Invariant sets for the wind-tree model (original) (raw)

Yuriy Tumarkin Institut für Mathematik, Universität Zürich, Winterthurerstrasse 190, CH-8057 Zürich, Switzerland yuriy.tumarkin@math.uzh.ch

Abstract.

We consider the wind-tree model, a ℤ2\mathbb{Z}^{2}-periodic billiard. In the case when the underlying compact translation surface lies on a periodic orbit of the Teichmüller geodesic flow, and at least one of the two homology classes defining the cover is unstable for the Kontsevich-Zorich cocycle, we prove that every orbit closure of the billiard has Hausdorff dimension strictly smaller than 2. The proof relies on a construction of explicit invariant functions, which along the way gives a new proof of non-ergodicity and non-transitivity of the wind-tree model for all parameters and almost all directions, as first shown by Fra̧czek and Ulcigrai in [FU14].

1. Introduction

The Ehrenfest wind-tree model is a billiard in the plane with rectangular obstacles. First introduced by P. Ehrenfest and T. Ehrenfest in 1911 [EE11] to model gas diffusion, the periodic version as first studied by Hardy and Weber [HW80] has attracted a lot of attention in the last decade and a half. It is defined as follows: a point (the ‘wind’) moves along straight lines in the plane ℝ2\mathbb{R}^{2} and bounces elastically off rectangular obstacles (the ‘trees’), which are translates of the rectangle [0,a]×[0,b][0,a]\times[0,b] for some parameters 0<a,b<10<a,b<1, centered at each element of ℤ2\mathbb{Z}^{2}.

Most of the recent progress on the wind-tree model ([HLT11, CG12, Del13, DHL14, FU14, FH18, DZ19, AH20, CHLP25]) has been due to the following observation. Following the classical unfolding construction of Katok and Zemlyakov [KZ75], the billiard flow on the wind-tree model is equivalent to the straight-line flow on an infinite translation surface X∞X_{\infty} (see Section˜2 for definitions), which turns out to be a ℤ2\mathbb{Z}^{2}-cover of a genus 5 compact translation surface XX, which in turn is a fourfold cover of a genus 2 translation surface YY. This reduction to compact translation surfaces allows one to apply deep results from Teichmüller dynamics (such as the formula of Eskin, Kontsevich and Zorich [EKZ14] for the Lyapunov exponents of the Kontsevich-Zorich cocycle and the work of Chaika and Eskin [CE15] on Oseledets genericity).

Among the most celebrated results on the wind-tree model is the result of Fra̧czek and Ulcigrai [FU14], showing that for any parameters a,ba,b, the billiard flow in almost any direction θ\theta is non-ergodic, and further even non-transitive, meaning that it has no dense orbits. This is in stark contrast to well-known results about compact translation surfaces, where by the work of Kerckhoff, Masur and Smillie [KMS86], for any compact translation surface the straight-line flow is ergodic (and even uniquely ergodic) in almost every direction.

The work of Fra̧czek and Ulcigrai opens many natural questions which need to be answered to obtain a full understanding of the dynamics of the wind-tree model:

• •
  We know that Lebesgue measure is not ergodic for the flow, so what are its ergodic components?
• •
  Further, we know that generically no orbits are dense, so what are their closures?

The goal of this article is to take a first step towards answering these questions, with a hope of spurring further progress in this direction.

Denote by X​(a,b)X(a,b) the compact translation surface underlying the wind-tree model with parameters (a,b)(a,b), and by rπ/2−θr_{\pi/2-\theta} the rotation by π/2−θ\pi/2-\theta (see Section˜2 for precise definitions). We discuss the case when rπ/2−θ​X​(a,b)r_{\pi/2-\theta}X(a,b) is periodic for the Teichmüller geodesic flow. The cover rπ/2−θ​X∞​(a,b)r_{\pi/2-\theta}X_{\infty}(a,b) is defined by two homology classes γh,γv∈H1​(X,ℤ)\gamma_{h},\gamma_{v}\in H_{1}(X,\mathbb{Z}) (see Section˜3).

The main new result that we obtain is as follows:

Theorem 1.1.

Let a,b∈(0,1)a,b\in(0,1) be any parameters, let θ\theta be a direction such that rπ/2−θ​X​(a,b)r_{\pi/2-\theta}X(a,b) is periodic for the Teichmüller geodesic flow and at least one of γh\gamma_{h} and γv\gamma_{v} is unstable. Then for the billiard flow on the wind-tree model with parameters a,ba,b in direction θ\theta, the closure of every orbit has Hausdorff dimension strictly less than 2.

Remark 1.2.

By unstable we mean that γh\gamma_{h} or γv\gamma_{v} belongs to the the unstable space Eω+E^{+}_{\omega}. (See Section˜2).

Since γh\gamma_{h} and γv\gamma_{v} are integer vectors, they can never be stable, but they can in some cases be central.

Let Ψ\Psi be the pseudo-Anosov diffeomorphism of X​(a,b)X(a,b) corresponding to the closed Teichmüller geodesic containing rπ/2−θ​X​(a,b)r_{\pi/2-\theta}X(a,b). If both γh\gamma_{h} and γv\gamma_{v} are invariant under the action of Ψ\Psi, this corresponds to the case when Ψ\Psi lifts to the cover X∞​(a,b)X_{\infty}(a,b). This case was studied in [Tum25], where it was shown that in this case the flow is ergodic, and further one has a full classification of Radon ergodic invariant measures as the Maharam measures.

As far as we are aware, other central cases, when Ψ\Psi does not lift, are still open.

Fraçzek and Hubert extended the results of [FU14] to more general translation surfaces and higher dimensional covers in [FH18], and in fact our result holds in this more general setting.

The setup and large part of the technical input for our proof is along the same lines as that of [FU14, FH18]. We also use the observation that the homology of XX splits into several invariant blocks for the Kontsevich-Zorich cocycle, and that the homology classes defining the cover X∞X_{\infty} are complemented by homology classes which are stable for the cocycle. These stable classes determine cocycles over the first-return map of the flow which are coboundaries. From here, while [FU14] and [FH18] use the theory of essential values to conclude non-ergodicity, we rely on a result of Marmi, Moussa and Yoccoz [MMY05] which states that these coboundaries have continuous transfer functions. This allows us to find continuous invariant functions for the flow, which provide the main ingredient in the proof of ˜1.1.

Another application of the explicit form of the invariant functions is that one can use them to plot computational approximations of invariant sets for the billiard flow. An example with θ\theta a gtg_{t}-periodic direction is shown in Figure˜1.

Figure 1. A 400×400400\times 400 window of an invariant set for the billiard in the wind-tree model, in the gtg_{t}-periodic case.

The outline of the paper is as follows. In Section˜2 we introduce the necessary background on translation surfaces and interval exchange transformations. In Section˜3 we discuss covers of translation surfaces and the geometry of the wind-tree model and the relevant translation surfaces. In Section˜4 we prove the existence of continuous invariant functions for the straight-line flow. Finally in Section˜5 we show that in the unstable gtg_{t}-periodic case the invariant sets given by level sets of the aforementioned invariant functions have Hausdorff dimension less than 2, proving ˜1.1. In addition, in Appendix˜A we say a few words on how one can plot the invariant sets as in Figure˜1.

1.1. Acknowledgments

I would like to thank my advisor Corinna Ulcigrai for teaching me about the Ehrenfest wind-tree model and encouraging me to work on this problem, and for her continuous support during this work. I am grateful to Krzystof Fra̧czek, Pascal Hubert and Anton Zorich for helpful discussions, and I would like to thank Andrea Ulliana for showing me how to simplify the proof of ˜5.6. This work was supported by the Swiss National Science Foundation through Grant 213663.

2. Background on translation surfaces and IETs

A translation surface is a tuple (M,ω)(M,\omega), where MM is an oriented surface (which may not be compact) and ω\omega is a complex structure on MM with an abelian differential, i.e. a non-zero 1-form which is holomorphic with respect to the complex structure. We denote by Σ⊂M\Sigma\subset M the set of zeros of ω\omega, also called the singular points of (M,ω)(M,\omega).

For a direction θ∈S1\theta\in S^{1}, denote by XθX^{\theta} the unit vector field in direction θ\theta on M∖ΣM\setminus\Sigma. Denote by (φtθ)t∈ℝ(\varphi^{\theta}_{t})_{t\in\mathbb{R}} the corresponding flow, which is called the straight-line flow in direction θ\theta. We denote the straight-line flow in the vertical direction by φtv\varphi^{v}_{t}. The flow (φtθ)t∈ℝ(\varphi^{\theta}_{t})_{t\in\mathbb{R}} preserves the natural volume form on (M,ω)(M,\omega) given by ℜ⁡ω∧ℑ⁡ω\Re\omega\wedge\Im\omega.

2.1. Teichmüller flow

2.1.1. Teichmüller space and moduli space

Denote by Diff+​(M)\mathrm{Diff}^{+}(M) the group of orientation-preserving diffeomorphisms of MM, and by Diff0+​(M)\mathrm{Diff}_{0}^{+}(M) the subgroup of diffeomorphisms isotopic to the identity. The quotient Γ​(M):=Diff+​(M)/Diff0+​(M)\Gamma(M):=\mathrm{Diff}^{+}(M)/\mathrm{Diff}^{+}_{0}(M) is the mapping class group of MM.

The group Diff+​(M)\mathrm{Diff}^{+}(M) acts on the space of abelian differentials on MM by pre-composition. Denote by 𝒯​(M)\mathcal{T}(M) the Teichmüller space of abelian differentials, the space of orbits of the action restricted to Diff0+​(M)\mathrm{Diff}_{0}^{+}(M). Denote by ℳ​(M)\mathcal{M}(M) the moduli space of abelian differentials, the space of orbits of the action of the entire group Diff+​(M)\mathrm{Diff}^{+}(M). Then ℳ​(M)=𝒯​(M)/Γ​(M)\mathcal{M}(M)=\mathcal{T}(M)/\Gamma(M).

Defining the area of an abelian differential ω\omega by A​(ω)=∫Mℜ⁡ω∧ℑ⁡ωA(\omega)=\int_{M}\Re\omega\wedge\Im\omega, we define also the Teichmüller and moduli spaces of unit-area abelian differentials by 𝒯1​(M):={[ω]∈𝒯​(M):A​(ω)=1},ℳ1​(M):={[ω]∈ℳ​(M):A​(ω)=1}\mathcal{T}_{1}(M):=\{[\omega]\in\mathcal{T}(M):A(\omega)=1\},\,\mathcal{M}_{1}(M):=\{[\omega]\in\mathcal{M}(M):A(\omega)=1\}.

