0 ) and C \ Dr (z0 ). On the unbounded component, the
winding number is zero. On the bounded component, the winding number
is constant; it is easy to compute at w = z0 , where the continuous choice of
argument is ϕ(t) = 2πt, so n(γ, z0 ) = (ϕ(1) − ϕ(0))/(2π) = 1. □
The integral representation of winding number. For rectifiable
curves, there is an integral representation of winding number:
Lemma 7.13. Let γ : [a, b] → C be a closed rectifiable curve and z ∈
/
Ran γ. Then I
1 dζ
n(γ, z) = .
2πi γ ζ − z
 2. THE ARGUMENT PRINCIPLE AND ROUCHÉ’S THEOREM 119

Proof. Shifting by z, we can assume that z = 0. Similarly as before,

we denote
( I )
1 dζ
x = sup t ∈ [a, b] | ϕ(s) − ϕ(a) = Im ∀s ∈ [a, t]
2πi γ|[a,s] ζ
On an interval [x − δ, x + δ] ⊂ (x − ϵ(x), x + ϵ(x)), since z1 = (log z)′ ,
I
1 dζ
log γ(t) − log γ(x − δ) = .
2πi γ|[x−δ,t] ζ
By the same reasoning as before, this shows that
I
dζ
log|γ(s)| − log|γ(a)| + i(ϕ(s) − ϕ(a)) = , ∀s ∈ [a, b].
γ|[a,s] ζ
In particular, taking s = b, we conclude the proof. □

2. The argument principle and Rouché’s theorem

If an analytic function f : Ω → C is not identically zero, its set of zeros
is discrete. Thus, for any S such that S is a compact subset of Ω, f has
finitely many zeros on S. Summing the multiplicities of those zeros gives the
number NS (f ), called the number of zeros on S (counted with multiplicity).
We will prove a formula for this number of zeros:
Theorem 7.14 (Argument principle). Let f be analytic in Ω, Dr (z0 ) ⊂
Ω, and f ̸= 0 on ∂Dr (z0 ). Then
f ′ (z)
I
1
NDr (z0 ) (f ) = dz. (7.3)
2πi |z−z0 |=r f (z)
Proof. Denoting the zeros in S by z1 , . . . , zk and their multiplicities by
m1 , . . . , mk , we can use Lemma 5.23 to factor
k
Y
f (z) = g(z) (z − zj )mj
j=1

where g is analytic on Ω and nonzero on Dr (z0 ). Since the set of zeros of g

is closed, there exists R > r such that DR (z0 ) ⊂ Ω and g has no zeros on
DR (z0 ).
On DR (z0 ) \ {z1 , . . . , zk },
k
f′ g ′ X mj
= + (7.4)
f g z − zj
j=1
′
Since g is holomorphic and nonzero on DR (z0 ), gg is holomorphic there, and
since DR (z0 ) is simply connected, Cauchy’s integral theorem gives
g ′ (z)
I
dz = 0.
|z−z0 |=r g(z)
120 7. THE ARGUMENT PRINCIPLE AND CONSEQUENCES

By the Cauchy integral formula, for any j,

I
mj
dz = 2πimj .
|z−z0 |=r z − zj

Summing up these equalities gives (7.3). □

Multiplying (7.4) by a function q analytic on a neighborhood of Dr (z0 )
and then applying similar arguments gives the following generalization:
Theorem 7.15 (Generalized argument principle). In the setting of The-
orem 7.14, if q is analytic on Ω, then
k
q(z)f ′ (z)
I
1 X
dz = mj q(zj ). (7.5)
2πi |z−z0 |=r f (z)
j=1

For instance, using this with q(z) = z gives a value for the sum of zeros;
this special case will be important later.
The argument principle has a geometric interpretation in terms of wind-
ing numbers:
Theorem 7.16 (Argument principle). Let f be analytic in Ω, Dr (z0 ) ⊂
Ω, and f ̸= 0 on ∂Dr (z0 ). Then
NDr (z0 ) (f ) = n(f ◦ γ, 0), (7.6)
where γ(t) = z0 + re2πit , t ∈ [0, 1].
Proof. Since the circle γ is a C 1 curve, so is the curve f ◦ γ. By
Theorem 7.14,
I ′ Z 1 ′
1 f (z) 1 f (γ(t)) ′
NDr (z0 ) (f ) = dz = γ (t)dt.
2πi γ f (z) 2πi 0 f (γ(t))
Using the chain rule and Lemma 7.13, we can write this as
(f ◦ γ)′ (t)
Z 1
1
NDr (z0 ) (f ) = dt = n(f ◦ γ, 0). □
2πi 0 (f ◦ γ)(t)
The simplest illustration of this is the function f (z) = z n and unit circle
γ(t) = e2πit : the curve f ◦ γ winds around the origin n times.
Theorem 7.17 (Rouché’s theorem for the disk). Let f, g be analytic in
a neighborhood Ω of Dr (z0 ) and suppose that
|f (z) − g(z)| < |f (z)| + |g(z)|, ∀z ∈ ∂Dr (z0 ). (7.7)
Then
NDr (z0 ) (f ) = NDr (z0 ) (f ) = NDr (z0 ) (g) = NDr (z0 ) (g).
Proof. Note, first of all, that (7.7) implies that f and g have no zeros
on the circle ∂Dr (z0 ). Using again the curve γ(t) = z0 + re2πit , t ∈ [0, 1] and
applying Lemma 7.8 to the curves f ◦ γ and g ◦ γ, we conclude n(f ◦ γ, 0) =
n(g ◦ γ, 0) so by the argument principle, NDr (z0 ) (f ) = NDr (z0 ) (g). □
 3. HURWITZ’S THEOREM 121

Theorem 7.18 (Fundamental theorem of algebra). Any polynomial of

degree d with complex coefficients has precisely d zeros in C (counted with
multiplicity).
Pd k
Proof. Consider a polynomial P (z) = k=0 ak z with ad ̸= 0. Fix
r > 0 and estimate for |z| = r
d−1
X d−1
X
d k
|P (z) − ad z | = ak z ≤ |ak |rk .
k=0 k=0

Since
d−1 d−1
1 X k
X |ak |
lim d
|ak |r = lim = 0,
r→∞ |ad |r r→∞ |ad |rd−k
k=0 k=0
for all r large enough, this quantity is smaller than 1 so
d−1
X
|P (z) − ad z d | ≤ |ak |rk ≤ |ad |rd < |P (z)| + |ad z d |.
k=0

Thus, by Rouché’s theorem, for all r large enough,

NDr (0) (P ) = NDr (0) (z d ) = d.
It follows that NC (P ) = d. □
It is an algebraic fact that if P (z1 ) = 0, then P can be factorized as
P (z) = (z − z1 )Q(z) for some polynomial Q with deg Q = deg P − 1. Thus,
factoring linear factors from P , we obtain:
Corollary 7.19. Every nontrivial polynomial P with complex coeffi-
cients has a factorization
d
Y
P (z) = C (z − zj )
j=1

for some C ∈ C \ {0} and z1 , . . . , zd ∈ C.

3. Hurwitz’s theorem
Assume that fn → f in A(Ω). If zn are zeros of fn such that zn → z in
Ω, then f (z) = 0 by uniform convergence. Hurwitz’s theorem is a kind of
converse: it tells us that for any zero z of f , there are zeros of fn converging
to z, with matching multiplicities.
Theorem 7.20 (Hurwitz’s theorem). Let fn → f in A(Ω) where Ω is
a domain in C. Assume f is not identically zero. Let z0 be a zero of f of
order m; we allow the case m = 0 meaning f (z0 ) ̸= 0. Then there exists
r > 0 such that Dr (z0 ) ⊂ Ω, f is nonzero on Dr (z0 ) \ {z0 }, and for every
δ ∈ (0, r], there exists n0 (δ) such that for all n ≥ n0 (δ), NDδ (z0 ) (fn ) = m.
122 7. THE ARGUMENT PRINCIPLE AND CONSEQUENCES

Proof. If m > 0, since zeros of f are isolated, there exists r > 0 such
that
∀z ∈ Dr (z0 ) \ {z0 } f (z) ̸= 0.
If m = 0, the same holds by continuity of f .
Fix δ ∈ (0, r]. Since fn → f , for large enough n,
|fn − f | < |f |
on the circle |z − z0 | = δ. By Rouché’s theorem, for large enough n,
NDδ (z0 ) (fn ) = NDδ (z0 ) (f ). □
Corollary 7.21. If fn are injective maps and fn → f , then f is injec-
tive or f is constant.
Proof. Assume that f is nonconstant and takes some value w at least
twice. Pick z1 , z2 such that f (z1 ) = f (z2 ) = w and z1 ̸= z2 . Fix r > 0 such
that Dr (z1 ) and Dr (z2 ) are mutually disjoint, contained in Ω, and f (z) ̸= w
for |z − zj | = r (this holds for all small enough r > 0).
By the Hurwitz theorem, for all large enough n, fn − w has a zero in
Dr (zj ) for j = 1, 2, so fn is not injective. □
Corollary 7.22. If fn are nonvanishing functions on Ω and fn → f ,
then f = 0 or f is nonvanishing.
Proof. If f is not identically zero but has a zero, then fn have nearby
zeros, causing a contradiction. □
Theorem 7.20 is a local statement, since it applies to small disks. It can
be used to prove a global version of Hurwitz’s theorem (Exercise 6).

4. Inverse function theorem

We will now describe more precisely the local behavior of analytic func-
tion f : Ω → C near a point z0 ∈ Ω. A point z0 ∈ Ω is said to be noncritical
if f ′ (z0 ) ̸= 0; we begin by considering this case.
Theorem 7.23 (Inverse function theorem). Let f : Ω → C be analytic
and z0 ∈ Ω such that f ′ (z0 ) ̸= 0. Denote w0 = f (z0 ). There exists ϵ > 0
and an open set U , z0 ∈ U ⊂ Ω such that
f |U : U → Dϵ (w0 )
is a bijection, and the inverse function g = (f |U )−1 is analytic and obeys
1
g ′ (w) = ′
f (g(w))
Proof. Since zeros of f − w0 are isolated, we can pick δ > 0 such that
f (z) − w0 ̸= 0 ∀z ∈ Dδ (z0 ) \ {z0 }
and then choose
ϵ= inf |f (z) − w0 |.
z∈∂Dδ (z0 )
 4. INVERSE FUNCTION THEOREM 123

With these choices, for all w ∈ Dϵ (w0 ) and z ∈ ∂Dδ (z0 ), we have
|(f (z) − w) − (f (z) − w0 )| < ϵ ≤ |f (z) − w0 |
Thus, Rouché’s theorem implies for any w ∈ Dϵ (w0 ) that
NDδ (z0 ) (f − w) = NDδ (z0 ) (f − w) = NDδ (z0 ) (f − w0 ) = 1.
We then choose
U = Dδ (z0 ) ∩ f −1 (Dϵ (w0 )),
which is open as an intersection of two open sets. For any w ∈ Dϵ (w0 ), f −w
has one zero in Dδ (z0 ) and that zero is simple; thus, f : U → Dϵ (w0 ) is a
bijection and f ′ ̸= 0 on U .
Since we have established above that f (z) ̸= w for |z − z0 | = δ and
|w − w0 | < ϵ, we can define a function g : Dϵ (w0 ) → C by
zf ′ (z)
I
1
g(w) = dz. (7.8)
2πi |z−z0 |=δ f (z) − w
By Theorem 6.8, g is an analytic function.
By the generalized argument principle (Theorem 7.15) with q(z) = z,
g(w) is the sum of all zeros of f (z)−w in Dδ (z0 ), counted with their multiplic-
ities; in our situation, this means that g(w) is the unique zero of f (z) − w in
Dδ (z0 ). Thus, f (g(w)) = w for w ∈ Dϵ (w0 ), or in other words, g = (f |U )−1 .
Applying the chain rule to f (g(w)) = w gives f ′ (g(w))g ′ (w) = 1 so
g (w) = 1/f ′ (g(w)).
′ □
Using the inverse function theorem, we can also describe the local be-
havior of an analytic function at a critical point:
Theorem 7.24. Let f : Ω → C be an analytic function. Let z0 ∈ Ω,
w0 = f (z0 ), and let z0 be a zero of multiplicity m for f − w0 , i.e.,
f (z0 ) = w0 , f ′ (z0 ) = . . . f (m−1) (z0 ) = 0, f (m) (z0 ) ̸= 0.
Then there exist ϵ > 0, an open set U with z0 ∈ U ⊂ Ω, and an analytic
bijection h : U → Dϵ (0) such that f (z) − w0 = h(z)m for all z ∈ U .
The inverse of h is an analytic function g : Dϵ (0) → U with g(0) = z0 ,
′
g (0) ̸= 0 such that for every w ∈ Dϵ (w0 ) \ {w0 }, there are exactly m distinct
zeros of f − w in U , and they are given by the formula
z = g((w − w0 )1/m )
by using the m possible values of (w − w0 )1/m .
Proof. Since f (z) − w0 has a zero of multiplicity m, factorizing gives
f (z) − w0 = (z − z0 )m q(z)
with q analytic on Ω and nonzero at z0 . Pick r > 0 so that q is nonvanishing
on Dr (z0 ); on that disk, there is an analytic function q1 such that q = q1m ,
so by denoting h(z) = (z − z0 )q1 (z), we have
f (z) − w0 = h(z)m , h(z0 ) = 0, h′ (z0 ) ̸= 0.
124 7. THE ARGUMENT PRINCIPLE AND CONSEQUENCES

Then f (z) = w is equivalent to w − w0 = h(z)m and to h(z) = (w − w0 )1/m .

By the inverse function theorem applied to h at its noncritical point z0 ,
there exist δ ∈ (0, r) and ϵ > 0 such that for every η ∈ Dϵ (0), h − η
has exactly one zero in Dδ (z0 ). Moreover, restricting to the open set U =
h−1 (Dϵ (0)) ∩ Dδ (z0 ), the map h becomes an analytic bijection. The other
properties follow. □
Since
P∞ g is nanalytic at 0, it has a power series representation g(η) =
z0 + n=1 bn η with radius of convergence at least ϵ. Combining this with
the previous theorem, the solutions of f (z) = w near z0 are given by a
so-called Puiseux series,
∞
X
z = z0 + bn ((w − w0 )1/m )n
n=1

(m possible values of (w − w0 )1/m correspond to m solutions).

