Flat on Riemann surfaces and Teichm¨uller dynamics in moduli spaces

Anton Zorich

Obergurgl, December 15, 2008

From Billiards to Flat Surfaces 2 Billiardinapolygon ...... 3 Closedtrajectories...... 4 Challenge...... 5 Motivationforstudyingbilliards ...... 6 Gasoftwomolecules...... 7 Unfoldingbilliardtrajectories ...... 8 Flatsurfaces...... 9 Rationalpolygons...... 10 Billiardinarectangle ...... 11 Unfoldingrationalbilliards ...... 13 Flatpretzel...... 14 Surfaceswhich are more flat thanthe others...... 15

Very flat surfaces 16 Very flat surfaces: construction from a polygon...... 17 Propertiesofveryflatsurfaces...... 19 Conicalsingularity ...... 20 Familiesofflatsurfaces...... 21 Familyofflattori...... 22

Holomorphic 1-forms versus very flat surfaces 23 Fromflattocomplexstructure ...... 24 Fromcomplextoflatstructure ...... 25 Concise geometro-analyticdictionary ...... 26 Flat surfaces and quadratic differentials...... 27 Volumeelement...... 28 Groupaction...... 29 Masur—VeechTheorem ...... 30 Hopeforamagicwand ...... 31 RelationtotheTeichm¨ullermetric ...... 32

Generic geodesics on very flat surfaces 33 Asymptoticcycleforatorus...... 34

1 Asymptoticcycle for generalsurfaces ...... 35 Asymptotic flag: empirical description ...... 36 Lyapunovexponents ...... 41

Saddle connections and closed geodesics 42 Countingofclosedgeodesics ...... 43 Exactquadraticasymptotics...... 44 Phenomenon of multiple saddle connections...... 45 Homologoussaddleconnections...... 46

Billiards in rectangular polygons 49 L-shapedbilliard...... 50 Rectangularpolygons ...... 51 Billiards versus quadratic differentials ...... 52 Numberofgeneralizeddiagonals...... 53 Naiveintuitiondoesnothelp...... 54 Billiardplayers ...... 55

2 From Billiards to Flat Surfaces 2/55

Billiard in a polygon

Moon Duchin plays billiard on an L-shaped table

3/55

Closed trajectories

It is easy to find a closed billiard trajectory in an acute triangle.

Exercise. Prove that the broken joining the bases of hights of an acute triangle is a billiard trajectory (it is called Fagnano trajectory). Show that it realizes an inscribed triangle of minimal perimeter. 4/55

3 Challenge

It is difficult to believe, but a similar problem for an obtuse triangle is open.

Open problem. Is there at least one closed trajectory for (almost) any obtuse triangle?

It seems like the answer is affirmative (see an extensive computer search performed by R. Schwartz and P. Hooper: www.math.brown.edu/ res/Billiards/index.html) ∼ Then, one can ask further questions: Estimate the number N(L) of periodic trajectories of length bounded by L 1 when L + . • ≫ → ∞ Is the billiard flow ergodic for almost any triangle? • 5/55

Motivation to study billiards: gas of two molecules in a one-dimensional cham- ber

Consider two elastic beads (“molecules”) sliding along a rod. They are bounded from two sides by solid walls. All collisions are ideal — without loss of energy. x2 a m1 m2

x1 x2 x 0 a | a x1

Neglecting the sizes of the beads we can describe the configuration space of our system using coordinates 0 < x1 x2 a of the beads, where a is the distance between the walls. This gives a right isosceles triangle.≤ ≤ 6/55

4 Gas of two molecules

Rescaling the coordinates by square roots of masses x˜1 = √m1x1 (x˜2 = √m2x2 we get a new right triangle ∆ as a configuration space.

x˜2 = √m2 x2

m1 m2

x1 x2 x 0 x˜1 = √m1 x1

Lemma In coordinates (˜x1, x˜2) trajectories of the system of two beads on a rod correspond to billiard trajectories in the triangle ∆.

