Introduction to Teichmüller theory

Lecture notes
Bram Petri

Version: February 11, 2025

 Contents

Preface 5
Lecture 1. Reminder on surfaces 7
1.1. Preliminaries on surface topology 7
1.2. Riemann surfaces 9
1.3. The uniformization theorem and automorphism groups 13
1.4. Quotients of the three simply connected Riemann surfaces 14
Lecture 2. Quotients, metrics, conformal structures 17
2.1. More on quotients 17
2.2. Riemannian metrics and Riemann surfaces 21
2.3. Conformal structures 23
Lecture 3. The Teichmüller space of the torus 27
3.1. Riemann surface structures on the torus 27
3.2. The Teichmüller and moduli spaces of tori 28
3.3. T 1 as a space of marked structures 30
3.4. Markings by diffeomorphisms 32
3.5. The Teichmüller space of Riemann surfaces of a given type 34
Lecture 4. Markings, mapping class groups and moduli spaces 35
4.1. Teichmüller space in terms of markings 35
4.2. The mapping class group 38
4.3. Moduli space 38
4.4. Elements and examples of mapping class groups 39
Lecture 5. Beltrami differentials and a topology on Teichmüller space 45
5.1. Wirtinger derivatives 45
5.2. Beltrami coefficients 45
5.3. Quasiconformal mappings 49
5.4. The Teichmüller metric 51
5.5. Grötschz’s theorem 52
Lecture 6. Measured foliations and quadratic differentials 55
6.1. Measured foliations 55
6.2. Holomorphic quadratic differentials 59
Lecture 7. Quadratic differentials, Teichmüller mappings, hyperbolic geometry 63
7.1. Quadratic differentials 63
7.2. Teichmüller mappings 64
3
4 CONTENTS

7.3. Hyperbolic surfaces 65

7.4. The universal cover of a hyperbolic surface with boundary 70
7.5. Pairs of pants and gluing 71
Lecture 8. Pants decompositions and annuli 73
8.1. Non-compact pairs of pants 73
8.2. The geometry of isometries 74
8.3. Geodesics and conjugacy classes 75
8.4. Every hyperbolic surface admits a pants decomposition 77
8.5. Annuli 78
Lecture 9. Fenchel–Nielsen coordinates 81
9.1. More on annuli 81
9.2. Fenchel-Nielsen coordinates 82
Lecture 10. Teichmüller’s theorem and the Nielsen–Thurston classification 87
10.1. Teichmüller’s theorem 87
10.2. The Nielsen–Thurston classification on the torus 90
10.3. Definitions and statement of the classification theorem 90
10.4. An overview of the proof 91
10.5. The space of measured foliations 92
10.6. Compactifying Teichmüller space 94
Lecture 11. Landmark results, part I 95
11.1. Compactifying Teichmüller space, part II 95
11.2. Classifying mapping classes 97
Lecture 12. Landmark results, part II 101
12.1. The Nielsen realization problem 101
12.2. Thurston’s hyperbolization of mapping tori 103
12.3. The number of short simple closed geodesics 106
Bibliography 113
 Preface

These are the lecture notes for a course called Introduction to Teichmüller theory, taught
in January and February 2024 and 2025 in the master’s program M2 de Mathématiques
fondamentales at Sorbonne University.
There are many references on various aspects of Teichmüller theory, like [IT92, Bus10,
GL00, Zor06, Hub06, FM12, Baa21, Wri15]. All of these treat a lot more material
than what we will have time for in the course, whence the present notes. Most of the
material presented here is adapted from these references.

5
 LECTURE 1

Reminder on surfaces

The Teichmüller space of a surface S is the deformation space of complex structures on S

and can also be seen as a space of hyperbolic metrics on S. The aim of this course will
be to study the geometry and topology of this space and its quotient: the moduli space of
Riemann surfaces.
Before we get to any of this, we need to talk about surfaces themselves. So, today we will
recall some of the basics on surfaces.

1.1. Preliminaries on surface topology

1.1.1. Examples and classification. A surface is a smooth two-dimensional manifold.
We call a surface closed if it is compact and has no boundary. A surface is said to be of
finite type if it can be obtained from a closed surface by removing a finite number of points
and (smooth) open disks with disjoint closures. In what follows, we will always assume
our surfaces to be orientable.

Example 1.1.1. To properly define a manifold, one needs to not only describe the set but
also give smooth charts. In what follows we will content ourselves with the sets.
(a) The 2-sphere is the surface
S2 = (x, y, z) ∈ R3 : x2 + y 2 + z 2 = 1 .


(b) Let S1 denote the circle. The 2-torus is the surface

T2 = S1 × S1

(c) Given two (oriented) surfaces S1 , S2 , their connected sum S1 #S2 is defined as
follows. Take two closed sets D1 ⊂ S1 and D2 ⊂ S2 that are both diffeomorphic
to closed disks, via diffeomorphisms
φi : (x, y) ∈ R2 : x2 + y 2 ≤ 1 → Di , i = 1, 2,


so that φ1 is orientation preserving and φ2 is orientation reversing.

Then  
S1 #S2 = S1 ∖ D̊1 ⊔ S2 ∖ D̊2 / ∼
where D̊i denotes the interior of Di for i = 1, 2 and the equivalence relation ∼ is
defined by
φ1 (x, y) ∼ φ2 (x, y) for all (x, y) ∈ R2 with x2 + y 2 = 1.
The figure below gives an example.
7
8 1. REMINDER ON SURFACES

Figure 1. A connected sum of two tori.

Like our notation suggests, the manifold S1 #S2 is independent (up to diffeomor-
phism) of the choices we make (the disks and diffeomorphisms φi ). This is a
non-trivial statement, the proof of which we will skip. Likewise, we will also not
prove that the connected sum of surfaces is an associative operation and that
S2 #S is diffeomorphic to S for all surfaces S.

A classical result from the 19th century tells us that the three simple examples above are
enough to understand all finite type surfaces up to diffeomorphism.

Theorem 1.1.2 (Classification of closed surfaces). Every closed orientable surface is dif-
feomorphic to the connected sum of a 2-sphere with a finite number of tori.

Indeed, because the diffeormorphism type of a finite type surface does not depend on where
we remove the points and open disks (another claim we will not prove), the theorem above
tells us that an orientable finite type surface is (up to diffeomorphism) determined by a
triple of positive integers (g, b, n), where
- g is the number of tori in the connected sum and is called the genus of the surface.
- b is the number of disks removed and is called the number of boundary components
of the surface.
- n is the number of points removed and is called the number of punctures of the
surface.

Definition 1.1.3. The triple (g, b, n) defined above will be called the signature of the
surface. We will denote the corresponding surface by Σg,b,n and will write Σg = Σg,0,0 .

1.1.2. Euler characteristic. The Euler characteristic is a useful topological invariant

of a surface. There are multiple ways to define it. We will use triangulations. A tri-
angulation T = (V, E, F ) of a closed surface S will be the data of a finite set of points
V = {v1 , . . . .vk } ⊂ S (called vertices), a finite set of arcs E = {e1 , . . . , el } with endpoints
in the vertices (called edges) so that the complement S ∖ (∪vi ∪ ej ) consists of a collection
of disks F = {f1 , . . . , fm } (called faces) that all connect to exactly 3 edges.
Note that a triangulation T here is a slightly more general notion than that of a simplicial
complex (it’s an example of what Hatcher calls a ∆-complex [Hat02, Page 102]). Figure 2
below gives an example of a triangulation of a torus that is not a simplicial complex.
 1.2. RIEMANN SURFACES 9

Figure 2. A torus with a triangulation

Definition 1.1.4. S be a closed surface with a triangulation T = (V, E, F ). The Euler

characteristic of S is given by
χ(S) = |V | − |E| + |F | .

Because χ(S) can be defined entirely in terms of singular homology (see [Hat02, Theorem
2.4] for details), it is a homotopy invariant. In particular this implies it should only depend
on the genus of our surface S. Indeed, we have

Lemma 1.1.5. Let S be a closed connected and oriented surface of genus g. We have
χ(S) = 2 − 2g.

Proof. Exercise: prove this using your favorite triangulation. □

For surfaces that are not closed, we can define

χ(Σg,b,n ) = 2 − 2g − b − n.
This can be computed with a triangulation as well. For surfaces with only boundary
components, the usual definition still works. For surfaces with punctures there no longer
is a finite triangulation, so the definition above no longer makes sense. There are multiple
ways out. The most natural is to use the homological definition, which gives the formula
above. Another option is to allow some vertices to be missing, that is, to allow edges to
run between vertices and punctures. Both give the formula above.

1.2. Riemann surfaces

For the basics on Riemann surfaces, we refer to the lecture notes from the course by Elisha
Falbel [Fal23] or any of the many books on them, like [Bea84, FK92]. For a text on
complex functions of a single variable, we refer to [SS03].

1.2.1. Definition and first examples. A Riemann surface is a one-dimensional complex

manifold. That is,

Definition 1.2.1. A Riemann surface X is a connected Hausdorff topological space X,

equipped with an open cover {Uα }α∈A of open sets and maps φα : Uα → C so that
(1) φα (Uα ) is open and φα is a homeomorphism onto its image.
10 1. REMINDER ON SURFACES

(2) For all α, β ∈ A so that Uα ∩ Uβ ̸= ∅ the map

φα ◦ (φβ )−1 : φβ (Uα ∩ Uβ ) → φα (Uα ∩ Uβ )
is holomorphic.
The pairs (Uα , φα ) are usually called charts and the collection ((Uα , φα ))α∈A is usually
called an atlas.

Note that we do not a priori assume a Riemann surface X to be a second countable space. It
is however a theorem by Radó that every Riemann surface is second countable (for a proof,
see [Hub06, Section 1.3]). Moreover every Riemann surface is automatically orientable
(see for instance [GH94, Page 18]).

Example 1.2.2. (a) The simplest example is of course X = C equipped with one
chart: the identity map.
(b) We set X = C∪{∞} = C b and give it the topology of the one point compactification
of C, which is homeomorphic to the sphere S2 . The charts are
U0 = C, φ0 (z) = z
and
U∞ = X \ {0}, φ∞ (z) = 1/z.

So U0 ∩ U∞ = C \ {0} and
φ0 ◦ (φ∞ )−1 (z) = 1/z for all z ∈ C \ {0}
which is indeed holomorphic on C \ {0}. C
b is usually called the Riemann sphere.

(c) C
b can also be identified with the projective line

.
P1 (C) = C2 \ {(0, 0} C∗ ,

where C∗ ↷ C2 \{(0, 0)} by λ·(z, w) = (λ·z, λ·w), for λ ∈ C∗ , (z, w) ∈ C2 \{(0, 0)}.
Indeed, we may equip P1 (C) with two charts
U0 = { [z : w] : w ̸= 0 } , φ0 ([z : w]) = z/w
and
U1 = { [z : w] : z ̸= 0 } , φ1 ([z : w]) = w/z.
The map

̸ 0
z/w if w =
[z : w] 7→
∞ if w = 0
then defines a biholomorphism P1 (C) → C.
b

(d) Recall that a domain D ⊂ C b is any connected and open set in C.

b Any such
domain inherits the structure of a Riemann surface from C.
b
 1.2. RIEMANN SURFACES 11

1.2.2. Automorphisms. To get a larger set of examples, we will consider quotients.

First of all, we need the notion of a holomorphic map:

Definition 1.2.3. Let X and Y be Riemann surfaces, equipped with atlasses {(Uα , φα )}α∈A
and {(Vβ , ψβ )}β∈B respectively. A function f : X → Y is called holomorphic if
ψβ ◦ f ◦ φ−1
α : φα (Uα ∩ f
−1
(Vβ )) → ψβ (f (Uα ) ∩ Vβ )
is holomorphic for all α ∈ A, β ∈ B so that f (Uα ) ∩ Vβ ̸= ∅. A bijective holomorphism is
called a biholomorphism or conformal. Aut(X) will denote the automorphism group of X,
the set of biholomorphisms X → X.

The automorphism group of the Riemann sphere is

.  
1 λ 0
Aut(P (C)) = PGL(2, C) = GL(2, C) : λ ̸= 0 .
0 λ
It acts on P1 (C) through the projectivization of the linear action of GL(2, C) on C2 \{(0, 0)}.
We can also descibe the action on C. b We have:
   az+b
a b if z ̸= −d/c
(1.2.1) ·z = cz+d
c d ∞ if z = −d/c
and    a
a b if c ̸= 0
·∞= c
c d ∞ if c = 0.
These maps are called Möbius transformations.
Finally, we observe that
  .  
a b 1 0
PGL(2, C) ≃ PSL(2, C) = : a, b, c, d ∈ C, ad − bc = 1 ±
c d 0 1

1.2.3. Quotients. Many subgroups of Aut(P1 (C)) give rise to Riemann surfaces:

Theorem 1.2.4. Let D ⊂ C

b be a domain and let G < PSL(2, C) such that

(1) g(D) = D for all g ∈ G

(2) If g ∈ G \ {e} then the fixed points of g lie ourside of D.
(3) For each compact subset K ⊂ D, the set
{ g ∈ G : g(K) ∩ K ̸= ∅ }
is finite.
Then the quotient space
D/G
has the structure of a Riemann surface.

A group that satisfies the second condition is said to act freely on D and a group that
satisfies the thirs condition is said to act properly discontinuously on D. The proof of this
theorem will be part of the exercises.
12 1. REMINDER ON SURFACES

1.2.4. Tori. The theorem from the previous section gives us a lot of new examples. The
first is that of tori. Consider the elements
   
1 1 1 τ
g1 := , gτ := ∈ PSL(2, C),
0 1 0 1

for some τ ∈ C with Im(τ ) > 0, acting on the domain C ⊂ C

b by

g1 (z) = z + 1 and gτ (z) = z + τ

for all z ∈ C.
We define the group
Λτ = ⟨g1 , gτ ⟩ < PSL(2, C).
A direct computation shows that
    
1 p + qτ 1 r + sτ 1 p + q + (r + s)τ
= ,
0 1 0 1 0 1
for all p, q, r, s ∈ Z, from which it follows that
  
1 n + mτ
Λτ = : m, n ∈ Z ≃ Z2 .
0 1

Let us consider the conditions from Theorem 1.2.4. (1) is trivially satisfied: Λτ preserves
C. Any non-trivial element in Λτ is of the form
 
1 n + mτ
0 1

and hence only has the point ∞ ∈ C

b as a fixed point, which gives us condition (2). To check
condition (3), suppose K ⊂ C is compact. Write dK = sup { |z − w| : z, w ∈ K } < ∞.
Given g ∈ Λτ , write
Tg = inf { |gz − z| : z ∈ C }
for the translation length of g. Note that Tg = |gz − z| for all z ∈ C (this is quite special
to quotients of C). We have
{ g ∈ Λτ : g(K) ∩ K ̸= ∅ } ⊂ { g ∈ Λτ : Tg ≤ 2dK }
and the latter is finite. So C/Λz is indeed a Riemann surface.
We claim that this is a torus. One way to see this is to note that the quotient map
π : C → C/Λτ restricted to the convex hull

F = conv({0, 1, τ, 1 + τ )
:= { λ1 + λ2 τ + λ3 (1 + τ ) : λ1 , λ1 , λ3 ∈ [0, 1], λ1 + λ2 + λ3 ≤ 1 }

is surjective. Figure 3 shows a picture of F. On F̊, π is also injective. So to understand

what the quotient looks like, we only need to understand what happens to the sides of F.
Since the quotient map identifies the left hand side of F with the right hand side and the
top with the bottom, the quotient is a torus.
 1.3. THE UNIFORMIZATION THEOREM AND AUTOMORPHISM GROUPS 13

τ 1+τ

0 1

Figure 3. A fundamental domain for the action Λτ ↷ C.

We can also prove that C/Λτ is a torus by using the fact that for all z ∈ C there exist
unique x, y ∈ R so that
z = x + yτ.
1 1
The map C/Λτ → S × S given by
[x + yτ ] 7→ (e2πix , e2πiy )
is a homeomorphism.
Note that we have not yet proven whether all these tori are distinct as Riemann surfaces.
But it will turn out later that many of them are.
1.2.5. Hyperbolic surfaces. Set H2 = { z ∈ C : Im(z) > 0 }, the upper half plane. It
turns out that the automorphism group of H2 is PSL(2, R). We will see a lot more about
this later during the course, but for now we will just note that there are many subgroups
of PSL(2, R) that satisfy the conditions of Theorem 1.2.4.
It also turns out that PSL(2, R) is exactly the group of orientation preserving isometries
of the metric
dx2 + dy 2
ds2 = .
y2
This is a complete metric of constant curvature −1. So, this means that all these Riemann
surfaces naturally come equipped with a complete metric of constant curvature −1. We will
prove some of these statements and treat a first example in the first problem sheet.

1.3. The uniformization theorem and automorphism groups

The Riemann mapping theorem tells us that any pair of simply connected domains in C
that are both not all of C are biholomorphic. In the early 20th century this was generalized
by Koebe and Poincaré to a classification of all simply connected Riemann surfaces:

Theorem 1.3.1 (Uniformization theorem). Let X be a simply connected Riemann surface.

Then X is biholomorphic to exactly one of
C,
b C or H2 .

Proof. See for instance [FK92, Chapter IV]. □

14 1. REMINDER ON SURFACES

This theorem implies that we can see obtain every Riemann surface as a quotient of one of
three Riemann surfaces. Before we formally state this, we record the following fact:

Proposition 1.3.2. • Aut(C)

b = PSL(2, C) acting by Möbius transformations,

• Aut(C) = { φ : z 7→ az + b : a ∈ C \ {0}, b ∈ C } ≃ C ⋊ C∗ ,
• Aut(H2 ) = PSL(2, R) acting by Möbius transformations.

Proof. See for instance [Bea84, Chapter 5] or [IT92, Section 2.3]. □

Note that in all three cases, we have

n o
Aut(X) = g ∈ Aut(C)
b : g(X) = X ,

that is, all the automorphisms of C and H2 extend to C.

b However, not all automorphisms
2
of H extend to C.

Corollary 1.3.3. Let X be a Riemann surface. Then there exists a group G < Aut(D),
b or H2 so that
where D is exactly one of C, C
• G acts freely and properly discontinuously on D and
• X = D/G as a Riemann surface.

Proof. Let X e denote the universal cover of X and π1 (X) its fundamental group.
The fact that X is a Riemann surface, implies that Xe can be given the structure of a
Riemann surface too, so that π1 (X) acts freely and properly discontinuously on X
e by
biholomorphisms (see for instance [IT92, Lemma 2.6]) and such that
X/π
e 1 (X) = X.
Since X b or H2 .
e is simply connected, it must be biholomorphic to exactly one of C, C □

1.4. Quotients of the three simply connected Riemann surfaces

Now that we know that we can obtain all Riemann surfaces as quotients of one of three
simply connected Riemann surfaces, we should start looking for interesting quotients.
1.4.1. Quotients of the Rieman sphere. It turns out that for the Riemann sphere
there are none:

Proposition 1.4.1. Let X be a Riemann surface. The universal cover of X is biholomor-

phic to C
b if and only if X is biholomorphic to C.
b

Proof. The “if” part is clear. For the “only if” part, note that every element in
PSL(2, C) has at least one fixed point on C
b (this either follows by direct computation or
from the fact that orientation-preserving self maps of the sphere have at least one fixed
point, by the Brouwer fixed point theorem [Mil65, Problem 6]). Since, by assumption
X = C/G,
b
 1.4. QUOTIENTS OF THE THREE SIMPLY CONNECTED RIEMANN SURFACES 15

where G acts properly discontinuously and freely, we must have G = {e}. □

1.4.2. Quotients of the plane. In Section 1.2.4, we have already seen that in the case
of the complex plane, the list of quotients is a lot more interesting: there are tori. This
however turns out to be almost everything:

Proposition 1.4.2. Let X be a Riemann surface. The universal cover of X is biholomor-

phic to C if and only if X is biholomorphic to either C, C \ {0} or
   
1 λ 1 µ
C/ ,
0 1 0 1
for some λ, µ ∈ C \ {0} that are linearly independent over R.

Proof. First suppose X = C/G. We claim that, since G acts properly discontinuously,
G is one of the following three forms:
(1) G = {e}
(2) G = ⟨φb ⟩, where φb (z) = z + b for some b ∈ C \ {0}
(3) G = ⟨φb1 , φb2 ⟩ where b1 , b2 ∈ C are independent over R.
To see this, we first prove that G cannot contain any automorphism z 7→ az + b for a ̸= 1.
Indeed, if a ̸= 1 then b/(1−a) is a fixed point for this map, which would contradict freeness
of the action. Moreover, since z 7→ z + b1 and z 7→ z + b2 commute for all b1 , b2 ∈ C, G is
a free abelian group and
G · z = { z + b : φb ∈ G } .
In particular, if G contains {z 7→ z + b1 , z 7→ z + b2 , z 7→ z + b3 } for b1 , b2 , b3 ∈ C that are
independent over Q, then spanZ (b1 , b2 , b3 ) is dense in C. This means that we can find a
sequence ((ki , li , mi ))i such that
φkb1i ◦ φlbi2 ◦ φm
b3 (z) → z
i
as i → ∞,
thus contradicting proper discontinuity. On a side note, we could have also used the
classification of surfaces (of potentially infinite type) in the last step: there is no surface
that has Zk for k ≥ 3 as a fundamental group.
We have already seen that the third case gives rise to tori. In the second case, the surface
is biholomorphic to C \ {0}. Indeed, the map
[z] ∈ C/⟨φb ⟩ 7→ e2πiz/b ∈ C \ {0}
is a biholomorphism.
Now let us prove the converse. For X = C the statement is clear. Likewise, for X = C\{0},
we have just seen that the composition
C → C/(z ∼ z + 1) ≃ C \ {0}
is the universal covering map. Finally, in the proposition, the tori are given as quotients
of C. □
 LECTURE 2

Quotients, metrics, conformal structures

2.1. More on quotients

2.1.1. Quotients of the complex plane, continued. We saw last time that any quo-
tient Riemann surface of C is either C, C − {0} or a torus. It turns out that moreover
every Riemann surface structure on the torus comes from the complex plane. We have seen
above that the universal cover cannot be the Riemann sphere, which means that (using
the uniformization theorem) all we need to prove is that it cannot be the upper half plane
either.
The fundamental group of the torus is isomorphic to Z2 , so what we need to prove is
that there is no subgroup of Aut(H2 ) = PSL(2, R) that acts properly discontinuously and
freely on H2 and is isomorphic to Z2 . We will state this as a lemma (in which we don’t
unnecessarily assume that the action is free, even if in our context that would suffice):

Lemma 2.1.1. Suppose G < PSL(2, R) acts properly on H2 and suppose furthermore that
G is abelian. Then either G ≃ Z or G is finite and cyclic.

Proof. We will use the classification of isometries of H2 that we shall prove in the
exercises: an element g ∈ PSL(2, R) has either
• a single fixed point in H2 , in which case it’s called elliptic and can be conjugated
into SO(2)
• a single fixed point
  on R ∪
 {∞}, inwhich case it’s called parabolic and can be
1 t
conjugated into : t∈R
0 1
• or two fixed points on R∪{∞},
 in whichcase it’s called
 hyperbolic (or loxodromic)
λ t
and can be conjugated into : λ>0 .
0 λ1
If g1 , g2 ∈ PSL(2, R) commute and p ∈ H2 ∪ R ∪ {∞} is a fixed point of g1 , then
g1 (g2 (p)) = g2 ◦ g1 (p) = g2 (p).
That is, g2 (p) is also a fixed point of g1 .
So if G contains an elliptic element g, then all other g ′ ∈ G \ {e} are elliptic as well, with
the same fixed point. Moreover, by proper discontinuity (and compactness of SO(2)), the
angles of rotation of all elements in G must be rationally related rational multiples of π.
This means that G is a finite cyclic group.
17
18 2. QUOTIENTS, METRICS, CONFORMAL STRUCTURES

Now suppose G contains a parabolic element g. Then all other g ′ ∈ G \ {e} are parabolic
as well, with the same fixed point (which we may assume to be ∞). If
   
1 t1 1 t2
, ∈G
0 1 0 1
for some t1 , t2 ∈ R that are not rationally related, then G is not discrete, which contradicts
proper discontinuity (see the exercises). So G ≃ Z.
The argument in the hyperbolic case is essentially the same as in the parabolic case. □

Combining this lemma with the uniformization theorem, we obtain:

Corollary 2.1.2. Let X be a Riemann surface that is diffeomorphic to a torus. Then the
universal cover of X is biholomorphic to C.

2.1.2. Quotients of the upper half plane. It will turn out that the richest family of
Riemann surfaces is that of quotients of H2 . Indeed, looking at the clasification of closed
orientable surfaces, we note that we have so far only seen the sphere and the torus. It turns
out that all the other closed orientable surfaces also admit the structure of a Riemann
surface. In fact, they all admit lots of different such structures. The two propositions
above imply that they must all arise as quotients of H2 .
We will not yet discuss how to construct all these surfaces but instead discuss an exam-
ple (partially taken from [GGD12, Example 1.7]). Fix some distinct complex numbers
a1 , . . . , a2g+1 and consider the following subset of C2 :

X̊ = (z, w) ∈ C2 : w2 = (z − a1 )(z − a2 ) · · · (z − a2g+1 ) .