5. Bonus: implicit function theorem

The inverse function theorem is closely related to the implicit function
theorem; to state the implicit function theorem, we need to define analytic
functions of 2 variables.
Definition 7.25. Consider an open set S ⊂ C × C. The function F :
S → C is analytic at (z0 , w0 ) if there exists r > 0 and anm ∈ C such that
∞ X
X ∞
|anm |rn+m < ∞
n=0 m=0
and
∞ X
X ∞
F (z, w) = anm (z − z0 )n (w − w0 )m , ∀(z, w) ∈ Dr (z0 ) × Dr (w0 ).
n=0 m=0

Theorem 7.26 (Analytic implicit function theorem). If F is analytic at

(z0 , w0 ) and
F (z0 , w0 ) = 0, (∂z F )(z0 , w0 ) ̸= 0,
then there exist ϵ, δ > 0 and analytic g : Dϵ (w0 ) → Dδ (z0 ) such that Dδ (z0 )×
Dϵ (w0 ) ⊂ S and
F (g(w), w) = 0, ∀w ∈ Dϵ (w0 ).
Moreover,
∂w F (g(w), w)
g ′ (w) = − .
∂z F (g(w), w)
Proof. Since F (·, w0 ) has a first order zero at z0 , there exists δ > 0
such that
|z − z0 | ≤ δ, z ̸= z0 =⇒ |F (z, w0 )| > 0.
 6. OPEN MAPPING, THE MAXIMUM PRINCIPLE, AND THE SCHWARZ LEMMA125

This implies on the boundary

|z − z0 | = δ =⇒ |F (z, w0 )| > 0.
By uniform continuity of F , there exists ϵ > 0 so that
|z − z0 | = δ, |w − w0 | ≤ ϵ =⇒ |F (z, w)| > 0.
Use this to define on w ∈ Dδ (w0 )
I
1 (∂z F )(z, w)
N (w) = dz
2πi |z−z0 |=δ F (z, w)
I
1 z(∂z F )(z, w)
g(w) = dz
2πi |z−z0 |=δ F (z, w)
Similarly as before, since N is continuous in w and N (w0 ) = 1, it follows that
N (w) = 1 for all w. Thus, F (z, w) = 0 has a unique zero with |z − z0 | < δ
and that zero is simple.
Then g(w) gives the value of that zero. Analyticity of g follows from
that of F . Applying the chain rule to ∂w [F (g(w), w)] = ∂w 0 = 0 gives the
formula for g ′ . □

6. Open mapping, the maximum principle, and the Schwarz

lemma
Theorem 7.27 (Open mapping theorem). Let f : Ω → C be a noncon-
stant analytic function. Then f (Ω) is an open set.
Proof. If f (z0 ) = w0 , by the local behavior of analytic functions, there
exists ϵ > 0 such that f takes all values in Dϵ (w0 ). Thus, f (Ω) is open. □
Theorem 7.28 (Maximum principle). Let f : Ω → C be a nonconstant
analytic function. Then |f | has no local maximum, i.e., for any z0 ∈ Ω and
δ > 0 there exists z ∈ Dδ (z0 ) such that |f (z)| > |f (z0 )|.
Proof. Write f (z0 ) = reiφ . By the open mapping theorem, f (Dδ (z0 ))
is open, so it contains (r + κ)eiφ for small enough κ > 0. □

There is a corresponding minimum principle, which is slightly different:

Theorem 7.29 (Minimum principle). Let f : Ω → C be a nonconstant
analytic function. Every local minimum of |f | is a zero of f , i.e., if z0 ∈ Ω
and δ > 0 are such that for all z ∈ Dδ (z0 ), |f (z)| ≥ |f (z0 )|, then f (z0 ) = 0.
Proof. If f (z0 ) ̸= 0, applying the maximum principle to 1/f in a neigh-
borhood of z0 shows that it cannot have a local minimum there. □

Taking contrapositives of these principles, they can be turned into cri-

teria for when an analytic function is constant. The maximum principle is
often used in the following form:
126 7. THE ARGUMENT PRINCIPLE AND CONSEQUENCES

Theorem 7.30. Let f be analytic on a bounded domain Ω, and contin-

uous on Ω. Then
∥f ∥Ω = ∥f ∥∂Ω .
Moreover, if for any z ∈ Ω, |f (z)| = ∥f ∥∂Ω , then f is constant.
Proof. Since Ω is bounded, the sets Ω, ∂Ω are compact, so |f | achieves
a global maximum on each of them. By the maximum principle, if f is
nonconstant, |f | cannot have a local maximum on Ω. □
Lemma 7.31 (Schwarz lemma). If f is a bounded analytic function on
D and f (0) = 0, then f has a factorization
f (z) = zg(z)
where g is a bounded analytic function on D and
∥g∥D = ∥f ∥D . (7.9)
In particular,
|f (z)| ≤ |z|∥f ∥D , ∀z ∈ D (7.10)
and
|f ′ (0)| ≤ ∥f ∥D . (7.11)
Equality holds if and only if f (z) = cz for some c ∈ C.
Proof. Denote M = ∥f ∥D . Since f (0) = 0, the function
(
f (z)/z z ∈ D \ {0}
g(z) =
f ′ (0) z=0
is analytic on D. For any r ∈ (0, 1), we know that |z| = r implies |g(z)| ≤
M/r, so by the maximum principle,
M
sup |g(z)| ≤ .
z:|z|≤r r
For any z ∈ D, applying this with |z| ≤ r < 1 and taking the limit r ↑ 1
shows
|g(z)| ≤ M
which proves the inequalities (7.10), (7.11).
In particular, g is also bounded on D with ∥g|D ≤ ∥f ∥D . Moreover, since
|f (z)| = |z||g(z)| ≤ |g(z)|, ∥g∥D ≥ ∥f ∥D , so we conclude (7.9).
If equality holds in (7.10) or (7.11), this means that |g(z)| = M for some
z ∈ D. In other words, g has a local maximum, so it is constant. □
The Schwarz lemma is often formulated as the conclusion (7.10) and,
sometimes, (7.11). However, thinking of it as a factorization of f which
doesn’t affect the supremum norm is beneficial if we wish to iterate the
Schwarz lemma (Exercise 10).
Analytic functions f : D → D are sometimes called Schur functions. By
rescaling, the previous result can be trivially reformulated as:
 7. A FIRST LOOK AT CONFORMAL BIJECTIONS 127

Lemma 7.32. If f is a Schur function and f (0) = 0, then f (z)/z is a

Schur function.

7. A first look at conformal bijections

If Ψ : Ω → C is analytic and injective, we know that
• Π = Ψ(Ω) is open and connected, so a domain
• Ψ′ ̸= 0 on Ω, by the local behavior of Ψ near critical points
• Ψ−1 : Π → Ω is analytic
A map Ψ : Ω → Π is called a conformal bijection if it is holomorphic
and a bijection.
If Ψ is a conformal bijection, then Ψ−1 : Π → Ω is a conformal bijection.
A composition of conformal bijections is a conformal bijection.
Definition 7.33. If there exists a conformal bijection from Ω to Π,
then Ω and Π are said to be conformally equivalent. This is an equivalence
relation.
Conformal bijections from Ω to itself are called automorphisms of Ω.
They form a group denoted Aut(Ω).
Conformal bijections are, in a sense, geometric transformations which
preserve analyticity. They can be viewed as a natural class of substitutions
(changes of variables) in the complex analytic setting, so it should not be
surprising that they are a useful tool for analysis.

Examples.
Example 7.34. For any α ∈ C \ {0}, β ∈ C, ψ(z) = αz + β is an
automorphism of C.
Of course, we can get additional conformal bijections by restriction. For
example:
Example 7.35. For any α > 0, β ∈ R, ψ(z) = αz+β is an automorphism
of C+ .
Example 7.36. The function ψ(z) = −1/z is an automorphism of C+ .
Example 7.37 (Cayley transform). The function
z−i
ψ(z) = (7.12)
z+i
is an analytic bijection from C+ to D. Its inverse is
w+1
ψ −1 (w) = −i . (7.13)
w−1
Proof. The function ψ is well defined and holomorphic on C \ {−i}.
Moreover, ψ(z) ̸= 1 for all z ∈ C \ {−i}, so let’s view it as a map ψ :
C \ {−i} → C \ {1}.
128 7. THE ARGUMENT PRINCIPLE AND CONSEQUENCES

For any z ∈ C \ {−i}, w ∈ C \ {1},

z−i i + iw w+1
= w ⇐⇒ z−i = zw+iw ⇐⇒ (1−w)z = i+iw ⇐⇒ z = = −i
z+i 1−w w−1
so ψ takes every value exactly once, i.e., it is a bijection.
Finally,
|z + i|2 − |z − i|2 4 Im z
1 − |ψ(z)|2 = 2
=
|z + i| |z + i|2
so z ∈ C+ if and only if ψ(z) ∈ D. □
√
Example 7.38. With the right choice of branch, z is a conformal map
from C+ to {z | Re z > 0, Im z > 0}. More generally, for any α ∈ (0, 2), z α
is a conformal map from C+ to {z | 0 < arg z < απ}.
For 0 ≤ r < R ≤ ∞, we define the annulus
Ar,R = {z ∈ C | r < |z| < R}.
For α < β ≤ α + 2π, we define the sector
Sα,β = {reiϕ | r > 0, α < ϕ < β}.
Theorem 7.39. For any 0 ≤ r < R ≤ ∞, α < β ≤ α + 2π, the region
Π = Ar,R ∩ Sα,β
is conformally equivalent to the rectangle
Ω = {z ∈ C | log r < Re z < log R and α < Im z < β}
with the map z 7→ ez (this allows r = 0 and R = ∞ with the conventions
log 0 = −∞, log ∞ = ∞).
Proof. We use the conformal equivalence z 7→ ez from
Ω = {z ∈ C | log r < Re z < log R and α < Im z < β}
to Ar,R ∩ Sα,β . This can be done even if r = 0, with the convention log r =
−∞, and even if R = ∞, with the convention log R = +∞. □
Example 7.40. tan maps {z ∈ C | − π2 < Re z < π
2} bijectively to
C \ {iy | y ∈ R, |y| ≥ 1}.
Proof. We rewrite
eiz −e−iz
sin z 2i e2iz − 1 2i
tan z = = eiz +e−iz
= −i 2iz
= −i + 2iz .
cos z e +1 e +1
2
The map z 7→ e2iz maps the strip {z ∈ C | − π2 < Re z < π2 } bijectively to
C\(−∞, 0]. The map z 7→ 2/(z +1) maps that bijectively to C+ ∪(0, 2)∪C− .
Subtracting 1 and multiplying by i completes the proof. □
Example 7.41. The Zhukovsky map
1
Ψ(z) = z +
z
is a conformal bijection from C± to C+ ∪ (−2, 2) ∪ C− .
 7. A FIRST LOOK AT CONFORMAL BIJECTIONS 129

Proof. The function Ψ : C \ {0} → C is analytic. For any w ∈ C,

√
2 w ± w2 − 4
Ψ(z) = w ⇐⇒ z − wz + 1 = 0 ⇐⇒ z = .
2
Since Ψ(z) = Ψ(1/z), if one solution is in C± , the other is in C∓ . The two
solutions are real if and only if w ∈ (−∞, −2] ∪ [2, ∞).
Thus, Ψ is a bijection from C+ to C+ ∪ (−2, 2) ∪ C− . □

Example 7.42. The function sin z is a conformal map from the strip
{z ∈ C | −π/2 < Re z < π/2} to C+ ∪ (−1, 1) ∪ C− .

Proof. In terms of the Zhukovsky map, sin can be expressed as

eiz − e−iz 1
sin z = = Ψ(−ieiz ).
2i 2
Using properties of the exponential function, we see that z 7→ −ieiz
maps the half-strip {z ∈ C | −π/2 < Re z < π/2} bijectively to C− .
Zhukovsky’s map is a bijection C− → C+ ∪ (−2, 2) ∪ C− . Composing
the two bijections and dividing by 2 completes the proof. □

For further insight about the Zhukovsky map, note that z = et+iθ maps
to
w = z + 1/z = 2 cosh t cos θ + 2i sinh t sin θ.
In particular, fixing t and varying θ, we see that circles centered at 0 map
to ellipses, and that rays starting from 0 map to hyperbolas.

Obstacles.

Proposition 7.43. If Ω is conformally equivalent to D, it is simply

connected.

Proof. Assume that there exists a conformal bijection f : Ω → D. For

any closed curves γ0 , γ1 in Ω, the loop homotopy F of their images f ◦ γ0 ,
f ◦ γ1 generates a loop homotopy f −1 ◦ F of γ0 , γ1 . This only uses continuity
of f , f −1 . Thus, Ω is simply connected. □

The contraposition of this statement provides a topological obstacle for

conformal equivalence: if Ω is not simply connected, it is not conformally
equivalent to D. There is also an analytical obstacle:

Theorem 7.44. D is not conformally equivalent to C.

Proof. If it were, there would be a conformal bijection Ψ : C → D.

Then Ψ would be a bounded entire function, so a constant, so not a bijection,
which is a contradiction. □
130 7. THE ARGUMENT PRINCIPLE AND CONSEQUENCES

Automorphisms of the disk. When a conformal bijection between

domains exists, it is usually not unique. We illustrate this by describing all
bijections of the disk to itself:
Theorem 7.45. A function f is a conformal bijection f : D → D if and
only if it is of the form
z−w
f (z) = eiϕ (7.14)
1 − wz
for some w ∈ D and ϕ ∈ R.
Proof. Let us denote
z−w
βw (z) = .
1 − wz
This is an analytic function on the domain C \ {1/w}, which contains D.
Moreover, for |z| = 1,
2
z−w |z|2 + |w|2 − 2 Re(zw)
|βw (z)|2 = = = 1.
1 − wz 1 + |z|2 |w|2 − 2 Re(zw)
By the maximum principle, |z| < 1 implies |βw (z)| < 1.
Since βw has a simple zero at w and no other zeros, by the argument
principle, the curve γ(t) = βw (eit ), t ∈ [0, 2π] has winding number n(γ, 0) =
1.
Since γ takes only values on ∂D, it has constant winding number on D,
i.e., for every η ∈ D,
ND (βw − η) = ND (βw ) = 1.
Thus, βw is a bijection from D to D.
It follows that eiϕ βw is also a conformal bijection D → D for any ϕ ∈ R.
Conversely, let f : D → D be a conformal bijection; then it has exactly
one zero, and we denote the zero by w. Then g = f ◦ βw −1 is also a conformal

bijection. Moreover,
−1
g(0) = f (βw (0)) = f (w) = 0.
Thus, the Schwarz lemma applies to both functions g and g −1 , and gives
|g ′ (0)| ≤ 1 and |(g −1 )′ (0)| ≤ 1. Due to the chain rule, the second inequality
turns into |g ′ (0)| ≥ 1, and the two inequalities combine into the case of
equality |g ′ (0)| = 1. By the Schwarz lemma, the cases of equality are g(z) =
cz, and since g ∈ Aut(D), c = eiϕ for some ϕ ∈ R. Thus, f = g ◦ βw is of
the form (7.14). □
If we wish to describe a unique automorphism of the disk, we have to
impose some additional conditions. For instance:
Corollary 7.46. For any w ∈ D, there is a unique f ∈ Aut(D) with
f (w) = 0 and f ′ (w) > 0.
Proof. Such f must be of the form f = eiϕ βw , and the value of ϕ
is uniquely determined by f ′ (w) = eiϕ βw ′ (w) > 0 (recall that conformal
′
bijections have no critical points, so βw (w) ̸= 0). □
 8. THE RIEMANN MAPPING THEOREM 131

8. The Riemann mapping theorem

In this section, we will completely answer the question: which domains
Ω are conformally equivalent to D?
Definition 7.47. A domain Ω ⊂ C is said to have the logarithm prop-
erty if for every analytic f : Ω → C \ {0}, there exists g ∈ A(Ω) such that
f = eg .
A domain Ω ⊂ C is said to have the square root property if for every
analytic f : Ω → C \ {0}, there exists h ∈ A(Ω) such that f = h2 .
Recall that any domain in which integrals of holomorphic functions over
closed curves are zero has the logarithm property, and that the logarithm
property implies the square root property.
Theorem 7.48 (Riemann mapping theorem). If Ω ⊂ C has the square
root property and Ω ̸= C, then there exists a conformal bijection
Θ : Ω → D.
For any z0 ∈ Ω, there is a unique such Θ with Θ(z0 ) = 0 and Θ′ (z0 ) > 0.
The proof uses the family of functions
R = {f ∈ A(Ω) | f is injective, ∥f ∥Ω ≤ 1, f (z0 ) = 0, f ′ (z0 ) > 0},
and we begin by studying the properties of this family; throughout this sec-
tion, we assume Ω is as in the statement of the Riemann mapping theorem.
Lemma 7.49. R is nonempty.
Note that if Ω is bounded, it suffices to take f (z) = a(z − z0 ) for small
enough a > 0; if Ω is unbounded, we should reduce to the bounded case by
using the square root property and Ω ̸= C.
Proof. Fix a point w ∈ C \ Ω. Then the function
z−w
h(z) =
z0 − w
is nonzero on the domain Ω, so it has a square root, an analytic function
g : Ω → C such that g 2 = h. Since h(z0 ) = 1, without loss of generality
we can choose g(z0 ) = 1. Then there exists ϵ > 0 such that Dϵ (1) ⊂ g(Ω).
Since h = g 2 is injective, it follows that Dϵ (−1) ∩ g(Ω) = ∅. Therefore, the
map
ϵ
f (z) =
g(z) + 1
is analytic on Ω, injective, and |f (z)| ≤ 1 for all z ∈ Ω. For a suitable
β ∈ Aut(D), β ◦ f ∈ R, so R is nonempty. □
Lemma 7.50. There exists a function Θ ∈ R which maximizes Θ′ (z0 ),
i.e., for every f ∈ R, f ′ (z0 ) ≤ Θ′ (z0 ).
132 7. THE ARGUMENT PRINCIPLE AND CONSEQUENCES

First proof, using the compactness formulation of Montel’s theorem.