2.1.2. The S​L​(2,ℝ)SL(2,\mathbb{R}) action

The group S​L​(2,ℝ)SL(2,\mathbb{R}) has a natural action on 𝒯​(M)\mathcal{T}(M) and ℳ​(M)\mathcal{M}(M) in the following way. Given a translation structure ω\omega, its primitives define charts to ℂ\mathbb{C}. Post-composition of these charts by an element gg of S​L​(2,ℝ)SL(2,\mathbb{R}) gives new charts that define a new complex structure and a new holomorphic abelian differential, thus defining a new translation structure g⋅ωg\cdot\omega.

The Teichmüller geodesic flow (gt)t∈ℝ(g_{t})_{t\in\mathbb{R}} is the restriction of the S​L​(2,ℝ)SL(2,\mathbb{R}) action to the diagonal subgroup{gt:=diag​(et,e−t):t∈ℝ}\{g_{t}:=\mathrm{diag}(e^{t},e^{-t}):t\in\mathbb{R}\}.

Say that ω\omega is gtg_{t}-periodic if it is periodic under the Teichmüller geodesic flow on the moduli space, meaning that for some T0T_{0}, gT0⋅ω=ωg_{T_{0}}\cdot\omega=\omega in ℳ​(M)\mathcal{M}(M).

We will also sometimes refer to the restriction of the action to S​O​(2,ℝ)SO(2,\mathbb{R}), denoting by rθr_{\theta} the matrix of rotation by θ\theta.

2.1.3. The Kontsevich-Zorich cocycle

Define the homological Hodge bundle over ℳ​(M)\mathcal{M}(M) by

Here Γ​(M)\Gamma(M) acts on 𝒯1​(M)\mathcal{T}_{1}(M) by pre-composition and on H1​(M,ℝ)H_{1}(M,\mathbb{R}) by the natural pushforward action.

The Kontsevich-Zorich cocycle (GtK​Z)t∈ℝ(G^{KZ}_{t})_{t\in\mathbb{R}} is the quotient by Γ​(M)\Gamma(M) of the trivial cocycle

to ℋ1​(M,ℝ)\mathcal{H}_{1}(M,\mathbb{R}).

Note that the Kontsevich-Zorich cocycle is usually defined on the cohomological Hodge bundle ℋ1​(M):=(𝒯1​(M)×H1​(M,ℝ))/Γ​(M)\mathcal{H}^{1}(M):=(\mathcal{T}_{1}(M)\times H^{1}(M,\mathbb{R}))/\Gamma(M), but we can identify the homological and cohomological bundles via the Poincaré duality 𝒫:H1​(M,ℝ)→H1​(M,ℝ)\mathcal{P}:H_{1}(M,\mathbb{R})\to H^{1}(M,\mathbb{R}). This identification allows us to define the Hodge norm on the homological bundle by pulling back the Hodge norm on the cohomological bundle.

The Hodge norm on ℋ1​(M,ℝ)\mathcal{H}^{1}(M,\mathbb{R}) is defined as follows. Given c∈H1​(M,ℝ)c\in H^{1}(M,\mathbb{R}), there exists a unique one-form η\eta on MM which is holomorphic with respect to the complex structure of ω\omega and satisfies c=[ℜ⁡η]∈H1​(M,R)c=[\Re\eta]\in H^{1}(M,R). Then the Hodge norm of cc is given by

| ‖c‖ω:=(12​∫Mη∧η¯)12.\|c\|_{\omega}:=\left(\frac{1}{2}\int_{M}\eta\wedge\bar{\eta}\right)^{\frac{1}{2}}. | | ---------------------------------------------------------------------------------------------------------------------- |

We will denote the Hodge norm on ℋ1​(M,ℝ)\mathcal{H}_{1}(M,\mathbb{R}) by ‖σ‖ω:=‖𝒫​(σ)‖ω\|\sigma\|_{\omega}:=\|\mathcal{P}(\sigma)\|_{\omega}.

For a compact orientable surface MM, the homology H1​(M,ℝ)H_{1}(M,\mathbb{R}) has a natural symplectic structure arising from the algebraic intersection number. This symplectic structure is preserved by the action of Γ​(M)\Gamma(M) and hence the Kontsevich-Zorich cocycle acts symplectically.

2.1.4. Orbit closures and S​L​(2,ℝ)SL(2,\mathbb{R})-invariant subbundles

Let ω∈ℳ1​(M)\omega\in\mathcal{M}_{1}(M) and let 𝒩=S​L​(2,ℝ)⋅ω¯\mathcal{N}=\overline{SL(2,\mathbb{R})\cdot\omega} be its orbit closure. By the work of Eskin, Mirzakhani and Mohammadi in [EM18, EMM15], 𝒩\mathcal{N} is an affine S​L​(2,ℝ)SL(2,\mathbb{R})-invariant submanifold of ℳ1​(M)\mathcal{M}_{1}(M), and there is a probability measure νℳ\nu_{\mathcal{M}} supported on 𝒩\mathcal{N} which is invariant for the S​L​(2,ℝ)SL(2,\mathbb{R}) action and ergodic for the Teichmüller geodesic flow.

Suppose that V⊂H1​(M,ℝ)V\subset H_{1}(M,\mathbb{R}) is a symplectic subspace. An S​L​(2,ℝ)SL(2,\mathbb{R})-invariant subbundle over 𝒩\mathcal{N} with fibre VV is a subbundle 𝒱=⋃ω∈𝒩{ω}×V​(ω)\mathcal{V}=\bigcup_{\omega\in\mathcal{N}}\{\omega\}\times V(\omega) of ℋ1​(M,ℝ)\mathcal{H}_{1}(M,\mathbb{R}), such that for all ω∈𝒩\omega\in\mathcal{N}, V​(ω)≅VV(\omega)\cong V, and for any g∈S​L​(2,ℝ)g\in SL(2,\mathbb{R}), V​(g⋅ω)=V​(ω)V(g\cdot\omega)=V(\omega).

We denote the Kontsevich-Zorich cocycle restricted to 𝒱\mathcal{V} by (Gt𝒱)t∈ℝ(G^{\mathcal{V}}_{t})_{t\in\mathbb{R}}.

2.2. Generic properties of a translation surface

Let 𝒩=S​L​(2,ℝ)⋅ω¯\mathcal{N}=\overline{SL(2,\mathbb{R})\cdot\omega}, and let ν𝒩\nu_{\mathcal{N}} be the corresponding S​L​(2,ℝ)SL(2,\mathbb{R})-invariant measure on 𝒩\mathcal{N}. Fix a symplectic subspace V⊂H1​(M,ℝ)V\subset H_{1}(M,\mathbb{R}) and consider an invariant subbundle 𝒱\mathcal{V} over 𝒩\mathcal{N} with fibre VV.

We state three important properties satisfied by a ν𝒩\nu_{\mathcal{N}}-generic translation surface (M,ω)∈𝒩(M,\omega)\in\mathcal{N}. We first state the definitions of the properties, then the theorems which say that these properties are indeed generic. In fact stronger than holding for ν𝒩\nu_{\mathcal{N}}- almost every (M,ω)(M,\omega), by the results of Chaika-Eskin [CE15], these properties hold for every surface when rotated by almost every θ∈S1\theta\in S^{1}.

Definition 2.1.

We say that (M,ω)(M,\omega) is Birkhoff generic if the Birkhoff ergodic theorem holds for the Teichmüller geodesic flow at ω\omega, i.e. for every continuous integrable function ϕ\phi on 𝒩\mathcal{N},

Definition 2.2.

We say that (M,ω)(M,\omega) is Oseledets generic if there exists an Oseledets splitting for the Kontsevich-Zorich cocycle Gt𝒱G^{\mathcal{V}}_{t} with respect to ν𝒩\nu_{\mathcal{N}}, i.e. there exist Lyapunov exponents

and a direct splitting V=⨁i=1sVi​(ω)V=\bigoplus_{i=1}^{s}V_{i}(\omega), such that or all ξ∈Vi\xi\in V_{i},

| limt→∞1t​log⁡‖ξ‖gt​ω=λi.\lim_{t\to\infty}\frac{1}{t}\log\|\xi\|_{g_{t}\omega}=\lambda_{i}. | | ---------------------------------------------------------------------------------------------------------- |

If ω\omega is Oseledets generic, VV has a direct splitting into stable, central and unstable subspaces,

where

| Eω−:={ξ∈V:limt→∞1t​log⁡‖ξ‖gt​ω<0},\displaystyle E^{-}_{\omega}:=\{\xi\in V:\lim_{t\to\infty}\frac{1}{t}\log\|\xi\|_{g_{t}\omega}<0\}, | | ---------------------------------------------------------------------------------------------------------------------------------------------------------- | | Eω0:={ξ∈V:limt→∞1t​log⁡‖ξ‖gt​ω=0},\displaystyle E^{0}_{\omega}:=\{\xi\in V:\lim_{t\to\infty}\frac{1}{t}\log\|\xi\|_{g_{t}\omega}=0\}, | | Eω+:={ξ∈V:limt→∞1t​log⁡‖ξ‖gt​ω>0}.\displaystyle E^{+}_{\omega}:=\{\xi\in V:\lim_{t\to\infty}\frac{1}{t}\log\|\xi\|_{g_{t}\omega}>0\}. |

Definition 2.3.

We say that (M,ω)(M,\omega) is Masur generic if both the vertical and the horizontal flows are uniquely ergodic on (M,ω)(M,\omega).

Say that (M,ω)(M,\omega) is BOM generic if it is generic in all three senses above.

Theorem 2.4 (Theorems 1.1 and 1.2 in [CE15]).

For every (M,ω)(M,\omega), and almost every θ∈S1\theta\in S^{1}, (M,rθ​ω)(M,r_{\theta}\omega) is Birkhoff and Oseledets generic.

Theorem 2.5 (Theorem 2 in [KMS86]).

For every (M,ω)(M,\omega), and almost every θ∈S1\theta\in S^{1}, (M,rθ​ω)(M,r_{\theta}\omega) is Masur generic.

2.3. IETs

Fix a transverse segment II to the direction θ\theta on a translation surface. Since the straight-line flow (φtθ)t∈ℝ(\varphi^{\theta}_{t})_{t\in\mathbb{R}} is volume preserving, its Poincaré map on II is an orientation-preserving piecewise isometry of the interval. Such maps are called interval exchange transformations.

An Interval Exchange Transformation (IET for short) of nn intervals is a map T:[a,b)→[a,b)T:[a,b)\to[a,b) which is a piecewise orientation-preserving isometry, with all discontinuities contained in a set of n−1n-1 marked points {a<x1<⋯<xn−1<b}\{a<x_{1}<\dots<x_{n-1}<b\}. Let x0=a,xn=bx_{0}=a,x_{n}=b and let IiI_{i} be the interval [xi−1,xi)[x_{i-1},x_{i}) for 1≤i≤n1\leq i\leq n. Denote the intervals of continuity of TT by {Ij:j∈𝒜}\{I_{j}:j\in\mathcal{A}\}, where 𝒜={1,…,n}\mathcal{A}=\{1,\dots,n\}.

