on simple closed geodesics on hyperbolic surfaces

Overview

Mirzakhani proved that the number of simple closed geodesics of length at most $L$ on a complete finite-area hyperbolic surface grows polynomially, not exponentially (Mirzakhani, 2008). If $X$ has type $(g,n)$ and

$$d=6g-6+2n,$$

then each mapping-class-group type of rational multicurve $\gamma$ has an asymptotic

$$s_X(L,\gamma )\sim \frac{c(\gamma )B(X)}{b_{g,n}}{L}^d.$$

The constant $B(X)$ is analytic and metric-dependent: it is the Thurston measure of the unit ball in measured lamination space for the length norm of $X$. The constant $c(\gamma )$ is topological: it is the leading coefficient extracted from Weil-Petersson volume polynomials of moduli spaces obtained by cutting the surface along $\gamma$.

The proof has a distinctive architecture. First, integrate the counting problem over moduli space and use Mirzakhani’s formulas for Weil-Petersson volumes of bordered surfaces. This computes the average density. Second, reinterpret the original count as a value of a rescaled orbit measure on measured lamination space. Mapping-class-group ergodicity forces every subsequential limiting density to be proportional to Thurston measure, and the averaged computation determines the proportionality constant.

Counting Simple Types

Let $S_{g,n}$ be an oriented topological surface of genus $g$ with $n$ punctures or boundary components, and assume

$$2g-2+n>0.$$

Let ${\mathcal{M}}_{g,n}$ denote the moduli space of complete finite-area hyperbolic surfaces of this type, with punctures represented by cusps. For

$$X\in {\mathcal{M}}_{g,n},$$

write $c_X(L)$ for the number of primitive closed geodesics on $X$ of length at most $L$. The classical prime geodesic theorem gives

$$c_X(L)\sim \frac{e^L}{L},$$

with refinements going back to Delsarte, Huber, Selberg, and Margulis (Buser, 1992; Margulis, 1969).

The word primitive is the geodesic analogue of avoiding powers in a group. If a closed geodesic wraps $m$ times around a shorter closed geodesic, it should not be counted as new for the primitive count. This count is exponential because a typical conjugacy class in ${\pi }_1(S)$ is represented by a self-intersecting geodesic, and the number of conjugacy classes with length at most $L$ grows like the volume of a ball in a negatively curved group.

Mirzakhani studies the much thinner set of simple closed geodesics. A closed geodesic is simple if it has no self-intersection. Birman and Series proved that geodesics with bounded self-intersection are sparse in the unit tangent bundle, and they also gave algorithmic structure for simple curves on surfaces (Birman & Series, 1984; Birman & Series, 1985). Sparse, however, does not by itself give a sharp asymptotic count.

The topological type of a simple closed curve is its orbit under the mapping class group

$${\mathrm{Mod}}_{g,n}.$$

Equivalently, for a connected simple closed curve $\gamma$, the type is determined by the homeomorphism type of the cut surface

$$S_{g,n}\setminus \gamma .$$

For example, on a closed genus two surface there are two connected types: nonseparating curves and separating curves cutting the surface into two one-holed tori.

Fix an isotopy class of a simple closed curve, or more generally a rational multicurve, $\gamma$. The counting function for its type is

$$s_X(L,\gamma )=\#\{\alpha \in {\mathrm{Mod}}_{g,n}\cdot \gamma :{\ell }_{\alpha }(X)\le L\}.$$

There are only finitely many connected simple-curve types on $S_{g,n}$, so the total simple geodesic count is a finite sum over types:

$$s_X(L)=\sum _{\gamma }{s}_X(L,\gamma ).$$

Multicurves

A multicurve is a formal sum

$$\gamma =\sum _{i=1}^k{a}_i{\gamma }_i,$$

where the ${\gamma }_i$ are pairwise disjoint, essential, nonperipheral simple closed curves, no two are isotopic, and $a_i>0$. It is integral if $a_i\in \mathbb{N}$ and rational if $a_i\in \mathbb{Q}$. Its length on $X$ is

$${\ell }_{\gamma }(X)=\sum _{i=1}^k{a}_i{\ell }_{{\gamma }_i}(X).$$

Allowing rational multicurves is not cosmetic. Rational multicurves are dense in measured lamination space, and their mapping-class-group orbits are the discrete objects whose asymptotic distribution controls the count.

This is already a strong conceptual reduction. The main problem is no longer to distinguish simple elements inside ${\pi }_1(S)$ one by one. It is to understand how one mapping-class-group orbit of a multicurve sits inside the cone of all measured laminations.