For any f ∈ R, n1 f → 0, so 0 ∈ R.
If fn ∈ R∪{0} converge to f and f ̸= 0, then fn ∈ R for all large enough
n. Since limits of analytic functions are analytic, f ∈ A(Ω). Trivially,
∥f ∥Ω ≤ 1, f (z0 ) = 0, and f ′ (z0 ) ≥ 0. Since f ̸= 0 and f (z0 ) = 0, f is
nonconstant, so by a corollary to Hurwitz’s theorem (Corollary 7.21), f is
injective. Injectivity implies f ′ (z0 ) ̸= 0. Thus, f ∈ R.
Thus, R ∪ {0} is closed and contained in R, so it is equal to R. By
Montel’s theorem, this set is compact.
Since differentiation and point evaluation are continuous on A(Ω), the
image of R under f 7→ f ′ (z0 ) is compact, so it has a maximum. □
Second, more elementary proof. For some r > 0, Dr (z0 ) ⊂ Ω, so
by a Cauchy estimate,
1 1
f ′ (z0 ) ≤ ∥f ∥∂Dr (z0 ) ≤ .
r r
Thus, M = supf ∈R f ′ (z0 ) < ∞. There exist fn ∈ R such that fn′ (z0 ) > M −
1/n. By Theorem 6.2, the sequence has a subsequence fnk converging uni-
formly on compacts to some function Θ. Since Θ′ (z0 ) = limk→∞ fnk (z0 ) ≥
M , we conclude Θ is nonconstant. As a limit of injective functions, it is
injective. Thus, Θ ∈ R, so Θ′ (z0 ) ≤ M . We conclude Θ′ (z0 ) = M . □
Lemma 7.51. Let f ∈ R and f (Ω) ̸= D. Then there exists g ∈ R and
φ : D → D with φ′ (0) < 1 such that
f = φ ◦ g.
z−w0
Proof. Let w0 ∈ D \ f (Ω). Let β(z) = 1−w 0z
. Then β ◦ f is nonva-
nishing, so it has a branch of square root which we denote r. Note that
f (z) = β −1 (r(z)2 ). Since f is injective, so is r. In particular, r′ is nonvan-
ishing.
For any α ∈ Aut(D), the function
g =α◦r
is injective and maps Ω to D. If we pick
φ(w) = β −1 (α−1 (w)2 ),
then φ : D → D and f = φ ◦ g.
In particular, if we choose α to be
|r′ (z0 )| w − r(z0 )
α(w) =
r′ (z0 ) 1 − r(z0 )w
then g(z0 ) = α(r(z0 )) = 0 and g ′ (z0 ) = α′ (r(z0 ))r′ (z0 ) > 0. Thus, g ∈ R.
Moreover, this implies φ(0) = 0 and φ′ (0) > 0 by the chain rule.
Obviously α−1 (w)2 has at least a double zero at α(0), so φ(z) − φ(α(0))
has at least a double zero at α(0). In particular, φ(z) is not a constant
multiple of z, so φ′ (0) < 1 by the Schwarz lemma. □
 9. EXERCISES 133

Proof of the Riemann mapping theorem. By the maximum prin-

ciple, Θ(Ω) ⊂ D. If Θ was not surjective, by the previous lemma, it could
not be the maximizer of Θ′ (z0 ).
If Θ̃ is another function with the same properties, then f = Θ ◦ Θ̃−1 ∈
Aut(D) obeys f (0) = 0 and f ′ (0) > 0, so f is the identity map and Θ̃ =
Θ. □
Corollary 7.52. For a domain Ω in C, the following are equivalent:
(a) Ω is simply connected
1
(b) H holomorphic f : Ω → C and every closed piecewise C curve γ
for every
on Ω, γ f (z) dz = 0
(c) Ω has the logarithm property
(d) Ω has the square root property
Proof. We proved (a) =⇒ (b) =⇒ (c) =⇒ (d) before.
(d) =⇒ (a): If Ω = C this is trivial. Otherwise, by the Riemann mapping
theorem, there is a conformal bijection Θ : D → Ω. By Prop. 7.43, Ω is
simply connected. □
One purely topological consequence of the Riemann mapping theorem
is:
Theorem 7.53. Let Ω be a simply connected domain. For any z0 ∈ Ω,
two curves are loop homotopic in Ω \ {z0 } if and only if they have the same
winding number about z0 .
Proof. If Ω = C, use the map Θ(z) = z − z0 . Otherwise, use a confor-
mal bijection Θ : Ω → D with Θ(z0 ) = 0. This reduces the statement to the
special case C \ {0}, D \ {0}, see Theorem 7.9. □

9. Exercises
1. How many zeros (counted with multiplicity) does f (z) = z 12 −4z 8 +9z 5 −
2z + 1 have in the disk {z | |z| < 1}?
2. How many zeros (counted with multiplicity) does f (z) = z 4 − 6z + 3 have
in the annulus {z | 1 < |z| < 2}?
3. How many zeros does f (z) = 9z 10 − e2z have in D? What are their
multiplicities?
4. Let f be analytic in some open set containing D. Prove that
1
sup f (z) − ≥ 1.
z∈∂D z
5. Let f be analytic on D and supz∈D |f (z)| < 1. Prove that f has a unique
fixed point (a unique point z ∈ D such that f (z) = z).
Hint: count fixed points on disks Dr (0), with supz∈D |f (z)| < r < 1.
6. Let fn → f in A(Ω) where Ω is a domain in C. Assume f is not identically
zero. If K ⊂ Ω is compact and f has no zeros on ∂K, denote by m the
number of zeros of f in K, counted with multiplicity; then there exists
 134 7. THE ARGUMENT PRINCIPLE AND CONSEQUENCES

n0 such that for all n > n0 , fn has precisely m zeros in K, counted with
multiplicity.
7. Let f ∈ A(Ω) be nonconstant and z0 a zero of multiplicity m ≥ 1.
Prove that there exists an open neighborhood U of z0 and m open curves
γ1 , . . . , γm in U (curves parametrized by open intervals) such that:
(a) curves γj all contain z0
(b) have pairwise no other intersections
(c) for z ∈ U , f (z) ∈ R if and only if z lies on the union of curves
8. Let f : Ω1 → Ω2 , g : Ω2 → C be nonconstant analytic functions. Prove
that g ◦ f : Ω1 → C is also nonconstant.
9. Let f : Ω → C be a nonconstant analytic function. Prove that for every
m ∈ N, the set
{w ∈ C | f −1 (w) contains at least m points}
is open.
Hint: emulate the proof of the open mapping theorem.
10. Let f : D → D be an analytic function with f (0) = f ′ (0) = 0. Prove that
|f (z)| ≤ |z|2 , ∀z ∈ D
and |f ′′ (0)|
≤ 2, and that if equality holds in one of these equalities (for
a single z), then f (z) = cz 2 for some c ∈ ∂D.
11. For any w ∈ D \ {0} define
|w| w − z
βw (z) = .
w 1 − wz
(note: the factor |w|/w is a unimodular constant not important in this
problem; for infinite products, it helps their convergence)
(a) Prove that βw is analytic in a neighborhood of D and that |z| = 1
implies |βw (z)| = 1.
(b) Prove that βw is a bijection from D to D.
Hint: Denote γ(t) = e2πit , t ∈ [0, 1]. Prove that n(βw ◦ γ, z) =
n(βw ◦γ, 0) for all z ∈ D and rule out other values by other arguments.
(c) Prove that
lim min |βz0 (z)| = 1.
r↑1 z:|z|=r

(d) If f is a Schur function (i.e. an analytic function f : D → D) and

f (z0 ) = 0, prove that
f (z) = βz0 (z)g(z)
for some Schur function g.
Hint: emulate the proof of the Schwarz lemma.
12. Use the Blaschke factors βw from the previous problem. If f is analytic
on some neighborhood of D and z ∈ ∂D implies f (z) ∈ ∂D, prove that f
is of the form
Yn
f =c βwj
j=1
 9. EXERCISES 135

for some n ∈ N0 , some c ∈ ∂D, and some w1 , . . . , wn ∈ D.

13. Find a conformal bijection from D to the half-plane {z ∈ C | Re z > 0}.
14. Prove that sin is a conformal map of {z ∈ C | −π/2 < Re z < π/2 and Im z >
0} onto C+ .
15. Find a conformal bijection from the region
{z ∈ C | Re z > 0} \ (0, 1]
to the region
{z ∈ C | Re z > 0}
Hint: construct this map as a composition of several simpler conformal
bijections.
16. Find a conformal bijection from {z ∈ C | Re z > 0, Im z > 0, |z| > 1} to
{z ∈ C | Re z > 0, 0 < Im z < 1}.
17. Find a conformal bijection from {z ∈ C | Re z > 0, Im z > 0, |z| > 1} to
{z ∈ C | Re z > 0, Im z > 0}.
18. Let f : D → D be an analytic function. Prove that
f (z) − f (w) z−w
≤ , ∀z, w ∈ D
1 − f (w)f (z) 1 − wz
and that equality holds if and only if f ∈ Aut(D).
Hint: use automorphisms of D to reduce to the case w = 0, f (w) = 0.
Remark: the hyperbolic metric on D is the function
z−w
ρ(z, w) = 2 arctanh .
1 − z̄w
In this problem, you proved that analytic functions f : D → D can only
decrease the hyperbolic metric, and automorphisms preserve it.
19. If f : C+ → C+ is analytic and f (i) = i, prove that |f ′ (i)| ≤ 1. Find the
largest possible value of |f (i/2)|.
20. Let D denote the open unit disk in C and consider a sequence of holomor-
phic functions fn : D → D. Assume that fn (0) → 0 and |fn (1/2)| → 1/2
as n → ∞. Prove that for each w ∈ D, there exists n0 such that w ∈ fn (D)
for all n ≥ n0 .
21. Let Ω ⊊ C be a simply connected domain. If f : Ω → Ω is analytic and,
for some z0 ∈ Ω, f (z0 ) = z0 and |f ′ (z0 )| = 1, prove that f is a bijection.
Hint: use a Riemann mapping Θ : Ω → D with Θ(z0 ) = 0.
22. Let f be a conformal mapping of D to the ellipse {z ∈ C | (Re z)2 /a2 +
(Im z)2 /b2 < 1} with the conditions f (0) = 0, f ′ (0) > 0. Prove that
f (n) (0) ∈ R for all n and f (n) (0) = 0 for even n.
 CHAPTER 8

Meromorphic functions and the residue theorem

We formulated many results with respect to a circle contour. Since disks

are neighborhoods of an arbitrary point in a domain, this was particularly
well suited for the study of local properties of analytic functions.
However, we naturally encounter functions which are infinite at certain
points, such as rational functions P/Q with P, Q polynomials, or tan. In-
stead of viewing them as analytic functions and working on the domain
on which they are finite, it will be beneficial to have a consistent way of
handling a class of functions which may be infinite at certain points; these
functions will be called meromorphic.
We will also take a global perspective on the Cauchy integral theorem,
Cauchy integral formula, and related results. This entails working with
more complicated contours rather than just circles, and it requires us to use
multiple contours, leading to the notion of chains.

1. Cauchy integral formula for the annulus

In an earlier chapter, we derived a Cauchy integral formula on a disk
Dr (z0 ). Since open disks are neighborhoods of a point z0 ∈ Ω, this was well
suited for the study of local behavior of f around a point in the domain.
We will now need a more general version, which uses an annulus. We need
to consider both circles |z| = R and |z| = ρ as parts of the boundary, and
integrate over both of them; the integral over the smaller circle naturally
appears with the opposite orientation. The precise statement is:
Theorem 8.1. Assume that Aρ,R ⊂ Ω for some 0 < ρ < R < ∞. If f
is analytic on Ω, then for any z ∈ Aρ,R ,
I I
1 f (ζ) 1 f (ζ)
f (z) = dζ − dζ. (8.1)
2πi |ζ|=R ζ − z 2πi |ζ|=ρ ζ − z
We will provide several different proofs, illustrating different tools de-
veloped so far.
Proof using star-shaped domains. Since Aρ,R ⊂ Ω is compact, it
is at nonzero distance from Ωc , i.e., for some ρ1 , R1 ,
Aρ,R ⊂ Aρ1 ,R1 ⊂ Ω.
We denote z = reiϕ and observe the domains
Ω± = {z ∈ C | ρ1 < |z| < R1 , ±ie−iϕ z ∈
/ [0, ∞)} \ {ze∓it | t ∈ [0, π/2)}
137
138 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

on which log is single-valued; moreover, log(Ω± ) is star-shaped with hub

log z ± iπ/2, so Ω± is simply connected. On Ω± , we observe closed contours
γ± consisting of semicircles and line segments, such that applying the Cauchy
integral theorem to γ± and summing gives
I I I
f (ζ) f (ζ) f (ζ)
dζ − dζ = dζ,
|ζ|=R ζ − z |ζ|=ρ ζ − z |ζ−z|=δ ζ − z

with 0 < δ < min{|z| − ρ, R − |z|}. Then we take δ → 0 as in the proof of

the Cauchy integral formula on the disk. □

Proof using loop homotopy. Denote z = reiϕ . Fix 0 < δ < min{|z|−
ρ, R − |z|}. Consider the closed contour γ0 consisting of two circles and a
line segment [−reiϕ , −Reiϕ ] in both directions. Parametrize it by t ∈ [0, 1]
so that γ0 (0) = Reiϕ and γ0 (1/2) = reiϕ .
We can represent γ0 uniquely in the form
γ0 (t) = eσ0 (t) ,
where σ0 is a rectangular curve with Im σ ∈ [ϕ − π, ϕ + π].
Likewise, we can parametrize the circle γ1 (t) = z + δe2πit = eσ1 (t) so that
σ1 is a curve in the punctured rectangle
K = {w | Re w ∈ [log r, log R], Im w ∈ [ϕ − π, ϕ + π]} \ {log z + iϕ}.
Since Im σ0 − ϕ and Im σ1 − ϕ have the same sign, linear interpolation be-
tween σ0 , σ1 is a loop homotopy in K, and exponentiating, we obtain a loop
homotopy between γ0 , γ1 . Thus, the contour integral is equal to contour
integral over |ζ − z| = δ, which is equal to 2πif (z) by the Cauchy integral
formula on the disk. □

Proof using factorization. By factorization, there is an analytic

function g on Ω such that
f (ζ) − f (z) = (ζ − z)g(ζ).
By the loop homotopy F (t, s) = (ρ + s(R − ρ))e2πit , we conclude
I I
g(ζ) dζ = g(ζ) dζ,
|ζ|=R |ζ|=ρ

which we rewrite as
f (ζ) − f (z) f (ζ) − f (z)
I I
1 1
dζ − dζ = 0. (8.2)
2πi |ζ|=R ζ −z 2πi |ζ|=ρ ζ −z
Meanwhile, using winding numbers of circles, we see that
I I
1 1 1 1
dζ − dζ = 1.
2πi |ζ|=R ζ − z 2πi |ζ|=ρ ζ − z
Multiplying this by f (z) and adding to (8.2), we conclude (8.1). □
 2. LAURENT SERIES 139

2. Laurent series
The goal of this section is to derive a Laurent series representation for
an analytic function in an annulus
Aρ,R = {z ∈ C | ρ < |z| < R}.
For notational convenience, we chose the center of the annulus to be 0. Un-
like a power series, a Laurent series contains negative powers of the complex
variable. To derive this representation, we need a few preliminaries.
The next preliminary are series representations of a Cauchy transform
inside and outside the circle:
Lemma 8.2. Let h : ∂Dr (0) → C be continuous and, for z ∈
/ ∂Dr (0),
I
1 h(ζ)
g(z) = dζ.
2πi |ζ|=r ζ − z
(a) There exist (an )∞
n=0 such that
∞
X
g(z) = an z n , |z| < r
n=0

with the series converging absolutely and uniformly for |z| ≤ ρ, for any
ρ<r
(b) There exist (bn )∞
n=0 such that
∞
X
g(z) = bn z −n−1 , |z| > r (8.3)
n=0

with the series converging absolutely and uniformly for |z| ≥ R, for any
R>r
Proof. (a) follows as in the proof of the Cauchy power series theorem
(Lemma 5.4).
(b): note that the substitution η = −1/ζ maps the positively oriented
contour |ζ| = r to the positively oriented contour |η| = 1/r. Using the
substitution rule for contour integrals (Theorem 3.48) with η = −1/ζ and
denoting w = −1/z, we have dζ = η −2 dη and
I
1 f (−1/η) dη
g(−1/w) = −
2πi |η|=ρ−1 (− η1 ) − (− w1 ) η 2
I
1 f (−1/η)/η
=w dη
2πi |η|=ρ−1 η − w
∞
X
=w bn wn
n=0

for |w| < ρ−1 , by (a). Rewriting in terms of z gives the required claims. □
140 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

Part (b) can also be proved by emulating the proof of the Cauchy power
series theorem, but instead of
∞
X
(ζ − z)−1 = ζ −1 (1 − zζ −1 )−1 = z n ζ −n−1 ,
n=0
using the geometric expansion
∞
X ∞
X
−1 −1 −1 −1 −1 n −n
(ζ − z) = −z (1 − ζz ) = −z ζ z = ζ n z −n−1
n=0 n=0

which is convergent for |z| > |ζ|.

