Riemann Surfaces

Henry Wilton∗
Michaelmas 2020

Contents

1 Analytic and meromorphic functions 3

1.1 Analytic functions and their zeroes . . . . . . . . . . . . . . . . 3
1.2 Meromorphic functions and singularities . . . . . . . . . . . . . 5
1.3 Analytic continuation . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 The complex logarithm . . . . . . . . . . . . . . . . . . . . . . . 9

2 Natural boundary, gluing constructions and roots 11

2.1 Natural boundary . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 A gluing construction . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Complex roots . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Riemann surfaces and analytic maps 16

3.1 Covering maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Abstract Riemann surfaces . . . . . . . . . . . . . . . . . . . . . 17
3.3 Analytic maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4 Examples of conformal structures and analytic functions 22

4.1 Covering maps and analyticity . . . . . . . . . . . . . . . . . . . 22
4.2 Analytic functions . . . . . . . . . . . . . . . . . . . . . . . . . . 24
∗
Comments and corrections are always welcome. Please send them to
[Link]@[Link].

1
5 Complex tori and the open mapping theorem 25
5.1 Complex tori . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.2 The open mapping theorem . . . . . . . . . . . . . . . . . . . . 27
5.3 Harmonic functions . . . . . . . . . . . . . . . . . . . . . . . . . 27

6 Meromorphic functions and a worked example 30

6.1 Meromorphic functions . . . . . . . . . . . . . . . . . . . . . . . 30
6.2 A worked example . . . . . . . . . . . . . . . . . . . . . . . . . . 31

7 Covering-space theory 34
7.1 Covering-space theory . . . . . . . . . . . . . . . . . . . . . . . . 34

8 The monodromy group and the space of germs 38

8.1 The monodromy group . . . . . . . . . . . . . . . . . . . . . . . 38
8.2 The space of germs . . . . . . . . . . . . . . . . . . . . . . . . . . 40

9 Uniqueness of analytic continuation and gluing 43

9.1 Analytic continuation revisited . . . . . . . . . . . . . . . . . . 43
9.2 The classical monodromy theorem . . . . . . . . . . . . . . . . 44
9.3 Gluing Riemann surfaces . . . . . . . . . . . . . . . . . . . . . . 45

10 More gluing and branching 47

10.1 A more detailed gluing example . . . . . . . . . . . . . . . . . . 47
10.2 Branching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

11 The valency and Riemann–Hurwitz theorems 52

11.1 The valency theorem . . . . . . . . . . . . . . . . . . . . . . . . . 52
11.2 Euler characteristic . . . . . . . . . . . . . . . . . . . . . . . . . 53
11.3 The Riemann–Hurwitz theorem . . . . . . . . . . . . . . . . . . 55

12 Applications of Riemann–Hurwitz 56
12.1 Immediate consequences . . . . . . . . . . . . . . . . . . . . . . 56
12.2 Higher-genus Riemann surfaces . . . . . . . . . . . . . . . . . . 58

13 Rational and periodic functions 60

13.1 Rational functions . . . . . . . . . . . . . . . . . . . . . . . . . . 60
13.2 Simply periodic functions . . . . . . . . . . . . . . . . . . . . . . 61
13.3 Doubly periodic functions . . . . . . . . . . . . . . . . . . . . . . 63

2
14 The Weierstrass ℘-function 65
14.1 The definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
14.2 Branching properties of ℘Λ . . . . . . . . . . . . . . . . . . . . . 68
14.3 An algebraic relation . . . . . . . . . . . . . . . . . . . . . . . . . 69

15 More ℘-function and quotients 70

15.1 An elliptic curve . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
15.2 Classification of elliptic functions . . . . . . . . . . . . . . . . . 71
15.3 Quotients of Riemann surfaces . . . . . . . . . . . . . . . . . . . 73

16 Uniformisation and its consequences 75

16.1 The uniformisation theorem . . . . . . . . . . . . . . . . . . . . 75
16.2 Classification of Riemann surfaces . . . . . . . . . . . . . . . . . 77
16.3 Consequences of uniformisation . . . . . . . . . . . . . . . . . . 80

Lecture 1: Analytic and meromorphic functions

This course picks up where IB Complex Analysis leaves off. The IB course
contains many classical results about complex-differentiable functions of one
variable. However, it also raises some natural questions, such as:

‘What do we really mean by multi-valued functions, such as the

complex logarithm or mth roots?’

The reader may like to keep this question in mind as this course begins. We
will see that these questions naturally lead us to define and study a whole new
class of mathematical objects, the Riemann surfaces of the course title. As
we shall see, Riemann surfaces exhibit a beautiful interplay between analysis
and geometry.
Since the course leans heavily on some of the results of IB Complex Anal-
ysis, we will start by recalling some of the definitions and results from that
course.

1.1 Analytic functions and their zeroes

The functions we study will be defined on domains. A domain is an open,
connected subset of the complex plane C. Two of the most important kinds

3
of domains are the (open) disc

D(z0 , r) ∶= {z ∈ C ∣ ∣z − z0 ∣ < r}

and the (open) punctured disc

D∗ (z0 , r) ∶= {z ∈ C ∣ 0 < ∣z − z0 ∣ < r} .

More generally, a neighbourhood U of a point z0 in a domain D is any subdo-

main U ⊆ D containing z0 , and a punctured neighbourhood of z0 is any open
set of the form U ∖ {z0 }, where U is a neighbourhood of z0 .
Definition 1.1. Let D ⊆ C be a domain. A function f ∶ D → C is called
holomorphic or analytic if either of the following two equivalent definitions
are satisfied:
(i) f is C-differentiable at every z0 ∈ D; or
(ii) for any z0 ∈ D, there is r > 0 such that f has a power-series expansion
∞
f (z) = ∑ an (z − z0 )n
n=0

for any z ∈ D(z0 , r) ⊆ D.

More precisely, Definition 1.1(i) is the definition of a holomorphic func-
tion, and Definition 1.1(ii) is the definition of an analytic function. It is
a theorem of IB Complex Analysis that the two definitions coincide. As a
consequence, analytic functions are surprisingly rigid: many behaviours that
we are used to from real analysis are impossible in the complex setting.
Proposition 1.2 (Principle of isolated zeroes). Let f ∶ D → C be an analytic
function on a domain D ⊆ C. If f (z0 ) = 0, then either f is identically zero
in a neighbourhood of z0 , or f is non-zero on a punctured neighbourhood of
z0 .
Proof. Unless f ≡ 0 in a neighbourhood of z0 , there is a minimal m ≥ 0 such
that am ≠ 0. Hence,
f (z) = (z − z0 )m g(z)
for some analytic function g, defined on an open disc D(z0 , ρ), with g(z0 ) ≠ 0.
By continuity of g, there is r > 0 with g(z) ≠ 0 for all z ∈ D(z0 , r), and the
result follows.

4
 It follows that, if two analytic functions on a domain agree fairly often,
then they are equal. Here, ‘fairly often’ means, precisely, on a non-discrete
subset. A subset A of C is called discrete if the subspace topology induced
on A is the discrete topology. Equivalently, A is discrete unless some a ∈ A
is a limit of a sequence in its complement A ∖ {a}.

Corollary 1.3 (Identity principle). Let f, g be analytic functions defined on

a domain D in C. Unless the set

{z ∈ D ∣ f (z) = g(z)}

is discrete, f ≡ g on D.

Proof. The proof is an easy application of the fact that D is connected.

Let A be the set of points z ∈ D on which f and g agree in a punctured
neighbourhood of z: that is,

A ∶= {z ∈ U ∣ ∃r > 0, ∀z ∈ D(z, r) , f (z) = g(z)} .

Likewise, let D be the set of points on which f and g disagree in some

punctured neighbourhood: that is,

B ∶= {z ∈ D ∣ ∃r > 0, ∀z ∈ D∗ (z, r) , f (z) ≠ g(z)} .

The sets A and B are open and disjoint by definition. More surprisingly, the
principle of isolated zeros applied to the analytic function f − g implies that
A and B together cover D. Therefore, either A or B must be empty, because
D is connected. If f and g agree on a non-discrete subset, then it follows
that B does not cover D, so A = D and the result follows.

1.2 Meromorphic functions and singularities

Singularities arise when analytic functions are defined on punctured discs.

Definition 1.4. An analytic function f ∶ D∗ (z0 , r) → C is said to have an

isolated singularity at z0 .

Just as holomorphic functions have Taylor series, so you saw in IB Com-

plex Analysis that analytic functions have Laurent series at their singularities.

5
Proposition 1.5 (Laurent series). If an analytic function f has an isolated
singularity at z0 , then
∞
f (z) = ∑ an (z − z0 )n
n=−∞

on D∗ (z0 , r) for some r > 0.

The coefficients an lead to a natural classification of singularities.

Definition 1.6 (Classification of singularities). Suppose that f has an iso-

lated singularity at z0 , so
∞
f (z) = ∑ an (z − z0 )n
n=−∞

on a punctured neighbourhood of z0 .

(i) If an = 0 for n < 0 then z0 is a removable singularity. In this case, f can

be extended to an analytic function g(z) defined on a neighbourhood
of z0 .

(ii) If there is m > 0 such that am = 0 for all n < −m but a−m ≠ 0, then f is
said to have a pole of order m at z0 . In this case,

f (z) = (z − z0 )−m g(z)

on a neighbourhood of z0 , for some analytic function g with g(z0 ) ≠ 0.

(iii) Otherwise (that is, if an ≠ 0 for infinitely many n < 0), then f is said to
have an essential singularity at z0 .

Fortunately, we don’t need to classify singularities by calculating Laurent

series. The following two important results are also proved in IB Complex
Analysis.

Theorem 1.7 (Removable singularities). An analytic function f has a re-

movable singularity at z0 if and only if f is bounded on some punctured disc
D∗ (z0 , r).

The corresponding theorem for essential singularities plays a very impor-

tant role in this course.

6
Theorem 1.8 (Casorati–Weierstrass). An analytic function f on a domain
D has an essential singularity at z0 if and only if f (D∗ (z0 , r)) is dense in
C, for any r > 0 such that D∗ (z0 , r) ⊆ D.

Proof. For the ‘only if’ direction, notice that f (D∗ (z0 , r)) is not dense if z0
is removable or a pole. Indeed, if z0 is removable, then some f (D∗ (z0 , r))
is bounded, and in particular not dense, by Theorem 1.7. If z0 is a pole of
order m, then
f (z) = (z − z0 )−m g(z)
where g is analytic and non-zero in a neighbourhood of z0 . There is an  > 0
so that
∣g(z)∣ ≥  > 0
on some punctured disc D∗ (z0 , r), and therefore, on that disc,
∣g(z)∣ 
∣f (z)∣ ≥ > m.
∣z − z0 ∣ m r
So f is bounded away from 0 on D∗ (z0 , r), and not dense.
For the other direction, if f (D∗ (z0 , r)) is not dense, then there is some
open disc D(w0 , ) disjoint from f (D∗ (z0 , r)). Consider the function
1
h(z) =
f (z) − w0
defined on D∗ (z0 , r). Since ∣f (z) − w0 ∣ ≥  whenever z ∈ D∗ (z0 , r), it follows
that ∣h(z)∣ ≤ 1/. By Theorem 1.7, h has a removable singularity at z0 , and
can be extended across z0 . Writing
1
f (z) = + w0 ,
h(z)
it follows that f has a removable singularity at z0 , unless h(z0 ) = 0, in which
case f has a pole.
Poles are almost as nice a removable singularities, so it is useful to have
terminology for functions with poles.
Definition 1.9. Let D be a domain in C. If there is a discrete subset A of
D, and f is a holomorphic function on D ∖ A with poles at the points of A,
then f is said to be a meromorphic function on D.

7
 We close with an example.
Example 1.10. Let
1
f (z) = ,
−1 e1/z
which is analytic on the complement of A = {0} ∪ {1/2πin ∣ n ∈ Z ∖ {0}}.
From the power series expansion of the exponential function, we see that
g(w) = ew − 1
has a simple zero at w = 1, and hence, by periodocity, at every point of 2πiZ.
Therefore, h(z) = e1/z − 1 has a simple zero at 1/2πin, for each non-zero
integer n, and so f has simple poles at those points.
However, 0 must be an essential singularity of f . Indeed, h(z) = 0 at a
sequence of points tending to 0 so, by the principle of isolated zeroes, if 0 were
a removable singularity then h would be 0 in a neighbourhood of 0, which is
absurd. Furthermore, if h has a pole at 0, then ∣h(z)∣ > 0 in a neighbourhood
of 0, which is also a contradiction. Therefore, h has an essential singularity
at 0, and f does too.

1.3 Analytic continuation

The identity principle leads to a very surprising rigidity for analytic functions.
Given analytic f defined on a small open disc D1 = D(z1 , r1 ), and another
small open disc D2 that intersects D1 , there is at most one way to extend
f analytically across D2 ! This move from one disc to a neighbouring one is
called analytic continuation.
Definition 1.11. Let D be a domain in C. A function element on D is a
pair (f, U ), where U is a subdomain in D and f is an analytic function on
U . If (g, V ) is another function element on D, then
(f, U ) ∼ (g, V )
means that U ∩ V ≠ ∅ and f ∣U ∩V = g∣U ∩V . In this case, (g, V ) is said to be a
direct analytic continuation of (f, U ). If there is a finite sequence of direct
analytic continuations
(f, U ) = (f1 , U1 ) ∼ ⋯ ∼ (fn−1 , Un−1 ) ∼ (fn , Un ) = (g, V )
then (g, V ) is said to be an analytic continuation of (f, U ), and we write
(f, U ) ≈ (g, V ).

8
 Note that ≈ is an equivalence relation.

Definition 1.12. A ≈-equivalence class F of function elements on a domain

D is called a complete analytic function on D.

1.4 The complex logarithm

If (f, U ) ∼ (g, V ) then, by the identity principle, g is determined by f . It is
tempting to think that this observation extends from direct analytic continu-
ations to all analytic continuations. However, that is not quite correct, as we
shall illustrate with one of the most famous motivating examples in complex
analysis. The notation C∗ denotes the punctured complex plane C ∖ {0}.
Example 1.13 (The complex logarithm). The complex logarithm log arises
from attempting to invert the exponential function exp ∶ C → C∗ . This isn’t
possible globally, because exp is not injective: for instance,

exp(0) = 1 = exp(2πi) .

Therefore, to define log as a function, we need to restrict attention to a

subdomain of C on which exp is injective. In the language of IB Complex
Analysis, we could do this explicitly using branch cuts. In the language of
this course, any definition of log on a subdomain U of C defines a function el-
ement. In total, although log isn’t strictly speaking a function, this collection
of function elements defines log as a complete analytic function.
To be concrete, let’s exhibit some explicit function elements that enable
us to make sense of all possible choices of log, everywhere on C∗ .
Given any (α, β) ⊆ R with ∣α − β∣ < 2π, define

U(α,β) = {reiθ ∣ r > 0, α < θ < β} .

For instance, U(0,2π) = C ∖ R≥0 . For z ∈ U(α,β) , which we think of as z = reiθ

with θ ∈ (α, β), define
f(α,β) (z) = log r + iθ ,
where log is the real logarithm. It is not hard to check that f(α,β) is analytic:
for instance, noting that the real part is independent of θ and the imaginary
part is independent of r, the Cauchy–Riemann equations in polar coordinates
reduce to
∂u 1 ∂v
=
∂r r ∂θ
9
which is true in this case. Therefore, together, these define function elements
F(α,β) = (f(α,β) , U(α,β) ).
To understand the complex logarithm, we just need to choose a collection
of open intervals that cover R. For instance, we can take the intervals

I(n) = ((n − 1)π/2, (n + 1)π/2)

across all n ∈ Z. Next, we need to analyse what happens when these intervals
overlap, to determine when FI(m) ∼ FI(n) .
There are four cases to consider, depending on m − n modulo 4.

(i) If m ≡ n modulo 4 the UI(m) = UI(n) , but I(m) and I(n) are disjoint
unless m = n. Therefore, FI(m) ∼ FI(n) if and only if m = n.

(ii) If m ≡ n + 1 modulo 4 then UI(m) ∩ UI(n) is a quadrant of C, but I(m)

and I(n) are disjoint unless m = n + 1. In this last case, fI(m) and fI(n)
agree on UI(m) ∩ UI(n) . Therefore, FI(m) ∼ FI(n) if and only if m = n + 1.

(iii) If m ≡ n+2 modulo 4 then UI(m) and UI(n) are disjoint, so FI(m) ≁ FI(n) .

(iv) If m ≡ n + 3 modulo 4 then, as in the case (ii), FI(m) ∼ FI(n) if and only
if m = n − 1.

In conclusion, FI(m) ∼ FI(n) if and only if ∣m − n∣ ≤ 1.

Applying these direct analytic continuations iteratively, all of the function
elements FI(n) are in the same ≈-equivalence class, so they all define the same
complete analytic function. This is the complex logarithm.
This example shows us that analytic continuation may not be unique.
Indeed, in the notation of Example 1.13, UI(0) = UI(4) , but fI(0) (1) = 0
whereas fI(4) (1) = 2πi. This is of course the familiar phenomenon that,
as we follow a clockwise loop around the origin, the value of the logarithm
changes by 2πi. To make this precise, we introduce the idea of analytically
continuing along a curve.

Definition 1.14 (Analytic continuation along a path). Let (f, U ) be a

function element in the domain D, and consider an analytic continuation
(f, U ) ≈ (g, V ), exhibited by a sequence of direct analytic continuations as
in the definition.

(f, U ) = (f1 , U1 ) ∼ . . . ∼ (fn−1 , Un−1 ) ∼ (fn , Un ) = (g, V ) ;

10
Let γ ∶ [0, 1] → D be a continuous path. If there is a dissection

0 = t0 < t1 < . . . < tn−1 < tn = 1

such that γ([ti−1 , ti ]) ⊆ Ui for each 1 ≤ i ≤ n, then (g, V ) is an analytic

continuation of (f, U ) along γ, and we write (f, U ) ≈γ (g, V ).
Remark 1.15. If (f, U ) ≈ (g, V ) then (f, U ) ≈γ (g, V ) for some path γ.
Although we saw above that analytic continuations are not always unique,
it turns out that they are determined the path γ: if (f, U ) ≈γ (g, V ), then
g(γ(1)) only depends on (f, U ) and γ. This can be proved directly, although
it also follows easily from later results in the course. Indeed, in due course
we shall see that much more is true.

