Counting closed geodesics in Moduli space

arXiv:0811.2362v3 [math.DS] 19 Mar 2011

Alex Eskin∗and Maryam Mirzakhani†

March 22, 2011

1 Introduction
Let Mg denote the moduli space of closed Riemann surfaces of genus g. We may write Mg = Tg /Γg ,
where Tg is the Teichmüller space of genus g surfaces, and Γg is the mapping class group. In this
paper, we investigate properties of Teichmüller geodesics on Mg . Let N (R) denote the number of
closed Teichmüller geodesics in Mg of length at most R. Then N (R) is also the number of conjugacy
classes of pseudo-Anosov elements of the mapping class group of translation length at most R. Our
main result is the following:
Theorem 1.1 As R → ∞, we have
ehR
N (R) ∼ ,
hR
where h = 6g − 6.
In the above theorem and below, the notation A ∼ B means that the ratio A/B tends to 1. Even
though we assume that the surface has no punctures here, most of the results also hold on Mg,n ,
provided h is replaced by 6g − 6 + 2n.
In the proof of Theorem 1.1, the key is to estimate the number of closed geodesics which stay
outside of compact sets. Let Nj (δ, R) denote the number of closed geodesics γ of length at most R
in Mg such that for each point X ∈ γ, X has at least j simple closed curves of hyperbolic length
less than δ.

Theorem 1.2 Given ǫ > 0 there exist δ > 0 and C = C(ǫ) such that for all j, 1 ≤ j ≤ 3g − 3, and
all R > 0,
Nj (δ, R) ≤ Ce(h−j+ǫ)R .

Let K be a compact subset of Mg . We let N K (R) denote the number of geodesics in Mg of length
less than R which never intersect K. Letting j = 1 in Theorem 1.2, we obtain the following:
Corollary 1.3 For every ǫ > 0 there exist a compact set K ⊂ Mg and C = C(ǫ) > 0 such that for
all R > 0,
N K (R) ≤ Ce(h−1+ǫ)R .
∗ partially supported by NSF grants DMS 0244542 and DMS 0604251 and the Clay foundation
† partially supported by the Clay foundation

1
Corollary 1.3 is complimentary to the following result:
Theorem 1.4 (Rafi, Hammenstadt) For any compact K ⊂ Mg , and sufficiently large R,

N K (R) ≥ e(h−1)R .

1.1 Previous Results

The first results on this problem are due to Veech [Ve]. He proved that there exists a constant c2
such that
log N (R) log N (R)
h ≤ lim inf ≤ lim sup ≤ c2
R→∞ R R→∞ R
and conjectured that c2 = h. In a remark in a paper by Ursula Hamenstadt [H1] (see also [H3]), in
which the main focus is different, she proves that c2 ≤ (6g − 6 + 2n)(6g − 5 + 2n).
Sasha Bufetov [Bu] proved the formula

log Ñ (R)
lim = h, (1)
R→∞ R

where Ñ (R) is the number of periodic orbits of the Rauzy-Veech induction such that the log of the
norm of the renormalization matrix is at most R. This is a closely related problem; essentially Ñ (R)
counts closed geodesics on a certain finite cover of Mg . However the equation (1) does not easily
imply
log N (R)
lim = h. (2)
R→∞ R
Very recently, Kasra Rafi [Ra1] proved Corollary 1.3 (which implies (2)) for the case of the five-
punctured sphere.
We note that (2) is an immediate consequence of Theorem 1.1, which is a bit more precise. In
order to prove Theorem 1.1 one needs Corollary 1.3 and certain recurrence results for geodesics,
which are based on [Ath].
Remarks.
• The problem of understanding the asymptotics of the number NM (R) of primitive closed
geodesics of length ≤ R on a given manifold M has been investigated intensively. An asymp-
totic formula for NM (R) on a compact hyperbolic surfaces was first proved by Huber. See §9
in [Bu] and references within for related work of Hejhal, Randol and Sarnak.
More generally, Margulis proved that on a compact n-manifold M of negative curvature

ehR
NM (R) ∼ ,
hR
where h is the topological entropy of the geodesic flow. In this case the techniques from
uniformly hyperbolic dynamics can be applied to study the geodesic flow on the unit tangent
bundle of M . See [Mar] for the proof of Margulis’ theorem, and related results on Anosov and
hyperbolic flows.

2
 • The main difficulty for proving Theorem 1.1 is the fact that the Teichmüller flow is not hyper-
bolic. In order to overcome this difficulty, first in §4 we use Minsky’s product region theorem
[Mi] to prove that the geodesic flow is biased toward the thick part of the moduli space. This
result implies Theorem 1.2. Then we use the basic properties of the Hodge norm [ABEM] to
prove a closing lemma for the Teichmüller geodesic flow in §6. But the Hodge norm behaves
badly near smaller strata, i.e. near points with degenerating zeros of the quadratic differential.
However, in §5 we show that the number of closed geodesics γ of length at most R such that
γ spends at least θ-fraction of the time outside of a compact subset of the principal stratum
is exponentially smaller than N (R) (see Theorem 5.2). Finally, we obtain the counting re-
sult using the fact that in any compact subset of the principal stratum, the geodesic flow is
uniformly hyperbolic ([Ve], [Fo], and [ABEM]).
• By results in [H2] the normalized geodesic flow invariant measure supported on the set of
closed geodesics of length ≤ R in Q1 Mg become equidistributetd with respect to the Lebesgue
measure µ (see §2.2) as R → ∞.
• In a forthcoming joint work with Kasra Rafi, we generalize the results in this paper to the case
of other strata of moduli spaces of Abelian and quadratic differentials.
Notation. In this paper, A ≈ B means that A/C < B < AC for some universal constant C which
only depends genus g. Also, A = O(B) means that A < BC, for some universal constant C, which
again could depend on g.

Acknowledgements. We would like to thank Kasra Rafi who suggested a major simplification of
the proof of Theorem 1.2. We would also like to thank Howard Masur for many useful discussions
(including the proof of Lemma 3.1) and also William Cavendish whose lecture notes were used for
part of this paper. We would like to thank the referee for helpful comments.

2 Background and notation

In this section, we recall definitions and known results about the Teichmüller geometry of Mg . For
more details see [Hu].

2.1 Teichmüller space

A point in the Teichmüller space Tg is a complex curve X of genus g equipped with a diffeomorphism
f : Sg → X. The map f provides a marking on X by Sg . Two marked surfaces f1 : Sg → X and
f2 : Sg → Y define the same point in Tg if and only if f1 ◦ f2−1 : Y → X is isotopic to a holomorphic
map. By the uniformization theorem, each point X in Tg has a metric of constant curvature −1.
The space Tg is a complex manifold of dimension 3g − 3, diffeomorphic to a cell. Let Γg denote the
mapping class group of Sg , or in other words the group of isotopy classes of orientation preserving
self homeomorphisms of Sg . The mapping class group Γg acts on Tg by changing the marking. The
quotient space
Mg = Tg /Γg

3
is the moduli space of Riemann surfaces homeomorphic to Sg . The space Mg is an orbifold. However,
it is finitely covered by a manifold, and

π1orb (Mg ) = Γg .

Let
p : Tg → Mg
denote the natural map from Tg to Mg = Tg /Γg . The Teichmüller metric on marked surfaces is
defined by
1
dT ((f1 : Sg → X1 ), (f2 : Sg → X2 )) = inf(log K(h)),
2
where h : X1 → X2 ranges over all quasiconformal maps isotopic to f1 of2−1 . Here K(h) ≥ 1 is the
dilatation of h.
Dilatation of pseudo-Anosov elements. According to the Nielsen-Thurston classification, every
irreducible mapping class element g ∈ Γg of infinite order has a representative which is a pseudo-
Anosov homeomorphism [Th]. By a theorem of Bers, every closed geodesic in Mg is the unique loop
of minimal length in its homotopy class. Given a pseudo-Anosov g ∈ Γg the dilatation of g is defined
by K(g). Then log(K(g)) is the translation length of g as an isometry of Tg [Be]. In other words,

L(Sg ) = {log(K(g)) |g ∈ Γg pseudo-Anosov }

is the length spectrum of Mg equipped with the Teichmüller metric.

By [AY] and [Iv] Lg is discrete subset of R. Hence the number N (R) of conjugacy classes of pseudo-
Anosov elements of the group Γg with dilatation factor K(g) ≤ R is finite. We remark that for any
pseudo-Anosov g ∈ Γg the number K(g) is an algebraic number. Moreover log(K(g)) is equal to the
minimal topological entropy of any element in the same homotopy class [FLP].

2.2 Moduli space of quadratic differentials

The cotangent space of Tg at a point X can be identified with the vector space Q(X) of holomorphic
quadratic differentials on X. Recall that given X ∈ Tg , a quadratic differential q ∈ Q(X) is a tensor
locally given by φ(z)dz 2 where φ is holomolphic. Then the space QTg = {(q, X) | X ∈ Tg , q ∈ Q(X)}
is the cotangent space of Tg . In this setting, the Teichmüller metric corresponds to the norm
Z
k q kT = |φ(z)| |dz|2
X

on QTg . Let QMg ∼ = QTg /Γg . Finally, let Q1 Tg denote the Teichmüller space of unit area quadratic
differentials on surfaces of genus g, and Q1 Mg ∼ = Q1 Tg /Γg . For simplicity, we let π denote the
natural projection maps
π : Q1 M g → M g ,
and
π : Q1 Tg → Tg .
Although the value of q ∈ Q1 (X) at a point x ∈ X depends on the local coordinates, the zero
set of q is well defined. As a result, there is a natural stratification of the space QMg by the

4
multiplicities of zeros of q. Define QMg (a1 , . . . , ak ) ⊂ QMg to be the subset consisting of pairs
(X, q) of holomorphic quadratic differentials on X with k zeros with multiplicities (a1 , . . . , ak ). Then
G
QMg = QMg (a1 , . . . , ak ).
(a1 ,...,ak )

It is known that each QMg (a1 , . . . , ak ) is an orbifold of dimension 4g − 4 + 2k. In particular

dim(QMg (1, . . . , 1)) = dim(QMg ).
We recall that when g > 1, the Teichmüller metric is not even Riemannian. However, geodesics
in this metric are well understood. A quadratic differential q ∈ QTg with zeros at x1 , . . . xk is
determined by an atlas of charts {φi } mapping open subsets of Sg − {x1 , . . . , xk } to R2 such that
the change of coordinates are of the form v → ±v + c. Therefore the group SL2 (R) acts naturally
on QMg by acting on the corresponding atlas; given A ∈ SL2 (R), A · q ∈ QM  g is determined by
t/2
e 0
the new atlas {Aφi }. The action of the diagonal subgroup gt = is the Teichmüller
0 e−t/2
geodesic flow for the Teichmüller metric. Moreover we have [Ve],[Mas]:
Theorem 2.1 (Veech-Masur) The space Q1 Mg carries a unique probability measure µ = µg in
the Lebesgue measure class such that :
• the action of SL2 (R) is volume preserving and ergodic;
• Teichmüller geodesic flow is mixing.
By this theorem, Q1 Tg carries a natural normalized smooth measure µ = µg , preserved by the action
of Γg . We set
m = π∗ µ. (3)
Remark. In fact, the Teichmüller flow on Q1 Mg is exponentially mixing with respect to µ [AR].
See also [AGY]. However, we will only use Theorem 2.1 in this paper.