7/55

Unfolding billiard trajectories

Identifying the boundary of two triangles we get a flat sphere. A billiard trajectory unfolds to a geodesic on this flat sphere. 8/55

5 Flat surfaces

The surface of the cube represents a flat sphere with eight conical singularities. The metric does not have singularities on the edges. After transport around a conical singularity a vector comes back pointing to a direction different from the initial one, so this flat metric has nontrivial holonomy. 9/55

Rational polygons

A polygon Π is called rational if all the of Π are rational multiples pi π of π. Properties of qi billiards in rational polygons are known much better. Consider a model case of a rectangular billiard.

As before instead of reflecting the trajectory we can reflect the billiard table. The trajectory unfolds to a straight line. Folding back the copies of the billiard table we project this line to the actual trajectory. 10 / 55

6 Billiard in a rectangle

Fix a (generic) trajectory. At any moment the ball moves in one of four directions. They correspond to four copies of the billiard table; other copies can be obtained from these four by a parallel translation:

D A A

C B B

A D A D

B C C

D D A 11 / 55

Billiard in a rectangle versus flow on a torus

Identifying the equivalent patterns by a parallel translation we obtain a torus; the billiard trajectory unfolds to a “straight line” on the corresponding torus.

12 / 55

7 Unfolding rational billiards

We can apply an analogous procedure to any rational billiard.

Consider, for example, a triangle with angles π/8, 3π/8,π/2. It is easy to check that a generic trajectory in such billiard table has 16 directions (instead of 4 for a rectangle). Using 16 copies of the triangle we unfold the billiard into a regular octagon with opposite sides identified by parallel translations. 13 / 55

Flat surface of genus two

Identifying the pair of horizontal sides and then the pair of vertical sides of a regular octagon we get a torus with a single square hole. Identifying two opposite sides of the hole we get a torus with two distinct holes. Identifying the holes we get a surface of genus two. 14 / 55

8 Surfaces which are more flat than the others

Note that the flat metric on the resulting surface has trivial holonomy, since the identifications of the sides were performed by parallel translations. As before, a billiard trajectory is unfolded to a geodesic on the surface. But now geodesics resemble geodesics on the torus: they do not have self-!

We abandon rational billiards for a while and pass to a more systematic study of “very flat” surfaces. 15 / 55

Very flat surfaces 16 / 55

Very flat surfaces: construction from a polygon

Consider a broken line constructed from vectors ~v1,...,~vk.

~v3 ~v2 ~v4 ~v1

~v4 ~v1 ~v2 ~v3

and another one constructed from the same vectors taken in another order. If we are lucky enough the two broken lines do not intersect and form a polygon. 17 / 55

9 Very flat surfaces: construction from a polygon

~v3 ~v2 ~v ~v1 4

~v4 ~v1 ~v2 ~v3

Identifying the corresponding pairs of sides by parallel translations we get a closed surface endowed with a flat metric. 18 / 55

Properties of very flat surfaces

The flat metric is nonsingular outside of a finite number of conical singularities (inherited from the • vertices of the polygon).

The flat metric has trivial holonomy, i.e. parallel transport along any closed path brings a tangent • vector to itself.

In particular, all cone angles are integer multiples of 2π. • By convention, the choice of the vertical direction (“direction to the North”) will be considered as a • part of the “very flat structure”. For example, a surface obtained from a rotated polygon is considered as a different very flat surface.

A conical singularity with the cone 2π N has N outgoing directions to the North. • ·

19 / 55

10 Example: conical singularity with cone angle 6π

Locally a neighborhood of a conical looks like a “monkey saddle”.

A neighborhood of a conical point with a cone angle 6π can be glued from six metric half discs. At this conical point we have 3 distinct directions to the North. 20 / 55

Families of flat surfaces

The polygon in our construction depends continuously on the vectors ~vj. This means that the combinatorial geometry of the resulting flat surface (its genus g, the number m and types 2π(d1 + 1),..., 2π(dm + 1) of the resulting conical singularities) does not change under small deformations of the vectors ~vj. This allows to consider a flat surface as an element of a family of flat surfaces sharing common combinatorial geometry.