Let X denote the one point compactification of X̊ obtained by adjoining the point (∞, ∞).
As opposed to charts, we will describe inverse charts, or parametrizations around every
p ∈ X̊:
• Suppose p = (z0 , w0 ) ∈ X̊ is so that z0 ̸= ai for all i = 1, . . . , 2g + 1. Set
ε := min {|z0 − ai | /2}
i=1,...,2g+1

Then define the map φ−1 : { ζ ∈ C : |ζ| < ε } → X̊ by

 q 
−1
φ (ζ) = ζ + z0 , (ζ + z0 − a1 ) · · · (ζ + z0 − a2g+1 ) ,

where the branch of the square root is chosen so that φ−1 (0) = (z0 , w0 ), gives a
parametrization.
• For p = (aj , 0), we set
q
ε := min{ |aj − ai | /2}
i̸=j
 2.1. MORE ON QUOTIENTS 19

Then define the map φ−1 : { ζ ∈ C : |ζ| < ε } → X̊ by

 
sY
φ−1 (ζ) = ζ 2 + aj , ζ (ζ 2 + aj − ai ) .
i̸=j

The reason that we need to take different charts around these points is that
p
z − aj
is not a well defined holomorphic function near z = aj .
Also note that the choice of the branch of the root does not matter. By changing
the branch we would obtain a new parametrization φ e−1 that satisfies φ
e−1 (ζ) =
−1
φ (−ζ).
It’s not hard to see that X̊ is not bounded as a subset of C2 . This means in particular that
it’s not compact. We can however compactify it in a similar fashion to how we compactified
C in order to obtain the Riemann sphere. That is, we add a point (∞, ∞) and around this
point define a parametrization:
(  p 
−2 −(2g+1) 2 ) · · · (1 − a 2)
ζ , ζ (1 − a ζ ζ if ζ ̸= 0
φ−1
∞ (ζ) =
1 2g+1
(∞, ∞) if ζ = 0,
for all ζ ∈ {|ζ| < ε} and some appropriate ε > 0.
The reason that the resulting surface X is compact is that we can write it as the union of
the sets
n o n o 
(z, w) ∈ X̊ : |z| ≤ 1/ε2 ∪ (z, w) ∈ X̊ : |z| ≥ 1/ε2 ∪ {(∞, ∞)} ,

for some small ε > 0. The first set is compact because it’s a bounded subset of C2 . The
second set is compact because it’s φ−1
∞ ({|ζ| ≤ ε}).

To see that X is connected, we could proceed using charts as well. We would have to find
a collection of charts that are all connected, overlap and cover X. However, it’s easier to
use complex analysis. Suppose z0 ̸= ai for all i = 1, . . . , a2g+1 and z0 ̸= ∞. In that case,
we can define a path
 v 
u2g+1
uY
z(t) 7→ z(t), t (z(t) − ai ) 
i=1

where z(t) is some continuous path in C between z0 and ai and we pick a continuous branch
of the square root, thus connecting any point (z0 , w0 ) ∈ X to (0, ai ).

To figure out the genus of X, note that there is a map π : X → C

b given by

π(z, w) = z for all (z, w) ∈ X.

This map is two-to-one almost everywhere. Only the points z = ai , i = 1, . . . , 2g + 1 and
the point z = ∞ have only one pre-image.
20 2. QUOTIENTS, METRICS, CONFORMAL STRUCTURES

Now triangulate C b so that the vertices of the triangulation coincide with the points
a1 , . . . , a2g+1 , ∞. If we lift the triangulation to X using π, we can compute the Euler char-
acteristic of X. Every face and every edge in the triangulation of C b has two pre-images,
whereas each vertex has only one. This means that:

b − (2g + 2) = 2 − 2g.
χ(X) = 2χ(C)

Because X is an orientable closed surface, we see that it must have genus g (Lemma 1.1.5).
In particular, if g ≥ 2, these surfaces are quotients of H2 . Note that this also implies that
for g ≥ 1, the Riemann surface X̊ is also a quotient of H2 . Note that we could have also
used the Riemann–Hurwitz formula for this calculation. Incidentally, this formula can be
proved using a similar argument to what we just did above.

To get a picture of what X looks like, draw a closed arc α1 between a1 and a2 on C,b an arc
α2 between a3 and a4 that does not intersect the first arc and so on, and so forth. The last
arc αg+1 goes between a2g+1 and ∞. Figure 1 shows a picture of what these arcs might
look like.
∞

a2g+1 a4
a3
a2
a1

b with some intervals removed.

Figure 1. C

Let
g+1
!
[
b\
D=C αi .
i=1

The map
π|π−1 (D) : π −1 (D) → D
is now a two-to-one map. Moreover on the arcs, it’s two-to-one on the interior and one-
to-one on the boundary. Because it’s also smooth, this means that the pre-image of the
arcs is a circle. So, X may be obtained (topologically) by cutting C
b open along the arcs,
taking two copies of that, and gluing these along their boundary. Figure 2 depicts this
process.
 2.2. RIEMANNIAN METRICS AND RIEMANN SURFACES 21

Figure 2. Gluing X out of two Riemann spheres.

Finally, we note that our Riemann surfaces come with an involution ı : X → X, given
by 
−w if w ̸= ∞
ı(w) =
∞ if w = ∞.
This map is called the hyperelliptic involution and the surfaces we described are hence
called hyperelliptic surfaces. Note that π : X → C
b is the quotient map X → X/ı.

2.2. Riemannian metrics and Riemann surfaces

We already noted that every Riemann surface comes with a natural Riemannian metric.
Indeed the Riemann sphere has the usual round metric of constant curvature +1. Likewise,
C has a flat metric, its usual Euclidean metric Aut(C) does not act by isometries. However,
in the proof of Proposition 1.4.2, we saw that all the quotients are obtained by quotienting
by a group that does act by Euclidean isometries. This means that the Euclidean metric
descends. Finally, we proved in the exercises that Aut(H2 ) also acts by isometries of the
hyperbolic metric defined in Section 1.2.5. So every quotient of H2 comes with a natural
metric of constant curvature −1.
It turns out that we can also go the other way around. That is: Riemann surface structures
on a given surface are in one-to-one correpsondence with complete metrics of constant
curvature.
One way to see this uses the Killing-Hopf theorem. In the special case of surfaces, this states
that every oriented surface equipped with a Riemannian metric of constant curvature +1, 0
or −1 can be obtained as the quotient by a group of orientation preserving isometries acting
properly discontinuously and freely on S2 equipped with the round metric, R2 equipped
with the Euclidean metric or H2 equipped with the hyperbolic metric respectively (see
[CE08, Theorem 1.37] for a proof). For a Riemannian manifold M , let us write
Isom+ (M ) = { φ : M → M : φ is an orientation preserving isometry } .
So, we need the fact that
(1) Isom+ (S2 ) = SO(2, R) and this has no non-trivial subgroups that act properly
discontinuously on S2 .
(2) Isom+ (R2 ) = SO(2, R) ⋉ R2 , where R2 acts by translations. The only subgroups
of this group that act properly discontinuously and freely are the fundamental
groups of tori and cylinders.
22 2. QUOTIENTS, METRICS, CONFORMAL STRUCTURES

(3) Isom+ (H2 ) = PSL(2, R).

Given the above, we get our one-to-one correspondence:

Proposition 2.2.1. Given an orientable surface Σ of finite type with ∂Σ = ∅, the identi-
fication described above gives a one-to-one correspondence of sets
 
  
 Complete Riemannian 
 ,
Riemann surface metrics of constant
 
∼ ↔ ∼,
structures on Σ 
 curvature {−1, 0, +1} 

 on Σ 

where the equivalence on the left is biholomorphism and the equivalence on the right is
isometry (and homothety in the Euclidean case).

Proof sketch. From the above we see that a Riemann surface structure on Σ yields
a metric of constant curvature and vice versa. We only need to check that biholomorphic
Riemann surfaces yield isometric/homothetic metrics and vice versa.
Suppose h : X → Y is a biholomorphism. We may lift this to a biholomorphism e h:X e → Ye
of the universal covers X
e and Ye of X and Y respectively. There are three cases to treat:
e ≃ Ye ≃ C, C, 2
X b H . Because it’s the most interesting case, we will treat the former, i.e.
Xe ≃ Ye ≃ C. We will also assume X and Y are tori. If we write

X ≃ C/Λ1 and Y ≃ C/Λ2 ,

h ∈ Aut(C) is such that e
then we get that e h(Λ1 ) = Λ2 . Since all automorphisms of C are of
∗
the form z 7→ az + b for a ∈ C and b ∈ C, Λ2 is obtained from Λ1 by translating, scaling
and rotating. This means that the quotient metrics are homothetic.
The proof of the reverse direction and both directions of all the remaining cases are similar.
□

Whether the curvature is 0, +1 or −1 is determined by the topology of Σ. This for instance

follows from the discussion above. It can also be seen from the Gauss-Bonnet theorem.
Recall that in the case of a closed Riemannian surface X, this states that
Z
K dA = 2π χ(Σ),
X

where K is the Gaussian curvature on X and dA the area measure. For constant curvature
κ, this means that
κ · area(X) = 2π χ(X)
So χ(X) = 0 if and only of κ = 0 and otherwise χ(X) needs to have the same sign as κ.
This last equality generalizes to finite type surfaces and we obtain:

Lemma 2.2.2. Let X be a hyperbolic surface homeomorphic to Σg,b,n then

area(X) = 2π(2g + n + b − 2).
 2.3. CONFORMAL STRUCTURES 23

2.3. Conformal structures

There is another type of structures on a surface that is in one-to-one correspondence with
Riemann surface structures, namely conformal structures.
We say that two Riemannian metrics ds21 and ds22 on a surface X are conformally equivalent
is there exists a positive function ρ : X → R+ so that
ds21 = ρ · ds22 .
So a conformal equivalence class of Riemannian metrics can be seen as a notion of angles
on the surface.
We have already seen that a Riemann surface structure induces a Riemannian metric on
the surface, so it certainly also induces a conformal class of metrics.
So, we need to explain how to go back. We will also only sketch this. First of all, suppose
we are given a surface X with oriented charts (Uj , (uj , vj ))j equipped with a Riemannian
metric that in all local coordinates (uj , vj ) is of the form
ds2 = ρ(uj , vj ) · (du2j + dvj2 ),
where ρ : X → R+ is some smooth function. Consider the complex-valued coordinate
wj = uj + i vj .
We claim that this is holomorphic. Indeed, applying a coordinate change on Uj ∩ Uk , we
have
"  2  2 !  2  2 !
∂u j ∂v j ∂u j ∂v j
ds2 = ρ(uk , vk ) · + du2k + + dvk2
∂uk ∂uk ∂vk ∂vk
  #
∂uj ∂uj ∂vj ∂vj
+2 + duk dvk .
∂uk ∂vk ∂uk ∂vk
Our assumption implies that
 2  2  2  2
∂uj ∂vj ∂uj ∂vj
(2.3.1) + = +
∂uk ∂uk ∂vk ∂vk
and
∂uj ∂uj ∂vj ∂vj
(2.3.2) + = 0.
∂uk ∂vk ∂uk ∂vk
This last line can be written as
 
∂uj /∂uk ∂vj /∂vk
det = 0.
−∂vj /∂uk ∂uj /∂vk
So this implies that    
∂uj /∂uk ∂vj /∂vk
=λ·
−∂vj /∂uk ∂uj /∂vk
for some λ ∈ R. Filling this into (2.3.1), we obtain
 2  2 !  2  2
2 ∂uj ∂vj ∂uj ∂vj
λ · + = + .
∂vk ∂vk ∂vk ∂vk
24 2. QUOTIENTS, METRICS, CONFORMAL STRUCTURES

So λ ∈ {±1}. Now using that our surface is oriented, i.e. that the determinant of the
Jacobian of the chart transition is positive, we obtain that λ = 1. And hence
∂uj ∂vj ∂vj ∂uj
= and − = ,
∂uk ∂vk ∂uk ∂vk
the Cauchy-Riemann equations for the chart transition wk ◦ wj−1 , which means that these
coordinates are indeed holomorphic. The coordinates (Uj , wj ) are usually called isothermal
coordinates.
Also note that we have not used the factor ρ, so any metric that is conformal to our metric
will give us the same structure. Moreover, our usual coordinate ‘z’ on the three simply
connected Riemann surfaces is an example of an isothermal coordinate, so if we apply the
procedure above to the metric we obtain from our quotients, we find the same complex
structure back.
This means that what we need to show is that for each Riemannian metric (that is not
necessarily given to us in the form above), we can find a set of coordinates so that our
metric takes this form. So, suppose our metric is given by
ds2 = A dx2 + 2B dx dy + C dy 2
in some local coordinates (x, y).
Writing z = x + iy, we get that
ds2 = λ |dz + µdz|2 := λ(dz + µdz)(dz + µdz),
where
1 √  A − C + 2i B
λ= A + C + 2 AC − B 2 and µ = √ .
4 A + C + 2 AC − B 2

We are looking for a coordinate w = u + iv so that

2 2
∂w ∂w/∂z
ds2 = ρ(du2 + dv 2 ) = ρ |dw|2 = ρ · · dz + dz ,
∂z ∂w/∂z
where
∂w ∂w ∂w ∂w ∂w ∂w
= −i and = +i
∂z ∂x ∂y ∂z ∂x ∂y
are called Wirtinger derivatives, we will discuss these in slightly more detail in Section
5.1.
This means that isothermal coordinates exist if there is a solution to the partial differential
equation
∂w ∂w
=µ· .
∂z ∂z
It turns out this solution does indeed exist on a surface, which means that we obtain a
Riemann surface structure. Moreover, it turns out this map is one-to-one. In particular,
holmorphic maps are conformal. So we obtain
 2.3. CONFORMAL STRUCTURES 25

Proposition 2.3.1. Given an orientable surface Σ of finite type with ∂Σ = ∅, the identi-
fication described above gives a one-to-one correspondence of sets
 ,
   Conformal classes 
Riemann surface
biholom. ↔ of Riemannian diffeomorphism.
structures on Σ  metrics on Σ 

Combined with Proposition 2.2.1, the proposition above also implies that in every confor-
mal class of metrics there is a metric of constant curvature that is unique (up to scaling
if the metric is flat). This can also be proved without passing through the uniformization
theorem, which comes down to solving a non-linear PDE on the surface. This was treated
in Olivier Biquard’s course Introduction à l’analyse géométrique.
 LECTURE 3

The Teichmüller space of the torus

3.1. Riemann surface structures on the torus

The goal of the rest of this course is to understand the deformation spaces associated to
Riemann surfaces: Teichmüller and moduli spaces.
In general, the Teichmüller space associated to a surface will be a space of marked Riemann
surface structures on that surface and the corresponding moduli space will be a space of
isomorphism classes of Riemann surface structures. As such, the moduli space associated
to a surface will be a quotient of the corresponding Teichmüller space.
First of all, note that the uniformization theorem tells us that there is only one Riemann
surface structure on the sphere. This means that the corresponding moduli space will be
a point. It turns out that the same holds for its Teichmüller space. This means that the
lowest genus closed surface for which we can expect an intersting deformation space is the
torus.
So, let us parametrize Riemann surface structures on the torus. Recall from Proposition
1.4.2 that every Riemann surface structure on the torus is of the form
   
1 λ 1 µ
C/ ,
0 1 0 1
for some λ, µ ∈ C \ {0} that are linearly independent over R.
First of all note that every such torus is biholomorphic to a torus of the form
Rτ := C/Λτ ,
for some τ ∈ H2 , where
   
1 1 1 τ
Λτ = , .
0 1 0 1
Indeed, rotating and rescaling the lattice induce biholomorphisms on the level of Riemann
surfaces (as we have already noted in the proof sketch of Proposition 2.2.1)
However, there are still distinct τ, τ ′ ∈ H2 that lead to holomorphic tori. We have:

Proposition 3.1.1. Let τ, τ ′ ∈ H2 . The two tori Rτ and Rτ ′ are biholomorphic if and
only if
aτ + b
τ′ =
cτ + d
for some a, b, c, d ∈ Z with ad − bc = 1.
27
28 3. THE TEICHMÜLLER SPACE OF THE TORUS

Proof. First assume Rτ and Rτ ′ are biholomorphic and let f : Rτ ′ → Rτ be a biholo-

morphism. Lift f to a biholomorphism fe : C → C. This means that
fe(z) = αz + β
for some α, β ∈ C. By postcomposing with a biholomorphism of C, we may assume that
fe(0) = 0.
Because fe is a lift, we know that both fe(1) and fe(τ ′ ) are equivalent to 0 under Λτ . So
fe(τ ′ ) = ατ ′ = aτ + b
fe(1) = α = cτ + d
for some a, b, c, d ∈ Z. So
aτ + b
τ′ = .
cτ + d
So we only need to show that ad − bc = 1. Moreover, since fe(Λτ ′ ) = Λτ , f (τ ′ ) = aτ + b
and f (1) = cτ + d generate Λτ . This means that the map
mτ + n 7→ m · (aτ + b) + n · (cτ + d)
 
a b
is an automorphism of Λ, and hence ∈ GL(2, Z). So, we obtain ad − bc = ±1.
c d
Since
ad − bc
Im(τ ′ ) = > 0,
|cτ + d|2
we get ad − bc = 1.
Conversely, if
aτ + b
τ′ =
cτ + d
Then
f ([z]) = [(cτ + d)z]
gives a biholomorphic map f : Rτ ′ → Rτ . □

3.2. The Teichmüller and moduli spaces of tori

Looking at Proposition 3.1.1, we see that we can parametrize all complex structures on
the torus with the set
M1 = SL(2, Z)\H2 = PSL(2, Z)\H2 .
Moreover this set is the quotient of the hyperbolic place by a group (PSL(2, Z)) of isometries
that acts properly discontinuously on it. However, the group doesn’t quite act freely, so
it’s not directly a hyperbolic surface.
So, let us investigate the structure of this quotient. One way of doing this is to find a
fundamental domain for the action of PSL(2, Z) on H2 . Set
 
2 1 1
F = z ∈ H : |z| ≥ 1 and − ≤ Re(z) ≤ .
2 2
Figure 1 shows a picture of F.
 3.2. THE TEICHMÜLLER AND MODULI SPACES OF TORI 29

H2 F

√ √
−1+ 3i 1+ 3i
2 i 2

− 12 1
2
Figure 1. A fundamental domain for the action of PSL(2, Z) on H2 .

We claim

Proposition 3.2.1. For all τ ∈ H2 there exists an element g ∈ PSL(2, Z) so that gτ ∈ F.

Moreover,

• if τ ∈ F̊ then
 
PSL(2, Z) · τ ∩ F = {τ },

1
• if Re(τ ) = 2
then
 
PSL(2, Z) · τ ∩ F = {gτ, gτ + 1},

• if Re(τ ) = − 21 then
 
PSL(2, Z) · τ ∩ F = {gτ, gτ − 1}.

• and if |τ | = 1 then
 
PSL(2, Z) · τ ∩ F = {gτ, −1/τ },

The proof of this proposition is part of this week’s exercises.

Since T : z 7→ z + 1 maps √ −1/2 to the line Re(z) = 1/2 and S : z 7→ −1/z

√the line Re(z) =
fixes i and swaps (−1 + 3i)/2 and (1 + 3i)/2 (which are in turn the fixed points of ST ).
These turn out to be the only side identifications and thus the quotient looks like Figure
2:
30 3. THE TEICHMÜLLER SPACE OF THE TORUS

i π

2π
3

√
±1+ 3i
2
Figure 2. A cartoon of M1 .

So M1 is a space that has the structure of a hyperbolic surface

√ near almost every point.
The only problematic points are the images of i and (±1 + 3i)/2, where the M1 looks
like a cone. The technical term for such a space is a hyperbolic orbifold.
M1 is called the moduli space of tori. T 1 = H2 is called the Teichmüller space of tori.
Our next intermediate goal is to generalize this to all surfaces. To this end, we will introduce
a different perspective on T 1 , that generalizes naturally to higher genus surfaces.

3.3. T 1 as a space of marked structures

Our objective in this section is to understand what the information is that is parametrized
by T 1 .
3.3.1. Markings
 asa choice of generators for π1 (R). So, suppose τ ∈ H2 and τ ′ = gτ
a b
for some g = ∈ PSL(2, Z). Let f : Rτ ′ → Rτ denote the biholomorphism from the
c d
proof of Proposition 3.1.1. We saw that we can find a lift fe : C → C so that fe(z) = (cτ +d)z.
In particular, using the relation between τ and τ ′ , we see that
fe({1, τ ′ }) = {cτ + d, aτ + b}.
So, the biholomorphism corresponds to a base change (i.e. the change of a choice of
generators) for Λτ .
Let us formalize this idea of a base change. First we take a base point p0 = [0] ∈ Rτ for
the fundamental group π1 (Rτ , p0 ). The segments between 0 and 1 and between 0 and τ
project to simple closed curves on Rτ and determine generators
[Aτ ], [Bτ ] ∈ π1 (Rτ , p0 ).
This now also gives us a natural choice of isomorphism Λτ ≃ π1 (Rτ , p0 ), mapping
1 7→ [Aτ ] and τ 7→ [Bτ ].
 3.3. T 1 AS A SPACE OF MARKED STRUCTURES 31

Likewise, for Rτ ′ we also obtain a natural system of generators [Aτ ′ ], [Bτ′ ] ∈ π1 (Rτ ′ , p0 ).
Moreover, if f∗ : π1 (Rτ ′ , p0 ) → π1 (Rτ , p0 ) denotes the map f induces on the fundamental
group, then
f∗ ([Aτ ′ ]) ̸= [Aτ ] and f∗ ([Bτ ′ ]) ̸= [Bτ ].

Let us package these choices of generators:

Definition 3.3.1. Let R be a Riemann surface homeomorphic to T2 .

(1) A marking on R is a generating set Σp ⊂ π1 (R, p) consisting of two elements.
(2) Two markings Σp and Σ′p′ are called equivalent if there exists a continuous curve
α from p to p′ so that the corresponding isomorphism Tα : π1 (R, p) → π1 (R, p′ )
satisfies
Tα (Σp ) = Σ′p′ .

Two pairs (R, Σ) and (R′ , Σ′ ) of marked Riemann surfaces homeomorphic to T2 are called
equivalent if there exists a biholomorphic mapping h : R → R′ such that
h∗ (Σ) ≃ Σ′ .

Note that above we have not proved that (Rτ , {[Aτ ], [Bτ ]}) and (Rτ ′ , {[Aτ ′ ], [Bτ ′ ]}) are
equivalent as marked Riemann surfaces, because our map f∗ did not send the generators
to each other, and in fact, they are not equivalent:

Theorem 3.3.2. Let τ, τ ′ ∈ T 1 . Then the marked Riemann surfaces

(Rτ , {[Aτ ], [Bτ ]}) and (Rτ ′ , {[Aτ ′ ], [Bτ ′ ]})
are equivalent if and only if τ ′ = τ . Moreover, we have an identification
 .
R a Riemann surface homemorphic to T2
T 1 = (R, Σp ) : ∼.
p ∈ R, Σp a marking on R

Proof. We begin by proving part of the second claim: every marked complex torus
is equivalent to a marked torus of the form (Rτ , {[Aτ ], [Bτ ]}). So, suppose (R, Σ) is a
marked torus. We know that R is biholomorphic to Rτ for some τ ∈ T 1 . Moreover,
since Σ = {[A], [B]} is a minimal generating set for Λτ , we can find a lattice isomorphism
φ : Λτ → Λτ so that
φ([A]) = 1.
Potentially switching the roles of [A] and [B], we can assume φ is an element of SL(2, Z)
and hence that φ([B]) lies in H2 . The torus Rφ([B]) is biholomorphic to Rτ . So (R, Σ) is
equivalent to
(Rφ([B]) , {Aφ([B]) , Bφ([B]) }).

So, to prove the theorem, we need to show that (Rτ , {[Aτ ], [Bτ ]}) and (Rτ ′ , {[Aτ ′ ], [Bτ ′ ]})
are equivalent if and only if τ = τ ′ . Of course, if τ = τ ′ then the two corresponding marked
surfaces are equivalent, so we need to show the other direction.
32 3. THE TEICHMÜLLER SPACE OF THE TORUS

So let h : Rτ ′ → Rτ be a biholomorphism that induces the equivalence. We may assume

h : C → C so that
that h([0]) = [0] and take a lift e
h(0) = 0.
e

h(z) = αz for some α ∈ C \ {0}. Hence 1 = e

This means that e h(1) = α, which implies that
′ ′
τ = h(τ ) = τ .
e □

Note that so far, our alternate description of Teichmüller space only recovers the set T 1
and not yet it topology. Of course we can use the bijection to define a topology. However,
there is also an intrinsic defintion. We will discuss how to do this later.

3.4. Markings by diffeomorphisms

First, we give a third interpreation of T 1 . This goes through another (equivalent) way of
marking Riemann surfaces.
To this end, once and for all fix a surface S diffeomorphic to T2 . We define:

Definition 3.4.1. Let R and R′ be Riemann surfaces and let

f : S → R and f ′ : S → R′
be orientation preserving diffeomorphisms. We say that the pairs (R, f ) and (R′ , f ′ ) are
equivalent if there exists a biholomorphism h : R → R′ so that
(f ′ )−1 ◦ h ◦ f : S → S
is homotopic to the identity.

Note that if we pick a generating set {[A], [B]} for the fundamental group π1 (S, p) then
every pair (R, f ) as above defines a point
(R, {f∗ ([A]), f∗ ([B])}) ∈ T 1 .

It turns out that this gives another description of the Teichmüller space of tori:

Theorem 3.4.2. Fix S and [A], [B] ∈ π1 (S, p) as above. Then the map
 .
R a Riemann surface, f : S → R
(R, f ) : ∼ → T1
an orientation preserving diffeomorphism
given by
(R, f ) 7→ (R, {f∗ ([A]).f∗ ([B])}),
is a well-defined bijection.

Proof. We start with well-definedness. Meaning, suppose (R, f ) and (R′ , f ′ ) are
equivalent. By definition, this means that there exists a biholomorphic map h : R → R′ so
that
h ◦ f : S → R′ and f ′ : S → R′
 3.4. MARKINGS BY DIFFEOMORPHISMS 33

are homotopic. Now if α is a continuous arc between f ′ (p) and h(f (p)), we see that Tα
induces an equivalence between the markings

{f∗′ ([A]), f∗′ ([B])} and {(h ◦ f )∗ ([A]), (h ◦ f )∗ ([B])},

making (R, {f∗ ([A]), f∗ [B]}) and (R′ , {f∗′ ([A]), f∗′ ([B])}) equivalent. This means that they
correspond to the same point by the previous theorem. So, the map is well defined.
Moreover, the map is surjective. For any τ ∈ T 1 we can find an orientation preserving
diffeomorphism f : S → Rτ . Indeed, we know that there exists some τ0 ∈ T 1 such
that (S, {[A], [B]}) ∼ (Rτ0 , {[Aτ0 ], [Bτ0 ]}) as marked surfaces. One checks that the map
fτ : C → C given by
(τ − τ 0 )z − (τ − τ0 )z
fτ (z) =
τ0 − τ 0
descends to an orientation preserving diffeomorphism Rτ0 → Rτ that induces the marking
{[Aτ ], [Bτ ]} on Rτ .
For the injectivity, suppose that
h i h i
′ ′ ′
(R, {f∗ ([A]), f∗ ([B])}) = (R , {f∗ ([A]), f∗ ([B])}) .