Lecture 2: Natural boundary, gluing construc-

tions and roots
2.1 Natural boundary
In the last lecture we saw that, when we try to analytically continue, we may
build very large complete analytic functions, taking infinitely many values
at any point. On the other hand, sometimes we simply cannot analytically
continue very far. This phenomenon is the subject of the next subsection.
Let’s introduce some new notation for some of our favourite subsets of C:
D is the open unit disc D(0, 1), and T is its boundary, the unit circle.
Consider a power series

f (z) = ∑ an z n
n≥0

with radius of convergence 1 (without loss of generality). In particular, the

series converges absolutely and uniformly on any closed disc D ⊆ D.
Definition 2.1. A point z0 ∈ T is called regular for f if there is an open
neighbourhood U of z0 and an analytic function g on U such that g ≡ f on
U ∩ D. Otherwise, z0 is called singular for f .
The point is, of course, that we may analytically continue to the regular
points, but not to the singular points.

11
Remark 2.2. The set of regular points in ∂T is open by definition; hence, the
set of singular points is closed.
Beware! It is not the case that z0 is a regular point if and only if ∑n≥0 an z0n
converges; indeed, neither implication is true.
Example 2.3. Consider the power series
1
f (z) = = ∑ zn ;
1 − z n≥0

evidently, every point of T ∖ {1} is regular. However, the power series does
not converge at z0 = −1.
Example 2.4. Consider the power series
zn
g(z) = ∑ .
n≥2 n(n − 1)

The series
1
g(1) = ∑
n≥2 n(n − 1)

is convergent, but if 1 were a regular point for g, then it would follow that it
is also regular for g ′′ = f , which is absurd since 1 is a pole of f .
There is always at least one singular point.

Proposition 2.5. If a power series

f (z) = ∑ an z n
n≥0

has radius of convergence 1, then some point of T is singular for f .

Proof. If not, for each z ∈ T there is z > 0 such that f extends analytically
over D(z, z ). Finitely many of these discs cover T by compactness, so f
extends analytically over some D(0, 1 + δ) with δ > 0. But this implies that
the radius of convergence is at least 1 + δ, which is a contradiction.
The concept of natural boundary refers to the extreme case where the
number of singular points is as large as possible.

Definition 2.6. If every point of T is singular, then T is said to be a natural

boundary for f .

12
 This can happen in practice, as the following example shows.
Example 2.7. Let
f (z) = ∑ z n! ,
n≥0

which has radius of convergence 1 by the ratio test, and consider a qth root
of unity ω = exp(2πip/q). We will show that ω is singular, and conclude that
f has natural boundary T by Remark 2.2, since roots of unity are dense in
T.
Whenever 0 < r < 1, we have
q−1
f (rω) = ∑ rn! ω n! + ∑ rn! .
n=0 n≥q

The second term diverges to ∞ as r → 1: indeed,

q+M
lim ∑ rn! = M + 1 ,
r→1 n=q

for any real number M , so for r large enough,

q+M
∑ rn! > ∑ rn! > M .
n≥q n=q

However, if ω were a regular value then f (rω) would converge to f (ω) as

r → 1, so ω must be singular, as claimed.
There are easy extensions of the notion of natural boundary to other
curves in the plane; for instance, it is not difficult to adapt Example 2.7 to
define a function with natural boundary equal to R. However, in general,
the natural boundary of an analytic function can be extremely complicated,
and a complete treatment is beyond the scope of this course.

2.2 A gluing construction

In the previous lecture, we constructed the complex logarithm as a collection
of function elements. An important idea in this course is that these collections
can be glued together to give geometric objects, whose geometry encodes the
function we are interested in. In this subsection, we will briefly outline that
construction for the complex logarithm.

13
 The idea is to take a quotient of the disjoint union of the domains of the
function elements. In the notation of Example 1.13, let

R = (∐ UI(n) ) / ∼
n∈Z

where ∼ identifies z1 ∈ UI(m) and z2 ∈ UI(n) if and only if z1 = z2 in C and,

furthermore, fI(m) (z1 ) = fI(n) (z2 ); in other words, z1 and z2 are the same
point in UI(m) ∩UI(n) , and FI(n) is a direct analytic continuation of FI(m) . By
giving R the quotient topology, it becomes a topological space. Furthermore,
R is path-connected, because all of the function elements F● are in the same
complete analytic function.
Informally, this construction can be thought of as an ‘infinite multi-storey
car park’: standing on the level UI(n) , one can go up through level UI(n+1)
to the level above, UI(n+1) ; alternatively, one can go down through the level
UI(n−1) to the level below, UI(n−2) .
As well as being an abstract space, R is equipped with two well-defined
functions to C. The first comes from the function elements F● = (f● , U● ).
Since fA and fB agree wherever points of UA and UB are identified in R,
there is a well-defined map
f ∶R→C
defined as follows: given [z] ∈ R, choose any representative z ∈ UA (where
A = I(n) or A = J(n), for some n) and set f ([z]) = fA (z). The second is
defined in the same way, but using the natural inclusion maps π● ∶ U● ↪ C.
Again, these agree whenever two points are identified in R, and so together
give a well-defined map π ∶ R → C.
Remark 2.8. Since each f● is defined as an inverse to the exponential map,
it follows that exp ○f● ≡ π● , and so exp ○f ≡ π.
The functions f and π can be used together to define a continuous map

Φ ∶ R → C2
z ↦ (π(z), f (z))

which can be used to study R. Looking back at the definition of the relation
that defines R, we see that

z1 ∼ z2 ⇔ z1 = z2 and f (z1 ) = f (z2 ) ⇔ Φ(z1 ) = Φ(z2 )

14
so Φ is injective. In particular, R is homeomorphic to a subset of C2 , so is
Hausdorff. In fact, with a little thought, we can see that this map identifies
R with the graph of the exponential function,

{(z, w) ∈ C2 ∣ w = exp(z)} .

We shall see much more of this as the course goes on!

The moral of this construction is that, as we try to analytically continue,
we may build functions whose domain (the ‘infinite multi-storey car park’)
is much larger than the original domain that we started with (C∗ ).

2.3 Complex roots

Along with the complex logarithm, kth roots provide a family of important
examples of multi-valued functions, arising, of course, in an attempt to invert
the kth power map pk ∶ z ↦ z k . These also define complete analytic functions,
just as we saw the complex logarithm does in §1.4. Having already handled
the complex logarithm, it is easy to deal with kth roots by thinking of them
as
√ 1
k
z = exp ( log z) .
k
Example 2.9. Consider intervals I(n) ⊆ R and open sets UI(n) as in Example
1.13, with each U● equipped with an analytic branch of log, f● . Define
1
g● (z) = exp ( f● (z))
k
to obtain a set of function elements GI(n) ∶= (gI(n) , UI(n) ). These function
elements only depend on n modulo 4k, and so this time we may as well
n ∈ Z/4kZ. With that modification, the analysis proceeds as before: we have

GI(m) ∼ GI(n) ,

if and only if m − n = 0, ±1 modulo 4k, so the function elements G● define a

complete analytic function, which is the complex kth root.
Furthermore, as in §2.2, we may proceed to define a path-connected,
Hausdorff topological space Rk , equipped with a pair of functions π, g ∶ Rm →
C, satisfying π(z) = g(z)k .

15
Lecture 3: Riemann surfaces and analytic maps
3.1 Covering maps
In §2.2, we constructed a topological space R that fitted into a commutative
diagram of maps.
f
R C
π exp

C∗
If π were a bijection, then the function f ○ π −1 would be a single-valued
complex logarithm. In fact, π isn’t a bijection, but from a topological point
of view it is the next best thing.
Definition 3.1 (Covering map). Let X and X ̃ be path-connected, Hausdorff,
̃
topological spaces. A covering map π ∶ X → X is a local homeomorphism:
̃ has an open neighbourhood U
that is, each x̃ ∈ X ̃ such that π∣ ̃ is a homeo-
U
morphism onto its image.
A covering map π ∶ X ̃ → X is regular if, for each x ∈ X, there is an open
neighbourhood U of x and a discrete set ∆x such that π −1 (U ) is homeomor-
phic to the direct product U × ∆x and the diagram
≅
π −1 (U ) U × ∆x
π

U
commutes, where the map U × ∆x → U is projection onto the first factor.
Remark 3.2. Beware! The regular covering maps of Definition 3.1 are just
called ‘covering maps’ in II Algebraic Topology.
The definition of a regular covering map is often illustrated with a ‘stack
of records’ picture.
Example 3.3. The map π ∶ R → C∗ defined in §2.2 is a covering map, indeed a
regular covering map. Each z ∈ C∗ lives in at least one UI(n) – the left, right,
upper or lower half-plane – and in each case the preimage is of the claimed
form:
π −1 (UI(n) ) = ∐ UI(m) ≅ UI(n) × Z .
m≅n mod 4

16
 This example is closely related to the exponential map itself.
Example 3.4. For each open interval I ⊆ R, let

ṼI = R + iI = {x + iy ∣ x ∈ R, y ∈ I} .

As long as I is of diameter at most 2π, the exponential map restricts to a

homeomorphism ṼI → UI , with inverse provided by the map fI of Example
1.13. As in the previous example, every z ∈ C∗ is contained in some UI(n) ,
and
exp−1 (UI(n) ) = ∐ ṼI(m) ≅ UI(n) × Z ,
m≅n mod 4

so exp ∶ C → C∗ is a regular covering map.

However, not all covering maps are regular. Easy examples of irregular
covering maps can be constructed by restricting the size of the domain. For
instance, the inclusion D ↪ C is a covering map, but not regular, since the
preimage of a small disc D(1, ) is a proper subset of that disc. For a more
interesting example see, for instance, question 7 of Example Sheet 3.
Example 3.5. The maps π ∶ Rk → C∗ constructed in Example 2.9 and associ-
ated to kth roots are also regular covering maps, just as in Example 3.3.

3.2 Abstract Riemann surfaces

Intuitively, a Riemann surface is a space in which, at every point, there are
systems of coordinates that look like the complex plane C. The notion of
systems of coordinates is made precise in the next definition. Throughout
this section, R is a connected, Hausdorff, topological space.
Definition 3.6. A chart on R is a pair (φ, U ), where U is an open subset
of R and φ ∶ U → D is a homeomorphism to an open subset of C. A set of
charts A is called an atlas on R if the following two conditions hold:
(i) ⋃(φ,U )∈A U = R;

(ii) if (φ1 , U1 ), (φ2 , U2 ) ∈ A and U1 ∩ U2 ≠ ∅, then

φ1 ○ φ−1
2 ≡ (φ1 ∣U1 ∩U2 ) ○ (φ2 ∣U1 ∩U2 )
−1

is analytic on φ2 (U1 ∩ U2 ).
The composition φ1 ○ φ−1
2 is called a transition function.

17
 The reader may like to compare this with the definition of an abstract
surface from IB Geometry. The main difference is the requirement that the
transition functions should be analytic. This extra hypothesis will enable
us to use notions from the complex plane when we work in the coordinates
provided by charts in A.
Remark 3.7. A few remarks are in order.

(i) Transition functions are always invertible, since (φ1 ○ φ−1

2 )
−1 = φ ○ φ−1 .
2 1

(ii) The space R is connected and locally path-connected, hence path-

connected.

Since Riemann surface are locally modelled on C, it should come as no

surprise that C is a Riemann surface.
Example 3.8. Taking
A = {id ∶ C → C}
defines an atlas on C. Many others are possible; for instance, A′ = {z ↦ z +1}
is an atlas, and so is the union A ∪ A′ .
Example 3.8 illustrates a problem with the definition of an atlas: the two
different atlases A and A′ carry the same information, because they are both
contained in a common larger atlas, A ∪ A′ . Thus, we should only consider
atlases that are as large as possible.

Definition 3.9. A conformal structure on R is an atlas A on R which is

maximal in the following sense: if (ψ, V ) is a chart on R such that, for any
(φ, U ) ∈ A, the transition function φ ○ ψ −1 is analytic, then (ψ, V ) ∈ A.

We now have all the concepts we need to define Riemann surfaces.

Definition 3.10. A Riemann surface is a pair (R, A), where A is a conformal

structure on R. By abuse of notation, the notation R will usually denote the
Riemann surface (R, A).

This definition presents a technical problem: because conformal struc-

tures are necessarily huge, it is rarely practical to exhibit an entire conformal
structure. Fortunately, by the next result, it suffices to exhibit any atlas.

Lemma 3.11. Every atlas A is contained in a unique conformal structure

̂
A.

18
Proof. Let  be the set of all charts (ψ, V ) on R such that ψ ○ φ−1 is analytic
for every (φ, U ) ∈ A. This is necessarily maximal so, to prove existence,
it remains to show that it is an atlas. Take (ψ1 , V1 ), (ψ2 , V2 ) ∈ A, ̂ and an
arbitrary point p ∈ V1 ∩ V2 . Since A is an atlas, there is (φ, U ) ∈ A with p ∈ U .
Therefore,
ψ1 ○ ψ2−1 = (ψ1 ○ φ−1 ) ○ (φ ○ ψ2−1 )
is analytic at ψ2 (p), since it is a composition of analytic maps. Since ψ2 (p)
is an arbitrary point in the domain of ψ1 ○ ψ2−1 , it follows that ψ1 ○ ψ2−1 is
analytic, so  is an atlas, and hence a conformal structure.
Uniqueness is clear: if A′ is any atlas containing A then, by the definition
of an atlas, every chart (φ′ , U ′ ) in A′ has analytic transition function with
every (φ, U ) ∈ A, so A′ ⊆ A.̂ In particular, Â contains any other conformal
structure that contains A, so is unique.
Example 3.12. The atlas
A = {id ∶ C → C}
from Example 3.9 is contained in a unique conformal structure  on C. Thus,
C is a Riemann surface.
Beware! This is not the only Riemann surface structure on C.
Example 3.13. The atlas
A = {z ↦ z̄}
on C is not contained in the conformal structure A, ̂ so defines a different
conformal structure. The Riemann surface obtained by equipping C with
this conformal structure is denoted by C.
̂ is called the canonical
Although other conformal structures on C exist, A
conformal structure on C, and is the one we will always use.
Domains in C provide many more examples.
Example 3.14. If S is any open subset of a Riemann surface R, then re-
stricting the charts of the conformal structure on R to S defines a conformal
structure on S. In particular, every domain in C is a Riemann surface.
Particular favourites include C∗ , the punctured plane, and D, the unit disc.
However, the real power of the definition stems from being able to equip
larger spaces with conformal structures. The next example is one of the most
famous Riemann surfaces, and you have probably encountered it before.

19
Example 3.15 (Riemann sphere). Let C∞ denote the 2-sphere S 2 , identified
with C ∪ {∞} via stereographic projection. We will define an atlas with two
charts. The first chart is (φ, U ), where U = C and φ ∶ U → C is the identity.
The second is (ψ, V ), where V = C∞ ∖ {0} and ψ(z) = 1/z (taking, of course,
ψ(∞) = 0). Setting
A = {(φ, U ), (ψ, V )} ,
both transition functions are z ↦ 1/z on C∗ , which is analytic. Therefore, A
is an atlas and defines a conformal structure on C∞ . This Riemann surface
is called the Riemann sphere.
Remark 3.16. The very alert reader may have noticed that, in contrast to
the definition of abstract surfaces in IB Geometry, Riemann surfaces are not
assumed to be second countable. In fact, a remarkable theorem of Radó
asserts that every Riemann surface is second countable!

3.3 Analytic maps

As mentioned in the previous section, the fact that the transition functions
in an atlas are analytic makes it possible to import notions from the com-
plex plane to R. Most importantly, the notion of an analytic map between
Riemann surfaces makes sense.
Definition 3.17. Let R and S be Riemann surfaces. A continuous map
f ∶R→S
is analytic or holomorphic if, for all charts (φ, U ) on R and (ψ, V ) on S, the
map ψ ○ f ○ φ−1 is analytic on φ(U ∩ f −1 V ).
Because transition functions are analytic, it follows that this definition
can be checked at each point, just like the usual notion of an analytic map
on the complex plane.
Lemma 3.18. A continuous map of Riemann surfaces f ∶ R → S is analytic
if and only if the following holds: for each p ∈ R, there is a chart (φp , Up ) on
R with p ∈ Up and a chart (ψp , Vp ) on S with f (p) ∈ Vp such that the map of
open subsets of C
ψp ○ f ○ φ−1 −1
p ∶ φp (Up ∩ f Vp ) → ψp (Vp )

is analytic at φp (p).