As an example of such family one can consider a family of flat tori of area one, which can be identified with the space of lattices of area one:

SL(2, R) / H2/ SO(2, R)\ SL(2, Z) = SL(2, Z)

21 / 55

11 Family of flat tori

neighborhood of a cusp = subset of tori having short closed geodesic

The corresponding “modular surface” is not compact: flat tori representing points, which are close to the cusp, are almost degenerate: they have a very short closed geodesic. 22 / 55

Holomorphic 1-forms and quadratic differentials versus very flat sur- faces 23 / 55

Holomorphic 1-form associated to a flat structure

Consider the natural coordinate z in the complex , where lives the polygon. In this coordinate the parallel translations which we use to identify the sides of the polygon are represented as z′ = z + const. Since this correspondence is holomorphic, our flat surface S with punctured conical points inherits the complex structure. This complex structure extends to the punctured points.

Consider now a holomorphic 1-form dz in the complex plane. The coordinate z is not globally defined on the surface S. However, since the changes of local coordinates are defined as z′ = z + const, we see that dz = dz′. Thus, the holomorphic 1-form dz on C defines a holomorphic 1-form ω on S which in local coordinates has the form ω = dz. The form ω has zeroes exactly at those points of S where the flat structure has conical singularities. 24 / 55

12 Flat structure defined by a holomorphic 1-form

Reciprocally a pair (Riemann surface, holomorphic 1-form) uniquely defines a flat structure: • z = ω.

In a neighborhoodR of a zero a holomorphic 1-form can be represented as wd dw, where d is a • degree of the zero. The form ω has a zero of degree d at a conical point with cone angle 2π(d + 1). Moreover, d1 + + d =2g 2. ··· m − g The moduli space g of pairs (complex structure, holomorphic 1-form) is a C -vector bundle • H over the moduli space of complex structures. Mg

The space g is naturally stratified by the strata (d1,...,dm) enumerated by unordered • H H partitions d1 + + d =2g 2. ··· m −

Any holomorphic 1-form corresponding to a fixed stratum (d1,...,dm) has exactly m zeroes • H of degrees d1,...,dm.

25 / 55

Concise geometro-analytic dictionary

flat structure (including a choice complex structure and a choice of the vertical direction) of a holomorphic 1-form ω

conical point zero of degree d with a cone angle 2π(d + 1) of the holomorphic 1-form ω (in local coordinates ω = wd dw)

P side ~v of a polygon relative period j+1 ω = dz j Pj ~vj of the 1-form ω R R

family of flat surfaces sharing stratum (d1,...,d ) in the H m the same cone angles moduli space of holomorphic 1-forms 2π(d1 + 1),..., 2π(dm + 1)

local coordinates in the family: local coordinates in (d1,...,d ) : H m vectors ~vi relative periods of ω in 1 defining the polygon H (S, P1,...,P ; C) { m} 26 / 55

13 Flat surfaces and quadratic differentials

Identifying pairs of sides of this polygon by isometries we obtain a surface of genus g =1. Now the flat metric has holonomy group Z/2Z. The cone angles are multiples of π. Flat surfaces of this type correspond to quadratic differentials. For example, the quadratic differential representing the surface from the picture belongs to the stratum (2, 1, 1). Q − − 27 / 55

Volume element

1 Note that the vector space H (S, P1,...,Pm ; C) contains a natural integer 1 { } H (S, P1,...,P ; Z √ 1 Z). Consider a linear volume element dν normalized in such a way { m} ⊕ − that the volume of the fundamental domain in this lattice equals one. Consider now a real hypersurface 1(d1,...,d ) (d1,...,d ) defined by the equation area(S)=1. The volume element dν H m ⊂ H m can be naturally restricted to the hypersurface defining the volume element dν1 on 1(d1,...,d ). H m