Take τ0 ∈ T 1 such that

h i h i
(S, {[A], [B]}) = (Rτ0 , {([Aτ0 ]), [Bτ0 ]}) .

Moreover, let h : R → R′ be a holomorphism such that

h∗ {f∗ ([A]), f∗ ([B])} = {f∗′ ([A]), f∗′ ([B])}.

We choose lattices Λ, Λ′ ⊂ C, generated by (1, a) and (1, a′ ) respectively such that

R = C/Λ and R′ = C/Λ′ ,

and the generators induce the bases {f∗ ([A]), f∗ ([B])} and {f∗′ ([A]), f∗′ ([B])} respectively.

Now, let fe, fe′ , e

h : C → C be lifts. We may assume that

fe(0) = fe′ (0) = e

h(0) = 0, fe(1) = fe′ (1) = e
h(1) = 1,

and
fe(τ0 ) = a, fe′ (τ0 ) = a′ h(a) = a′
and e
So we obtain a homotopy Ft : C → C defined by

h ◦ fe(z) + t fe′ (z)

Ft (z) = (1 − t) e

between fe and fe′ that descends to a homotopy between f : S → R′ and f ′ : S → R′ . □

34 3. THE TEICHMÜLLER SPACE OF THE TORUS

3.5. The Teichmüller space of Riemann surfaces of a given type

The two description of the Teichmüller space of the torus above can be generalized to
different Riemann surfaces. We will take the second one as a definition, as this is the most
common definition in the literature. Moreover, it naturally leads to another key object in
Teichmüller theory: the mapping class group.

Definition 3.5.1. Let S be a surface of finite type. Then the Teichmüller space of S is
defined as
 .
X a Riemann surface , f : S → X
T (S) = (X, f ) : ∼,
an orientation preserving diffeomorphism
where
(X, f ) ∼ (Y, g)
if and only if there exists a biholomorphism h : X → Y so that the map
g −1 ◦ h ◦ f : S → S
is homotopic to the identity.

We will often write

T (Σg,n ) = T g,n and T (Σg ) = T g .
 LECTURE 4

Markings, mapping class groups and moduli spaces

4.1. Teichmüller space in terms of markings

In order to get to the analogous definition to the space of marked tori, we need to single
out particularly nice generating sets for the fundamental group, just like we did for tori.
We will stick to closed surfaces. Recall that the fundamental group of a surface of genus g
satisfies:
g
* +
Y
π1 (Σg , p) = a1 , . . . , ag , b1 , . . . , bg [ai , bi ] = e .
i=1
In what follows, a generating set A1 , . . . , Ag , B1 , . . . , Bg of π1 (Σg , p) that satisfies
g
Y
[Ai , Bi ] = e,
i=1

will be called a canonical generating set. Note that this includes the torus case.

Definition 4.1.1. Let R be a closed Riemann surface.

(1) A marking on R is a canonical generating set Σp ⊂ π1 (R, p).
(2) Two markings Σp and Σ′p′ are called equivalent if there exists a continuous curve
α from p to p′ so that the corresponding isomorphism Tα : π(R, p) → π1 (R, p′ )
satisfies
Tα (Σp ) = Σ′p′ .

Two pairs (R, Σ) and (R′ , Σ′ ) of marked closed Riemann surfaces are called equivalent if
there exists a biholomorphic mapping h : R → R′ so that
h∗ (Σ) ≃ Σ′ .

Just like in the case of the torus, the space of marked Riemann surfaces turns out to be
the same as Teichmüller space:

Theorem 4.1.2. Let S be a closed surface and Σ a marking on S. Then the map
 .
R a closed Riemann surface diffeomorphic to S
T (S) → (R, Σp ) : ∼.
p ∈ R, Σp a marking on R
given by
[(R, f )] 7→ [(R, f∗ (Σ)]
is a bijection.
35
36 4. MARKINGS, MAPPING CLASS GROUPS AND MODULI SPACES

Before we sketch the proof of this theorem, we state and prove a lemma that will be of use
in the study of mapping class groups as well:

Lemma 4.1.3 (Alexander Lemma). Let D be a 2-dimensional closed disk and ϕ : D → D

a homeomorphism that restricts to the identity on ∂D. Then ϕ is isotopic to the identity
D→D

Proof of the Alexander lemma. Identify D with the closed unit disk in R2 and
define the map F : D × [0, 1] → D by
  
x
 (1 − t) · ϕ (1−t)
 if ||x|| < 1 − t and t < 1
Ft (x) = x if ||x|| > 1 − t and t < 1

x if t = 1.


This yields the isotopy we want. □

We can make this lemma work in the smooth category as well, but its proof is significantly
less easy. It for instance follows from work by Smale [Sma59]. In this course we will
generally gloss over the difference between homeomorphisms and diffeomorphisms.

Proof sketch. Write Σ = {[A1 ], . . . , [Ag ], [B1 ], . . . , [Bg ]}, where Ai , Bi are simple
closed curves based at a point p0 ∈ S. Let us start with the injectivity. So, suppose
[(R, f∗ (Σ)] = [(R′ , f∗′ (Σ)].
This means that we can find a biholomorphic map h : R → R′ and a self-diffeomorphism
g0 : R′ → R′ that is homotopic to the identity and such that
g1 = g0 ◦ h ◦ f
corresponds with f ′ on the curves A1 , . . . , Ag , B1 , . . . , Bg . The domain obtained by deleting
these curves from S is a disk. This implies that f ′ and g1 are homotopic (using the
Alexander trick), which in turn means that
[(R, f )] = [(R′ , f ′ )] ∈ T (S).

For surjectivity, suppose we are given a marked Riemann surface (R, Σp ). So we need to
find an orientation preserving homeomorphism f : S → R so that f∗ (Σ) = Σp . So, let us
take simple closed smooth curves A′1 , . . . , A′g , B1′ , . . . , Bg′ such that
Σp = {[A′1 ], . . . , [A′g ], [B1′ ], . . . , [Bg′ ]}.
Moreover, we will set
g g
[ [
′
C= (Aj ∪ Bj ), C = (A′j ∪ Bj′ ), S0 = S \ C, and R0 = R \ C ′ .
j=1 j=1

R0 and S0 are diffeomorphic to polygons with 4g sides. So we can find a diffeomorphism

by extending a diffeomorphism R0 , S0 . For more details, see [IT92, Theorem 1.4]. □
 4.1. TEICHMÜLLER SPACE IN TERMS OF MARKINGS 37

4.1.1. Punctures and marked points. If n ≥ 1, we can think of T (Σg,n ) as a space of

surfaces with marked points (as opposed to punctures) as well:

Proposition 4.1.4. Let n ≥ 1 and fix n distinct points x1 , . . . , xn ∈ Σg . There is a

bijection
.
T (Σg,n ) −→ { (X, f ) : f : Σg → X an orientation preserving diffeomorphism } ∼,
where (X1 , f1 ) ∼ (X2 , f2 ) if and only if there exists a biholomorphism h : X1 → X2 such
that
f2−1 ◦ h ◦ f1 (xi ) = xi for i = 1, . . . , n
and f2−1 ◦ h ◦ f1 : Σg → Σg is homotopic to the identity through maps fixing x1 , . . . xn .

We leave the proof to the reader.

4.1.2. Basic examples. We have seen that the Teichmüller space of the torus can be
identified with H2 (as a set for now). We will treat some further examples in this sec-
tion.

Proposition 4.1.5. We have

(a) Let S be diffeomorphic to Σ0 , Σ0,1 , Σ0,2 or Σ0,3 , then T (S) is a point.
(b) Let T (Σ1,1 ) can be identified with T (Σ1 ).

Proof. For (a), suppose that [X1 , f1 ], [X2 , f2 ] ∈ T (Σ0,n ) with 0 ≤ n ≤ 3. We will
think of these two as surfaces with marked points, coming from at most three marked
points x1 , x2 , x3 on S2 . By the uniformization theorem, we can identify X1 and X2 with
the Riemann sphere C. b Moreover (using that n ≤ 3), there exists a Möbius transformation
φ : X1 → X2 such that
φ(f1 (xi )) = f2 (xi ), i = 1, . . . , n.
As such the diffeomorphism f2−1 ◦ φ ◦ f1 : S2 → S2 fixes x1 , . . . , xn . All we need to do, is
show that this map is homotopic to the identity. If n = 0, we can perform a homotopy
such that f2−1 ◦ φ ◦ f1 : S2 → S2 fixes a point, which we shall call x1 . This means that
f2−1 ◦ φ ◦ f1 : S2 → S2 can be restricted to a self homeomorphism of S2 − {x1 } ≃ R2 , that
we call f : R2 → R2 . The map F : R2 × [0, 1] → R2 , defined by
Ft (x) = (1 − t) · f (x) + t · x
is a homotopy between f and the identity R2 → R2 . Because both f2−1 ◦ φ ◦ f1 and the
identity fix x1 ∈ S2 , the homotopy above can be extended to S2 .
The proof for item (b) is similar. We again think in terms of surfaces with marked points.
We have a surjective map
T (Σ1,1 ) → T (Σ1 ),
mapping [X, f ] ∈ T (Σ1,1 ) to [X, f ] ∈ T (Σ1 ). What we need to show is that this map is
injective.
So, suppose [X1 , f1 ] = [X2 , f2 ] ∈ T (Σ1 ). So there exists a biholomorphism h : X1 → X2
such that f2−1 ◦ h ◦ f1 : Σ1 → Σ1 is homotopic to the identity. We need to show that we can
38 4. MARKINGS, MAPPING CLASS GROUPS AND MODULI SPACES

modify h in such a way that f2−1 ◦ h ◦ f1 remains homotopic to the identity and also fixes
our favorite point x1 ∈ Σ1 . To this end, let’s write X2 = C/Λ for some lattice Λ. Suppose
[p], [q] ∈ X2 . Observe that h0 : X2 → X2 , defined by
h0 ([z]) = h0 ([z + q − p])
is a biholomorphic map X2 → X2 that is homotopic to the identity and maps [p] to [q].
So, if we set [p] = h ◦ f1 (x1 ) and [q] = f2 (x1 ), then h0 ◦ h : X1 → X2 is the biholomorphic
map we’re looking for. □

4.2. The mapping class group

4.2.1. Defintion. Just like in the case of the torus, we have a natural group action on
the Teichmüller space of a surface, by a group called the mapping class group:

Definition 4.2.1. Let S0 be a compact surface of finite type and Σ ⊂ S0 a finite set. Set
S = S0 \ Σ. The mapping class group of S is given by
MCG(S) = Diff + (S, ∂S, Σ)/Diff +
0 (S, ∂S, Σ)

where
 
 f an orientation preserving diffeomorphism that 
Diff + (S, ∂S, Σ) = f : S0 → S0 : acts as the identity on the boundary components
 of S0 and preserves the elements of Σ pointwise 
and
 
 f homotopic to the identity 
Diff +
0 (S, ∂S, Σ) = f ∈ Diff + (S, ∂S, Σ) : through a homotopy preserving .
 the elements of Σ pointwise 

The group operation is induced by composition of functions.

Some authors let go of the condition that MCG(S) fixes the punctures. The group we
defined above is then often called the pure mapping class group.

4.3. Moduli space

Looking at Definition 3.5.1, we see there is a natural group action of the mapping class
group of a surface on the corresponding Teichmüller space.
[g] · [(R, f )] = [(R, f ◦ g −1 )].

The quotient is what will be called moduli space.

Definition 4.3.1. Let S be a surface of finite type . The moduli space of S is the space
M(S) = T (S)/ MCG(S).

We will write
M(Σg,n ) = Mg,n and M(Σg ) = Mg .
 4.4. ELEMENTS AND EXAMPLES OF MAPPING CLASS GROUPS 39

Remark 4.3.2. Note that by using the convention that the mapping class group fixes
boundary components and punctures, we leave these “marked”, i.e. if two surfaces are
isometric, but any isometry between them permutes the punctures, these surfaces represent
different points in moduli space.

4.4. Elements and examples of mapping class groups

4.4.1. Basic examples. We have:

Proposition 4.4.1. Let n ≤ 3, then

MCG(Σ0,n ) = {e}.

Proof. We start with the case n ≤ 1. This is very similar to some of what we did
in the proof of Proposition 4.1.5. Suppose f : Σ0 → Σ0 is an orientation preserving
diffeomorphism. We can (up to homotopy if n = 0) assume that f fixes a point x ∈ Σ0 .
This allows us to restrict f to Σ0 −{x} ≃ R2 and use a straight line homotopy to homotope
f |R2 to the identity. This extends to a homotopy between f and the identity on Σ0 , because
both fix x.
The proof of the cases n ∈ {2, 3} is part of this week’s exercises. □

4.4.2. Dehn twists and the mapping class group of the annulus. Before we move
on, let us describe some non-trivial elements of the mapping clas group. First, consider an
annulus
A := [0, 1] × R/Z.
Define a map T : A → A by
T (t, [θ]) = (t, [θ + t])
for all t ∈ [0, 1], θ ∈ R. This map is called a Dehn twist. Note that this map fixes ∂A
pointwise. Figure 1 shows that this map does to a segment connecting the two boundary
components of the annulus.

Figure 1. A Dehn twist on an annulus.

Before we show how to turn T into a non-trivial element of a mapping class group of a
different surface, we mention that T generates the mapping class group of the annulus:
40 4. MARKINGS, MAPPING CLASS GROUPS AND MODULI SPACES

Proposition 4.4.2. Let A = [0, 1] × R/Z. Then

MCG(A) ≃ Z = ⟨[T ]⟩.

We will prove below..

Now let α be an essential (i.e. not homotopically trivial and not homotopic into a puncture
or boundary component) simple closed curve on S. Let N be closed regular neighborhood
of α. Identifying N with A, we can define a map Tα : S → S by

T (p) if p ∈ N
Tα (p) =
p if p ∈ S \ N
Because T |∂A is the identity map, this is a continuous map. To obtain an element in
MCG(S), we need to start with a smooth map. There are multiple ways out at the moment.
We could smoothen T . Or we could use surface topology to argue that Tα is homotopic to
a smooth map. Since for the mapping class group, we only care about diffeomorphisms up
to homotopy, the element we get in MCG(S) will not depend on how we do this.
Figure 2 shows an example of a Dehn twist.

γ Tα (γ)

Figure 2. A Dehn twist on a surface of genus two.

We see that Tα maps a curve γ on the surface intersecting the defining curve α (of which we
have only drawn the regular neighborhood) transversely to a curve that is not homotopic
to γ. In particular, Tα is not homotopic to the identity and hence defines a non-trivial
element in MCG(S).

Proof of Proposition 4.4.2. We will first construct a homomorphism ρ : MCG(A) →

Z. Given an orientation preserving diffeomorphism f : A → A such that f |∂A = Id, we can
find a lift fe : [0, 1] × R → [0, 1] × R such that fe(0, 0) = (0, 0). This means that
fe|{0}×R = Id.
Because f |∂A = Id, the restriction fe|{1}×R is an integer translation. We let ρ(f ) be this
integer.
ρ is surjective, because ρ([T n ]) = n. Now suppose ρ([f ]) = 0. This means that fe restricts
to the identity on {0, 1} × R. We have that
 
fe n · (t, x) = f∗ (n) · fe(t, x), n ∈ Z, (t, x) ∈ [0, 1] × R,

where f∗ ∈ Aut(Z) = {±Id}. Because fe|{0,1}×R = Id, we need that f∗ = Id. Implying that
 
fe n · (t, x) = n · fe(t, x), n ∈ Z, (t, x) ∈ [0, 1] × R
 4.4. ELEMENTS AND EXAMPLES OF MAPPING CLASS GROUPS 41

and thus that the straight line homotopy

Fs ((t, x)) = (1 − s) · fe(x, t) + s · (x, t), s ∈ [0, 1]
is a Z-equivariant homotopy between fe and the identity, that hence descends to A. This
proves that ρ is injective and concludes the proof of the proposition. □

4.4.3. The mapping class group of the torus. We briefly return to the torus. The
question is of course whether the general definition on the mapping class group really
corresponds to what happens in the case of the torus. We recall that
M1 = PSL(2, Z)\H2 .
This makes one wonder whether the mapping class group of the torus is maybe PSL(2, Z).
This turns out to be almost correct. Indeed, we have the following theorem:

Theorem 4.4.3. We have

MCG(T2 ) ≃ SL(2, Z).
The action of MCG(T2 ) on T 1 is that given by
 
a b aτ − b
τ=
c d −cτ + d
 
a b
for all ∈ SL(2, Z) and τ ∈ T 1 .
c d

Proof. We will identify

T2 = R2 /Z2
First observe that very element A ∈ SL(2, Z) induces a linear diffeomorphism x 7→ A · x of
R2 . Moreover, since SL(2, Z) preserves Z2 ⊂ R2 , the action on R2 descends to an action
by diffeomorphisms
T2 → T2
that are orientation preserving because det(A) > 0.
Our goal is to show that every orientation preserving diffeomorphism ϕ : T2 → T2 is
homotopic to such a map. To this end, we may homotope ϕ so that it fixes [0] ∈ T2 and
we can take a lift ϕe : R2 → R2 that fixes the origin of R2 . We have
ϕ(x
e + (m, n)) = ϕ(x)
e + ϕ∗ (m, n),
for all (m, n) ∈ Z2 where ϕ∗ : Z2 → Z2 is an isomorphism, i.e. an element of GL(2, Z).
For a general surface S, the map [ϕ] ∈ MCG(S) 7→ ϕ∗ ∈ Aut(π1 (S)) does not yield a
homomorphism: we have chosen a homotopy to make ϕ fix a base point. Changing this
choice a priori changes ϕ∗ by an inner automorphism of π1 (S). So we only obtain a map
to Out(π1 (S)). However, because Z2 is abelian, we have Out(Z2 ) ≃ Aut(Z2 ). So in the
case of the torus, we obtain a homomorphism MCG(T2 ) → GL(2, Z).
Write Aϕ for the GL(2, Z) matrix corresponding to ϕ. Observe that
Ft (x) = tAϕ · x + (1 − t)ϕ(x),
e t ∈ [0, 1], x ∈ R2
42 4. MARKINGS, MAPPING CLASS GROUPS AND MODULI SPACES

gives a Z2 -equivariant homotopy between ϕe and the linear map x 7→ Aϕ · x. Since ϕe

is orientation preserving, det(Aϕ ) > 0, and hence Aϕ ∈ SL(2, Z). So we obtain a map
MCG(T2 ) → SL(2, Z). The map is surjective, because ϕA maps to A. Moreover, the map
is injective, because if Aϕ is the identity matrix, Ft gives a homotopy of ϕe to the identity.
Since the action of [ϕ] ∈ MCG(T2 ) on T (T2 ) is by precomposition with ϕ−1 , the action is
as described. □

Remark 4.4.4. Note that the theorem above implies that the mapping class group action
is not faithful. The kernel of the action is the center of SL(2, Z), i.e. the subgroup
  
1 0
± < SL(2, Z).
0 1

On the other hand, we do have

H2 / PSL(2, Z) = H2 / SL(2, Z).
This means that the mapping class group action is indeed a generalization of the situation
for the torus case.

4.4.4. Mapping class groups in higher genus. We proved in the exercises that SL(2, Z)
can be generated by the matrices
   
1 1 0 −1
T = and S = .
0 1 1 0
We can also generate SL(2, Z) by the matrices
   
1 1 1 0
T = and R = .
0 1 1 1
Indeed, a calculation shows that S = T −1 RT −1 .
Now identify T2 = R2 /Z2 again and write α and β for the closed curves in T2 that are the
images of the straight line segments between the origin and (0, 1) and (1, 0) respectively.
Tracing the proof of Theorem 4.4.3, we see that T = [Tα ] and R = [Tβ ]. That is, MCG(T2 )
can be generated by two Dehn twists.
It actually turns out that an analogous statement holds for all mapping class groups. In
the following theorem, a non-separating curve will be a curve α so that S \ α is connected.
Figure 3 shows an example.

α
β

Figure 3. A separating curve (α) and a non-separating curve (β).

 4.4. ELEMENTS AND EXAMPLES OF MAPPING CLASS GROUPS 43

Theorem 4.4.5 (Dehn - Lickorish theorem). Let S be a surface of finite type, the mapping
class group MCG(S) is generated by finitely many Dehn twists about nonseparating simple
closed curves.

4.4.5. The action on homology. If S is a surface, then MCG(S) acts on its homology
H1 (S). Indeed every diffeomorphism f : S → S induces an automorphism f∗ : H1 (S, Z) →
H1 (S, Z). In this section, we briefly descibe some aspects of this action. We will restrict
to closed surfaces.
First of all, it turns out the action preserves some extra structure: the algebraic intersection
number between oriented curves. In order to define it, let α and β be two oriented closed
curves on an oriented surface S that intersect each other transversely at every intersection
point. Then the algebraic intersection number between α and β is given by
X
i(α, β) = sgn(ω(vp (α), vp (β))),
p∈α∩β

where sgn : R → {±1} denotes the sign function, ω is any volume form that induces the
orientation and vp (α) and vp (β) denote the unit tangent vectors to α and β respectively
at p. Note that
i(β, α) = −i(α, β).
Figure 4 shows an exemple of a positive contribution to the intersection number.
β
p α

Figure 4. A positive contribution to i(α, β) if the orientation points out of

the page.

We note that this form descends to homology. That is, it induces a form
i : H1 (S, Z) × H1 (S, Z) → Z
called the intersection form, with the properties:
(1) i is bilinear.
(2) i is alternating, i.e.
i(a, b) = −i(b, a)
for all a, b ∈ H1 (S, Z).
(3) i is non-degenerate, i.e. if a ∈ H1 (S, Z) is such that
i(a, b) = 0 for all b ∈ H1 (S, Z)
then a = 0.
(see [FK92, Section III.1] for more details). Such a form is called a symplectic form.
First of all note that the image preserves the intersection form. Moreover, isotopic maps
give rise to the same automorphism. So this gives us a representation
MCG(S) → Aut(H1 (S, Z), i)
44 4. MARKINGS, MAPPING CLASS GROUPS AND MODULI SPACES

called the homology representation of the mapping class group. Recall that if S is a
closed orientable surface of genus g, then H1 (S, Z) ≃ Z2g . Choosing an identification, the
homology representation becomes a map
MCG(S) → Sp(2g, Z) = A ∈ Mat2g (Z) : i(Av, Aw) = i(v, w), ∀v, w ∈ Z2g .


It turns out that this representation is surjective (this can be proved using a finite gener-
ating set for Sp(2g, Z) consisting of transvections, which can be realized by Dehn twists),
but generally highly non-injective. A notable exception is the case of the torus, there is an
isomorphism
Sp(2, Z) ≃ SL(2, Z)
and indeed the the homology representation MCG(T2 ) → Sp(2, Z) is an isomorphism.
 LECTURE 5

Beltrami differentials and a topology on Teichmüller space

Our next goal is to put a topology on Teichmüller space (that coincides with the topology
of the upper half plane in the case of the torus and the once punctured torus). Of course,
the goal is that marked complex structures that are more similar should be closer to each
other. So that raises the question how one measures the how similar two marked complex
structures are.

5.1. Wirtinger derivatives

Let us first recall the notation for differentiation with respect to complex coordinates. If
(U, z) is a complex coordinate chart on a surface S, we wrote z = x + iy and
   
∂ 1 ∂ ∂ ∂ 1 ∂ ∂
= −i and = +i .
∂z 2 ∂x ∂y ∂z 2 ∂x ∂y
Moreover, the equation ∂f∂z(z) = 0 is equivalent to the Cauchy-Riemann equations for f (z),
i.e. this equation is equivalent to f being holomorphic. One readily verifies that in these
coordinates, the product rule takes the form
∂ ∂f ∂g
(f · g) = ·g+f · .
∂z ∂z ∂z
Moreover, the chain rules read
∂(f ◦ g) ∂f ∂g ∂f ∂g
(z0 ) = (g(z0 )) · (z0 ) + (g(z0 )) · (z0 )
∂z ∂z ∂z ∂z ∂z
and
∂(f ◦ g) ∂f ∂g ∂f ∂g
(z0 ) = (g(z0 )) · (z0 ) + (g(z0 )) · (z0 ).
∂z ∂z ∂z ∂z ∂z
5.2. Beltrami coefficients
In Section 2.3 on conformal structures we saw that when trying to find an isothermal
complex coordinate w on a Riemannian surface, equipped with a coordinate z, the frac-
tion
∂w/∂z
µ=
∂w/∂z
shows up as a measure of “how much” we need to “correct” the coordinate z in order
to make the metric conformal to the standard Euclidean metric in w (and thus yielding
holomorphic chart transitions).
In general, given a smooth map f : S → R between Riemann surfaces, we can use this
expression to measure how far f is from holomorphic. Let us first prove that this yields a
well defined object:
45
46 5. BELTRAMI DIFFERENTIALS AND A TOPOLOGY ON TEICHMÜLLER SPACE

Lemma 5.2.1. Let S and R be Riemann surfaces and f : S → R a smooth map. Suppose
that (U, z) is a holomorphic local coordinate on S and (V, w) one on R. Then the smooth
function µ : z(U ) → C defined by
   
∂F . ∂F
µ= ,
∂z ∂z
where F = w ◦ f ◦ z −1 : z(U ) → C is independent of the choice of coordinate (V, w).

Proof. Suppose (V ′ , w′ ) is a different holomorphic local coordinate with f (U ) ⊂ V ′ .

