Loading presentation...

Present Remotely

Send the link below via email or IM


Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.


Interval Exchange Maps, Translation Surfaces, and Lyapunov Exponents

Luminy Summer 2015

Jayadev Athreya

on 3 July 2015

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Interval Exchange Maps, Translation Surfaces, and Lyapunov Exponents

Motivation, Goals, and Sources
Kontsevich-Zorich, `Lyapunov Exponents and Hodge Theory', 1997:

We started from computer experiments with simple one-dimensional ergodic dynamical systems, and quite unexpectedly ended with topological string theory
Motivating Quote
Understand the connection between
deviation of ergodic averages for IETs
Lyapunov exponents for Teichmuller flow
Develop a strategy for
understanding asymptotics for counting special trajectories on translation surfaces
Computing Siegel-Veech constants
One of the rare examples where Lyapunov exponents can be calculated
The numbers are geometrically meaningful
Numerical experiments are crucial in discovering new phenomenon
Main Ideas:

G. Forni
On the Lyapunov exponents of the Kontsevich-Zorich cocycle
, Handbook of Dynamical Systems v. 1B Elsevier, 2006.
M. Kontsevich and A. Zorich
Lyapunov Exponents and Hodge Theory
, `The mathematical beauty of physics', Saclay, 1996.
H. Masur
Ergodic Theory of Translation Surfaces
, Handbook of Dynamical Systems v. 1B Elsevier, 2006.
A. Zorich
Deviation for interval exchange transformations
, Ergodic Theory & Dynam. Systems, v. 17, 1477--1499, 1997.

A. Eskin
Counting problems in moduli space
, Handbook of Dynamical Systems v. 1B Elsevier, 2006.
A. Eskin and H. Masur
Asymptotic Formulas on Flat Surfaces
, Ergodic Theory & Dynam. Systems, v.21, 443-478, 2001.
Counting Problems
Lyapunov Exponents and Teichmuller Flow:
A orientation preserving, piecewise isometery of an interval.
What is an IET?
What parameters do we need to specify such a map?
is a permutation on n letters and
is a vector in
Connection to (moduli spaces of)
geometric structures on surfaces
Why do we care about IETs?
Rotations of the circle
: the map
Examples of IETs
mod 1
on the unit interval [0, 1).
Given a topological surface S of genus, a
translation structure
on S is an atlas of charts from S
(away from a finite set of points)
to the complex plane whose transition maps are translations.

Flat tori: given a lattice
Examples of Translation Surfaces
the quotient
is a flat torus.
What happens in higher genus?
Regular octagon:
Genus 2 Translation Surfaces
The regular octagon
genus 2,
one singular point

(angle = 1080 degrees)
Exercise: More generally, regular 4g-gon leads to genus g surface with one singular point, angle (4g-2)x180 degrees.
L-shaped table:
Again, get a genus 2 surface
with one singular point, angle 1080
L-shaped table
Identifying parallel sides by translation
get genus 2 surface with two singular
points, each angle 720 degrees
Complex Analysis
A translation surface is given by a collection of euclidean polygons (not necessarily convex) P_1, P_2, ..., P_K, sitting in the complex plane, so that the collection of sides can be divided into sets of parallel pairs of the same length, which are then identified by translation.
Euclidean geometry
Given a topological surface S of genus, a
translation structure
on S is an atlas of charts from S
(away from a finite set of points)
to the complex plane whose transition maps are translations.
A genus g translation surface is a pair (X, w), where X is a compact genus g Riemann surface, and w is a holomorphic one-form, w= f(z)dz in local coordinates.

The zeros of w correspond to singular points for the flat metric, and a zero of order k yields a cone angle of order (k+1)360 degrees
Integrate w to get charts
Pull back dz
"developing map"
Pull back dz
cut up surface
push forward Euclidean structure
Each translation surface yields a (family of) dynamical system(s),
the directional (or straight line) flows.
Dynamics on Translation Surfaces
The connection with IETs is given by taking the
first return map
of these flows to a transverse arc.
If you play billiards in a polygon with angles rational multiples of
You can
, and obtain a translation surface. Rationality condition yields that there will only be finitely many copies of your original polygon. The billiard flow is given by the flow on the translation surface.
For example, for the square torus, the return map for directional flow is a rotation.
Let (X, m) be a measure space, T: X -> X a measure preserving map.