P−1 k,
Remark 8.3. We will prefer to denote the series (8.3) by k=−∞ bk z
with the substitution k = −n − 1. Likewise, we will write
+∞
X ∞
X −1
X
k k
ak z := ak z + ak z k
k=−∞ k=0 k=−∞

in situations when both series on the right are absolutely convergent.

Theorem 8.4 (Laurent series). If f is analytic on Aρ,R for some 0 ≤
P∞ R ≤k ∞, there
ρ < P−1 exists ka sequence of ak , k ∈ Z such that the series
k=0 ak z and k=−∞ ak z converge absolutely and uniformly on compact
subsets of Aρ,R , and
+∞
X
f (z) = ak z k (8.4)
k=−∞
for all z ∈ Aρ,R . Moreover, the sequence with these properties is unique, and
given for any r ∈ (ρ, R) by
I
1
an = ζ −n−1 f (ζ) dζ. (8.5)
2πi |ζ|=r
Proof. Fix an interval [ρ1 , R1 ] ⊂ (ρ, R). Define
I
1 f (ζ)
f− (z) = dζ z ∈ DR1 (0)
2πi |ζ|=R1 ζ − z
I
1 f (ζ)
f+ (z) = − dζ z ∈ C \ Dρ1 (0)
2πi |ζ|=ρ1 ζ − z
By Theorem 8.1, f = f− + f+ on Aρ1 ,R1 .
By Lemma 8.2, there exist (an )∞ ∞
n=0 , (bn )n=0 , such that
∞
X ∞
X
f− (z) = an z n , f+ (z) = − bn z −n−1
n=0 n=0
uniformly on compact subsets of Aρ1 ,R1 . Denoting ak = −b−k−1 for k ≥ 0
and combining the two sums gives the Laurent series representation (8.4)
on Aρ1 ,R1 .
 3. CLASSIFICATION OF ISOLATED SINGULARITIES 141

Fix n ∈ Z and r ∈ (ρ, R). Since the two series converge uniformly and
absolutely on |z| = r, we can insert the Laurent series into a contour integral
and exchange the series and integral:
I XI X I
ζ −n−1 f (ζ) dζ = ζ −n−1 ak ζ k dζ = ak ζ k−n−1 dζ.
|ζ|=r k∈Z |ζ|=r k∈Z |ζ|=r

Recall that (
2πi m = −1
I
m
ζ dζ =
|ζ|=r 0 m ∈ Z \ {−1}
(for m ̸= −1 this follows from the fundamental theorem of calculus, since
z m+1 /(m + 1) is holomorphic on C \ {0}), so only the term k = n is nonzero
and leads to (8.5).
Thus, the sequence is unique. Since different choices of [ρ1 , R1 ] ⊂ (ρ, R)
give the same sequence, the same Laurent series (8.4) converges uniformly
on compact subsets of Aρ,R . □
Just as for power series, there are Cauchy estimates for coefficients of
Laurent series:
Corollary 8.5 (Cauchy estimates). For all ρ < r < R and n ∈ Z,
Z 2π
−n dθ
|an | ≤ r |f (reiθ )| ≤ r−n ∥f ∥∂Dr (0) .
0 2π
For |n| large, it is usually better to apply this for ρ near R if n > 0, and
for ρ near r if n < 0.

3. Classification of isolated singularities

Definition 8.6. An analytic function f : Ω → C is said to have an
isolated singularity at z0 if z0 ∈
/ Ω and there exists R > 0 such that DR (z0 ) \
{z0 } ⊂ Ω.
A punctured disk DR (z0 ) \ {z0 } is an annulus with inner radius ρ = 0,
so f has a Laurent series
+∞
X
f (z) = an (z − z0 )n , z ∈ DR (z0 ) \ {z0 }.
n=−∞
We denote
N = inf{n ∈ Z | an ̸= 0}
with the conventions inf ∅ = +∞ and inf A = −∞ if A is not bounded
below. Of course, when the inf is finite, it is a minimum, since Z is discrete.
This allows us to define the following classification. At z0 , the function
f is said to have:
(a) a removable singularity if N ≥ 0
(b) a pole if −∞ < N < 0. In this case, −N is called the order of the pole.
(c) an essential singularity if N = −∞
142 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

The term ”removable singularity” refers to the fact that we can then
define f (z0 ) = a0 and f becomes analytic on Ω ∪ {z0 }.
sin z
Example 8.7. f (z) = z has a removable singularity at 0.
Proof. Using the Taylor series for sin z, we obtain for all z ̸= 0
∞
X (−1)n z 2n
f (z) = .
(2n + 1)!
n=0
This is the Laurent series for f around 0, so we read off N = 0. □
z+1
Example 8.8. f (z) = z 3 (z 2 +1)
has a pole at 0 of order 3.

Proof. The function h(z) = zz+1

2 +1 is analytic in D so it has a Maclaurin
series
X∞
h(z) = an z n , ∀z ∈ D,
n=0
with a0 = h(0) ̸= 0. Thus, for z ∈ D \ {0},
∞
X ∞
X
n−3
f (z) = an z = bk z k
n=0 k=−3

with bk = ak+3 and, in particular, b−3 ̸= 0. Thus, N = −3. □

Example 8.9. f (z) = e1/z has an essential singularity at 0.
Proof. For all z ̸= 0, using the Maclaurin series of the exponential
function,
∞ 0
X 1 −n X 1
f (z) = z = zk ,
n! (−k)!
n=0 k=−∞
so we read off N = −∞. □
The first two examples can be generalized with the same proof:
Theorem 8.10. Assume that f (z) = z −n h(z) for all z ∈ DR (z0 ) \ {z0 },
where n ∈ Z and h is analytic on DR (z0 ) with a zero at z0 of multiplicity m
(we allow m = 0). Then f has a zero of multiplicity m − n if m ≥ n, and a
pole of order n − m if n > m.
Theorem 8.11 (Riemann removable singularity theorem). If f is bounded
on DR (z0 ) \ {z0 }, then f has a removable singularity at z0 .
Proof. For any k ∈ Z and r ∈ (0, R),
|ak | ≤ r−k sup |f (reiθ )|.
θ∈[0,2π]

For k < 0, in the limit r ↓ 0, we obtain |ak | = 0. □

Example 8.12. Aut(D \ {0}) = {z 7→ eiϕ z | ϕ ∈ [0, 2π)}.
 3. CLASSIFICATION OF ISOLATED SINGULARITIES 143

Proof. Let f ∈ Aut(D \ {0}). Since f is bounded, the singularity at

0 is removable. The open mapping theorem rules out the case |f (0)| ≥ 1.
Thus, |f (0)| < 1. If f (0) = w ∈ D \ {0}, then there would also be z ∈ D \ {0}
such that f (z) = w, so f would take values near w twice, once near 0 and
once near z. This would contradict f ∈ Aut(D \ {0}). Thus, f (0) = 0
and f ∈ Aut(D). All such functions are of the form f (z) = eiϕ z for some
ϕ ∈ [0, 2π). □
Note that the proof of Theorem 8.11 actually proves a slightly stronger
statement:
Theorem 8.13. If f (z) = o(|z−z0 |−1 ) as z → z0 , then f has a removable
singularity at z0 .
Analogously,
Theorem 8.14. For any n ∈ N0 , if f (z) = o(|z − z0 |−n−1 ) as z → z0 ,
then f has a removable singularity at z0 or a pole of order at most n.
Conversely, at an essential singularity, f must have much more varied
behavior:
Theorem 8.15 (Casorati–Weierstrass). If f has an essential singularity
at z0 , then for any δ > 0, the set of values of f in Dδ (z0 ) \ {z0 } is dense in
C.
Proof. If this was false, there would exist δ, ϵ > 0 and w ∈ C such that
0 < |z − z0 | < δ =⇒ |f (z) − w| > ϵ.
Then the function
1
g(z) =
f (z) − w
would be bounded on Dδ (z0 ) \ {z0 }, so it would have a removable singularity
at z0 . Even though g is nonzero on Ω, it may be zero at z0 .
If g(z0 ) ̸= 0, this would imply that f (z) = w + g(z)−1 also has a remov-
able singularity at z0 .
If g(z0 ) = 0, factor g(z) = (z − z0 )k h(z) with h(z0 ) ̸= 0. Then
1 −k

k −1

f (z) = w + = (z − z 0 ) w(z − z 0 ) + h(z)
(z − z0 )k h(z)
so f would have a pole of order k. □
Automorphisms of C.
Theorem 8.16. Every ψ ∈ Aut(C) is of the form ψ(z) = αz + β for
some α ∈ C \ {0}, β ∈ C.
Proof. Since ψ is a bijection and ψ −1 is continuous, there exists ϵ > 0
such that |ψ(z) − ψ(0)| < ϵ implies |z| < 1. Taking the contrapositive,
|z| ≥ 1 implies |ψ(z) − ψ(0)| ≥ ϵ.
144 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

Since ψ is an entire function, it has a Maclaurin series

∞
X
ψ(z) = an z n .
n=0
The function
∞
X
f (w) = ψ(1/w) = an w−n
n=0
is well-defined on C \ {0} and 0 < |w| ≤ 1 implies |f (w) − ψ(0)| ≥ ϵ, so by
Casorati–Weierstrass, f does not have an essential singularity at 0. Thus,
ψ is a polynomial.
If deg ψ > 1, then ψ ′ is a nonconstant polynomial, so ψ has critical
points; a conformal bijection cannot have critical points, so ψ must be a
polynomial of degree 1. □

4. Meromorphic functions
Definition 8.17. Let Ω be a domain and S ⊂ Ω a subset with no
accumulation points in Ω. If a function f : Ω \ S → C is analytic, and
every point in S is a removable singularity or a pole of f , we say that f is
a meromorphic function on Ω.
It is standard to define f at removable singularities p by
f (p) = lim f (z),
z→p

and to define f at poles p by f (p) = ∞. That extends f to a function

f : Ω → Ĉ, where
Ĉ = C ∪ {∞}.
Note that these conventions are natural, since if f has a pole at p, then
limz→p |f (z)| = ∞.
These extensions are used implicitly, and from now on we will think
of a meromorphic function as a map Ω → Ĉ which arose from the above
construction. In particular, two meromorphic functions on Ω are said to be
equal if they are equal as maps Ω → Ĉ.
, and from now on we will not make a distinction between the mero-
morphic function Ω → Ĉ and the analytic function on Ω \ S, with S a
discrete subset of Ω. For instance, if f, g are analytic, then f /g is analytic
on Ω \ g −1 (0). In fact:
Theorem 8.18. Let f, g be analytic functions on Ω, with g not identically
zero. Then f /g is meromorphic on Ω.
Proof. The set of zeros of g is discrete in Ω. On Ω\g −1 (0), the function
h(z) = f (z)/g(z) is analytic. At each zero z0 of g, by factoring g(z) =
(z − z0 )k g1 (z) with g1 (z0 ) ̸= 0, we obtain
1 f (z)
h(z) =
(z − z0 )k g1 (z)
 4. MEROMORPHIC FUNCTIONS 145

with f /g1 analytic at z0 , so z0 is a pole or removable singularity of h. □

Example 8.19. tan is a meromorphic function on C.
Using the Weierstrass product theorem, we can also prove the converse:
Theorem 8.20. Every meromorphic function on a domain Ω is the ratio
of two analytic functions on Ω.
Proof. If h is meromorphic on Ω, list its poles with multiplicity as
(zn )N
n=1 , where N is finite or ∞. The sequence of poles has no accumulation
points in Ω. Thus, by Theorem 6.15, there exists g ∈ A(Ω) with zeros zn .
Then f = gh is meromorphic on Ω and all its singularities are removable,
so f ∈ A(Ω). Thus, h = f /g. □

For what follows, we will need some further terminology:

Definition 8.21. At a pole p, a meromorphic function has a Laurent
expansion of the form
∞
X
f (z) = an (z − p)n
n=−m

where m is the order of the pole. The principal part of the pole is
−1
X
ρ(z) = an (z − p)n
n=−m

and the residue at p is

Res(f, p) = a−1 .
In practice, we can often compute principal parts by using Taylor ex-
pansions. For instance:
z+1
Example 8.22. The function f (z) = z 3 (z 2 +1)
has a pole at 0 with prin-
±1+i
cipal part z −3 + z −2 − z −1 and poles at ±i with principal parts 2(z∓i) .

Proof. Expanding around 0, we get

1 1+z 1 1 1 1 1
f (z) = 3 2
= 3 (1+z)(1−z 2 +O(z 4 )) = 3 (1+z−z 2 +O(z 3 )) = 3 + 2 − +O(1)
z 1+z z z z z z
so we read off principal part z −3 + z −2 − z −1 .
Factoring (z 2 + 1) = (z + i)(z − i) we see that for z → i,
  1+i
1 z+1 1 i+1 2
f (z) = = + O(z − i) = +O(1), z→i
z − i z 3 (z + i) z − i i3 (2i) z−i
with a similar calculation for z → −i. □
146 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

5. Bonus: partial fraction decompositions

We will now consider another kind of sum or series representation for a
meromorphic function, based on its asymptotic behavior at the poles. As
the first example, we revisit a calculus technique:
Theorem 8.23 (Partial fraction decomposition). For any rational func-
tion P (z)/Q(z) with deg P < deg Q,
P (z) X
= ρj (z)
Q(z)
j

with the sum taken over all poles of the rational function.
Proof. The difference of both sides has a removable singularity at all
the poles of P/Q, so it is an entire function. Since deg P < deg Q, it goes to
0 as z → ∞. By Liouville’s theorem, this difference is identically zero. □
A specialization of this to rational functions with real coefficients is used
in calculus to integrate rational functions with real coefficients (Exercise 8).
Example 8.24. The function π 2 / sin2 (πz) has poles at integers n ∈ Z
with principal part 1/(z − n)2 .
Proof. We know that sin(πz) has zeros at integers and no other zeros
in C. Using Taylor expansions around z = 0, note that sin(πz) = πz +O(z 3 )
implies
sin2 (πz) = π 2 z 2 (1 + O(z 2 )), z→0
so
π2 1 1
2 = 2 (1 + O(z 2 )) = 2 + O(1), z → 0.
sin (πz) z z
Since sin2 (πz) is 1-periodic, the principal part at n ∈ Z is 1/(z − n)2 . □
The last example also leads to a partial fraction decomposition; since
there are infinitely many poles, the sum of principal parts is now a series,
whose convergence must be checked:
Example 8.25. For all z ∈ C \ Z,
∞
π2 X 1
2 = .
sin (πz) k=−∞ (z − k)2

Proof. For any N ∈ N, for |z| ≤ N < |k|, we see

1 1
≤
|z − k|2 (k − N )2
which is summable in k, so by the Weierstrass test,
X 1
(z − k)2
k∈Z
|k|>N
 5. BONUS: PARTIAL FRACTION DECOMPOSITIONS 147

defines an analytic function on DN (0). Thus,

X 1
(z − k)2
k∈Z

defines a meromorphic function on DN (0), with poles at k and principal

parts 1/(z − k)2 . Since N is arbitrary, the series is meromorphic on C.
Since convergence is absolute for all z ∈ C \ πZ, it is independent of the
order of summation, so g(z + 1) = g(z). Thus, the function
h(z) = f (z) − g(z)
is 1-periodic and entire.
We have previously seen
lim inf |sin(x + iy)| = ∞.
|y|→∞ x∈[0,π]

Thus, h is 1-periodic, entire, and

lim sup |f (x + iy)| = 0.
|y|→∞ x∈[0,1]