Write F ′ = w′ ◦ f ◦ z −1 : z(U ) → C. By the chain rule,
∂F ′ ∂(w′ ◦ w−1 ◦ F ) ∂(w′ ◦ w−1 )
 
∂F
= = ◦F · ,
∂z ∂z ∂z ∂z
where the second term disappears because w′ ◦ w−1 is holomorphic. Likewise,
∂F ′ ∂(w′ ◦ w−1 ◦ F ) ∂(w′ ◦ w−1 )
 
∂F
= = ◦F · .
∂z ∂z ∂z ∂z
So when we divide the two, we obtain the same µ. □

Observe that µ does depend on the local coordinate (U, z). Indeed, if (U ′ , z ′ ) is a different
holomorphic local coordinate and we write F ′ = w ◦ F ◦ (z ′ )−1 , then

∂F ′ ∂(F ◦ z ◦ (z ′ )−1 ) ∂(z ◦ (z ′ )−1 )

 
∂F ′ −1
= = ◦ z ◦ (z ) ·
∂z ′ ∂z ′ ∂z ∂z ′
∂(z ◦ (z ′ )−1 )
   
∂F ′ −1
= ◦ z ◦ (z ) · ,
∂z ∂z ′
again using the fact that the coordinate change is holomorphic to conclude that the other
term disappears, and
∂F ′ ∂(F ◦ z ◦ (z ′ )−1 ) ∂(z ◦ (z ′ )−1 )
 
∂F ′ −1
= = ◦ z ◦ (z ) · ,
∂z ′ ∂z ′ ∂z ∂z ′
where we have used that
∂(z ◦ (z ′ )−1 ) ∂(z ◦ (z ′ )−1 )
 
= = 0.
∂z ′ ∂z ′
In conclusion
∂(z ◦ (z ′ )−1 ) . ∂(z ◦ (z ′ )−1 )
 
′
µ(z ) = µ(z) ·
∂z ′ ∂z ′
and thus a smooth map f : S → R yields a differential µf of type (−1, 1) associated to f .
That is, it makes sense to write
dz
µf = µ(z) · .
dz
So, we have a well defined differential µf associated to f that satisfies
µf = 0 ⇔ f is biholomorphic.
 5.2. BELTRAMI COEFFICIENTS 47

Indeed, if µf = 0 then ∂f /∂z = 0 everywhere, so f is holomorphic. Since f is also invertible

(and the inverse of a bijective holomorphic function is holomorphic), f is biholomorphic.
We will call µf the Beltrami coefficient of f .
We also observe that the coordinate transition above implies that z 7→ |µf (z)| is a well
defined function on S. Moreover, it is uniformly bounded by 1 for all orientation preserving
diffeomorphisms:

Lemma 5.2.2. Let S and R be Riemmann surface and let f : S → R be an orientation

preserving diffeomorphism, then
|µf (z)| < 1
for all z ∈ S.

Proof. Since this is a local question, we can just think of f as map U → V , where
U, V ⊂ C are open subsets. Writing f (x, v) = u(x, v)+iv(x, y), the fact that f is orientation
preserving means that
 
∂u(x, y)/∂x ∂u(x, y)/∂y
0 < det(Df (x, y)) = det
∂v(x, y)/∂x ∂v(x, y)/∂y
2 2
∂u(x, y) ∂v(x, y) ∂u(x, y) ∂v(x, y) ∂f ∂f
= − = − ,
∂x ∂y ∂y ∂x ∂z ∂z
hence proving the lemma. □

A differential µ of type (−1, 1) on a Riemann surface X whose L∞ -norm satisfies

||µ||∞ = sup{|µ(z)|} < 1
z∈X

is called a Beltrami differential.

The following transformation formula for Beltrami coefficients will also be useful to us:

Lemma 5.2.3. Let S, X1 and X2 be Riemann surfaces and let

f g
S −→ X1 −→ X2
be orientation preserving diffeomorphisms. Then
 !
∂f . ∂f µg◦f − µf
µg ◦ f = · .
∂z ∂z 1 − µf · µg◦f

Proof. This is part of this week’s exercises. □

Indeed, we can derive from it that Beltrami coefficients can recognize biholomorphisms:

Lemma 5.2.4. Let X1 and X2 be Riemann surfaces and let

f 1 : R → X1 and f 2 : R → X2
48 5. BELTRAMI DIFFERENTIALS AND A TOPOLOGY ON TEICHMÜLLER SPACE

be orientation preserving diffeomorphisms. Then the map f2 ◦ f1−1 : X1 → X2 is biholo-

morphic if and only if
µ f1 = µ f2 .

Proof. f2 ◦ f1−1 is biholomorphic if and only if µf2 ◦f1−1 = 0 as a Beltrami differential

on X1 . This is true if and only if, as a Beltrami differential on S,
∂f1 /∂z µ f2 − µ f1
0 = µf2 ◦f1−1 ◦ f1 = · ,
∂f1 /∂z 1 − µf1 · µf2
where we have used the previous lemma. Because neither the first factor on the right, nor
the denominator (because |µf | < 1 for an orientation preserving diffeomorphism) can be
0, we obtain that the equation µf2 = µf1 is equivalent to µf2 ◦f1−1 ◦ f1 being 0. □

5.2.1. Topologizing Teichmüller space. We are going the use Beltrami coefficients to
topologize Teichmüller space. First of all, the following theorem implies that Riemann
surface structures up to homotopy can be recognized using Beltrami differentials:

Theorem 5.2.5. Let S, X1 and X2 be Riemann surfaces and

f 1 : S → X1 and f2 : S → X 2
be orientation preserving diffeomorphisms. Then there exists a biholomorphic mapping
h : X1 → X 2
if and only if
µf1 = µf2 ◦φ−1
for some φ ∈ Diff + (S). Moreover, the map
(f2 )−1 ◦ h ◦ f1 : S → S
is homotopic to the identity if and only if φ ∈ Diff +
0 (S, Σ).

Proof. First suppose that there exists a biholormorphic map h : X1 → X2 . Then we

set
φ = (f2 )−1 ◦ h ◦ f1 : S → S.
Then
µf2 = µh◦f1 ◦φ−1 = µf1 ◦φ−1 ,
where we have used Lemma 5.2.3 for the second equality. Since φ = (f2 )−1 ◦ h ◦ f1 , the
second claim is immediate.
Conversely, suppose there exists some φ ∈ Diff + (S) such that
µf1 = µf2 ◦φ−1
Lemma 5.2.3 then shows that h = f2 ◦ φ ◦ f1−1 : X1 → X2 is biholomorphic. Again, the
second claim follows from the form of φ. Indeed f2−1 ◦ h ◦ f1 = φ. □
 5.3. QUASICONFORMAL MAPPINGS 49

So, given a Riemann surface S, we can define the space

 
 f : S → R an orientation 
B(S)1 := µf : preserving diffeomorphism,
 R a Riemann surface 

That we equip with the L∞ topology. This space admits an action of the group of orien-
tation preserving diffeomorphisms Diff + (S) by
φ · µf = µf ◦φ−1 , φ ∈ Diff + (S), µf ∈ B(S)1 .
A direct consequence is the following:

Corollary 5.2.6. The map from the set of marked Riemann surfaces defined by
(R, f ) 7→ µf
induces a bijections
T (S) → B(S)1 /Diff +
0 (S, Σ)
and
M(S) → B(S)1 /Diff + (S, Σ).

In particular, since B(S)1 is a topological space, these bijections equip T (S) and M(S)
with a topology. It is not so hard to see that the choice of Riemann surface structure
on S does not influence the topology on Teichmüller space. Indeed, if S and S ′ are two
Riemann surfaces and g : S ′ → S is any orientation preserving diffeomorphism between
them, then
[X, f ] ∈ T (S) 7→ [X, f ◦ g] ∈ T (S ′ )
is a homeomorphism.
5.3. Quasiconformal mappings
We will now first describe another viewpoint on the same topology: namely that of quasi-
conformal mappings and the Teichmüller metric.
Let us consider what it is that |µf (z)| measures. Let f : D → D denote an orientation
preserving diffeomorphism of some domain D ⊂ C containing 0. And write
∂f ∂f
Df (0) · z = (0) · z + (0) · z =: a · z + b · z
∂z ∂z
denote the first order Taylor expansion of f at 0 (so a = ∂f ∂z
(0) and b = ∂f
∂z
(0)).
Let us consider the inverse image of a unit circle under this map. I.e. the solutions of the
equation
|Df (0) · z| = 1.
Let us write
z = r eiθ , a = |a| eiα , b = |b| eiβ .
The equation then becomes
   
α−β α−β
r · (|a| + |b|) cos θ + + i(|a| − |b|) sin θ + = 1.
2 2
This is the equation of an ellipse with
50 5. BELTRAMI DIFFERENTIALS AND A TOPOLOGY ON TEICHMÜLLER SPACE

β−α 1
- minor axis at polar angle 2
of half length |a|+|b|

β−α+π 1
- major axis at polar angle 2
of half length |a|−|b|
.
Note that the latter is positive since the Jacobian
det(Df (0)) = |a|2 − |b|2 > 0,
as we have seen in the proof of Lemma 5.2.2. Figure 1 shows a picture of the situation.

1
|a|−|b| Df (0)

1
|a|+|b|

Figure 1. An ellipse that gets mapped to a circle.

The ratio of the axes of the ellipse is

|a| + |b| 1 + |µf (0)|
= .
|a| − |b| 1 − |µf (0)|
In particular, if µf (0) is close to 1, then the preimage is very far away from a circle and
if µf (0) = 0 (i.e. if f is biholomorphic) then the preimage is a circle. So this means
that
• Biholomorphic maps locally send circles to circles (i.e. they are conformal)
• The µf (0) measures how far away the preimage of a circle (an ellipse) is from a
circle.
For this reason, the number
1 + |µf (z)|
Kf (z) =
1 − |µf (z)|
is called the quasiconformal dillatation of f at z. Since the ratio between the major and
minor axis of the ellipse is a measure of how far f is from being conformal, this leads to
the following definition:

Definition 5.3.1. Let X and Y be Riemann surfaces and let f : X → Y be an orientation

preserving diffeomorphism. We say f : X → Y is a K-quasiconformal mapping if
 
1 + |µf (z)|
K ≥ Kf := sup .
z∈X 1 − |µf (z)|
Kf is called the quasiconformal dillatation of f .
 5.4. THE TEICHMÜLLER METRIC 51

Remark 5.3.2. The notion of a quasiconformal map can be generalized to maps of much
lower regularity (see eg. [IT92, Chapter 4]). We will consider slightly lower regularity
(but not as low as can be found in the literature) soon as well.

We record two useful properties of quasiconformal maps in a lemma:

Lemma 5.3.3. Suppose X, Y and Z are Riemann surfaces and f : X → Y and g : Y → Z

are orientation preserving diffeomorphisms. Then the following holds:
(a) We have that
Kf ≥ 1
with equality if and only if f is a biholomorphism.
(b) We have that
Kg◦f ≤ Kg · Kf .

(c) Finally,
Kf −1 = Kf .

Proof. This is part of this week’s exercises. □

5.4. The Teichmüller metric

Lemma 5.3.3 implies that the following defintion defines a metric:

Definition 5.4.1. Let S be a Riemann surface. Then the Teichmüller distance between
[X1 , f1 ], [X2 , f2 ] ∈ T (S) is
  
  1  g : X1 → X2 an orientation 
dT [X1 , f1 ], [X2 , f2 ] = log inf Kg : preserving diffeomorphism .
2  −1
homotopic to f2 ◦ f1


1
The factor 2
is a convention. The first thing to observe is that the Teichmüller distance is
a metric:

Lemma 5.4.2. Let S be a Riemann surface. Then the Teichmüller distance dT : T (S) ×
T (S) → [0, ∞) defines a metric.

Proof sketch. The fact that dT is symmetric and satisfies the triangle inequality are
direct from Lemma 5.3.3. In order to show non-degeneracy, suppose
 
dT [X1 , f1 ], [X2 , f2 ] = 0.
There are two ways to show that this implies that [X1 , f1 ] = [X2 , f2 ]:
• We can use Teichmüller’s theorem (that we will state, but not prove later on in
the course), which states that there is a map (of slightly lower regularity) that
realizes dT . This map must have Kg = 1 and hence is a biholomorphism.
52 5. BELTRAMI DIFFERENTIALS AND A TOPOLOGY ON TEICHMÜLLER SPACE

• We can run an approximation argument. Suppose gn : X1 → X2 is a sequence of

maps in the homotopy class of f2 ◦ f1−1 such that
1 n→∞
 
log(Kgn ) −→ dT [X1 , f1 ], [X2 , f2 ] = 0.
2
This means that
n→∞
Kgn −→ 1,
which in turn implies that gn locally uniformly converges to a holomorphic map
X1 → X2 (see [IT92, Proposition 4.36] for details).
□

We could have used dT to topologize Teichmüller space as well:

Lemma 5.4.3. The Teichmüller metric is compatible with the topology on T (S).

Proof. This is a matter of tracing the definitions. Two points in [X1 , f1 ], [X2 , f2 ] ∈
T (S) are close if we can make µf2 ◦f1−1 close to 0 in the L∞ topology by precomposing f1
with a homotopically trivial self-diffeomorphism X1 → X1 . This is the same as saying that
Kg is small for g in the homotopy class of f2 ◦ f1−1 . □

5.5. Grötschz’s theorem

Teichmüller proved that the Teichmüller distance is realized by a unique map called the
Teichmüller map (that has slightly lower regularity than the maps we’re taking the infimum
over, but we’ll get to that).
As a warm up, we’ll consider Grötschz’s problem, that goes as follows. Suppose that
R1 = [0, a] × [0, 1] and R2 = [0, K · a] × [0, 1] are two rectangles in the plane, where a > 0
and K ≥ 1. We can ask what the minimal quasi-conformal dillatation is of a map R1 → R2
that preserves vertical sides and horizontal sides. Grötschz proved the following:

Theorem 5.5.1. (Grötschz’s theorem) Suppose R1 and R2 are as above and f : R1 → R2

is a homeomorphism that is smooth and orientation preserving away from a finite number
of points. Then
Kf ≥ K
with equality if and only if f is the affine map
(x, y) ∈ R1 7→ (K · x, y) ∈ R2 .

Before we prove the theorem, we briefly comment on definitions. Formally, we defined Kf

only for orientation preserving diffeomorphisms. In particular, in the points z ∈ R1 where
f is not smooth, µf (z) is not defined. We can remedy this by simply ignoring these points
in the supremum.
 5.5. GRÖTSCHZ’S THEOREM 53

Proof sketch. Writing Kf (x, y) for the quasiconformal dilatation of f at (x, y) ∈ R1 ,

we first claim that
2
∂f
(5.5.1) (x, y) ≤ Kf (x, y) · det(Jf (x, y)),
∂x
where Jf (x, y) denotes the Jacobian matrix of f at (x, y) ∈ R1 . Indeed, writing
1

M = sup df(x,y) (v) : v ∈ T(x,y) R1
and
1

m = inf df(x,y) (v) : v ∈ T(x,y) R1 ,
∂f 2
We have Kf (x, y) = M/m, det(Jf (x, y)) = M · m and ∂x
(x, y) ≤ M 2 , which proves
(5.5.1).
The second inequality we will use is
Z
∂f
(5.5.2) (x, y) dxdy ≥ K · area(R1 )
R1 ∂x
Ra
This is the observation that for almost all y ∈ [0, 1], 0 ∂f ∂x
(x, y) dx ≥ K · a, by the
substitution rule. Integrating with respect to y gives the desired inequality.
We now have
(5.5.2)
Z 2
2 ∂f
(K · area(R1 )) ≤ (x, y) dxdy
R1 ∂x
(5.5.1)
Z q q 2
≤ Kf (x, y) · det(Jf (x, y))dxdy
R1
Cauchy–Schwarz
Z Z
≤ Kf (x, y)dxdy · det(Jf (x, y))dxdy
R1 R1
≤ area(R2 ) · Kf · area(R1 )
= K · area(R1 ) · Kf · area(R1 ).
Thus yielding that Kf ≥ K.
Equality for the affine map is a matter of computation, so now we need to show uniqueness.
This can be proved using the fact that we need to have equality in all the inequalities
above. □
 LECTURE 6

Measured foliations and quadratic differentials

6.1. Measured foliations

Teichmüller’s characterization of maps realizing the Teichmüller distance is a generalization
of Grötschz’s theorem: optimizers are analogues of affine maps. In order to define what
these are, we need to discuss measured foliations and quadratic differentials. We will start
with the former.

6.1.1. The torus. We start with our favorite example: the torus T2 = R2 /Z2 . Given any
e ℓ of R2 consisting of all lines (that we
straight line ℓ through (0, 0), we obtain a foliation F
will call the leaves of F
e ℓ ) parallel to ℓ.

Because F e ℓ is invariant under the Z2 -action, it defines a foliation F ℓ on T2 . We observe

that if ℓ has a rational slope, then all the leaves of F ℓ are simple closed curves. If the slope
is irrational, every leaf of F ℓ is dense in T2 .
The foliation F ℓ also naturally comes with a measure on trajectories that are transverse
to it. Indeed, let
νℓ : R2 → R
denote the (signed) distance to ℓ. The associated 1-form dνℓ is Z2 -invariant and hence
descends to a 1-form wℓ on T2 . This induces a measure µℓ on arcs α transverse to F ℓ ,
given by Z
µℓ (α) = |wℓ | .
α
By construction µℓ (α) is invariant under istopies of α that preserve the leaf that each point
of α lies in. µℓ is called a transverse measure to F ℓ . The pair (F ℓ , µℓ ) is called a measured
foliation.
Observe that the 1-form wℓ completely determines the measured foliation: the leaves of
F ℓ are the integral submanifolds corresponding to the subbundle ker(wℓ ) of the tangent
bundle T (R2 /Z2 ).
We have seen that every mapping class [ϕ] ∈ MCG(T2 ) admits a representative of the
form
ϕA : [z] 7→ [A · z], [z] ∈ T2 ,
for A ∈ SL(2, Z). If A is hyperbolic, i.e. has two real eigenvalues λ > 1 and 1/λ. The
matrix  
2 1
1 1
55
56 6. MEASURED FOLIATIONS AND QUADRATIC DIFFERENTIALS

Figure 1. A 3-pronged singularity.

is everyone’s favorite example. Then A has two eigenspaces E + , E − ⊂ R2 respectively.

These eigenspaces define two transverse measured foliations (F + , µ+ ) and (F − , µ− ) on T2 .
The map ϕA preserves these foliations and
1
(ϕA )∗ µ+ = λ · µ+ and (ϕA )∗ µ− = · µ− .
λ
So we can think of A as stretching by a factor λ in the direction of F + and contracting by
a factor 1/λ in the direction of F − . As such, this is a natural analogue to affine maps on
rectangles.
6.1.2. Higher complexity. For surfaces of higher complexity, the situation is more com-
plicated. First of all, such surfaces do not admit smooth foliations, so we have to allow for
singularities for the theory to work.

Definition 6.1.1. Let S be a closed surface. A singular foliation F on a S is a decompo-

sition of S into a disjoint union of subsets of S, called the leaves of F, and a finite set of
points of S, called singular points of F, such that the following conditions hold:
(1) Around each nonsingular point p ∈ S, there is a smooth chart to R2 that takes
leaves to horizontal line segments. The transition maps between any two of these
charts are smooth maps that take horizontal lines to horizontal lines.
(2) Around each singular point p ∈ S, there is a smooth chart to R2 that takes leaves
to the level sets of a k-pronged singularity for some k ≥ 3 (Figure 1 shows an
example).

We have the following formula for the relation between the singularities and the Euler
characteristic:

Proposition 6.1.2 (Euler–Poincaré formula). Given a singular foliation F on a closed

surface S, let Ps denote the number of prongs of the singular point s ∈ S. Then the Euler
characteristic of S can be computed as
X
2χ(S) = 2 − Ps .
s∈S
singular

Proof. The proof will be part of this week’s exercises. □

In particular, if S has negative Euler characteristic, any foliation of it necessarily has

singular points, thus justifying our claim above.
 6.1. MEASURED FOLIATIONS 57

Figure 2. A foliation on a surface of genus two.

In order to define transverse measures, we need a notion of arcs transverse to a singular

foliation. These are smooth arcs on the surface that do not go through the singular points
of F and are transverse to the leaves of F.

Definition 6.1.3. Let S be a closed surfae and let F be a singular foliation of S. A

transverse measure µ on F is an assignment of a positive real number µ(α) to each arc α
that is transverse to F such that
• µ is invariant under leaf-preserving isotopy of smooth arcs
• µ is regular, that is, there exists a smooth chart to R2 around each non-singular
point that maps the leaves of F to horizontal lines and such that µ is induced by
the line element |dy|.
The pair (F, µ) is called a measured foliation on S.

6.1.3. Example 1: polygons. It’s high time for an example. Let’s start with a concrete
one. Figure 2 shows a foliated octagon. We can turn this octagon into a surface X of genus
2 by using horizontal and vertical translations to identify its sides (an Euler characteristic
computation shows that the resulting genus is indeed 2.
Moreover, because the foliation is translation invariant, it glues into a singular foliation F
of X. The singularity can be found at the image of the vertices of the polygon (that are
all identified in a single vertex on X).
We claim that the singularity around this vertex is 6-pronged. One way to see this, is by
using Euclidean geometry. Indeed, the surface X comes equipped with the structure of
a Euclidean metric that is defined everywhere except at the image of the vertex. Indeed,
adding up the angles in the corners of the octagon, we see that the total angle around the
vertex is 6π instead of 2π. Every prong in the singularity adds an angle π (a 2-pronged
singularity being a regular point). So, we indeed have a 6-pronged singularity.
58 6. MEASURED FOLIATIONS AND QUADRATIC DIFFERENTIALS

A second way to see this, is just by using the Euler–Poincaré formula. If s ∈ X denotes
the image of the vertex, then we have

−4 = 2χ(X) = 2 − Ps ,

thus yielding Ps = 6.
Again the line-segment coming from the distance to any of the leaves yields a measure µ
on F.
Finally, none of this really used our explicit octagon (except of course the calculation of the
number of prongs of the singularity). So in general, we can build a surface X, by taking
a Euclidean polygon P ⊂ R2 and identifying pairs of parallel sides of the same length.
After that, we take any foliation of R2 by straight lines, from which we obtain a singular
foliation on X. Moreover, instead of only translations, we can also allow rotations of angle
π. These also preserve the foliation. They however don’t preserve any orientation on the
leaves, so we obtain a non-orientable foliation.

6.1.4. Example 2: natural coordinates. Suppose our surface S and P ⊂ S is a finite

set of points. If the surface S − P admits an atlas for which all the transition maps are of
the form:
(x, y) 7→ (f (x, y), c ± y)
for some smooth f : R2 → R and some constant c ∈ R (both depending on the pair of
charts). That is, the transition maps send horizontal lines in R2 to horizontal lines. This
means that the horizontal foliation of R2 induces a singular foliation on S. Since the line
element |dy| is preserved as well, this foliation comes with a measure too.
Let’s again look at our surface X from the previous example. We now equip it with the
horizontal foliation. Since the gluings are by translations, we can equip X − s with an
atlas for which the transition maps all take the form

(x, y) 7→ (x + c1 , y + c2 )

for some c1 , c2 ∈ R. In particular, not only the horizontal, but also the vertical foliation of
X falls under this example.

6.1.5. Example 3: simple closed curves. If we take a (4g + 2)-gon and glue opposite
sides together with orientation reversing homeomorphisms, we obtain a closed surface of
genus g (this can for instance be shown using the Euler characteristic of the result). We
can turn this (4g + 2)-gon into a rectangle, by drawing two opposite sides vertically and
all the other sides horizontally (see Figure 4 for an example of genus 2).
Again, since the identification is by translations, the foliation and the line element |dy|
descend to the quotient surface X. In this foliation, all the non singular leaves are simple
closed curves, that are all homotopic to each other. Conversely, any simple closed curve on
a surface defines such a foliation. We can change the measure by scaling the rectangle.
 6.2. HOLOMORPHIC QUADRATIC DIFFERENTIALS 59

Figure 3. Another foliation on a surface of genus two.

Figure 4. Yet another foliation on a surface of genus two. We identify

(combinatorially and not geometrically) opposite sides.

6.1.6. Boundary and punctures. We haven’t discussed yet what we require of a sin-
gular measured foliation when the surface S has boundary of punctures.
In our definition:
• We will allow the puncture to be a singular point or a regular point. If it’s singular
it will also be allowed to be a 1-pronged singularity (see Figure 5)
• At the boundary, the leaves of the folliation are allowed to be transverse or parallel.
Moreover, we allow for singularities on the boundary (see Figure 6 for what the
local picture is allowed to look like).

Figure 5. A one-pronged singularity at a puncture.

6.2. Holomorphic quadratic differentials

Next up in our discussion are holomorphic quadratic differentials. Indeed, if our surface
comes equipped with the structure of a Riemann surface, we can define measured foliations
on it using quadratic differentials.
60 6. MEASURED FOLIATIONS AND QUADRATIC DIFFERENTIALS

Figure 6. The behavior we allow on the boundary.

If X is a Riemann surface, then a holomomorphic quadratic differential on X is a section

of the symmetric square of the holomorphic cotangent bundle, i.e. a quadratic form on the
tangent bundle that varies holomorphically, or more concretely, an object that, in a local
coordinate (U, z), can be expressed as
φ(z) · dz 2 .
So if (V, w) is another holomorphic chart in which the quadratic differential takes the form
ψ(w) · dw2 , then in the overlap of V and U , we have
 2
dw
ψ(w) · = ϕ(z).
dz

It follows from the Riemann–Roch theorem (see [Fal23]) that, on a closed Riemann surface
of genus g, the space of holomorphic quadratic differentials has complex dimension 1 if g = 1
and 3g − 3 if g ≥ 2.
6.2.1. Example 0: the complex plane and squares of 1-forms. The complex plane
C needs a single chart, so dz 2 is a well defined quadratic differential. Likewise α ⊗ α is
quadratic differential for any holomorphic 1-form α on a Riemann surface X.
6.2.2. Example 1: polygons. Suppose P ⊂ R2 is a Euclidean polygon and we are given
a set of side pairings that are all (compositions of) translations and rotations or angle π.
Let X be the surface we obtain when we glue P together using these side pairings (just
like we’ve seen in the examples of measured foliations above).
First of all, the surface X has a natural Riemann surface structure. Indeed, if we let S ⊂ X
be the set of images of vertices of P . The subsurface X − S ⊂ X admits an atlas consisting
of charts whose transition maps are translations and rotations (with angle π, but that’s
irrelevant for this part). Since such maps are holomorphic, this equips X − S with the
structure of a Riemann surface. So, we only need to explain how to find charts around the
points in S that are compatible with this atlas.
First we observe that X comes with a natural metric that is Euclidean outside of S.
Because the side pairings are translations and rotations of angle π (now the angle does
matter), the total angle around any point in S is mπ for some integer m ≥ 1. Suppose s is
such a singularity, then we build a chart as follows. We take some small disk around s in
the piecewise Euclidean metric (small enough to avoid other singularities). We can think
of this disk as a gluing of m half disks (see Figure 7).
Our goal is to map this disk into a small disk around the origin in C. To this end, we label
the half disks in order: H0 , . . . , Hm−1 that we all parametrize
Hk = r · eiθ : 0 ≤ r < r0 , 0 ≤ θ ≤ π , k = 1, . . . , m

 6.2. HOLOMORPHIC QUADRATIC DIFFERENTIALS 61

Figure 7. A small disk around a singularity.