Measured Laminations and Thurston Measure

A geodesic lamination on a hyperbolic surface is a closed subset decomposed as a disjoint union of complete simple geodesics. A measured lamination is a geodesic lamination together with a transverse invariant measure. The transverse measure assigns a measure to every arc crossing the lamination, invariant under homotopies of the arc through transverse arcs.

The space of compactly supported measured laminations on $S_{g,n}$ is denoted

$${\mathcal{ML}}_{g,n}.$$

Although a geodesic representative uses a hyperbolic metric, the resulting space is topological: if two hyperbolic metrics are put on the same marked surface, their universal covers have naturally identified circles at infinity, and this identifies their measured lamination spaces.

The crucial structure is that ${\mathcal{ML}}_{g,n}$ is a piecewise-linear cone of dimension

$$d=6g-6+2n.$$

Train-track charts give local coordinates. In such a chart, a measured lamination carried by a train track is encoded by nonnegative weights on branches satisfying switch equations. Integral weights correspond to integral multicurves. These integral charts give a canonical Lebesgue-class measure, the Thurston measure

$${\mu }_{\mathrm{Th}}.$$

It is invariant under the mapping class group and homogeneous under dilation:

$${\mu }_{\mathrm{Th}}(tU)=t^d{\mu }_{\mathrm{Th}}(U)$$

for measurable $U\subset {\mathcal{ML}}_{g,n}$ and $t>0$ (Thurston, 1979; Fathi, Laudenbach & Poenaru, 1979; Harer & Penner, 1992).

This is the source of the exponent. If a full-rank lattice in ${\mathbb{R}}^d$ is counted in dilates of a bounded region, the main term is proportional to $L^d$. Mirzakhani’s theorem says that a mapping-class-group orbit of a multicurve behaves, after the correct normalization and averaging, like a rational-density subset of the integral points in ${\mathcal{ML}}_{g,n}$.

Every hyperbolic surface $X$ defines a continuous homogeneous length function

$${\ell }_X:{\mathcal{ML}}_{g,n}\to {\mathbb{R}}_{\ge 0},\lambda \mapsto {\ell }_{\lambda }(X),$$

extending the length of weighted multicurves. Define the unit ball

$$B_X=\{\lambda \in {\mathcal{ML}}_{g,n}:{\ell }_{\lambda }(X)\le 1\}$$

and its Thurston volume

$$B(X)={\mu }_{\mathrm{Th}}(B_X).$$

Homogeneity gives

$${\mu }_{\mathrm{Th}}(\{\lambda :{\ell }_{\lambda }(X)\le L\})=L^{d}B(X).$$

The function $B(X)$ is the geometric factor in the final answer. It changes with the hyperbolic metric. Near the boundary of moduli space, when some curve becomes short, the length ball in lamination space changes drastically; this is why Mirzakhani proves that $B(X)$ is proper and still Weil-Petersson integrable (Mirzakhani, 2008).

Main Asymptotic

Mirzakhani’s theorem separates metric information from topological information.

Mirzakhani's Counting Theorem

Let $\gamma$ be a rational multicurve on $S_{g,n}$ and let

$$d=6g-6+2n.$$

For every $X\in {\mathcal{M}}_{g,n}$,

$$\underset{L\to \infty }{\lim }\frac{s_X(L,\gamma )}{L^d}=\frac{c(\gamma )B(X)}{b_{g,n}},$$

where

$$b_{g,n}={\int }_{{\mathcal{M}}_{g,n}}B(X)dX<\infty .$$

The measure $dX$ is the Weil-Petersson volume form, $c(\gamma )\in {\mathbb{Q}}_{>0}$ depends only on the mapping-class-group orbit of $\gamma$, and $B(X)={\mu }_{\mathrm{Th}}(B_X)$ (Mirzakhani, 2008).

For connected simple curves, summing over the finitely many topological types gives

$$s_X(L)\sim n(X)L^d,$$

with

$$n(X)=\frac{B(X)}{b_{g,n}}\sum _{\gamma \text{connected type}}c(\gamma ).$$

Thus the exponent is universal for the topology, while the leading constant has a metric factor $B(X)$ and a finite topological sum.

The formula also gives frequencies of types. For rational multicurves ${\gamma }_1,{\gamma }_2$,

$$\underset{L\to \infty }{\lim }\frac{s_X(L,{\gamma }_1)}{s_X(L,{\gamma }_2)}=\frac{c({\gamma }_1)}{c({\gamma }_2)}.$$

The right side is rational and independent of $X$. This independence is striking: the individual counts depend on the metric through $B(X)$, but the relative frequencies of topological types do not.