The function g obeys the same property: for any sequence xn ∈ [0, 1] and
|yn | ≥ 1, |yn | → ∞,
1
|xn + iyn − k|−2 ≤
1 + (|k| − 1)2
so the right-hand side is a dominating ”function” and g(xn + iyn ) → 0 by
dominated convergence. Thus,
lim sup |h(x + iy)| = 0.
|y|→∞ x∈[0,1]

Thus, by Liouville’s theorem, h = 0. □

For a trickier example, consider
π cos(πz)
f (z) = π cot(πz) = .
sin(πz)
It has simple poles at k ∈ Z, with principal parts 1/(z − k). The series
X 1
z−k
k∈Z

is not absolutely convergent, since 1/(z−k) ∼ −1/k for large k and ∞ −1 =

P
k=1 k
∞. However, it can be renormalized in two ways. We will use the trick to
combine terms for ±k: this improves convergence since
1 1 2z
+ = 2
z−k z+k z − k2
and ∞ −2 = ∞. A more consistent general approach is to add constants
P
k=1 k
to the terms in a way that improves convergence, see Exercise 9.
148 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

Example 8.26.
∞
1 X 2z
π cot(πz) = + (8.6)
z z 2 − k2
k=1

Proof. Since for |z| ≤ N < k,

2z 2N
≤ 2 ,
z 2 − k2 k − N2
the right-hand side gives a meromorphic function on C. Away from the
poles, we can differentiate term by term to compute
∞
!′ +∞  
1 X 2z 1 X 1 1
+ =− 2 − +
z z 2 − k2 z (z − k)2 (z + k)2
k=1 k=1
π2
=−
sin2 (πz)
 ′
π cos(πz)
=
sin(πz)
so the two sides of (8.6) are equal up to an additive constant C ∈ C. Since
both sides of (8.6) are odd functions, C = 0. □
Shifting (8.6) by 1/2 gives an expansion of tan(πz), in which the principal
parts of poles at ±k + 1/2 are grouped together. A more elegant formula
is obtained by grouping together the principal parts of poles at ±(k + 1/2),
see Exercise 11.
Partial fraction decompositions are related to other types of represen-
tations we have seen before. For example, we use the previous example to
prove a product formula:
Theorem 8.27 (Euler’s product formula). For all z ∈ C,
∞ 
z2
Y 
sin(πz) = πz 1− 2
n
n=1
z2
Proof. Denote fn (z) = 1 − n2
. Then
∞ ∞
X X R2
∥1 − fn ∥DR (0) = <∞
n2
n=1 n=1
so the product
∞ 
z2
Y 
F (z) = πz 1− 2
n
n=1
is uniformly convergent in compact subsets of C, has simple zeros at integers
and no other zeros, and
∞
F ′ (z) 1 X 2z π cos(πz)
= + = .
F (z) z z 2 − n2 sin(πz)
n=1
 6. CHAINS OF CONTOURS AND THE GLOBAL CAUCHY INTEGRAL FORMULA149

Denote G(z) = sin(πz). Note that

F′ G′
=
F G
′ ′
implies F G − F G = 0 so F/G = C. Finally, since
∞
Y
fn (0) = 1,
n=1
we see that
F (z) G(z)
lim = lim =1
z→0 πz z→0 πz
so the limit F/G at 0 is 1, so C = 1. □
Corollary 8.28 (Euler reflection formula). For z ∈
/ Z,
π
Γ(z)Γ(1 − z) = .
sin(πz)
√
In particular, Γ(1/2) = π.
Proof. (a) Using properties of Γ and Euler’s product formula for sin,
1 π
Γ(z)Γ(1 − z) = −zΓ(z)Γ(−z) = Q∞ = .
z n=1 E1 (z/n)E1 (−z/n) sin(πz)
(b) By (a), Γ(1/2)2 = π. Obviously, Γ(z) > 0 for z ∈ (0, ∞). □

6. Chains of contours and the global Cauchy integral formula

We have seen two versions of the Cauchy integral formula: on the disk
and on the annulus. We will now derive a version which is more general in
two important ways.
The previous versions relied on loop homotopy in the domain. If a curve
γ is homotopic in Ω to a constant curve, it has zero winding number on
C \ Ω, by Theorem 7.9 applied to γ − z for each z ∈ C \ Ω. The converse
is false: having zero winding number on C \ Ω (Figure 8.1) does not imply
that the curve is homotopic to a constant curve. Thus, a version of the
Cauchy integral formula which doesn’t require loop homotopy would be a
generalization.

Figure 8.1. A curve in the domain Ω = C \ {−1, 1} which

has zero winding number on its complement but isn’t homo-
topic to a constant curve
150 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

The other generalization is in handling finite sums of contour integrals.

This is formulated in the following language. A chain Γ in Ω is a formal sum
of finitely many closed curves γ1 , . . . , γk in Ω. We define integration over a
chain Γ as the sum of contour integrals
I X k I
f (ζ) dζ = f (ζ) dζ,
Γ j=1 γj

We also define
k
[
Ran Γ = Ran γj
j=1
and the winding numbers of a chain as
k
X
n(Γ, z) = n(γj , z), z∈
/ Ran Γ.
j=1

Note that some properties follow by additivity, for example

I
1 1
n(Γ, z) = dζ. (8.7)
2πi Γ ζ − z
Tthis formalism can be used to represent the Cauchy integral formula
on the annulus in terms of the chain consisting of two circles γR and γρ← .
Note that in this example,
(
1 ρ < |z| < R
n(Γ, z) = n(γR , z) − n(γρ , z) =
0 |z| < ρ or |z| > R
We will use this formalism to state a more general, global version of the
Cauchy integral formula.
Definition 8.29. A chain Γ is null-homologous in Ω if
n(Γ, z) = 0, ∀z ∈ C \ Ω.
Theorem 8.30 (Global Cauchy integral formula). For an analytic func-
tion f : Ω → C, and a chain Γ null-homologous in Ω,
I
1 f (ζ)
dζ = n(Γ, z)f (z), ∀z ∈ Ω \ Ran Γ.
2πi Γ ζ − z
Proof. We begin by observing that the function
(
f (w)−f (z)
w−z z ̸= w
h(z, w) = ′
f (z) z=w
is jointly continuous on Ω × Ω. Namely, continuity at points (z0 , w0 ) with
z0 ̸= w0 follows from continuity of f . Continuity at points of the form
(z0 , z0 ) follows from the representation
Z 1
h(z, w) = f ′ (z + t(z − w)) dt
0
 6. CHAINS OF CONTOURS AND THE GLOBAL CAUCHY INTEGRAL FORMULA151

which is valid in Dr (z0 ) × Dr (z0 ) if Dr (z0 ) ⊂ Ω, because f ′ (z + t(z − w)) is

jointly continuous in Dr (z0 ) × Dr (z0 ) × [0, 1].
For fixed w ∈ Ω, the function f − f (w) has a zero at w, and factoring
that zero precisely gives
f (z) − f (w) = (z − w)h(z, w).
Thus, by Lemma 5.23, h(z, w) is analytic in z ∈ Ω. Thus, by Fubini and
Morera’s theorems, the function
I
1
g(z) = h(z, w) dw
2πi Γ
is analytic in z ∈ Ω. Our goal will be to show that g = 0.
Denote
Π = {z ∈ C \ Ran Γ | n(Γ, z) = 0}.
Since winding numbers are constant on connected components of C \ Ran Γ,
the set Π is open. Since Γ is null-homologous in Ω, Π contains all points in
the complement of Ω, so C = Ω ∪ Π. We will use this to obtain an analytic
continuation of g to C.
For z ∈ Ω \ Ran Γ,
I I
1 f (z) 1 1
dw = f (z) dw = f (z)n(Γ, z).
2πi Γ w − z 2πi Γ w − z
In particular, for z ∈ Ω ∩ Π, this is equal to 0, so
f (w) − f (z)
I I
1 1 f (w)
dw = dw.
2πi Γ w−z 2πi Γ w − z
Thus, we can extend the domain of definition of g to C by
(
1
H
h(z, w) dw z ∈ Ω
g(z) = 2πi1
HΓ f (w)
2πi Γ w−z dw z∈Π

and, since both definitions match on Ω ∩ Π and give analytic functions on

Ω and Π, g is now analytic on Ω ∪ Π = C.
Moreover, denote R = maxz∈Ran Γ |z|. For |z| > R, we have z ∈ Π and
ℓ(Γ) f (w) C
|g(z)| ≤ sup ≤
2π w∈Ran Γ w − z |z| − R
which converges to 0 as z → ∞. Since g is an entire function, by Liouville’s
theorem, g = 0. Applying this to z ∈ Ω \ Ran Γ gives
I I
1 f (w) 1 f (z)
dw = dw.
2πi Γ w − z 2πi Γ w − z
By factoring f (z) and using (8.7), the right-hand side is f (z)n(Γ, z). □
152 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

Global Cauchy integral theorem. As a corollary, we obtain a Cauchy

integral theorem in the same level of generality:
Theorem 8.31 (Global Cauchy integral theorem). Let Ω ⊂ C, f analytic
on Ω, and Γ a null-homologous chain in Ω. Then
I
f (ζ) dζ = 0.
Γ

Proof. Fix some point z ∈ Ω \ Ran Γ. Applying the Cauchy integral

formula to the function (ζ − z)f (ζ) gives the Cauchy integral theorem. □
This has a topological consequence:
Theorem 8.32. If Ω is a domain and every connected component of
C \ Ω is unbounded, then Ω is simply connected.
Proof. For any closed curve γ in Ω, the winding number is constant on
connected components of C \ Ω, and zero on each unbounded component.
In particular, γ is homologous to zero in Ω. Thus, by the global Cauchy
integral theorem, I
f (z) dz = 0
γ
for all f ∈ A(Ω). By a consequence of the Riemann mapping theorem, Ω is
simply connected. □
Bounded connected components of C \ Ω are sometimes called holes. We
have showed that if Ω has no holes, it is simply connected. The converse is
also true and will be proved later.

7. Residue theorem and argument principle

If f is analytic on Ω and f (w) ̸= 0, the function g(z) = f (z)/(z − w)
has a simple pole at w with residue Res(g, w) = f (w). In this sense, the
Cauchy integral theorem describes contour integrals of analytic functions,
and the Cauchy integral formula describes contour integrals of meromorphic
functions with a single pole. The natural generalization to meromorphic
functions is known as the residue theorem:
Theorem 8.33 (Residue theorem). Let f be a meromorphic function on
Ω, and let Γ be a null-homologous chain in Ω such that f has no poles on
Ran Γ. Then f has finitely many poles with nonzero winding number n(Γ, ·);
if we denote them by p1 , . . . , pN , then
I N
1 X
f (z)dz = n(Γ, pj ) Res(f, pj ).
2πi Γ
j=1

Proof. The set C \ Ran Γ is open and the winding number is constant
on its connected components.
 7. RESIDUE THEOREM AND ARGUMENT PRINCIPLE 153

Thus, the boundary of the set S = {w ∈ C \ Ran Γ | n(Γ, w) ̸= 0} is a

subset of Ran Γ, i.e.,
S ⊂ S ⊂ S ∪ Ran Γ.
It follows that S ∪ Ran Γ = S ∪ Ran Γ is closed. Since it is bounded, it is
a compact subset of Ω. Since poles are isolated, there are finitely many on
S ∪ Ran Γ; denote them p1 , . . . , pN and their principal parts by ρ1 , . . . , ρN .
Then we can write
N
X
f (z) = g(z) + ρj (z)
j=1

with g is meromorphic on Ω and has no polesH in S. As a consequence, g is

analytic in a neighborhood of S. Therefore Γ g(z)dz = 0 by the ultimate
Cauchy integral theorem.
Moreover, I
1 1
dz = n(Γ, pj )
2πi Γ z − pj
by (8.7) and I
1
k
dz = 0, k≥2
Γ (z − pj )
because the function
 ′
1 1
=
(z − pj )k (1 − k)(z − pj )k−1
has an antiderivative in Ω \ {pj }. From this we obtain
I
1
ρj (z) dz = n(Γ, pj ) Res(f, pj )
2πi Γ
and then summing the integrals concludes the proof. □
We can now also derive a very general version of an argument principle,
which works with respect to a chain and applies to meromorphic functions.
Whenever f is a meromorphic function and not identically zero, we note
that f ′ /f is a meromorphic function, with poles precisely at the zeros and
poles of f , and:

 ′  
f k w is a zero of order k
Res , w = −k w is a pole of order k
f 
0 else


Thus, applying the residue theorem to f ′ /f implies:

Theorem 8.34 (Global argument principle for meromorphic functions).
Let Ω ⊂ C, Γ a chain homologous to zero in Ω, and f meromorphic on Ω with
no zeros or poles on Ran Γ. Then f has finitely many zeros and poles with
nonzero winding number n(Γ, ·); if we denote those zeros by z1 , . . . , zNz , their
154 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

multiplicities by n1 , . . . , nNz , and those poles by p1 , . . . , pNp and multiplicities

by m1 , . . . , mNp , then
N Np
f ′ (z) z
I
1 X X
dz = n(Γ, zj )nj − n(Γ, pj )mj .
2πi Γ f (z)
j=1 j=1

There is also a corresponding version of Rouché’s theorem.

8. Applications of contour integration

Complex analysis provides a very useful tool for computing integrals.
For an integral over a closed contour, or an integral that can be rewritten
in this way (Exercise 12), the residue theorem can often be used to evaluate
it. In other examples, we have an integral over a contour which is not
closed, and we have to close the contour manually; this works particularly
well for improper integrals, and allows us to compute improper integrals in
many cases when we cannot find the antiderivative. Such improper integrals
over the entire real line are of great interest, since they appear as Fourier
transforms or appear in more applied subjects such as probability. Some
standard techniques are best learned through examples, some of which are
presented below.
Example 8.35. For any a > 0 and t ∈ R,
Z +∞ Z R
eitx eitx π
2 2
dx = lim 2 2
dx = e−a|t| .
−∞ x + a R→∞ −R x + a a
R R eitx
Strategy. We view −R x2 +a2 dx as a contour integral over [−R, R].
We close the contour by a semicircle and hope that the decay of the integrand
makes the semicircle negligible as R → ∞. Since |eitz | = eRe(itz) = e−t Im z ,
we use a semicircle in C+ if t ≥ 0 and a semicircle in C− if t ≤ 0. □

ia

−R R

Figure 8.2. Contour used in Example 8.35. For R > a, it

encloses the pole at ia

Proof. We use the meromorphic function on C

eitz
f (z) = ,
z 2 + a2
 8. APPLICATIONS OF CONTOUR INTEGRATION 155

which has poles at ±ia with residues

eit(±ia) e∓ta
Res(f, ±ia) = ± =± .
2ia 2ia
RR
The integral −R . . . is viewed as a contour integral over the line segment
[−R, R].
We first assume t ≥ 0 and “close the contour” by the semicircle γR (t) =
Reit , t ∈ [0, π]. For R > a, the residue theorem gives
e−at
I I
1 1
f (z)dz + f (z)dz = Res(f, ia) = . (8.8)
2πi [−R,R] 2πi γR 2ia
If we take the limit R → ∞, the crucial observation is that the semicircle
vanishes in the limit: for R > a,
|eitz | e−t Im z 1
|f (z)| = 2 2
= 2 2
≤ 2
|z + a | |z + a | R − a2
so I
1 Rπ
f (z)dz ≤ ℓ(γR ) = 2 → 0, R → ∞.
γR R2 −a 2 R − a2
Thus, taking R → ∞ in (8.8) gives
e−at
I
1
lim f (z)dz =
2πi R→∞ [−R,R] 2ia
so multiplying by 2πi completes the proof.
The case t < 0 is proved similarly by completing the contour by a semi-
circle in the lower half-plane, or by using the symmetry x 7→ −x to reduce
to the case of positive t. □
In probability, multiplying this by a/π gives the so-called Cauchy distri-
bution with parameter a and computes its characteristic function.
The case t = 0 could have also been solved by using arctan. However,
the technique above worked for any t, and the same technique applies in
more complicated cases.
In the following example, the integrand is not a meromorphic function
on C, so we will use a branch cut away from our contour:
Example 8.36. For any κ ∈ (−1, 1),
Z ∞ Z R
tκ tκ π/2
2
dt = lim lim 2
dt = . (8.9)
0 1+t ρ↓0 R→∞ ρ 1 + t cos(κπ/2)
Strategy. The integrand f (z) = z κ /(1 + z 2 ) is not meromorphic on
C; to handle z κ with κ ∈
/ Z, we must select a branch cut and use a contour
which avoids the branch cut. It is convenient to use
Z R
tκ
Z
f (z) dz = 2
dt,
[ρ,R] ρ 1+t
156 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

so we will choose a branch with arg z = 0 for z > 0. In addition, we notice

that, using z = −t and a change of orientation,
Z R Z R
(−t)κ tκ
Z
iπκ
f (z) dz = − 2
(−dt) = e 2
dt
[−R,−ρ] ρ 1 + (−t) ρ 1+t

so we can use both line segments [ρ, R] and [−R, −ρ] and close this contour
by two semicircles. □