Say that TT satisfies Keane’s condition if for any 0≤i≤n0\leq i\leq n, any m≥1m\geq 1, Tm​(xi)T^{m}(x_{i}) does not belong to the set of marked points{x1,…,xn−1}\{x_{1},\dots,x_{n-1}\}. Let 𝒳n\mathcal{X}_{n} be the space of IETs of nn intervals satisfying Keane’s condition. Let 𝒳n0⊂𝒳n\mathcal{X}^{0}_{n}\subset\mathcal{X}_{n} be the subspace of IETs defined on the unit interval I=[0,1)I=[0,1).

A key tool for studying IETs is the Rauzy-Veech renormalisation ℛ\mathcal{R}. We do not give an explicit definition of Rauzy-Veech renormalisation here, only recalling some basic properties, and refer the reader to the many excellent introductions available in the literature, such as [Via08], [Yoc10], [Zor06].

The Rauzy-Veech induction is a map ℛ^:𝒳n→𝒳n\hat{\mathcal{R}}:\mathcal{X}_{n}\to\mathcal{X}_{n}. Given T:[a,b)→[a,b)T:[a,b)\to[a,b), the Rauzy-Veech induction procedure produces a sequence of shrinking nested intervals [a,b)=I(0)⊃I(1)⊃I(2)⊃…[a,b)=I^{(0)}\supset I^{(1)}\supset I^{(2)}\supset\dots, which all have the same left endpoint. The map ℛ^k​(T)\hat{\mathcal{R}}^{k}(T) is defined as the induced map of TT on I(k)I^{(k)}, also denoted by T(k):I(k)→I(k)T^{(k)}:I^{(k)}\to I^{(k)}. The subintervals of I(k)I^{(k)} permuted by T(k)T^{(k)} are labelled by Ii(k)I^{(k)}_{i} for i∈𝒜i\in\mathcal{A}, where the order of these subintervals depends on TT.

The Rauzy-Veech renormalisation is a map ℛ:𝒳n0→𝒳n0\mathcal{R}:\mathcal{X}^{0}_{n}\to\mathcal{X}^{0}_{n}, defined byℛ​(T)=T(1)|I(1)|\mathcal{R}(T)=\frac{T^{(1)}}{|I^{(1)}|}.

For 0≤r<k0\leq r<k let qj(r,k)q^{(r,k)}_{j} be the return time of T(r)T^{(r)} to I(k)I^{(k)} on Ij(k)I^{(k)}_{j}, i.e.

The Rokhlin tower Zj(r,k)Z^{(r,k)}_{j} is the union

Then by definition of the induced map and qj(r,k)q^{(r,k)}_{j}, I(r)=⨆j=1nZj(r,k)I^{(r)}=\bigsqcup_{j=1}^{n}Z^{(r,k)}_{j}.

For 0≤l<qj(r,k)0\leq l<q^{(r,k)}_{j} we call (T(r))l​(Ij(k))\big(T^{(r)}\big)^{l}\left(I^{(k)}_{j}\right) the lthl^{\text{th}} floor of the tower Zj(r,k)Z^{(r,k)}_{j}, and Ij(k)I^{(k)}_{j} the base.

The Rauzy-Veech cocycle is defined as integer-valued d×dd\times d matrices A(r,k)A^{(r,k)}, with

Thus Ai​j(r,k)A^{(r,k)}_{ij} counts the number of floors in the tower Zj(r,k)Z^{(r,k)}_{j} which are subsets of Ii(r)I^{(r)}_{i}. It is a cocycle since for r<s<kr<s<k, Ai​j(r,k)=∑t=1nAi​t(r,s)​At​j(s,k)A^{(r,k)}_{ij}=\sum_{t=1}^{n}A^{(r,s)}_{it}A^{(s,k)}_{tj}, and so A(r,k)=A(r,s)​A(s,k)A^{(r,k)}=A^{(r,s)}A^{(s,k)}.

T(r)T^{(r)}T(r)T^{(r)}T(r)T^{(r)}T(r)T^{(r)}T(r)T^{(r)}I1(k)I^{(k)}_{1}I2(k)I^{(k)}_{2}I3(k)I^{(k)}_{3}(T(r))q3(r,k)−1​(I3(k))\Big(T^{(r)}\Big)^{q^{(r,k)}_{3}-1}\Big(I^{(k)}_{3}\Big) #\# of floors =q1(r,k)=q_{1}^{(r,k)} Z2(r,k)Z^{(r,k)}_{2}

Figure 2. An example of towers for an IET on 3 intervals. The union of the base intervals is I(k)I^{(k)}, and the union of all the floors (including the base) is I(r)I^{(r)}. From the towers we can not see exactly the image under T(r)T^{(r)} of the top floor of a tower, but we do know that it gets sent into I(k)I^{(k)}, hence somewhere in the union of the bases.

We say an IET TT is of periodic type if it is a fixed point of ℛN\mathcal{R}^{N} for some NN. For Keane IETs it follows automatically that some power of the matrix A(0,N)A^{(0,N)} is positive, hence we do not need to include this condition as part of the definition.

Given any function gg on II, denote its Birkhoff sums with respect to TT by SnT​g​(x):=∑k=0n−1g​(Tk​x)S^{T}_{n}g(x):=\sum_{k=0}^{n-1}g(T^{k}x).

Consider a piecewise constant function ϕ\phi on II, such that ϕ\phi is constant on every IjI_{j} for j∈𝒜j\in\mathcal{A}. The following is a standard fact, allowing one to compute the Birkhoff sums of ϕ\phi at special times using the Rauzy-Veech cocycle. For a proof, see for example [Tum25, Lemma 3.10].

Lemma 2.6.

Let ϕ\phi be a piecewise constant function, constant on each IjI_{j}, let ϕ(k)=(A(0,k))T​ϕ\phi^{(k)}=(A^{(0,k)})^{T}\phi. Then for any j∈𝒜j\in\mathcal{A} and any k>0k>0, for x∈Ij(k)x\in I^{(k)}_{j}, Sqj(0,k)T​ϕ​(x)=ϕ(k)​(x)S^{T}_{q_{j}^{(0,k)}}\phi(x)=\phi^{(k)}(x).

Rauzy-Veech renormalisation is a discretisation of the Teichmüller geodesic flow, and the Rauzy-Veech cocycle is a discretisation of the Kontsevich-Zorich cocycle. (See [Zor06].) In particular, a translation surface (M,ω)(M,\omega) being gtg_{t}-periodic implies that any IET which is a Poincaré section of the vertical flow is periodic under Rauzy-Veech renormalisation.

2.4. Stable space and coboundaries

Let II be a horizontal interval on M∖ΣM\setminus\Sigma that has no self-intersections. The first return map of the flow to II is an IET T:I→IT:I\to I, label its continuity intervals by {Iα:α∈𝒜}\{I_{\alpha}:\alpha\in\mathcal{A}\}. For α∈𝒜\alpha\in\mathcal{A} denote by ξα∈H1​(M,ℤ)\xi_{\alpha}\in H_{1}(M,\mathbb{Z}) the homology class of the loop obtained by following the flow (φtθ)t∈ℝ(\varphi^{\theta}_{t})_{t\in\mathbb{R}} from any point x∈Iαx\in I_{\alpha} until T​(x)T(x) and then returning to xx along the segment [x,T​(x)]⊂I[x,T(x)]\subset I. Given a homology class γ∈H1​(M,ℝ)\gamma\in H_{1}(M,\mathbb{R}), we can define a corresponding piecewise constant function ϕγ\phi_{\gamma} on II by

| ϕγ|Iα:=⟨γ,ξα⟩.\phi_{\gamma}|_{I_{\alpha}}:=\langle\gamma,\xi_{\alpha}\rangle. | | ------------------------------------------------------------------------------------------ |

Let Φ:H1​(M,ℝ)→ℝn\Phi:H_{1}(M,\mathbb{R})\to\mathbb{R}^{n} be the functions that assigns to the homology class γ\gamma the piecewise constant function ϕγ\phi_{\gamma} as defined above.

Note that if γ∈H1​(M,ℤ)\gamma\in H_{1}(M,\mathbb{Z}), then Φ​(γ)∈ℤn\Phi(\gamma)\in\mathbb{Z}^{n}.

The following combination of existing results, connecting negative Lyapunov exponents and coboundaries for the Poincaré section, is the key result that we apply.

In [MMY05], Marmi, Moussa and Yoccoz introduced a full measure Diophantine condition on IETs called Roth type. We will not reproduce the definition here since we will not need to use it directly and refer the interested reader to [MMY05].

Theorem 2.7 (See Theorem A in [MMY05] and Corollary 3.6 in [MMY12]).

Let TT be an IET of Roth type. Suppose ψ\psi is a piecewise function that is constant on the intervals of continuity of TT. Suppose that ψ\psi is a coboundary, then there exists a continuous function hh such that ψ=h∘T−h\psi=h\circ T-h.

Theorem 2.8.

Suppose that (M,ω)(M,\omega) is either BOM generic or gtg_{t}-periodic. Then there exists a horizontal interval I⊂M∖ΣI\subset M\setminus\Sigma with no self-intersections, such that the first return map T:I→IT:I\to I of the vertical flow to II is a minimal ergodic IET, and for every α∈Eω−\alpha\in E^{-}_{\omega}, the piecewise-constant function Φ​(α)\Phi(\alpha) on II is a coboundary, with a continuous transfer function h:I→ℝh:I\to\mathbb{R} such that Φ​(α)​(x)=h​(T​(x))−h​(x)\Phi(\alpha)(x)=h(T(x))-h(x).

Proof.

In the case that (M,ω)(M,\omega) is BOM generic, the fact that Φ​(α)\Phi(\alpha) is a coboundary is proven in [FU14, Thm 4.2] (see also [FH18, Thm 4.6]). It is left to check that the transfer function hh is in fact continuous, not just measurable. As shown in [CE15, Section 1.2.2], the IET TT is of Roth type, hence we can apply ˜2.7 to conclude that hh is continuous.

Consider now the case that (M,ω)(M,\omega) is gtg_{t}-periodic, with gT0​ω=ωg_{T_{0}}\omega=\omega. Then (M,ω)(M,\omega) is automatically Oseledets and Masur generic. Indeed in this case Oseledets genericity is immediate, as the Kontsevich-Zorich cocycle, at times which are multiples of T0T_{0}, is given by a constant matrix. Masur genericity in the periodic case is also classical, following from Masur’s criterion ([Mas92]).