20
Proof. The ‘only if’ direction is immediate. For the ‘if’ direction, given charts
(φ, U ) on R and (ψ, V ) on S, it suffices to show that ψ ○ f ○ φ−1 is analytic
at an arbitrary point φ(p) of its domain φ(U ∩ f −1 V ). The given hypothesis
provides charts (φp , Up ) on R and (ψp , Vp ) on S such that
ψp ○ f ○ φ−1 −1
p ∶ φp (U ∩ f V ) → ψp (V )

is analytic at φp (p). Hence

ψ ○ f ○ φ−1 = (ψ ○ ψp−1 ) ○ (ψp ○ f ○ φ−1 −1
p ) ○ (φp ○ φ )

is analytic at φ(p), since the transition functions ψ ○ ψp−1 and φp ○ φ−1 are
analytic.
To make use of the definition of analytic maps, we need to check that
various properties of analytic maps on C carry over to the setting of Riemann
surfaces. The proofs of these are usually easy: just work in local coordinates
provided by charts, and then appeal to standard results about the complex
plane. Lemma 3.18 makes it particularly easy.
Lemma 3.19. If f ∶ R → S and g ∶ S → T are analytic maps of Riemann
surfaces, then g ○ f ∶ R → T is analytic.
Proof. Let p ∈ R. Since f is analytic, there are charts (φp , Up ) about p and
(ψp , Vp ) about f (p) such that ψp ○f ○φp is analytic at φp (p). Likewise, since g
is analytic, there are charts (ψf (p) , Vf (p) ) about f (p) and (θf (p) , Wf (p) ) about
g ○ f (p) such that θf (p) ○ g ○ ψf−1(p) is analytic at ψf (p) (f (p)). Therefore,

θf (p) ○ (g ○ f ) ○ φ−1 −1 −1 −1
p = (θf (p) ○ g ○ ψf (p) ) ○ (ψf (p) ○ ψp ) ○ (ψp ○ f ○ φp )

is a composition of analytic maps and hence analytic at φp (p). Therefore,

g ○ f is analytic by Lemma 3.18.
In pure mathematics, we are always interested in finding the right notion
of equivalence for the objects we study.
Definition 3.20. A conformal equivalence or biholomorphism is an analytic
bijection of Riemann surfaces f ∶ R → S with an analytic inverse f −1 ∶ S → R.
By Lemma 3.19, conformal equivalence is an equivalence relation.
Example 3.21. Complex conjugation z ↦ z̄ defines analytic bijections in both
directions between the Riemann surfaces C and C. Thus, these two Riemann
surfaces are conformally equivalent.

21
Lecture 4: Examples of conformal structures
and analytic functions
To apply this theory to the object R we associated to the complex logarithm
in §2.2, we need to find a way to put a conformal structure on R. Fortunately,
the covering maps of §3.1 are a big help.

4.1 Covering maps and analyticity

Using covering maps, we can ‘pull back’ a conformal structure from the range
to the domain.

Lemma 4.1. If π ∶ R ̃ → R is a covering map and R is a Riemann surface

̃ such that π is analytic.
then there is a unique conformal structure on R

Proof. Each point p ∈ R̃ has an open neighbourhood N ̃p such that the restric-
̃p is a homeomorphism onto its image. Because R is a Riemann
tion of π to N
surface, there is a chart (φp , Up ) about π(p). Setting φ̃p = φp ○ π (suitably
restricted) and Ũp = N
̃p ∩ π −1 (Up ) defines a chart (φ̃p , U
̃p ) about p.
We will show that
̃p ) ∣ p ∈ R}
à = {(φ̃p , U ̃
̃ The sets U
is an atlas on R. ̃p cover R ̃ by construction, so it suffices to check
that the transition functions are analytic. Given two points p, q ∈ R, ̃ the map
π is invertible when restricted to the intersection U ̃p ∩ U ̃q by construction.
Therefore, the transition function of the corresponding charts is

φ̃p ○ φ̃−1
q = φp ○ π ○ π
−1
○ φ−1 −1
q = φp ○ φq

which is also a transition function of R, and so is analytic. Therefore à is

indeed an atlas, and defines a conformal structure.
̃ we may take the charts (φ̃p , U
At an arbitrary point p ∈ R, ̃p ) about p and
(φp , Up ) about π(p), and then

φp ○ π ○ φ̃−1
p = φp ○ π ○ π
−1
○ φ−1
p = id

which is certainly analytic at p, so this conformal structure does indeed make

π into an analytic map.

22
 For uniqueness, suppose that B̃ is some conformal structure that makes π
̃ be arbitrary, and let (ψ, V ) ∈ B̃ be any chart
into an analytic map, let p ∈ R
about p. Since π is analytic, the composition
φ̃p ○ ψ −1 = φp ○ π ○ ψ −1
̃
is analytic, so φ̃p has analytic transition function with every chart in B.
̃ ̃
Hence A = B by Lemma 3.11.
This makes it easy to put a conformal structure on the space we associated
to the complex logarithm in §2.2.
Example 4.2. Let R be as in §2.2. In Example 3.3 we saw that natural
map π ∶ R → C∗ is a covering map. Hence, by Lemma 4.1, there is a unique
conformal structure on R that makes π analytic; in particular, R is a Riemann
surface.
In fact, more is true. Recall that R and π fitted into a commutative
diagram.
f
R C
π exp

C∗
On any given domain UI(n) ⊆ R, in the standard charts, f can be written as
fI(n) ∶ UI(n) → C
which is analytic, so this conformal structure on R also makes f into an
analytic map.
Furthermore, recall from Example 3.4 that fI(n) ∶ UI(n) → ṼI(n) has an
−1
analytic inverse fI(n) = exp ∣ṼI(n) , for each n ∈ Z. These inverses agree on the
intersections ṼI(m) ∩ ṼI(n) , and so piece together to give a globally defined
analytic inverse f −1 ∶ C → R. In particular, f is a conformal equivalence.
A similar analysis applies to kth roots.
Example 4.3. Let Rk be as in Example 2.9, which fits into a commutative
diagram
g
Rk C∗
π pk

C∗

23
where pk is the power map z ↦ z k . In Example 3.5, we saw that π is a
covering map so, by Lemma 4.1, there is a unique conformal structure on Rk
that makes π analytic. As in Example 4.2, g takes the form
gI(n) ∶ UI(n) → C∗
on any open set UI(n) ⊆ Rk , so g is also analytic. Again, the functions gI(n)
−1
have analytic inverses gI(n) (defined by suitable restrictions of the power
maps), which piece together to define a global inverse g −1 ∶ C∗ → Rk , so g is
a conformal equivalance.
Unlike the complex logarithm, the kth root has an additional nice prop-
erty. The space Rk can be embedded in a compact Riemann surface R ̂k , and
g and π extended, to obtain a commutative diagram of compact Riemann
surfaces.
̂k
R
ĝ
C∞
π̂ pk

C∞
Compactifying loses the nice property that π is a covering map, since pk fails
to be a local homeomorphism. However, it is a huge advantage to work with
compact spaces, so this sacrifice is usually worth making. We will learn a lot
more about compact Riemann surfaces later in the course.

4.2 Analytic functions

An important way to study a Riemann surface R is via analytic maps from R
to well-understood Riemann surfaces. Since C is the prototypical Riemann
surface, maps to C are especially important.
Definition 4.4. An analytic function on a Riemann surface R is an analytic
map R → C.
To check that a function f ∶ R → C is analytic, we only need to find a
chart (φ, U ) about each point p ∈ R such that f ○ φ is analytic. Recall the
statement of the inverse function theorem from IB Complex Analysis.
Theorem 4.5 (Inverse function theorem). Let f be an analytic function on
a domain D ⊆ C. If f ′ (z0 ) ≠ 0 for z0 ∈ D, then there are open neighbourhoods
U of z0 and V of f (z0 ) such that f restricts to a biholomorphism U → V .

24
 One advantage of working with Riemann surfaces is that, in a suitable
chart, any analytic function can be put into a very simple local form. The
proof of this uses the inverse function theorem.
Proposition 4.6. Let f be a non-constant analytic function on a Riemann
surface R and let p ∈ R be a zero of f . There is a chart (φ, U ) about p with
φ(p) = 0 such that
f ○ φ−1 (z) = z m
for some integer m > 0.
Proof. Choose a chart (ψ, V ) about p; without loss of generality, ψ(p) = 0.
As in the proof of the principle of isolated zeroes, there is an analytic function
g with g(0) ≠ 0 such that
f ○ ψ −1 (z) = z m g(z)
on a neighbourhood of 0, for some natural number m. By the identity prin-
ciple for Riemann surfaces (question 10 of Example Sheet 1), m > 0 since f
is non-constant.
Since g(0) ≠ 0 and is continuous, there is δ > 0 such that g(D(0, δ)) ⊆
D(g(0), ∣g(0)∣). Since there is an analytic mth root defined on D(g(0), ∣g(0)∣)
(for instance, one could use one√of the function elements from Example 2.9),
there is an
√ analytic function m g(z) defined in a neighbourhood of 0. If
h(z) = z m g(z), then f ○ ψ −1 (z) = (h(z))m . Furthermore,
√
h′ (0) = m g(0) ≠ 0 ,
so h has an analytic inverse defined on some D(0, ), by the inverse function
theorem. Setting φ = h ○ ψ and U = φ−1 (D(0, )) gives the required chart,
because
f ○ φ−1 (z) = f ○ ψ −1 ○ h−1 (z) = (h ○ h−1 (z))m = z m
as required.

Lecture 5: Complex tori and the open map-

ping theorem
5.1 Complex tori
So far, the Riemann sphere is our only example of a compact Riemann sur-
face. In fact, the collection of Riemann surfaces is rich and diverse. In this

25
section, we construct a much more interesting family of examples.
Example 5.1 (Complex tori). Let τ1 , τ2 be complex numbers that are linearly
independent over R – that is, τi ∈ C∗ and τ2 /τ1 ∉ R. Let Λ = ⟨τ1 , τ2 ⟩, the
additive subgroup generated by them, let T be the quotient group C/Λ, and
let π ∶ C → T be the quotient map.
As a topological space, T is equipped with the quotient topology. This
can be studied via the fundamental parallelogram P ⊆ C, the parallelogram
with corners {0, τ1 , τ2 , τ1 + τ2 }: T is obtained by identifying opposite sides of
P . From this description, it is not hard to prove that T is homeomorphic to
the torus S 1 × S 1 ; in particular, T is compact and Hausdorff.
The map π is a regular covering map. Indeed, if
 < min{∣λ∣ ∣ λ ∈ Λ ∖ {0}}/2
then, for any z ∈ C, D(z, ) ∩ {z + Λ} = {z}. If p = π(z0 ) ∈ T , it follows that
U = π(D(z0 , )) is open with preimage
π −1 (U ) = ∐ D(z0 , ) + λ ≅ D(z0 , ) × Λ
λ∈Λ

as required.
Finally, for T to be a Riemann surface, we need to endow it with an
atlas. For any z0 ∈ C, let U = π(D(z0 , )), an open set in T as above. Then π
restricts to a homeomorphism D(z0 , ) → U , and so we may set φ = π∣−1 D(z0 ,)
to define a chart (φ, U ) on T . Since z0 was arbitrary, the set of all such charts
covers T . Furthermore, if (ψ, V ) is another such chart with ψ = π∣−1 D(z1 ,)
, then
the transition function φ ○ ψ is translation by some element of Λ. Since
−1

translation is analytic, this defines a conformal structure on T .

Remark 5.2. The construction of Example 5.1 is the reverse of the construc-
tion of Lemma 4.1. Rather than ‘pulling back’ the complex structure along
the covering map, the complex structure of C is ‘pushed forward’ along π to
define a complex structure on T . In general, it is more difficult to make this
sort of ‘push forward’ construction work than the ‘pull back’ construction of
Lemma 4.1. In this case, the construction works because T is a quotient of
C by a group (in this case, Λ), acting by analytic maps on C.
Topologically, these complex tori are all the same: they are all homeo-
morphic to S 1 × S 1 . However, in question 5 of Example Sheet 2, we shall see
that different lattices Λ often give rise to complex tori that are not confor-
mally equivalent. In particular, they will provide a source of infinitely many
different conformal equivalence classes of compact Riemann surfaces.

26
5.2 The open mapping theorem
As mentioned in §4.2, we often want to study Riemann surfaces via their
analytic functions. However, we shall see in this subsection that the analytic
functions on compact Riemann surfaces are not very interesting. This is an
easy consequence of the open mapping theorem for Riemann surfaces which,
like many other theorems about analytic functions, follows easily from the
open mapping theorem on the complex plane from IB Complex Analysis.

Theorem 5.3 (Open mapping theorem for Riemann surfaces). Any non-
constant, analytic map of Riemann surfaces f ∶ R → S is an open map.

Proof. Consider an open set W ⊆ R, and let p ∈ R be arbitrary. Take charts

(φ, U ) about p and (ψ, V ) about f (p). By the identity principle for Riemann
surfaces (question 10 of Example Sheet 1), ψ○f ○φ−1 is a non-constant function
from φ(U ∩ W ∩ f −1 V ) to ψ(V ), so ψ ○ f (U ∩ W ∩ f −1 V ) is open, by the
open mapping theorem for domains in C. Because ψ is a homeomorphism,
f (U ∩ W ∩ f −1 V ) is an open neighbourhood of f (p) in f (W ), and since p
was arbitrary, it follows that f (W ) is open, as required.
This has especially profound consequences for analytic maps of compact
Riemann surfaces, such as C∞ or the complex tori of Example 5.1.

Corollary 5.4. Let f ∶ R → S be a non-constant, analytic map of Riemann

surfaces. If R is compact, then f is surjective, and in particular, S is also
compact.

Proof. By the open mapping theorem, f (R) is open. Since R is compact,

f (R) is also compact and so closed. The result follows since S is connected.

This immediately rules out any non-constant analytic function on a com-

pact Riemann surface, since C is not compact.

Corollary 5.5. Every analytic function on a compact Riemann surface is

constant.

5.3 Harmonic functions

This subsection is a brief digression into real functions on Riemann surfaces.
Of course, it doesn’t make sense to ask for a real-valued function to be

27
complex-differentiable. Since we need to work with real coordinates, we
write z = x + iy for x, y ∈ R.

Definition 5.6. Let D ⊆ C be a domain. A smooth function u ∶ D → R is

called harmonic if
∂ 2u ∂ 2u
∆u ∶= 2 + 2 ≡ 0
∂x ∂y
on D. (Recall that ∆u is the Laplacian of u.)

The next lemma explains the connection between analytic and harmonic
functions.

Lemma 5.7. Consider a disc D ⊆ C. A function u ∶ D → C is harmonic if

and only if
u = Re(f )
for some analytic function f .

Proof. If f = u + iv is analytic, then it satisfies the Cauchy–Riemann equa-

tions:
∂u ∂v ∂u ∂v
= and =− .
∂x ∂y ∂y ∂x
Therefore,
∂ 2u ∂ 2u ∂ 2v ∂ 2v
+ = − =0
∂x2 ∂y 2 ∂x∂y ∂y∂x
so u is harmonic.
For the other direction, see question 11 of Example Sheet 1.
With this lemma in hand, we can make sense of harmonic functions on
Riemann surfaces.

Definition 5.8. Let R be a Riemann surface. A function u ∶ R → R is

harmonic if, for any chart (φ, U ) on R, the composition

u ○ φ−1 ∶ U → R

is harmonic.

28
Remark 5.9. Lemma 5.7 implies that Definition 5.8 can be checked on any
atlas. If u ○ φ−1 is harmonic then, by the lemma,

u ○ φ−1 = R(f ) .

For any overlapping chart (ψ, V ), we therefore have

u ○ ψ −1 = Re(f ○ φ ○ ψ −1 ) ,

which is the real part of a composition of analytic functions, and so harmonic.

Lemma 5.7 makes it quite easy to extend results about analytic functions
on Riemann surfaces to harmonic functions. For instance, they satisfy an
identity principle.

Proposition 5.10 (Identity principle for harmonic functions). Let u and v

be harmonic functions on a Riemann surface R. Either u ≡ v on R, or the
set where they coincide
{p ∈ R ∣ u(p) = v(p)}
has empty interior.

Proof. See Example Sheet 1, question 12.

Moreover, harmonic functions also satisfy an open mapping theorem.

Theorem 5.11 (Open mapping theorem for harmonic functions). Any non-
constant harmonic function u on a Riemann surface R is an open map.

Proof. Suppose W ⊆ R is open. Let p ∈ W , and consider a chart (φ, U )

about p. By Lemma 5.7, after shrinking U if necessary, there is an analytic
function f ∶ φ(U ) → C such that u ○ φ−1 = Re(f ). If f is constant on φ(U )
then u is constant on U , and hence on R by Proposition 5.10. Therefore f
is non-constant and so, by the open mapping theorem, f ○ φ(U ) is open.
Let f ○φ(p) = a+ib. Since the topology on C is identified with the product
topology on R2 , f ○ φ(U ) contains an open set of the form

(a − δ, a + δ) + i(b − , b + ) .

But a = u(p), so u(W ) contains (u(p)−, u(p)+). Since p ∈ W was arbitrary,

u(W ) is open, as required.

29
 Just as in the analytic case, it follows that all harmonic functions on
compact Riemann surfaces are constant.

Corollary 5.12. If R is a compact Riemann surface, all harmonic functions

on R are constant.

Harmonic functions don’t play much of a role in the rest of the course,
but the results of this section do at least indicate a connection with Laplace’s
equation. This connection can be exploited in both directions. Corollary 5.12
shows that Laplace’s equation doesn’t have interesting solutions on compact
Riemann surfaces. In the other direction, Lemma 5.7 indicates that we could
aim to construct analytic functions by solving Laplace’s equation.

Lecture 6: Meromorphic functions and a worked

example
6.1 Meromorphic functions
Let R be a compact Riemann surface. Corollary 5.5 shows us that the an-
alytic functions on R will not yield much information. However, Corollary
5.4 also points to a solution: the problem is that C is non-compact, and so
we can obtain a more interesting theory by compactifying C. This motivates
the next definition.

Definition 6.1. A meromorphic function on a Riemann surface R is an

analytic map f ∶ R → C∞ , where C∞ is the Riemann sphere, which is not
identically ∞.

This is an elegant definition, but to avoid confusion we need to check that

it coincides with the definition of a meromorphic function on the complex
plane that we already use!

Proposition 6.2. Let D ⊆ C be a domain. A function f ∶ D → C is mero-

morphic if and only if there is a discrete subset A ⊆ D such that f ∶ D∖A → C
is analytic, and f has a pole at each a ∈ A.

Proof. Let’s start with the ‘only if’ direction. Let A = f −1 (∞). By the
identity principle for Riemann surfaces A is discrete, since otherwise f would
be constant and equal to ∞, contrary to hypothesis. It remains to prove

30
that each a ∈ A is a pole. Considering the standard chart on C∞ about ∞,
any a ∈ A has a neighbourhood on which 1/f (z) is analytic and so, as in the
proof of the identity principle, takes the form
1/f (z) = (z − a)m g(z)
for some m ≥ 1 and some analytic g with g(a) ≠ 0. Therefore, on some
possibly smaller neighbourhood of a,
f (z) = (z − a)−m h(z)
for h(z) = 1/g(z) analytic. Thus a is a pole of f , as required.
For the ‘if’ direction, suppose that f has a pole of order m at a ∈ A,
meaning that, in a neighbourhood of a,
f (z) = (z − a)−m h(z)
for some m ≥ 1 and some analytic function h with h(a) ≠ 0. Passing to a
smaller neighbourhood on which h is never zero,
1/f (z) = (z − a)m g(z)
for g(z) = 1/h(z) analytic. Therefore, f extends at each a ∈ A to an analytic
map to C∞ .