2.3 Period coordinates on the strata

A saddle connection on q ∈ Q1 Tg is a geodesic segment which joins a pair of singular points without
passing through one in its interior. In general, a geodesic segment e joining two zeros of a quadratic
differential q = φdz 2 determines a complex number holq (e) (after choosing a branch of φ1/2 and an
orientation of e) by
holq (e) = Re(holq (e)) + Im(holq (e)),
where Z
Re(holq (e)) = Re(φ1/2 ),
e

and Z
Im(holq (e)) = Im(φ1/2 ).
e

We recall that the period coordinates give QTg (a1 , . . . , ak ) the structure of a piecewise linear mani-
fold. For notational simplicity we discuss the case of QTg (1, . . . , 1). Given q0 ∈ QTg (1, . . . , 1) there

5
is a triangulation E of the underlying surface by saddle connections, h = 6g − 6 directed edges
δ1 , . . . , δh of E, and an open neighborhood Uq0 ⊂ QTg (1, . . . , 1) of q0 such that the map

ψE,q0 : QTg (1, . . . , 1) → C6g−6

by
ψE,q0 (q) = (holq (δi ))hi=1
′
is a local homeomorphism. Also for any other geodesic triangulation E the map the map ψE ′ ,q0 ◦
−1
ψE,q 0
is linear. For a discussion of these coordinates see [MasSm]. The measure µg in Theorem 2.1
(up to a constant) is given by the piecewise linear structure of QMg . This measure, up to normaliza-
tion, also coincides with the measure defined by the Teichmüller norm on the unit cotangent bundle
of Tg . This measure is supported on Q1 Mg (1, 1, . . . , 1)); that is, µg (Q1 Mg −Q1 Mg (1, 1, . . . , 1)) = 0.

2.4 Extremal and hyperbolic lengths of simple closed curves

Given a homotopy class of a simple closed curve α on a topological surface Sg and X ∈ Tg , let ℓα (X)
be the length of the unique geodesic in the homotopy class of α with respect to the hyperbolic metric
on X. The extremal length of a simple closed curve α on X is defined by

ℓα (ρ)2
Extα (X) = sup , (4)
ρ Area(X, ρ)

where the supremum is taken over all metrics ρ conformally equivalent to X, and ℓα (ρ) denotes the
length of α in the metric ρ.
Given simple closed curves α and β on Sg , the intersection number i(α, β) is the minimum number
of points in which representatives of α and β must intersect. In general, we have:
p q
i(α, β) ≤ Extα (X) · Extβ (X).

The following result [Ker] relates the ratios of extremal lengths to the Teichmüller distance:
Theorem 2.2 (Kerckhoff) Given X, Y ∈ Tg , the Teichmüller distance between X and Y is given
by p !
Extβ (X)
dT (X, Y ) = sup log p ,
β Extβ (Y )
where β ranges over all simple closed curves on Sg .
The relationship between the extremal length and hyperbolic length is complicated; in general, by
the definition of extremal length
ℓα (X)2
≤ Extα (X).
4π(g − 1)
Also for any X ∈ Tg the extremal length can be extended continuously to the space of measured
laminations [Ker] such that
Extr·λ (X) = r2 Extλ (X).

6
As a result given X there exists a constant aX such that
1 p
ℓα (X) ≤ Extα (X) ≤ aX ℓα (X).
aX
However, by [M],
1 Extα (X) 1
≤ ≤ eℓα (X)/2 . (5)
π ℓα (X) 2
Hence as ℓα (X) → 0,
ℓα (X)
≈ 1.
Extα (X)

Bounded pants decompositions. Recall that by a theorem of Bers, there exists Lg depending
only on g such that for any surface Y ∈ Tg there exists a pants decomposition PY = {α1 , . . . , α3g−3 }
of Y satisfying ℓαi (Y ) ≤ Lg . By (5) the extremal length of each αi on Y is bounded from above by
Cg , where Cg = eLg Lg is independent of Y . We call such a pants decomposition a bounded pants
decomposition for Y.
Fix a small enough number ǫ0 > 0 such that:
1). Any two simple closed curves of extremal length ≤ ǫ20 are disjoint;
2). any simple closed curve intersecting a simple closed curve of extremal length ≤ ǫ20 on a surface
of genus g has extremal length ≥ 2Cg .
We say α is short on X if Extα (X) ≤ ǫ20 . Let CX be the set of simple closed curves β on the surface
X with Extβ (Y ) ≤ ǫ20 . Note that by the definition any bounded pants decomposition of X contains
all the elements of CX .
Now, define G : Tg → R+ by
Y 1
G(Y ) = 1 + p . (6)
β∈CY
Extβ (Y )

Note that for any bounded pants decomposition PX = {α1 , . . . , α3g−3 }

Y 1
G(X) ≈ p .
α∈PX
Extα (X)

Also, if dT (X, Y ) = O(1) then G(X) ≈ G(Y ). By the definition, G induces a proper function on Mg .

2.5 Estimating extremal length and Minsky’s product theorem

Let α1 , . . . αj be a collection of disjoint simple closed curves on Sg , and ǫ ≤ ǫ20 . Let

Pǫ (α1 , . . . , αj ) = {X ∈ Tg |ExtX (αi ) ≤ ǫ}.

Let H2 denote the upper half plane model of the hyperbolic plane. Then using the Fenchel-Nielsen
coordinates on Tg , we can define φ0 : Pǫ (α1 , . . . , αj ) → (H2 )j by
1 1
φ0 (X) = (θ1 (X), , . . . , θj (X), ),
ℓα1 (X) ℓαj (X)

7
where θi is the twist coordinate around αi . In fact, following Minsky, we get a map
φ : Pǫ (α1 , . . . , αj ) → (H2 )j × T ′ ,
where T ′ is the quotient Teichmüller space obtained by collapsing all the αi . The product region
theorem [Mi] states that for sufficiently small ǫ the Teichmüller metric on Pǫ (α1 , . . . , αj ) is within
an additive constant of the supremum metric on (H2 )j × T ′ . More precisely, let d′ (·, ·) denote the
supremum metric on (H2 )j × T ′ . Then :
Theorem 2.3 If ǫ > 0 is small enough, then there exists a constant b > 0 depending only on the
genus such that for all X, Y ∈ Pǫ (α1 , . . . , αj ),
|dT (X, Y ) − d′ (φ(X), φ(Y ))| < b.
Note that by Theorem 2.2 this statement can be rewritten in terms of the ratios of extremal
lengths of simple closed curves on X and Y .
Let Cg denote the set of all multicurves on Sg , and {α1 , . . . , α3g−3 } be a pants decomposition of
Sg . Consider the Dehn-Thurston parameterization [HP]
DT : Cg → (Z+ × Z)3g−3
defined by
3g−3
DT (β) = (i(β, αi ), tw(β, αi ))i=1 ,
where i(·, ·) denotes the geometric intersection number and tw(β, αi ) is the twisting parameter of β
around αi . See [HP] for more details.
The proof of Theorem 2.3 relies on the following estimates for the extremal lengths of arbitrary
simple closed curves on a surface, see [Mi] (Theorem 5.1, and equation (4.3)):
Theorem 2.4 (Minsky) Suppose Y ∈ Tg , and let P = PY = {α1 , . . . , α3g−3 } be a bounded pants
decomposition on Y . Then given a simple closed curve β on Sg , Extβ (Y ) is bounded from above and
below by  
i(β, αj )2
max + tw2 (β, αj )Extαj (Y ) (7)
1≤j≤3g−3 Extαj (Y )

up to a multiplicative constant depending only on g.

This result gives an upper bound for the Dehn-Thurston coordinates of a simple closed curve in
terms of its extremal length.
We remark that the definition of the twist used in equation (4.3) in [Mi] is different from the
definition we are using here. We follow the definition used in [HP]. In terms of our notation,
tw(hrα (β), α) = tw(β, α) + r · i(β, αi ), (8)
where hα ∈ Γg is the right Dehn twist around α.
Let PX = {α1 , . . . , αn } be a bounded pants decomposition of X ∈ Tg (see §2.4). Then:
• For any simple closed curve β on Sg ,
p q
Extβ (X) p
tw(β, αi ) ≤ c1 · p , i(β, αi ) ≤ Extβ (X) · Extαi (X), (9)
Extαi (X)
where c1 > 0 only depend on g.

8
 • Let α ∈ PX and r0 > 0. We claim that if

dT (hkα (X), X) ≤ r0 ,

then
e r0
|k| ≤ c ,
Extα (X)
where c > 0 only depends on g and hα ∈ Γg is the right Dehn twist around α. Note that by
[Mi], if α ∈ PX then there exists a simple closed curve β such that
1
Extβ (X) ≈ , (10)
Extα (X)
q p
and i(α, β) = 1. We can use Theorem 2.4 to estimate Exth−kα (β)
(X) = Extβ (hkα (X)). As a
result q
1 p
p + |k| Extα (X) ≤ c′ Exth−k
α (β)
(X),
Extα (X)
where c′ only depends on g. By Theorem 2.2, we have:
q q
r0
Exth−k
α (β)
(X) ≤ e Extβ (X).

Now from (10) we get

q  
1 p e r0
p + |k| Extα (X) = O(er0 Extβ (X)) =⇒ |k| = O . (11)
Extα (X) Extα (X)

In other words, the number of twists around αi ∈ PX which one can take and still stay in
BT (X, r) is O(er /ai ), where ai = Extαi (X). A generalization of this argument is used in the
proof of Lemma 7.3 [ABEM].
Remark. As a result of Theorem 7.2, for any X ∈ Tg the number of g ∈ Γg such that
Q3g−3
dT (g · X, X) ≤ r is O( i=1 r er /ai ). When r > 0 is fixed, one can obtain this bound by
showing that up to uniformly bounded index, the set of {g ∈ Γg | dT (g · X, X) ≤ r} only
consists of twists about αi ∈ PX .

3 Net points in Teichmüller space

Recall that a set N is a (c1 , c2 ) separated net on a metric space X if N ⊂ X , every point of X is
within c2 of a net point, and the minimal distance between net points is at least c1 . In §4 we will
use nets in order to discretize Tg .

3.1 Volumes of balls of fixed radius

Let BT (X, L) ⊂ Tg denote the ball of radius L with respect to the Teichmüller metric, and let m
be the smooth measure defined by (3) on Tg . Then we have:

9
Lemma 3.1 There exists L0 > 0 (depending only on g) and for every L > L0 there exist constants
0 < c1 < c2 such that for all X ∈ Tg ,

c1 ≤ m(BT (X, L)) ≤ c2

The constants c1 and c2 depend on L and g (but not on X).