Thus the only assumption that is not satisfied is Birkhoff genericity. But in the proof as written in [FH18, Thm 4.5], Birkhoff genericity is used only to show that for some compact section 𝒦\mathcal{K}, the consecutive hitting times (tk)k≥0(t_{k})_{k\geq 0} by the Teichmüller flow (gt​ω)t∈ℝ(g_{t}\omega)_{t\in\mathbb{R}} of 𝒦\mathcal{K} satisfy tkk→0\frac{t_{k}}{k}\to 0. But in the periodic case, if the closed geodesic enters 𝒦\mathcal{K} nn times, then tn+k=tk+T0t_{n+k}=t_{k}+T_{0}, and so the limit limk→∞tkk=0\lim_{k\to\infty}\frac{t_{k}}{k}=0 is satisfied. The rest of the proof carries through without alterations.

Finally, all periodic type IETs are Roth type (see [MMY05, Section 1.3.4]), so here we can also apply ˜2.7 to conclude. ∎

3. Covers

For a manifold MM, a ℤd\mathbb{Z}^{d}-cover of MM is a covering map p:M~→Mp:\tilde{M}\to M equipped with a Deck group action by ℤd\mathbb{Z}^{d}. We will denote the cover simply by M~\tilde{M}, with the covering map and the action of ℤd\mathbb{Z}^{d} implicit.

Denote by ⟨⋅,⋅⟩\langle\cdot,\cdot\rangle the intersection form on H1​(M,ℝ)H_{1}(M,\mathbb{R}).

Lemma 3.1 (See [HW12] or Section 2 in [FU14]).

The ℤd\mathbb{Z}^{d}-covers of MM are in one-to-one correspondence with H1​(M,ℤ)dH_{1}(M,\mathbb{Z})^{d}, in such a way that if we denote the cover corresponding to γ∈H1​(M,ℤ)d\gamma\in H_{1}(M,\mathbb{Z})^{d} by M~γ\tilde{M}_{\gamma}, then M~γ\tilde{M}_{\gamma} has the following property:

If σ\sigma is a closed curve on MM and

then any lift of σ\sigma, σ~:[t0,t1]→M~γ\tilde{\sigma}:[t_{0},t_{1}]\to\tilde{M}_{\gamma} satisfies

where ⋅\cdot denotes the Deck group action of ℤd\mathbb{Z}^{d}.

Given a compact translation surface (M,ω)(M,\omega) and a ℤd\mathbb{Z}^{d}-cover M~γ\tilde{M}_{\gamma} of MM, the pullback ω~γ=p∗​ω\tilde{\omega}_{\gamma}=p^{*}\omega determines a translation structure on M~\tilde{M}. We say that (M~γ,ω~γ)(\tilde{M}_{\gamma},\tilde{\omega}_{\gamma}) is a ℤd\mathbb{Z}^{d}-cover of the translation surface (M,ω)(M,\omega).

We consider the vertical flow (φ~t)t∈ℝv(\tilde{\varphi}_{t})^{v}_{t\in\mathbb{R}} on (M~γ,ω~γ)(\tilde{M}_{\gamma},\tilde{\omega}_{\gamma}).

As a consequence of Lemma˜3.1, the following lemma holds. Let I~=p−1​(I)≅I×ℤd\tilde{I}=p^{-1}(I)\cong I\times\mathbb{Z}^{d}. Denote by T~:I×ℤd→I×ℤd\tilde{T}:I\times\mathbb{Z}^{d}\to I\times\mathbb{Z}^{d} the Poincaré map of (φ~t)t∈ℝv(\tilde{\varphi}_{t})^{v}_{t\in\mathbb{R}} to I~\tilde{I}.

Lemma 3.2 (See Lemma 2.1 in [FU14]).

The map T~\tilde{T} is given by a skew-product T~=Tϕγ:I×ℤd→I×ℤd\tilde{T}=T_{\phi_{\gamma}}:I\times\mathbb{Z}^{d}\to I\times\mathbb{Z}^{d}, where

3.1. The wind-tree model

In this section we define the wind-tree model and discuss the splitting of its homology, following [DHL14].

Fix parameters a,b∈(0,1)a,b\in(0,1). The wind-tree model with parameters a,ba,b is defined as

Unfolding the billiard in W​(a,b)W(a,b), one gets a translation surface X∞=X∞​(a,b)X_{\infty}=X_{\infty}(a,b), which is a ℤ2\mathbb{Z}^{2}-cover as described below.

Let Y=Y​(a,b)Y=Y(a,b) be a genus two translation surface defined by gluing the opposite edges of the square with a rectangular hole([−12,12]×[−12,12])\([−a2,a2]×[−b2,b2])\left(\big[-\frac{1}{2},\frac{1}{2}\big]\times\big[-\frac{1}{2},\frac{1}{2}\big]\right)\,\big\backslash\,\left(\big[-\frac{a}{2},\frac{a}{2}\big]\times\big[-\frac{b}{2},\frac{b}{2}\big]\right).

cvc_{v}chc_{h}vvhh

Figure 3. The surface YY with the gluings and curves v,h,chv,h,c_{h} and cvc_{v}.

Let h,v,chh,v,c_{h} and cvc_{v} be the four curves on YY as in Figure˜3. Then let X=X​(a,b)X=X(a,b) be the ℤ2×ℤ2\mathbb{Z}^{2}\times\mathbb{Z}^{2} cover of YY corresponding to the homology classes ([ch]−[h],[cv]−[v])∈H1​(Y,ℤ/2​ℤ)2([c_{h}]-[h],[c_{v}]-[v])\in H_{1}(Y,\mathbb{Z}/2\mathbb{Z})^{2}. In Figure˜4 we show XX with its four singularities.

h00h_{00}h00h_{00}v00v_{00}v00v_{00}cv,0c_{v,0}ch,0c_{h,0}h10h_{10}h10h_{10}v10v_{10}v10v_{10}cv,1c_{v,1}h11h_{11}h11h_{11}v11v_{11}v11v_{11}h01h_{01}h01h_{01}v01v_{01}v01v_{01}ch,1c_{h,1}

Figure 4. The surface XX with its four singularities and the classes generating H1​(X,ℤ)H_{1}(X,\mathbb{Z}).

Define the homology classes hx​y,vx​y∈H1​(X,ℤ)h_{xy},v_{xy}\in H_{1}(X,\mathbb{Z}) for x,y∈{0,1}x,y\in\{0,1\} as in Figure˜4. If we define

then X∞X_{\infty} is the ℤ2\mathbb{Z}^{2}-cover arising from the classes (γh,γv)(\gamma_{h},\gamma_{v}).

One computes that XX has genus 5, and so H1​(X,ℤ)≅ℤ10H_{1}(X,\mathbb{Z})\cong\mathbb{Z}^{10}. The twelve homology classes hx​y,vx​y,ch,y,cv,xh_{xy},v_{xy},c_{h,y},c_{v,x} for x,y∈{0,1}x,y\in\{0,1\} generate H1​(X,ℤ)H_{1}(X,\mathbb{Z}) with the following two relations:

Now consider the Deck group action of the Klein group ℤ/2​ℤ×ℤ/2​ℤ\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/2\mathbb{Z} on XX, seen as a cover of YY. If we let τh\tau_{h} and τv\tau_{v} be the generators of the two ℤ/2​ℤ\mathbb{Z}/2\mathbb{Z} factors, they act on XX by permuting the four sheets. In particular, the induced action on homology is given by:

and symmetrically for τv\tau_{v}.

Since the action of KK commutes with the S​L​(2,ℤ)SL(2,\mathbb{Z}) action on ℳ1​(X)\mathcal{M}_{1}(X), H1​(X,ℝ)H_{1}(X,\mathbb{R}) splits into four invariant blocks,

where

These blocks have the following generators:

Each of these four blocks is symplectic, and the Lyapunov exponents for the Kontsevich-Zorich cocycle over the corresponding invariant subbundle are respectively {1,λ++,−λ++,−1}\{1,\lambda^{++},-\lambda^{++},-1\}, {λ+−,−λ+−}\{\lambda^{+-},-\lambda^{+-}\}, {λ−+,−λ−+}\{\lambda^{-+},-\lambda^{-+}\}, {λ−−,−λ−−}\{\lambda^{--},-\lambda^{--}\}.

Theorem 3.3 (See Corollary 1 in [DHL14]).

For any a,b∈(0,1)a,b\in(0,1), for almost all θ∈S1\theta\in S^{1}, the following holds.

Let 𝒩\mathcal{N} be the S​L​(2,ℝ)SL(2,\mathbb{R}) orbit closure of rπ/2−θ​X​(a,b)r_{\pi/2-\theta}X(a,b) and ν𝒩\nu_{\mathcal{N}} the corresponding invariant probability measure. Then the Lyapunov exponents for the Kontsevich-Zorich cocycle with respect to ν𝒩\nu_{\mathcal{N}} are given by λ++=λ−−=1/3\lambda^{++}=\lambda^{--}=1/3 and λ+−=λ−+=2/3\lambda^{+-}=\lambda^{-+}=2/3.

4. Construction of invariant sets and non-transitivity

Theorem 4.1.

Suppose that (M,ω)(M,\omega) is either BOM generic or gtg_{t}-periodic. Suppose that H1​(M,ℤ)=K⊕K⟂H_{1}(M,\mathbb{Z})=K\oplus K^{\perp} is an orthogonal splitting with respect to the symplectic form, and that V=K⊗ℝV=K\otimes\mathbb{R} defines an S​L2​(ℝ)SL_{2}(\mathbb{R})-invariant subbundle 𝒱\mathcal{V} of the homology Hodge bundle of (M,ω)(M,\omega). Let 2​d=dimK2d=\dim K, and let d+d_{+} be the number of positive Lyapunov exponents of 𝒱\mathcal{V}. Assume that d+>0d_{+}>0. Let m=2​d−d+m=2d-d_{+} be the number of non-positive Lyapunov exponents, and let γ1,…,γm∈K\gamma_{1},\dots,\gamma_{m}\in K be any linearly independent classes. Let (M~γ,ω~γ)(\tilde{M}_{\gamma},\tilde{\omega}_{\gamma}) be the ℤm\mathbb{Z}^{m}-cover corresponding to γ=(γ1,…,γm)\gamma=(\gamma_{1},\dots,\gamma_{m}).

Then there exists a lattice Λ⊂ℝd+\Lambda\subset\mathbb{R}^{d_{+}} and a non-constant continuous function h^:M~γ→ℝd+/Λ\hat{h}:\tilde{M}_{\gamma}\to\mathbb{R}^{d_{+}}/\Lambda, which is invariant for the vertical flow φ~tv\tilde{\varphi}^{v}_{t} on (M~γ,ω~γ)(\tilde{M}_{\gamma},\tilde{\omega}_{\gamma}).

Proof.