Theorem (H. Masur; W. A. Veech) The total volume Vol( 1(d1,...,d )) of every stratum is finite. H m The values of these volumes were computed by A. Eskin and A. Okounkov. 28 / 55

14 Group action

The subgroup SL(2, R) of area preserving linear transformations acts on the “unit hyperboloid” et 0 1(d1,...,d ). The diagonal subgroup SL(2, R) induces a natural flow on the H m 0 e−t ⊂ stratum, which is called the Teichm¨uller geodesic flow .

et 0 Key Theorem (H. Masur; W. A. Veech) The action of the groups SL(2, R) and 0 e−t   preserves the measure dν1. Both actions are ergodic with respect to this measure on each connected component of every stratum 1(d1,...,d ). H m 29 / 55

Masur—Veech Theorem

Theorem of Masur and Veech claims that taking an arbitrary octagon as below we can contract it horizontally and expand vertically by the same factor et to get arbitrary close to, say, regular octagon. There is no paradox since we are allowed to cut-and-paste!

= −→

The first modification of the polygon changes the flat structure while the second one just changes the way in which we unwrap the flat surface 30 / 55

15 Hope for a magic wand

Suppose that we need some information about geometry or dynamics of an individual flat surface S. Consider S as a point in the space (d1,...,dm) of flat surfaces. Let (S) be the closure of the + H K GL (2, R)-orbit of S in (d1,...,d ). Knowledge about the structure of (S) gives a H m K comprehensive information about geometry and dynamics on the initial flat surface S. Delicate numerical characteristics of S can be expressed as averages of simpler characteristics of (S). K + Optimistic hope: The closure (S) of the GL (2, R)-orbit of any flat surface S is a nice K complex-analytic variety. (Proved in genus g =2 by C. McMullen; under additional assumptions proved by M. Moller.)¨

+ Theorem (M. Kontsevich) Suppose that the closure (S) of a GL (2, R)-orbit of some flat surface K S is a complex-analytic subvariety. Then in cohomological coordinates it is represented by an affine subspace.

31 / 55

Relation to the Teichm ¨uller metric

Teichm ¨uller Theorem. For any two complex structures C1,C2 on the same closed topological surface one can find a unique “very flat metric” (= holomorphic quadratic differential) on C1 and a unique “very flat metric” (= holomorphic quadratic differential) on C2 such that the corresponding flat surfaces et 0 S1,S2 are obtained one from the other by contraction-expansion: S2 = S1. The 0 e−t   corresponding map realizes the extremal quasiconformal map and the value t /2 represents the | | distance between the two complex structures in the Teichm¨uller metric.

Projection of any orbit of GL(2, R) to the Teichm¨uller space (called a Teichm¨uller disc) is a hyperbolic plane H2 = GL(2, R)/C∗. Any such projection is an isometric embedding with respect to the Teichm¨uller metric. Thus, Teichm¨uller discs can be considered as complex geodesics for the Teichm¨uller metric. 32 / 55

16 Generic geodesics on very flat surfaces 33 / 55

Asymptotic cycle for a torus

Consider an “irrational” geodesic on a torus. Choose a short transversal segment X. Each time when the geodesic crosses X we join the crossing point with the point x0 along X obtaining a closed loop. Consecutive return points x1, x2,... define a sequence of cycles c1,c2,... .

cn 2 The asymptotic cycle is defined as lim = c H1(T ; R) n→∞ n ∈

The asymptotic cycle is the same for all points of T2. 34 / 55

Asymptotic cycle for general surfaces

Theorem (S. Kerckhoff, H. Masur, J. Smillie) For any flat surface directional flow in almost any direction is ergodic with respect to the natural measure.