Proof. Suppose X ∈ Tg . Let α1 , . . . , αk be the simple closed curves on X with extremal length
less than Cg2 . Let ai denote the extremal length of αi . Note that by (5) the hyperbolic length
ℓαi (X) ≈ ai . By Theorem 2.2 for all Y ∈ BT (X, L),

ExtY (αi ) ≤ L21 ExtX (αi ) ≤ C L21 ai ,

where L1 = eL , and C depends only on the genus. Then, in view of the definition of extremal length
(equation (4)), for any area 1 holomorphic quadratic differential q on Y ∈ BT (X, L),

ℓq (αi )2 ≤ ExtY (αi ) ≤ CL21 ai , (12)

where ℓq (·) denotes length in the flat metric defined by q, and C depends only on the genus. Thus
any flat √ metric in the conformal class of a surface in BT (X, L) has closed curves of flat length at
most L1 Cai .
Let F be a fundamental domain for the action of Γg on Tg . For k ≤ 3g − 3 and 0 ≤ c1 ≤
c2 ≤ . . . ck ≤ 1, let Q(c1 , . . . , ck ) denote the subset of X ∈ Tg for which there exist disjoint curves
β1 , . . . , βk with ExtX (βi ) ≤ ci . Then, for any area 1 holomorphic quadratic differential q ∈ π −1 (X),
√
the length of αi in the flat metric induced by q is at most ci . Then, by the definition of the measure
m(·),
k
Y
m(Q(c1 , . . . , ck ) ∩ F ) ≤ C ci . (13)
i=1

In view of (12),
BT (X, L) ⊂ Q(CL21 a1 , . . . , CL21 ak ).
Then, in view of (13), for any g ∈ Γg ,
k
Y
m(BT (X, L) ∩ gF ) ≤ C CL21 ai . (14)
i=1

Let IX,L denote the set of elements g ∈ Γg such that BT (X, L) ∩ gF is non-empty. We will now
estimate the size of IX,L .
Note that as Mg is finitely covered by a manifold, (11) implies that

|IX,L | ≤ CL G(X)2 ,

where
Yk
1
G(X) ≈ √ ,
i=1
ai

and CL only depends on L and g. Also, See Theorem 7.2 (for the case of B = ∅).

10
 On the other hand, by Theorem 2.4, the number of twists around αi which one can take and still
stay in BT (X, L) is ≈ L21 /ai . Thus,
Yk
′ 1
|IX,L | ≈ CL . (15)
a
i=1 i
See §2.5. The upper bound of Lemma 3.1 follows from (14).
We now briefly outline the proof of the lower bound. Let R ⊂ Q1 Mg /Γg be the set of flat
√
structures such that each S ∈ R has flat cylinders Ci with width (i.e. core curve) between ai
√ √ √
and ai /2 and height between 1/(g ai ) and 1/(2g ai ), and the rest of the arcs in a triangulation
of S ∈ R have length comparable to 1. The period coordinates give the structure of a piecewise
linear integral manifold on Q1 Mg . By work of Masur [Mas], (up to normalization) the measure µg
is the canonical measure defined by this piecewise linear integral structure (See §2.3). Hence, by the
definition of the measures µg and m we get
k
Y
m(π(R)) ≥ µ(R) > c ai , (16)
i=1

where c depends only on the genus. Note that by [Ra2] for any Y ∈ π(R), the only short simple
closed curves on Y (in the hyperbolic or extremal metric) are the core curves of the cylinders Ci ,
and the extremal length of the core curve of Ci is within a constant multiple of ai . Thus, there
exists a constant L′ depending only on the genus such that
π(R) ⊂ BT (X, L′ ).
Note that the above equation takes place in Tg /Γg . We may think of it as taking place in Tg if we
identify π(R) with a subset of the fundamental domain F . Then, for any g ∈ IX,L′ ,
gπ(R) ⊂ BT (X, 2L′ ).
Thus, in view of (15) and (16),
m(BT (X, 2L′ )) ≥ |IX,L′ |m(π(R)) ≥ c,
where c depends only on the genus. 

3.2 Choosing a net in Teichmüller space

Let c > 2L0 (the constant in Lemma 3.1), N e be a (c, 2c)−net in Mg , and N ⊂ p−1 (N
e ) be a net in
e
Tg such that p(N ) = N . As before, p : Tg → Mg is the natural projection to the moduli space. In
other words,
1. given X ∈ Tg there exists Y ∈ N such that dT (X, Y ) ≤ 2c , and
2. for any Y1 6= Y2 ∈ N , we have dT (Y1 , Y2 ) ≥ c.
In view of Lemma 3.1, for any (c, 2c) separated net N in any Teichmüller space (including H2 ), for
τ ≫ 1 and any X,
c1 |BT (X, τ ) ∩ N | ≤ m(BT (X, τ )) ≤ c2 |BT (X, τ ) ∩ N |, (17)
where c1 , c2 depend only on c and the genus.

11
Lemma 3.2 Let c > 2L0 . Then for any (c, 2c) net N e in Mg , there exists a constant C2 > 0 such
that for any X ∈ Tg
e | ≤ C2 τ 3g−3 .
|p(BT (X, τ )) ∩ N (18)
Proof. Fix
Cg2

τ0 = log
ǫ20
(see §2.4).

Step 1. First we assume that τ ≥ τ0 . Let AX = {α1 , α2 , . . . , αk } be the set of simple closed curves
of extremal length ≤ ǫ20 e−2τ on X ∈ Tg . Let P1 , . . . , Pm be the set of all combinatorially distinct
pants decompositions of Sg containing α1 , . . . , αk .
Given a pants decomposition Pj = {α1 , . . . , αk , αjk+1 , . . . , αj3g−3 } and i = (i1 , . . . , i3g−3 ) ∈ Z3g−3
choose Zj,i ∈ Tg such that:
• for each 1 ≤ l ≤ k
Extαl (Zj,i )
eil −1 ≤ ≤ eil ,
Extαl (X)

• for k + 1 ≤ l ≤ 3g − 3
Cg eil −1 ≤ Extαj (Zj,i ) ≤ Cg eil .
l

Let

I = {(i1 , . . . , i3g−3 ) ∈ Z3g−3 | for 1 ≤ l ≤ k, −2τ ≤ il ≤ 2τ , and for k+1 ≤ l ≤ 3g−3 , −5τ +1 ≤ il ≤ 0 }.

Define the set Z (with |Z| = O(τ 3g−3 )) by

Z = {Zj,(i1 ,i2 ,...,i3g−3 ) | i ∈ I} ⊂ Tg .

Claim. Given Y ∈ BT (X, τ ), there exists Zj,i ∈ Z such that p(Zj,i ) ∈ p(BT (Y, C)), where C is a
constant which only depends on g. This is a simple corollary of Theorem 2.4 and Theorem 2.2. Note
that given Y ∈ BT (X, τ ) and α ∈ AX , α is a short simple closed curve on Y and

Extα (Y )
e−2τ ≤ ≤ e2τ .
Extα (X)

Moreover, for any other simple close curve β in a bounded pants decomposition of Y since dT (X, Y ) <
τ, if β 6∈ AX then
ǫ20 e−4τ ≤ Extβ (Y ) ≤ Cg .
Now since any bounded pants decomposition PY on Y has the combinatorial type of Pj for some
1 ≤ j ≤ m, the claim follows easily from Theorem 2.4. 

As a result, [
e⊂
p(BT (X, τ )) ∩ N e ).
(p(BT (Zi , C)) ∩ N (19)
Zi ∈Z

12
Step 2. Since N ⊂ Tg is a (c, 2c) net with c > 2L0 , by Lemma 3.1 for any Z ∈ Tg
Vol(BT (Z, C + c))
|BT (Z, C)) ∩ N | ≤ = O(1),
Vol(BT (Z, c))
and hence
e | = O(1).
|p(BT (Z, C)) ∩ N (20)
3g−3
Now since |Z| = O(τ ) the result follows from (19) and (20).
Note that τ0 only depends on g, and therefore if τ ≤ τ0 the result follows from step 2. 

4 Geodesics in the thin part of moduli space

In this section we prove Theorem 1.2. The main idea, which is due to Margulis, is to prove a system
of inequalities, which shows that the flow (or more precisely an associated random walk) is biased
toward the thick part of Teichmüller space. Variations on this theme have been used in [EMM],
[EM] and [Ath].

4.1 A system of inequalities

Suppose 0 < s < 1 (in fact we will be using s = 1/2 only). Let τ ≫ 1 be a parameter to be chosen
later. (In particular we will assume e−(1−s)τ < 1/2.) Let Aτ be the operator of averaging over a
ball of radius τ in Tg with respect to the Teichmüller metric. So if f is a real-valued function on
Teichmüller space, then
Z
1
(Aτ f )(X) = f (Y ) dm(Y ).
m(BT (X, τ )) BT (X,τ )

Remark. In [EM], [Ath] and [EMM] the average is over spheres. In this context, we use balls, since
Minsky’s product region theorem gives us much more precise information about balls than about
spheres.
Let m be the maximal number of disjoint simple closed curves on a closed surface of genus g.
Choose C > e2mτ , and pick constants ǫ1 < ǫ2 < · · · < ǫm < 1/C3 such that for all 1 ≤ i ≤ m − 1,
ǫi+1
ǫi < . (21)
C2 (2mC2 )2/s
Note that C and ǫ1 , . . . , ǫm are constants which depend only on τ and the genus g.
For 1 ≤ i ≤ m and X ∈ Tg let Ei (X) denote the extremal length of the i’th shortest simple
closed curve on X. Let f0 = 1 and for 1 ≤ j ≤ m let
Y  ǫi s
fj (X) = .
Ei (X)
1≤i≤j

Note that fj is invariant under the action of the mapping class group, and thus descends to a function
on Mg . Let
m
X
u(X) = fj (X). (22)
k=1

13
Let ǫ′j = ǫj /(mC2 ). Let
Wj = {X ∈ Tg : Ej+1 (X) > ǫ′j }.
Note that W0 is compact, and on Wj there are at most j short simple closed curves. If X 6∈ Wj−1
then X has at least j short simple closed curves, and thus if X ∈ Wj \ Wj−1 then X has exactly j
short simple closed curves.
In this subsection, we prove the following:
Proposition 4.1 Set s = 1/2. Then we may write

(Aτ u)(X) ≤ c(X)u(X) + b(X), (23)

where b(X) is a bounded function which vanishes outside the compact set W0 , and for all j and for
all X 6∈ Wj−1 ,
c(X) ≤ Cj′ τ j e−jτ ,
where Cj′ depends only on the genus.
We now begin the proof of Proposition 4.1. If α1 , . . . , αj are disjoint simple closed curves let
P (α1 , . . . , αj ) denote the product region where for all 1 ≤ i ≤ j, the extremal length of αi is at most
Cǫi .

Lemma 4.2 Suppose 1 ≤ j ≤ m, 1/2 ≤ s < 1, and suppose X and τ are such that BT (X, τ ) is
completely contained in P (α1 , . . . , αj ). Also suppose that for all Y ∈ BT (X, τ ), the set {α1 , . . . , αj }
coincides with the set of the j shortest curves on y. Then

(Aτ fj )(X) ≤ cj fj (X),

where for 1/2 < s < 1,

cj = Cj (s)e−2j(1−s)τ . (24)
j −jτ
and if s = 1/2, cj = Cj τ e .

Sketch of proof. We will use Minsky’s product region theorem as stated in Theorem 2.3. Let m′
denote the product measure on (H2 )j × T ′ , and let A′τ denote the averaging operator with respect
to the product measure m′ , i.e. for a real-valued function f ,
Z
′ 1
(Aτ f )(X) = ′ f (Y ) dm′ (Y ).
m (BT (X, τ )) BT (X,τ )

We first establish the lemma with Aτ replaced by A′τ .