−R −ρ ρ R

Figure 8.3. Contour used in Example 8.36. It avoids the

branch cut [0, −i∞). For ρ < 1 < R, it encloses the pole at i

Proof. Consider the meromorphic function

zκ
f (z) = , z ∈ C \ [0, −i∞)
1 + z2
with arg(z κ ) = κ arg z for arg z ∈ (−π/2, 3π/2). The function f has a pole
at i with residue
eiκπ/2
Res(f, i) = .
2i
Let 0 < ρ < 1 < R < ∞ and consider the closed contour γ = γ4 ∗ γ3 ∗ γ2 ∗ γ1 ,
where
γ1 (t) = t, t ∈ [ρ, R]
it
γ2 (t) = Re , t ∈ [0, π]
γ3 (t) = t, t ∈ [−R, −ρ]
γ4 (t) = ρei(π−t) , t ∈ [0, π]
so by residue calculus, the contour integral of f over γ is
I
f (z) dz = 2πi Res(f, i) = πeiκπ/2 .
γ

Integrating over each part of the contour and combining the integrals over
the two line segments gives
Z R Z π Z π 1+κ iκθ
iκπ tκ R1+κ eiκθ ρ e
(1+e ) 2
dt+i iθ 2
dθ −i iθ )2
dθ = πeiκπ/2 .
ρ 1 + t 0 1 + (Re ) 0 1 + (ρe
 8. APPLICATIONS OF CONTOUR INTEGRATION 157

Since κ ∈ (−1, 1), letting ρ ↓ 0 and R ↑ ∞ gives

Z ∞
iκπ tκ
(1 + e ) dt = πeiκπ/2
0 1 + t2
which implies (8.9). □
In the previous example, due to the lucky coincidence 1 + (−t)2 = 1 + t2 ,
we were able to use both line segments [ρ, R] and [−R, −ρ]. In a more
complicated example it may be necessary to use a more gradual approach,
using a ”keyhole contour” such as γ4 ∗ γ3 ∗ γ2 ∗ γ1 ,
γ1 (t) = eiϕ t, t ∈ [ρ, R]
γ2 (t) = Reit , t ∈ [ϕ, 2π − ϕ]
γ3 (t) = ei(2π−ϕ) (R − t), t ∈ [0, R − ρ]
i(2π−t)
γ4 (t) = ρe , t ∈ [ϕ, 2π − ϕ]
and then letting ϕ → 0, ρ → 0, R → ∞.

ρeiϕ Reiϕ

ρe−iϕ Re−iϕ

Figure 8.4. A Rkeyhole contour used to evaluate many im-

∞
proper integrals 0 . . . by using an integrand with a branch
cut [0, ∞)

Example 8.37. For any a > 0,

Z +∞ Z R
sin(ax) sin(ax)
dx = lim dx = π.
−∞ x R→∞ −R x
Strategy. The function sin(az)/z is entire. However, sin grows ex-
ponentially in both the upper and lower half planes, so we wouldn’t be
able to close the contour with a negligible semicircle. Instead, we will use
sin(ax) = Im eiax , so it suffices to compute
Z +∞ iax
e
dx
−∞ x
158 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

and take the imaginary part. Now we can hope to close the contour in C+ .
Since eiaz /z doesn’t have a removable singularity at 0, we will avoid the
point 0 by using a semicircle. Depending on whether we choose a semicircle
in C+ or C− , we may or may not enclose the pole at 0; this hints that we
should expect an interesting contribution near 0. □
Proof of Example 8.37. The function f (z) = eiaz /z is meromorphic
on C. We use the contour consisting of line segments
[ρ, R], [−R, −ρ]
and the semicircles
γρ (t) = ρei(π−t) , γR (t) = Reit , t ∈ [0, π].
Controling the semicircle contours is also a bit more delicate in this example.
Instead of bounding the integrand everywhere by its maximum value, we
have to bound it pointwise: by bounding f (Reit ) we get
I Z π −aR sin t
e
f (z)dz ≤ R dt.
γR 0 R
By using sin(π − t) = sin t and sin t ≥ π2 t for t ∈ [0, π/2], we have
Z π Z π/2 Z π/2 Z ∞
−aR sin t −aR sin t − 2aR 2aR 2π
e dt = 2 e dt ≤ 2 e π
t
dt ≤ 2 e− π t dt = .
0 0 0 0 2aR
Thus, this integral vanishes as R → ∞.
For the integral over γρ , use the fact that
eiaz 1
= + h(z)
z z
for some h analytic in a neighborhood of 0. In particular h is bounded on a
neighborhood of 0 so
I
lim h(z)dz ≤ ℓ(γρ )C → 0, ρ ↓ 0.
ρ↓0 γρ

If we work with branches of log, arg on C \ [0, −i∞), then

I
1
dz = ln(ρ) − ln(−ρ) = (|ρ| + i arg ρ) − (|−ρ| + i arg(−ρ)) = −πi.
γρ z

Thus, we have
!
eiaz eiaz
Z Z
lim lim dz + dz = πi.
R→∞ ρ↓0 [−R,−ρ] z [ρ,R] z
Writing as Riemann integrals and taking imaginary parts we obtain
Z −ρ Z R 
sin(ax) sin(ax)
lim lim dx + dx = π
R→∞ ρ↓0 −R x ρ x
 8. APPLICATIONS OF CONTOUR INTEGRATION 159

but now the integrand sin(ax)/x is continuous at 0 so we can remove the

cut near 0 and conclude
Z R
sin(ax)
lim dx = π. □
R→∞ −R x

Note that even though the contour didn’t enclose the pole at 0, its
residue still showed up due to the contour γρ which partly circles the pole.
The residue at 0 is 1 but γρ only goes half-way around the pole so we get
a contribution of πi. This result, stated more generally, is called Jordan’s
lemma:
Lemma 8.38 (Jordan). Let γρ (t) = z0 + ρeit , t ∈ [α, β]. Assume that f
has a simple pole at z0 . Then
I
lim f (z) dz = i(β − α) Res(f, z0 ).
ρ↓0 γρ

In the next example, we will use the Gaussian integral

Z ∞ √
−t2 π
e dt = (8.10)
0 2
as a known fact. This is often derived by using polar coordinates and a
double integral in R2 . It also follows from the value of Γ(1/2) derived in
Cor. 8.28, by a change of variables t2 = u.
Since the integrand in (8.10) is even, and the variable can be rescaled,
the Gaussian integral is often written in the more symmetric form
Z +∞
1 2
√ e−x /2 dx = 1.
2π −∞
The 1/2 is commonly placed in the exponent because this gives the Gauss-
ian/normal probability density with mean 0 and variance 1. The following
example computes its Fourier transform (characteristic function of the nor-
mal distribution, in probabilistic language).
Example 8.39. For any t ∈ R,
Z +∞
1 2 2
√ eitx e−x /2 dx = e−t /2 .
2π −∞
Proof. Consider the entire function
2 /2
f (z) = e−z
and the rectangular contour with vertices −R, R, R − it, −R − it. Since f is
entire, there are no residues and we obtain
I I I I
f (z)dz+ f (z)dz+ f (z)dz+ f (z)dz = 0.
[−R,R] [R,R−it] [R−it,−R−it] [−R−it,−R]
160 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

Along the horizontal segments, parametrizing by z = x and by z = x − it,

and paying attention to the orientation, we get
Z R Z R I I
−x2 /2 −(x−it)2 /2
e dx − e dx + f (z)dz + f (z)dz = 0.
−R −R [R,R−it] [−R−it,−R]
Next, let us show that vertical segments vanish as R → ∞. This will use
2 )/2 2 −(Im z)2 )/2 2 +t2 )/2
|f (z)| = e− Re(z = e−((Re z) ≤ e(−R , Re z = ±R, |Im z| ≤ |t|.
This goes to 0 as R → ∞, and the length of the segment is constant, so
vertical segments vanish as R → ∞. Thus,
Z R Z R 
−x2 /2 −(x−it)2 /2
lim e dx − e dx = 0.
R→∞ −R −R

Since (x − it)2 = x2 − itx − t2 , expanding this square in the second exponent

gives
Z +∞
t2 /2 itx −x2 /2
Z +∞
2 /2 √
e e e dx = e−x dx = 2π
−∞ −∞
√ 2 /2
and dividing by 2πet gives the final answer. □

9. Exercises
1. Let f : C \ {0} → C be holomorphic. Prove that for any closed rectifiable
curve γ in C \ {0},
I I
f (z) dz = n(γ, 0) f (z) dz.
γ |z|=1

Prove that there exists c ∈ C such that the function f (z) − zc has an
antiderivative on C \ {0}.
2. For each of these functions, determine whether it has a removable singu-
larity, pole, or essential singularity at 0. Justify your answers.
(a) 1/ sin z
(b) sin(1/z)
 
sin z
(c) sin
z
3. If f : DR (z0 ) \ {z0 } → C+ is analytic, prove that it has a removable
singularity at z0 and that f (z0 ) ∈ C+ .
4. Find the group Aut(D \ {0, 21 }).
5. If f : Dr (z0 ) \ {z0 } → Π is analytic where Π ⊊ C is simply connected,
prove that f has a removable singularity at z0 and that limz→z0 f (z) ∈ Π.
Hint: use the Riemann mapping theorem.
6. Let f be analytic on Ω ⊂ C+ . Assume that f has a continuous extension
to an open subset I of R, for each z ∈ I there exists δ > 0 such that
Dδ (z) ∩ C+ ⊂ Ω, and |f (x)| = 1 for x ∈ I ⊂ R. Prove that f has an
meromorphic continuation to Π = Ω ∪ I ∪ {z | z ∈ Ω} given by
f (z) = 1/f (z̄) z̄ ∈ Ω.
 9. EXERCISES 161

7. Prove that the Gamma function Γ is a meromorphic function on C and

that, for n = 0, 1, 2, . . . , Γ has a simple pole at −n with residue
(−1)n
Res(Γ, −n) = .
n!
8. Let P, Q ∈ R[x] with deg P < deg Q. Prove that P/Q can be expressed
uniquely as a sum of terms of two types: C/(x−p)k where (x−p)k divides
Q(x) and (Ax + B)/(x2 + bx + c)k where x2 + bx + c has no real zeros
and (x2 + bx + c)k divides Q(x).
9. Prove that the series
1 X  1 1

+ +
z z−n n
n∈Z\{0}

is absolutely convergent on C \ Z and that its limit is π cot(πz).

10. Prove that for z ∈ C \ Z,
∞
π 1 X 2z
= + (−1)k 2 .
sin(πz) z z − k2
k=1

11. (a) Prove that for all z ∈ C \ { k + 21 π | n ∈ Z},

∞
X 2z
tan z = −
k=0
z2 − (k + 12 )2 π 2
(b) Prove that for any n ∈ N,
∞
!(n)
X 2z
(tan z)(n) = π2
k=0
z − (k + 12 )2
2

and derive a Maclaurin expansion for tan,

∞ ∞
X
2n+1
X 22n+3
tan z = an z , an =
(2k + 1)2n+2 π 2n+2
n=0 k=0
(c) Prove that for any n ∈ N0 , the sums
∞ ∞
X 1 X 1
,
(2k + 1)2n+2 k 2n+2
k=0 k=0

are rational multiples of π 2n+2and find their values for n = 0 and

n = 1.
12. For a, b ∈ R such that a > |b|, compute
Z 2π
1 dθ
.
0 a + b cos θ 2π
Hint: check that this is equal to the contour integral
I
1 2
2
dz
2πi |z|=1 bz + 2az + b
 162 8. MEROMORPHIC FUNCTIONS AND THE RESIDUE THEOREM

and compute that contour integral by the residue theorem.

13. Compute the improper integral
Z +∞ Z R
dx dx
4
= lim .
−∞ 1 + x R→∞ −R 1 + x4

Aim to provide your final answer in a form which is obviously a real,

positive number.
14. (a) Find the value of the improper integral
Z ∞ Z R
it2 2
e dt = lim eit dt
0 R→∞ 0

This is called the Fresnel integral; it is important in optics.

2
Hint: use f (z) = eiz and the contour consisting of the line segment
γ1 (t) = t, t ∈ [0, R], the arc γ2 (t) = Reit , t ∈ [0, π/4], and the line
segment γ3 (t) = (R − t)eiπ/4 , t ∈ [0, R]. You can use the Gaussian
integral Z ∞ √
2 π
e−t dt = .
0 2
(b) Find the values of
Z ∞ Z ∞
2
cos(t ) dt, sin(t2 ) dt.
0 0
R∞ 2
15. For w ∈ C, compute −∞ e−t +tw
dt.
16. Let a ∈ (0, 1). Compute
Z ∞ a−1 Z R a−1
x x
dx = lim lim dx.
0 1+x ϵ↓0 R→∞ ϵ 1 + x

Hint: consider the function

z a−1
f (z) =
1+z
on the region C \ [0, ∞). Use a “keyhole contour” obtained by concate-
nating
γ1 (t) = eiδ t, t ∈ [ϵ, R]
γ2 (t) = Reit , t ∈ [δ, 2π − δ]
γ3 (t) = ei(2π−δ) (R − t), t ∈ [0, R − ϵ]
i(2π−t)
γ4 (t) = ϵe , t ∈ [δ, 2π − δ]
and then let δ → 0, ϵ → 0, R → ∞. You can use the consequence of
uniform continuity that for fixed ϵ, R,
I Z R
lim f (z) dz = lim f (xeiδ ) dx
δ↓0 [ϵeiδ ,Reiδ ] ϵ δ↓0

and similarly for the line segment [ϵei(2π−δ) , Rei(2π−δ) ].

 9. EXERCISES 163

17. Let a ∈ (0, 1) and w ∈ C \ [0, ∞). Compute the improper integral
Z ∞ a−1
x
dx.
0 x−w
If your answer involves a multivalued function, state the branch.
 CHAPTER 9

Riemann surfaces

In this brief chapter we discuss the Riemann sphere and collect a few
introductory observations about Riemann surfaces.