Weil-Petersson Coordinates

The averaged part of the proof takes place on moduli spaces of bordered hyperbolic surfaces. Let

$${\mathcal{T}}_{g,n}(L_1,\dots ,L_n)$$

be the Teichmuller space of marked hyperbolic surfaces with $n$ geodesic boundary components of specified lengths $L_1,\dots ,L_n$. The quotient by the mapping class group is

$${\mathcal{M}}_{g,n}(L_1,\dots ,L_n).$$

The convention $L_i=0$ turns the corresponding boundary component into a cusp, so ${\mathcal{M}}_{g,n}$ is the all-cusped case.

Choose a pants decomposition

$$P=\{{\alpha }_1,\dots ,{\alpha }_{3g-3+n}\}.$$

Fenchel-Nielsen coordinates on Teichmuller space consist of length and twist coordinates

$$({\ell }_{{\alpha }_1},\dots ,{\ell }_{{\alpha }_{3g-3+n}},{\tau }_{{\alpha }_1},\dots ,{\tau }_{{\alpha }_{3g-3+n}}).$$

The length coordinate records the length of the geodesic ${\alpha }_i$; the twist coordinate records how the two sides are glued after cutting along ${\alpha }_i$.

The Weil-Petersson symplectic form has a simple expression in these coordinates:

$${\omega }_{\mathrm{WP}}=\sum _{i}d{\ell }_{{\alpha }_i}\wedge d{\tau }_{{\alpha }_i}.$$

This formula, due to Wolpert, is the reason cutting along curves produces tractable volume integrals (Wolpert, 1982). It is also compatible with Goldman’s symplectic viewpoint on surface group representations (Goldman, 1984).

For moduli spaces with geodesic boundary lengths $b=(b_1,\dots ,b_n)$, Mirzakhani proved that the Weil-Petersson volume

$$V_{g,n}(b)={\mathrm{Vol}}_{\mathrm{WP}}({\mathcal{M}}_{g,n}(b_1,\dots ,b_n))$$

is a polynomial in the variables $b_i^2$ (Mirzakhani, 2007). In a companion result, the coefficients are identified with intersection numbers of tautological classes on the Deligne-Mumford compactification (Mirzakhani, 2007b).

These volume polynomials are not auxiliary decoration. They are the engine that computes the constants $c(\gamma )$.

Averaging Over Moduli Space

Define the averaged counting function

$$P(L,\gamma )={\int }_{{\mathcal{M}}_{g,n}}{s}_X(L,\gamma )dX.$$

The integral is with respect to the Weil-Petersson volume form. Pointwise counting is hard because a fixed surface has a complicated length spectrum. Averaging over moduli space makes the mapping-class-group orbit unfold.

First consider a connected simple curve $\gamma$. Cutting $S_{g,n}$ along $\gamma$ gives a surface $S_{g,n}(\gamma )$ with two new boundary components of equal length

$$x={\ell }_{\gamma }(X).$$

The twist parameter along $\gamma$ ranges over an interval of length $x$. In Fenchel-Nielsen coordinates, Wolpert’s formula says that the Weil-Petersson volume element contains

$$dx\wedge d\tau ,$$

so the twist contributes a factor comparable to $xdx$ after passing to the quotient. The remaining coordinates describe the moduli space of the cut surface with boundary length $x$ on both new components.

For a multicurve

$$\gamma =\sum _{i=1}^k{a}_i{\gamma }_i,$$

cutting along all components produces $2k$ new boundary components with paired lengths

$$x_1,x_1,\dots ,x_k,x_k.$$

The length constraint becomes

$$\sum _{i=1}^k{a}_i{x}_i\le L.$$

The unfolded integral has the schematic form

$$P(L,\gamma )=\text{(finite symmetry factor)}{\int }_{\sum a_i{x}_i\le L}V(S_{g,n}(\gamma );x_1,x_1,\dots ,x_k,x_k)\prod _{i=1}^k{x}_{i}d{x}_i.$$

Here $V(S_{g,n}(\gamma );\cdots )$ is the product of Weil-Petersson volume polynomials for the connected components of the cut surface. The exact formula includes automorphism factors for $\gamma$ and factors accounting for components interchanged by symmetries; these affect the rational constant but not the mechanism (Mirzakhani, 2008).