We may write X = (X1 , . . . , Xj , X ′ ) where Xk is in the k’th copy of the hyperbolic plane and
X ∈ T ′ . Because of the product region theorem BT (X, τ ) is essentially BT (X1 , τ )×. . .×BT (Xj , τ )×
′

B ′ (X ′ , τ ) where for 1 ≤ k ≤ j, BT (Xk , τ ) is a ball of radius τ in the hyperbolic plane and B ′ (X, τ )
a ball of radius τ in T ′ . Also, by assumption, for any Y ∈ B(X, τ ) the set of j shortest simple closed
curves on Y is {α1 , . . . , αj }. Thus, for Y ∈ B(X, τ ), with Y = (Y1 , . . . , Yj , Y ′ ), we have
j
Y
s
fj (Y ) ≈ (ǫ1 . . . ǫj ) ℓmin (Yk )−2s
k=1

14
where for Yk ∈ H2 , ℓmin (Yk ) is the flat length of the shortest simple closed curve in the torus
parametrized by Yk . We remark that the exponent is 2s instead of s because on the torus extremal
length is the square of flat length. As a result, we have
j Z !
Y 1
(A′τ fj )(X) ≈ (ǫ1 . . . ǫj )s ℓmin (Y )−2s d Vol(Y ) , (25)
Vol(B(Xk , τ ) B(Xk ,τ )
k=1

where Vol is the standard volume form on H2 . Now the integral in the parenthesis, i.e. an average
of ℓ−2s over a ball in a hyperbolic plane is essentially done in [EM, Lemma 7.4] (except that there
the average is over spheres, but to get the average over balls one just makes an extra integral over
the radius). One gets for 1/2 < s < 1,
Z
1
ℓmin (Y )−2s d Vol(Y ) ≤ c(s)e−2(1−s)τ ℓmin (X)
Vol(B(X, τ ) B(X,τ )

and for s = 1/2, Z

1
ℓmin (Y )−1 d Vol(Y ) ≤ c′ τ eτ ℓmin (X).
Vol(B(X, τ ) B(X,τ )

Substituting these expressions into (25) completes the proof of Lemma 4.2 with A′τ instead of Aτ .
In view of the form of the function fj ,
X Z X
c1 fj (Y ) ≤ fj (Y ) dm(Y ) ≤ c2 fj (Y ). (26)
B(X,τ )∩N B(X,τ ) B(X,τ )∩N

Choose L ≫ b (where b is as in Theorem 2.3), and choose a (L, 2L)-separated net Nk in each factor.
Let N be the product of the Nk . Then N is an (L − b, 2L + b)-separated net in P (α1 , . . . , αj ). Now
in view of (17),
j
Y j
Y
m(B(X, τ )) ≈ |B(X, τ ) ∩ N | ≈ |B(Xk , τ ) ∩ Nk | ≈ m′ (B(Xk , τ )) ≈ m′ (B(X, τ )),
k=1 k=1

where, as before, A ≈ B means that the ratio A/B is bounded by two constants depending only on
b, L and g, and thus ultimately only on g. Similarly, using (26), we can show that
Z Z
fj (Y ) dm(Y ) ≈ fj (Y ) dm′ (Y ).
B(X,τ ) B(X,τ )

Thus (A′τ fj )(X) ≈ (Aτ fj )(X). 

Remark. The proof works even if at some point Y ∈ B(X, τ ) there are short simple closed curves
other then {α1 , . . . , αj } (but these other curves are longer then the maximum of the lengths of the
αj at Y ). This is used in the next lemma.
Pm
Lemma 4.3 For 1 ≤ j ≤ m, let uj (X) = k=j fj (X). Suppose Ej (X) < ǫj . Then (assuming τ is
large enough),  
1
(Aτ uj )(X) ≤ cj + uj (X), (27)
2C

15
where cj are as in (24). In particular, letting j = 1, and noting that the set {X ∈ Mg : E1 (X) > ǫ1 }
is compact, we have for all X ∈ Tg ,
 
1
(Aτ u1 )(X) ≤ c1 + u1 (X) + b(τ ).
2C

Proof. Note that for any 1 ≤ i ≤ m, and any X ∈ Tg ,

1
fi (X) ≤ (Aτ fi )(X) ≤ Cfi (X) (28)
C
(this is because in B(X, τ ) the extremal length of any simple closed curve cannot change by more
than e2τ ).
We divide the set {j, j +1, . . . m} into two disjoint subsets: Let I1 be the set of k ∈ {j, j +1, . . . m}
such that
uj (X)
fk (X) ≤ , (29)
2mC2
and let I2 be the set of k ∈ {j, j + 1, . . . m} such that the opposite inequality to (29) holds. Suppose
1
k ∈ I1 . Then, by (28), (Aτ fk )(X) ≤ 2mC fk (X), and thus
X 1
(Aτ fk )(X) ≤ uj (X). (30)
2mC
k∈I1

Now suppose k ∈ I2 . We claim that

Ek (X) < (2mC2 )1/s ǫk . (31)

Indeed, if k = j then (31) is true by assumption. If k > j then

 s  s
Ek (X) uj (X) Ek (X)
uj (X) ≥ fk−1 (X) = fk (X) ≥ ,
ǫk 2mC2 ǫk

where we have used the inequality opposite to (29) in the last estimate. Thus (31) follows.
We now claim that under the assumption that k ∈ I2 we have

Ek+1 (X) ≥ C2 Ek (X). (32)

If k = m this is clear from (31) (since in the case where Em (X) is small, there are no other short
simple closed curves on X). Now if k < m, then
 s  s
ǫk+1 uj (X) ǫk+1
uj (X) ≥ fk+1 (X) = fk (X) ≥
Ek+1 (X) 2mC2 Ek+1 (X)
where again we used the inequality opposite to (29) in the last estimate. Thus,
ǫk+1
Ek+1 (X) ≥ . (33)
(2mC2 )1/s

Now (32) follows from (33), (31) and (21).

16
 Now in view of (31) and (32), Lemma 4.2 can be applied to fk . Thus, for k ∈ I2 ,

(Aτ fk )(X) ≤ ck fk (X) ≤ cj fk (X) (34)

where for the last inequality we assumed that τ was large enough so that ck < cj for k > j. Now
(27) follows from (30) and (34). 

Proof of Proposition 4.1. Suppose X 6∈ Wj−1 . We may write

j−1
X
u(X) = uj (X) + fk (X).
k=1

Note that for X 6∈ Wj−1 and k < j, fk (X) ≤ uj (X)/(mC2 ). Hence, (23) follows from (27) and (28).


4.2 A uniform estimate for the measure of a ball

Given X and Y in Tg and τ > 0, let

Fτ (X, Y ) = {Z | Z ∈ Γg · Y, dT (X, Z) ≤ τ } ⊂ Tg .

In the Appendix we show:

Proposition 4.4 Given ǫ > 0, there exists τ0 > 0 such that for any τ > τ0 and X, Y ∈ Tg we have

|Fτ (X, Y )| ≤ e(h+ǫ)τ G(Y )2 .

In order to prove this statement we show that in general

|Fτ (X, Y )| = O(τ 3g−3 ehτ G(X)G(Y )). (35)

Using these estimates we show that:

Proposition 4.5 There exists a constant C2 such that for any X, any δ ′′ > 0 and any sufficiently
′′
large τ , the volume of any B(X, τ ) is bounded by C2 e(h+δ )τ .

Proof. See Appendix. 

Remark. Lemma 7.2 in the appendix is a stronger version of (35). The proof of this lemma
uses Theorem 2.4. This statement can be generalized for other strata of moduli space of quadratic
differentials.

4.3 Proof of Theorem 1.2

As in 3.2, we discretize Teichmüller space by fixing a (c, 2c) separated net N ⊂ Tg . We note that
there exists a constant κ0 such that for all X ∈ Tg ,
1
u(X) ≤ G(X) ≤ κ0 u(X),
κ0

17
where κ0 depends on τ and the ǫi (and thus ultimately only on τ and the genus). Also, as in (6),
Y
G(X) = Ei (X)−1/2 .
1≤i≤m

Trajectories of the random walk. Suppose R ≫ τ and let n be the integer part of R/τ . By a
trajectory of the random walk we mean a map λ : {0, n − 1} → Tg such that :
• for all 0 < k ≤ n − 1 we have d(λk , λk−1 ) ≤ τ , and
• λk belongs to the net N .
Let Pτ (X, R) denote the set of all trajectories for which d(λ0 , X) = O(1). It is a corollary of
Proposition 4.5 that ′′
|Pτ (X, R)| ≤ C2 e(h+δ )R , (36)
where | · | denotes the cardinality of a set.
We say that a trajectory is almost closed in the quotient if the distance in Mg between the
projection to Mg of λ(0) and the projection to Mg of λn−1 is O(1).
Let δ > 0 be a constant to be chosen later. (We will have δ < ǫ′j for 1 ≤ j ≤ m, where the ǫ′j are
as in §4.1). For j ∈ N, let Pj,τ (X, δ, R) denote the set of all trajectories starting within O(1) of X
for which at any point, there are at least j simple closed curves of length at most δ. Let P ej (X, δ, R)
denote the subset of these trajectories which are almost closed in the quotient. We use the systems
of inequalities discussed in §4.1 to show:

Lemma 4.6 Let j ∈ N. For any ǫ′ > 0, there exists τ0 > 0 such that for τ > τ0 , δ > 0 small enough
(depending on τ , ǫ′ and g), and any sufficiently large R (depending only on ǫ′ and τ ) we have:
′
ej,τ (X, δ, R)| ≤ e(h−j+ǫ )R .
|P (37)

Proof. By the definition of u for s = 1/2 (see §4.1), it is easy to check that for any Y ∈ Tg

1
u(Y ) ≤ G(Y ) ≤ κ0 u(Y ), (38)
κ0
where
κ 0 ≤ e m0 τ .
Here m0 is a universal constant independent of τ and Y . Also, by the definition of the function u if
dT (X1 , X2 ) = O(1), then
u(X1 )
= O(1). (39)
u(X2 )
ej (X, δ, R) then G(X) ≈ G(λn−1 ) ≤ κ0 u(λn−1 ). Let R = nτ , and let
Note that if λ ∈ P
X
qj (X, R) = u(λn−1 )
λ∈Pj (X,δ,R)

18
Therefore,
C1 X C1 κ0
ej (X, δ, R)| ≤
|P κ0 u(λn−1 ) = qj (X, R), (40)
G(X) G(X)
ej (X,δ,R)
λ∈P

where C1 = O(1). For 0 < r = kτ < R let

X
qj (X, R, r) = u(λk−1 ),
λ∈Pj (X,δ,R,r)

where the elements of Pj (X, δ, R, r) are the trajectories λ belonging to Pj (X, δ, R) but truncated
after k = r/τ steps. Then X
qj (X, R, r + τ ) = u(λk ) (41)
λ∈Pj (X,δ,R,r+τ )

X X
= u(λk )
λ∈Pj (X,δ,R,r) λk ∈N ∩B(λk−1 ,τ )
X Z
≤C u(Y ) dm(y)
λ∈Pj (X,δ,R,r) B(λk−1 ,τ )
X
=C m(B(λk , τ )) (Aτ u)(λk−1 )
λ∈Pj (X,δ,R,r)

where in the next to last line we use (39) and Lemma 3.1 to estimated a sum over N ∩ B(λk−1 , τ )
by a constant C times an integral over B(λk−1 , τ ).
Note that for λ ∈ Pj (X, δ, R), the number of simple closed curves shorter then δ on λk−1 is at
least j. Thus, if δ is small enough, (depending on the the ǫ′j and thus ultimately only on τ and
the genus), λk−1 6∈ Wj−1 . Then, from Proposition 4.1, and assuming τ is large enough so that
Proposition 4.5 holds with δ ′′ < ǫ′ /3, we get
X ′ ′
qj (X, R, r + τ ) ≤ CC2 e(h+ǫ /3)τ Cj′ τ e−jτ u(λk−1 ) = CCj′ C2 τ e(h+(ǫ /3)−j)τ qj (X, R, r).
λ∈Pj (X,δ,R,r)
(42)
Now iterating (42) n = R/τ times we get
′ τ)
log(CC2 Cj
′ ′
qj (X, R) ≤ u(X)(CC2 Cj′ )n τ n e(h−j+ǫ /3)nτ = u(X)e(h−j+ǫ /3+ τ )R
. (43)

Hence, in view of (40) and (38), we get

′ τ)
log(CC2 Cj ′ τ)
log(CC2 Cj
′ ′ τ
|Pej (X, δ, R)| ≤ C1 e2m0 τ × e(h−j+ǫ /3+ τ )R
≤ C1 e(h−j+ǫ /3+ τ +2m0 R )R
. (44)
log(C C2 Cj′ τ0 ) 6m0 τ +3 log(C1 )
Now it is enough choose τ0 so that τ0 < ǫ′ /3. Then for any τ ≥ τ0 and R ≥ ǫ′
(44) implies (37).