This implies that for almost any direction the asymptotic cycle exists and is the same for almost all points of the surface. 35 / 55

17 Asymptotic flag: empirical description

S

Direction of the asymptotic cycle cN

x2g To study a deviation of cycles x5 cN from the asymptotic cycle consider their projections to an orthogonal hyperscreen x4 2g H1(S; R) R ≃ x3 x1 x2

36 / 55

Asymptotic flag: empirical description

S

Direction of the asymptotic cycle cN

x2g The projections accumulate x5 along a straight line inside the hyperscreen

x4 2g H1(S; R) R ≃ x3 x1 x2

37 / 55

18 Asymptotic flag: empirical description

S

Direction of the asymptotic cycle cN

x2g

x5

Asymptotic plane L2 x4 2g H1(S; R) R ≃ x3 x1 x2

38 / 55

Asymptotic flag: empirical description

S

ν2 cN Direction of the k k asymptotic cycle cN c ν3 k N k x2g

x5

Asymptotic plane L2 x4 2g H1(S; R) R ≃ x3 x1 x2

39 / 55

19 Asymptotic flag

Theorem For almost any surface S in any stratum 1(d1,...,d ) there exists a flag of subspaces H m L1 L2 L H1(S; R) such that for any j =1,...,g 1 one has ⊂ ⊂···⊂ g ⊂ −

log dist(cN , Lj) lim sup = νj+1 N→∞ log N and dist(c , L ) const , N g ≤ where the constant depends only on S and on the choice of the Euclidean structure in the homology space.

The numbers 2 > 1+ ν2 > > 1+ ν are the top g Lyapunov exponents of the Teichm¨uller ··· g geodesic flow on the corresponding connected component of the stratum (d1,...,d ). H m Remark The fact that the inequalities are strict is not trivial at all! It was proved partly by G. Forni and in full generality by A. Avila and M. Viana. 40 / 55

Lyapunov exponents

1 We have a natural vector bundle with a fiber H (S, R) over a point S of a stratum (d1,...,d ). H m This vector bundle is endowed with a flat Gauss—Manin connection. Let us carry a fiber along a et 0 trajectory of the diagonal group . When at some time t the trajectory passes near the 0 e−t k   starting point we can evaluate the eigenvalues λ1(t ) λ2 (t ) of the monodromy matrix, take k ≥···≥ g k their logarithms and normalize by the length tk of our piece of trajectory.

log λj(tk) By Oseledets theorem the limits νj = limk→+∞ | | exist and are shared for almost all tk starting points. These logarithms of eigenvalues of a “mean monodromy matrix along the flow” are called Lyapunov exponents. In our case they are symmetric since the monodromy matrix is symplectic: ν = ν2 +1 for j =1,...,g. j − g−j 41 / 55

20 Saddle connections and closed geodesics 42 / 55

Counting of closed geodesics

A saddle connection is a geodesic segment joining a pair of conical singularities or a conical singularity to itself without any singularities in its interior. Similar to the torus case regular closed geodesics on a flat surface always appear in families; any such family fills a maximal cylinder bounded on each side by a closed saddle connection or by a chain of parallel saddle connections.

Let Nsc(S, L) be the number of saddle connections of length at most L on a flat surface S. Let Ncg(S, L) be the number of maximal cylinders filled with closed regular geodesics of length at most L on S. 43 / 55

Exact quadratic asymptotics

Theorem (A. Eskin and H. Masur) For almost all flat surfaces S in any stratum (d1,...,d ) the H m counting functions Nsc(S, L) and Ncg(S, L) have exact quadratic asymptotics

N (S, L) c (S) πL2 N (S, L) c (S) πL2 sc ∼ sc · cg ∼ cg ·

where the constants csc(S) and ccg(S) depend only on the connected component of the stratum.