6.2 A worked example

Our next topic will give a theoretical proof that analytically continuing any
function element defines a Riemann surface. However, before seeing the
theoretical construction, we provide a detailed worked example.
We have already seen this in action in a few cases: Example 1.13 asso-
ciated a Riemann surface to the complex logarithm, while Example 2.9 did
the same for kth roots. In this subsection, we will look at a function element
for √
w = z3 − z .
There are two approaches to constructing this Riemann surface. One is to
put a conformal structure on the graph
{(w, z) ∈ C2 ∣ w2 = z 3 − z} .
However, here we will construct this Riemann surface by gluing together
function elements.

31
Example 6.3. Consider f (z) = z 3 − z = z(z − 1)(z + 1). Since the complex
√ 0 and ∞, we are most concerned
square root fails to be locally defined at
with the points z = 0, 1, −1, ∞, where f (z) is not locally defined. This
motivates us to define a domain D by joining these points with branch cuts.

D = C ∖ ([−1, 0] ∪ [0, ∞))

Our goal is to construct analytic functions g± (z) on D that satisfy g± (z)2 =

f (z). Pick any base point z0 ∈ D, and let g± (z0 ) be the two values such that
g± (z0 )2 = f (z0 ). For any z ∈ D, choose a path γ in U from z0 to z. The
function g± (z) is then defined by a path integral.

1 f ′ (ζ)
g± (z) ∶= g± (z0 ) exp ( ∫ dζ)
2 γ f (ζ)

We claim that this definition is independent of the choice of path. To prove

this, it suffices to show that

1 f ′ (ζ)
exp ∮ dζ = 1
2 γ f (ζ)

for any closed curve γ. By the argument principle from IB Complex analysis,

f ′ (ζ) ⎛ ⎞
∮γ f (ζ) dζ = 2πi ∑ n(γ, Z) − ∑ n(γ, P )
⎝zeroes of f poles of f ⎠
= 2πi(n(γ, 1) + n(γ, 0) + n(γ, −1)) ,

where n(γ, z) denotes winding numbers. In question 1 of Example Sheet 1,

we saw that n(γ, 1) = 0 and n(γ, 0) = n(γ, −1). Therefore,

1 f ′ (ζ)
dζ ∈ 2πiZ
2 ∮γ f (ζ)

and so the claim follows and g± (z) is well-defined.

Next, let’s check that g± is continuous. Indeed, since f ′ /f is continuous
at any z ∈ D, there is δ such that

f ′ (z) f ′ (τ )
∣ − ∣≤1
f (z) f (τ )

32
for any τ ∈ D(z, δ). Therefore,
τ f ′ (ζ) z f ′ (ζ) τ f ′ (ζ)
∣∫ dζ − ∫ dζ∣ = ∣∫ dζ∣ ≤ M ∣z − τ ∣ → 0
z0 f (ζ) z0 f (ζ) z f (ζ)
(ζ) z ′
as τ → z, where M = ∣f ′ (z)/f (z)∣ + 1. This shows that ∫z0 ff (ζ) dζ is a contin-
uous function of z, and so g± is√ also continuous.
Given continuity, g± (z) = f (z) locally, for a suitable choice of square
root, and so g± is analytic.
This construction gives two function elements, denoted by

(g+ , D+ ) and (g− , D− )

where D± = D. The Riemann surface will be defined by gluing these two

function elements together. Before we do this, let’s discuss the topology of
D.
The domain D is obtained by deleting two closed intervals from the Rie-
mann sphere C∞ , which lie on the equator R ∪ {∞}. From this, it follows
that D is topologically an open annulus S 1 × R.
For any z0 in the intervals (−1, 0) ∪ (0, ∞), let z → z0+ denote the ap-
proaching z0 from the upper half-plane, and let z → z0− denote approaching
z0 from the lower half-plane. It is not hard to see that

lim g+ (z0 ) = lim+ g− (z0 )

z→z0− z→z0

and
lim g+ (z0 ) = lim− g− (z0 ) ,
z→z0+ z→z0

for any z0 in the interiors of the branch cuts. Thus, we can glue the two
function elements together along the open intervals (−1, 0) ∪ (0, ∞) to obtain
a Riemann surface R and a well-defined function

g ∶ R → C.

Since R is obtained by gluing two annuli along four open intervals, we can see
that R is homeomorphic to the torus S 1 × S 1 with four points deleted. The
Riemann surface R is also equipped with another function π ∶ R → C, defined
by the inclusions of D± into C. By construction two functions together satisfy
the equation
g(p)2 = f ○ π(p)

33
at every p ∈ R. It is in this sense that g solves the equation w2 = f (z).
To be fully rigorous about this gluing, we need to describe the complex
structure at the glued points z0 ∈ (−1, 0) ∪ (0, ∞). This can be done by
considering function elements on small discs D(z0 , ). Alternatively, one can
make the opposite choice of branch cuts

E = C ∖ ((−∞, −1] ∪ [0, 1])

define function elements (h± , E± ) similarly to above, and then construct R by

gluing together D± and E± wherever they overlap and the function elements
agree. However, we will usually suppress this detail.

Lecture 7: Covering-space theory

This section introduces some tools from topology that will be have many
applications in this course, especially when we come to give a complete treat-
ment of analytic continuation.

7.1 Covering-space theory

Lifts are the fundamental objects of study here.
Definition 7.1. Suppose π ∶ X ̃ → X is a covering map and γ ∶ [0, 1] → X is
̃ such that π ○ γ̃ = γ.
a path. A lift of γ along π is a path γ̃ ∶ [0, 1] → X
Example 7.2. In Example 3.4 we saw that exp ∶ C → C∗ is a covering map.
Consider the anticlockwise loop around the unit circle

γ(t) = e2πit .

Then
γ̃1 (t) = 2πit
and
γ̃2 (t) = 2πi(1 + t)
are both lifts of γ.
As this example shows, lifts are not unique. However, the two different
lifts given start in different places, and it turns out that this extra information
– the start point – is all that’s needed to determine a lift.

34
Proposition 7.3 (Uniqueness of lifts). Suppose γ̃1 , γ̃2 are both lifts of γ along
a covering map π. If γ̃1 (0) = γ̃2 (0) then γ̃1 ≡ γ̃2 .
Proof. The result follows from the claim that the subset I = {t ∈ I ∣ γ̃1 (t) =
γ̃2 (t)} is both open and closed in [0, 1]. Since I is hypothesised to contain
0, it follows that I = [0, 1] by connectedness.
The fact that I is closed follows from very general facts. Because X is
Hausdorff, the diagonal X is a closed subset of the product X × X, endowed
with the product topology. Since the map
γ1 × γ2 ∶ I → X × X
sending t ↦ (γ1 (t), γ2 (t)) is continuous, it follows that I, the preimage of the
diagonal, is also closed.
To prove that I is open, consider any t ∈ I. Since π is a covering, γ̃1 (t) =
γ̃2 (t) has an open neighbourhood U ̃ such that π∣ ̃ is a a homeomorphism
U
onto its image U . By continuity, there is δ > 0 such that γ̃1 (t − δ, t + δ) and
γ̃2 (t − δ, t + δ) are both contained in U ̃ . For all t − δ < s < t + δ, we therefore
have
γ̃1 (s) = π∣−1
̃ ○ γ(s) = γ̃2 (s) ,
U
so s ∈ I as required.
Mathematicians like to complement uniqueness results with existence re-
sults. However, for arbitrary coverings, lifts need not exist, even if the cov-
ering map is surjective.
Example 7.4. Let X ̃ = {z ∈ C ∣ −π < Im(z) < 2π}, and consider the surjective
covering map
π∶X ̃ → C∗
which is the restriction of the exponential map. Let γ(t) = e2πit , as in Exam-
ple 7.2. Since π −1 (1) = {0}, any lift γ̃ of γ must start at 0. But such a lift
γ̃ is also a lift of γ along the exponential map so, by uniqueness of lifts, it
must coincide with γ̃1 from Example 7.2. However, γ̃1 (1) = 2πi ∉ X, ̃ so this
is impossible. Hence, there is no lift of γ along π.
This is where regular coverings come into their own. The proof of the
next lemma follows a similar strategy to the proof of uniqueness of lifts.
Proposition 7.5 (Path-lifting lemma). Let π ∶ X ̃ → X be a regular covering
map. Let γ ∶ [0, 1] → X be a path, and suppose that π(x̃) = γ(0). Then there
is a (unique) lift γ̃ of γ such that γ̃(0) = x̃.

35
 ̃
Proof. Let I be the set of t ∈ [0, 1] such that there exists a lift γ̃ ∶ [0, t] → X
of γ∣[0,t] along π with γ̃(0) = x̃. Again, the result follows from the claim that
I is both open and closed, since 0 ∈ I by hypothesis.
To show that I is closed, consider a sequence tn ∈ I converging to some
τ ∈ [0, 1]. Since π is a regular covering map and X is locally path-connected,
γ(τ ) has a path-connected open neighbourhood U such that

π −1 (U ) = ∐ Uδ ≅ U × ∆
δ∈∆

for some discrete set ∆. Let N be large enough that γ(tn ) ∈ U for all n ≥ N .
It follows that γ̃(tn ) are in the same path component Uδ of π −1 (U ) for all
n ≥ N . Therefore, setting

γ̃(τ ) = (π∣Uδ )−1 ○ γ(τ )

extends γ̃ to a continuous lift at τ , so τ ∈ I as required.

To show that I is open, consider τ ∈ I, and again, let U be a path-
connected open neighbourhood of γ(τ ) such that

π −1 (U ) = ∐ Uδ .
δ∈∆

There is a unique δ such that γ̃(τ ) ∈ Uδ . Let  > 0 be such that γ(t) ∈ U
whenever ∣t − τ ∣ < . By uniqueness of lifts,

γ̃(t) = (π∣Uδ )−1 ○ γ(t)

for every t ∈ (τ − , τ + ) ∩ I. This shows that γ̃ can be extended to an open

neighbourhood of τ , as required.
Rather than just being interested in individual curves, we often want to
deform one into another.
Definition 7.6. Let X be a a topological space and α, β ∶ [0, 1] → X paths
with α(0) = β(0) and α(1) = β(1). This pair of paths is homotopic (conve-
niently written α ≃ β) if there exists a family of paths (αs )s∈[0,1] such that:
(i) α0 ≡ α and α1 ≡ β;

(ii) αs (0) = α(0) and αs (1) = α(1) for all s;

(iii) the map (t, s) ↦ αs (t) is a continuous map [0, 1]2 → X.

36
 In particular, this enables us to make rigorous sense of the notion of a
‘space with no holes’.

Definition 7.7. A topological space X is simply connected if:

(i) X is path-connected; and

(ii) every pair of paths α, β ∶ [0, 1] → X with the same endpoints (i.e.
α(0) = β(0) and α(1) = β(1)) is homotopic.

Remark 7.8. Let D ⊆ C be a convex domain. The formula

αs (t) = (1 − s)α(t) + sβ(t)

defines a homotopy between any two paths α, β in D with equal endpoints,

so D is simply connected.
The next result is an important statement about the existence of homo-
topies. In due course, it will enable us to prove uniqueness results about
analytic continuation.
̃ → X be a covering map and
Theorem 7.9 (Monodromy theorem). Let π ∶ X
let α, β be paths in X. Suppose that:

(i) α ≃ β in X;

(ii) there are lifts α̃ of α and β̃ of β with α̃(0) = β̃(0);

̃ with γ(0)
(iii) every path γ in X with γ(0) = α(0) = β(0) has a lift to X ˜ =
α̃(0) = β̃(0).

Then α̃ ≃ β̃; in particular, α̃(1) = β̃(1).

This result is not proved here. In II Algebraic topology, an essentially

identical result is called the homotopy lifting lemma, and a proof is given.
Remark 7.10. Note that hypotheses (ii) and (iii) of Theorem 7.9 are auto-
matically satisfied if π is a regular covering map.

37
Lecture 8: The monodromy group and the space
of germs
8.1 The monodromy group
In this section, we shall see that the monodromy theorem can be used to
define an invariant of regular covering maps – the monodromy group. It’s an
important object, corresponding to the Galois group in Galois theory and to
the deck group in algebraic topology. We won’t discuss its properties much
in this course, but you will be asked to compute it on example sheets.
̃ → X be a regular covering map, and let’s pick a base point
Let π ∶ X
x0 ∈ X. Now consider a loop γ ∶ [0, 1] → X based at x0 ; that is, a path with
γ(0) = γ(1) = x0 . Path lifting enables us to associate to γ a self-map σγ of
the preimage π −1 (x0 ).
Definition 8.1. Let x̃ ∈ π −1 (x0 ), and let γ̃x̃ be the unique lift of γ starting
at x̃. Since it is a lift,
π(γ̃x̃ (1)) = γ(1) = x0
so γ̃x̃ (1) ∈ π −1 (x0 ). Therefore, we may define σγ ∶ π −1 (x0 ) → π −1 (x0 ) by

σγ (x̃) ∶= γ̃x̃ (1)

for any x̃ ∈ π −1 (x0 ).

The next remarks collect together a number of nice properties of this
definition.
̃ → X, x0 and γ be as above.
Remark 8.2. Let π ∶ X
(i) For the constant loop ι ∶ t ↦ x0 , the corresponding map σι is the identity.
(ii) Let γ̄ be the loop
γ̄(t) ∶= γ(1 − t) .
Using uniqueness of lifts, it is easy to see that σγ̄ = σγ−1 . In particular,
σγ is always a bijection, i.e. a permutation.
(iii) If α and β are both loops based at x0 , we can define the concatenation
⎧
⎪
⎪ α(2t) 0 ≤ t ≤ 1/2
α ⋅ β(t) ∶= ⎨
⎪
⎩ β(2t − 1) 1/2 ≤ t ≤ 1
⎪

38
 which is also a loop based at x0 . If x̃1 ∈ π −1 (x0 ) and the lift α̃x̃1 ends
at x̃2 then,
(α̃⋅ β)x̃1 = α̃x̃1 ⋅ β̃x̃2
by uniqueness of lifts. In particular,

σα⋅β )(x̃1 ) = σβ (x̃2 ) = σβ ○ σα (x̃1 ) .

Taking all of these properties together, we see that the set of all permuta-
tions σγ of π −1 (x0 ) corresponding to loops based at x0 form a subgroup Hx0
of the symmetric group Sym(π −1 (x0 )). This is the monodromy group of the
regular covering map π. The definition appears to also depend on the choice
of base point, but in fact this dependence is an illusion.
Remark 8.3. Let π ∶ X ̃ → X be as above, with X path connected.

(i) Let x0 and x1 be different choices of base point in X. Concatenation

with α can be used to transform a loop γ based at x1 into a loop base
at x0 , via the map
γ ↦ α ⋅ γ ⋅ ᾱ .
This map defines a homomorphism of monodromy groups

θα ∶ Hx1 → Hx0

and, by uniqueness of lifts again, it is not hard to see that θᾱ = θα−1 . In
particular, the isomorphism class of the monodromy group is indepen-
dent of the choice of base point.
(ii) Computation of the monodromy group seems daunting, since there are
usually uncountably many loops based at a point. However, the mon-
odromy theorem implies that σα = σβ if α ≃ β; that is, σα only de-
pends on the homotopy class of α. This makes computation much
more tractable.
This section finishes with the easiest example.
Example 8.4. The power map pn (z) = z n defines a regular covering map
C∗ → C∗ , for any natural number n. Let γ(t) = e2πit , the standard clockwise
loop around the unit circle in C∗ . Let ζn be a primitive nth root of unity,
and let γ̃k be the unique lift of γ at ζnk . By uniqueness of lifts,

γ̃k (t) = exp (2πi(k + t)/n) ;

39
in particular, γ̃k (1) = ζnk+1 . Therefore, γ naturally defines a permutation
σγ ∈ Sym(n) via
σ (k)
γ̃k (1) = ζn γ .
Any loop in C∗ starting and ending at 1 is homotopic to γ n for some n ∈ Z.
Therefore, the regular covering map pn and the choice of base point 1 defines
a subgroup of Sym(n), namely ⟨σγ ⟩ ≅ Cn .
See question 9 of Example Sheet 2 for an example of a similar computa-
tion.

8.2 The space of germs

The main construction of this section associates a space to any domain –
the space of germs. The path components of the space of germs are Rie-
mann surfaces, which can be thought of as the result of all possible analytic
continuations. The construction makes sense for any domain D ⊆ C.

Definition 8.5. Let (f, U ) and (g, V ) be function elements on D. For any
z ∈ D ∩ E, write
(f, U ) ≡z (g, V )
if f and g agree on a neighbourhood of z.

Note that ≡z is an equivalence relation on the set of function elements

(f, U ) such that z ∈ U .

Definition 8.6 (Germ). Let (f, U ) be a function element and z ∈ U . The

equivalence class of (f, U ) under ≡z is called the germ of f at z, and is
denoted by [f ]z . In summary, two germs [f ]z and [g]w are equal if and only
if z = w and f = g on a neighbourhood of z = w.

Our main task is to give the collection of all possible germs the structure
of a nice topological space – ultimately, a disjoint union of Riemann surfaces.

Definition 8.7 (The space of germs). The space of germs over D is

G ∶= {[f ]z ∣ z ∈ D, (f, U ) a function element with z ∈ U }

as a set.