Remark. One can use Lemma 7.2 to prove Lemma 4.6.

19
 Let Nj (X, δ, R) be the number of conjugacy classes of closed geodesics of length at most R which
pass within O(1) of the point X and always have at least j simple closed curves of length at most
δ.
Lemma 4.7 For any ǫ′ > 0 we may choose τ large enough (depending only on ǫ′ ) so that for any
X ∈ Tg , any δ < 1/c0 and any sufficiently large R (depending only on ǫ′ and τ ) we have

Nj (X, δ, (1 − ǫ′ )R) < C|Pej (X, c0 δ, R)|, (45)

where C and c0 are constants which only depend on N and g.

Proof. Let
IX = {g ∈ Γg |BT (X, g · X) ≤ 1}.
In other words, IX is the subset of the mapping class group which moves X by at most O(1). In
fact, |IX | ≈ G(X)2 (See also Theorem 7.2).
Now consider a closed geodesic γ in Mg which passes within O(1) of p(X) (recall that p denotes
the projection map p : Tg → Mg ). Let [γ] denote the corresponding conjugacy class in Γg . Then
there are approximately |IX | lifts of [γ] to Tg which start within O(1) of X. Each lift γ̃ is a geodesic
segment of length equal to the length of γ.
We can mark points distance τ apart on γ̃, and replace these points by the nearest net points.
(This replacement is the cause of the ǫ′ ). This gives a map Ψ from lifts of geodesics to trajectories.
If the original geodesic γ has length at most (1 − ǫ′ )R and always has j simple closed curves shorter
then δ, then by Theorem 2.2 the resulting trajectory belongs to P ej (X, c0 δ, R), where c0 only depends
on N .
If two geodesic segments map to the same trajectory, then the segments fellow travel within O(1)
of each other. In particular if g1 and g2 are the pseudo-Anosov elements corresponding to the two
geodesics, then dT (g−1 −1
2 g1 X, X) = O(1), thus g2 g1 ∈ IX .
We now consider all possible geodesics contributing to Nj (X, δ, (1 − ǫ′ )R); for each of these we
consider all the possible lifts which pass near X, and then for each lift consider the associated random
walk trajectory. We get:
ej (X, c0 δ, R)|
Nj (X, δ, (1 − ǫ′ )R)|IX | ≤ C|IX ||P

(the factor of |IX | on the left hand side is due to the fact that we are considering all possible lifts
which pass near X, and the factor of |IX | on the right is the maximum possible number of times a
given random walk trajectory can occur as a result of this process). Thus, the factor of |IX | cancels,
and the lemma follows. 
The following lemma is due to Veech [Ve].
Lemma 4.8 Suppose γ is a closed geodesic of length at most R on Mg . Then for any X ∈ γ the
extremal length of the shortest simple closed curve on X is at least ǫ′0 e−(6g−4)R , where ǫ′0 depends
only on g.

Proof. We reproduce the proof for completeness. Let X̃ be some point in Tg with p(X̃) = X.
Suppose the estimate is false, and let α be a simple closed curve on X̃ with extremal length less

20
than ǫ′0 e−(6g−4)R . Let g be the element of the mapping class group associated to the lift of γ passing
through X̃. Extα (X̃) ≤ ǫ′0 e−(6g−4)R . Therefore, by Theorem 2.2, for j ∈ N,

Extgj α (X̃) = Extα (g−j X̃) ≤ ǫ′0 e−(6g−4+2j)R

In particular, for 1 ≤ j ≤ 3g − 2, Extgj α (X̃) ≤ ǫ′0 . Therefore (assuming that ǫ′0 is small enough), for
1 ≤ j ≤ 3g − 2,the simple closed curves gj α are disjoint. This is a contradiction. 
Remark. Recall that by (5) for very short curves the hyperbolic and the extremal lengths are
comparable. Therefore hyperbolic length of the shortest closed geodesic on X ∈ γ is at least
ǫ′′0 e−(6g−6)R , where ǫ′′0 depends only on g.
Proof of Theorem 1.2. Let ǫ′ = ǫ/8c0, where c0 is the constant in Theorem 4.7. By Lemma 4.7
and Lemma 4.6 we can choose τ and δ so that (45) holds and also (37) holds with δ replaced by c0 δ.
We get, for sufficiently large R,
′
Nj (X, δ, R) < C ′ e(h−j+ǫ c0 /4)R . (46)
P
Finally Nj (δ, R) is at most X Nj (X, δ, R), where we have to let X vary over the net points within
distance 1 of a fundamental domain for the action of the mapping class group. In view of Lemma 4.8,
the number of relevant points in the net is at most polynomial in R. Thus Theorem 1.2 follows. 

5 Recurrence of Geodesics
In this section, we discuss basic recurrence properties of closed geodesics in Q1 Mg . Note that a
Teichmüller geodesic γ is in fact a path in the the space of unit area holomophic quadratic differentials
on surfaces of genus g. As in §2.2, let Q1 Mg (1, . . . , 1) ⊂ Q1 Mg denote the principal stratum, i.e.
the set of pairs (X, q) where q is a holomoprhic quadratic differential on X with simple zeroes.
Notation. For a compact subset K of Mg and a number θ > 0 let N K (θ, R) denote the number of
closed geodesics γ of length at most R such that γ spends at least θ-fraction of the time outside K.
Similarly, for a compact subset K of Q1 Mg (1, . . . , 1) and θ > 0 we denote by N K (θ, R) the
number of closed geodesics γ of length at most R such that γ spends at least θ-fraction of the time
outside K.
We prove the following theorems:

Theorem 5.1 Suppose θ > 0. Then there exists a compact subset K of Mg and δ > 0 such that for
sufficiently large R,
N K (θ, R) ≤ e(h−δ)R .

Theorem 5.2 Suppose θ > 0. Then there exists a compact subset K ⊂ Q1 Mg (1, . . . , 1) and δ > 0
such that for sufficiently large R,
N K (θ, R) ≤ e(h−δ)R .

Proof of 5.1. In view of Corollary 1.3, there exists a compact set K1 , such that the number of
geodesics which do not return to K1 is O(e(h−0.99)R ). But then using Proposition 4.1 and [Ath,
Corollary 2.7] (cf. [Ath, Theorem 2.3]) , there exists a compact set K depending on K1 , θ and there

21
exists δ ′ > 0, such that the number of random walk trajectories which start in K1 and spend at
′
least θ-fraction of the time outside of K is at most e(h−δ )R . (Note that even though [Ath, Corollary
2.7] is not stated with an exponential bound, the proof does in fact imply this, as is done in the
statement and proof of [Ath, Theorem 2.3]). It follows that the same kind of estimate is true for the
number of geodesics (see the proof of Lemma 4.7). 
The rest of this subsection will consist of the proof of Theorem 5.2.

5.1 Choosing hyperbolic neighborhoods of points

Notation. Given q ∈ Q1 Mg we let ℓmin (q) denote the length of the shortest saddle connection
on q in the flat metric defined by q. Suppose K ⊂ Mg is a compact set. For simplicity, we denote
the preimage of K in Tg by K e = P −1 (K). Also, to simplify the notation, let qX,Y ∈ Q1 (X) be the
quadratic differential such that gdT (X,Y ) (q) ∈ Q1 (Y ).
As in [ABEM, §2], we denote the strong unstable, unstable, stable and strong stable foliations
of the geodesic flow by F uu , F u , F s and F ss respectively; for a given quadratic differential q,

F ss (q) = {q1 ∈ Q1 Tg | Re(q1 ) = Re(q)}, F s (q) = {q1 ∈ Q1 Tg | [Re(q1 )] = [Re(q)]},

and
F uu (q) = {q1 ∈ Q1 Tg | Im(q1 ) = Im(q)}, F u (q) = {q1 ∈ Q1 Tg | [Im(q1 )] = [Im(q)]}.
We consider the distance function defined by the modified Hodge norm dH on each horosphere
F ss . This is closely related to the Hodge norm studied by Forni [Fo]. We also consider dE (·, ·) the
Euclidean metric on each horosphere as defined in [ABEM, §3.5]. In fact, the Euclidean norm on
the tangent space of Q1 Mg is defined using period coordinates (see §2.3). We remark that this
norm depends on the choice of a triangulation on the surface. However, in a given compact subset of
Q1 Mg it is well-defined up to a multiplicative constant. So we can use it to measure the ”distance”
between two quadratic differentials in a compact subset of Q1 Mg . We will show:
Lemma 5.3 Suppose K ⊂ Mg is compact. Given ρ, ǫ > 0, and 1 > b > 0, there exists ρ0 > 0
e dT (p0 , p1 ) < ρ0 ,
(depending only on K, and b), and L0 = L0 (K, ǫ, ρ, b) such that if X, p0 ∈ K,
dT (X, p1 ) = L > L0 , and

{s |s ∈ [0, L] , ℓmin (gs (qX,p0 )) ≥ ǫ} > bL,

then
dE (q, qX,p0 ) < ρ,
where q is the unique quadratic differential in F uu (qX,p0 ) ∩ F s (qX,p1 ).
Remark. Note that given q1 , q2 ∈ Q1 Tg , |F uu (q1 ) ∩ F s (q2 )| ≤ 1. In general, this set can be empty,
but if q2 is close enough to q1 then F uu (q1 ) ∩ F s (q2 ) 6= ∅. The proof of Lemma 5.3 implies that
when ρ0 is small and L0 is large F uu (qX,p0 ) ∩ F s (qX,p1 ) 6= ∅.
We will show:
Lemma 5.4 Suppose K ⊂ Mg is compact. Given ǫ > 0, there exists a constants L0 , depending only
on ǫ and K, and c0 depending only on K with the following property.
If

22
 • γ : [0, L] → Q1 Tg is a geodesic segment (parametrized by arclength) such that (p ◦ π)(γ(0)) ∈
K, (p ◦ π)(γ(L)) ∈ K, and L > L0 ,
• γ̂ : [0, L′ ] → Q1 Tg is the geodesic segment connecting p1 , p2 ∈ Tg such that

dT (p1 , π(γ(0)) < c0 , dT (p2 , π(γ(L))) < c0 ,

and
•
|{t ∈ [0, L] : ℓmin (γ(t)) > ǫ}| > (1/2)L,

then
|{t ∈ [0, L′ ] : ℓmin (γ̂(t)) > ǫ/4}| > (1/3)L′ .

5.2 Hodge and Euclidean distance functions on the stable and unstable
foliations
First, we briefly recall some useful decay properties of the Hodge and Euclidean distances proved in
[ABEM, §3.4, §3.5]:

• P1: Given ǫ > 0, there exists ǫ′ > 0 such that if ℓmin (q1 ) ≥ ǫ and dE (q1 , q2 ) ≤ ǫ′ then
ℓmin (q2 ) ≥ ǫ/2.