Because each volume factor is polynomial in the squared boundary lengths, the integral over the simplex

$$\sum a_i{x}_i\le L$$

is a polynomial in $L$. Mirzakhani proves

$$P(L,\gamma )$$

is a polynomial of degree $d=6g-6+2n$, and defines

$$c(\gamma )=\underset{L\to \infty }{\lim }\frac{P(L,\gamma )}{L^d}.$$

For rational multicurves,

$$c(\gamma )\in {\mathbb{Q}}_{>0}.$$

The rationality is a topological consequence of the integral structure on multicurves and the leading coefficients of the relevant Weil-Petersson volume polynomials (Mirzakhani, 2008; Mirzakhani, 2007b).

This averaged computation determines the global density but does not yet prove that each individual surface $X$ sees the same density profile. That step is dynamical.

Orbit Measures in Lamination Space

Let ${\mu }_{\gamma }$ be the counting measure on the distinct orbit

$${\mathrm{Mod}}_{g,n}\cdot \gamma \subset {\mathcal{ML}}_{g,n}.$$

For $T>0$, define the rescaled measure

$${\mu }_{T,\gamma }(U)=\frac{\#(T\cdot U\cap {\mathrm{Mod}}_{g,n}\cdot \gamma )}{T^d}$$

for suitable measurable sets $U\subset {\mathcal{ML}}_{g,n}$, where $T\cdot U=\{T\lambda :\lambda \in U\}$. Equivalently, ${\mu }_{T,\gamma }$ is obtained by dilating the orbit down by a factor $T$ and normalizing by $T^d$.

The relation to counting is immediate:

$${\mu }_{T,\gamma }(B_X)=\frac{\#(T{B}_X\cap {\mathrm{Mod}}_{g,n}\cdot \gamma )}{T^d}=\frac{s_X(T,\gamma )}{T^d}.$$

Thus the desired asymptotic is the evaluation of ${\mu }_{T,\gamma }$ on the length unit ball $B_X$.

Mirzakhani proves the weak convergence

$${\mu }_{T,\gamma }⟶\frac{c(\gamma )}{b_{g,n}}{\mu }_{\mathrm{Th}}$$

as $T\to \infty$ (Mirzakhani, 2008). Once this is known, the counting theorem follows by evaluating both sides on $B_X$:

$$\frac{s_X(T,\gamma )}{T^d}={\mu }_{T,\gamma }(B_X)⟶\frac{c(\gamma )}{b_{g,n}}{\mu }_{\mathrm{Th}}(B_X)=\frac{c(\gamma )B(X)}{b_{g,n}}.$$

The boundary of $B_X$ has zero Thurston measure, so weak convergence applies to this set.

The proof of measure convergence has three parts.

First, the measures ${\mu }_{T,\gamma }$ are locally uniformly bounded. Given a compact set $K\subset {\mathcal{ML}}_{g,n}$, choose a hyperbolic surface $X_0$ and $R>0$ such that

$$K\subset R{B}_{X_0}.$$

Counting bounds for multicurves imply

$$\underset{T}{\sup }{\mu }_{T,\gamma }(K)<\infty .$$

Therefore every sequence $T_j\to \infty$ has a weakly convergent subsequence.

Second, any subsequential limit is absolutely continuous with respect to the Thurston measure. In train-track coordinates, Mirzakhani compares orbit counts in convex chart domains with counts of all integral multicurves in the same domains. The latter are controlled by the integral lattice defining ${\mu }_{\mathrm{Th}}$. This gives, for suitable chart domains $U$,

$$\nu (U)\le {\mu }_{\mathrm{Th}}(U)$$

up to the normalization used in the argument, and hence sets of Thurston measure zero have zero $\nu$-measure.

Third, each subsequential limit is mapping-class-group invariant, because the original orbit counting measures are invariant. Masur’s theorem says that the mapping class group acts ergodically on ${\mathcal{ML}}_{g,n}$ with respect to the Thurston measure class (Masur, 1985). Absolute continuity plus ergodicity implies that the limit must be a scalar multiple of Thurston measure:

$$\nu =k{\mu }_{\mathrm{Th}}.$$

The only remaining question is the scalar $k$.

Determination of the Density

Let

$${\mu }_{T_j,\gamma }\to k{\mu }_{\mathrm{Th}}$$

along a subsequence. Evaluating on $B_X$ gives

$$\frac{s_X(T_j,\gamma )}{T_j^d}\to kB(X)$$

for each $X$ at which the boundary issue is harmless; Mirzakhani handles the required uniformity using her multicurve bounds and the integrability of $B(X)$.