40
 To endow G with a topology, it is convenient to write
[f ]U ∶= {[f ]z ∣ z ∈ U }
for any function element U . The open sets of the topology are all unions of
all sets of the form [f ]U , for all function elements (f, U ) on D. Let us check
that this is indeed a topology.
Lemma 8.8. Unions of sets of the form [f ]U define a topology on G.
Proof. Taking the empty union shows that ∅ is open. By definition, each
point [f ]z ∈ [f ]U for some U , whence G is also open. The topology is closed
under taking unions by definition, so it only remains to check that it is closed
under finite intersections.
Since the intersection of two unions is the union of the intersections,
to check that the topology is closed under finite intersections, it suffices to
check that a set of the form [f ]U ∩ [g]V is open. Consider any germ [h]z ∈
[f ]U ∩ [g]V . This means that z ∈ U ∩ V and that h agrees with both f and g
on a neighbourhood W of z. Thus [h]W provides an open neighbourhood of
[h]z in the intersection, so the intersection is open claimed.
If the components of G are to be Riemann surfaces, they must in particular
be Hausdorff.
Lemma 8.9. The space of germs G is Hausdorff.
Proof. Consider two distinct germs [f ]z ≠ [g]w , and choose representative
function elements (f, U ) and (g, V ). If z ≠ w then, by shrinking U and V , we
may assume that U and V are disjoint, whence [f ]U ∩ [g]V = ∅, as required.
The case z = w is all that remains, in which case we may take represen-
tative function elements (f, U ) and (g, U ) for U connected. Suppose that
[h]x ∈ [f ]U ∩ [g]U . This implies that x has a neighbourhood W in U on
which f and g both agree with h, and hence with each other. By the iden-
tity principle, f and g agree on U , so [f ]U = [g]U ] whence [f ]z = [g]z , a
contradiction.
To define a conformal structure on each component of G, we will use a
useful extra piece of structure – the natural ‘forgetful’ map to D.
Definition 8.10. Let G be the space of germs over a domain D. The forgetful
map π ∶ G → D is defined by
π([f ]z ) = z .

41
Lemma 8.11. For each component G ⊆ G, the restriction of the forgetful
map
π∶G→D
is a covering map

Proof. Let U ⊆ D be open. The preimage of U under π is

π −1 (U ) = ⋃ [f ]V
V ⊆U

where the union ranges over all function elements on U . In particular, π is

continuous.
The restriction of π to any open set of the form [f ]U has an inverse,
namely z ↦ [f ]z . Furthermore, this inverse is continuous, since the preimage
of an open set [f ]V is the open set V ∩ U . Since the sets [f ]U cover G, π is
indeed a local homeomorphism, and so its restriction to any component is a
covering map.
By Lemma 4.1, each component of G has a unique conformal structure
that makes π into an analytic map. An explicit atlas is easy to write down,
with each chart of the form (π∣U , [f ]U ).
As well as the forgetful map, the space of germs carries another naturally
defined map.

Definition 8.12. Let G be the space of germs on a domain D. The evaluation

map E ∶ G → C is defined by

E([f ]z ) = f (z) .

Remark 8.13. With respect to a standard chart (π∣U , [f ]U ), the evaluation

map takes the form
E ○ (π∣U )−1 (z) = f (z)
which is analytic, since f is. Hence E (or, more precisely, its restriction to
each component of G) is analytic.

42
Lecture 9: Uniqueness of analytic continuation
and gluing
9.1 Analytic continuation revisited
When combined with the theory of covering spaces, the space of germs con-
structed in the last lecture gives a very clean account of analytic continuation.
Theorem 9.1. Let (f, U ) and (g, V ) be function elements on a domain D ⊆
C, and let γ ∶ [0, 1] → D be a path starting in U and ending in V . Then
(f, U ) ≈γ (g, V ) if and only if the lift γ̃ to (a component of ) G starting at
[f ]γ(0) exists, and ends at [g]γ̃(1) .
Proof. For the ‘only if’ direction, suppose that (f, U ) ≈γ (g, V ). That is,
there is a sequence of direct analytic continuations as in the definition.

(f, U ) = (f1 , U1 ) ∼ . . . ∼ (fn−1 , Un−1 ) ∼ (fn , Un ) = (g, V ) ,

a continuous path γ ∶ [0, 1] → D, and a dissection

0 = t0 < t1 < . . . < tn−1 < tn = 1

such that γ([ti−1 , ti ]) ⊆ Ui for each 1 ≤ i ≤ n. Define a lift by

γ̃(t) = [fi ]γ(t)

whenever t ∈ [ti−1 , ti ], which is well-defined since [fi ]γ(ti ) = [fi+1 ]γ(ti ) for each
0 < i < n. For continuity, note that

γ̃(t) = (π∣[fi ] )−1 ○ γ(t)

for t ∈ [ti−1 , ti ], so γ̃ is continuous on each interval in the dissection and hence

continuous on [0, 1].
For the ‘if’ direction, suppose that there is a lift γ̃ of γ to G such that
γ̃(0) = [f ]γ(0) and γ̃(1) = [g]γ(1) . By the compactness of [0, 1], there is a
finite sequence of function elements (fi , Ui ) for 1 ≤ i ≤ n and a dissection

0 = t0 < t1 < . . . < tn−1 < tn = 1

such that γ̃([ti−1 , ti ]) ⊆ [fi ]Ui for each 1 ≤ i ≤ n. Indeed, we may assume that
each Ui is an open disc in C. Applying the forgetful map π, it follows that

43
γ([ti−1 , ti ]) ⊆ Ui , so it remains only to prove that (fi−1 , Ui−1 ) ∼ (fi , Ui ) for
1 < i ≤ n. For each such i,
[fi−1 ]γ(ti−1 ) = γ̃(ti−1 ) = [fi ]γ(ti−1 )
so fi and fi−1 agree on a neighbourhood of γ(ti−1 ) ∈ Ui−1 ∩Ui . Since the Ui are
discs, Ui−1 ∩Ui is connected, so it follows by the identity principle that fi−1 and
fi agree on the whole intersection Ui−1 ∩ Ui . Therefore (fi−1 , Ui−1 ) ∼ (fi , Ui )
as required.
Theorem 9.1 immediately implies that we can give a much more concrete
description of complete analytic functions.
Corollary 9.2. Let F be a complete analytic function on a domain D ⊆ C.
Then
GF ∶= ⋃ [f ]U
(f,U )∈F

is a path component of G.
Since the space of germs is equipped with the evaluation map E, this result
tells us that a complete analytic function on a domain D is essentially the
same thing as a Riemann surface R equipped with a covering map π ∶ R → D
and an analytic function R → C.
Definition 9.3. The component GF is the Riemann surface associated to F.

9.2 The classical monodromy theorem

With these tools in hand, it becomes easy to prove the uniqueness results we
want about analytic continuation. Specifically, analytic continuation along a
path only depends on the homotopy class of the path.
Theorem 9.4 (Classical monodromy theorem). Let D ⊆ C be a domain.
Suppose that (f, U ) is a function element in D and can be continued along
any path in D starting in U . If (f, U ) ≈α (g1 , V ) and (f, U ) ≈β (g2 , V ) and
α ≃ β then g1 ≡ g2 on V .
Proof. Let α̃, β̃ be the lifts to G of α, β respectively, starting at [f ]α(0) =
[f ]β (0). By Theorem 9.1 and the monodromy theorem,
[g1 ]α(1) = α̃(1) = β̃(1) = [g2 ]β(1)
so g1 and g2 agree on a neighbourhood of α(1) = β(1). Therefore, g1 and g2
agree on V by the identity principle.

44
 In particular, analytic continuation is unique on any simply connected
domain.

Corollary 9.5. Let D be a simply connected domain and (f, U ) a function

element on D. If (f, U ) can be analytically continued along every path in D
starting in U then (f, U ) extends to an analytic function f ∶ D → C.

9.3 Gluing Riemann surfaces

In Example 4.3, we analysed kth roots on C by constructing the associated
Riemann surface Rk , together with a covering map π ∶ Rk → C∗ and an
analytic function g ∶ Rk → C∗ . We also noted that Rk can be embedded in
a compact Riemann surface R ̂k . But how can we do this in practice? The
answer is by gluing.

Definition 9.6. Let X, Y be topological spaces. Suppose we have subspaces

X ′ ⊆ X and Y ′ ⊆ Y , and a homeomorphism Φ ∶ X ′ → Y ′ . The quotient space

Z ∶= (X ⊔ Y )/ ∼

where ∼ is the finest equivalence relation such that x ∼ Φ(x) for all x ∈ X ′ ,
is called the result of gluing X and Y along Φ. It may also sometimes be
denoted by X ∪Φ Y , or even by X ∪X ′ Y if the map Φ is implicit.

In the context of this course, we particularly want to know when the

result of gluing produces a Riemann surface.

Proposition 9.7. Let R1 , R2 be Riemann surfaces. Suppose that Sj ⊆ Rj

(for j = 1, 2) are non-empty, connected, open subsets, and Φ ∶ S1 → S2 is
a conformal equivalence of Riemann surfaces. There is a unique conformal
structure on
R = R1 ∪Φ R2
such that the inclusion maps ij ∶ Rj ↪ R are analytic. In particular, if the
resulting gluing R is Hausdorff then it is a Riemann surface.

Proof. For j = 1, 2, every chart (φj , Uj ) on Rj gives rise to a chart (φj ○

i−1
j , ij (U )) on R. By construction these charts cover R. The transition func-
tions between two charts arising from Rj are just transition functions of Rj ,
hence analytic. If (φ1 , U1 ) and (φ2 , U2 ) are charts on R1 and R2 respectively,

45
giving rise to charts (φ1 ○ i−1 −1
1 , i1 (U )) and (φ2 ○ i2 , i2 (U )), then the resulting
transition function is

φ2 ○ i2−1 ○ i1 ○ φ−1 −1
1 = φ2 ○ Φ ○ φ1 ,

which is analytic because Φ is a conformal equivalence.

To prove the uniqueness statement, suppose that A is any conformal
structure on R that makes ij analytic. If (φj , Uj ) is a chart on Rj and
(ψ, V ) ∈ A, then
ψ ○ ij ○ φ−1
j

is analytic. Thus (φj ○i−1

j , ij (U )) has analytic transition functions with every
chart in A, so is contained in A by maximality. This proves uniqueness.
Finally, note that, since R1 and R2 are path-connected and the Si are
non-empty, the gluing R is also path-connected. Since we have assumed that
R is Hausdorff, it follows that it is a Riemann surface.
It is an irritation that non-Hausdorff spaces are quite easy to construct
by gluing, as the following example shows.
Example 9.8. Let R1 = R2 = C, let S1 = S2 = C∗ , and let Φ ∶ S1 → S2 be the
identity map. The gluing
R = C ∪C∗ C
is not Hausdorff, since the two copies of 0 do not have disjoint open neigh-
bourhoods.
Therefore, we always have to check Hausdorffness of a gluing by hand.
Gluing is a convenient way to construct compactifications. For instance,
we can use gluing to compactify the complex plane to the Riemann sphere.
Example 9.9. Let R1 = R2 = C, let S1 = S2 = C∗ , and let Φ ∶ S1 → S2 be the
inversion map z ↦ 1/z. Every pair of points in the gluing

R = C ∪Φ C

is contained in either R1 or R2 , except for the pair {i1 (0), i2 (0)}. Therefore,
to check that R is Hausdorff, we only need to check that this pair have
disjoint open neighbourhoods. Indeed, i1 (D) and i2 (D) is a disjoint pair of
open neighbourhoods, so R is Hausdorff. By Proposition 9.9, R is therefore
a Riemann surface, and indeed it is easily seen to be the Riemann sphere
C∞ . Furthermore, this approach makes it clear that R is compact: R =
i1 (D) ∪ i2 (D), so is the continuous image of two closed discs.

46
Lecture 10: More gluing and branching
10.1 A more detailed gluing example
In this section, we give a more involved example of compactification by glu-
ing. In Example 6.3, we constructed
√ the Riemann surface R associated to
the complete analytic function z − z, and noticed that the result was a
3

torus with four points removed. We would like to be able to compactify this
Riemann surface to a compact torus.
In Example Sheet 1, question 14, you saw that the graph

R1 = {(z, w) ∈ C2 ∣ w2 = z 3 − z} ,

is conformally equivalent to R with three points added.1 To complete the

compactification of R, it remains to ‘add a point at infinity’.
Example 10.1. First, we perform a change of coordinates to send the point
at infinity to a finite point, as in the example of the Riemann sphere. Setting
u = 1/z and v = z/w, we can rearrange w2 = z 3 − z to obtain the equation
u = v 2 (1 − u2 ).2 Therefore, consider the graph

R2 = {(u, v) ∈ C2 ∣ u = v 2 (1 − u2 )} .

We next need to check that R2 is indeed a Riemann surface and carries a

conformal structure. As in question 14 of Example Sheet 1, the idea is to
define an atlas on R2 using the two coordinate projections, π(u, v) = u and
τ (u, v) = v.
Noting that u ≠ ±1, the defining equation of R2 can be rewritten as
√
u
v = v(u) = ± .
1 − u2
The projection π therefore has local inverses defined by

π −1 (u) = (u, v(u))

1
In the example sheet, the graph is denoted by R, and R0 is the result of deleting three
points.
2
We won’t discuss how we found this change of coordinates further in this course, so
you will have to think of it as found by trial and error. In questions, changes of coordinates
will usually be given. If you take Part II Algebraic Geometry, you will learn more about
where these kinds of changes of coordinates come from.

47
for suitable choices of branches of the square root. These branches can be
chosen unless u = 0, in which case v = 0 too. Thus, restrictions of π provide
charts on the whole of R2 ∖ {(0, 0)}.
Thus, it remains to provide a chart about (0, 0), and for this we use the
other coordinate projection, τ . The defining equation of the graph can also
be rewritten as
v 2 u2 + u − v 2 = 0 ,
so τ has local inverses
τ −1 (v) = (u(v), v)
where √
1 + 4v 4
−1 ±
u(v) =
2v 2
for a suitable choice of branch of square root. This is not defined at v = 0,
but has a removable singularity there if we make the correct choice of branch.
Indeed, about v = 0, taking the positive branch of the square root, we have
the power series expansion
1 + 4v 4 /2 + o(v 8 ) − 1
u(v) = = v 2 + o(v 6 ) .
2v 2
Thus, τ has an inverse in a neighbourhood of {(0, 0)}, and so defines a chart
there. Taken together with restrictions of π, we have charts about every point
of R2 , and the transition functions for this collection of charts, apart from
the identity, are the analytic maps u(v) and v(u). Therefore, they define
an atlas, and hence a conformal structure. Furthermore, R2 is a subspace
of C2 , hence is Hausdorff. Below, we shall see that R2 is obtained from R1
by deleting one point and adding one point; from this it follows that R2 is
connected. Hence, R2 is a Riemann surface.
The map Φ from a subset of R1 to a subset of R2 is given by the change
of coordinates, which is to say that
Φ(z, w) = (1/z, z/w) and Φ−1 (u, v) = (1/u, 1/uv) .
The domain of Φ is
S1 = {(z, w) ∈ R1 ∣ z, w ≠ 0} = R1 ∖ {(0, 0), (1, 0), (−1, 0)} ,
and the domain of Φ−1 is
S2 = {(u, v) ∈ R2 ∣ u, v ≠ 0} = R2 ∖ {(0, 0)} .

48
Since the atlases on R1 and R2 are given by coordinate projections, Φ and
Φ−1 are given in local charts by their coordinate formulae, which are analytic
functions, so Φ is a conformal equivalence.
Using all of the above, we may define the gluing
̂ = R1 ∪Φ R2 .
R

We will shortly check that this is Hausdorff, but first, notice that the mero-
morphic functions π̂1 (z, w) = z on R1 and π̂2 (u, v) = 1/u on R2 satisfy

π̂2 ○ Φ = π̂1

on S1 , and so descend to a meromorphic function π̂ ∶ R ̂ → C∞ . This function

makes it easy to prove that R ̂ is Hausdorff. Indeed, since S1 and S2 are
̂
identified in R, it suffices to find a disjoint pair of open neighbourhoods of
i1 (0, 0), i1 (±1, 0) and i2 (0, 0), and π̂ −1 ({∣z∣ < 2}) and π̂ −1 ({∣z∣ > 2} ∪ {∞}) are
such a pair.
Therefore, by Proposition 9.7, R ̂ is a Riemann surface. Let’s prove that it
is (sequentially) compact. For any sequence of points (pn )n≥0 of R1 , if π̂(pn )
remains bounded then, after passing to a subsequence, we may assume that
π̂(pn ) converges to some limit z0 ∈ C. But there are at most two points in R1
with z-coordinate equal to z0 so, after passing to a further subsequence, it
follows that (pn ) converges in R1 . On the other hand, if π̂(pn ) is unbounded
then, after passing to a subsequence, π̂(pn ) → ∞, so pn converges to i2 (0, 0),
which is the unique point p of R ̂ with π̂(p) = ∞.
In summary, we have compactified R1 to a Riemann surface R, ̂ in such a
way that π extends to a meromorphic function on R. ̂ Similarly, the function
g on R1 also extends to a meromorphic function on R; ̂ the details are left to
the reader.
In the case of Example 10.1, we can understand the topology of R ̂ di-
rectly, since we visualised it via a gluing construction in Example 6.3: it is
homeomorphic a torus. In general, however, these kinds of computations are
difficult to perform. It turns out that the key to determining the topology
of a Riemann surface R ̂ is to understand the points where a meromorphic
function like π̂ fails to be a local homeomorphism. This phenomenon is called
branching, and is the topic of our next section.