• P2: Assume that K e 1 ⊂ Tg is the preimage of a compact subset K1 ⊂ Mg . Given ρ1 > 0, there
e 1 , and dE (q1 , q2 ) > ρ1 , then dT (π(q1 ), π(q2 )) > ρ0 .
exists ρ0 such that if q1 ∈ F ss (q2 ), π(q1 ) ∈ K
• P3: There exists C1 > 0 such that if dH (q1 , q2 ) < 1, q1 ∈ F ss (q2 ) and s ≥ 0

dE (gs q1 , gs q2 ) ≤ dH (gs q1 , gs q2 ) < C1 dH (q1 , q2 ); (47)

• P4: Moreover, given ǫ, b > 0 , there exist C0 , C0′ and a > 0 such that for any q1 ∈ F ss (q2 )
with dH (q1 , q2 ) < 1, and s ≥ 0 if

|{t| t ∈ [0, s], ℓmin (gt q1 ) > ǫ}| > b s, (48)

e 1 , then
and π(gs q1 ) ∈ K

dE (gs q1 , gs q2 ) ≤ C0′ dH (gs q1 , gs q2 ) < C0 e−as dH (q1 , q2 ). (49)

e 1 , b and ǫ) such that for

Note that in this case, by (49), there exists L0 (depending only on K
s > L0 , (48) implies that :

b
|{t| t ∈ [0, s], ℓmin (gt q2 ) > ǫ/2}| > s. (50)
2

23
 f (q) γ(0)
F0

S0

Figure 1. Proof of Lemma 5.3.

Proof of Lemma 5.3. Let BE (q, r) denote the ball of radius r with center q in the Euclidean
metric. Let S0 = S(X) denote the sphere at X, i.e. the set of unit area holomorhic quadratic
differentials on the surface X, and γ(t) = gt (qX,p0 ). For q ∈ S0 near γ(0) = qX,p0 , let
f (q) = F uu (qX,p0 ) ∩ F s (q);
in other words, we can choose t(q) ∈ R be such that f (q) ∈ F uu (qX,p0 ) and gt(q) q and f (q) are on the
same leaf of F ss . Then clearly f (qX,p0 ) = qX,p0 , and there exists a number ρ3 > 0 depending only on
K such that the restriction of f to BE (qX,p0 , ρ3 ) ∩ S0 is a homeomorphism onto a neighborhood of
qX,p0 in F uu (qX,p0 ). In particular, for any q ∈ BE (qX,p0 , ρ3 )∩S0 , we know that F uu (qX,p0 )∩F s (q) 6=
∅. Let [
U= {gt q : q ∈ BE (qX,p0 , ρ3 /2) ∩ S0 }.
|t|<ρ3

e 0 depending on ǫ and K such that for L > L

Claim: there exists ρ0 > 0 depending only on K, and L e0 ,

π(gL U ) ⊃ BT (π(γ(L)), ρ0 ). (51)

This is straightforward (in view of the non-uniform hyperboliclity as in (47) and (49)) but
somewhat tedious argument. Let
[
V = {gt+t(q) f (q) : q ∈ BE (qX,p0 , ρ3 /2) ∩ S0 } ⊂ F u (qX,p0 ).
|t|<ρ3

24
Then V is relatively open as a subset of F u (qX,p0 ). Let ∂V denote the boundary of V viewed as a
subset of F u (qX,p0 ). Therefore we can choose ρ2 > 0 depending only on K such that for all q ′ ∈ ∂V ,

dE (q ′ , qX,p0 ) ≥ ρ2 .

By (47) and the fact that ∂V ⊂ F u (qX,p0 ), this implies that for some constant ρ′2 , (depending only
on K), all q ′ ∈ ∂V and all L > 0,
dE (gL q ′ , γ(L)) ≥ ρ′2 . (52)
S
Note that U ⊂ t∈R gt S0 ∼ = R × S0 . Let ∂U denote the boundary of U viewed as a subset of
R × S0 . Suppose q1 ∈ ∂U . We may write q1 = gt q for some q ∈ S0 . Then the fact that q1 ∈ ∂U
implies that either dE (q, qX,p0 ) = ρ3 /2 or |t| = ρ3 . In either case, let q2 = gt+t(q) f (q). Then q2 ∈ ∂V ,
q1 and q2 are on the same leaf of F ss , and

dE (q1 , q2 ) ≤ C,

where C depends only on K. Hence, by (47), we have

dE (gL q1 , gL q2 ) ≤ C · C0 . (53)

In order to prove the claim, we show that there exists L0 (depending only on K and ǫ) such that
for L > L0 ,
dE (gL q1 , γ(L)) ≥ ρ′2 /2. (54)
Suppose that (54) fails. Then:
• by (53), dE (gL q2 , γ(L)) ≤ C2 , where C2 only depends on K. On the other hand, we can choose
|t0 | ≤ ρ3 such that gt0 q2 ∈ F ss (qX,p0 ). Using (49) for gL+t0 q2 ∈ F ss (γ(L)), we get that

|{t| t ∈ [0, L], ℓmin (gt q2 ) > ǫ′ }| > ǫ0 s,

where ǫ′ , ǫ0 only depend on K, and ǫ.

• Now we can apply (49) for q1 ∈ F ss (q2 ), and we get dE (gL q1 , gL q2 ) ≤ C0 e−aL dH (q1 , q2 ).
Therefore there exists L e 0 such that if L > L
e 0 , dE (gL q1 , gL q2 ) ≤ ρ′ /2. Then, in view of (52),
2
′
dE (gL q2 , γ(L)) ≥ ρ2 . So for all q1 ∈ ∂U and all L > L e 0 , dE (qL q1 , γ(L)) ≥ ρ′ /2.
2

e 0 and all q1 ∈ ∂U ,
By P2, there exists ρ0 depending only on K such that for L > L

dT (π(gL q1 ), π(γ(L))) ≥ ρ0 . (55)

Let φ : U → Tg denote the map φ(q) = π(gL (q)). Then φ is continuous, and φ(qX,p0 ) = π(γ(L)). Now
e 0 , F uu (qX,p0 )∩F s (qX,p1 ) = q
in view of (55), (51) holds. In other words, if dT (p0 , p1 ) < ρ0 , and L > L
is well defined, and dE (q, qX,p0 ) < ρ3 . On the other hand, dH (gL qX,p0 , gL q) < C, where C only
depends on K. Therefore, we can use P4 to get

dE (q, qX,p0 ) < C0 e−aL C,

e 0 (depending on ǫ, ρ, and
where C0 , a > 0 depend on ǫ, b and K. This shows that we can find L0 > L
K) such that if L > L0 we have
dE (q, qX,p0 ) < ρ.

25
 
Proof of Lemma 5.4. It is enough to show that given ǫ > 0 and 1 > m > m′ ≥ 1/2, there exist
c0 > 0 (depending only on K, m and m′ ) and L0 > 0 (depending only on ǫ, m, m′ and K) such that
e dT (p1 , p2 ) < c0 , for i = 1, 2, dT (X, pi ) = ri > L0 , and
if x, p1 ∈ K,

{s |0 < s ≤ r1 , ℓmin (gs (qX,p1 )) ≥ ǫ} > r1 m,

then
{s |0 < s ≤ r2 , ℓmin (gs (qX,p2 )) ≥ ǫ/2} > r2 m′ .
Let q1 = qX,p1 , and (as in the previous lemma) let q = f (qX,p2 ) be the unique point in F uu (q1 ) ∩
F s (qX,p2 ). Now let q2 be a quadratic differential of area 1 on the geodesic joining X to p2 such that
q2 ∈ F ss (q); in particular, we have

q ∈ F uu (q1 ), q ∈ F ss (q2 ). (56)

By P1, we can choose ǫ0 > 0 such that if dE (q ′ , q ′′ ) < ǫ0 , and ℓmin (q ′ ) > ǫ then ℓmin (q ′′ ) > ǫ/2.
Then by Lemma 5.3, P4, and P2, we can choose c0 < 1/2 (depending only on K) and L0 > 0 such
that for dT (X, p1 ) = r > L0 , and dT (p1 , p2 ) < c0 the following hold:

1. As in P4 (for b = m), there exist C0 > 0 and a > 0 such that (49) holds. Then L0 should
satisfy:
C0 e−aL0 /6 < ǫ0 /4,

2. Also, by Lemma 5.3, if L0 is large enough

dE (q1 , q) < ǫ0 /4C0 , dE (q, q2 ) ≤ ǫ0 /4C0 ,

and
dE (gr q1 , gr q2 ) ≤ 1.

As a result, from (47) we get

dE (gr q, gr q2 ) < ǫ0 /4 and dE (gr q1 , gr q) < C1 ,

where C1 only depends on K, and ǫ.

Consider the map between the points on the geodesic [xp1 ] to the points on [xp2 ] as follows:

[xp1 ] → [xp2 ]

gs q1 → gs q2 .
We can choose, 0 ≤ s0 ≤ r such that |{t |s0 < t < r , ℓmin (gt q1 ) ≥ ǫ}| = r(m − m′ ), and let

A = {t |0 < t < s0 , ℓmin (gs q1 ) ≥ ǫ}.

It is easy to check that |A| > rm′ . We claim that for s ∈ A, we have

dE (gs q1 , gs q2 ) ≤ ǫ0 . (57)

This is because:

26
 • For any s ∈ A, (48) holds for gr q and gr q1 and the interval (0, r − s). Hence, by (49),

dE (gs q1 , gs q) = dE (gs−r (gr q1 ), gs−r (gr q)) < ǫ0 /4;

• by (47),
dE (gs q, gs q2 ) < ǫ0 /4;

• Finally, since q1 ∈ F uu (q), dE (gs q1 , gs q2 ) ≤ ǫ0 .

Hence, for s ∈ A, ℓmin (gs q1 ) ≥ ǫ, (57), and P1 imply that ℓmin (gs q2 ) ≥ ǫ/2. 

5.3 Proof of Theorem 5.2

Choose θ1 > 0 . Let K ⊂ Mg be such that Theorem 5.1 holds for K, and θ1 = θ. Let K2 be a
compact subset of Q1 Tg (1, . . . , 1)/Γg such that K2 ⊂ p−1 (K), and let K3 be a subset of the interior
of K2 . We may choose these sets so that µ(K3 ) > (1/2), where µ is defined in §2.2. We also choose
K2 and K3 to be symmetric, i.e if q ∈ K2 then −q ∈ K2 (and same for K3 ). Then there exists ǫ > 0
such that for X ∈ K3 , ℓmin (X) > ǫ. Let c0 be as in Lemma 5.4. We now choose a (c1 , c2 ) separated
net N on Tg , which c1 < c0 , c2 < c0 . We may assume that N ∩ p−1 (K) is invariant under the action
of the mapping class group.
Suppose X ∈ Tg , and as before, let S(X) denote the set of area 1 holomoprhic quadratic differ-
entials on X. Let

U (X, T ) = {q ∈ S(X) : |{t ∈ [0, T ] : gt q ∈ K2c }| > (1/2)T }.

Let [
V (X, T ) = π(gt U (X, t)),
0≤t≤T

so that V (X, T ) is the subset of BT (X, T ) consisting of points Y ∈ BT (X, T ) such that the geodesic
from X to Y spends more than half the time outside K2 .
By [ABEM, Theorem 2.6], for any θ1 > 0, there exists T > 0 such that for any τ > T and any
X ⊂ N ∩ p−1 (K),
m(Nbhdc2 (V (X, τ ) ∩ p−1 (K))) ≤ θ1 ehτ ,
where Nbhda (A) denotes the set of points within Teichmüller distance a of the set A. Then, since
K is compact and θ1 is arbitrary, this implies that for any θ2 > 0 there exists T > 0 such that for
any τ > T and any X ∈ N ∩ p−1 (K),

|N ∩ V (X, τ ) ∩ p−1 (K)| ≤ θ2 ehτ .