We map these half disks into sectors of total angle 2π/m in C, by

reiθ ∈ Hk 7→ r2/m ei·(2θ+2kπ)/m ∈ C.
So locally, each of these maps is a branch of the map z 7→ z 2/m . These maps glue together
into a chart around the singularity. We claim that the atlas we obtain from such charts,
combined with the charts of X − S whose transition functions are translations combined
with rotations of angle π. Indeed, the chart transitions are of the form z 7→ z 2/m , that
are holomorphic functions away from 0. Since 0 corresponds to the singular point, which
does not lie in the overlap of two charts, the chart transitions are holomorphic where
they’re supposed to be. X, equipped with this Riemann surface structure, is called a half-
translation surface. If the side pairings are translations (so no rotations of angle π are
involved) we call X a translation surface.
The quadratic differential dz 2 on C is preserved by tranlations and rotations of angle π, so
it descends to a quadratic differential q on X −S. In order to obtain a quadratic differential
on X, we need to say how we define q in the charts around the singularities. Let’s write
(U, w) for the singular chart around a singularity of total angle mπ and (V, z) for some
non-singular chart that has a non-trivial overlap with it. So q(z) = dz 2 . Let’s write
q(w) = ϕ(w) · dw2
for the form q takes in the w-coordinate. By the transformation rule for quadratic differ-
entials,
 2  2
dz dz
ϕ(w) = · ψ(z) = .
dw dw
Because the transition function is of the form z(w) = wm/2 , we obtain that we need to
set 2
m2 m−2 2

2 d m/2
q(w) = ϕ(w)dw = w = w dw .
dw 4
If m = 1, this leads to a pole. We can either allow such poles, or leave these singularities
out.
In conclusion, from a polygon and a set of side identifications given by translations and
rotations of angle π, we obtain a half translation surface equipped with a quadratic differ-
ential and a singular flat metric. Moreover, the singularities of total angle m · π correspond
to zeroes (or poles is m = 1) of order m − 2 of q.
Finally, we observe that if we use only translations for the gluings (and X is a translation
surface), then all the angles are integer multiples of 2π (as opposed to π) and instead of
62 6. MEASURED FOLIATIONS AND QUADRATIC DIFFERENTIALS

only dz 2 , the 1-form dz descends to X − S and can be completed to a globally defined

1-form on X. The quadratic differential we just constructed is its square in this case.
 LECTURE 7

Quadratic differentials, Teichmüller mappings, hyperbolic

geometry

7.1. Quadratic differentials

7.1.1. Measured foliations coming from quadratic differentials. We’ve already
seen that, besides differentials, polygon gluings yield foliations too. This is a general phe-
nomenon: quadratic differentials induce measured foliations and the charts we’ve described
above are natural coordinates.
Indeed, we can define two smooth vector fields V h and V v , where at p ∈ X, the vectors
Vph , Vpv ∈ Tp X are such that
q(Vph ) ∈ (0, ∞) and q(Vpv ) ∈ (−∞, 0)
if p is not a zero or pole of q, and the zero vector if p is. The foliations F h and F v consisting
of the integral lines of V h and V v are called the horizontal and vertical foliations associated
to q.
These foliations come with transverse measures. Indeed, suppose q(z) = ϕ(z)dz 2 in some
local coordinate (U, z) on X, then we can define
Z p  Z p 
h v
µ (α) = Im ϕ(z) dz and µ (α) = Re ϕ(z) dz .
α α

Example 7.1.1. If q is a quadratic differential coming from a half translation structure as

described above, then F h and F v are the foliations induced by the horizontal and vertical
foliation of C respectively. Moreover, the horizontal and vertical transverse measure are
induced by |dy| and |dx| respectively.

A natural question is whether all measured foliations on X appear as the horizontal folia-
tions of quadratic differentials. A theorem due to Hubbard–Masur [HM79] says that this
is essentially (up to a certain equivalence of foliations) the case.
7.1.2. Natural coordinates. Just like measured foliations, quadratic differentials admit
natural coordinates.

Definition 7.1.2. Let X be a Riemann surface. A natural coordinate (U, z) for a quadratic
differential q on X is a holomorphic chart such that q(z) = z k dz 2 for some k ≥ −1.

Such coordinates exist. For example, if p ∈ X is a point at which q does not vanish and
let (U, z) be a chart such that z(p) = 0 and U is small enough such that it doesn’t contain
63
64 7. QUADRATIC DIFFERENTIALS, TEICHMÜLLER MAPPINGS, HYPERBOLIC GEOMETRY

any zeroes of q. Moreover, suppose q(z) = ϕ(z) · dz 2 . We can then compose z : U → C

with the function Z zp
η(z) = ϕ(w) dw,
0
where we take the branch of the square root that gives the correct value at 0. η is a
well-defined biholomorphic function onto its image. Moreover
 2
2 dη
dη = dz 2 = ϕ(z)dz 2 .
dz
So it indeed does what we want. We refer to [Str84, Section 6] for details.

7.1.3. Euclidean metrics. The final fact we will record about quadratic differentials is
that they always come with a singular Euclidean metric.
Indeed, if, locally, q(z) = ϕ(z)dz 2 then
1
|ϕ(z)| dz ∧ dz = |ϕ(z)| dx ∧ dy
2i
and p
|ϕ(z)|1/2 |dz| = |ϕ(z)|1/2 dx2 + dy 2
give a well-defined length and area element respectively, that vanish only at the zeroes of
q.

7.2. Teichmüller mappings

We are now ready to define Teichmüller mappings:

Definition 7.2.1. Let X and Y be two closed Riemann surfaces. We say that a home-
omorphism f : X → Y is a Teichmüller mapping if there are holomorphic quadratic
differentials qX and qY on X and Y respectively, and K > 0 such that:
(1) f maps the zeroes of qX to the zeroes of qY .
(2) If p ∈ X is not a zero of qX , then with respect to the natural coordinates for qX
and qY based at p and f (p), f satisfies
√ 1
f (x + iy) = K x + i √ y.
K

In complex coordinates, the last equation reads:

 
1 K +1 K −1
f (z) = √ ·z+ √ ·z .
2 K K
In particular, f is holomorphic as a function of z if and only if K = 1.

Lemma 7.2.2. If X and Y are closed Riemann surfaces and f : X → Y is a Teichmüller

mapping, then the total area of (the singular Euclidean metrics corresponding to) qX and
qY are the same.
 7.3. HYPERBOLIC SURFACES 65

Proof. In the natural coordinates for qX and qY , the area forms associated to these
quadratic differentials are the usual Euclidean area forms. In these same coordinates, the
Jacobian of f has determinant 1, which means that f is area preserving with respect to
these two area forms. □

We are now ready to state Teichmüller’s theorem:

Theorem 7.2.3 (Teichmüller’s theorem). Let X and Y be closed Riemann surfaces and
let f : X → Y be a homeomorphism. Then the following holds.
• The homotopy class of f contains a Teichmüller mapping h : X → Y .
• If f is quasiconformal then
Kf ≥ Kh
with equality if and only if f ◦ h−1 is a biholomorphism. In particular, if g ≥ 2
this means that f = h.

If time permits, we will prove this in a future lecture. We will however first continue
studying the topology of Teichmüller space.

7.3. Hyperbolic surfaces

Hyperbolic geometry is a powerful tool in the study of Teichmüller spaces of surfaces of
higher genus. Indeed, we will use it to prove that Teichmüller space is a ball. Before we get
to that, we start by recalling some of the basics of the hyperbolic geometry of surfaces. We
have already seen some of what follows in the first problem set, so we refer to this problem
set for some of the proofs. Our main goal will be to show how to use pants decompositions
to build hyperbolic surfaces
7.3.1. The hyperbolic plane. Hyperbolic geometry originally developed in the early
19th century to prove that the parallel postulate in Euclidean geometry is independent of
the other postulates. From this perspective, the hyperbolic plane can be seen as a geomet-
ric object satisfying a collection of axioms very similar to Euclid’s axioms for Euclidean
geometry, but with the parallel postulate replaced by something else. From a more mod-
ern perspective, hyperbolic geometry is the study of manifolds that admit a Riemannian
metric of constant curvature −1.
7.3.2. The upper half plane model. From the classical point of view, any concrete
description of the hyperbolic plane is a model for two-dimensional hyperbolic geometry, in
the same way that R2 is a model for Euclidean geometry.
As we’ve already mentioned in Lecture 1, the hyperbolic plane can be fined as follows.

Definition 7.3.1. The hyperbolic plane H2 is the complex domain

H2 = { z ∈ C : Im(z) > 0 }
equipped with the Riemannian metric given by
1
ds2 = 2 dx2 + dy 2


y
66 7. QUADRATIC DIFFERENTIALS, TEICHMÜLLER MAPPINGS, HYPERBOLIC GEOMETRY

at x + iy ∈ H2

The group of orientation preserving isometries of (H2 , ds2 ) coincides with the group of
complex automorphisms of H2 . That is,
Isom+ (H2 ) = PSL(2, R).
Moreover, we’ve already seen during the exercises that the associated distance function is
given by !
2
|z − w|
d(z, w) = cosh−1 1 + .
2 · Im(z) · Im(w)

Finally, we have also seen in the exercises what geodesics look like:

Proposition 7.3.2. The image of a geodesic γ : R → H2 is a vertical line or a half circle

orthogonal to R. Moreover, every vertical line and half circle orthogonal to the real line
can parameterized as a geodesic.

We will often forget about the parametrization and call the image of a geodesic a geodesic
as well. Note that it follows from the proposition above that given any two distinct points
z, w ∈ H2 there exists a unique geodesic γ ⊂ H2 so that both z ∈ γ and w ∈ γ. Further-
more, it also follows given a point z ∈ H2 and a geodesic γ that does not contain it, there
is a unique perpendicular from z to γ (a geodesic γ ′ that intersects γ once perpendicularly
and contains z)
The final fact we will need about the hyperbolic plane is:

Proposition 7.3.3. Let z ∈ H and let γ ⊂ H2 be a geodesic so that z ∈

/ γ. Then
d(z, γ) := inf { d(z, w) : w ∈ γ }
is realized by the intersection point of the perpendicular from z to γ.
Likewise, any two geodesics that don’t intersect and are not asymptotic to the same point
in R ∪ {∞} have a unique common perpendicular. Moreover, this perpendicular minimizes
the distance between them.

Proof. The first claim follows from Pythagoras’ theorem for hyperbolic triangles.
Indeed, given three points in H2 so that the three geodesics through them form a right
angled hyperbolic triangle with sides of length a, b and c (where c is the side opposite the
right angle), we have
cosh(a) cosh(b) = cosh(c).
Indeed, this can be computed directly by putting the triangle in some standard position
and then using the distance formula, a computation that we leave to the reader. This
means in particular that c > b.
So, any other point on γ is further away from z than the point w realizing the perpendicular.
Because that other point forms a right angled triangle with w and z.
The second claim follows from the first. □
 7.3. HYPERBOLIC SURFACES 67

Figure 1. The Farey tesselation.

7.3.3. The disk model. Another useful model, especially if one likes compact pictures,
is the disk model of the hyperbolic plane. Set
∆ = { z ∈ C : |z| < 1 } .
The map f : H2 → ∆ given by
z−i
f (z) =
z+i
is a biholormorphism. We can also use it to push forward the hyperbolic metric to ∆. A
direct computation tells us that the metric we obtain is given by
dx2 + dy 2
ds2 = 4 .
(1 − x2 − y 2 )2
Since f is conformal, the angles in the disk model are still the same as Euclidean an-
gles.
Using the fact that the map f above is a Möbius transformation and thus sends circles and
lines to circles and lines, one can prove:

Proposition 7.3.4. The hyperbolic geodesics in ∆ are

• straight diagonals through the origin 0 ∈ ∆
• C ∩ ∆ where C ⊂ C is a circle that intersects ∂∆ orthogonally.

For example, Figure 1 shows a collection of geodesics in ∆, known as the Farey tesselation.

7.3.4. Hyperbolic surfaces. A hyperbolic surface will be a finite type surface equipped
with a metric that locally makes it look like H2 .
Because we will want to deal with surfaces with boundary later on, we need half spaces.
Let γ ⊂ H2 be a geodesic. H2 ∖ γ consists of two connected components C1 and C2 . We
will call Hi = Ci ∪ γ a closed half space (i = 1, 2). So for example
z ∈ H2 : Re(z) ≤ 0

68 7. QUADRATIC DIFFERENTIALS, TEICHMÜLLER MAPPINGS, HYPERBOLIC GEOMETRY

is a closed half space.

We formalize the notion of a hyperbolic surface as follows:

Definition 7.3.5. A finite type surface S with atlas (Uα , φα )α∈A is called a hyperbolic
surface if φα (Uα ) ⊂ H2 for all α ∈ A and
1. for each p ∈ S there exists an α ∈ A so that p ∈ Uα and
- If p ∈ ∂S then
φα (Uα ) = V ∩ H
for some open set V ⊂ H2 and some closed half space H ⊂ H2 .
- If p ∈ S̊ then φα (Uα ) ⊂ H2 is open.
2. For every α, β ∈ A and for each connected component C of Uα ∩ Uβ we can find
a Möbius transformation A : H2 → H2 so that
φα ◦ φ−1
β (z) = A(z)

for all z ∈ φβ (C) ⊂ H2 .

Note that every hyperbolic surface comes with a metric: every chart is identified with an
open set of H2 which gives us a metric. Because the chart transitions are restrictions of
isometries of H2 , this metric does not depend on the choice of chart and hence is well
defined.

Definition 7.3.6. A hyperbolic surface S is called complete if the induced metric is com-
plete.

We have seen in Lecture 2 that complete hyperbolic surfaces without boundary (considered
up to isometry) correspond one-to-one to Riemann surfaces (considered up to biholomor-
phism).
7.3.5. Right angled hexagons. Even though Definition 7.3.5 is a complete definition,
it is not very descriptive. In what follows we will describe a concrete cutting and pasting
construction for hyperbolic surfaces.
We start with right angled hexagons. A right angled hexagon H ⊂ H2 is a compact simply
connected closed subset whose boundary consists of 6 geodesic segments, that meet each
other orthogonally. The picture to have in mind is displayed in Figure 2.
It turns out that the lengths of three non-consecutive sides determine a right angled
hexagon up to isometry.

Proposition 7.3.7. Let a, b, c ∈ (0, ∞). Then there exists a right angled hexagon H ⊂
H2 with three non-consecutive sides of length a, b and c respectively. Moreover, if H ′ is
another right angled hexagon with this property, then there exists a Möbius transformation
A : H2 → H2 so that
A(H) = H ′ .
 7.3. HYPERBOLIC SURFACES 69

H2

γ6

γ1
H

γ2 γ3
γ4 γ5

Figure 2. A right angled hexagon H.

Proof. Let us start with the existence. Let γim denote the positive imaginary axis
and set
B = z ∈ H2 : d(z, γim ) = c .


B is a one-dimensional submanifold of H2 . Because the map z 7→ λz is an isometry that

preserves γim for every λ > 0, it must also preserve B. This means that B is a (straight
Euclidean) line.
Now construct the following picture:

H2

B
a γ α
x

Figure 3. Constructing a right angled hexagon H(a, b, c).

That is, we take the geodesic though the point i ∈ H2 perpendicular to γim and at distance
a draw a perpendicular geodesic γ. furthermore, for any p ∈ B, we draw the geodesic α
that realizes a right angle with the perpendicular from p to γim . Now let
x = d(α, γ) = inf { d(z, w) : z ∈ γ, w ∈ α } .
Because of Proposition 7.3.3, x is realized by the common perpendicular to α and γ. By
moving p over B, we can realize any positive value for x and hence obtain our hexagon
H(a, b, c).
70 7. QUADRATIC DIFFERENTIALS, TEICHMÜLLER MAPPINGS, HYPERBOLIC GEOMETRY

We also obtain uniqueness from the picture above. Indeed, given any right angled hexagon
H ′ with three non-consequtive sides of length a, b and c, apply a Möbius transformation
A : H2 → H2 so that the geodesic segment of length a starts at i and is orthogonal to
the imaginary axis. This implies that the geodesic after a gets mapped to the geodesic γ.
Furthermore, one of the endpoints of the geodesic segment of length c needs to lie on the
line B. We now know that the the geodesic α before that point needs to be tangent to
B. Because α and β have a unique common perpendicular. The tangency point of α to
B determines the picture entirely. Because the function that assigns the length x of the
common perpendicular to the tangency point is injective, we obtain that there is a unique
solution. □

7.4. The universal cover of a hyperbolic surface with boundary

It will be useful to have a description of the Riemannian universal cover of a surface with
boundary. To this end, we first prove:

Proposition 7.4.1. Let X be a hyperbolic surface with non-empty boundary that consists
of closed geodesics. Then there exists a complete hyperbolic surface X ∗ without boundary
in which X can be isometrically embedded so that X is a deformation retract of X ∗ .

Proof. For each ℓ ∈ (0, ∞), we define a hyperbolic surface

Fℓ = [0, ∞) × R/{t ∼ t + 1},

equipped with the metric

ds2 = dρ2 + ℓ2 cosh2 (ρ) · dt2

for all (ρ, t) ∈ Fℓ . We will call such a surface a funnel. One can check that this is a metric
of constant curvature −1, in which the boundary is totally geodesic. Alternatively, we can
identify
  
 2 eℓ/2 0
Fℓ = z ∈ H : Re(z) ≥ 0 .
0 eℓ/2

We can glue funnels of the right length along the boundary components, in a similar way
to Example 8.1.1. Figure 4 shows an example.
 7.5. PAIRS OF PANTS AND GLUING 71

Fℓ2
Fℓ1
X

Fℓ3

Figure 4. Attaching funnels

Since both Fℓ and X are complete, the resulting surface X ∗ is complete.

Moreover, since Fℓ retracts onto its boundary component, X is a deformation retract of
X ∗. □

See [Bus10, Theorem 1.4.1] for a version of the above to surfaces with more general types
of boundary components.
Recall that a subset C ⊂ M of a Riemannian manifold M is called convex if for all p, q ∈ C
there exists a length minimizing geodesic γ : [0, d(p, q)] → M such that
γ(0) = p, γ(d(p, q)) = q and γ(t) ∈ C ∀ t ∈ [0, d(p, q)].

As a result of this construction we obtain:

Proposition 7.4.2. Let X be a complete hyperbolic surface with non-empty boundary that
consists of closed geodesics. Then the universal Riemannian cover of X
e of X is isometric
2
to a convex subset of H whose boundary consists of complete geodesics.

Proof. The Killing-Hopf theorem tells us that the universal cover of X ∗ is the hyper-
bolic plane H2 . Here X ∗ is the surface given by Proposition 7.4.1.
Let us denote the covering map by π : H2 → X ∗ . Now let C be a connected component of
π −1 (X). The boundary of C consists of the lifts of ∂X and hence of a countable collection
of disjoint complete geodesics in H2 . As such, it’s a countable intersection of half spaces
(which are convex) and hence convex. □

7.5. Pairs of pants and gluing

One of our main building blocks for hyperbolic surfaces is the following:

Definition 7.5.1. Let a, b, c ∈ (0, ∞). A pair of pants is a hyperbolic surface that is
diffeomorphic to Σ0,3,0 such that the boundary components have length a, b and c respec-
tively.
72 7. QUADRATIC DIFFERENTIALS, TEICHMÜLLER MAPPINGS, HYPERBOLIC GEOMETRY

Proposition 7.5.2. Let a, b, c ∈ (0, ∞) and let P and P ′ be pairs of pants with boundary
curves of lengths a, b and c. Then there exists an isometry φ : P → P ′ .

Proof sketch. There exists a unique orthogonal geodesic (this essentially follows
from Proposition 7.3.3 below, the proof of Proposition 8.3.1 that we will do in full during
the exercises, is similar) between every pair of boundary components of P .
These three orthogonals decompose P into right-angled hexagons out of which three non-
consecutive sides are determined. Proposition 7.3.7 now tells us that this determines the
hexagons up to isometry and this implies that P is also determined up to isometry. □

Note that it also follows from the proof sketch above that the unique perpendiculars cut
each boundary curve on P into two geodesic segments of equal length.
 LECTURE 8

Pants decompositions and annuli

8.1. Non-compact pairs of pants

In order to deal with non-compact surfaces, we will need non-compact polygons. To this
end, we note that, looking at Proposition 7.3.2, complete geodesics in H2 are parametrized
by their endpoints: pairs of distinct point in
∂H2 := R ∪ {∞}
(or S1 if we use the disk model).
A (not necessarily compact) polygon now is a closed connected simply connected subset
P ⊂ H2 , whose boundary consists of geodesic segments.
If two consecutive segments “meet” at a point in ∂H2 , this point will be called an ideal
vertex of the boundary. Note that the angle at an ideal vertex is always 0. A polygon all
of whose vertices are ideal is called an ideal polygon.
We can also make sense of a pair of pants where some of the boundary components have
“length” 0. In this case, we obtain a complete hyperbolic structure on a surface with
boundary and punctures so that
#punctures + #boundary components = 3.
Such pairs of pants can be obtained by gluing either
• two pentagons with one ideal vertex each and right angles at the other vertices,
• two quadrilaterals with two ideal vertices each right angles at the other vertices
or
• two ideal triangles.
Along the sides of infinite length there however is a gluing condition. We will come back
to this later (see Proposition 8.3.3). Moreover, we obtain a similar uniqueness statement
to the proposition above. As always in the non-compact case, the adjective “complete”
does need to be added.

Example 8.1.1. If P is a pair of pants and δ ⊂ ∂P is one of its boundary components, let
us write ℓ(δ) for the length of δ. Given two pairs of pants P1 with boundary components
δ1 , δ2 and δ3 and P2 with boundary components γ1 , γ2 and γ3 so that
ℓ(δ1 ) = ℓ(γ1 ),
73
74 8. PANTS DECOMPOSITIONS AND ANNULI

we can choose an orientation reversing isometry φ : δ1 → γ1 and from that obtain a

hyperbolic surface
S = P1 ⊔ P2 / ∼,
where φ(x) ∼ x for all x ∈ δ1 . One way to see that this surface comes with a well defined
hyperbolic structure, is that locally it’s obtained by gluing two half spaces in H2 together
along their defining geodesics. Note that S is diffeomorphic to Σ0,4,0 .

Repeating the construction above, we can build hyperbolic surfaces of any genus and any
number of boundary components. In what follows we will prove that every hyperbolic
surface can be obtained from this construction.

8.2. The geometry of isometries

Recall that we can classify isometries in PSL(2, R) into three different types:

Definition 8.2.1. Let g ∈ PSL(2, R).

(1) If tr (g)2 < 4 then g is called elliptic.
(2) If tr (g)2 = 4 then g is called parabolic.
(3) If tr (g)2 > 4 then g is called hyperbolic.

Note that, since trace is conjugacy invariant, conjugate elements in PSL(2, R) are of the
same type. It turns out (as we will see below) that closed geodesics correspond exactly to
conjugacy classes of hyperbolic elements.
We’ve seen during the exercises that the classification above can equivalently be described
as:

Lemma 8.2.2. Let g ∈ PSL(2, R). Then

(1) g is elliptic if and only if g has a single fixed point inside H2 .
(2) g is parabolic if and only if g has a single fixed point on R ∪ {∞}.
(3) g is hyperbolic if and only if g has two distinct fixed points on R ∪ {∞}.

Given a hyperbolic isometry, we can define its translation distance as follows:

Definition 8.2.3. Let g ∈ PSL(2, R) be hyperbolic. Then its translation distance is given
by
Tg := inf z ∈ H2 : d(z, gz) .


Moreover, its axis is defined as

z ∈ H2 : d(z, gz) = Tg .

αg :=

We have:
 8.3. GEODESICS AND CONJUGACY CLASSES 75

Lemma 8.2.4. Let g ∈ PSL(2, R) be hyperbolic with fixed points x1 , x2 ∈ ∂H2 . Then its
axis αg is the unique geodesic between x1 and x2 and its translation length is given by
 
−1 |tr (g)|
Tg = 2 cosh .
2

Proof. Since the claim is conjugacy invariant, we can conjugate g so that x1 = 0 and
x2 = ∞. Which means that we can assume without loss of generality that
 
λ 0
g=
0 λ−1
√
for some λ ∈ (0, ∞). Using the fact that 2 cosh( 21 cosh−1 (x)) = 2x + 2, We get that
s r
(λ2 − 1)2 · (Im(z)2 + Re(z)2 ) (λ2 − 1)2 1
2 cosh(d(z, gz)/2) = 4 + 2 2
≥ 4 + 2
=λ+ ,
λ Im(z) λ λ
with equality if and only Re(z) = 0, thus proving the lemma. □

8.3. Geodesics and conjugacy classes

Recall that there is a one-to-one correspondence
   
Conjugacy classes of free homotopy classes of
↔ .
non-trivial elements in π1 (X) non-trivial closed curves on X
We will call a curve puncture parallel if it can be homotoped into a puncture.
It turns out that on a hyperbolic surface (or more generally a negatively curved Rie-
mannian manifold), every free homotopy class of essential closed curves contains a unique
geodesic:

Proposition 8.3.1. Let X be a complete hyperbolic surface with totally geodesic boundary.
(1) Then in every homotopy class of non-puncture parallel closed curves γ on X,
there exists a unique geodesic that minimizes the length among all classes in the
homotopy class.
(2) Moreover, if the free homotopy class contains a simple closed curve, then the
corresponding geodesic is also simple.
(3) More generally, if γ and γ ′ are non-homotopic non-puncture parallel and non-
trivial closed curves, then
• The number of self-intersections of the unique geodesic γ homotopic to γ is
mimimal among all closed curves homotopic to γ and
• #γ ∩ γ ′ is minimal among all pairs of curves homotopic to γ and γ ′ respec-
tively.