49
10.2 Branching
Compactifying Riemann surfaces often leads to meromorphic functions that
are no longer covering maps. For instance, although the power map pk (z) = z k
is a covering map C∗ → C∗ , its natural extension to the Riemann sphere
p̂k ∶ C∞ → C∞ fails to be a local homeomorphism at 0 and ∞ if k ≥ 2.
Definition 10.2. Let f ∶ R → S be an analytic map of Riemann surfaces
and let p ∈ R. By Proposition 4.6, there are choices of charts (φ, U ) about p
and (ψ, V ) about f (p), with φ(p) = 0, such that

ψ ○ f ○ φ−1 (z) = z mf (p)

for some integer mf (p) ≥ 0. Note that the integer mf (p) is equal to the
number of preimages in a sufficiently small neighbourhood of p of any point in
a sufficiently small punctured neighbourhood of f (p), and so is independent
of the choice of charts. This is the multiplicity of f at p.
Unless f is constant, ‘most’ points have ramification index equal to 1.
The remaining points are especially interesting, and so we introduce special
terminology for them.
Definition 10.3. If mf (p) > 1 then p is called a ramification point and f (p)
is called a branch point of f . In this case, the multiplicity mf (p) is also
sometimes called the ramification index of p.
Example 10.4. Let p̂k ∶ C∞ → C∞ be the power map z ↦ z k for k ≥ 2, thought
of as meromorphic function on the Riemann sphere. The only points with
multiplicity greater than 1 are 0 and ∞, so these are the ramification points,
each of which has ramification index equal to k. The branch points are their
images, which are also 0 and ∞.
Arbitrary polynomials behave similarly.
Example 10.5. Let f ∶ C∞ → C∞ be any polynomial

f (z) = ad z d + . . . + a1 z + a0

with ad ≠ 0. If w = φ(z) = 1/z is the chart about infinity then

φ ○ f ○ φ−1 (w) = 1/(ad w−d + ad−1 w1−d + . . . + a0 )

= wd /(ad + ad−1 w + . . . + a0 wd )

50
which shows that mf (∞) = d, since 1/(ad +. . .+a0 wd ) is analytic and non-zero
in a neighbourhood of 0. That is, for any polynomial, the multiplicity of ∞
is the degree d.
For more complicated maps it may not be easy to immediately see which
points are ramification points. However, differentiation gives a convenient
method for analytic functions.
Remark 10.6. Let f be a non-constant analytic function on a Riemann surface
R, let p ∈ R and let (φ, U ) be any chart about p with φ(p) = z0 . Then
F = f ○ φ−1 (z) = (z − z0 )mf (p) g(z)
for some analytic function g with g(z0 ) ≠ 0. Hence,
F ′ (z) = (mf (p)g(z) + (z − z0 )g ′ (z))(z − z0 )mf (p)−1
by the product rule. Therefore, we have
F ′ (z0 ) = g(z0 ) ≠ 0
if mf (p) = 1, whereas we have F ′ (z0 ) = 0 if mf (p) > 1. In summary, the
ramification points of an analytic function are exactly the points where the
derivative vanishes, in any choice of local coordinates.
The next lemma tells us that multiplicity behaves as you might hope
under composition of maps.
Lemma 10.7. If f ∶ R → S and g ∶ S → T are analytic functions of Riemann
surfaces then
mg○f (p) = mg (f (p))mf (p)
for any point p ∈ R.
Proof. Fix any chart (θ, W ) about g ○ f (p), with θ ○ g ○ f (p) = 0. By Propo-
sition 4.6, there is a chart (ψ, V ) about f (p) such that
θ ○ g ○ ψ −1 (z) = z mg (f (p))
on a neighbourhood of 0. Likewise, there is a chart (φ, U ) about p such that
ψ ○ f ○ φ−1 (z) = z mf (p)
on a neighbourhood of 0, by the same proposition. Therefore,
θ ○ (g ○ f ) ○ φ−1 (z) = (θ ○ g ○ ψ −1 ) ○ (ψ ○ f ○ φ−1 )(z) = (z mf (p) )mg (f (p))
and the result follows.

51
Lecture 11: The valency and Riemann–Hurwitz
theorems
11.1 The valency theorem
The next theorem, the main result of this section, is our first step towards
using analytic maps to determine the topology of surfaces. It says that any
non-constant map f between compact Riemann surfaces is n-to-1, for some
well-defined integer n, as long as we count with multiplicities.
Theorem 11.1 (Valency theorem). Suppose that f ∶ R → S is a non-
constant, analytic map between compact Riemann surfaces. The function
n ∶ S → N defined by
n(q) ∶= ∑ mf (p)
p∈f −1 (q)

is constant on S.
Proof. Since R is compact, each q ∈ S only has finitely many pre-images in
R, by the identity principle, so n is indeed a well-defined map to N.
Since R is connected, it suffices to prove that n is locally constant. There-
fore, taking an arbitrary q0 ∈ S with n(q0 ) = n0 , it suffices to find an open
neighbourhood of q0 on which n(q) = n0 . Let
f −1 (q0 ) = {p1 , . . . , pk } ,
and fix a chart (ψ, V ) about q0 . By Proposition 4.6, there is a chart (φi , Ui )
about each pi , such that
ψ ○ f ○ φ−1
i (z) = z
mf (pi )

on Ui . Passing to smaller charts if necessary, we may furthermore assume

that the {Ui } are disjoint. Now R ∖ ⋃i Ui is closed, hence compact so the
image K = f (R ∖ ⋃i Ui ) is also compact, hence closed. Therefore, there is a
connected open neighbourhood V ′ ⊆ V of q0 that is disjoint from K. Thus
f −1 (V ′ ) ⊆ R ∖ f −1 (K) ⊆ R ∖ (R ∖ ⋃ Ui ) = ⋃ Ui .
i i

Taking Ui′ = f −1 (V ′ )∩Ui , we obtain charts (φi , Ui′ ) about pi and (ψ, V ′ ) about
q such that the local form of f is a power map everywhere on the preimage
of V ′ . In particular, n(q) = n for all q ∈ V ′ , as required.

52
 Evidently the constant n is an important invariant of f .
Definition 11.2. The constant n that Theorem 11.1 associates to a non-
constant analytic map f of compact Riemann surfaces is called the degree or
valency of f , and denoted by deg(f ).
Note that, by Example 10.5, this notion of degree coincides with the
usual one when f is a polynomial. In any course on complex analysis It is
traditional to give a proof of the fundamental theorem of algebra. We could
have done so several times already, but this is a particularly good moment,
since it is now an easy corollary of the valency theorem.
Corollary 11.3 (Fundamental theorem of algebra). Any polynomial f of
degree d has exactly d zeroes, counted with multiplicity.
Proof. By Example 10.5, the valency of f is d. The result now follows from
the valency theorem.

11.2 Euler characteristic

In this section, we build on the valency theorem, and see how to use the
branching data of analytic maps to determine the topology of compact Rie-
mann surfaces. First, we have to briefly review the topological classification
of compact surfaces. The crucial invariant is the Euler characteristic.
Definition 11.4. Let S be a compact Riemann surface. A topological trian-
gle in S is a continuous embedding ∆ ↪ S, where ∆ is a closed triangle in the
plane R2 . A triangulation of S is a finite collection of topological triangles
{∆i } satisfying the following conditions.
(i) The union ⋃i ∆i is the whole of S.
(ii) Unless i = j, the intersection ∆i ∩ ∆j is either empty, a vertex or an
edge.
(iii) Each edge is an edge of exactly two triangles.
The Euler characteristic of the triangulation {∆i } is defined to be
χ=V −E +F
where F is the number of triangles ∆i , E is the number of edges, and V is
the number of vertices.

53
 These ideas can be applied to Riemann surfaces, because of two important
facts, which we do not prove in this course. They are summarised in the
following theorem.

Theorem 11.5. Every compact Riemann surface S admits a triangulation,

and the Euler characteristic χ is independent of the choice of triangulation.

The second assertion – that χ is independent of the choice of triangulation

– is proved in Part II Algebraic Topology. In light of the theorem, we may
write χ(S) for the Euler characteristic of S.
Example 11.6. The Riemann sphere is homeomorphic to a tetrahedron. There-
fore
χ(C∞ ) = 4 − 6 + 4 = 2 .
Example 11.7. Subdivide the square into nine squares, then divide each of
these into two triangles. Identifying opposite sides of the square in the usual
way, this exhibits a triangulation of the torus S 1 × S 1 . Counting the number
of faces, edges and vertices, it follows that

χ(C/Λ) = 9 − 18 + 9 = 0 ,

for any complex torus C/Λ.

The sphere and the torus are, in a sense, the simplest compact Riemann
surfaces. The number of ‘holes’ is called the genus: the sphere is the surface
of genus 0, and the torus is the surface of genus 1. A third important fact,
which again will remain unproved, is that every compact Riemann surface is
homeomorphic to some surface of genus g, for some g.

Theorem 11.8 (Topological classification of Riemann surfaces). Every com-

pact Riemann surface S is homeomorphic to the surface of genus g, for some
g. The Euler characteristic of S is then

χ(S) = 2 − 2g .

In particular, the Euler characteristic determines the topological type of

any compact Riemann surface. That is, if R and S are Riemann surfaces
such that χ(R) = χ(S), then R and S are homeomorphic.

54
11.3 The Riemann–Hurwitz theorem
In this section we will state, and sketch the proof of, the Riemann–Hurwitz
theorem. This is the key tool that enables us to deduce topological informa-
tion from the branching data of an analytic map.

Theorem 11.9 (Riemann–Hurwitz). Let f ∶ R → S be any non-constant

analytic map of compact Riemann surfaces. Then

χ(R) = deg(f )χ(S) − ∑ (mf (p) − 1) .

p∈R

Note that mf (p) − 1 > 0 only if p is a ramification point of f . Since R

is compact and ramification points are isolated, there are only finitely many
such points, and so the sum in the theorem is finite.
Sketch proof. As in the proof of the valency theorem, each q ∈ S has a neigh-
bourhood U such that f takes the form of a power map on each component
of the preimage of U . These neighbourhoods form an open cover of S so, by
compactness, there is a finite subcover

{U1 , . . . , Uk } ,

where Ui is the neighbourhood associated the point qi . The only point of Ui

that can be a branch point is qi itself, so there are only finitely many branch
points.
Take a triangulation of S. Each triangle contains at most finitely many
branch points, so after subdividing further, we may assume that each triangle
contains at most one branch point. Subdividing further still, we may assume
that each branch point is a vertex of the triangulation.
Now refine the triangulation further still, until each triangle is contained
in some Vi . In particular, the preimages of the triangles in R together form a
triangulation of R. To relate the Euler characteristics of these triangulations,
let’s introduce some notation: n = deg(f ); VS , ES , FS respectively denote the
number of vertices, edges and triangles in S; likewise, VR , ER , FR respectively
denote the number of vertices, edges and triangles in R.
We now count preimages in R.

(i) Each triangle in S has exactly n preimages in R, so FR = nFS .

(ii) Each edge in S has exactly n preimages in R, so ER = nES .

55
(iii) Each vertex q in S has exactly

n− ∑ (mf (p) − 1)
p∈f −1 (q)

preimages in R. (Note that the sum is zero unless q is a branch point.)

Therefore,

VR = nVS − ∑ ∑ (mf (p) − 1) = nVS − ∑ (mf (p) − 1) .

q∈S p∈f −1 (q) p∈R

Putting these together completes the proof:

χ(R) = FR − ER + VR
= nFS − nES + nVS − ∑ (mf (p) − 1)
p∈R

= nχ(S) − ∑ (mf (p) − 1)

p∈R

as required.

Lecture 12: Applications of Riemann–Hurwitz

12.1 Immediate consequences
Last time, we sketched the proof of the Riemann–Hurwitz theorem. Writing
gR and gS for the genera of R and S respectively, and n = deg(f ), the
Riemann–Hurwitz equation can be rewritten as

2gR − 2 = n(2gS − 2) + ∑ (mf (p) − 1) ,

p∈R

which is how we will usually use it.

The most important application is to the computation of the topological
types of compact Riemann surfaces. Let’s complete the story of our running
example by using Riemann–Hurwitz to compute its genus.
Example 12.1. In Example 10.1, we compactified
√ the Riemann surface associ-
ated to the complete analytic function z 3 − z, building a compact Riemann
̂ equipped with a meromorphic function π̂. Furthermore, π̂ has four
surface R

56
branch points – 0, ±1, ∞ – each ‘totally ramified’, meaning that their preim-
ages each consist of exactly one ramification point, of index equal to deg(π̂).
Since deg(π̂) = 2, Riemann–Hurwitz therefore gives

2gR̂ − 2 = 2 × −2 + 4 × (2 − 1) ,

which rearranges to give

gR̂ = −2 + 2 × (2 − 1) + 1 = 1 .
̂
This is consistent with our observation from the gluing construction that R
is a torus.
Usually, it will be much easier to apply Riemann–Hurwitz than a specific
gluing construction! Riemann–Hurwitz also has many general consequences,
which are worth noting. Here is one.
Remark 12.2. The correction term ∑p∈R mf (p) − 1 is even.
This observation is especially useful when looking at meromorphic func-
tions of degree 2. Let’s take one last look at our favourite example.
Example 12.3. Let
̂ = R1 ∪Φ R2
R
be as in Example 10.1, but let’s not use any information about R2 – that
is, let’s not use any explicit information about π̂ −1 (∞). Knowing that π̂ is
a map of degree two, with three totally ramified branch points in C, the
correction term in Riemann–Hurwitz is

3 × (2 − 1) + C ,

where C = ∑p∈π̂−1 (∞) (mπ̂ (p) − 1). In particular, C is odd. Since deg(π̂) = 2,
there are only two possibilities for π̂ −1 (∞).

(i) If #π̂ −1 (∞) = 2 then both preimages are unramified, so ∞ is not a

branch point and C = 0.

(ii) If #π̂ −1 (∞) = 1 then ∞ is totally ramified and C = 1.

Knowing that C must be even, only the second case is possible, and the
computation that gR̂ = 1 follows as before.

57
 The moral here is that, for meromorphic functions of degree two, we do
not need to explicitly construct the compactification at ∞ to deduce the
branching data there – it is enough to know that the compactification exists.
Riemannn–Hurwitz also places restrictions on the kinds of covering maps
between Riemann surfaces that can occur.
Remark 12.4. If f is a covering map (also called unramified ) then the cor-
rection term coming from the branching data is zero, so

gR − 1 = n(gS − 1) .

(i) If gS = 0 then we must have n = 1 and gR = 0. In particular, n is a

conformal equivalence.

(ii) If gS = 1 then we also have gR = 1, and there is no constraint on n.

(iii) If gS > 1 then either n = 1, in which case f is a conformal equivalence,

or gR > gS .

We do not have the tools to construct examples realising the third item,
but our understanding of complex tori makes the second item easy to realise.
Example 12.5. For any integer n ≥ 1, consider the lattice Λn = ⟨n, i⟩ in C.
The natural inclusion Λ1 ↪ Λn induces a quotient map of complex tori

C/Λn → C/Λ1

which is a covering map of degree n.

12.2 Higher-genus Riemann surfaces

We haven’t yet seen any example of Riemann surfaces of genus greater than
1. We’ll do this as with the previous example – by writing down an explicit
graph and then compactifying.
Example 12.6. Consider the Fermat curve of degree d, namely the graph

Fd′ ∶= {(x, y) ∈ C2 ∣ xd + y d = 1}

so-called because rational points on this graph correspond to integer solutions

to the famous Fermat equation xd + y d = z d .

58
 The projection πx ∶ (x, y) ↦ x has local inverse
√
πx−1 (x) = 1 − xd
d

unless x is a dth root of unity ζdi , in which case y = 0. Symmetrically, the

projection πy to the y coordinate is a local homeomorphism except at (0, ζdi ).
This collection of charts
√ covers Fd′ , and the transition functions are either
the identity or x ↦ 1 − xd , hence analytic. Therefore, this defines an atlas,
d

and hence a conformal structure. It is Hausdorff as a subset of C2 .

To check path-connectedness, define a simply-connected domain D in the
x-plane by making branch cuts joining the dth roots of unity to ∞. That is:
d
D = C ∖ ⋃{tζdi ∣ t ≥ 1} .
i=1

Since D is simply connected, there are well-defined branches of

√
y(x) = 1 − xd
d

defined on D, which extend continuously to the branch points ζdi . Let

(x0 , y0 ) ∈ Fd′ with x0 ∈ D, and choose the branch of y(x) so that y(x0 ) = y0 .
Since D is path-connected, there is a continuous path γ in D from x0 to 1,
so the path
γ̃(t) ∶= (γ(t), y(γ(t)))
joins (1, 0) to some (x0 , y0 ).
In summary, every point (x0 , y0 ) ∈ Fd′ is in the same path component of
(1, 0) unless x0 ∉ D. If x0 ∉ D, the same argument on a small disc about x0
shows that a short path joins (x0 , y0 ) to some (x1 , y1 ) ∈ Fd′ such that x1 ∈ D.
Therefore, Fd′ is path connected.
On question 13 of Example Sheet 3, you show that Fd′ can be compactified
to a Riemann surface Fd by adding d points, and πx extends to a meromorphic
function π̂x , in such a way that such that

π̂x−1 (∞) = Fd ∖ Fd′ .

For a generic choice of x0 ∈ C, there are p points in Fd′ satisfying πx (p) = x0 ,

so deg(π̂x ) = d. Thus, ∞ is not a branch point of π̂x , and our computations
above showed that the only branch points of πx are the roots of unity ζdi ,
which are totally ramified. In summary, π̂x has d ramification points, each
of multiplicity d.

59
 From Riemann–Hurwitz applied to π̂x we now obtain the formula

2gd − 2 = d × −2 + d(d − 1)

for the genus gd of Fd . Rearranging this gives

gd = (d − 1)(d − 2)/2 .

Thus, this construction gives us compact Riemann surfaces of arbitrarily

large (but not all) genera.

Lecture 13: Rational and periodic functions

Corollary 9.2 tells us that understanding analytic continuation is equivalent
to understanding certain functions on Riemann surfaces. In general this is
an impossible task, but there is hope when the Riemann surface is compact.
In this lecture, we will classify the meromorphic functions on the Riemann
sphere, and then move on try to understand the meromorphic functions on
complex tori.

13.1 Rational functions

Riemann–Hurwitz gives us a much better understanding of compact Riemann
surfaces, by enabling us to compute their topology. As well as compact Rie-
mann surfaces themselves, we are also interested in analytic and meromorphic
functions on those surfaces. Corollary 5.5 tells us that analytic functions will
be constant, but the set of meromorphic functions on a compact Riemann
surface R can still be very interesting, and we would like to classify them if
possible. In this section, we will do that for the simplest possible compact
Riemann surface – the Riemann sphere, C∞ .