By the compactness of K and [ABEM, Theorem 1.2 and Theorem 5.1] there exists C1 > 1 such that
for τ sufficiently large and any X ∈ N ∩ p−1 (K),

C1−1 ehτ ≤ |N ∩ BT (X, τ )| ≤ C1 ehτ .

Thus, for any θ3 > 0 there exists T > 0 such that for τ > T ,

|N ∩ V (X, τ ) ∩ p−1 (K)| < θ3 |N ∩ BT (X, τ )|. (58)

27
From now on we assume that τ is sufficiently large so that (58) holds.
Let K′1 = Nbhdc2 (K), and let G(R) denote the set of closed geodesics in Mg of length at most
′ ′
R, and let G K1 (θ3 , R) ⊂ G(R) denote the subset which contributes to N K1 (θ3 , R). In view of
Theorem 5.1, it is enough to show that there exists δ0 > 0 such that for R sufficiently large,
′
|G(R) \ G K1 (θ3 , R)| ≤ e(h−δ0 )R . (59)

As in §4, we associate a random walk trajectory Φ(γ) to each closed geodesic γ ∈ G(R). Let
′
P1 (R) = Φ(G(R) \ G K1 (θ3 , R)) denote the set of resulting trajectories. Note that by construction,
every trajectory in P1 spends at most θ3 fraction of the time outside K.
Suppose Λ = (λ1 , . . . , λn ) be a trajectory in P1 (R). Let J(Λ) denote the number of j, 1 ≤ j ≤ n
such that
λj ∈ p−1 (Kc ) or λj+1 ∈ V (λj , τ ) ∪ p−1 (Kc ). (60)
Let P2 (R) ⊂ P1 (R) denote the trajectories for which J(Λ) < (θ/2)n. Then, as long as θ3 is chosen
sufficiently small , the Law of Large Numbers implies that there exists δ1 > 0 such that for n = R/τ
sufficiently large,
|P1 (R) \ P2 (R)| ≤ e(h−δ1 )R .
Since every trajectory in P1 (R) intersects the compact set K, for any Λ ∈ P1 (R), the cardinality of
Φ−1 (Λ) is bounded by a constant C depending only on K. Thus,

|Φ−1 (P1 (R) \ P2 (R))| ≤ Ce(h−δ1 )R .

To complete the proof we will show that there exists a compact set K ⊂ Q1 Mg (1, . . . , 1) such that
any geodesic γ ∈ Φ−1 (P2 (R)) spends at least 1 − θ fraction of the time in K. Indeed, suppose
Λ = (λ1 , . . . , λn ) ∈ P2 (R) and γ ∈ Φ−1 (Λ). Then there exist points qj = γ(tj ) ∈ Q1 Mg such
that qj ∈ γ and for 1 ≤ j ≤ n, dT (π(qj ), λj ) ≤ c2 < c0 . Now suppose j 6∈ J(Λ). Then by
Lemma 5.4, there exists ǫ′ > 0 depending only on K and ǫ, and a point tj < t′ < tj+1 such that
ℓmin (γ(t′ )) > ǫ′ . But then for all tj < t < tj+1 , ℓmin (γ(t)) ≥ e−τ ǫ′ , so the entire geodesic γ spends
at least (1 − θ)-fraction of the time in the compact set {q ∈ Q1 Mg : ℓmin (q) > e−τ ǫ′ }. 

6 A Closing Lemma
In this section, we use the properties of the geodesic flow discussed in §5 to prove the following
closing lemma:
Lemma 6.1 (Closing Lemma) Let K be a compact subset of Q1 Mg (1, . . . , 1) consisting of non-
orbifold points of Q1 Mg . Given q ∈ Q1 Mg (1, . . . , 1) and ǫ > 0, there exist constants L0 > 0, ǫ′ > 0,
and open neighborhoods U ⊂ U ′ of q with the following property.
Suppose that γ : [0, L] → Q1 Mg is a geodesic segment (parametrized by arclength) such that
(a) γ(0), γ(L) ∈ U and
(b)
|{t ∈ [0, L] : γ(t) ∈ K}| > (1/2)L,

(c) L > L0 .

28
Let γ1 be the closed path in Mg which is the union of γ and the geodesic segment from π(γ(L)) to
π(γ(0)) in π(U ). Then there exists a unique closed geodesic γ ′ with the following properties:
(a’) γ ′ and γ1 have lifts in Tg which stay ǫ close (with respect to the Teichmüller metric on Tg ).
(b’) The length of γ ′ is within ǫ of L,
(c’) γ ′ passes through U ′ .
Remark. A stronger version of this statement can be found in [H2] (§6).
Outline of Proof. Consider the stable and unstable foliations for the geodesic flow as in §5.1.
Our goal is to show that if U is small enough, the first return map on these foliations will define a
contraction with respect to the modified Hodge distance function. As a result, we find a fixed point
for the first return map in U ′ (which is the same as a closed geodesic going through U ′ ).
From Lemma 5.4, we can find open neighborhoods U ⊂ U ′′ ⊂ U ′ of q, and L0 > 0 such that the
following properties hold :
• if γ : [0, L] → Q1 Mg satisfies properties (a), (b) and (c) then in view of the hyperbolicity
statement (49) the time L geodesic flow restricted to the neighborhood U ′′ expands along the
leaves of F uu and contracts along the leaves of F ss , in the metric dH ,
• for any q1 , q2 ∈ U ′ , if q1 ∈ F ss (q2 ) or q1 ∈ F uu (q2 ) then dH (q1 , q2 ) ≤ ǫ.
We can apply the contraction mapping principle to F ss to find q0 ∈ U ′ such that gL (q0 ) ∈ F uu q0 .
Now we can consider the first return map of the map g−t on F uu (q0 ). Again, we can use (49),
Lemma 5.4 and the contraction mapping principle to find a fixed point for the geodesic flow in U ′ .


6.1 Remark on proof of Theorem 1.1

Note that by the bound proved in Theorem 5.2, we only need to consider the set of closed geodesics
going through a fixed compact subset of Mg . We remark that in Lemma 6.1 if we remove the
assumption that K consists of non-orbifold points then there are at most c0 closed geodesics satisfying
conditions (a′ ), (b′ ) and (c′ ), where c0 is a constant depending only on g.
Roughly speaking, since
• by Theorem 2.1, the geodesic flow on Q1 Mg (1, . . . , 1) is mixing, and
• on a fixed compact subset of Q1 Mg (1, . . . , 1) the geodesic flow is uniformly hyperbolic,
the derivation of Theorem 1.1 from Lemma 6.1 can be done following the work of Margulis [Mar].
See also §20.6 in [KH]. We skip the argument since it has been done carefully in §5 and §6 of [H2].

7 Appendix: Proof of Proposition 4.5

Given X and Y in Tg and τ > 0, let Fτ (X, Y ) denote the intersection of BT (X, τ ) with Γg · Y .
Proposition 4.5 will follow from the following:

29
Proposition 7.1 Given ǫ > 0, there exists τ0 > 0 such that for any τ > τ0 and X, Y ∈ Tg we have

|Fτ (X, Y )| ≤ e(h+ǫ)τ G(Y )2 ,

where G is defined in (6).

Proof of Proposition 4.5 assuming Proposition 7.1. Suppose X ∈ Tg is arbitrary, and suppose
Y ∈Ne . Then, by Proposition 7.1, for τ sufficiently large,

|BT (X, τ ) ∩ Γg Y | ≤ e(h+ǫ/2)τ G(Y )2 .

Note that by Theorem 2.4 (see the calculation in (11))

|Γg · Y ∩ BT (Y, c2 )| ≥ c G(Y )2 . (61)

Thus, since distinct points in N are are least c2 apart,

|BT (X, τ ) ∩ Γg Y ∩ N | ≤ c e(h+ǫ/2)τ .

In view of (18), this implies that

|BT (X, τ ) ∩ N | ≤ c′ τ 6g−6 e(h+ǫ/2)τ ≤ e(h+ǫ)τ .

Now this implies Proposition 4.5 in view of Lemma 3.1. 

Given X, Y ∈ Tg , and B ⊂ CX let

FR (X, Y, B) = {g · Y |g ∈ Γg , dT (X, g · Y ) ≤ R, B ⊂ CX ∩ Cg·Y } ⊂ FR (X, Y ).

Here, as in §2, given Z ∈ Tg the set CZ consists of simple closed curves of extremal length ≤ ǫ20 on
Z.
In general, we have:
Theorem 7.2 Given X, Y ∈ Tg

|FR (X, Y, B)| ≤ C1 R3g−3 e(h−2|B|)R G(X)G(Y ), (62)

where C1 only depends on g.

In the proof we use the following lemma from [ABEM]:
Lemma 7.3 Let α = {α1 , . . . , α3g−3 } be a bounded pants decomposition on X and suppose Y0 ∈
m3g−3
BT (X, R). If hm
α1 · · · hα3g−3 (Y0 ) ∈ BT (X, R) then for 1 ≤ i ≤ 3g − 3
1

p eR
|mi | · Extαi (Y0 ) ≤ C p .
Extαi (X)

Here C is a constant which only depends on g.

30
Proof of Theorem 7.2. To simplify the notation, given X, Y ∈ Tg define
s !
Extα (X)
Dα (X, Y ) = log .
Extα (Y )

Choose bounded pants decompositions PX = {α1 , . . . , α3g−3 } and PY = {β1 , . . . , β3g−3 } on X and
Y . Without loss of generality, we can assume that

α1 = β1 , . . . , αb = βb ∈ B,

where b = |B|.
Given r = (r1 , . . . , r3g−3 ), s = (s1 , . . . , s3g−3 ) ∈ Z3g−3 define

MR,(r,s) (X, Y, B) = MR (X, Y, B) ∩ {Z = g · Y | Dαi (X, gY ) ∈ [si , si + 1], Dβi (g−1 X, Y ) ∈ [ri , ri + 1]}.

By the definition, if MR,(r,s) (X, Y, B) 6= ∅ then for 1 ≤ i ≤ 3g − 3 we have |ri |, |si | ≤ R.

Claim. Given r, s ∈ Z3g−3 with |ri |, |si | ≤ R

|MR,(r,s) (X, Y, B)| ≤ e(h−2|B|)R G(X)G(Y ). (63)

In order to prove this claim, fix Y0 = g0 ·Y ∈ MR,(r,s) (X, Y, B), and consider the pants decomposition
3g−3
α = ∪i=1 αi . Note that given Z = g · Y ∈ MR,(r,s) (X, Y, B),
3g−3
α(Z) = g0 g−1 (PX ) = ∪i=1 g0 g−1 (αi )

defines a pants decomposition. Moreover, we have:

• g0 PY is a bounded pants decomposition on Y0 = g0 · Y.