Let II be a horizontal Poincaré section for the vertical flow on (M,ω)(M,\omega) satisfying the conclusion of ˜2.8, and let T:I→IT:I\to I be the Poincaré map. Let ϕ:I→ℤm\phi:I\to\mathbb{Z}^{m} be the piecewise constant function given by ϕ=(Φ​(γ1),…,Φ​(γm)).\phi=(\Phi(\gamma_{1}),\dots,\Phi(\gamma_{m})).By Lemma˜3.2 the vertical flow on (M~γ,ω~γ)(\tilde{M}_{\gamma},\tilde{\omega}_{\gamma}) is a special flow over the skew-product TϕT_{\phi}, hence to construct an invariant function for the vertical flow we construct an invariant function for TϕT_{\phi} on I×ℤmI\times\mathbb{Z}^{m} and then extend it to M~γ\tilde{M}_{\gamma} along vertical leaves.

As the numbers of positive and negative exponents are the same by the symplectic structure of the cocycle, the space of stable vectors in VV, V∩Eω−V\cap E^{-}_{\omega} is d+d_{+}-dimensional. Let α1,…,αd+\alpha_{1},\dots,\alpha_{d_{+}} be a basis for this stable subspace of VV.

Extend the set {γ1,…,γm}\{\gamma_{1},\dots,\gamma_{m}\} to a basis {γ1,…,γm,σ1,…,σd+}\{\gamma_{1},\dots,\gamma_{m},\sigma_{1},\dots,\sigma_{d_{+}}\} of KK.

Express the αi\alpha_{i} in terms of this basis as

with bi​j,ci​k∈ℝb_{ij},c_{ik}\in\mathbb{R}.

Since the αi\alpha_{i} are stable, by ˜2.8 the functions ψi=Φ​(αi)\psi_{i}=\Phi(\alpha_{i}) are coboundaries, i.e. there exist continuoush1,…,hd+:I→ℝh_{1},\dots,h_{d_{+}}:I\to\mathbb{R} such that

Now define the function h~:I×Zm→ℝd+\tilde{h}:I\times Z^{m}\to\mathbb{R}^{d_{+}}, where for a=(a1,…,am)∈ℤma=(a_{1},\dots,a_{m})\in\mathbb{Z}^{m},

Then observe that

Now note that since σ∈H1​(M,ℤ)\sigma\in H_{1}(M,\mathbb{Z}), Φ​(σk)​(x)∈ℤ\Phi(\sigma_{k})(x)\in\mathbb{Z} for all k,xk,x.

Hence if C∈Md+×d+C\in M_{d_{+}\times d_{+}} is the matrix with entries ci​kc_{ik}, we see that for all (x,a)∈I×ℤm(x,a)\in I\times\mathbb{Z}^{m},

Let Λ=C⋅ℤd+\Lambda=C\cdot\mathbb{Z}^{d_{+}}, which is a lattice in ℝd+\mathbb{R}^{d_{+}}. Then if we define h^:I×Zm→ℝd+/Λ\hat{h}:I\times Z^{m}\to\mathbb{R}^{d_{+}}/\Lambda by composing h~\tilde{h} with the quotient map, h^\hat{h} is a continuous function and it is TϕT_{\phi}-invariant.

Finally, let us show that h^\hat{h} can not be constant on I×ℤmI\times\mathbb{Z}^{m}. Indeed, suppose h^\hat{h} is constant. Then since h~\tilde{h} is continuous, also h~\tilde{h} must be constant. By the definition of h~\tilde{h}, considering a fixed a∈ℤma\in\mathbb{Z}^{m}, we see that every hih_{i} must be constant on II. But this implies that αi=0\alpha_{i}=0, which contradicts our assumption. Thus we conclude that h^\hat{h} is a non-constant, continuous, TϕT_{\phi}-invariant function on I×ℤmI\times\mathbb{Z}^{m}.

If we extend h^\hat{h} to M~γ\tilde{M}_{\gamma}, by defining, for every x~∈I\tilde{x}\in I, the value on the vertical segment joining x~\tilde{x} and Tϕ​(x~)T_{\phi}(\tilde{x}) to be equal to the value at x~\tilde{x}, we obtain a non-constant continuous function on M~γ\tilde{M}_{\gamma} that is invariant under the vertical flow. ∎

Corollary 4.2.

Under the assumptions of ˜4.1, the vertical flow on (M~γ,ω~γ)(\tilde{M}_{\gamma},\tilde{\omega}_{\gamma}) is non-transitive.

Proof.

Observe that by ˜4.1, for any (x,a)∈I×ℤm(x,a)\in I\times\mathbb{Z}^{m},

As h^\hat{h} is continuous, in fact the level set h^−1​(h^​(x,a))\hat{h}^{-1}(\hat{h}(x,a)) is closed. Since h^\hat{h} is not constant, for every (x,a)(x,a) this level set is not all of I×ℤmI\times\mathbb{Z}^{m}, and so no orbit of TϕT_{\phi} is dense. Hence also no orbit of φ~tv\tilde{\varphi}^{v}_{t} is dense on (M~γ,ω~γ)(\tilde{M}_{\gamma},\tilde{\omega}_{\gamma}). ∎

5. Hausdorff dimension in the periodic case

Here we consider the case where the IET TT is of periodic type. Let NN be such that ℛN​(T)=T\mathcal{R}^{N}(T)=T, and the corresponding matrix A=A(0,N)A=A^{(0,N)} is positive. Denote by TkT_{k} the induced IET ℛ^N​k​(T)\hat{\mathcal{R}}^{Nk}(T), defined on an interval that we call J(k)J^{(k)}. Denote by x0x_{0} the left endpoint of II, which is also the left end-point of J(k)J^{(k)}.

Denote by Zj:=Zj(0,N)Z_{j}:=Z_{j}^{(0,N)} the Rokhlin tower for TT over Jj(1)J_{j}^{(1)}, let qjq_{j} be its height. By the periodicity, ZjZ_{j} is naturally identified with the Rokhlin tower for Tk−1T_{k-1} over Jj(k)J^{(k)}_{j} for any kk.

Suppose that ψ\psi is an eigenvector of ATA^{T} with eigenvalue λ<1\lambda<1. We denote by ψ\psi also the piecewise constant function ψ​(x)=ψα\psi(x)=\psi_{\alpha} for x∈Iαx\in I_{\alpha}. Let hh be the continuous transfer function satisfying ψ​(x)=h​(T​(x))−h​(x)\psi(x)=h(T(x))-h(x), we assume without loss of generality that h​(x0)=0h(x_{0})=0.

Lemma 5.1.

For x∈Jj(1)x\in J_{j}^{(1)}, h​(T1​(x))−h​(x)=λ​ψjh(T_{1}(x))-h(x)=\lambda\psi_{j}.

Proof.

Observe that T1​(x)=Tqj​(x)T_{1}(x)=T^{q_{j}}(x), hence

where in the third equality we used Lemma˜2.6. ∎

5.1. The Vershik adic coding for IETs

Here we discuss the Vershik coding for IETs, introduced in [Buf13] and [LT16]. We give only the brief definitions necessary for the following, for a more detailed explanation and proofs see the above references or [Tum25].

The idea is to code a point x∈Ix\in I by the floor of the tower Zjk(0,N​k)Z_{j_{k}}^{(0,Nk)} in which it appears. Taking larger kk gives a finer partition of II, and recording the floor for all k∈ℕk\in\mathbb{N} allows one to identify xx uniquely.

Let ℰ\mathcal{E} be the set of all floors of the towers {Zj:j∈𝒜}\{Z_{j}:j\in\mathcal{A}\}, where we denote the floor Tℓ​(Jj(1))T^{\ell}(J^{(1)}_{j}) of ZjZ_{j} by (j,ℓ)(j,\ell):

The set ℰ\mathcal{E} is the set of edges in a graph called the Bratteli diagram, hence by convention we will sometimes call the elements of ℰ\mathcal{E} edges, and talk about their start s​(e)s(e) and terminus t​(e)t(e).

For an edge e=(j,l)e=(j,l) we define t​(e)=jt(e)=j and s​(e)=is(e)=i, where i∈𝒜i\in\mathcal{A} is such that Ii⊃Tℓ​(Jj(1))I_{i}\supset T^{\ell}(J^{(1)}_{j}). Importantly, by periodicity, also for any k≥1k\geq 1, Ji(k−1)⊃Tk−1ℓ​(Jj(k))J^{(k-1)}_{i}\supset T_{k-1}^{\ell}(J^{(k)}_{j}).

Denote by Σk⊂ℰk\Sigma_{k}\subset\mathcal{E}^{k} the admissible paths of length kk, i.e. sequences (e1,…,ek)(e_{1},\dots,e_{k}) such that t​(e1)=s​(e2),…,t​(ek−1)=s​(ek)t(e_{1})=s(e_{2}),\dots,t(e_{k-1})=s(e_{k}).

Given a path pk=((j1,ℓ1),…,(jk,ℓk))∈Σkp_{k}=((j_{1},\ell_{1}),\dots,(j_{k},\ell_{k}))\in\Sigma_{k}, we define a corresponding subinterval of II, by

Lemma 5.2 (See Lemma 4.7 and Lemma 4.8 in [Tum25]).

1. (1)
   The map 𝒥\mathcal{J} is a bijection between the set Σk\Sigma_{k} of admissible paths of length kk and floors of the towers for TT over J(k)J^{(k)}.
2. (2)
   If pk+1∈Σk+1p_{k+1}\in\Sigma_{k+1} and pkp_{k} is the path consisting of the first kk entries of pk+1p_{k+1}, then 𝒥​(pk+1)⊂𝒥​(pk)\mathcal{J}(p_{k+1})\subset\mathcal{J}(p_{k}).

Hence if Σ\Sigma denotes the space of infinite admissible paths,

and denoting by pk∈Σkp_{k}\in\Sigma_{k} the first kk entries of p∈Σp\in\Sigma, one can define the map 𝒥:Σ→I\mathcal{J}:\Sigma\to I by

The map 𝒥\mathcal{J} is surjective, it is 2-to-1 on countably many sequences in Σ\Sigma and one-to-one everywhere else.

5.2. Computing the transition function hh using the coding

Define the function f:ℰ→ℝf:\mathcal{E}\to\mathbb{R} to be the partial Birkhoff sum of ψ\psi from the base of the tower ZjZ_{j} until the floor corresponding to the edge (j,ℓ)(j,\ell):

| f​((j,ℓ)):=SℓT​ψ|Ij.f((j,\ell)):=S^{T}_{\ell}\psi|_{I_{j}}. | | ------------------------------------------------------------------ |

Proposition 5.3.

If a point xx has coding x=𝒥​(p)x=\mathcal{J}(p) for p=(e1,e2,…)∈Σp=(e_{1},e_{2},\dots)\in\Sigma, the value of hh at xx is given by the series

Proof.