Proposition 13.1. Every meromorphic function on the Riemann sphere f ∶

C∞ → C∞ is a rational function. That is, it is of the form

(z − a1 ) . . . (z − am )
f (z) = c
(z − b1 ) . . . (z − bn )

for some integers m, n ≥ 0 and constants ai , bj , c ∈ C.

60
Proof. Clearly we may assume that f is non-constant. After replacing f by
1/f if necessary, we may also assume that f (∞) ∈ C. Now f −1 (∞) is a finite
set of poles b1 , . . . , bn′ ∈ C, and f takes the form
∞
f (z) = ∑ cj,l (z − bj )l
l=−kj

in a punctured neighbourhood of each pole bj . Defining Qj to be the principal

part
−1
Qj (z) = ∑ cj,l (z − bj )l ,
l=−kj

we see that all the singularities of the function

n′
g(z) = f (z) − ∑ Qj (z)
j=1

are removable. Hence g(z) is non-surjective self-map from C∞ , and hence

constant. Since the Qj are all rational functions, the result follows.
Of course, we should assume that the representation
(z − a1 ) . . . (z − am )
f (z) = c
(z − b1 ) . . . (z − bn )
is chosen so that m and n are minimal, or equivalently that the ai are all dis-
tinct from the bj . The hypothesis that f (∞) ∈ C in the proof of Proposition
13.1 is equivalent to assuming that m ≤ n, and in this case the proof shows
that deg(f ) = n.
Remark 13.2. For a rational function
(z − a1 ) . . . (z − am )
f (z) = c
(z − b1 ) . . . (z − bn )
as above, we have deg(f ) = max{m, n}.

13.2 Simply periodic functions

Having classified the meromorphic functions on C∞ , we would like to carry
out something similar for other Riemann surfaces. The fact that many Rie-
mann surfaces can be constructed as quotients is useful here, because it
enables us to give an alternative description of the meromorphic functions as
periodic functions on domains in C.

61
Definition 13.3. Let f ∶ C → C∞ be meromorphic. A period of f is a
complex number ω ∈ C such that f (z + ω) = f (z) for all z ∈ C.

Note that the set of periods Ω of any meromorphic function f on C is an

additive subgroup of C. We may therefore ask what the subgroup of periods
can be, and it turns out that there are not too many possibilities.

Lemma 13.4. Let Ω be the set of periods of a meromorphic function f on

C. One of the following holds:

(i) Ω = {0};

(ii) Ω = ⟨ω⟩ ≅ Z for some ω ≠ 0;

(iii) Ω = ⟨ω1 ⟩ ⊕ ⟨ω2 ⟩ ≅ Z2 for some ω1 , ω2 linearly independent over R;

(iv) Ω = C.

Proof. See question 1 of Example Sheet 3.

The case of Ω = {0} is of course the ‘generic’ case, and such f are too
complicated to classify. By contrast, the case of Ω = C is very simple: it
occurs exactly when f is constant. Before moving on to the case of Ω ≅ Z2 ,
let’s briefly consider the case Ω ≅ Z.

Definition 13.5. A meromorphic function f on C for which the group of

periods is isomorphic to Z is called simply periodic.

Some of our favourite examples are simply periodic.

Example 13.6. The exponential function has periods ⟨2πi⟩.
Another way to think about the exponential function is as a regular cov-
ering map exp ∶ C → C∗ , as in Example 3.4. This suggests an alternative
interpretation of simply periodic functions.

Proposition 13.7. If f is a meromorphic function on C and the periods of

f contain an infinite cyclic subgroup ⟨ω⟩, then there is a unique meromorphic
function f¯ on C∗ such that

f (z) = f¯ ○ exp((2πi/ω)z)

for all z ∈ C.

62
Proof. On a small open neighbourhood of any point in C∗ , choose a branch
of the complex logarithm and define

f¯(w) = f ((ω/2πi) log(w)) ,

which is evidently a locally defined analytic function that satisfies

f¯ ○ exp((2πi/ω)z) = f ((ω/2πi) log(exp((2πi/ω)z))) = f (z)

as required. If fˆ is defined in the same way by choosing a different choice of

branch, differing from the original choice by 2πin for some n ∈ Z, then

fˆ(w) = f ((ω/2πi)(log(w) + 2πin))

= f ((ω/2πi) log(w) + nω))
= f ((ω/2πi) log(w))
= f¯(w)

since nω ∈ ⟨ω⟩ is a period of f . Therefore the definition of f¯ is independent

of the choice of branch of logarithm, which proves the result.
The moral of Proposition 13.7 is that simply periodic functions are es-
sentially the same things are functions on C∗ . Since C∗ is non-compact,
classifying meromorphic functions on C∗ is still too much to ask for. How-
ever, the idea that functions on a Riemann surface are equivalent to periodic
functions on a covering space is very useful, as we shall see in the next section.

13.3 Doubly periodic functions

Definition 13.8. A meromorphic function f on C with periods Ω = ⟨ω1 ⟩ ⊕
⟨ω2 ⟩ ≅ Z2 is said to be doubly periodic or elliptic.
In this case the period group is a lattice Λ, as in Example 5.1, where we
constructed the complex tori C/Λ. Indeed, replacing exp by the quotient
map π ∶ C → C/Λ, there is a precise analogue to Proposition 13.7.
Proposition 13.9. If f is a meromorphic function on C and the periods of
f contain a lattice Λ, then there is a unique meromorphic function f¯ on the
complex torus such that
f (z) = f¯ ○ π
for all z ∈ C.

63
Proof. The proof is identical to the proof of Proposition 13.7, using π instead
of the exponential function and local inverses to π instead of branches of the
logarithm.
Thus, we can study functions on complex tori by studying doubly periodic
functions on C. We will often abuse notation by identifying a doubly periodic
function f with the induced function f¯ on a complex torus. The next result
follows immediately from Proposition 13.9 and Corollary 5.5.
Corollary 13.10. Any analytic function f on C that is doubly periodic is
constant.
Here’s another example of an application of the theory of compact Rie-
mann surfaces, this time using Riemann–Hurwitz. If f is doubly periodic,
we define deg(f ) to be the degree of the associated function f¯ on a complex
torus.
Corollary 13.11. If f is a doubly periodic, non-constant, meromorphic func-
tion then deg(f ) ≥ 2.
Proof. If deg(f ) = 1 then in particular f has no ramification points, and
hence mf (p) = 1 for all p. The Riemann–Hurwitz theorem then gives
(2 × 1 − 2) = 1 × (2 × 0 − 2)
which is absurd.
A doubly periodic function f is determined by its values on any period
parallelogram
P ∶= {z0 + t1 ω1 + t2 ω2 }
where t1 , t2 ∈ [0, 1]. (We will usually take z0 = 0, but sometimes it will be
convenient to perturb P slightly.) This point of view offers a more concrete
approach to studying doubly periodic functions. To illustrate this, we give
another proof of the last result.
Alternative proof of Corollary 13.11. Let P be a period parallelogram, cho-
sen so that no zeroes or poles lie on the boundary ∂P. (We can perturb P
to make this true, since there are only deg(f ) zeroes or poles by the valency
theorem.) The residue theorem now gives
1
∑ resz (f ) = f (z)dz .
z∈poles(f )∩P 2πi ∮∂P

64
If the path α is the straight line from z0 to z0 + ω1 and β is the straight
line from z0 + ω1 to z0 + ω1 + ω2 then the boundary ∂P decomposes as the
concatenation
α ⋅ β ⋅ (ᾱ + ω2 ) ⋅ (β̄ − ω1 ) .
Now
∫α f (z)dz + ∫ᾱ+ω f (z)dz = ∫α f (z)dz + ∫ᾱ f (z)dz = 0
2

since ω2 is a period of f . The same holds to the sides parallel to β so we

conclude that the sum of the residues is zero. This is only possible if f has
at least two poles in P, counted with multiplicity.

Lecture 14: The Weierstrass ℘-function

14.1 The definition
Let Λ = ⟨ω1 , ω2 ⟩ be a lattice in C. Can we construct any non-constant func-
tions on a complex torus C/Λ, or equivalently, any doubly periodic functions
with periods Λ? By Corollary 5.5 such a function must have a pole, and
Corollary 13.11 tells us that it must have either two poles or a pole of degree
2. This suggests that we could try to construct a function that is approx-
imately 1/z 2 in a small disc about 0. A few minutes’ thought about the
constraints we have imposed on ourselves suggests the following definition.
Definition 14.1. Let Λ be a lattice in C2 . The associated Weierstrass ℘-
function is defined by
1 1 1
℘Λ (z) ∶= + ∑ ( − 2) .
ω∈Λ∖{0} (z − ω)
z 2 2 ω

We will usually write ℘ ≡ ℘Λ when the lattice Λ is understood.

We now need to check that this series really does converge. To do this,
we use the following lemma.
Lemma 14.2. Let Λ = ⟨ω1 , ω2 ⟩ be a lattice in C and t ∈ R. The sum
1
∑
ω∈Λ∖{0} ∣ω∣
t

converges if and only if t > 2.

65
Proof. The set {(t1 , t2 ) ∈ R2 ∣ ∣t1 ∣ + ∣t2 ∣ = 1} is compact, so the function

(t1 , t2 ) ↦ ∣t1 ω1 + t2 ω2 ∣

achieves its maximum M and minimum m, which are both non-negative

numbers. If m = 0 then
t1 ω1 + t2 ω2 = 0
for some non-zero t1 and t2 , which contradicts the hypothesis that ω1 and ω2
are linearly independent over R. In summary,

0 < m ≤ ∣t1 ω1 + t2 ω2 ∣ ≤ M < ∞

whenever ∣t1 ∣ + ∣t2 ∣ = 1. Now let (k, l) ∈ Z2 . Setting t1 = k/(∣k∣ + ∣l∣) and
t2 = l/(∣k∣ + ∣l∣) gives

m(∣k∣ + ∣l∣) ≤ ∣kω1 + lω2 ∣ ≤ M (∣k∣ + ∣l∣) .

Therefore the sum we’re interested in, namely

1
∑ ,
(k,l)∈Z2 ∖{0} ∣kω1 + lω2 ∣
t

is bounded above and below by constant multiples of the series

1
∑ .
(k,l)∈Z2 ∖{0} (∣k∣ + ∣l∣)
t

Setting n = ∣k∣ + ∣l∣ and noting that there are precisely 4n pairs (k, l) with
∣k∣ + ∣l∣ = n > 0, this sum can be rewritten as
∞ ∞
1 4n 1
∑ = ∑ = 4 ∑ ,
(k,l)∈Z2 ∖{0} (∣k∣ + ∣l∣)
t t t−1
n=1 n n=1 n

so the sum converges if and only if t > 2, as required.

Using this result, we can prove that our definition of the ℘-function is
well defined.

Theorem 14.3. The function ℘Λ is a well-defined elliptic function with pe-

riods Λ. Moreover, ℘Λ is even and of degree 2.

66
Proof. First, let’s prove convergence of the sum for a fixed z. Start by esti-
mating the summands:

1 1 ω 2 − (z − ω)2
∣ − ∣ = ∣ ∣
(z − ω)2 ω 2 ω 2 (z − ω)2
z(2ω − z)
= ∣ ∣
ω 2 (z − ω)2
z (2ω − z)
= ∣ ∣∣ ∣
ω 2 (z − ω)2
∣z∣ 2∣ω − z∣ + ∣z∣
≤ ⋅
∣ω∣2 ∣z − ω∣2
∣z∣ 2 ∣z∣
= ( + )
∣ω∣ ∣z − ω∣ ∣z − ω∣2
2

Fix some R ≥ ∣z∣. For all but finitely many ω ∈ Λ, ∣ω∣ ≥ 2R, whence ∣ω − z∣ ≥
∣ω∣/2 ≥ R. These summands are therefore bounded above by

R 4 R 6R
( + )= 3
∣ω∣ ∣ω∣ R∣ω∣/2
2 ∣ω∣

The exponent of 3 implies that the sum defining ℘Λ converges absolutely and
uniformly on compact subsets, by Lemma 14.2.
The definition implies immediately that ℘Λ is even.
To prove that ℘Λ is elliptic, we need to prove that each ω0 ∈ Λ is a period
of ℘Λ . Let’s start by considering the derivative:
−2
℘′Λ (z) = ∑ .
ω∈Λ (z − ω)
3

Note that his expression as a sum implies immediately that ω0 is a period of

℘′Λ . Therefore, the function

℘Λ (z + ω0 ) − ℘Λ (z) = c

for some constant c, because its derivative is zero. Setting z = −ω0 /2 and
using the fact that ℘Λ is even gives

℘Λ (ω0 /2) = c + ℘Λ (−ω0 /2) = c + ℘Λ (ω/2) ,

67
so c = 0, and ω0 is indeed a period of ℘Λ as required.
Finally, note that, by Lemma 14.2, the only poles of ℘′Λ are at the lattice
points Λ, so the periods are precisely Λ. In particular, ℘Λ has a unique pole
of order 2 on C/Λ, so deg(℘Λ ) = 2 as claimed.
Remark 14.4. The proof shows that ℘Λ has the following properties:

(i) ℘Λ is meromorphic with periods Λ;

(ii) ℘Λ has poles only at Λ;

(iii) limz→0 (℘Λ (z) − 1/z 2 ) = 0.

Indeed, these properties uniquely characterise ℘Λ . If f (z) is any other func-

tion satisfying (i) and (ii), then f (z) − ℘Λ (z) is an elliptic function on C with
poles only at the lattice points Λ. But item (iii) implies that f (z)−℘Λ (z) → 0
as z → 0, so the poles are in fact removable singularities, and f (z) − ℘Λ (z) is
constant by Corollary 13.10. This constant is 0 by item (iii) again, so f = ℘Λ .

14.2 Branching properties of ℘Λ

Our goal for this section is to investigate the branching behaviour of ℘Λ . To
do this, Remark 10.6 tells us that we should investigate the derivative ℘Λ .
Remark 14.5. Since
−2
℘′Λ (z) = ∑
ω∈Λ (z − ω)
3

we see that ℘′Λ has poles precisely at the lattice points Λ, and these poles
are of order 3. Hence the periods are precisely Λ and deg(℘′Λ ) = 3. It also
follows immediately that ℘′Λ is odd.
Therefore, for any ω ∈ Λ,

℘′Λ (ω/2) = −℘′Λ (−ω/2) = −℘′Λ (ω/2) ,

by oddness and periodicity, so ℘′Λ (ω/2) = 0. Thus, ℘′Λ has zeroes at the three
half-lattice points ω1 /2, ω2 /2 and (ω1 + ω2 )/2 in the fundamental parallelo-
gram. By the valency theorem, these are all the zeroes and each is simple,
since deg(℘′Λ ) = 3.
This information about the zeroes of ℘′Λ translates into the information
we want about the branching of ℘Λ .

68
Remark 14.6. By Remark 10.6, aside from its poles, the ramification points of
℘Λ are the points where the derivative vanishes. Therefore, the ramification
points of ℘Λ in the fundamental parallelogram are 0 and the three half-lattice
points, and each has multiplicity 2. The corresponding branch points are ∞
and three finite values ℘Λ (ω1 /2), ℘Λ (ω2 /2) and ℘Λ ((ω1 + ω2 )/2). Note that
these three values are distinct, by the valency theorem. In the future, we
shall let e1 , e2 and e3 denote them.
Finally, note that this is consistent with our expectations: Riemann–
Hurwitz predicts that, for any elliptic function f of degree 2,

0 = 2 × −2 + ∑ (mf (p) − 1)
p∈C/Λ

which in turn implies that there must be exactly four ramification points.

14.3 An algebraic relation

Although the we derived ℘′Λ from ℘Λ by an analytic procedure – differen-
tiation – it turns out that there is an algebraic relation between the two
functions.
Proposition 14.7. There are constants g1 , g2 ∈ C, depending only on Λ,
such that
(℘′Λ )2 ≡ 4℘3Λ − g2 ℘Λ − g3 .
Proof. Since ℘ is even, every term in its Laurent expansion about 0 has an
even exponent. By item (iii) of Remark 14.4 the constant term is 0, and so

℘(z) = 1/z 2 + az 2 + o(z 4 )

for some a ∈ C. Cubing this gives

(℘(z))3 = 1/z 6 + f (z)

where f is analytic in a neighbourhood of 0. Differentiating ℘ gives

℘′ (z) = −2/z 3 + 2az + o(z 3 )

and squaring this expression gives

(℘′ (z))2 = 4/z 6 − 8a/z 2 + g(z)

69
where g is another analytic function in a neighbourhood of 0. Therefore,

(℘′ (z))2 − 4(℘(z))3 = −8a/z 2 − h(z)

for another analytic function h. Setting g2 = 8a, we get that

(℘′ (z))2 − 4(℘(z))3 + g2 ℘(z)

is doubly periodic, with the only possible poles at Λ, but analytic in a neigh-
bourhood of 0. Hence it is equal to some constant −g3 and the result fol-
lows.
In fact, the constants g2 and g3 are related to the branch points e1 , e2
and e3 of ℘.
Remark 14.8. Recall that e1 , e2 and e3 are the images of the half-lattice
points under ℘, which are the zeroes of ℘′ = 0. Therefore, the cubic equation

4x3 − g2 x − g3 = 0

has three distinct zeroes, namely e1 , e2 and e3 , and so we can factor it:

4x3 − g2 x − g3 = 4(x − e1 )(x − e2 )(x − e3 ) .

Thus, the relation from Proposition 14.7 can be rewritten as follows.

(℘′ )2 ≡ 4(℘ − e1 )(℘ − e2 )(℘ − e3 )

Since the cubic equation had zero coefficient in x2 , this also tells us that
e1 + e2 + e3 = 0.

Lecture 15: More ℘-function and quotients

15.1 An elliptic curve
In this course, we have seen two constructions of complex tori. On the one
hand, we constructed them as quotients C/Λ. On the other hand, complex
tori have also arisen as compactifications of certain Riemann surfaces defined
by graphs, as in Example 10.1. It is reasonable to ask whether these con-
structions are connected in any way. The next result, remarkably, shows us
that every torus constructed as a quotient also arises as a compactification
of a graph.