• Also, p p
ExtY0 (α(Z)) = ExtZ (α) = O(eR ). (64)

• Note that
∪bi=1 αi ⊂ α(Z),
and hence for i ≤ b
i(g0 βi , α(Z)) = 0. (65)

• On the other hand, since Z = g · Y is in MR,(r,s) (X, Y, B) we have:

p
Extβ (g−1 X)
p j = O(eri ).
Extβj (Y )
As a result, we get
p q
i(g0 g−1 (αi ), g0 βj ) = i(αi , g(βj )) ≤ Extαi (X) Extβj (g−1 X) = O(erj ),

and hence
i(α(Z), g0 · βj ) = O(erj ). (66)

31
 • By (64) Theorem 2.4 gives rise to an upper bound on twisting parameter of α(z) around g0 βj
on Y0 :
eR
twY0 (α(Z), g0 βj ) = twY (g−1 (α), βj ) ≤ p . (67)
Extβj (Y )

Now, in view of the Dehn-Thurston parametrization of multicurves, (65), (66) and (67) imply:
3g−3
Y
rb+1 +...rn ( h
2 −b)R
1
|{α(Z) |Z ∈ MR,(r,s) (X, Y, B)}| ≤ e e p (68)
i=1
Extβi (Y )

• On the other hand given Z1 = g1 · Y, Z2 = g2 · Y ∈ Γg · y, we have α(Z1 ) = α(Z2 ) if and only

if g−1 −1
1 g2 (PX ) = PX . If g1 g2 (αi ) = αi for 1 ≤ i ≤ 3g − 3 then there are m1 , . . . m3g−3 ∈ Z
such that
g 1 = hm m3g−3
α1 · · · hα3g−3 · g2 .
1

Since for any Z ∈ MR,(r,s) (X, Y, B)

1 1
p = O( si ),
Extαi (Z) e

from Lemma 7.3 we get

3g−3
!
e( 2 −b)R Y
h
1
|{Z ∈ MR,(r,s) (X, Y, B) | α(Z) = α(Z0 )}| = O p . (69)
e|s| i=1
Extαi (X)

Therefore from (68) and (69) we get

 
|MR,(r,s) (X, Y, B)| = O e|r|−|s|G(X)G(Y )e(h−2|B|)R .

On the other hand, since MR,(r,s) (X, Y, B) = MR,(s,r) (Y, X, B), we have
 
|MR,(r,s) (X, Y, B)| = O e|s|−|r|G(X)G(Y )e(h−2|B|)R .

Therefore, we get (63).

Finally, we have [
MR (X, Y, B) = MR,(r,s) (X, Y, B),
where r = (r1 , . . . , r3g−3 ) and s = (s1 , . . . , s3g−3 ) with |ri |.|si | ≤ R. Hence, in view of (63) we get
(62).

Proof of Proposition 7.1. To simplify the notation, given X ∈ Tg let
p
eα (X) = Extα (X).

As before, we say α is short on X if eα (X) ≤ ǫ0 , and let CX denote the set of all short simple closed
curves on X. Choose a (c, 2c) net N in Tg as in §3.2.

32
 e , the inequality (62) implies that:
Then in view of (61), given Z ∈ Tg , B ⊂ CZ , r > 0 and W ∈ N

|BT (Z, τ ) ∩ p−1 (W ) ∩ NB | ≤ C1 e(h−|B|)r r3g−3 G(Z)G(W )/G(W )2

≤ Cr3g−3 e(h−|B|)r G(Z)/G(W ), (70)

where NB = {Z1 ∈ N |∀α ∈ B, eα (Z1 ) ≤ ǫ0 } ⊂ N , and C is a universal constant independent of X,

Z and r.
We order the elements in CX , (α1 , . . . , αk ), so that

eα1 (X) ≥ . . . ≥ eαs (X) ≥ e−τ > . . . eαk (X).

Then 3g − 3 ≥ k ≥ s. Let τ0 = 0, and for i ≥ 1,

τi = − log(eαi (X)).

Also for 1 ≤ i ≤ s, di = τi − τi−1 , mi = h − k + i − 1, ds+1 = τ − τs , and ms+1 = h − k. Then it is

easy to verify that
1.
s+1
X s+1
X
di = τ, mi di = (h − k + s)τ − τ1 . . . − τs , G(X) = eτ1 +...+τk ;
i=1 i=1

2. By Theorem 2.2 , for 1 ≤ i ≤ s, αi , . . . , αk are all short in BT (X, τi ).

3. Since αs+1 , . . . αk stay short in BT (X, τ ), again by Kerckhoff’s theorem, for any Y ∈ BT (X, τ )

eτs+1 +...+τk ≤ G(Y )e(k−s)τ . (71)

Consider

P = {(X = Z0 , Z1 , . . . , Zs , g · Y = Zs+1 ) |Z1 , . . . , Zs ∈ N , ∀ i, dT (Zi , Zi−1 ) ≤ di , g ∈ Γg },

e ∩ p(BT (X, τ ))},

Z = {(W1 , . . . , Ws ) |Wi ∈ N
and finally
P(W1 , . . . Ws ) = {(X, Z1 , . . . , Zs , g · Y ) ∈ P |π(Zi ) = Wi }.
Note that for any (X, Z1 , . . . , Zs , Zs+1 = g · Y ) ∈ P, Zi ∈ BT (X, τi ).
On the other hand, since we can approximate a geodesic by points in the net N , we have

|Fτ (X, Y )| ≤ |P|,

also by the definition, [

P⊂ P(W ).
W ∈Z

By equation (18) we have

|Z| = O(τ cg ),
where cg ≤ (12g − 12)s < (12g − 12)2 .

33
 Let Z0 = X. Then for a given Z = Zi−1 ∈ N ∩ BT (X, τi−1 ), BT (Z, di ) ⊂ BT (X, τi ). This implies
that h − mi simple closed curves (i.e, the curves in Bi = {αi , . . . , αk }) are short on BT (Z, di ).
Now, since di ≤ τ, equation (70) for B = Bi , W = Wi , and Z = Wi−1 implies that

|p−1 (Wi )∩N ∩BT (Zi−1 , di )| = |p−1 (Wi )∩NBi ∩BT (Zi−1 , di )| ≤ Cτ 3g−3 emi di G(Wi−1 )/G(Wi ). (72)

Given (Z1 , . . . , Zs ) ∈ Z, we can apply (72) for (X, W1 ), . . . , (Ws−1 , Ws ), and apply (62) for
(Ws , Y ). We get

|P(W1 , . . . , Ws )| ≤ Cτ 3g−3 em1 d1 G(X)/G(W1 ) . . . Cτ 3g−3 ems ds G(Ws−1 )/G(Ws )×

e(h−k+s)τ
×Cτ 3g−3 ems+1 ds+1 G(Ws )G(Y ) = C s τ (3g−3)s G(X)G(Y ) =
eτ1 +...τs
2
= C s τ (3g−3)s e(h−k+s)τ × eτs+1 +...+τk G(Y ) ≤ C 3g−3 τ (3g−3) ehτ G(Y )2 .
We are using (71) to obtain the last inequality. Now we have,
′
|P| ≤ |Z| ehτ G(Y )2 ≤ C ′ τ cg ehτ G(Y )2 ,

where c′g = cg + (3g − 3)2 = O(g 2 ), and C ′ = C 3g−3 . 

References
[AY] P. Arnoux and J. Yoccoz, Construction de diff ’eomorphismes pseudo-Anosov. C. R. Acad.
Sci. Paris Ser. I Math., 29(1981), 75–78.
[AGY] A. Avila, S. Gouezel, and J.-C. Yoccoz, Exponential mixing for the Teichmüller flow.
Publ. Math. IHES, 104(2006), 143–211.
[AR] A. Avila, and M. Resende, Exponential Mixing for the Teichmüller flow in the Space of
Quadratic Differentials. Preprint.
[Ath] J. Athreya, Quantitative recurrence and large deviations for Teichmüeller geodesic flow.
Geom. Dedicata 119(2006), 121–140.
[ABEM] J. Athreya, A. Bufetov, A. Eskin and M. Mirzakhani, Lattice Point Asymptotics and
Volume Growth on Teichmüller space. Preprint
[Be] L. Bers, An extremal problem for quasiconformal maps and a theorem by Thurston. Acta
Math. 141(1978), 73–98.
[Bu] A. Bufetov, Logarithmic asymptotics for the number of periodic orbits of the Teichmueller
flow on Veech’s space of zippered rectangles. Mosc. Math. J. 9(2009), no. 2, 245–261.
[Bu] P. Buser, Geometry and spectra of compact Riemann surfaces, Progr. Math., 106,
Birkhuser, Boston, 1992.
[EMM] A. Eskin, G. Margulis, and S. Mozes, Upper bounds and asymptotics in a quantitative
version of the Oppenheim conjecture. Ann. of Math. (2) 147(1998), no. 1, 93–141.

34
[EM] A. Eskin and H. Masur, Asymptotic formulas on flat surfaces. Ergodic Theory Dynam.
Systems. 21(2001), 443–478.
[FaMa] B. Farb and D. Margalit. A primer on mapping class groups.
http://www.math.utah.edu/ margalit/primer.
[FLP] A. Fathi, F. Laudenbach, and V. Ponaru, Travaux de Thurston sur les surfaces. Aster-
isque, Vols. 66 and 67 (1979).
[Fo] G. Forni, Deviation of ergodic averages for area-preserving flows on surfaces of higher
genus. Ann. of Math. (2) 155(2002), no. 1, 1–103.
[Hu] J. Hubbard, Teichmüller theory and applications to geometry, topology, and dynamics
I, Matrix Editions.
[H1] U. Hamenstädt, Bernoulli Measures for the Teichmüller Flow. Preprint.
[H2] U. Hamenstädt, Bowen’s construction for the Teichmüeller flow. Preprint.
[H3] U. Hamenstädt, Dynamics of the Teichmueller flow on compact invariant sets. J. Mod.
Dynamics 4(2010), 393–418.
[HP] J. L. Harer and R. C. Penner, Combinatorics of Train Tracks, volume 125, Annals of
Math. Studies. Princeton University Press, 1992.
[Iv] N. V. Ivanov, Coefficients of expansion of pseudo-Anosov homeomorphisms. Zap. Nauchn.
Sem. Leningrad. Otdel. Mat. Inst. Steklov. (LOMI), 167(1988), 111–116.
[KH] A. Katok, B. Hasselblat, Introduction to the Modern Theory of Dynamical Systems.
Cambridge University Press, 1995.
[Ker] S. Kerckhoff, The asymptotic geometry of Teichmüller space. Topology. 19(1980), 23–41.
[Mar] G.A. Margulis, On some aspects of the theory of Anosov flows, Ph.D. Thesis, 1970,
Springer, 2003.
[M] B. Maskit, Comparison of hyperbolic and extremal lengths. Ann. Acad. Sci. Fenn.,
10(1985), 381–386.
[Mas] H. Masur, Interval exchange transformations and measured foliations. Ann. of Math. (2)
115(1982), no. 1, 169–200.
[MasSm] H. Masur and J. Smillie, Hausdorff Dimension of Sets of Nonergodic Measured Foliations.
Ann. of Math., Second Series, 134(1991), 455–543.
[Mi] Y. Minsky, Extremal length estimates and product regions in Teichmüller space. Duke
Math. J. 83(1996), no. 2, 249–286.
[Ra1] K. Rafi, Closed geodesics in the thin part of moduli space. In preparation.
[Ra2] K. Rafi, Thick-thin decomposition of quadratic differentials. preprint, 2005.

35
[Th] W. Thurston. On the geometry and dynamics of diffeomorphisms of surfaces. Bull. Amer.
Math. Soc. (N.S.), 19(1988), 417–431.
[Ve] W. Veech, The Teichmüller geodesic flow. Ann. of Math. (2) 124(1986), no. 3, 441–530.

36