1. The Riemann sphere and Möbius transformations

In real analysis, two ideal points ±∞ often appear in discussions of
limits. They are often treated by some special definitions, but they can also
be adjoined to the real line to give an extended real line R ∪ {−∞, +∞}. In
complex analysis, it is natural to consider a single infinite point ∞ and the
set
Ĉ = C ∪ {∞}.
This is called the Riemann sphere.
To motivate this name, we denote the unit sphere in R3 ,
S 2 = {x ∈ R3 | ∥x∥ = 1},
where q
∥x| = x21 + x22 + x23
denotes the Euclidean norm on R3 .
The line containing points (0, 0, 1) and x = (x1 , x2 , x3 ) ∈ S 2 \ {(0, 0, 1)}
intersects the plane x3 = 0 at the point
 
x1 x2
, ,0 ,
1 − x3 1 − x3
and the resulting correspondence between the punctured plane S 2 \{(0, 0, 1)}
and the plane is called stereographic projection. If we identify the complex
plane C with the plane x3 = 0 in R3 by identifying z ∈ C with the point
(Re z, Im z, 0) ∈ R3 , we obtain:
Lemma 9.1. The map Q : C → R3 given by
2 Re z 2 Im z |z|2 − 1
 
Q(z) = , , , z∈C
|z|2 + 1 |z|2 + 1 |z|2 + 1
maps C bijectively to the punctured sphere S 2 \ {(0, 0, 1)}. Its inverse Q−1 :
S 2 \ {(0, 0, 1)} → C is given by
x1 + ix2
Q−1 (x1 , x2 , x3 ) = ,
1 − x3
and Q, Q−1 are continuous.
165
166 9. RIEMANN SURFACES

Proof. For any z ∈ C, by a direct calculation, ∥Q(z)∥ = 1 and Q(z) ̸=

(0, 0, 1). Moreover, for any x ∈ S 2 \ {(0, 0, 1)}, the equation Q(z) = x is
equivalent to
2 Re z 2 Im z |z|2 − 1
= x1 , = x2 , = x3 .
|z|2 + 1 |z|2 + 1 |z|2 + 1
The third equation is equivalent to x3 = 2/(|z|2 + 1) so then the first two
are equivalent to Re z = x1 /(1 − x3 ), Im z = x2 /(1 − x3 ).
Continuity of Q, Q−1 is evident from the explicit formulas. □

We extend Q by setting
Q(∞) = (0, 0, 1)

so that it becomes a bijection Q : Ĉ → S 2 . This induces a metric on Ĉ as

the pullback of the Euclidean metric from the sphere:

Lemma 9.2. Ĉ is a compact metric space with the chord length metric
d(z, w) = ∥Q(z) − Q(w)∥
When restricted to C, they give the same topology as the Euclidean metric
on C. Moreover, for any sequence zn ∈ C, d(zn , ∞) → 0 as n → ∞ if and
only if |zn | → ∞ as n → ∞.
Proof. Since Euclidean norm generates a metric, d obeys the triangle
inequality and is symmetric; it is a metric because Q is injective. It makes
Ĉ a compact metric space because S 2 is a compact subset of R3 .
Since Q : C → S 2 \{(0, 0, 1)} and its inverse are continuous, for zn , z ∈ C,
zn → z in C if and only if Q(zn ) → Q(z) in S 2 , which proves that the
restriction to C is equivalent to the Euclidean metric.
By a direct calculation, for z ∈ C,
2
d(z, ∞) = p .
|z|2 + 1
Thus, d(zn , ∞) → 0 if and only if |zn | → ∞. □

In particular, meromorphic functions on Ω are continuous as maps Ω →

Ĉ.
A more intrinsic notion of distance is obtained by measuring angles on
the sphere, instead of chord lengths:

Lemma 9.3. On the Riemann sphere Ĉ, the arclength metric

˜ w) = arccos (Q(z) · Q(w))
d(z,
is a metric. It is equivalent to the chord length metric, in the sense that they
˜ n , z) → 0.
generate the same topology, i.e., d(zn , z) → 0 if and only if d(z
 1. THE RIEMANN SPHERE AND MÖBIUS TRANSFORMATIONS 167

Proof. To prove that d˜ is a metric, the only nontrivial part is the

triangle inequality. Let u, v, w ∈ S 2 and denote α = arccos(u · v), β =
arccos(v · w). Then
u = cos αv + u⊥ , u⊥ · v = 0, w = cos βv + w⊥ , w⊥ · v = 0,
and ∥u⊥ ∥ = sin α, ∥w⊥ ∥ = sin β. Then
u·w = cos α cos β+u⊥ ·w⊥ ≥ cos α cos β−∥u⊥ ∥∥w⊥ ∥ = cos α cos β−sin α sin β = cos(α+β).
Since cos is strictly decreasing on [0, π], applying arccos, we obtain
arccos(u · w) ≤ α + β = arccos(u · v) + arccos(v · w).
That the two metrics give the same topology follows from
˜ w).
d(z, w)2 = 2 − 2Q(z) · Q(w) = 2 − 2 cos d(z, □
Möbius transformations. On C, we sometimes use affine transforma-
tions z 7→ αz + β with α ̸= 0 as useful bijections of C to itself. On Ĉ, we
use the following class of transformations:
Example 9.4 (Möbius transformations). For any invertible 2 × 2 matrix
 
a b
A=
c d
a meromorphic function fA : Ĉ → Ĉ is given by
az + b
fA (z) = .
cz + d
Note that by invertibility of A,
    
a b z 0
̸=
c d 1 0
so the fraction is never of the form 0/0, and fA is well-defined.
Lemma 9.5. (a) fA = id if and only if A = λI for some λ ∈ C \ {0}
(b) For any invertible 2 × 2 matrices A and B, fAB = fA ◦ fB
(c) For any invertible 2 × 2 matrix A, the map fA is a bijection of Ĉ to itself
Proof. (a) is immediate from the definition.
(b) This follows by a calculation:
A11 B11 z+B12
B21 z+B22 + A12 A11 B11 z + A11 B12 + A12 B21 z + A12 B22 (AB)11 z + (AB)12
= =
A21 B11 z+B12
B21 z+B22 + A22 A21 B11 z + A21 B12 + A22 B21 z + A22 B22 (AB)21 z + (AB)22
(c) follows from (a) and (b) using fA ◦ fA−1 = fA−1 ◦ fA = fI = id. □
Remark 9.6. The multiplicative property of Möbius transformations
can be seen more systematically if we use the projective relation on C2 \ {0},
       
z1 w1 z w1
≃ ⇐⇒ ∃λ ∈ C : 1 = λ
z2 w2 z2 w2
168 9. RIEMANN SURFACES

Then we can write    

fB (z) z
≃B
1 1
so      
fA (fB ((z)) z fAB (z)
≃ AB ≃
1 1 1
This idea gives a new interpretation of the Riemann sphere as the projective
space
CP [1] = (C2 \ {0})/≃ .
Two invertible matrices A, B give the same Möbius transformation if and
only if A = λB for some λ ̸= 0. Note that even if we restrict to matrices
with det A = det B = 1, this still leaves an ambiguity of λ ∈ {±1}. In other
words, Möbius transformations are in 1-1 correspondence with the group
PSL(2, C) = SL(2, C)/{±I}.
Lemma 9.7. If a Möbius transformation has three fixed points, it is the
identity map.
Proof. If fA has fixed point z ∈ C, then A has eigenvector z1 . If fA

has fixed point ∞, then A has eigenvector 10 .

Thus, if fA has three fixed points, A has three eigenvectors, each two of
which are linearly independent. This can only happen if A is a multiple of
the identity. □
Lemma 9.8. For any three distinct points z1 , z2 , z3 ∈ Ĉ, there is a unique
Möbius transformation f such that f (z1 ) = 0, f (z2 ) = 1, f (z3 ) = ∞.
Proof. Uniqueness follows from the previous lemma. Existence can be
obtained by a direct construction; if z1 , z2 , z3 are not infinity, the Möbius
transformation is given by
z2 − z3 z − z1
f (z) = . □
z 2 − z1 z − z3
This is where the cross-ratio comes from.

2. Riemann surfaces
For motivation, we first prove:
Lemma 9.9. A function f : Ω → Ĉ is meromorphic if and only if f
is continuous, not identically infinite, f is analytic on f −1 (C), and 1/f is
analytic on f −1 (Ĉ \ {0}).
Proof. If f is meromorphic, then it is continuous and not identically
infinite. Moreover, if f is meromorphic, then so is 1/f , and both functions
are analytic where they are finite.
For the opposite direction, since f is not identically infinite, 1/f is not
identically zero. Its set of zeros is discrete in Ω. Thus, f −1 (∞) is discrete
 2. RIEMANN SURFACES 169

in Ω, and for every p with f (p) = ∞, the function 1/f has a zero of some
multiplicity m. Thus, on Ω\f −1 (∞), the function f is analytic, and at every
p ∈ f −1 (∞), f has a pole of some multiplicity m. Thus, f is meromorphic.
□
Definition 9.10. A Riemann surface is a connected Hausdorff topolog-
ical space M, with a collection of “coordinate charts” φα : Uα → C, which
obey:
(a) Uα ⊂ M is open
(b) φα : Uα → C is an injective, continuous open mapping
(c) ∪α Uα = M
(d) If Uα ∩ Uβ ̸= ∅, then the transition map
φα ◦ φ−1
β : φβ (Uα ∩ Uβ ) → φα (Uα ∩ Uβ )

is analytic.
The idea is that a Riemann surface ”locally” looks like an open subset
of the complex plane.
Remark 9.11. (a) If we didn’t require connectedness, we could divide
every Riemann surface into connected components and study them sep-
arately. Thus, there is no loss of generality.
(b) It can be proved that every Riemann surface is second-countable, i.e.,
has a countable base of the topology (Rado’s theorem). This is often
included in the definition.
(c) It can be proved that every Riemann surface has a metrizable topology.
This follows from the uniformization theorem.
Example 9.12. Any open connected domain Ω ⊂ C is a Riemann sur-
face, with the single chart id : Ω → Ω.
Example 9.13. The Riemann sphere Ĉ is a Riemann surface with two
charts,
φ1 : C → C, φ1 (z) = z
and
φ2 : Ĉ \ {0} → C, φ2 (z) = 1/z.
Proof. Images of φ1 , φ2 intersect on C \ {0}, and
φ2 ◦ φ−1
1 : C \ {0} → C \ {0}

is analytic because it is simply (φ2 ◦ φ−1

1 )(z) = 1/z. □
Definition 9.14. f : M → N is an analytic map between Riemann
surfaces if for every chart φ : U → C on M and every chart ψ : V → C on
N,
ψ ◦ f ◦ φ−1 : φ(U ∩ f −1 (V )) → ψ(f (U ) ∩ V )
is an analytic function (from one subset of C to another).
170 9. RIEMANN SURFACES

This extends our previous terminology. An analytic function on Ω is an

analytic map f : Ω → C. By Lemma 9.9, a meromorphic function on Ω is
an analytic map f : Ω → Ĉ which is not identically infinite.
This also allows us to consider conformal maps between Riemann sur-
faces and the group Aut(M) of analytic bijections f : M → M for any
Riemann surface M.

3. Meromorphic maps on the Riemann sphere

Recall that a rational function is a ratio P/Q of two polynomials, where
Q is not the zero polynomial.
Lemma 9.15. Any rational function P/Q is a meromorphic function
Ĉ → Ĉ. At ∞, it has:
(a) a removable singularity if deg Q ≥ deg P
(b) a zero of order deg Q − deg P if deg Q > deg P
(c) a pole of order deg P − deg Q if deg Q < deg P
Proof. As a ratio of analytic functions on C, it is meromorphic on C.
Moreover, P (1/z)/Q(1/z) is also a rational function, so it is meromor-
phic on C. The asymptotic behavior
P (1/z)
∼ cz deg Q−deg P , z→0
Q(1/z)
where c is the ratio of leading coefficients of P, Q, implies the other claims.
□
Lemma 9.16. For a rational function f = P/Q where P, Q have no
common factors, f has max{deg P, deg Q} zeros on Ĉ (counted with multi-
plicity).
Proof. Since P, Q have no common factors, Q is nonzero at zeros of P ,
so f has deg P zeros on C (counted with multiplicity). Combining with the
previous lemma concludes the proof.
Since deg(1/f ) = deg f , the function has the same number of poles. □
Conversely:
Lemma 9.17. Every meromorphic function on Ĉ is a rational function.
Proof. By considering f or f −1 , we can assume without loss of gen-
erality that f (∞) ̸= ∞. Since f is meromorphic on Ĉ, its set of poles is
discrete. Since Ĉ is compact, the set of poles is finite. Thus, there exists a
polynomial Q such that f Q is analytic on C.
Taking d = deg Q, we have |f (z)Q(z)| = O(|z|d ) as z → ∞, so by a
consequence of Cauchy estimates, f Q is a polynomial. □
Theorem 9.18. Aut(Ĉ) = PSL(2, C).
 4. EXERCISES 171

Proof. Any f ∈ Aut(Ĉ) is meromorphic on Ĉ, so it is a rational func-

tion. Moreover, it must have deg f = 1, since deg f = 0 means that f is
constant and deg f ≥ 2 means that it is not injective. □
The Zhukovsky map. Note that one can talk about conformal maps
between Riemann surfaces. For instance, by the conformal bijection z 7→
1/z, the domain Ĉ \ [−2, 2] is conformally equivalent to C \ ((−∞, − 21 ] ∪
[ 12 , ∞)), so it is simply connected. Thus, it is conformally equivalent to D.
We can explicitly give a conformal bijection:
Example 9.19. The Zhukovsky map
1
Ψ(z) = z +
z
is a conformal bijection from D to Ĉ \ [−2, 2].
Solution. As a rational function, Ψ : Ĉ → Ĉ is an analytic map.
Clearly
Ψ(z) = ∞ ⇐⇒ z = 0 or z = ∞.
For any w ∈ C,
√
2 w ± w2 − 4
Ψ(z) = w ⇐⇒ z − wz + 1 = 0 ⇐⇒ z = .
2
Since Ψ(z) = Ψ(1/z), the quadratic equation can be in two cases:
(a) solutions have unit absolute value; then they are of the form z = eiϕ and
1/z = e−iϕ so w = 2 cos ϕ. Thus, this happens precisely for w ∈ [−2, 2].
(b) One solution is in D and the other is in C \ D. This happens for all
w ∈ C \ [−2, 2].
In conclusion, we see that Ψ is a bijection from D to Ĉ \ [−2, 2]. □

4. Exercises
1.
 APPENDIX A

Some prerequisites

1. Upper and lower limits (lim sup and lim inf)

The goal of this section is to describe an intuitive way to think about the
limiting behavior of a sequence of real numbers, by identifying the smallest
closed interval which contains the set {xn | n ≥ k} and following how that
interval shrinks as k → ∞. In particular, this leads to notions known as
lim inf, lim sup.
Since these notions rely only on order, they are naturally considered in
the extended real line
R̂ = {−∞} ∪ R ∪ {+∞},
which is a totally ordered set with the obvious definition −∞ < x < +∞
for x ∈ R. On the extended real line, any nonempty subset A ⊂ R̂ has a
supremum; in particular, sup A = +∞ if and only if the set A is not bounded
above.
We can begin by enclosing the set {xn | n ∈ N} by the smallest closed
interval which contains it. This interval is given by [a1 , b1 ] where
a1 = inf xn , b1 = inf xn .
n∈N n∈N

If we choose to ignore the first several terms of the sequence, we instead

consider
ak = inf xn , bk = inf xn .
n≥k n≥k
Since A ⊂ B implies sup A ≤ sup B, we conclude bn+1 ≤ bn . Similarly an ≤
an+1 . Moreover, the infimum of a set is smaller or equal to its supremum.
Thus
an ≤ an+1 ≤ bn+1 ≤ bn ∀n ∈ N.
Thus, the limits of these monotone sequences are given by
a = sup an , b = inf bn
n∈N n∈N

and obey a ≤ b. We call the coefficients a, b the lim inf and lim sup of the
sequence:
Definition A.1. For an arbitrary sequence (xn )∞
n=1 in R̂, we define

lim sup xn = inf sup xn (A.1)

n→∞ k∈N n≥k

173
174 A. SOME PREREQUISITES

lim inf xn = sup inf xn (A.2)

n→∞ k∈N n≥k

Example A.2.
lim sup(−1)n = 1, lim inf (−1)n = −1.
n→∞ n→∞

Solution. For every n,

sup(−1)n = 1, inf (−1)n = −1
n≥k n≥k

so the claim follows. □

Above we have proved:
Theorem A.3. For any bounded sequence (xn )∞
n=1 in R,
lim inf xn ≤ lim sup xn .
n→∞ n→∞

Recall that convergence to x requires that for all ϵ > 0,

x − ϵ < xn < x + ϵ for all large enough n.
In this sense, lim sup and lim inf each capture half of what it takes to be the
limit.
Theorem A.4. A bounded sequence (xn )∞n=1 in R is convergent if and
only if
lim inf xn = lim sup xn
n→∞ n→∞
and in that case, the limit is equal to the lim inf and lim sup.
Proof. We use the notation an , bn , a, b introduced before.
Assume a = b and fix ϵ > 0. There exists n0 such that for all n ≥ n0 ,
an > a − ϵ, bn < b + ϵ.
Combining this with an ≤ xn ≤ bn shows that for all n ≥ n0 ,
a − ϵ < xn < b + ϵ.
Since a = b and this holds for any ϵ > 0, it shows that
lim xn = a = b.
n→∞
Conversely, assume that x = limn→∞ xn exists. Fix ϵ > 0. Then there
exists n0 such that for all n ≥ n0 ,
x − ϵ < xn < x + ϵ.
Thus, taking inf and sup over n ≥ k, we see
x − ϵ ≤ ak ≤ bk ≤ x + ϵ ∀k ≥ n0 .
Thus, taking limits as k → ∞,
x − ϵ ≤ a ≤ b ≤ x + ϵ.
Since ϵ > 0 is arbitrary, this implies a = b = x. □
 1. UPPER AND LOWER LIMITS (lim sup AND lim inf) 175

Theorem A.5. For any sequence,

lim inf xn = − lim sup(−xn ) (A.3)
n→∞ n→∞

Proof. For any k,

inf xn = − sup(−xn ).
n≥k n≥k

Taking the limit of both sides gives (A.3). □

Due to this, properties of lim inf are similar to those of lim sup and we
will not always list them separately.
Theorem A.6. If xn ≤ yn for all large enough n, then
lim inf xn ≤ lim inf yn
n→∞ n→∞
lim sup xn ≤ lim sup yn
n→∞ n→∞

Proof. If xn ≤ yn for all n ≥ n0 , then

sup xn ≤ sup yn ∀k ≥ n0
n≥k n≥k
so
lim sup xn ≤ lim sup yn .
k→∞ n≥k k→∞ n≥k

This proves the inequality for lim sup; the inequality for lim inf is analogous.
□
Theorem A.7 (Squeeze theorem). Consider three sequences such that
xn ≤ yn ≤ zn for all large enough n ∈ N. If
lim xn = lim zn ,
n→∞ n→∞
then the sequence (yn )∞
n=1 is also convergent and
lim xn = lim yn = lim zn .
n→∞ n→∞ n→∞

Proof. Denote
t = lim xn = lim zn .
n→∞ n→∞
Then
t = lim inf xn ≤ lim inf yn ≤ lim sup yn ≤ lim sup zn = t.
n→∞ n→∞ n→∞ n→∞
Thus,
lim inf yn = lim sup yn = t
n→∞ n→∞
which implies limn→∞ yn = t. □
lim sup and lim inf have useful properties for addition:
Theorem A.8. For any bounded sequences (xn )∞ ∞
n=1 , (yn )n=1 ,
lim sup(xn + yn ) ≤ lim sup xn + lim sup yn .
n→∞ n→∞ n→∞
176 A. SOME PREREQUISITES

Combining this with the reflection formula (A.3) allows us to extract

further properties, culminating in the following useful result:
Theorem A.9. For any bounded sequence (xn )∞
n=1 and convergent se-
∞
quence (yn )n=1 ,
lim sup(xn + yn ) = lim sup xn + lim yn .
n→∞ n→∞ n→∞

We leave the proof of these facts as an exercise.