70
Corollary 15.1. Let C/Λ be a complex torus. There are constants g2 , g3
such that C/Λ is biholomorphic to a one-point compactification of the graph
X ′ ∶= {(x, y) ∈ C2 ∣ y 2 = 4x3 − g2 x − g3 } .
Sketch proof. Take g2 , g3 as in Proposition 14.7. As in previous examples,
such as Question 14 of Example Sheet 1 or Example 10.1, the x- and y-
coordinate projections turn out to define an atlas on X ′ . Furthermore, as
in Example 10.1 there is a compactification X ′ ⊆ X with a single point at
infinity, and the coordinate projections extend to meromorphic functions.
Because the conformal structure on X includes the coordinate projections,
the map F ∶ C → X defined by
F (z) ∶= (℘(z), ℘′ (z))
is analytic. Since Λ is the group of periods of ℘ and ℘′ , F descends to a
well-defined analytic map
Φ ∶ C/Λ → X ,
sending the coset 0 + Λ to ∞.
It remains to show that Φ is a biholomorphism. As a non-constant ana-
lytic map of compact Riemann surfaces, Φ is certainly surjective.
To show injectivity, take a period parallelogram P centred on 0, and
suppose that z and z ′ are P and that F (z) = F (z ′ ). Then
℘(z) = ℘(z ′ )
which implies that z ′ = ±z, because ℘ is even and of degree 2. But then
℘′ (z) = ℘′ (±z) = ±℘′ (z)
since ℘′ is odd. As long as z ∉ Λ/2, we have that ℘′ (z) ∈ C∗ and so this
implies that z ′ = z.
In summary, away from the ramification points of ℘, the map Φ has
degree 1 on C/Λ. Therefore, deg(Φ) = 1 by the valency theorem, and so Φ is
a conformal equivalence.

15.2 Classification of elliptic functions

Finally, let’s see that we can in fact classify all meromorphic functions on a
complex torus, and indeed they can all be written in terms of ℘. The result
is phrased equivalently in terms of elliptic functions with periods Λ.

71
Theorem 15.2. Let f be an elliptic function with periods Λ. There exist
rational functions Q1 , Q2 such that

f (z) = Q1 (℘(z)) + Q2 (℘(z))℘′ (z) .

Furthermore, if f is even, then we can take Q2 = 0.

Proof. First, assume f is even. Next, let’s observe that we may assume that
f only has simple zeroes and poles.
Indeed, let B1 be the set of branch points of f . This is equal to the set of
branch points of the induced function f¯ on the complex torus C/Λ, which is
finite. Likewise, let B2 be the set of branch points of ℘, which is also finite.
Therefore, we may choose distinct c, d ∈ C ∖ (B1 ∪ B2 ). Now the function
f (z) − c
f (z) − d
is still even, neither its zeroes nor its poles are ramification points of f or ℘.
So we can and will assume that f has this property.
Since f is even, its zeroes are of the form

±a1 , . . . , ±an

and likewise its poles are of the form

±b1 , . . . , ±bn

(where 2n = deg(f )). Therefore, the function

(℘(z) − ℘(a1 ))⋯(℘(z) − ℘(an ))
g(z) =
(℘(z) − ℘(b1 ))⋯(℘(z) − ℘(bn ))
is also an elliptic function with periods Λ, and with the same zeroes and
poles as f . Therefore the function f (z)/g(z) is a non-zero elliptic function
with all singularities removable, and hence is equal to some constant c by
Corollary 13.10. Therefore
f (z) = cg(z) ,
which is a rational function as required.
If f is odd, then f (z)/℘′ (z) is an even elliptic function, so equal to a
rational function of ℘(z) by the above. Therefore f (z) = Q2 (℘(z))℘′ (z) as
required.

72
 Finally, an arbitrary function f can be written as as sum of its even and
odd parts,
f (z) + f (−z) f (z) − f (−z)
f (z) = ( )+( ),
2 2
so the general result follows.

15.3 Quotients of Riemann surfaces

Hopefully it is now clear that we were able to learn a great deal about complex
tori C/Λ by thinking of them as quotients of C. We would like to be able to
think about all Riemann surfaces that way, but in order to make a reasonable
statement, we need to make some definitions.
Definition 15.3. Let a group G act by homeomorphisms on a space X. The
action is said to be properly discontinuous if, for every compact K ⊆ X, the
set
{g ∈ G ∣ g(K) ∩ K ≠ ∅}
is finite. If, for each x ∈ X, the stabiliser StabG (x) is trivial, then the action
is said to be free.
This definition generalises the action of a lattice Λ on C.
Example 15.4. If Λ is a lattice in C then the action of Λ on C by translation
is properly discontinuous and free.
These slightly awkward-looking conditions turn out to be exactly what is
needed to show that the quotient space is Hausdorff and the quotient map is
a covering map.
Lemma 15.5. Let G be a group acting freely and properly discontinuously
by homeomorphisms on a Riemann surface R. The quotient space G/R is
Hausdorff and the quotient map π ∶ R → G/R is a regular covering map.
Proof. To prove that G/R is Hausdorff, let p, q ∈ R with π(p) ≠ π(q). By
taking charts about p and q, and using small discs in C, we may find disjoint
open sets U ∋ p and V ∋ q such that the closures U and V are compact.
Therefore, setting K = U ∩ V in the definition of proper discontinuity, the
set of g ∈ G such that g(U ) ∩ V ≠ ∅ is a finite set {g1 , . . . , gn }. Because R
is Hausdorff, for each gi there are disjoint open neighbourhoods Ui ∋ x and
Vi ∋ gi .y. Now
U ′ = U ∩ ⋂ Ui
i

73
and
V ′ = V ∩ ⋂ gi−1 (Vi )
i

are neighbourhoods of p and q respectively, with the property that G.U ′ is

disjoint from G.V ′ . Therefore, π(U ′ ) and π(V ′ ) are the required disjoint
open neighbourhoods of π(p) and π(q).
The argument that π is a covering map is similar. For p ∈ R, as above take
an open neighbourhood U ∋ p such that the closure U is compact. Setting
K = U in the definition of proper discontinuity, the set of g ∈ G such that
g(U ) ∩ U ≠ ∅ is a finite set {1, g1 , . . . , gn }, with each gi ≠ 1. Since the action
is free, gi .x ≠ x for each i. Therefore, for each i, there are disjoint open
neighbourhoods Ui ∋ x and Vi ∋ gi x. Now

U ′ = U ∩ ⋂(Ui ∩ gi−1 Vi )
i

has the property that g(U ′ ) ∩ U ′ = ∅ for each g ∈ G ∖ 1. This can be restated
as
π −1 (π(U ′ )) = ∐ g(U ′ ) ,
g∈G

which shows that π is a regular covering map.

Just as in the case of lattices, free and properly discontinuous actions are
a convenient way of constructing conformal structures on the quotients.
Proposition 15.6. Let R be a Riemann surface, and let G be a group acting
freely and properly discontinuously by conformal equivalences on R. Then
the quotient S = G/R is a Riemann surface, and the quotient map

π∶R→S

is analytic and a regular covering map.

Proof. Since R is connected, its continuous image S must be too. By Lemma
15.5, S is Hausdorff and π is a regular covering map. Finally, the construction
of the conformal structure on S is exactly the same as in Example 5.1, where
the result was proved for complex tori.
As an example of the application of these ideas, we will proof a simple
case of a famous theorem of Hurwitz about the automorphisms of Riemann
surfaces.

74
Theorem 15.7 (Hurwitz theorem). Let R be a compact Riemann surface
of genus gR ≥ 2, and suppose that a group G acts freely and properly dis-
continuously on R by conformal equivalences. Then G is finite, and indeed
∣G∣ ≤ gR − 1.

Proof. By Proposition 15.6, the quotient S = G/R is a Riemann surface, and

the quotient map π ∶ R → S is an analytic regular covering map; in particular,
it is unramified. By construction, deg(π) = ∣G∣. If gS is the genus of S then
the Riemann–Hurwitz theorem gives

2gR − 2 = ∣G∣(2gS − 2)

so, since the left-hand side is positive, gS ≥ 2 and so

∣G∣ ≤ ∣G∣(gS − 1) = gR − 1

as required.
Finally, notice that the result fails if gR = 1.
Example 15.8. The complex torus C/Λ is an infinite group, acting faithfully
on itself by conformal equivalences. Any finite subgroup acts freely and
properly discontinuously, and there are arbitrarily large finite subgroups.

Lecture 16: Uniformisation and its consequences

16.1 The uniformisation theorem
In this final lecture, we state without proof some theorems that classify all
Riemann surfaces, and then deduce some consequences. The first, and most
difficult, task is to address the simply connected case.

Theorem 16.1 (Uniformisation theorem). Every simply connected Riemann

surface is conformally equivalent to one of:

(i) the Riemann sphere, C∞ ;

(ii) the complex plane C; or

(iii) the unit disc D.

75
Remark 16.2. The three Riemann surfaces listed are not conformally equiv-
alent to each other. Indeed, C∞ is compact, so not even homeomorphic to
the other two. The complex plane C and the unit disc D are homeomorphic,
but any analytic map C → D is constant by Liouville’s theorem, so they are
not conformally equivalent.
The proof is beyond the scope of this course. With the uniformisation
theorem in hand, we can make progress on classifying all Riemann surfaces.
Let’s start with the case of surfaces of genus 0.
Corollary 16.3. The conformal structure on S 2 is conformally equivalent to
C∞ .
Proof. Let R be a Riemann surface defined by a conformal structure on S 2 .
By Theorem 16.1, R is conformally equivalent to one of C∞ , C or D; in
particular, it is homeomorphic to one of these. But only C∞ is compact.
For Riemann surfaces of positive genus, we need to invoke the tools of
algebraic topology.
Theorem 16.4. Every Riemann surface R has a regular covering map π ∶
̃ → R such that R
R ̃ is simply connected. Furthermore, there is a group G
̃
acting freely and properly discontinuously by conformal equivalences on R,
and the covering map π descends to a conformal equivalence G/R ̃ ≅ R.

Sketch proof. The existence of a simply connected regular covering space R̃

(the universal cover of R) is proved in Part II Algebraic Topology; further-
more, the results of that course also show that the fundamental group G acts
freely and properly discontinuously on R, ̃ so that π descends to a homeo-
morphism G/R ̃ ≅ R. By Lemma 4.1, there is a unique conformal structure
on R̃ making π into an analytic map. In the standard local coordinates on
̃
R each g ∈ G just acts as the identity, so G does indeed act by conformal
equivalences. The induced homeomorphism G/R ̃ → R is now an analytic
map of degree one, and hence a conformal equivalence.
Combining Theorem 16.4 with the uniformisation theorem now gives a
classification of all Riemann surfaces as quotients.
Corollary 16.5. Every Riemann surface R is conformally equivalent to a
quotient
R ≅ G/R̃,

76
 ̃ is one of C∞ , C or D, and G is a properly discontinuous group of
where R
̃
conformal equivalences of R.
̃
We say that R is uniformised by R.
Remark 16.6. From the point of view of this course, the group G can be
defined as the group
̃ ∣ π ○ φ = π} ,
G = {φ ∈ Aut(R)
̃ is the group of conformal equivalences R
where Aut(R) ̃ → R.
̃

16.2 Classification of Riemann surfaces

Corollary 16.5 suggests a strategy for classifying Riemann surfaces R, by
studying properly discontinuous groups of conformal equivalences of R ̃ =
̃ The simplest
C∞ , C, D. It divides naturally into three cases, depending on R.
̃ = C∞ .
case is R

Proposition 16.7. If a Riemann surface R is uniformised by C∞ then R is

conformally equivalent to C∞ .

Proof. By hypothesis, R = G/C∞ for some G acting properly discontinuously

by conformal equivalences. In Question 7 of Example Sheet 1, you showed
that the group of conformal equivalences of C∞ is the group of Möbius trans-
formations P SL2 (C). But every Möbius transformation g ∈ P SL2 (C) has
at least one fixed-point on C∞ (by the fundamental theorem of algebra, for
instance). Therefore, no non-trivial subgroup of P SL2 (C) can act freely on
C∞ . So G = 1 and R ≅ C∞ .
The proof of Proposition 16.7 relies on understanding the properly dis-
continuous subgroups of the group of conformal automorphisms of R. ̃ We
have a similarly good understanding of this group for C, and we can use this
classify the Riemann surfaces uniformised by C.

Proposition 16.8. If a Riemann surface R is uniformised by C then one of

the following holds:

(i) G = 1 and R ≅ C;

(ii) G ≅ Z and R ≅ C∗ ;

77
(iii) G ≅ Z2 and R ≅ C/Λ for some lattice Λ.

Proof. In Question 7 of Example Sheet 1 (again), you showed that the group
of conformal equivalences of C is the group

C ⋊ C∗ ≅ {z ↦ az + b ∣ a ∈ C∗ , b ∈ C} .

Again, since the action is free, non-trivial elements of G cannot have fixed
points. But
z = az + b
has a solution in C unless a = 1, so G consists entirely of translations

z ↦ z + b.

In Question 1 of Example Sheet 3, you prove that such a group G must

be either 1 (the trivial subgroup), or be isomorphic to Z and generated by
translation by some non-zero ω, or be isomorphic to Z2 and generated by a
pair of translations by a pair of non-zero numbers ω1 , ω2 with ω2 /ω1 ≅ C2 ,
The first case is the same as item (i) of the proposition, the second case
corresponds to item (ii) by Proposition 13.7, and the final case immediately
corresponds to item (iii).
In summary, the Riemann surfaces uniformised by C∞ and by C are
some of our favourite simple invariants that we have seen in this course!
Everything else is uniformised by D. To see this, we notice that the three
different uniformising spaces are mutually exclusive. This in turn relies on
the following important fact.

Lemma 16.9. Let f ∶ R → S be an analytic map of Riemann surfaces.

Suppose that R is simply connected, and let π ∶ S̃ → S be the uniformising
map of S. Then there is an analytic map F ∶ R → S̃ such that f = π ○ F .

Proof. See Question 3 on Example Sheet 2, where you proved the result for
tori. The general proof is the same.
Using this, we can easily show that the different uniformising surfaces are
mutually exclusive.

Proposition 16.10. A Riemann surface R is uniformised by at most one of

C∞ , C and D.

78
Proof. Propositions 16.7 and 16.8 together show that no surface is uni-
formised by both C∞ and C. Suppose therefore that R is uniformised by
both D and R̃ = C or C∞ , so we have uniformising maps π ∶ D → R and
̃ → R. By Lemma 16.9, f lifts to an analytic map F ∶ S → D. But now
f ∶R
F is a bounded analytic function on C or C∞ , hence constant by Liouville’s
theorem. Therefore f = π ○ F is also constant, which is a contradiction.
In particular, any Riemann surface not conformally equivalent to one of
the surfaces mentioned in Propositions 16.7 or 16.8 must be uniformised by
D.
Any attempt to classify the Riemann surfaces uniformised by D must
rely on understanding the group of conformal self-equivalences of D. The
following result was proved in IB Complex Analysis.

Proposition 16.11. The group of conformal equivalences of D is the group

of Möbius transformations
z−a
{z ↦ eiθ ∣ a ∈ C, θ ∈ R} ,
1 − āz
acting on D in the natural way.

Recall also that this group action is more easily understood if we use the
Möbius transformation
1 + iz
µ∶z↦
1 − iz
to conjugate D to the upper half-plane H ∶= {z ∈ C ∣ Im, z > 0}. The conformal
equivalences of the upper half-plane are just the group P SL2 (R) of Möbius
transformations with real coefficients.
Remark 16.12. There is a connection here with IB Geometry: both D and
H provide models of the hyperbolic plane, and in both cases the group of
conformal equivalences is equal to the group of orientation-preserving isome-
tries.

Definition 16.13. A subgroup of P SL2 (R) that acts properly discontinu-

ously on H is called a Fuchsian group.

In summary, the problem of classifying the Riemann surfaces uniformised

by D involves first classifying Fuchsian groups, and then understanding all
their free and properly discontinuous actions.

79
16.3 Consequences of uniformisation
Finally, we will give a few consequences of uniformisation.
Corollary 16.14. If R is a compact Riemann surface and the genus of R is
at least 2 then R is uniformised by D.
Proof. The compact Riemann surfaces uniformised by C∞ or C have genus
at most 1. Therefore, by Proposition 16.10, R is uniformised by D .
Even in the case of domains in C, the uniformisation theorem is non-
trivial. This instance of it is called the Riemann mapping theorem.
Corollary 16.15 (Riemann mapping theorem). If D ⊊ C is a simply con-
nected, proper subdomain of C, then D is conformally equivalent to D.
Proof. By Theorem 16.1, it suffices to prove that D is not conformally equiv-
alent to C∞ or C. The case of C∞ is clear, because D is not compact whereas
C∞ is.
The case of C is similar to part of Question 7 from Example Sheet 1.
Suppose f ∶ C → D be a conformal equivalence. If the singularity at ∞ is
essential then, by the Casorati–Weierstrass theorem, f ({∣z∣ > 1}) is dense. On
the other hand, f (D) is open by the open mapping theorem, so f (D)∩f ({∣z∣ >
1}) ≠ ∅, which contradicts the fact that f injective. Therefore, ∞ is a
removable singularity or a pole, so f extends to an analytic map
f¯ ∶ C∞ → C∞ .
But f¯ is non-constant and hence surjective, so C∞ = D ∪ {f¯(∞)}. This
contradicts the claim that D is a proper subdomain of C.
We finish where we started the course – with complex analysis – by giving
a proof of Picard’s theorem.
Corollary 16.16 (Picard’s theorem). Any analytic function
f ∶ C → C ∖ {0, 1}
is constant.
Proof. In Question 9 of Example Sheet 3, you will prove that C ∖ {0, 1}
is uniformised by D. Therefore, by Lemma 16.9, there is an analytic map
F ∶ C → D with f = π ○ F , where π is the uniformising map D → C ∖ {0, 1}.
But F is constant by Liouville’s theorem, so f is too.

80