Integrate this relation over moduli space:

$${\int }_{{\mathcal{M}}_{g,n}}\frac{s_X(T_j,\gamma )}{T_j^d}dX\to k{\int }_{{\mathcal{M}}_{g,n}}B(X)dX.$$

The left side is

$$\frac{P(T_j,\gamma )}{T_j^d},$$

and the averaged computation gives

$$\frac{P(T_j,\gamma )}{T_j^d}\to c(\gamma ).$$

Therefore

$$k{b}_{g,n}=c(\gamma ),$$

or

$$k=\frac{c(\gamma )}{b_{g,n}}.$$

The scalar is independent of the subsequence, so the full family ${\mu }_{T,\gamma }$ converges.

This is the main logical loop of the paper. The moduli-space average supplies a numerical density, while ergodicity proves that there is no room for a different pointwise density profile.

Frequencies and Examples

Because the factor $B(X)/b_{g,n}$ is common to every rational multicurve type, it cancels in ratios:

$$\frac{s_X(L,{\gamma }_1)}{s_X(L,{\gamma }_2)}\to \frac{c({\gamma }_1)}{c({\gamma }_2)}.$$

The ratio is independent of the hyperbolic metric. Mirzakhani notes that the same frequency result holds for compact surfaces of variable negative curvature, because the measure-convergence statement in $\mathcal{ML}$ is topological and the length function remains continuous and homogeneous (Mirzakhani, 2008).

On a closed genus two surface, let ${\alpha }_1$ be a nonseparating simple closed curve and let ${\alpha }_2$ be a separating simple closed curve. Mirzakhani computes the relevant constants from the bordered volume polynomials:

$$\frac{c({\alpha }_1)}{c({\alpha }_2)}=6.$$

Hence

$$\frac{s_X(L,{\alpha }_1)}{s_X(L,{\alpha }_2)}\to 6.$$

Equivalently, among long connected simple closed geodesics on a genus two hyperbolic surface, the limiting probability of being separating is

$$\frac{1}{7}.$$

Earlier work had obtained special cases and bounds. McShane and Rivin treated the once-punctured torus using a norm on homology (McShane & Rivin, 1995). Rivin proved polynomial upper and lower bounds for simple curves on general surfaces (Rivin, 2001). Rees obtained related bounds for pants decompositions and measured foliations (Rees, 1981). Mirzakhani’s result identifies the exact exponent, the exact metric factor, and the topological frequency constants.

Scope and Limitations

The theorem is asymptotic and qualitative at the leading-order level. The Annals paper does not give an effective error term in $L$, and the proof uses compactness, dominated convergence, and ergodicity in a way that is not designed for explicit rates.

The constants are explicit in principle but not elementary in general. Computing $c(\gamma )$ requires the topology of the cut surface and the appropriate Weil-Petersson volume polynomial. Mirzakhani’s recursion and intersection-theoretic formulas make these constants computable, but the expressions grow quickly with genus and number of components (Mirzakhani, 2007a; Mirzakhani, 2007b).

The proof is also tailored to orientable finite-type surfaces. The measured lamination space, Thurston measure, mapping-class-group action, and Weil-Petersson geometry all have analogues in related settings, but the exact statement depends on this package of structures.

Finally, the result counts topological types through mapping-class-group orbits. It does not distinguish finer arithmetic or homological constraints inside a type unless those constraints can be encoded by a different invariant measure or by a refined counting problem.

Transferable Mechanisms

The first mechanism is the replacement of a direct spectral count by a count in a homogeneous geometric cone. Simple geodesics are difficult to isolate inside the full closed-geodesic spectrum, but integral multicurves are natural lattice points in ${\mathcal{ML}}_{g,n}$, and the length condition is a homogeneous ball (Thurston, 1979; Fathi, Laudenbach & Poenaru, 1979; Mirzakhani, 2008).

The second mechanism is averaging before proving pointwise convergence. The averaged count $P(L,\gamma )$ unfolds over moduli space and becomes a Weil-Petersson volume integral. This converts an orbit-counting problem into a polynomial-volume computation (Wolpert, 1982; Mirzakhani, 2007a; Mirzakhani, 2008).

The third mechanism is the separation of metric and topological constants. The metric dependence is entirely in $B(X)$, while the topological type is entirely in $c(\gamma )$. This separation is what makes frequency ratios independent of the hyperbolic metric (Mirzakhani, 2008).