Proof. The proof will be part of this week’s exercises. □

We will use the proposition above to prove:

76 8. PANTS DECOMPOSITIONS AND ANNULI

Proposition 8.3.2. Let X be a complete hyperbolic surface with totally geodesic boundary.
Then there are one-to-one correspondences between the following three sets:
 
Non-trivial free homotopy classes of
,
non puncture-parallel closed curves on X
 
Conjugacy classes of
hyperbolic elements in Γ
and  
Oriented, unparametrized
.
closed geodesics on H2 /Γ

Proof. The correspondence between the last and the first set is given by the previ-
ous proposition, so we only need to show that conjugacy classes of hyperbolic elements
correspond one-to-one to oriented, unparametrized geodesics.
In order to make our lives a little easier, we will assume X to be closed. The argument for
the general case is similar. We will hence not worry about the assumption that the curve
is non puncture parallel, nor about boundary components.
First of all consider a conjugacy class K ⊂ Γ of hyperbolic elements. Let us pick an element
g ∈ K, with axis αg ⊂ H2 . The projection map π : H2 → X sends αg to a closed geodesic
of length Tg . Moreover, since
     
π αhgh −1 = π hαg = π αg ,

the resulting geodesic does not depend on the choice of g.

In the opposite direction, a closed geodesic on H2 /Γ lifts to a countable union of geodesics in
H2 (the orbit of a single such geodesic under Γ), each invariant under a cyclic group of deck
transformations. These transformations need to fix the endpoints of the given geodesic, so
they are hyperbolic. The action of Γ on the geodesics corresponds to conjugation of these
hyperbolic elements. □

Before we get to pants decompositions, we record what happens to curves that are parallel
to a puncture.

Proposition 8.3.3. Let X be a complete hyperbolic surface. So that X = C/Γ where C

is a convex subset of H2 , bounded by complete geodesics and Γ < PSL(2, R) acts prorperly
discontinuously and freely on C. Then there are a one-to-one correspondences
   
Conjugacy classes of Oriented, unparametrized
↔ .
parabolic elements in Γ puncture-parallel closed curves H2 /Γ

This proposition gives us the gluing condition we spoke about in Section 7.5: the gluing
needs to be so that the resulting puncture parallel curves give rise to parabolic elements,
this turns out to uniquely determine the gluing.
 8.4. EVERY HYPERBOLIC SURFACE ADMITS A PANTS DECOMPOSITION 77

8.4. Every hyperbolic surface admits a pants decomposition

As an immediate consequence to Proposition 8.3.1 we get that hyperbolic surfaces admit
pants decompositions.

Definition 8.4.1. Let X be a complete, orientable hyperbolic surface of finite area. A

pants decomposition of X is a collection of pairwise disjoint simple closed geodesics P =
{α1 , . . . , αk } in X so that each connected component of
k
!
[
X\ αi
i=1
consists of hyperbolic pairs of pants whose boundary components have been removed.

We have the following:

Lemma 8.4.2. Let P be a pants decomposition of a hyperbolic surface X that is homeo-

morphic to Σg,b,n then
• P contains 3g + n + b − 3 closed geodesics and
• X \ P consists of 2g + n + b − 2 pairs of pants.

Proof. This can be proved using the Euler characteristic. □

Proposition 8.4.3. Let X be a complete, orientable hyperbolic surface of finite area and
totally geodesic boundary. Then X admits a pants decomposition.

Proof. Take any collection of simple closed curves on Σg,b,n that decompose it into
pairs of pants. Proposition 8.3.1 tells us that these curves can be realized by unique
geodesics. □

Note that we actually get countably many such pants decompositions: given a pants decom-
position we can apply a diffeomorphism not isotopic to the identity (of which we already
know there are many) to obtain a new topological pants decomposition, that is realized by
different geodesics.
Finally, we remark, that lengths alone are not enough to determine the hyperbolic met-
ric:

Example 8.4.4. φ in Example 8.1.1 is determined up to ‘twist’. That is, if we parameterize

δ1 by a simple closed geodesic x : R/(ℓ(δ1 )Z) → δ1 and φ′ : δ1 → γ1 is a different orientation
reversing isometry, then there exists some t0 ∈ R so that
φ′ (x(t)) = φ(x(t0 + t))
for all t ∈ R/(ℓ(δ1 )Z) → δ1 .

Summarizing the discussion above, we get the following parametrization of all hyperbolic
surfaces:
78 8. PANTS DECOMPOSITIONS AND ANNULI

Theorem 8.4.5. Let (g, b, n) be so that

χ(Σg,b,n ) < 0.
If we fix a pants decomposition P of Σg,b,n and vary the lengths ℓi ∈ (0, ∞) and twist
τi ∈ [0, ℓi ], we obtain all complete hyperbolic surfaces homeomorphic to Σg,b,n .

Note however that there is no guarantee that we don’t obtain the same surface multiple
times (and in fact we do).

8.5. Annuli
8.5.1. Hyperbolic annuli. Our goal is to use pants decompositions to define global
coordinates on Teichmüller space. In order to prove continuity of the coordinates we obtain,
we need to understand (to some degree) how the complex structure and the hyperbolic
metric depend on each other. It turns out that understanding this for annuli will suffice.
So, before we get to Fenchel–Nielsen coordinates, we will discuss annuli.
If g ∈ PSL(2, R) is a hyperbolic or parabolic isometry then the group ⟨g⟩ ≃ Z acts on H2
properly disconinuously and freely. This means that
Ng = H2 /⟨γ⟩
is an orientable hyperbolic surface with fundamental group Z and hence an annulus. First
we note that the geometry of the annulus only depends on the translation length of g. We
record this as a lemma, the proof of which we leave to the reader.

Lemma 8.5.1. Let g, h ∈ PSL(2, R) be either both hyperbolic or both parabolic elements
so that their translation lengths satisfy Tg = Th . Then the annuli Ng and Nh are isomet-
ric. Moreover, every complete hyperbolic annulus is isometric to Ng for some parabolic or
hyperbolic g ∈ PSL(2, R).

Note that this includes the case where Tg = Th = 0.

8.5.2. Complex annuli. The complex parametrization of annuli we will need is:
Am := { z ∈ C : 0 < Im(z) < m } /Z
for all m > 0. Here the Z-action is given by k · z = z + k for all k ∈ Z, z ∈ C.
We also record a version of Grötzsch’s theorem for these annuli (the proof of which is a
variation of the proof we saw in Section 5.5).

Theorem 8.5.2 (Grötzsch’s theorem). Let f : Am → Am′ be a K-quasiconformal map.

Then
1 m
≤ ′ ≤ K.
K m
Moreover, equality is realized if and only if f can be lifted to a map
fe : { z ∈ C : 0 < Im(z) < m } → { z ∈ C : 0 < Im(z) < m }
given by
′
m
fe(x + iy) = b + x + i y
m
 8.5. ANNULI 79

for some b ∈ R.

We observe that this theorem also implies that Am and Am′ are biholomorphic if and only
if m = m′ . The number m is called the modulus of the annulus.
 LECTURE 9

Fenchel–Nielsen coordinates

9.1. More on annuli

The question now becomes whether Ng is biholomorphic to Am for some m and if so, to
which. In order to solve this question, we introduce a new (somewhat uncommon) model
for the hyperbolic plane the band model. Set
n πo
B = z ∈ C : |Im(z)| < ,
2
equipped with the metric
2 dx2 + dy 2
ds = .
cos2 (y)
This is another model for the hyperbolic plane, moreover the real line is a geodesic in B.
Maps of the from φb : B → B defined by
z 7→ z + b
for some b > 0 are isometries for this metric. Moreover ⟨φb ⟩ ≃ Z acts on B properly
discontinuously, which means that
Mb = B/⟨φb ⟩
is a hyperbolic annulus. Moreover, the translation length of φb is b, so using Lemma 8.5.1,
we see that
Mb ≃ Ng
as hyperbolic surfaces, where g ∈ PSL(2, R) is any hyperbolic element with translation
length b.
We now claim that:

Lemma 9.1.1. Let m > 0. The annuli Am and Mπ/m are biholomorphic.

Proof. Since the map z 7→ z − i m/2 is a biholormophism of C that commutes with

the Z-action. Am is biholomorphic to
n mo
z ∈ C : |Im(z)| < /Z.
2
The map z ∈ C : |Im(z)| < m2 → B given by z 7→ m π

z is a Z-equivariant biholomorphism
and hence descends to a biholomorphism
n mo
z ∈ C : |Im(z)| < /Z ≃ Mπ/m .
2
□
81
82 9. FENCHEL–NIELSEN COORDINATES

For the parabolic case we have:

Lemma 9.1.2. let g ∈ PSL(2, R) be parabolic. The annuli Ng and D \{0} are biholomorphic.

Proof. Using Lemma 8.5.1, we may assume that

 
1 1
g= .
0 1
The map H2 → D given by
z 7→ e−2πiz
induces the biholomorphism. □

9.2. Fenchel-Nielsen coordinates

Now we’re ready to introduce Fenchel–Nielsen coordinates on Teichmüller spaces of hyper-
bolic surfaces. In particular, in this section, we will assume that our base surface S admits
a complete hyperbolic metric. Moreover, we will fix a (topological) pants decomposition
P on S.
9.2.1. Lengths. Given any essential closed curve γ on S, we obtain a function
ℓγ : T (S) → R+
called a length function, defined as follows. Each [R, f ] ∈ T (S) can be seen as a marked
hyperbolic surface. So, Proposition 8.3.1 implies that the homotopy class of f (γ) on R
contains a unique geodesic. ℓγ ([R, f ]) is the length of this geodesic.
Hence, given S and P as above, we obtain a map
ℓP : T (S) → R3g−3+n
+

defined by  
ℓP ([R, f ]) = ℓγ ([R, f ]) .
γ∈P

We have:

Lemma 9.2.1 (Wolpert). Let S and γ be as above. The function

log ◦ℓγ : T (S) → R
is 2-Lipschitz with respect to the Teichmüller metric, i.e.
|log(ℓγ ([R, f ])) − log(ℓγ ([R′ , f ′ ]))| ≤ 2 dT ([R, f ], [R′ , f ′ ])
for all [R, f ], [R′ , f ′ ] ∈ T (S).

Proof. Fix a basepoint p ∈ S so that we can identify γ with an element of π1 (S, p),
that we will also denote by γ. The infinite cyclic subgroup of π1 (S, p) generated by γ
induces a Z-cover
Sγ → S.
We will write A and A for the corresponding covering spaces of R and R′ . Just like in the
′

proof of Proposition 8.3.1, these are annuli and by Lemma 9.1.1, they are biholomorphic
 9.2. FENCHEL-NIELSEN COORDINATES 83

to Aπ/ℓγ ([R,f ]) and Aπ/ℓγ ([R′ ,f ′ ]) respectively. K-quasiconformal maps between R and R′ lift
to K-quasiconformal maps of A and A′ . So we have

2 dT ([R, f ], [R′ , f ′ ]) = inf log(Kg )

g homotopic
to ′ f ◦f −1

≥ inf log(Kge:A→A′ )
g homotopic
to f ′ ◦f −1
 
ℓγ ([R, f ])
≥ log ,
ℓγ ([R′ , f ′ ])

where the last line follows from Grötzsch’s theorem (Theorem 8.5.2). □

9.2.2. Twists. So, given S and P as above, we have a continuous map

ℓP : T (S) → R3g−3+n
+ .

It’s however not quite injective. The problem is that we can still rotate the hyperbolic met-
ric along the curves in the pants decomposition. Twist coordinates will remedy this.

First we pick a collection of disjoint simple closed curves Γ so that for each pair of pants
P in S \ P, Γ ∩ P consists of three arcs, each connecting a different pair of boundary
components of P . Figure 1 shows an example.

Figure 1. A pants decomposition P with a set of curves Γ.

Regardless of our choice of pants decomposition P, such a system of curves Γ always

exists.

Now let γ ∈ P be a pants curve. Then γ bounds either one P or two pairs of pants P1
and P2 in the decomposition. Let us assume the latter for simplicity, the other case is
analogous. The left hand side of Figure 2 shows an example of such a curve γ.
84 9. FENCHEL–NIELSEN COORDINATES

η α1
P1 f
γ δ
P2 α2

Figure 2. The image of an arc under a diffeomorphism.

If f : S → R is an orientation preserving diffeomorphism, then it maps P to some pants

decomposition of R. Moreover, if η is one of the (two) components of (P1 ∪ P2 ) ∩ Γ that
intersects γ, then f (η) is some arc between boundary components of f (P1 ) and f (P2 ) (like
on the right hand side of Figure 2). Now
• δ will be the unique simple closed geodesic in the free homotopy class of f (γ) on
R.
• α1 and α2 the two unique perpendiculars between the boundary components be-
tween which f (η) runs and δ (see Figure 2).
Then relative to the boundary of f (P1 ∪ P2 ), the arc f (η) is freely homotopic to α2 · δ k · α1
for some k ∈ Z.
The twist along γ is now
τγ ([R, f ]) = k · ℓγ ([R, f ]) ± d(p1 , p2 ) ∈ R
where
• p1 and p2 are the points where α1 and α2 hit δ.
• The signs are determined by the orientation of R in the following way. The
orientation of R gives a notion of “twisting to the left” along δ. Left twists are
counted positively and right twists negatively.
Let us prove that twists are continuous:

Lemma 9.2.2. Let S and γ be as above. The function

τγ : T (S) → R
is continuous.

Proof sketch. Suppose that

dT ([R, f ], [R′ , f ′ ])
is small. This means that the map f ′ ◦ f −1 : R → R′ is close to an isometry. Since it maps
the curves and arcs used to define τγ ([R, f ]) to those used to define τγ ([R′ , f ′ ]). So, this
 9.2. FENCHEL-NIELSEN COORDINATES 85

′ ◦ f −1 : H2 → H2 that is close to conformal and hence close to an

map lifts to a map f^
isometry. This means that the numbers τγ ([R, f ]) and τγ ([R′ , f ′ ]) are close. □

Putting the above together, we obtain a continuous map

3g−3+n
FNP : T (S) → R+ × R3g−3+n
defined by  
FNP ([R, f ]) = ℓγ ([R, f ]), τγ ([R, f ]) .
γ∈P

It turns out that the Fenchel Nielssen map is a homeomorphism:

Theorem 9.2.3. Let S be a surface of finite type such that χ(S) < 0 and let P be a pants
decomposition of S. Then the map
FNP : T (S) → R3g−3+n
+ × R3g−3+n ,
is a homeomorphism.

Proof. Since we have already proved that lengths and twists are continuous, we only
need to provide a continuous inverse to the map FNP .
Given a vector (ℓγ , τγ )γ∈P , we can use the gluing construction we discussed above in order
to produce a hyperbolic surface R. The lengths give us the geometry of the pairs of pants
and the gluing along a curve γ is determined by
τγ(0) = τγ + k · ℓγ ,
 
(0)
where k is such that τγ ∈ [0, ℓγ ). Call this surface R (ℓγ , [τγ ])γ . In particular, by varying
the twist τγ , we obtain the same surface countably many times.
 
The question however is what the marking, i.e. the map f : S → R (ℓγ , [τγ ])γ , should
be. In order to do this, we fix open regular neighborhoods NγS of the curves γ ∈ P on S
so that [
S\ Nγ
γ∈P

consists of disjoint pairs of pants P1S , . . . , PkS .

We will once and for all parametrize the
annuli  
NγS = R/Z × (−1, 1).

On R(ℓγ , [τγ ]) we pick such neighborhoods too and obtain neighborhoods NγR and pairs of
pants PiR . We will assume that
n   o
NγR = x ∈ R (ℓγ , [τγ ])γ : d(x, γ) < ε

for some ε small enough. Moreover, we assume ε varies continuously as a function of

(ℓγ , [τγ ])γ .
86 9. FENCHEL–NIELSEN COORDINATES

In order to build f , we now pick a parametrization

 
R
Nγ = R/ℓγ Z × (−1, 1)
where the subset  
R/ℓγ Z × {t} ⊂ NγR
is one of the (one or two) components of
n   o
x ∈ R (ℓγ , [τγ ])γ : d(x, γ) = |t| · ε ,
parametrized by a constant multiple (depending on t) of arclength for all t ∈ (−1, 1).
The map fγ : NγS → NγR is now given by
 
t+1
fγ (θ, t) = ℓγ · θ + τγ · ,t .
2
The awkward (t + 1)/2 is an artifact of choosing the interval (−1, 1) instead of (0, 1) (the
latter would have made some of the previous equations more awkward).
For the complements of the annuli we choose arbitrary homeomorphisms and fiP : PiS →
PiR that smoothly extend the fγ .
This map is clearly an inverse and since we can make everything depend on the input
continuously, it’s continuous. □

Remark 9.2.4. Looking at the proof above, it’s a natural question to ask whether we
maybe get a fundamental domain for moduli space by only considering τγ ∈ [0, ℓγ ).
However, this not the case. To see this, take any f ∈ Diff + (S, Σ) (where S = S0 \ Σ, S0
is closed and Σ a finite set)that is not homotopic to the identity. Then we get a surface
isometric to R (ℓγ , [τγ ])γ∈P if we assign the lengths of the curves in f (P) to the curves
in P instead (the isometry will be induced by f ).
 LECTURE 10

Teichmüller’s theorem and the Nielsen–Thurston classification

10.1. Teichmüller’s theorem

10.1.1. Building Teichmüller maps. The first question is how one builds Teichmüller
maps. Given a Riemann surface X, a quadratic differential qX and K > 0, let us de-
scribe how to build a Riemann surface Y , equipped with a quadratic differential qY and a
Teichmüller map X → Y corresponding to this data.
Let X ′ denote the Riemann surface we obtain if we remove the zeroes of qX from X. We
may equip X ′ with an atlas consisting of natural coordinates for qX . Now compose all of
these coordinates with affine maps
√ 1
f (x + iy) = K · x + √ · iy
K
This yields a new Riemann surface Y ′ (homeomorphis to X ′ through a map we will also
call f ). Moreover, by Riemann’s removable singularities theorem [SS03, Theorem 3.1], we
may extend the Riemann surface structure to a closed Riemann surface Y homeomorphic
to X. qX induces a quadratic differential qY on Y and we obtain an Teichmüller map that
we will also denote f : X → Y .
Fixing X and qX , but varying K ∈ (0, ∞) yields a one parameter family of Riemann
surfaces. If [X, ϕ] ∈ T (S), the resulting family of points in T (S) is called a Teichmüller
line.

10.1.2. An exponential map. Recall from Section 7.1.3 that, if in a local coordinate z,
the quadratic differential q on X is given by q(z) = ϕ(z)dz 2 , then
1
|q(z)| = |ϕ(z)| dz ∧ dz = |ϕ(z)| dx ∧ dy
2i
is a well-defined area form on X. We set
Z
||q|| = |q| .
X

Set
QD1 (X) = { q a quadratic differential on X : ||q|| < 1 } .
By the Riemann–Roch theorem, this has the topology of a ball in C3g−3 .
Now given q ∈ QD1 (X), we set
1 + ||q||
Kq = .
1 − ||q||
87
88 10. TEICHMÜLLER’S THEOREM AND THE NIELSEN–THURSTON CLASSIFICATION

So, using the procedure from the previous section, a quadratic differential q ∈ QD1 (X)
now yields a point [Y, f ] ∈ T (X), where f is a Kq -quasiconformal Teichmüller map. We
will denote this map by
E : QD1 (X) → T (X).
This map will play an important role in our proof of Teichmüller’s theorem.
10.1.3. A proof sketch. We now have all the set-up we need in order to prove Te-
ichmüller’s theorem, that we repeat here:

Theorem 7.2.3 (Teichmüller’s theorem). Let X and Y be closed Riemann surfaces and
let f : X → Y be a homeomorphism. Then the following holds.
• The homotopy class of f contains a Teichmüller mapping h : X → Y .
• If f is quasiconformal then
Kf ≥ Kh
with equality if and only if f ◦ h−1 is a biholomorphism. In particular, if g ≥ 2
this means that f = h.

Proof sketch. The uniqueness part can be proved using a similar string of inequali-
ties to the proof of Grötschz’s theorem (Theorem 5.5.1) and we will skip it (see for instance
[FM12, Section 11.6] for this).
We observe that existence is equivalent to the map E : QD1 (X) → T (X) being surjective.
Indeed, if we want to find a Teichmüller map in the homotopy class of f : X → Y we need
to show that the map E hits the point [Y, f ] ∈ T (X).
The idea is to use Brouwer’s invariance of domain:

Theorem 10.1.1. Let n ≥ 1, then any proper injective continuous map Rn → Rn is a

homeomorphism.

Indeed, E is a map from one homeomorphic copy of R6g−6 to another. So we need to show
that it’s proper, injective and continuous, which will then yield that it’s surjective.
Uniquess of the Teichmüller map in each homotopy class implies injectivity, so we need to
show properness and continuity, the last of which is the hard part.
Indeed, there is no clear geometric reason why nearby quadratic differentials should yield
geometrically similar Teichmüller maps: the behavior of the associated foliations can vary
drastically: arbitrarily close to a quadratic differential with a horizontal foliation all of
whose leaves are closed we can find quadratic differentials with horizontal foliations all
of whose leaves are infinite (think of the rational versus irrational slope examples on the
torus).
Continuity: In order to prove continuity of E, we will decompose it into two maps E 1 and
E 2 . The idea is to first transform our quadratic differential into a Beltrami differential. It
makes sense to do this, because we’ve seen that Teichmüller space can be embedded in a
space of Beltrami differentials. The second step is to go back from Beltrami differentials
to points in Teichmüller space.
 10.1. TEICHMÜLLER’S THEOREM 89

Let us detail both maps. So, the first map turns our quadratic differential q ∈ QD1 (X)
into an almost everywhere defined essentially bounded Beltrami differential on C. Since we
need only one chart to cover C, we can think of the space of such differentials as L∞ (C).
Let us assume that X has genus at least two and write it as X = Γ\H2 (the case of tori
is similar). So we can lift q to a Γ-invariant quadratic differential qe on H2 . We now define
E(q) ∈ L∞ (C)Γ by
(
||q|| · qe(z)/ |e
q (z)| if Im(z) > 0
E 1 (q)(z) = .
||q|| · qe(z)/ |e
q (z)| if Im(z) < 0

In order to obtain a point in Teichmüller space from this Beltrami differntial we now need
to solve the Beltrami equation. Indeed, the idea is to find the quasiconformal map that
induces E 1 (q) as a Beltrami differential. Even if we identified Teichmüller space with a
space of Beltrami differentials before, this step is necessary. Indeed, in our identification
(Corollary 5.2.6), we identified Teichmüller space with a set of Beltrami differentials coming
from quasiconformal maps. In order to go back, we need to know that we can go back:
given a Beltrami differential, it comes from a quasiconformal map. To this end, we state
the Riemann mapping theorem that we already mentioned once in the second lecture:

Theorem 10.1.2. Let µ ∈ L∞ (C) with ||µ||∞ < 1. There exists a unique quasiconformal
homeomorphism f µ : Cb→C b fixing 0, 1 and ∞ that almost everywhere satisfies the Beltrami
equation
∂f µ /∂z
µ= .
∂f µ /∂z
Moreover, f µ is smooth wherever µ is, and f µ varies complex analytically with respect to
µ.

Since ||E 1 (q)||∞ = ||q|| < 1, this theorem gives us a quasiconformal map f E 1 (q) : C b → C.b
E 1 (q)
By uniqueness, f has the same symmetries as E 1 (q) and hence restricts to a Γ-invariant
Beltrami differential on H2 and hence yields a Beltrami differential µf E 1 (q) on X and hence
a point in Teichmüller space T (X) (see Corollary 5.2.6), which we will cal E 2 (E 1 (q)).
E 2 ◦ E 1 is continuous because E 1 is given by a continuous expression and continuity of
E 2 follows from the Riemann mapping theorem. What we still need to explain is that
E = E 2 ◦ E 2.
To this end, suppose at a point q is given by q = reiθ dz 2 then E 1 (q) = ||q|| eiθ . So the
Beltrami equation looks for a function f such that
∂f /∂z
= ||q|| eiθ
∂f /∂z
1+||q||
This is a stretch of 1−||q||
in the direction eiθ (dictated by q), which is exactly what the
map E(q) does.
Properness: Because E is continuous, we know that the pre-image of a closed set is closed.
As such, we just need to check that the pre-image of a compact set in T (X) is bounded in
QD1 (X). This is a direct consequence of the fact that the topology of T (X) is defined in
90 10. TEICHMÜLLER’S THEOREM AND THE NIELSEN–THURSTON CLASSIFICATION

terms of Beltrami differentials. Indeed, on a compact set in T (X), the minimal quasicon-
formal dillatation of maps homtopic to marking changes is uniformly bounded (because
the ∞-norm of their Beltrami differentials is uniformly bounded away from 1). So the
pre-image yields a set of quadratic differentials whose norm is uniformly bounded away
from 1. □

10.2. The Nielsen–Thurston classification on the torus

The next goal of this class will be to describe Thurston’s proof of the Nielsen–Thurston
classification, based on a compactification of Teichmüller space using projective measured
foliations. We note that there is also a proof using the Teichmüller metric, due to Bers,
this can for instance be found in [FM12, Section 13.6] and [Hub16, Chapter 8].
First, we need to discuss the statement of the theorem. For some intuition, recall that if T
denotes the torus, then MCG(T ) ≃ SL(2, Z) and elements in this group are either elliptic,
parabolic or hyperbolic.
 
1 0
Thought of as self maps of the torus, both elliptic elements and − are homotopic
0 1
to a map that can be realized as a homotopically non-trivial isometry/biholomorphism of
the torus corresponding to their fixed point in T (T ) ≃ H2 . We will call such mapping
classes periodic.
 
1 1
Parabolic elements correspond to Dehn twists. Indeed, the matrices ± represent
0 1
the only conjugacy classes of parabolics in SL(2, Z) and they are both represented by Dehn
twists. Moreover, if α is a simple closed curve on T and [φ] ∈ MCG(T ), then the conjugate
of the Dehn Twist Tα by [φ] satisfies
[φ] ◦ Tα ◦ [φ−1 ] = Tφ(α) .
As such all parabolics in SL(2, Z) correspond to Dehn twists. In particular, as mapping
classes they are reducible: they fix a curve.
we have seen in Section 6.1 that if A ∈ SL(2, Z) is hyperbolic, like for instance
Finally, 

2 1
, then it fixes two transverse smooth measured foliations, stretches among one
1 1
and contracts by the same amount along the other. Such maps are called Anosov. They
are also Teichmüller maps from a torus represented by a point on the axis of A in H2
(now thought of as a Riemann surface) to itself with equal initial and terminal quadratic
differential.