Ergodic Theory
is the study of long-term behavior of orbits of T, and more generally of the study of orbits of measure-preserving actions of general groups

We will work with
probability spaces
: spaces (X, m) where m(X) = 1.

We say a measure-preserving transformation T: X -> X is ergodic if the space cannot be decomposed non-trivially (in the measure category) into T-invariant pieces. Formally, if A is a T-invariant subset of X, we want either

m(A) = 0 or m(A) = 1.

The most important theorem in Ergodic Theory is the
Birkhoff Ergodic Theorem
Let T: (X, m) -> (X, m) be an ergodic transformation of a probability space X. Then for any integrable function f, we have that for a randomly chosen point, the
time averages converge to the space average
. That is, for m-almost every x,
The Birkhoff Ergodic Theorem
In particular, if f is the indicator function of a set A, this says that the
amount of time the orbit spends in A is proportional to the measure of A
This theorem is fantastic, but has one niggling problem: it can't be improved, in general, to a statement about
every point.
Unique Ergodicity
So what can we do?
Let (X, d) be a metric space. We say a Borel-measurable map T: X -> X is unqiuely ergodic if there is a unique T-invariant Borel probability measure m on X.
Exercise: Prove that for a continuous function f on a compact metric space X, the conclusion of Birkhoff's theorem holds for every point x in X. Use weak-* compactness of the set of Borel probability measures P(X).
IETs and Unique Ergodicity
When are IETs uniquely ergodic? That is, when do they preserve
Lebesgue measure?
Exercise: A rotation is uniquely ergodic if and only if it is irrational.
Exercise: More generally, regular (4g+2)-gon leads to genus g surface with two singular points, angle 2g x 180 degrees.
Obvious Obstructions
What are some things to rule out?
If the permutation fixes some leading subset {1, 2, ..., k} of {1, 2, ..., n}, we'll be able to split the interval into invariant pieces, each supporting invariant Lebesgue measure. We call a permutation that does not fix any such subset

We'll only work with irreducible IETs from now on.
If there are points without dense orbits, then there can be invariant measures constructed on the orbit closures.
We say that an IET T satisfies the
if the backward orbit of every discontinuity point of T is infinite and distinct.
Keane's I.D.O.C.
Infinite Distinct Orbit Condition:
Theorem (Keane)
If T satisfies the i.d.o.c., every well-defined forward orbit is dense. We say in this case that T is
Exercise: prove this theorem, and prove that if the lengths of intervals are linearly independent over the rationals, then T satisfies the i.d.o.c.
The Boshernitzan Criterion
If T is an irreducible IET of n intervals,
T^k is an irreducible IET of roughly kn intervals.
Theorem (Boshernitzan):
Let m_k denote the shortest interval for T^k. Then if T is not uniquely ergodic,
km_k -> 0
as k tends to infinity.
This implies the following, a solution to a conjecture of Keane (resolved originally by Masur and Veech, independently):
Generic IETs
Given an irreducible permutation on n letters, for Lebesgue almost-every choice of parameters, the associated IET is uniquely ergodic.
There is a crucial `renormalization' relationship between
dynamics on a translation surface
, and the
orbit of the translation surface in an appropriate dynamical system on the space of translation surfaces
k = (k_1, k_2, ..., k_m)
denote a positive integer partition of 2g-2. The stratum
consists of all translation surfaces of genus g with singular points with angles (k_i +1)x360 degrees.

Equivalently, it consists of the space of pairs (X, w), where X is a genus g Riemann surface and w a holomorphic 1-form with zeros of order (k_1, k_2, ..., k_m). We say two pairs (X_1, w_1) and (X_2, w_2) are equivalent if there is a biholomorphism f: X_1 -> X_2 with f*w_2 = w_1.

Konstevich-Zorich showed that each stratum contains at most 3 connected components, and that `most' are connected.
Strata of Translation Surfaces
Real and Imaginary Foliations
In complex analytic language, the orbits of the straight line flows are:
vertical: leaves of Re(w) =0
horizontal: leaves of Im(w) = 0
Direction t: vertical on exp(it)w
There is a natural
SL(2, R) action
on the space of translation surfaces, preserving connected components of strata, given simply by linear post-composition with charts.