The fourth mechanism is measure classification at the correct level of generality. Mirzakhani does not need a classification of every orbit closure in ${\mathcal{ML}}_{g,n}$; she needs absolute continuity of subsequential limits and Masur’s ergodicity theorem to force those limits to be scalar multiples of Thurston measure (Masur, 1985; Mirzakhani, 2008).

The fifth mechanism is the use of auxiliary moduli spaces created by cutting. Cutting along $\gamma$ introduces boundary lengths and twist parameters, and Wolpert’s formula turns those parameters into the factors that make the integral computable. This is a robust pattern: introduce boundary data so that a constrained object becomes an unconstrained point in a larger moduli space (Wolpert, 1982; Mirzakhani, 2007a).

The sixth mechanism is the extraction of discrete frequencies from continuous volume polynomials. The leading coefficient of a continuous Weil-Petersson volume integral becomes the density of a discrete mapping-class-group orbit. This bridge between continuous symplectic volume and discrete topological counting is one of the main reasons the paper is useful beyond the original counting problem (Mirzakhani, 2007b; Mirzakhani, 2008).

See Also

• on orbit closures in moduli space — both posts use invariant-measure rigidity to turn subsequential limits into structured limiting measures. Here the acting group is the mapping class group on measured laminations; in Eskin-Mirzakhani-Mohammadi it is $SL(2,\mathbb{R})$ and its subgroups on strata of Abelian differentials.
• on almost geodesic surfaces in hyperbolic three-manifolds — both arguments use hyperbolic geometry and curve decompositions, but the organizing dynamics differ: Mirzakhani uses mapping-class-group ergodicity on $\mathcal{ML}$, while Kahn-Markovic use exponential mixing of the frame flow.
• on min-max theory and the Willmore conjecture — both papers convert a geometric problem about surfaces into a structured variational or measure-theoretic framework, then identify a global constant through compactness and deformation arguments.
• on distinct distances and polynomial partitioning — the analogy is methodological: a hard discrete count becomes tractable after embedding it into a geometric ambient space with the correct scaling dimension.

References

🐻  Birman, J.S. & Series, C. 1984. An algorithm for simple curves on surfaces. Journal of the London Mathematical Society 29(2), 331–342.

🐻  Birman, J.S. & Series, C. 1985. Geodesics with bounded intersection number on surfaces are sparsely distributed. Topology 24(2), 217–225.

🐻  Buser, P. 1992. Geometry and Spectra of Compact Riemann Surfaces, Birkhauser Boston,p.

🐻  Fathi, A., Laudenbach, F. & Poenaru, V. 1979. Travaux de Thurston sur les surfaces, Societe Mathematique de France,p.

🐻  Goldman, W.M. 1984. The symplectic nature of fundamental groups of surfaces. Advances in Mathematics 54(2), 200–225.

🐻  Harer, J.L. & Penner, R.C. 1992. Combinatorics of Train Tracks, Princeton University Press,p.

🐻  Margulis, G.A. 1969. Applications of ergodic theory to the investigation of manifolds of negative curvature. Functional Analysis and Its Applications 4, 335–336.

🐻  Masur, H. 1985. Ergodic actions of the mapping class group. Proceedings of the American Mathematical Society 94(3), 455–459.

🐻  McShane, G. & Rivin, I. 1995. Simple curves on hyperbolic tori. Comptes Rendus de l’Academie des Sciences. Serie I. Mathematique 320, 1523–1528.

🐻  Mirzakhani, M. 2007a. Simple geodesics and Weil-Petersson volumes of moduli spaces of bordered Riemann surfaces. Inventiones Mathematicae 167(1), 179–222.

🐻  Mirzakhani, M. 2007b. Weil-Petersson volumes and intersection theory on the moduli space of curves. Journal of the American Mathematical Society 20(1), 1–23.

🐻  Mirzakhani, M. 2008. Growth of the number of simple closed geodesics on hyperbolic surfaces. Annals of Mathematics 168(1), 97–125.

🐻  Rees, M. 1981. An alternative approach to the ergodic theory of measured foliations on surfaces. Ergodic Theory and Dynamical Systems 1(4), 461–488.

🐻  Rivin, I. 2001. Simple curves on surfaces. Geometriae Dedicata 87(1), 345–360.

🐻  Thurston, W.P. 1979. The geometry and topology of three-manifolds.

🐻  Wolpert, S. 1982. The Fenchel-Nielsen deformation. Annals of Mathematics 115(3), 501–528.