10.3. Definitions and statement of the classification theorem

We now first need to generalize the definitions to higher genus surfaces. Singular foliations
will be called transverse if their singularities cooincide and their non-singular leaves are
transverse to each other. Moreover, we generalize the terminology from the torus in the
following way:

Definition 10.3.1. Let [φ] ∈ MCG(Σg ). Then

 10.4. AN OVERVIEW OF THE PROOF 91

• [φ] is called periodic if it has finite order in MCG(Σg ),

• [φ] is called reducible if there exists some finite collection of disjoint pairwise non-
homotopic homotopy classes of simple closed curves γ1 , . . . , γk on Σg such that
[φ]({γ1 , . . . , γk }) = {γ1 , . . . , γk },
and
• [φ] is called pseudo-Anosov if there exists a representative ϕ ∈ [φ], a real number
λ > 1 and a pair of transverse measured foliations (F u , µu ) and (F s , µs ) called the
unstable and the stable foliation respectively, such that
ϕ · (F u , µu ) = (F u , λµu ) and ϕ · (F s , µs ) = (F s , λ−1 µs ).
We note that the pseudo-Anosov representative is not unique: we can isotope the
foliations around and conjugate the map by that isotopy. However, it turns out
that any two representatives are related by an isotopy [FLP79, Exposé 12, Thm
III].

Again, the latter can be phrased in terms of Teichmüller mappings. [φ] is pseudo-Anosov
if and only if it contains a representative ϕ : Σg → Σg for which there exists a point
[X, f ] ∈ T (Σg ) such that
f ◦ ϕ ◦ f −1 : X → X
is a Teichmüller mapping.
The content of the Nielsen–Thurston classification is that the options above are the only
options:

Theorem 10.3.2 (Nielsen–Thurston classification of mapping classes). Let g ≥ 1 and

[φ] ∈ MCG(Σg ). Then [φ] is either periodic, reducible or pseudo-Anosov.

10.4. An overview of the proof

Thurston’s full proof of Theorem 10.3.2 is worked out in [FLP79, FLP12] (see also
[Baa21]). We will content ourselves with sketching the main ideas.
Let us start with an overview of the proof. The idea will be to try to mimic the classifica-
tion on elements of MCG(Σ1 ) ≃ SL(2, Z) acting on T (Σ1 ) ≃ H2 into elliptic, parabolic and
hyperbolic elements. This classification can be phrased in terms of the fixed points of ele-
ments of SL(2, Z) on H2 = H2 ∪∂H2 , which is homeomorphic to a closed disk. In particular,
Brouwer’s fixed point theorem guarantees fixed points for all elements of SL(2, Z).
We have already seen that in the higher genus case, T (Σg ) is homeomorphic to a closed
ball. So, in order to run a similar argument, we need to find a way to attach a boundary
to T (Σg ) to which the mapping class group action continuously extends.
This boundary will be the space of projective measured foliations on Σg . We will first define
the space of measured foliations M F(Σg ), which as a set will be measured foliations up
to isotopy and Whitehead moves (see Figure 1) that either collapse two singularities into a
92 10. TEICHMÜLLER’S THEOREM AND THE NIELSEN–THURSTON CLASSIFICATION

single one or split a singulartiy into two singularities. Note that because of the invariance
of the associated measure under leafwise isotopy, the measure is preserved by such a move.
The projectivization P M F(Σg ) = M F(Σg )/R>0 of the space of measured foliations,

Figure 1. A Whitehead move, turning two 3-pronged singularities into a

single 4-pronged singularity

where R>0 acts by multiplying the measure, turns out to be homeomorphic to a 6g − 7

dimensional sphere.
After that we will consider the set
n o.
C = essential simple closed curves on Σg isotopy

and the space P RC>0 . It turns out that T (Σg ) and P M F(Σg ) both inject into P RC>0 .



Moreover, in the image, P M F(Σg ) appears as the boundary of T (Σg ) and the union is a
closed (6g − 6)-dimensional ball, on which MCG(Σg ) acts continuously. So, we get what
we wanted: by Brouwer’s fixed point theorem, every element will have a fixed point. Like
in the case of SL(2, Z), we obtain the classification by analyzing these fixed points.

10.5. The space of measured foliations

10.5.1. Measured foliations on a pair of pants. Let P be a pair of pants and let us
define the set
 
no boundary component of
M F 0 (P ) = (F, µ) a measured foliation on P : .
P is a non-singular leaf
We claim that M F 0 (P ) can be identified with R3≥0 − {0}. Indeed, the claim is that for
any (a, b, c) ∈ R3≥0 , there exists a unique measured foliation (F, µ), up to isotopy and
Whitehead moves, such that the µ-lengths of the boundary components equals a, b and c
respectively.
Indeed, up to Whitehead moves and isotopy and permuting the roles of the boundary
components, the only two options for the foliation are shown in Figure 2.
Using the invariance of µ under leafwise isotopy, we see that a + b ≥ c for the foliation on
the left and a + b ≤ c for the foliation on the right.
To see that all elements of M F 0 (P ) are indeed of this form, we use the Euler–Poincaré
formula (Proposition 6.1.2), which implies that any foliation of P has at most two three-
pronged singularity or one four-pronged singularity, which can be split into two three-
pronged singularities using a Whitehead move. Afterwards it’s a matter of where the
singular leaves go.
 10.5. THE SPACE OF MEASURED FOLIATIONS 93

Figure 2. The two options for the foliation underlying an element of

M F 0 (P ).

δi εi
γi

Figure 3. A pants curve γi and two arcs δi and εi .

We obtain all elements of M F(P ) by attaching annuli foliated by parallel circles to the
elements of M F 0 (P ).
10.5.2. Coordinates on M F(Σg ). In order to obtain global coordinates on M F(Σg ),
we will use a construction not dissimilar to that of Fenchel–Nielsen coordinates. Indeed,
3g−3
we will fix a pants decomposition (γi )i=1 of Σg .
Now fix an annulus Ai parallel to each pants curve γi such that γi is one of the boundary
components of Ai . Moreover, fix two arcs δi and εi that triangulate Ai (see Figure 3).
Now given a measured foliation (F, µ), we can put it in normal form with respect to this
choice of annuli (using isotopy and Whitehead moves). This means that:
• the foliation restricts to an element of M F 0 (P ) on the complements of the Ai ,
and
• in the annuli Ai , the foliation is either transverse to γi or it consists of leaves
parallel to γi . In particular, it contains no singularities in Ai .
9g−9
We obtain an injective map M F(Σg ) → R≥0 by recording the measures of the γi , δi and
εi . Because any arc transverse to γi hits either εi of δi however, we have that
µ(γi ) = µ(δi ) + µ(εi ),
so the image lands in a copy of R6g−6 . This endows M F(Σg ) with a topology and indeed
gives P M F(Σg ) the topology of S6g−7 .
94 10. TEICHMÜLLER’S THEOREM AND THE NIELSEN–THURSTON CLASSIFICATION

10.6. Compactifying Teichmüller space

10.6.1.
Two maps. Like we said above, we will map both T (Σg ) and P M F(Σg ) into
P RC≥0 and will identify the latter as the image of the former.
The map from Teichmüller

space is given by lengths of curves. That is, we will use the
C
map ℓ : T (Σg ) → P R≥0 , given by
h i
ℓ([X, f ]) = (ℓα ([X, f ]))α∈C .
To show that this is a homeomorphism onto its image, one can use Fenchel–Nielsen co-
ordinates. Indeed, if we fix any pants decomposition, we immediately recover the lengths
corresponding to it from the image, up to a constant factor. Using enough curves that
intersect the curves in the pants decomposition, we can also recover the constant factor
and the twists (it turns out that a finite number of curves suffices).
In order to define the map from P M F(Σg ) into P RC≥0 , we need to define the length of


a simple closed curve with respect to a measured foliation. We will define this by
 
I (F, µ), α = inf { µ(eα) : αe isotopic to α } , (F, µ) ∈ M F(Σg ), α ∈ C .
This defines a map
I : P M F(Σg ) → P RC≥0 .

This map is a homeomorphism onto its image. This again comes down to proving that we
can recover the coordinates we described above (up to a scaling factor) from the image of
a point.
 LECTURE 11

Landmark results, part I

11.1. Compactifying Teichmüller space, part II

We start this lecture by continuing our discussion of the Nielsen–Thurston classification
theorem (Theorem 10.3.2). Just like last time, this is only meant to be a sketch of the
ideas. Details can be found in [FLP79, FLP12, Baa21].
 
Now we need to show that the union of the images of the two maps ℓ : T (Σg ) → P RC≥0
 
and I : M F(Σg ) → P RC≥0 described above is a closed (6g − 6)-dimensional ball whose
interior corresponds to T (Σg ) and whose boundary is the image of P M F(Σg ).
We will first prove that the images are disjoint. To see this, we note that the systole – the
length of the shortest closed geodesic
– of a closed hyperbolic surface is a strictly positive
number. As such, if a point in P RC≥0 comes from a point in T (Σg ) then there is a uniform
lower bound on the coordinates of any representative in RC≥0 of that point.
On the other hand, we claim that this is not the case for points in the image of P M F(Σg ).
Indeed, suppose [(F, µ)] ∈ P M F(Σg ) there are two cases to consider:
(1) If F has a closed leaf λ. This is a simple closed curve, the intersection of which
with F equals 0.
(2) If F doesn’t have any closed leaves, then take any non-singular leaf of F. Because
Σg is compact, this leaf needs to come back arbitrarily close to itself. As such, we
can create simple closed curves with arbitrarily small intersection with (F, µ) by
tracking such a non singular leaf and closing up the tracjectory once we’re very
close to a point on the leaf we have already visited.
11.1.1. Projective measured laminations
 as the boundary
  of Teichmüller space.
In order to see that I P M F(Σg ) is the boundary of ℓ T (Σg ) , we first construct a map
from
p : T (Σg ) −→ M F(Σg ).
3g−3
To do this, we pick a pants decomposition (αi )i=1 of Σg and measure Fenchel-Nielssen
coordinates with respect to this pants decomposition.
The triple of lengths corresponding to a pair of pants give us an element of M F 0 (P ). We
use the twists to decide how to glue these foliations together into a foliation of Σg .
The key observation is the following. Given ε > 0, define
U (ε) = { [X, f ] ∈ T (Σg ) : ℓαi ([X, f ]) > ε for i = 1, . . . , 3g − 3 } .
95
96 11. LANDMARK RESULTS, PART I

Now for every ε > 0 and every α ∈ C, there exists a C(α, ε) > 0 such that
I(p([X, f ]), α) ≤ ℓα ([X, f ]) ≤ I(p([X, f ]), α) + C(α, ε)
   
In particular, if a sequence [Xn , fn ] ∈ U (ε) is such that it’s image ℓ [Xn , fn ] ∈ RC≥0
n
tends to infinity, then the projectivized secquence
h  i
ℓ [Xn , fn ] ∈ P RC≥0

converges if and only if the sequence

h  i
I p([Xn , fn ]) ∈ P RC≥0

converges. Moreover, if they do, then the limits are the same. So this implies that we can
identify limits of of sequences of points in U (ε) with points in P M F(Σg ). By changing
the pants decomposition, we can contain any sequence in U (ε) for some ε > 0. This proves
the identification.
We will write    
T (Σg ) = ℓ T (Σg ) ∪ I P M F(Σg ) .

11.1.2. The topology of T (Σg ). It turns out that T (Σg ) is a manifold. Moreover,
a theorem by Brown implies that P M F(Σg ) admits a collar neighborhood in T (Σg ).
The boundary of this collar is homeomorphic to S6g−7 . Since it lies in T (Σg ), which is
homeomorphic to R6g−6 , the generalized Schönflies theorem, proved by Brown, Mazur and
Morse, the interior of this copy of S6g−7 is a ball (there is a regularity condition to check
here, that we skip over). This implies that T (Σg ) is homeomorphic to a ball.

11.1.3. The case of the torus. In order to have a concrete example, we will treat the
case of the torus. In the higher genus case, we used the fact that the lengths of finitely
many curves suffice to determine a point in Teichmüller space.
In the case of the torus, three lengths suffice. That is, we identify Σ1 = R2 /Z2 and will
let α1 , α2 and α3 be the curves corresponding to (1, 0), (0, 1), (1, 1) ∈ Z2 respectively. We
obtain a map
(ℓα1 , ℓα2 , ℓα3 ) : T (Σ1 ) −→ R3>0 .
Moreover, because, when we base them at the same point, α1 , α2 and α3 form a (non-
degenerate) triangle, we have
ℓαi ([X, f ]) < ℓαj ([X, f ]) + ℓαk ([X, f ]) for all distinct i, j, k ∈ {1, 2, 3}
and for all [X, f ] ∈ T (Σ1 ). Moreover, we have also seen that we may rescale say the
 length

3 3
of α1 to be 1. This is equivalent to saying that the projectivization map R>0 → P R>0 is
 
injective on the image of T (Σ1 ). We can identify the image P R3>0 of the projectivization
map with ∆ = { (x1 , x2 , x3 ) ∈ R3>0 : x1 + x2 + x3 = 1 }. Teichmüller space sits in here as
the subset
{ (x1 , x2 , x3 ) ∈ ∆ : xi < xj + xk for all i, j, k } .
 11.2. CLASSIFYING MAPPING CLASSES 97

The map M F(Σ1 ) → R3≥0 defined by

[F, µ] 7→ (µ(α1 ), µ(α2 ), µ(α3 ))
is injective. Moreover, for the same reason as in Section 10.5.2,
µ(αi ) = µ(αj ) + µ(αk ) for one choice of distinct i, j, k ∈ {1, 2, 3}.
So P M F(Σ1 ) can be identified with
{ (x1 , x2 , x3 ) ∈ ∆ : xi = xj + xk for some choice of i, j, k } ,
i.e. the boundary of the image of Teichmüller space. Moreover, their union is indeed a
closed disk.

11.2. Classifying mapping classes

11.2.1. The mapping class group action. The mapping class group MCG(Σg ) acts
both on Teichmüller space and on projective measured foliations. It turns out that these
two action combine into a continuous action on T (Σg ). Indeed, in the image, this is action
is the action by permuting the elements of C.
Since T (Σg ) Brouwer’s fixed point theorem implies that every element of MCG(Σg ) has
at least one fixed point in T (Σg ). Now we need to analyze the fixed points of elements of
MCG(Σg )

11.2.2. Analyzing the fixed points. Let [φ] ∈ MCG(Σg ). We will distinguish the
following cases:
(1) [φ] · [X, f ] = [X, f ] for some [X, f ] ∈ T (Σg ).
(2) The fixed point is some element [F, µ] ∈ P M F(Σg ). Since MCG(Σg ) acts on
M F(Σg ) as well, we may lift to a representative [F, µ] ∈ M F(Σg ). This point
is no longer necessarily a fixed point, because [φ] might multiply the measure. As
such, we make a further case distinction:
(a) [φ] · [F, µ] = [F, λ · µ] for some λ ∈ [1, ∞) and F has at least one closed leaf.
(b) [φ] · [F, µ] = [F, µ] and F has no closed leaves.
(c) [φ] · [F, µ] = [F, λ · µ] for some λ > 1 and F has no closed leaves.
In the above, we have assumed that λ ≥ 1, which we may, by potentially replacing
[φ] by its inverse.

11.2.3. Case 1. In this case (X, f ) and (X, f ◦ φ−1 ) are Teichmüller equivalent. This
means that there exists an orientation preserving isometry (or equivalently a complex
automorphism) m : X → X such that
(f ◦ φ−1 )−1 ◦ m ◦ f = φ ◦ (f −1 ◦ m ◦ f )
is isotopic to the identity on Σg . Since m : X → X is a periodic map (because auto-
morphism groups of higher genus surfaces are finite, by Hurwitz’s theorem – see Theorem
12.1.1 below), this means that [φ] is homotopic to a periodic map.
98 11. LANDMARK RESULTS, PART I

Figure 1. A full system of transversals near a three-pronged singularity.

11.2.4. Case 2a. The foliation F will have at most finitely many isotopy classes of closed
leaves. If φ preserves it, then it needs to permute these finitely many istopy classes. In
particular, this means that [φ] is reducible.
11.2.5. Case 2b. If both the foliation F and the measure µ are preserved by [φ], then
we can build a finite system of rectangles on Σg that are permuted by [φ], which implies
that [φ] is periodic. This goes as follows.
First of all, a horizontal rectangle R ⊂ Σg with respect to a singular foliation F is an
immersion ϕ : [0, 1]2 ,→ Σg such that:
• The restriction of ϕ to (0, 1)2 is an embedding.
• For all t ∈ [0, 1], the set ϕ([0, 1] × {t}) is contained in a finite union of leaves of F.
• For all s ∈ [0, 1], the set ϕ({s} × [0, 1]) is transverse to F.
We will write ∂ T R for the transverse boundary of R: the set ϕ({0, 1} × [0, 1]).
We now choose transversal arcs τ1 , . . . , τn to F such that:
• all these arcs have exactly one endpoint in the singularity,
• they have no other endpoints in common,
• there is exactly one arc between every pair of prongs of each singularity and
• the arcs based at a given singularity all have the same µ-length.
See Figure 1 for a local picture near a singularity. We will call this collection of arcs a full
system of transversals for F and we set τ = τ1 ∪ . . . ∪ τn .
We now claim:

Lemma 11.2.1. Given a singular foliation F on Σg without any closed leaves and a full
system of transversals τ , we can find a finite collection of horizontal rectangles R1 , . . . , RN
with respect to F such that
N
[ N
[
Ri = Σg and ∂ T Ri = τ
i=1 i=1

Proof. Using the full system of transversals τ = ∪i τi , we define a finite set V of

“vertices” as follows:
• Both endpoints of τi are elements V for all i,
 11.2. CLASSIFYING MAPPING CLASSES 99

• the first intersection point with τ of any leaf ℓ of F starting at any of the endpoints
of any of the τi is an element of V .
The leaves running between the vertices (this includes all the finitely many singular leaves
of F) V delimit a finite system of rectangles, the union of whose transverse boundaries
equals τ . Moreover, the union of these rectangles necessarily equals Σg . Indeed, if it didn’t,
the subsurface spanned by these rectangles would have a boundary component consisting
of finitely many leaves. This boundary component would be homotopic to a closed curve
contained in τ . This is impossible, because the components of τ need to be trees. The
reason for this is that F does not have any closed leaves. Indeed, any closed cycle in
τ would need to be homotopically non-trivial, which would mean F has a closed leave
homotopic to this cycle. □

Now we go back to our mapping class [φ] that preserves [F, µ] ∈ M F(Σg ). We may take
some power [φn ] that preserves the singularities and the singular leaves running between
them. Since φ preserves the measure, we can assume that φn (τ ) = τ . This means that φn
maps each rectangle Ri to itself, fixing the boundary. By the Alexander lemma (Lemma
4.1.3), this means that φn is isotopic to the identity and hence that φ is periodic.

11.2.6. Case 2c. This, to some degree is the “generic” case and will yield pseudo-Anosov
maps. In particular, we need to construct a singular measured foliation (F, b µ
b) transverse
1
to F such that [φ] · (F, µ
b b) = (F, λ µ
b b). The idea is to divide Σg into horizontal rectangles
for F which we will then foliate in the vertical direction to obtain F.
b In order to determine
µ
b, we need to assign widths to these rectangles.
We can still construct a full system τ of transversals to the foliation F. However, because
λ > 1, we can only assume that φ(τ ) ⊂ τ as opposed to the equality we had before. Indeed,
φ∗ µ = λ · µ, so φ needs to contract in the transverse direction to F. We will not directly
build our rectangles based on this, but transform the system of transversals first. This will
guarantee that the rectangles we obtain have the right properties with respect to φ.
Again we let n ≥ 1 be minimal such that φn preserves all the singularities of F and maps
the sectors of each singularity to themselves.
Because the leaves of F are not closed, they are everywhere dense in Σg (compare to the
case of foliations of the torus discussed in Section 6.1.1), once they do not run between two
singularities of the foliation. In particular, every arc τi ⊂ τ is intersected in its interior by
some singular leaf ℓi – the existence of this leaf also uses that τ is a tree. We let αi denote
the segment of ℓi that runs between the singularity at which ℓi is based and the first point
at which ℓi intersects τi .
This allows us to define
L = ∪i ∪nk=1 φk (αi ).
Because φ stretches in the horizontal direction,
L ⊂ φ(L).
We will also shorten the arcs τi such that their endpoints lie on L.
100 11. LANDMARK RESULTS, PART I

Now we use Lemma 11.2.1 to fill Σg with a finite collection of horizontal rectangles
R1 , . . . , RN , using the arcs τi . We have that
N
[
T
φ(∂ Ri ) ⊂ ∂ T Ri .
i=1
P
Likewise, if we let ∂ Ri be the horizontal part of the boundary of Ri (parallel to the
foliation), we have
[N
−1 P
φ (∂ Ri ) ⊂ ∂ P Ri .
i=1

Now write
aij = the number of connected components of φ(int(Ri )) ∩ int(Rj )
and we let A = (aij )ij denote the corresponding N-valued N × N matrix. Because the
leaves of F are dense in Σg , there is some K ∈ N such that AK is a strictly positive matrix.
By the Perron–Frobenius theorem, this means A has a unique eigenvalue ρ of maximal
modulus. We let v = (v1 , . . . , vN ) be any eigenvector corresponding to that eigenvalue, so
by the same theorem, v1 , . . . , vN > 0.
Now set the width of the rectangle Ri equal to vi . This gives us a transverse measure µ
b on
the vertical foliation F
b coming from the decomposition into rectangles. So,
φ · [F,
b µb] = [F,
b α·µb].
1
To prove that α = λ
, we observe that µ and µ
b together yield an area measure ν on Σg
such that
φ∗ ν = α · λ · ν.
Since Σ is compact and φ is a diffeomorphism, this is only possible if α · λ = 1. So indeed
φ is a pseudo-Anosov map.
 LECTURE 12

Landmark results, part II

12.1. The Nielsen realization problem

We have seen above that if a non-trivial mapping class [φ] ∈ MCG(Σg ) has a fixed point
[X, f ] ∈ T (Σg ), then f ◦ φ ◦ f −1 : X → X is isotopic to an orientation preserving isometry
(or equivalently a conformal automorphism) of X. In particular, if X has a non-trivial
isometry group G, then this yields a subgroup of the mapping class group. Such a subgroup
is always finite because of the following classical result.

Theorem 12.1.1 (Hurwitz). Let X be a closed Riemann surface of genus g ≥ 2 then

# Aut(X) = # Isom+ (X) ≤ 84(g − 1).
Here Isom+ (X) denotes the group of orientation preserving isometries of the hyperbolic
metric on X.

Proof. We will use the hyperbolic metric for this proof. First we observe that if Y is
an orientable hyperbolic 2-orbifold (like PSL(2, Z)\H2 ) then the area of Y equals
k
X 1
area(Y ) = 2π(2g − 2 + n) + 2π · k − ,
v
i=1 i

where g is the genus of Y , n its number of cusps, k its number of cone points and 2π/vi
the angle at the ith cone point. We will prove this formula (in the closed case) in the
exercises. It’s also a consequence of the Gauss-Bonnet theorem. A case by case analysis of
this formula shows that
π
area(Y ) ≥
21
and equality is uniquely realized by the sphere with three cone points such that (v1 , v2 , v3 ) =
(2, 3, 7).
Our goal will be to show that Isom+ (X)\X is a closed orbifold. Indeed, then we can argue
that
  π
4π(g − 1) = area(X) = #Isom+ (X) · area Isom+ (X)\X ≥ #Isom+ (X) · ,
21
which yields the claim.
In order to show that Isom+ (X)\X is indeed an orbifold, we need to show that Isom+ (X)
is finite. This we do by looking at the action of Isom+ (X) on closed geodesics on X.
Indeed, there can only be finitely many closed geodesics on X of bounded length – this
is for instance a consequence of the proper discontinuity of the action of the fundamental
101
102 12. LANDMARK RESULTS, PART II

group of X on its universal cover, combined with the fact that every free homotopy class
of non-trivial closed loops contains a unique geodesic.
So, because isometries preserve the lengths of geodesics, if γ is a closed geodesic on X,
the orbit Isom+ (X) · γ consists of finitely many closed geodesics. Now we just need to
prove that the stabilizer of a given geodesic is finite. To this end, we will use a geodesic γ
with a single self intersection (like in Figure 1). Any isometry stabilizing γ needs to map

Figure 1. A closed curve with a single self intersection

its unique intersection point to itself. Moreover, it needs to map each of the four prongs
emanating from the intersection point to another prong. In fact, the entire isometry is
determined by where the prong goes. So, the size of the stabilizer is at most 4. □

So, we conclude that isometry groups of surfaces yield finite subgroups of their mapping
class groups. The Nielsen realization problem asks whether all finite subgroups of the
mapping class group come from this construction. Indeed this is not obvious from the
proof sketch above, because we had a second source of periodic maps (Case 2a).
Nielsen himself proved that his question has a positive answer if the finite subgroup is
cyclic. The general case was settled by Kerckhoff [Ker83]:

Theorem 12.1.2 (Kerckhoff). Let g ≥ 2 and let G < MCG(Σg ) be a finite subgroup. Then
G can be lifted to a finite subgroup of Diff + (Σg ) and moreover there exists a hyperbolic
metric on Σg for which G acts by isometries.

Proof sketch. Let’s start with proving the case in which G < MCG(Σg ) is a finite
cyclic group. General theorems from topology (for instance using the fact that the group
homology of G is non-trivial in infinitely many degrees) imply that G cannot act freely
on a finite-dimensional contractible cell complex. In particular, if h ∈ G is a generator,
then there is some point [X, f ] ∈ T (Σg ) that is fixed by h (because Teichmüller space is
a contractible cell complex). But since h is a generator, this means that all elements of G
fix [X, f ] and hence that G can be lifted to a group of isometries of X.
For more general finite subgroups G < MCG(Σg ), the topological theorem is still true. So
every element h ∈ G has at least one fixed point. However, for the argument to work,
some of these fixed points need to coincide. The mapping class group preserves much more
structure on Teichmüller space than just its topology.
 12.2. THURSTON’S HYPERBOLIZATION OF MAPPING TORI 103

Figure 2. A filling collection of curves.