Consider some saddle connection γ1 = [P1P2] with an endpoint at P1. Memorize its direction, say, let it be the North-West direction. Let us launch a geodesic from the same starting point P1 in one of the remaining k 1 North-West directions. Let us study how big is the chance to hit P2 ones again, and how big is the− chance to hit it after passing the same distance as before. 44 / 55

21 Phenomenon of multiple saddle connections

Theorem (A. Eskin, H. Masur, A. Zorich) For almost any flat surface S in any stratum and for any pair P1,P2 of conical singularities on S the function N2(S, L) counting the number of pairs of parallel saddle connections of the same length joining P1 to P2 also has exact quadratic asymptotics

N2(S, L) lim = c2 > 0. L→∞ πL2

For almost all flat surfaces S in any stratum one cannot find neither a single pair of parallel saddle connections on S of different length, nor a single pair of parallel saddle connections joining different pairs of singularities.

45 / 55

Homologous saddle connections

S3 S2

′ S3 ′ S2

S3 γ3 S2

γ2 γ1

′ S1 S1

Saddle connections γ1,γ2,γ3 are homologous. Hence, under any small deformation of the flat metric this configuration persists and deformed saddle connections share the same length and direction. Hence such configuration is present on almost any flat surface in the embodying stratum. 46 / 55

22 Siegel–Veech constants

Typical degenerations of flat surfaces (analogs of “typical stable curves”) correspond to configurations of homologous saddle connections. C Theorem (A. Eskin, H. Masur, A. Z.) A constant c in the quadratic asymptotics N(S, L) c πL2 for the number of configurations of a ∼ · combinatorial type of homologous saddle connections of length at most L on a flat surface S (called the Siegel–Veech constantC ) can be evaluated by the following formula:

1 Vol(“ε-neighborhood of the cusp ”) c( ) = lim comp C = C ε→0 πε2 Vol (β) H1 k ′ j=1 Vol 1(βk) =(explicit combinatorial factor) H · Vol comp(β) Q H1 Remark. Sums ν1 + + νg of the Lyapunov exponents can be evaluated in terms of Siegel–Veech constants. ··· 47 / 55

More artistic picture of a decomposition

48 / 55

23 Applications: billiards in rectangular polygons. (With J. Athreya and A. Eskin) 49 / 55

L-shaped billiard

Moon Duchin playing billiard on a L-shaped table.

50 / 55

Rectangular polygons

P4P 3

P7P 6

P5

P1 P2

By a rectangular polygon Π we call a topological disc endowed with a flat metric, such that the boundary ∂Π is presented by a finite broken line of geodesic segments and such that the angle between any two consecutive sides equals kπ/2, where k N. ∈ 51 / 55

24 1 Billiards in rectangular polygons versus quadratic differentials on CP

We want to count trajectories emitted from a corner of such billiard and getting to some other corner. We also want to count closed billiard trajectories. In both cases we count trajectories of length bounded by L 1 and study the asymptotics as L tends to infinity. ≫ We have already seen that a topological sphere obtained by gluing two copies of the billiard table by the boundary is endowed with a flat metric. In the case of a “rectangular polygon” the flat metric has holonomy in Z/(2Z), or, in other words, it corresponds to a meromorphic quadratic differential with 1 simple poles on CP .

We have seen that geodesics on this flat sphere project to billiard trajectories. Thus, to count billiard trajectories we may count geodesics on flat spheres! 52 / 55

Number of generalized diagonals

Pj Pj

Pi Pi

We prove that for a generic rectangular polygon with angles π/2 and 3π/2 the number of trajectories joining any two fixed corners with right angles is “approximately” the same as for a rectangle:

1 (bound for the length)2 2π · area of the table and does not depend on the shape of the polygon. 53 / 55

25 Naive intuition does not help... P 4 P5

P 0 P1

P3 P2

However, say, for a typical L-shaped polygon the number of trajectories joining the corner with the angle 3π/2 to some other corner is “approximately”

2 (bound for the length)2 π · area of the table which is 4 times (and not 3) times bigger than the number of trajectories joining a fixed pair of right corners... 54 / 55

Billiard in a rectangular polygon, artistic view

Varvara Stepanova. Billiard players. Thyssen museum, Madrid

55 / 55

26