Consider the finite truncations pk=(e1,…,ek)p_{k}=(e_{1},\dots,e_{k}) for k≥1k\geq 1. Let xkx_{k} be the left endpoint of the interval in II corresponding to pkp_{k}. Since 𝒥​(pk)\mathcal{J}(p_{k}) is a floor of the tower above Jjk(k)J^{(k)}_{j_{k}}, the point xkx_{k} is in the orbit of the left endpoint yjk(k)y^{(k)}_{j_{k}} of Jjk(k)J^{(k)}_{j_{k}},

Denote by xk,rx_{k,r} the point xk,r=Trℓr+1∘Tr+1ℓr+2∘⋯∘Tk−1ℓk​(yjk(k))x_{k,r}=T_{r}^{\ell_{r+1}}\circ T_{r+1}^{\ell_{r+2}}\circ\dots\circ T_{k-1}^{\ell_{k}}(y^{(k)}_{j_{k}}), so that xk,0=xkx_{k,0}=x_{k}, andxk,k=yjk(k)x_{k,k}=y^{(k)}_{j_{k}}. As xk,i=Tiℓi+1​(xk,i+1)x_{k,i}=T_{i}^{\ell_{i+1}}(x_{k,i+1}), applying Lemma˜5.1, we see that

| h​(xk,i)−h​(xk,i+1)=Sℓi+1T​λi​ψ|Iji+1=λi​f​(ei+1).h(x_{k,i})-h(x_{k,i+1})=S^{T}_{\ell_{i+1}}\lambda^{i}\psi|_{I_{j_{i+1}}}=\lambda^{i}f(e_{i+1}). | | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- |

Hence

Now observe that limk→∞xk,k=x0\lim_{k\to\infty}x_{k,k}=x_{0}, and so limk→∞h​(xk,k)=0\lim_{k\to\infty}h(x_{k,k})=0. Similarly, as limk→∞xk=x\lim_{k\to\infty}x_{k}=x,

∎

5.3. Bounding the Hausdorff dimension

Lemma 5.4.

By taking a multiple of the period NN if needed, we can assume that for any i,ji,j, there exist two edges e1,e2∈ℰe_{1},e_{2}\in\mathcal{E} with s​(er)=i,t​(er)=js(e_{r})=i,t(e_{r})=j for r=1,2r=1,2, and f​(e1)≠f​(e2)f(e_{1})\neq f(e_{2}).

Proof.

Fix a letter ii. Note that hh can not be constant on IiI_{i}, since then it would be piecewise constant everywhere, and since it is continuous it would have to be constant on II, which is not possible as ψ≠0\psi\neq 0. Then there must exist two subintervals Ui,1U_{i,1} and Ui,2U_{i,2} of IiI_{i}, such that h​(Ui,1)∩h​(Ui,2)=∅h(U_{i,1})\cap h(U_{i,2})=\emptyset.

By minimality of TT, there are points yi,1∈Ui,1y_{i,1}\in U_{i,1} and yi,2∈Ui,2y_{i,2}\in U_{i,2}, such that yi,2=Tk​(yi,1)y_{i,2}=T^{k}(y_{i,1}) for some kk. Since by assumption h​(Ui,1)∩h​(Ui,2)=∅h(U_{i,1})\cap h(U_{i,2})=\emptyset, it must hold that SkT​ψ​(yi,1)≠0S^{T}_{k}\psi(y_{i,1})\neq 0, since it is equal to the difference h​(yi,2)−h​(yi,1)h(y_{i,2})-h(y_{i,1}).

Let Ji,1J_{i,1} be the interval containing yi,1y_{i,1} on which TkT^{k} is continuous, and let Ji,2=Tk​(Ji,1)J_{i,2}=T^{k}(J_{i,1}). Take NiN_{i} to be a sufficiently high multiple of the period NN such that, for each jj, in the tower Zj(0,Ni)Z^{(0,N_{i})}_{j} there is a floor which is a subinterval of Ji,1J_{i,1} and kk floors above it a floor that is a subinterval of Ji,2J_{i,2}. Then these two floors correspond to two edges e1,e2e_{1},e_{2} with start ii and end jj, and f​(e2)−f​(e1)=Sk​ψ​(yi,1)≠0f(e_{2})-f(e_{1})=S_{k}\psi(y_{i,1})\neq 0. Doing this for each ii and taking N=maxi⁡NiN=\max_{i}N_{i} will give an NN that works for all ii and jj. ∎

For fixed i,ji,j, call the two edges e1,e2e_{1},e_{2} as above an alternate pair for i,ji,j. From now on we assume that NN is large enough to satisfy the conclusion of Lemma˜5.4. For each i,ji,j, let δi​j\delta_{ij} be the difference |f​(e2)−f​(e1)|>0|f(e_{2})-f(e_{1})|>0 for the alternate pair e1,e2e_{1},e_{2}. Let δ=mini,j⁡δi​j\delta=\min_{i,j}\delta_{ij}.

Let F=max(e,e′)∈ℰ2⁡|f​(e)−f​(e′)|F=\max_{(e,e^{\prime})\in\mathcal{E}^{2}}|f(e)-f(e^{\prime})| and let bb be an integer such that

Lemma 5.5.

Let (e1,e1′)(e_{1},e^{\prime}_{1}) be an alternate pair, and let (e1,…,eb)(e_{1},\dots,e_{b}) be an admissible finite path. Then h​(𝒥​(e1,e2,…,eb))∩h​(𝒥​(e1′,e2,…,eb))=∅h(\mathcal{J}(e_{1},e_{2},\dots,e_{b}))\cap h(\mathcal{J}(e^{\prime}_{1},e_{2},\dots,e_{b}))=\emptyset.

Proof.

Observe that if p=(e~1,e~2,…),q=(e^1,e^2,…)p=(\tilde{e}_{1},\tilde{e}_{2},\dots),q=(\hat{e}_{1},\hat{e}_{2},\dots) are two paths such that e~i=e^i\tilde{e}_{i}=\hat{e}_{i} for i=1,…,bi=1,\dots,b, then

| |h​(p)−h​(q)|≤F​∑i=b+1∞λi−1≤F​λb1−λ<δ.|h(p)-h(q)|\leq F\sum_{i={b+1}}^{\infty}\lambda^{i-1}\leq\frac{F\lambda^{b}}{1-\lambda}<\delta. | | ------------------------------------------------------------------------------------------------------------------------------------------------- |

Now take any path p∈Σp\in\Sigma starting with the finite path (e1,e2,…,eb)(e_{1},e_{2},\dots,e_{b}) and any path p′∈Σp^{\prime}\in\Sigma starting with (e1′,e2,…,eb)(e^{\prime}_{1},e_{2},\dots,e_{b}). Then if qq is the path obtained from p′p^{\prime} by replacing the first edge e1′e^{\prime}_{1} with e1e_{1}, then |h​(q)−h​(p′)|=|f​(e1)−f​(e1′)|>δ|h(q)-h(p^{\prime})|=|f(e_{1})-f(e^{\prime}_{1})|>\delta, whereas |h​(p)−h​(q)|<δ|h(p)-h(q)|<\delta. Thus h​(p)≠h​(p′)h(p)\neq h(p^{\prime}), and henceh​(𝒥​(e1,…,eb))∩h​(𝒥​(e1′,e2,…,eb))=∅h(\mathcal{J}(e_{1},\dots,e_{b}))\cap h(\mathcal{J}(e^{\prime}_{1},e_{2},\dots,e_{b}))=\emptyset. ∎

In particular, for any given value zz, at least one of the two intervals 𝒥​(e1,…,eb)\mathcal{J}(e_{1},\dots,e_{b}) and 𝒥​(f1,e2,…,eb)\mathcal{J}(f_{1},e_{2},\dots,e_{b}) must be in the complement of h−1​(z)h^{-1}(z). This allows us to find gaps in the level set h−1​(z)h^{-1}(z).

Theorem 5.6.

Let T,hT,h as above. Then for any z∈ℝz\in\mathbb{R}, the Hausdorff dimension of the level set h−1​(z)h^{-1}(z) is strictly less than 1.

Proof.

Fix some z∈ℝz\in\mathbb{R}. We will show that h−1​(z)h^{-1}(z) is contained in a Cantor set of Hausdorff dimension less than 1. As shown inLemma˜5.5, picking any alternate pair (e1,f1)(e_{1},f_{1}) and admissible finite path (e1,…,eb)(e_{1},\dots,e_{b}) gives us a gap in h−1​(z)h^{-1}(z), as either𝒥​(e1,e2,…,eb)\mathcal{J}(e_{1},e_{2},\dots,e_{b}) or 𝒥​(f1,e2,…,eb)\mathcal{J}(f_{1},e_{2},\dots,e_{b}) must be in the complement of the level set.

For any k≥1k\geq 1, recall that we can partition II as I=⊔p∈Σk​b𝒥​(p)I=\sqcup_{p\in\Sigma_{kb}}\mathcal{J}(p), where Σk​b\Sigma_{kb} are the finite admissible paths of length k​bkb. For any p=(e1,e2,…,ek​b)∈Σk​bp=(e_{1},e_{2},\dots,e_{kb})\in\Sigma_{kb}, there exists an alternate pair (ek​b+1,fk​b+1)(e_{kb+1},f_{kb+1}) and an admissible path (ek​b+1,ek​b+2,…,e(k+1)​b)(e_{kb+1},e_{kb+2},\dots,e_{(k+1)b}). To this pair corresponds as above a gap in h−1​(z)h^{-1}(z), which we will denote by GpG_{p}. Thus there is a gap Gp⊂I∖h−1​(z)G_{p}\subset I\setminus h^{-1}(z) for every element pp of the partition. The choice of alternate pair and admissible path that gives a gap is very far from unique, but we just need to find one. We will construct our Cantor set CC inductively as C=∩k≥1CkC=\cap_{k\geq 1}C_{k}, where Ck+1C_{k+1} is obtained from CkC_{k} by removing the gap GpG_{p} from every element of the partition I=⊔p∈Σk​b𝒥​(p)I=\sqcup_{p\in\Sigma_{kb}}\mathcal{J}(p), which is still contained in CkC_{k}.

More formally, for k=0k=0, denote by GiG_{i} a gap inside IiI_{i}, given by an alternate pair (e1,e1′)(e_{1},e^{\prime}_{1}) with s​(e1)=s​(e1′)=is(e_{1})=s(e^{\prime}_{1})=i, and some admissible path (e1,…,eb)(e_{1},\dots,e_{b}). Let C1=I∖∪i∈𝒜GiC_{1}=I\setminus\cup_{i\in\mathcal{A}}G_{i}. For k≥1k\geq 1, let 𝒫k\mathcal{P}_{k} be the set of admissible paths p=(e1,…,ek​b)p=(e_{1},\dots,e_{kb}) such that 𝒥​(p)⊂Ck\mathcal{J}(p)\subset C_{k}, and defineCk+1=Ck∖∪p∈𝒫kGpC_{k+1}=C_{k}\setminus\cup_{p\in\mathcal{P}_{k}}G_{p}. Let C=∩k≥1CkC=\cap_{k\geq 1}C_{k}. Then h−1​(z)⊂Ch^{-1}(z)\subset C.