For instance it preserves the Teichmüller metric. If this were non-positively curved1 then
that would help us. For instance, let’s consider the Euclidean metric. if some finite group
G acts by isometries on Rn then we the circumcenter –the center of the unique ball of
minimal radius that contains the orbit– of any orbit is a fixed point. This idea generalizes
to more general non-positively curved spaces (including the hyperbolic plane and its higher-
dimensional generalizations). Unfortunately, the Teichmüller metric is not non-positively
curved.
So we need something else. The idea is to use sums of length functions over a filling
collection of curves on Σg . Here, a set of non-trivial free homotopy classes of closed curves
γ1 , . . . , γk on Σg is said to be filling if the complement of the curves consists of disjoint
disks (see Figure 2 for an example).
Given a finite subgroup G < MCG(Σg ), we take a G-invariant filling collection of curves
Γ = {γ1 , . . . , γk } on Σg . Such a collection can be obtained by taking the orbit under G of
a collection of curves that is already filling. Now we define ℓΓ : T (Σg ) → (0, ∞) by
k
X
ℓΓ ([X, f ]) = ℓγi ([X, f ]).
i=1

The main theorem Kerckhoff proved is that such a function admits a unique minimum on
Teichmüller space. Because ℓΓ itself is G-invariant, this minimum is G-invariant, i.e. a
fixed point.
The way Kerckhoff proves that ℓΓ admits a unique minimum is by proving that it’s a strictly
convex function along deformations that are called earthquakes. These deformations are a
generalization of twist deformations. Moreover, between any pair of points in Teichmüller
space, there is a unique earthquake path. This implies that ℓΓ has a unique minimum. □

12.2. Thurston’s hyperbolization of mapping tori

The next big result we’ll discuss is about 3-manifolds. Given an orientation preserving
diffeomorphism f : Σg → Σg , we can build a 3-manifold that we will call the mapping
torus of f . This is the manifold
 .
Mf = Σg × [0, 1] (x, 0) ∼ (f (x), 0) ∀x ∈ Σg .

1There is something to say about how to formalize this, because the Teichmüller metric is not a
Riemannian.
104 12. LANDMARK RESULTS, PART II

Observe that such a manifold admits a map to the circle given by [(x, t)] 7→ [t] ∈ R/Z. We
also observe that if f and f ′ are isotopic, then Mf and Mf′ are homeomorphic. Indeed,
the isotopy is a map F = Σg × [0, 1] → Σ such that F (·, 0) = f and F (·, 1) = f ′ . So the
homeomorphism can be built by attaching an extra “slab” homeomorphic to Σg × [0, 1] to
the manifold before gluing.
Just like in dimension two, one can ask which type of geometric structures one can put on
such a manifold. In dimension two, we had three types of geometries: spherical, euclidean
and hyperbolic.
In dimension three, the situation is more complicated: there are eight types of geometries
(three among them are still spherical, euclidean and hyperbolic) and it’s no longer true
that every manifold can be equipped with a geometric structure. Instead, every closed 3-
manifold can be cut along spheres and tori in such a way that the pieces admit a geometric
structure. This is the content of Perelman’s geometrization theorem (which had been
conjectured by Thurston).
We will discuss an important special case of this theorem, namely mapping tori. Let us
first very briefly discuss 3-dimensional hyperbolic geometry. Like in dimension two, there
are multiple models for hyperbolic 3-space, for instance
dx2 + dy 2 + dz 2
H3 = (x, y, z) ∈ R3 : z > 0 , ds2H3 =

z2
and
dx2 + dy 2 + dz 2
B3 = (x, y, z) ∈ R3 : x2 + y 2 + z 2 < 1 , ds2H3 = 4

(1 − x2 − y 2 − z 2 )2
are the 3-dimensional versions of the upper half plane model and the disk model respec-
tively. More generally, the Killing–Hopf theorem also applies in three dimensions, so every
complete simply connected Riemannian manifold is isometric to H3 . The boundary of H3
can be identified with the Riemann sphere C. b It turns out that the isometry group of
3
H can be identified with the group of complex automorphisms of its boundary. That
is Isom+ (H3 ) ≃ PSL(2, C). Moreover, the non-trivial elements of PSL(2, C) can still be
classified into three categories:
• elliptic elements that have at least one fixed point in H3 ,
• parabolic elements that have infinite order, no fixed points in H3 and exactly one
fixed point in C
b

• hyperbolic elements that have positive translation length, no fixed points in H3

and exactly two fixed points in C.
b

Moreover, the type of an element is again determined by its trace: if tr(g) ∈ (−2, 2) then
g is elliptic, if tr(g) ∈ {−2, 2} then g is parabolic and if neither of the two is true then g
is hyperbolic.
So if M is a closed hyperbolic 3-manifold, then there exists some Γ < PSL(2, C) that
is discrete and torsion free, such that M is isometric to Γ\H3 . In every non-trivial free
homotopy class of closed curves in our closed manifold M , there exists a unique geodesic
and the corresponding conjugacy class of Γ consists of hyperbolic elements.
 12.2. THURSTON’S HYPERBOLIZATION OF MAPPING TORI 105

In general, the study of hyperbolic 3-manifolds is very different from that of hyperbolic
surfaces. Indeed, in dimension 3 (and higher) there is the Mostow rigidity theorem which
states that if M and N are two closed hyperbolic 3-manifolds such that π1 (M ) ≃ π1 (N )
as abstract groups, then M and N are isometric. In other words, the moduli space of
hyperbolic metrics on any closed 3-manifold is a point.
Now we get back to our question of what kind of geometric structure our manifold Mf
admits. We will ask when it admits a hyperbolic structure. First of all we observe:

Lemma 12.2.1. If [f ] ∈ MCG(Σg ) is not pseudo-Anosov, then Mf does not admit a hyper-
bolic metric.

Proof. Our first claim is that, if Mf is not pseudo-Anosov, then Mf admits a finite
cover N → Mf such that π1 (N ) has a subgroup that is isomorphic to Z2 .
Indeed, by the Nielsen–Thurston classification, if f is not pseudo-Anosov, then f is either
periodic or reducible. We start with the periodic case. Suppose that [f ]n is the identity
in MCG(Σg ). Observe that in general Mf n is an n-fold cover of Mf . If f n is the identity
map, that means that Mf n = Σg × S1 . So its fundamental group is π1 (Σg ) × π1 (S1 ), which
contains plenty of copies of Z2 (generated by any pair of non trivial elements g ∈ π1 (Σg )
and h ∈ π1 (S1 ) ≃ Z.
If [f ] is reducible, that implies it has some power [f ]n that preserves a curve γ ⊂ Σg .
Essentially for the same reason as before, any group element in the conjugacy class of
π(Σg ) determined by γ and any non trivial element of π1 (S1 ) together generate a copy of
Z2 .
Now we need to show that the fundamental group Γ of a closed hyperbolic 3-manifold
Γ\H3 cannot contain a subgroup isomorphic to Z2 . There are many ways to do this. We
will use the fact that all non-trivial elements of are hyperbolic. Now if two hyperbolic
elments g, h ∈ PSL(2, C) commute, then they need to preserve each others fixed points.
This means that they translate along the same axis. But then the group ⟨g, h⟩ generated
by g and h can only be discrete if there exist some m, n ∈ Z such that g m = hn . That
means that ⟨g, h⟩ ̸≃ Z2 . □

The surprising fact is that this “obvious” obstruction is the only one:

Theorem 12.2.2 (Thurston’s hyperbolization for mapping tori). Let g ≥ 2 and let [f ] ∈
MCG(Σg ). Then Mf admits a hyperbolic metric if and only if [f ] is pseudo-Anosov.

Otal [Ota96] was the first to provide a full proof of this theorem, based on the ideas by
Thurston. We will not go into the proof, but will just mention that much more recently
it was proved that mapping tori (i.e. 3-manifolds that fiber over the circle) are not as
restrictive a class of manifolds as they may seem at first sight. Indeed, Agol [Ago13], using
work by Wise and many others proved that every closed hyperbolic 3-manifold (by many
measures the wildest class of 3-manifolds) admits a cover of finite degree by a mapping
torus, thus confirming another conjecture by Thurston.
106 12. LANDMARK RESULTS, PART II

12.3. The number of short simple closed geodesics

The final big result we will discuss is Mirzakhani’s result on the number of simple closed
geodesics of bounded length on a hyperbolic surface.
In her thesis, Mirzakhani proved a sequence of groundbreaking results [Mir07a, Mir07b,
Mir08]. She found a method to compute volumes of moduli spaces of hyperbolic surfaces
(with respect to a volume form called the Weil–Petersson volume form), she found a new
proof of Witten’s conjecture on intersection numbers of tautological classes on Mg,n and
she determined the asymptotics of the number of simple closed geodesics of bounded length
on a hyperbolic surface. The first two results were discussed in Nalini Anantharaman’s
class on random hyperbolic surfaces. We will discuss Mirzakhani’s result on counting
simple closed geodesics.
For some context, let us first describe what happens when we count all short closed
geodesics. For a hyperbolic surface X and L > 0, set
 
γ a closed geodesic on X
cX (L) = # ,
of length ≤ L
where we do not distinguish between geodesics γ and γ ′ if one can be obtained by reparametriz-
ing the other, but do distinguish between two geodesics with opposite orientations.
A classical result, originally due to Huber [Hub56], states that on a closed hyperbolic
surface2
eL
cX (L) ∼ as L → ∞.
L
This is a somewhat surprising result, because it is completely independent of the geometry
and topology of X. Huber used the Selberg trace formula to derive his result.
Now let us discuss simple closed geodesics. Given a complete hyperbolic surface X of finite
area, we will set:  
γ a simple closed geodesic
sX (L) = # .
on X of length ≤ L
In the function sX , we will not distinguish between different orientations on our closed
geodesics. This is just because this fits better with Mirzakhani’s methods. In any event,
the difference between the two functions is a factor 2.
It turns out that the number of simple geodesics grows much slower than all geodesics
[Mir08, Theorem 1.1]:

Theorem 12.3.1. Let X ∈ Mg,n be a hyperbolic surface. Then there exists a constant
bX > 0 so that
sX (L) ∼ bX · L6g−6+2n as L → ∞.

The proof of this theorem is based on the convergence of a sequence of measures on the
space of measured geodesic laminations on Σg,n . In particular, the methods behind which
are wildly different from those behind Huber’s result.
2Recall that the notation “f (x) ∼ g(x) as x → ∞” means that limx→∞ f (x)/g(x) = 1.
 12.3. THE NUMBER OF SHORT SIMPLE CLOSED GEODESICS 107

12.3.1. Measured laminations. We will very briefly sketch what measured lamina-
tions and their key properties are. For a more background, we refer to [CB88, PH92,
AL17].
A geodesic lamination on a hyperbolic surface X is a closed subset that is a disjoint union
of complete simple geodesics that do not intersect each other.
One example of a geodesic lamination is a collection of disjoint simple closed geodesics
(a multicurve), like for instance a pants decomposition. Figure 3 shows another exam-
ple.

Figure 3. A lamination consisting of two simple closed geodesics and one

geodesic arc. The left hand side is consists of three lifts of the components
to H2

A measured geodesic lamination on X is a geodesic lamination on X, together with an

assignment
α 7→ λα
that assigns a Radon measure λα to each arc α that is transverse to the lamination. This
assignment has to satisfy the following conditions:
• Subarcs: If α′ ⊂ α is a subarc, then λα′ is the restriction of λα to α′
• Sliding: If α and α′ are homotopic via a homotopy Ft so that F1 : α → α′ is a
homeomorphism and Ft (α) is transverse to the leaves of the lamination for all t,
then
λα′ = (F1 )∗ λα .

Note that it follows that the support of the measure λα is contained in the intersection of
α with the lamination.
The set of measured laminations ML(X) can be topologized using weak star convergence.
It turns out that this topology does not depend on the geometry of X. So we will often
write MLg,n for the space of measured laminations. There are coordinates, called Dehn-
Thurston coordinates, that are somewhat similar to Fenchel-Nielsen coordinates and that
provide a homeomorphism
MLg,n → R3g−3+n
+ × R3g−3+n .
Unlike Fenchel-Nielssen coordinates, these coordinates do however not give MLg,n the
structure of a smooth manifold.
108 12. LANDMARK RESULTS, PART II

Note that R+ acts on MLg,n , by multiplying the measures with a scalar. MCG(Σg,n ) acts
on MLg,n as well.
Every multicurve induces a measured lamination. The lamination is a multicurve and the
measure assigned to an arc α is the sum Dirac masses on the intersections of α and the
multicurve. We will denote this element of MLg,n by
X
γi ∈ MLg,n ,
i

where {γi }i are the geodesics on X that form the multicurve.

For us, the most important example of a measured lamination is a weighted multicurve,
i.e. we scale the measures on arcs transverse to γi with a number ti ∈ R+ . These measured
laminations will be denoted X
ti γi ∈ MLg,n .
i
It turns out that the set of such measures is actually dense in MLg,n .
The length function on geodesics can naturally be extended to weighted multicurves, by
setting X X
ℓ( ti γi ) = ti ℓ(γi ).
i i
It turns out that ℓ extends to a continuous function
ℓ : MLg,n × T g,n → R+
that satisfies:
• For any simple closed curve γ
ℓ(γ, X) = ℓγ (X).

• For all t ∈ R+ , λ ∈ MLg,n , X ∈ T g,n

ℓ(tλ, X) = tℓ(λ, X).

• For all φ ∈ MCG(Σg,n ), λ ∈ MLg,n

ℓ(φλ, φX) = ℓ(λ, X)

Finally, MLg,n can be equipped with a MCG(Σg,n )-invariant measure µTh called the
Thurston measure. This measure satisfies
µTh (tU ) = t6g−6+2n µTh (U )
and [Mas85]:

Theorem 12.3.2 (Masur). Suppose ν is a measure on MLg,n that is absolutely continuous3

with respect to µTh and MCG(Σg,n )-invariant. Then ν is a multiple of µTh .
3recall
that a measure λ is absolutely continuous with repsect to another measure ν if ν(A) = 0 implies
that λ(A) = 0 for all measurable sets A.
 12.3. THE NUMBER OF SHORT SIMPLE CLOSED GEODESICS 109

12.3.2. The proof. The proof of Theorem 12.3.1 now goes as follows. Once and for all
fix a simple closed curve γ on Σg,n and for all L > 0, define a measure µL,γ on MLg,n
by:
1 X
µL,γ = 6g−6+2n δ 1 γ,
L L
α∈MCG(Σg,n )·γ

where for λ ∈ MLg,n , δλ denotes the Dirac mass on λ.

For X ∈ T g,n , write
BX = { λ ∈ MLg,n : ℓ(λ, X) ≤ 1 }
We have
1 X
µL,γ (BX ) = 1 1 γ has length ≤1
L6g−6+2n L
α∈MCG(Σg,n )·γ
1 X
= 1γ has length ≤L
L6g−6+2n
α∈MCG(Σg,n )·γ
1
=: sX (L, γ).
L6g−6+2n
There are finitely many mapping class group orbits of simple closed curves on Σg,n , which
means that sX (L) splits as a finite sum of measures of the form above.
So, if we can prove that, as L → ∞ the measures µL,γ converge to fixed measures µγ , we
would get
sX (L, γ) ∼ µγ (BX )L6g−6+2n
as L → ∞. Then, since sX (L) is a finite sum of the sX (L, γ)’s, we would be done.
The proof of this goes as follows.
Step 1: One proves that there exists a constant C(X, γ) > 0 so that
sX (L, γ)
µL,γ (BX ) = ≤ C(X, γ).
L6g−6+2n
Since any compact set K lies in L · BX for L large enough, this proves that for any compact
set K,
µL,γ (K)
is bounded as a function of L and as such (by Banach-Alaoglu) there is a subsequential
limit
µLn ,γ → ν.

Step 2: We need to prove that all such limits are multiples of the Thurston measure. It
is not so hard to see that the limit is MCG(Σg,n )-invariant. So, we need to show that
it is absolutely continuous with respect to µTh , so that we can invoke Masur’s theorem
(Theorem 12.3.2). For this, we refer to [Mir08].
Step 3: Now that we know all our subsequential limits are multiples of the Thurston mea-
sure, we need to prove that which multiple it is, does not depend on the subsequence. This
is where Weil–Petersson volumes come in.
110 12. LANDMARK RESULTS, PART II

Let (Lk )k be so that

µLk ,γ → R(Lk )k µTh ,
as n → ∞.
Now let P be a pants decomposition of Σg,n and let (ℓαi , ταi )i be the corresponding Fenchel–
Nielsen coordinates on T (Σg,n ). Wolpert [Wol82] proved that the volume form
3g−3+n
^
dℓαi ∧ dταi
i=1

on Teichmüller space descends to a form of finite total volume on Mg,n . Moreover, similar
forms can be defined on Teichmüller and moduli spaces of surfaces with boundary whose
boundary lengths are fixed.
In her thesis, Mirzakhani proved how to integrate a certain type of functions that she
calls geometric functions. This construction is very similar to the Siegel transform in the
context of the moduli space of lattices. So, it also makes sense to call this the Mirzakhani
transform. Let F : Rk≥0 → R be a function and write Mg,n (L1 , . . . , Ln ) for the moduli
space of hyperbolic surfaces with n boundary components of length L1 , . . . , Ln respectively
(we allow Li = 0, in which case the corresponding boundary component is a cusp). Given
an ordered k-tuple of simple closed curves Γ = (γ1 , . . . , γk ) on Σg,n , we can define the
function F Γ : Mg,n (L1 , . . . , Ln ) → R by
X
F Γ (X) = F (ℓα1 (X), . . . , ℓαk (X)), X ∈ Mg,n (L1 , . . . , Ln ).
(α1 ,...,αk )∈MCG(Σg,n )·(γ1 ,...,γk )

Note that the length functions ℓαi are only defined on Teichmüller space. However, because
we sum over the whole mapping class group orbit, the function above is well-defined on
moduli space.
Our function X 7→ sX (L) is a finite sum such functions. Indeed, we again split our set of
isotopy classes of simple closed curves on Σg,n into finite number of mapping class group
orbits. Given such an orbit, we pick a curve γ in it and set function F = χ[0,L] , the indicator
function of the interval [0, L], such that sX (L, γ) = F γ (X).
Using a very nice covering argument, Mirzakhani proved:

Theorem 12.3.3. With the notation from above and for L = (L1 , . . . , Ln ):
Z Z  
Γ
F (X)d volWP (X) = CΓ F (x)·volWP M(Σg,n −Γ, x, x, L) ·x1 · · · xk dx1 · · · dxk ,
Mg,n (L) Rk≥0

where CΓ > 0 is a constant depending on Γ only, that can be computed effectively and
M(Σg,n − Γ, x, x) denotes the moduli space of hyperbolic metrics on Σg,n − Γ whose bound-
ary components corresponding to Γ have lengths given by (x, x) (they come in pairs) and
those corresponding to Σg,n have lengths given by L.
 12.3. THE NUMBER OF SHORT SIMPLE CLOSED GEODESICS 111

Using this, we can compute:

Z Z
R(Lk )k µTh (BX )d volWP (X) = lim µLk ,γ (BX )d volWP (X)
Mg,n Mg,n k→∞
Z
sX (Lk , γ)
= lim 6g−6+2n d volWP (X)
Mg,n k→∞ Ln
Z
sX (Lk , γ)
= lim 6g−6+2n d volWP (X)
g,n Ln
n→∞ M
Z Lk
1
= lim 6g−6+2n volWP (M(Σ \ γ, x, x)) x d volWP (X).
k→∞ 0 Lk
Now we use that volWP (M(Σ \ γ, x, x)) is a polynomial of the correct degree (another
theorem from Mirzakhani’s thesis) in x and see that the right hand side converges to a
constant that is independent of the sequence Lk . The integral
Z
µTh (BX )d volWP (X)
Mg,n

also doesn’t depend on the sequence, so neither can R(Lk )k , which finishes the proof.
 Bibliography

[Ago13] Ian Agol. The virtual Haken conjecture. Doc. Math., 18:1045–1087, 2013. With an appendix by
Agol, Daniel Groves, and Jason Manning.
[AL17] Javier Aramayona and Christopher J. Leininger. Hyperbolic structures on surfaces and geodesic
currents. In Algorithmic and geometric topics around free groups and automorphisms, Adv.
Courses Math. CRM Barcelona, pages 111–149. Birkhäuser/Springer, Cham, 2017.
[Baa21] Sebastian Baader. Geometry and topology of surfaces. Zurich Lectures in Advanced Mathematics.
EMS Press, Berlin, [2021] ©2021.
[Bea84] A. F. Beardon. A primer on Riemann surfaces, volume 78 of London Mathematical Society
Lecture Note Series. Cambridge University Press, Cambridge, 1984.
[Bus10] Peter Buser. Geometry and spectra of compact Riemann surfaces. Modern Birkhäuser Classics.
Birkhäuser Boston, Inc., Boston, MA, 2010. Reprint of the 1992 edition.
[CB88] Andrew J. Casson and Steven A. Bleiler. Automorphisms of surfaces after Nielsen and Thurston,
volume 9 of London Mathematical Society Student Texts. Cambridge University Press, Cam-
bridge, 1988.
[CE08] Jeff Cheeger and David G. Ebin. Comparison theorems in Riemannian geometry. AMS Chelsea
Publishing, Providence, RI, 2008. Revised reprint of the 1975 original.
[Fal23] Elisha Falbel. Introduction to Riemann surfaces. Lecture notes, available at: https://
webusers.imj-prg.fr/~elisha.falbel/Surfaces/RS.html, 2023.
[FK92] H. M. Farkas and I. Kra. Riemann surfaces, volume 71 of Graduate Texts in Mathematics.
Springer-Verlag, New York, second edition, 1992.
[FLP79] Travaux de Thurston sur les surfaces, volume 66-67 of Astérisque. Société Mathématique de
France, Paris, 1979. Séminaire Orsay, With an English summary.
[FLP12] Albert Fathi, Fran¸cois Laudenbach, and Valentin Poénaru. Thurston’s work on surfaces, vol-
ume 48 of Mathematical Notes. Princeton University Press, Princeton, NJ, 2012. Translated from
the 1979 French original by Djun M. Kim and Dan Margalit.
[FM12] Benson Farb and Dan Margalit. A primer on mapping class groups, volume 49 of Princeton
Mathematical Series. Princeton University Press, Princeton, NJ, 2012.
[GGD12] Ernesto Girondo and Gabino González-Diez. Introduction to compact Riemann surfaces and
dessins d’enfants, volume 79 of London Mathematical Society Student Texts. Cambridge Univer-
sity Press, Cambridge, 2012.
[GH94] Phillip Griffiths and Joseph Harris. Principles of algebraic geometry. Wiley Classics Library.
John Wiley & Sons, Inc., New York, 1994. Reprint of the 1978 original.
[GL00] Frederick P. Gardiner and Nikola Lakic. Quasiconformal Teichmüller theory, volume 76 of Math-
ematical Surveys and Monographs. American Mathematical Society, Providence, RI, 2000.
[Hat02] Allen Hatcher. Algebraic topology. Cambridge University Press, Cambridge, 2002.
[HM79] John Hubbard and Howard Masur. Quadratic differentials and foliations. Acta Math., 142(3-
4):221–274, 1979.
[Hub56] Heinz Huber. Über eine neue Klasse automorpher Funktionen und ein Gitterpunktproblem in
der hyperbolischen Ebene. I. Comment. Math. Helv., 30:20–62 (1955), 1956.
[Hub06] John Hamal Hubbard. Teichmüller theory and applications to geometry, topology, and dynamics.
Vol. 1. Matrix Editions, Ithaca, NY, 2006. Teichmüller theory, With contributions by Adrien
Douady, William Dunbar, Roland Roeder, Sylvain Bonnot, David Brown, Allen Hatcher, Chris
Hruska and Sudeb Mitra, With forewords by William Thurston and Clifford Earle.

113
114 BIBLIOGRAPHY

[Hub16] John Hamal Hubbard. Teichmüller theory and applications to geometry, topology, and dynamics.
Vol. 2. Matrix Editions, Ithaca, NY, 2016. Surface homeomorphisms and rational functions.
[IT92] Y. Imayoshi and M. Taniguchi. An introduction to Teichmüller spaces. Springer-Verlag, Tokyo,
1992. Translated and revised from the Japanese by the authors.
[Ker83] Steven P. Kerckhoff. The Nielsen realization problem. Ann. of Math. (2), 117(2):235–265, 1983.
[Mas85] Howard Masur. Ergodic actions of the mapping class group. Proc. Amer. Math. Soc., 94(3):455–
459, 1985.
[Mil65] John W. Milnor. Topology from the differentiable viewpoint. University Press of Virginia, Char-
lottesville, VA, 1965. Based on notes by David W. Weaver.
[Mir07a] Maryam Mirzakhani. Simple geodesics and Weil-Petersson volumes of moduli spaces of bordered
Riemann surfaces. Invent. Math., 167(1):179–222, 2007.
[Mir07b] Maryam Mirzakhani. Weil-Petersson volumes and intersection theory on the moduli space of
curves. J. Amer. Math. Soc., 20(1):1–23, 2007.
[Mir08] Maryam Mirzakhani. Growth of the number of simple closed geodesics on hyperbolic surfaces.
Ann. of Math. (2), 168(1):97–125, 2008.
[Ota96] Jean-Pierre Otal. Le théorème d’hyperbolisation pour les variétés fibrées de dimension 3.
Astérisque, (235):x+159, 1996.
[PH92] R. C. Penner and J. L. Harer. Combinatorics of train tracks, volume 125 of Annals of Mathe-
matics Studies. Princeton University Press, Princeton, NJ, 1992.
[Sma59] Stephen Smale. Diffeomorphisms of the 2-sphere. Proc. Amer. Math. Soc., 10:621–626, 1959.
[SS03] Elias M. Stein and Rami Shakarchi. Complex analysis, volume 2 of Princeton Lectures in Anal-
ysis. Princeton University Press, Princeton, NJ, 2003.
[Str84] Kurt Strebel. Quadratic differentials, volume 5 of Ergebnisse der Mathematik und ihrer Gren-
zgebiete (3) [Results in Mathematics and Related Areas (3)]. Springer-Verlag, Berlin, 1984.
[Wol82] Scott Wolpert. The Fenchel-Nielsen deformation. Ann. of Math. (2), 115(3):501–528, 1982.
[Wri15] Alex Wright. Translation surfaces and their orbit closures: an introduction for a broad audience.
EMS Surv. Math. Sci., 2(1):63–108, 2015.
[Zor06] Anton Zorich. Flat surfaces. In Frontiers in number theory, physics, and geometry. I, pages
437–583. Springer, Berlin, 2006.