Equivalently given a matrix G and a collection of polygons (P_1, P_2, ..., P_k) representing a translation surface, we build a new translation surface out of (GP_1, GP_2, ..., GP_k).

Why SL(2, R)? We work with
translation surfaces.
SL(2, R)-action
Slight lie: up to a homothety.
Examples of the action:
Below, the two surfaces are the same: the matrix acts by an affine diffeomorphsim. In fact, for the torus,
matrix in SL(2, Z) works.
A saddle connection
Saddle Connections
on a translation surface (X, w) is a geodesic in the flat metric connecting two singular points (zeros) with none in its interior.
The holonomy vector of a saddle connection
is the integral
Given a translation surface (X, w), define
to be the set of holonomy vectors of saddle connections on (X, w)
This assignment is SL(2, R)-equivariant. Given G in SL(2, R):
V_Gw = G V_w
Exercise: the set V_w is a discrete subset of the plane.
Given a stratum H, let l: H -> R^+ be the function
l(X, w) = length of the shortest vector in V_w.
For any c>0, the set
{(X, w): l(X, w)>c}
is pre-compact.
Teichmuller Geodesic Flow
The action of g_t on a stratum is called
Teichmuller geodesic flow
, since the projection of an orbit to the moduli space of Riemann surfaces is a geodesic in the Teichmuller metric.
g_t is a renormalization flow for the flow on a translation surface- it `speeds up' the vertical flow.
An important example of a stratum is
H() = SL(2, R)/SL(2, Z).
Masur's Criterion (weak version):
Masur's Criterion
Suppose the vertical flow on (X, w) is not uniquely ergodic. Then l(X_t, w_t) -> 0 as t tends to infinity.
Write (X_t, w_t) for g_t(X,w)
That is, non-divergence
the orbit of the translation surface to unique ergodicity
the translation surface.
In the case of the torus, H(), Masur's criterion checks if the vertical flow has closed leaves- the only obstruction to having unique ergodicity.

In general, however, you can have minimal, non-uniquely ergodic behavior. For example, for the billiard table with a barrier, for appropriate choices of the height, you can have a big set of minimal, non-uniquely ergodic directions (Veech, Masur, Cheung, Hubert-Cheung-Masur)
Suppose T is a uniquely ergodic IET. Then for any subinterval J, we know that for any point x with a well-defined forward orbit,
|{0<=i<=n: T^i x in J}| = n|J| + o(N)
Zorich was motivated by the following:
Leaves of 1-forms
Zorich asked:
What is the behavior of the error term?
How do the leaves of a closed 1-form wind around a surface?
In our setting, take w to be a holomorphic 1-form, and consider the vertical foliation. Start at a point x, follow the leaf through x until time T. Close the trajectory up with some bounded length segment , to get a closed curve and a homology class.
How does this vector in homology grow with T?
Zorich observed, using computer experiments, that the error term seemed to satisfy a
power law:
There was a number
Zorich's Observations:
so that
In fact, Zorich noticed `turtles all the way down'- if you kept subtracting previous error terms, you got more power laws until you finally got a bounded error.
Motivating question:
What are these numbers? What is their geometric meaning?
Gave a Translation Surface version of
Boshernitzan's criterion
Masur's criterion
Explained what
Rauzy Induction
Explained what
Lyapunov Exponents
Hinted at connection between
deviations of ergodic averages for IETs
Lyapunov exponents for KZ cocycle

Renormalization Review:
Luminy Summer School 2015
Jayadev S. Athreya
Mathematics Department, University of Illinois