Now let us show that the Hausdorff dimension of CC is strictly less than 1. First let us bound the number of gaps in CkC_{k}. Let m=maxi​j⁡Ai​jbm=\max_{ij}A^{b}_{ij}. Then the number of all admissible paths of length k​bkb is bounded above by n​mknm^{k}, where n=|𝒜|n=|\mathcal{A}|. Hence #​𝒫k≤n​mk\#\mathcal{P}_{k}\leq nm^{k}, and so the number of gaps in CkC_{k} is bounded above by n​∑r=0k−1mr<c​mkn\sum_{r=0}^{k-1}m^{r}<cm^{k} for some uniform constant cc.

To bound the Hausdorff dimension of CC from above, we consider for each kk the cover of CC given by taking the set of intervals that are the connected components of CkC_{k}. Denote the lengths of these intervals by ℓk,1,…,ℓk,Nk\ell_{k,1},\dots,\ell_{k,N_{k}}, where Nk<c​mkN_{k}<cm^{k} is their number. Observe that the mesh of this cover goes to 0 as kk goes to infinity, since ℓk,j≤2​maxi∈𝒜⁡|Ji(k)|\ell_{k,j}\leq 2\max_{i\in\mathcal{A}}|J^{(k)}_{i}|.

If μ=mini∈𝒜⁡|Ii|maxi∈𝒜⁡|Ii|\mu=\frac{\min_{i\in\mathcal{A}}|I_{i}|}{\max_{i\in\mathcal{A}}|I_{i}|}, then for any p∈𝒫k−1p\in\mathcal{P}_{k-1}, |Gp|≥m−1​μ​|𝒥​(p)||G_{p}|\geq m^{-1}\mu|\mathcal{J}(p)|, which means that at every step the total length of gaps that we remove has proportion at least m−1​μm^{-1}\mu of the length of CkC_{k} and hence by induction, for all kk,

| ∑i=1Nkℓk,i=|Ck|≤(1−m−1​μ)k.\sum_{i=1}^{N_{k}}\ell_{k,i}=|C_{k}|\leq(1-m^{-1}\mu)^{k}. | | ------------------------------------------------------------------------------------------------ |

Now take β<1\beta<1. Then by the power means inequality,

| (∑i=1Nkℓk,iβNk)1β≤∑i=1Nkℓk,iNk=|Ck|Nk,\left(\frac{\sum_{i=1}^{N_{k}}\ell_{k,i}^{\beta}}{N_{k}}\right)^{\frac{1}{\beta}}\leq\frac{\sum_{i=1}^{N_{k}}\ell_{k,i}}{N_{k}}=\frac{|C_{k}|}{N_{k}}, | | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |

and hence

| ∑i=1Nkℓk,iβ≤|Ck|β​Nk1−β≤c​((1−m−1​μ)β​m1−β)k.\sum_{i=1}^{N_{k}}\ell_{k,i}^{\beta}\leq|C_{k}|^{\beta}N_{k}^{1-\beta}\leq c\left((1-m^{-1}\mu)^{\beta}m^{1-\beta}\right)^{k}. | | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |

From this we deduce that if β0<1\beta_{0}<1 is such that (1−m−1​μ)β0​m1−β0=1(1-m^{-1}\mu)^{\beta_{0}}m^{1-\beta_{0}}=1, then for β>β0\beta>\beta_{0}, the β\beta-dimensional Hausdorff measure of CC is 0, and hence we deduce that the Hausdorff dimension of CC is at most β0\beta_{0}. ∎

Corollary 5.7.

Suppose that (M,ω)(M,\omega), γ\gamma satisfy the conditions of the statement of ˜4.1, with ω\omega gtg_{t}-periodic. Then for every x~∈M~γ\tilde{x}\in\tilde{M}_{\gamma}, the closure of the orbit of x~\tilde{x} under the vertical flow has Hausdorff dimension strictly less than 2.

Proof.

By assumption, Gt𝒱G_{t}^{\mathcal{V}} has some negative Lyapunov exponents. Hence there exists a class α1∈K⊗ℝ\alpha_{1}\in K\otimes\mathbb{R} and a λ<0\lambda<0, such that Gt0𝒱​α1=eλ​α1G^{\mathcal{V}}_{t_{0}}\alpha_{1}=e^{\lambda}\alpha_{1}, where t0t_{0} is the length of the closed Teichmüller geodesic containing (M,ω)(M,\omega), i.e. such that gt0​(M,ω)=(M,ω)g_{t_{0}}(M,\omega)=(M,\omega).

Let T:I→IT:I\to I be a Poincaré section for the vertical flow on MM as before. Then for some NN that corresponds to a multiple of the time t0t_{0}, we have ℛN​(T)=T\mathcal{R}^{N}(T)=T, and we can assume, by taking a multiple of t0t_{0} and NN if necessary, that the corresponding matrix AA is positive.

Then if ψ1\psi_{1} is the piecewise constant function on II defined by ψ1​(x)=⟨α1,ξ​(x)⟩\psi_{1}(x)=\langle\alpha_{1},\xi(x)\rangle, ψ1\psi_{1} when seen as a vector is an eigenvector of ATA^{T} with eigenvalue eλ<1e^{\lambda}<1. Let h1:I→ℝh_{1}:I\to\mathbb{R} be the transfer function corresponding to ψ1\psi_{1}. By ˜5.6 we know that for any z∈ℝz\in\mathbb{R}, h1−1​(z)h_{1}^{-1}(z) has Hausdorff dimension strictly less than 1.

Extend {α1}\{\alpha_{1}\} to a basis {α1,…,αd+}\{\alpha_{1},\dots,\alpha_{d_{+}}\} for the stable subspace of Gt𝒱G_{t}^{\mathcal{V}}, and define hih_{i} as in the proof of ˜4.1. In the notation of that theorem,

is invariant by TϕT_{\phi} up to an additive factor of Λ=C⋅ℤd+\Lambda=C\cdot\mathbb{Z}^{d_{+}}.

Consider a fixed a∈ℤma\in\mathbb{Z}^{m} and a fixed z=(z1,…,zd+)∈ℝd+z=(z_{1},\dots,z_{d_{+}})\in\mathbb{R}^{d_{+}}. Let us show that

Indeed, as each hih_{i} is bounded, (I×{a})∩h~−1​(z+Λ)(I\times\{a\})\cap\tilde{h}^{-1}(z+\Lambda) is the finite union

for some finite G⊂ΛG\subset\Lambda. As the Hausdorff dimension of a finite union of sets is the maximum of the individual dimensions, it is enough to show that for any zz,

Now since our aa is fixed, h~​(x,a)=z⇔(h1​(x),…,hd+​(x))=(z1′,…,zd+′)\tilde{h}(x,a)=z\iff(h_{1}(x),\dots,h_{d_{+}}(x))=(z^{\prime}_{1},\dots,z^{\prime}_{d_{+}}), for zi′=z+∑j=1mbi​j​ajz^{\prime}_{i}=z+\sum_{j=1}^{m}b_{ij}a_{j}. Hence in particular,

and by ˜5.6, the latter set has Hausdorff dimension strictly less than 1.

Thus we conclude by ˜4.1 that for any (x,a)∈I×ℤm(x,a)\in I\times\mathbb{Z}^{m},

and hence for any x~∈M~γ\tilde{x}\in\tilde{M}_{\gamma},

∎

5.4. Application to the wind-tree model

Proof of ˜1.1.

Recalling Section˜3.1, we see that for any parameters a,ba,b, and any θ\theta such that rπ/2−θ​X​(a,b)r_{\pi/2-\theta}X(a,b) is gtg_{t}-periodic and at least one of γh\gamma_{h} or γv\gamma_{v} is unstable, H1​(X,ℤ)H_{1}(X,\mathbb{Z}) has an orthogonal splitting satifying the conditions of ˜4.1. Indeed, without loss of generality suppose that γh\gamma_{h} is unstable. Then if K=E−+K=E^{-+} and K⟂=E++⊕E+−⊕E−−K^{\perp}=E^{++}\oplus E^{+-}\oplus E^{--}, then this is an orthogonal splitting. By the assumption that γh\gamma_{h} is unstable, KK has one positive and one negative Lyapunov exponent. Hence applying ˜5.7 for the ℤ\mathbb{Z}-cover of X​(a,b)X(a,b) given by γh\gamma_{h}, we deduce that the closure of any orbit of (φtθ)t∈ℝ(\varphi^{\theta}_{t})_{t\in\mathbb{R}} on X~​(a,b)γh\tilde{X}(a,b)_{\gamma_{h}} has Hausdorff dimension less than 2. Since X∞​(a,b)X_{\infty}(a,b) is a ℤ\mathbb{Z}-cover of X~​(a,b)γh\tilde{X}(a,b)_{\gamma_{h}}, orbit closures of (φtθ)t∈ℝ(\varphi^{\theta}_{t})_{t\in\mathbb{R}} on X∞​(a,b)X_{\infty}(a,b) are contained within the lifts of orbit closures on X~​(a,b)γh\tilde{X}(a,b)_{\gamma_{h}}, and hence also have Hausdorff dimension less than 2. Since X∞​(a,b)X_{\infty}(a,b) is the unfolding of the wind-tree model W​(a,b)W(a,b), we obtain the desired result. ∎

Appendix A Plotting invariant sets

Here we briefly describe some of the steps of plotting examples of the invariant sets given by ˜4.1, as in Figure˜1. We do this in the gtg_{t}-periodic case, where the computation can be done more rigorously.

The code used to generate the figure, using the surface-dynamics library in Sage ([CDD+25]) is contained in the notebook Invariant_sets_for_periodic_type_windtrees.ipynb, which is available on ArXiv, as well as on the website

https://yuriytumarkin.github.io/windtree/

The main steps are:

1. (0)
   Find parameters a,b,θa,b,\theta for which rπ/2−θ​X​(a,b)r_{\pi/2-\theta}X(a,b) is gtg_{t}-periodic.
2. (1)
   Find the stable vectors αi\alpha_{i} and compute the transfer functions hih_{i}.
3. (2)
   Find the level sets of hih_{i}.