1.1. lim inf and lim sup for unbounded sequences.

Definition A.10. A sequence (xn )∞ n=1 in R is said to be bounded above
if there exists C ∈ R such that xn ≤ C for all n.
It is said to be bounded below if there exists C ∈ R such that xn ≥ C
for all n.
Lemma A.11. Let (xn )∞n=1 be a sequence in R.
The sequence is not bounded above in R if and only if lim supn→∞ xn =
+∞.
The sequence is not bounded below in R if and only if lim inf n→∞ xn =
−∞.
Then by simply considering some additional cases, we can generalize
all previous results. We omit the discussion of cases and merely state the
results. In particular:
Theorem A.12. For any sequence (xn )∞
n=1 in R̂,
lim inf xn ≤ lim sup xn .
n→∞ n→∞
The limit limn→∞ xn exists if and only if
lim inf xn = lim sup xn
n→∞ n→∞
and in that case, the lim inf and lim sup are equal to the limit.
Example A.13. For an arbitrary sequence of real numbers, find the
negation of the statement
lim xn = +∞.
n→∞

Solution. It is not limn→∞ xn < ∞, since in general the sequence may

not be convergent.
If we rewrite the statement as
lim inf xn = ∞ and lim sup xn = +∞
n→∞ n→∞
and note that the second equality follows from the first, we see that the
negation is
lim inf xn < ∞. □
n→∞
 2. CONNECTED SETS 177

2. Connected sets
We begin by considering some topological notions such as connectedness,
for which we have good geometric intuition; our first goal is to make that
intuition precise. In this section, X denotes a metric space.
Definition A.14. Let X be a metric space.
Two subsets A, B ⊂ X are separated if A ∩ B and A ∩ B are empty.
A subset E ⊂ X is connected if, for any separated sets A, B ⊂ X,
E ⊂ A ∪ B implies E ⊂ A or E ⊂ B.
Since A ∩ B = ∅, an equivalent characterization is: a subset E ⊂ X is
connected if, for any separated sets A, B ⊂ X, E ⊂ A ∪ B implies E ∩ A = ∅
or E ∩ B = ∅.
Lemma A.15. Let E ⊂ Y ⊂ X. Then E is connected in X if and only
if it is connected in Y .
Proof. For A ⊂ Y ⊂ X, let us denote by clY A the closure of A in Y .
Recall that clY A = clX A ∩ Y .
Thus, if A, B ⊂ Y ⊂ X, clY A ∩ B = clX A ∩ B.
Thus, if A, B ⊂ Y ⊂ X, sets A, B are separated in Y if and only if they
are separated in X.
Now assume that E is not connected in Y . Then there exist separated
sets A, B ⊂ Y such that E ⊂ A ∪ B, E ̸⊂ A, E ̸⊂ B. Then A, B are also
separated sets in X, so E is not connected in X.
Conversely, assume that E is not connected in X. Then there exist
separated sets A, B ⊂ X such that E ⊂ A ∪ B, E ̸⊂ A, E ̸⊂ B. It follows
that the sets A′ = A ∩ Y , B ′ = B ∩ Y are separated and E ⊂ A′ ∪ B ′ ,
E ̸⊂ A′ , E ̸⊂ B ′ . Thus, E is not connected in Y . □
For this reason, instead of speaking of a connected subset of X, we speak
of a connected set and there is no ambiguity (as long as the metric is known).
To prove an important result about continuous images of connected sets,
we need a prerequisite: a general fact about continuous functions.
Lemma A.16. If f : X → Y is continuous and G ⊂ X, A ⊂ Y , then
f (G) ⊂ A implies f (G) ⊂ A.
This is a standard fact and it can be proved in several ways. It can be
proved using only properties of closures. However, a shorter proof is based
on sequences; recall that x ∈ G if and only if there exists a sequence of
xn ∈ G such that xn → x.
Proof. Let x ∈ G. Then there exists a sequence of xn ∈ G with
xn → x. Then f (xn ) → f (x) and f (xn ) ∈ f (G) ⊂ A, so f (x) ∈ A. □
We can now state and prove the main property of connected sets:
Theorem A.17. The continuous image of a connected set is connected:
if f : X → Y is continuous and E ⊂ X is connected, then f (E) is connected.
178 A. SOME PREREQUISITES

Proof. We will prove the contrapositive. Assume that f (E) is not

connected. Then, for some separated sets A, B in Y , f (E) ⊂ A ∪ B, E ∩ A ̸=
∅, E ∩ B ̸= ∅.
Denote G = f −1 (A) and H = f −1 (B). In particular, f (G) ⊂ A. Thus,
by Lemma A.16, f (G) ⊂ A. Since f (Ḡ) ⊂ Ā, f (H) ⊂ B, and Ā ∩ B = ∅, it
follows that Ḡ ∩ H = ∅. Analogously G ∩ H = ∅, so G, H are separated in
X.
Moreover, E ⊂ f −1 (f (E)) ⊂ G ∪ H.
There exists y ∈ f (E) ∩ A. Then y = f (x) for some x ∈ E and f (x) =
y ∈ A, so x ∈ E ∩ f −1 (A). Thus, E ∩ G ̸= ∅. Analogously, E ∩ H ̸= ∅. It
follows that E is not connected. □
Connected subsets of R.
Example A.18. Any interval I ⊂ R is connected.
Proof. We prove this by contradiction. Assume I is an interval and
that I ⊂ A ∪ B, where A, B are separated and intersect I. Pick some
elements x ∈ I ∩ A and y ∈ I ∩ B and assume without loss of generality that
x < y. Then in particular, [x, y] ⊂ I.
The set A ∩ [x, y] is nonempty (since it contains x) and bounded above
by y. Denote
z = sup(A ∩ [x, y]).
The supremum of a set is a closure point, so
z ∈ A ∩ [x, y] ⊂ A.
This implies z ∈
/ B. In particular, z < y. Since I is an interval, the definition
of z now implies (z, y) ⊂ B and therefore z ∈ B. It follows that z ∈ / A. □
Lemma A.19. Assume E ⊂ R is connected. If x is not a lower bound
for E and x is not an upper bound for E, then x ∈ E.
Proof. The intervals (−∞, x) and (x, ∞) are separated. If x ∈ / E,
then E ⊂ (−∞, x) ∪ (x, ∞). Thus, by connectedness, E ⊂ (−∞, x) or
E ⊂ (x, ∞), which implies that x is an upper bound for E or a lower bound
for E, respectively. □
Theorem A.20. The connected sets E ⊂ R are: ∅, single point sets
{x} for all x ∈ R, and all intervals: (a, b), (a, b], [a, b), [a, b] for all a < b,
(a, ∞), [a, ∞), (−∞, b), (−∞, b], R.
Proof. If E = ∅ or E = {x} for some x, it is trivially connected. From
now on assume that E contains at least two points.
Denote a = inf E and b = sup E, allowing infinite values if E is un-
bounded. Then E ⊂ [a, b] (by definition of inf and sup) and a < b (be-
cause E contains at least two points). For any x ∈ (a, b), x is not an
upper bound of E and not a lower bound of E, so x ∈ E. This proves
(a, b) ⊂ E. Now (a, b) ⊂ E ⊂ [a, b] implies that E is one of the intervals
(a, b), (a, b], [a, b), [a, b].
 2. CONNECTED SETS 179

If E is bounded below but not above, denote a = inf E and similarly

prove (a, ∞) ⊂ E ⊂ [a, ∞). The other cases are proved similarly.
Conversely, as we have already shown, every interval is connected. □
For instance, this implies the intermediate value theorem from real anal-
ysis:
Theorem A.21 (Intermediate value theorem). Let f : [a, b] → R be a
continuous function. If f (a) < f (b), then for any c such that f (a) < c <
f (b), there exists x ∈ (a, b) such that f (x) = c.
Proof. Since [a, b] is an interval, it is connected. Since f is continuous,
f ([a, b]) is connected.
If f (a) < c < f (b), then c is not a lower bound for f ([a, b]) and not an
upper bound, so there exists x ∈ (a, b) such that f (x) = c. □
Corollary A.22 (Intermediate value theorem). Let f : [a, b] → R be a
continuous function. Then f ([a, b]) is an interval or a single point set.
 APPENDIX B

Class schedule Fall 2025

Homework is due Wednesdays at 11:59 PM.

Week 1:
08/27: Introduction, cubic equations, field of complex numbers, abs. value

Week 2:
09/01: no class (Labor day)
09/03: Euclidean metric, convexity, affine maps, inversion

Week 3:
09/08: holomorphic functions
09/10: power series and term-by-term differentiation

Week 4:
09/15: exponential and trigonometric functions
09/17: polar form, FTC for curves, rectifiable curves

Week 5:
09/22: piecewise C 1 curves, contour integrals, antiderivatives
09/24: arclength reparametrization. connectedness and path-connectedness

Week 6:
09/29: Goursat argument on triangles. Cauchy integral theorem
10/01: Cauchy integral theorem on simply connected domains. Branch of log

Week 7:
10/06: Branches of log and roots, Cauchy integral formula
10/08: The Cauchy power series theorem and applications

Week 8:
10/13: no class (Fall break)
10/15: Fundamental theorem of algebra, zeros of analytic functions, Mor-
era’s theorem

Week 9:
10/20: Reflection principle, limits of analytic functions
181
182 B. CLASS SCHEDULE FALL 2025

10/22: Montel’s theorem, parametric integrals of analytic functions

Week 10:
10/27: Products of analytic functions, Weierstrass theorem
10/29: Gamma function, winding numbers

Week 11:
11/03: Argument principle, Rouché theorem
11/05: Hurwitz theorem, inverse function theorem, local behavior at critical
points

Week 12:
11/10: open mapping theorem, maximum principle, Schwarz lemma
11/12: Conformal bijections, Riemann mapping theorem

Week 13:
11/17: Riemann mapping theorem, Cauchy integral formula on the annulus
11/19: Laurent series and classification of isolated singularities

Week 14:
11/24: Casorati–Weierstrass, Aut(C), meromorphic functions, principal parts
11/26: no class (Thanksgiving)

Week 15:
12/01: Residue theorem and applications
12/03: Riemann sphere and a glimpse of Riemann surfaces

Week 16:
12/08: Harmonic functions and a glimpse of applications
 APPENDIX C

Some superfluous notes

1. Rectifiability and improper integrals

Lemma C.1. For any curve γ : [a, b] → C,


ℓ(γ) = sup ℓ(γ|[a,τ ] ) + |γ(τ ) − γ(b)| .
τ ∈[a,b)

Proof. The quantity ℓ(γ|[a,τ ] ) + |γ(τ ) − γ(b)| is precisely the supremum

over those partitions for which tn−1 = τ . □
Lemma C.2. For any curve γ : [a, b] → C,
ℓ(γ) = lim ℓ(γ|[a,τ ] ).
τ ↑b

Proof. We define f on (a, b] by

f (τ ) = ℓ(γ|[a,τ ] ) + |γ(τ ) − γ(b)|.
Note that f (b) = ℓ(γ).
If τ1 < τ2 , then by the previous lemma,
ℓ(γ|[a,τ1 ] ) + |γ(τ1 ) − γ(τ2 )| ≤ ℓ(γ|[a,τ2 ] )
Adding to this the triangle inequality
|γ(τ1 ) − γ(b)| ≤ |γ(τ1 ) − γ(τ2 )| + |γ(τ2 ) − γ(b)|
we conclude that f (τ1 ) ≤ f (τ2 ). Thus, f is an increasing function, so
ℓ(γ) = sup f (τ ) = lim f (τ ).
τ ∈[a,b) τ ↑b

By continuity, γ(τ ) → γ(b) as τ ↑ b, which concludes the proof. □

√
Example C.3. The curve γ(t) = 1 − t2 + it, t ∈ [0, 1], is not C 1 , but
it is rectifiable and Z τ
1
ℓ(γ) = lim √ dt. (C.1)
τ ↑1 0 1 − t2
√
Its length satisfies 2 ≤ ℓ(γ) ≤ 2.
Proof. Since γ|[0,τ ] is C 1 for each τ < 1 and |γ ′ (t)| = (1−t2 )−1/2 , using
the formula for the length of C 1 curves and we obtain the representation
(C.1). Using 1 − t2 ≥ 1 − t we obtain
√
Z τ
1
ℓ(γ) ≤ lim √ dt = 2 1 − t|10 = 2,
τ ↑1 0 1−t
183
184 C. SOME SUPERFLUOUS NOTES

and using 1 − t2 = (1 + t)(1 − t) ≤ 2(1 − t) we obtain

Z τ
1 √ √ √
ℓ(γ) ≥ lim p dt = 2 1 − t|10 = 2. □
τ ↑1 0 2(1 − t)
Denote π = 2ℓ(γ) where γ is the curve in the previous example. Assume
that cos has no positive zeros in (0, π/2). Denote
L = min ({π/2} ∪ {t > 0 | cos t = 0}) .
Obviously 0 < L ≤ π/2.
Consider the curve
γ1 (s) = eis , s ∈ [0, L].
Since cos is positive on [0, L), sin is strictly increasing, and Im γ1 : [0, L] →
[0, sin L] is a bijection. We conclude
γ1 (s) = γ(sin s).
Since sin is strictly increasing, it maps partitions of [0, L] to partitions of
[0, sin L], and we conclude that
Z L
ℓ(γ|[0,sin L] ) = ℓ(γ1 ) = |γ1′ (s)| ds = L.
0
From this line, we read off that L = π/2 if and only if sin L = 1, and this is
if and only if cos L = 0. Thus, by our definition of L, all these statements
must be true. In particular, L = π/2 is the smallest positive zero of cos.
Remark C.4. Another way to parametrize this arc is
1 − t2 2t
γ2 (t) = +i , t ∈ [0, 1].
1 + t2 1 + t2
This is a C 1 curve and |γ2 (t)| = 2/(1 + t2 ) so
Z 1
dt
ℓ(γ2 ) = 2 .
0 1 + t2