Counting Agenda

Translation Surfaces
moduli space of translation surfaces
Ergodicity and Unique Ergodicity
Boshernitzan's Criterion
Two Invariant Measures
If not minimal, have vertical saddle connection.
If minimal, not uniquely ergodic, we have (at least)
two invariant measures.
We have two orbits (or leaves) that
behave differently
The fact that
lim inf l(g_t w)>0
means there is a limit surface
Limit surfaces and Rectangles
The existence of two measures means there is a rectangle Q so that
There are leaves through points points x, y that visit Q with frequencies m_1(Q), m_2(Q) respectively
Applying g_t `speeds up' the vertical flow.
On limit surface, can find points corresponding to x,y which bound a rectangle with no singularities.
So leaves have to hit Q in the same way!
We got (qualitative) information about flows on translation surfaces from Teichmuller flow, which `renormalized' them.
Is there a similar strategy that can gives us
information for IETs?
Can build a surface out of an IET, by putting rectangles on top of intervals, zipping up, and zipping down.
Building surfaces
To speed up IET, cut off an interval at the end, look at first return to what's left
We get a new IET by inducing, what happens to the data ?
Induction Type 1
Induction Type 2
Depends on which interval we cut off. Our procedure is to cut off the shortest interval possible at the right, that is, either of length
Change in Data
changes by a linear transformation, ones on the diagonal and one off-diagonal -1
can be easily computed as well. The set of permutations you can get is called the
Rauzy class
To each IET, by iteration, get path in Rauzy graph, and sequence of 0-1 matrices A_1, A_2, ... A_n so that the original lengths are related to the lengths at stage N by the product
A_1 A_2... A_N
Paths, Matrices and the Kontsevich-Zorich cocycle
The `limiting' eigenvalues of this matrix are deviation exponents for IET.
compute Rauzy induction for rotations.
compute the Rauzy graph above
What are these limiting eigenvalues? What does this even mean?
Let M_1, M_2,... M_k denote your favorite collection of matrices in SL(m, R). Let p_1, p_2, ...,p_k be probabilities,
p_1 + p_2 + ... + p_k = 1
Let X_1, X_2, ... X_n, ... be an i.i.d. sequence with P(X_1 = M_i) = p_i. Let || . || denote some matrix norm.

Furstenberg-Kesten showed that, with probability 1, as N goes to infinity,
1/N log ||X_n... X_1||
has (the same) limit.
Random Matrix Products
Vastly, and beautifully, generalized by Oseledets:
Oseledet's Theorem
Actually, any matrix-valued cocycle A satisfying the finite first moment condition works:
The crucial lemma in both of these is:
Kingman's Subadditive Ergodic Theorem
These numbers are called
Lyapunov Exponents
Rauzy Induction is a map
Invariant measures and speed-ups
from the simplex (cross the Rauzy class) to itself. Veech constructed an
(infinite) invariant measure,
and showed the map was
Zorich `sped up' the map, by doing all type I or type II transformations in one giant step. He showed this has a
finite (ergodic) invariant measure
, and that the cocycle has
finite first moment,
which yielded existence of Lyapunov exponents!
Exercise: work out these invariant measures for case of rotations.
Key Relationship:
Rauzy Induction and Teichmuller Flow
Rauzy induction is a `discrete version' of Teichmuller flow. Precisely, the Teichmuller flow is a
suspension flow
over the
natural extension
of (fast) Rauzy induction.
For a concrete example, the geodesic flow on
SL(2, R)/SL(2,Z) = H()
is a suspension flow over the natural extension of the
Farey Map
F: [0, 1) -> [0, 1),
Gauss Map
G: [0,1) -> [0, 1)
G(x) = {1/x} = 1/x - [1/x]
Definitions Review
Renormalization Agenda
Give a Translation Surface version of
Boshernitzan's criterion
Masur's criterion
Exlplain what
Rauzy Induction
Explain what
Lyapunov Exponents
Hinted at connection between
deviations of ergodic averages for IETs
Lyapunov exponents for KZ cocycle
counting problems
on translation surfaces.
Discuss Eskin-Masur/Veech strategy to link to
equidistribution on
volumes of
State miraculous connection to
Lyapunov exponents.
Given translation surface (X, w), compute asymptotics of

N(w, R) = |{v in V_w: |v| < R}|
For the square torus T, putting a fake singularity at 0, saddle connections are
primitive integer vectors
, so