0. Finding periodic parameters

One method to find periodic parameters is to look for loops in the Rauzy graph (see [SU05]). The Rauzy graph for IETs corresponding to the flow on XX is too large for this to be effective, but one can look for loops in the Rauzy graph corresponding to the flow on the genus 2 surface YY. If one find parameters a,b,θa,b,\theta such that rπ/2−θ​Y​(a,b)r_{\pi/2-\theta}Y(a,b) is gtg_{t}-periodic, then the fourfold cover rπ/2−θ​X​(a,b)r_{\pi/2-\theta}X(a,b) is also gtg_{t}-periodic, with possibly a longer period.

1. Finding stable vectors and computing the transfer function

In the periodic case, taking the matrix A=A(0,N)A=A^{(0,N)} corresponding to the period NN, the stable vectors can be found simply as the eigenvectors of ATA^{T} with eigenvalues less than 1, in the relevant blocks E+−E^{+-} and E−+E^{-+}. Knowing the stable vectors αi\alpha_{i} we can find the corresponding cocycles ψi\psi_{i}, which allows one to compute the transfer functions hih_{i} on the orbit of 0 using the cohomological equation h​(Tk​(0))=h​(0)+sk​ψ​(0)h(T^{k}(0))=h(0)+s_{k}\psi(0). However this isn’t enough to determine whether hih_{i} takes a certain value on a given interval or not.

2. Finding level sets

In the periodic case, one can explicitly compute the values of the function ff defined in Section˜5.2 and then use the expression with geometric tails as derived in ˜5.3 to get precise upper and lower bounds for hh on each interval IiI_{i}. Let 𝒫(1)\mathcal{P}^{(1)} be the dynamical partition of II into intervals, where each interval is a floor of some tower Zj,j∈𝒜Z_{j},j\in\mathcal{A}. Using the self-similarity of hh, this allows one to compute upper and lower bounds for hh on each interval belonging to 𝒫(1)\mathcal{P}^{(1)}, if one can compute the value of hh at its left endpoint. To do this, one can use the following observation:

Lemma A.1.

Let TT be a periodic type IET, with ℛN​(T)=T\mathcal{R}^{N}(T)=T. Let A=A(0,N)A=A^{(0,N)} be the corresponding Rauzy-Veech matrix. Suppose ψ\psi is an eigenvector of ATA^{T} with eigenvalue λ<1\lambda<1, and let hh be the corresponding continuous transfer function. Let 0=x0<x1<⋯<xn=10=x_{0}<x_{1}<\dots<x_{n}=1 be the discontinuities of TT. Then if the vector τ∈ℝn\tau\in\mathbb{R}^{n} is defined by

it satisfies A​τ=λ−1​τ.A\tau=\lambda^{-1}\tau.

Proof.

If the discontinuities of the induced map T1T_{1} are 0=x0(1)<x1(1)<⋯<xn(1)0=x_{0}^{(1)}<x_{1}^{(1)}<\dots<x_{n}^{(1)}, then by self-similarity the vector τ(1)\tau^{(1)} given by τi(1)=h​(xi(1))−h​(xi−1(1))\tau^{(1)}_{i}=h(x^{(1)}_{i})-h(x^{(1)}_{i-1}) is simply τ(1)=λ​τ\tau^{(1)}=\lambda\tau. Indeed, if (kr)r∈ℕ(k_{r})_{r\in\mathbb{N}} is an increasing sequence of integers such that Tkr​(0)→xiT^{k_{r}}(0)\to x_{i}, then T1kr​(0)→xi(1)T_{1}^{k_{r}}(0)\to x^{(1)}_{i}, and applying Lemma˜5.1 we see that h​(xi(1))=λ​h​(xi)h(x^{(1)}_{i})=\lambda h(x_{i}).

Since for each jj, the interval IiI_{i} consists of Ai​jA_{ij} floors of the tower ZjZ_{j}, and the difference between the value of hh at the left and right endpoints of each floor is τj(1)\tau^{(1)}_{j}, it follows that τi=∑j∈𝒜Ai​j​τj(1)\tau_{i}=\sum_{j\in\mathcal{A}}A_{ij}\tau^{(1)}_{j}, and thus τ=A​τ(1)=λ​A​τ.\tau=A\tau^{(1)}=\lambda A\tau.∎

This may not be enough by itself to identify τ\tau, since the λ−1\lambda^{-1}-eigenspace of AA may have higher dimension, but one can also use additional considerations such as the values of hh at T−1​(0)T^{-1}(0) and T−1​(1)T^{-1}(1). Thus for example, if β=πb−1​(1)\beta=\pi_{b}^{-1}(1), so that T−1​(0)T^{-1}(0) is the left endpoint of IβI_{\beta}, then

Combining this with Lemma˜A.1 one can identify the vectors τ\tau. Knowing also the substitution corresponding to AA, for any interval K∈𝒫(1)K\in\mathcal{P}^{(1)}, one can locate KK inside II, compute the value of hh at the left endpoint of KK, and then use the upper and lower bounds to check whether hh takes a certain value on KK or not.

References

• [AH20] Artur Avila and Pascal Hubert, Recurrence for the wind-tree model, Annales de l’Institut Henri Poincaré C, Analyse non linéaire37 (2020), no. 1, 1–11.
• [Buf13] Alexander I. Bufetov, Limit theorems for suspension flows over Vershik automorphisms, Russian Mathematical Surveys 68 (2013), no. 5, 789.
• [CDD+25] Frédéric Chapoton, Diana Davis, Vincent Delecroix, Oscar Fontaine, Charles Fougeron, Luke Jeffreys, Samuel Lelièvre, Julian Rüth, Ivan Yakovlev, and Christopher Zhang, surface-dynamics, February 2025.
• [CE15] Jon Chaika and Alex Eskin, Every flat surface is Birkhoff and Oseledets generic in almost every direction, Journal of Modern Dynamics 9(2015), no. 1, 1–23.
• [CG12] Jean-Pierre Conze and Eugene Gutkin, On recurrence and ergodicity for geodesic flows on non-compact periodic polygonal surfaces, Ergodic Theory and Dynamical Systems 32 (2012), no. 2, 491–515.
• [CHLP25] Sylvain Crovisier, Pascal Hubert, Erwan Lanneau, and Angel Pardo,Diffusion rate in non-generic directions in the wind-tree model, Advances in Mathematics 479 (2025), 110419.
• [Del13] Vincent Delecroix, Divergent trajectories in the periodic wind-tree model, Journal of Modern Dynamics 7 (2013), no. 1, 1–29.
• [DHL14] Vincent Delecroix, Pascal Hubert, and Samuel Lelièvre, Diffusion for the periodic wind-tree model, Annales Scientifiques de l’École Normale Supérieure 47 (2014), no. 6, 1085–1110.
• [DZ19] Vincent Delecroix and Anton Zorich, Cries and whispers in wind-tree forests, What’s Next? The Mathematical Legacy of William P. Thurston (Dylan Thurston, ed.), Princeton University Press, Princeton, 2019, pp. 83–115.
• [EE11] Paul Ehrenfest and Tatjana Ehrenfest, Begriffliche Grundlagen der statistischen Auffassung in der Mechanik, Encyklopädie der mathematischen Wissenschafte IV 32 (1911), 1–90.
• [EKZ14] Alex Eskin, Maxim Kontsevich, and Anton Zorich, Sum of lyapunov exponents of the hodge bundle with respect to the teichmüller geodesic flow, Publications mathématiques de l’IHÉS 120 (2014), no. 1, 207–333.
• [EM18] Alex Eskin and Maryam Mirzakhani, Invariant and stationary measures for the action on moduli space, Publications mathématiques de l’IHÉS127 (2018), no. 1, 95–324.
• [EMM15] Alex Eskin, Maryam Mirzakhani, and Amir Mohammadi, Isolation, equidistribution, and orbit closures for the S​L​(2,ℝ)SL(2,\mathbb{R}) action on moduli space, Annals of Mathematics (2015), 673–721.
• [FH18] Krzysztof Frączek and Pascal Hubert, Recurrence and non-ergodicity in generalized wind-tree models, Mathematische Nachrichten 291 (2018), no. 11-12, 1686–1711.
• [FU14] Krzysztof Frączek and Corinna Ulcigrai, Non-ergodic-periodic billiards and infinite translation surfaces, Inventiones mathematicae197 (2014), no. 2, 241–298.
• [HLT11] Pascal Hubert, Samuel Lelievre, and Serge Troubetzkoy, The Ehrenfest wind-tree model: periodic directions, recurrence, diffusion, Journal für die reine und angewandte Mathematik 656 (2011), 223–244.
• [HW80] Jean Hardy and Jaques Weber, Diffusion in a periodic wind-tree model, Journal of Mathematical Physics 21 (1980), no. 7, 1802–1808.
• [HW12] W. Patrick Hooper and Barak Weiss, Generalized Staircases: Recurrence and Symmetry, Annales de l’Institut Fourier 62 (2012), no. 4, 1581–1600 (en). MR 3025751
• [KMS86] Steven Kerckhoff, Howard Masur, and John Smillie, Ergodicity of billiard flows and quadratic differentials, Annals of Mathematics 124(1986), no. 2, 293–311.
• [KZ75] Anatole B Katok and Aleksandr N Zemlyakov, Topological transitivity of billiards in polygons, Mathematical Notes of the Academy of Sciences of the USSR 18 (1975), no. 2, 760–764.
• [LT16] Kathryn Lindsey and Rodrigo Treviño, Infinite type flat surface models of ergodic systems, Discrete and Continuous Dynamical Systems 36(2016), no. 10, 5509–5553.
• [Mas92] Howard Masur, Hausdorff dimension of the set of nonergodic foliations of a quadratic differential, Duke Mathematical Journal 66 (1992), no. 3, 387 – 442.
• [MMY05] Stefano Marmi, Pierre Moussa, and Jean-Christophe Yoccoz, The cohomological equation for roth type interval exchange maps, Journal of the American Mathematical Society 18 (2005), 823 – 872.
• [MMY12] by same author, Linearization of generalized interval exchange maps, Annals of mathematics (2012), 1583–1646.
• [SU05] Yakov G. Sinai and Corinna Ulcigrai, Weak mixing in interval exchange transformations of periodic type, Letters in Mathematical Physics74 (2005), 111–133.
• [Tum25] Yuriy Tumarkin, Ergodic measures for periodic typeℤm\mathbb{Z}^{m}-skew-products over interval exchange transformations, Journal d’Analyse Mathématique (2025), 1–43.
• [Via08] Marcelo Viana, Dynamics of interval exchange transformations and Teichmüller flows, Lecture notes available online athttps://w3.impa.br/˜viana/, 2008.
• [Yoc10] Jean-Christophe Yoccoz, Interval exchange maps and translation surfaces, Homogeneous flows, moduli spaces and arithmetic, vol. 10, Amer. Math. Soc., Providence, RI, 2010, pp. 1–69.
• [Zor06] Anton Zorich, Flat surfaces, Frontiers In Number Theory Physics And Geometry (2006), 439–585.