N(T, R) = vol(B(o,R))/zeta(2) + lower order terms
This can be proved by tiling arguments + a little elementary number theory.
For more general surfaces, we need a different strategy.
The Siegel-Veech Machine
Let Q be the
with vertices at (1, 1), (-1, 1), (1/2, 1/2), (-1/2, -1/2).
The Trapezoid Trick
(True) Mathematical Folklore:
This strategy was developed by
after his mind wandered when he was attending a lecture of
, while
extremely jetlagged
The Siegel-Veech transform:
So N(w, R) is the S-V transform of the indicator function of the ball of radius R.
Then we have, for t>>0
So if we put R^2 = e^t, then this integral gives a weight of R^-2 to each vector v with R/2 < |v| < R
So N(R) - N(R/2) is roughly the sum of the above integral over all holonomy vectors, multiplied by R^2.
If we could show that the measures supported on the pieces of orbit
`spread out evenly'
, that is,
according to some measure on the stratum H, we could solve our counting problem... almost.
Integrals on Strata
That is, we have:
Illustrating the Trapezoid Trick
We first need a measure on strata!
We can put coordinates on the stratum H(k_1, ..., k_m) by
fixing a basis a_1, ... a_g, b_1, ..., b_g, r_1, ... r_m for homology
to the singular points.
Taking periods
That is, we view w as an element of
relative cohomology
Masur-Veech Measure
In these coordinates,
area is a quadratic form.
Theorem (Masur, Veech):
The Lebesgue measure on this area 1 `hypersurface' is a finite ergodic SL(2,R)-invariant measure.
Let SL(2, R) act ergodically on a (topological) probability space (X, m). Then for almost every x in X, and every continuous compactly supported function f on X,
Nevo's Ergodic Theorem
How do we use this?
So using Nevo's Theorem, we get, for m-almost every w in a stratum H,
Applying Nevo's Theorem
where m is Masur-Veech measure.
Actually, this works for any ergodic
SL(2, R)- invariant probability measure on the stratum.
But how do we evaluate the integral?
Let m be an SL(2, R) invariant measure on a stratum H. Suppose that for
The Siegel-Veech formula
Theorem (Veech):
Then there is a constant c = c_m (
the Siegel-Veech Constant
) so that
Corollary (Eskin-Masur):
Let m be Masur-Veech measure. Then for m-almost every w in H,
N(w, R)/R^2 --> c_H
as R --> 00. Here c_H denotes the S-V constant for the Masur-Veech measure m.
Proof: N(R) = (N(R)- N(R/2)) + (N(R/2)- n(R/4)) + ...
Applying our machine, the terms are each behaving like
2^-k x c_H x area(Q) = 2^-k x 3/4 c_H.
Summing the series in k gives a 4/3, so we get c_H, as desired.
To prove the Siegel-Veech formula, consider the map
Proving Siegel-Veech
This is a
linear, SL(2,R)-invariant functional
on C_c(R^2).

So it comes from integrating f against an
SL(2, R)-invariant measure on R^2
. Only two (up to scaling)-
delta measure at 0
, and
. Can rule out delta measure at 0, so get scalar multiple of Lebesgue.
Can you compute the constant c_H?
In many cases, yes.
For some examples (like the torus), we have a big stabilizer in SL(2, R). Following McMullen, we use SL(X,w) to denote this group.
Lattice Surfaces
If SL(X,w) is a lattice, we say (X, w) is a
lattice surface.
Exercise: SL(X, w) is a discrete subgroup of SL(2, R)
Smillie showed that the SL(2, R) orbit of (X,w) is closed if and only if (X,w) is a lattice surface.
In this case, we can take our measure to be Haar measure on SL(2, R)/SL(X, w) and our S-V constant is the natural volume. Applied to the torus, we get our zeta(2) factor.
What about the constant c_H for Masur-Veech measure? Using a slightly different constant c_area (associated to counting cylinders weighted by area), Eskin-Kontsevich-Zorich proved the amazing:
SV Constants and Lyapunov Exponents
Thus, counting problems are also related to Lyapunov Exponents!!!!!!!
Thanks to
Alex Eskin
Samuel Lelievre
Thierry Monteil,
Anton Zorich
, for use of images.

Thanks to Pascal, Erwan, and Anton for an incredible job organizing the meeting.
Full transcript