Hyperfiniteness of boundary actions of relatively hyperbolic groups

This paper studies equivalence relations induced by boundary actions of relatively hyperbolic groups. The study of boundary actions began with the work of Connes, Feldman and Weiss in [5] and Vershik in [21] who studied the actions of free groups on their boundaries. They showed that for a free group, its action on the Gromov boundary is $\mu$-hyperfinite for every Borel quasi-invariant probability measure $\mu$ on the boundary. Adams [1] later generalized this result to all hyperbolic groups.

Relatively hyperbolic groups were introduced by Gromov [10]; see also the monograph of Osin [17]. Given a relatively hyperbolic group $G$ with a collection of parabolic subgroups $\mathcal{P}$ there is a natural boundary called the Bowditch boundary, denoted $\partial(G,\mathcal{P})$, which is a compact metrizable space on which $G$ acts naturally by homeomorphisms.

In [18], Ozawa generalized the work of Adams [1] to the actions of relatively hyperbolic groups on their Bowditch boundary under the assumptions that the parabolic subgroups are exact. When the parabolic subgroups of $G$ in $\mathcal{P}$ are amenable, Ozawa [18] proved that the action of $G$ on $\partial(G,\mathcal{P})$ is topologically amenable, and, more generally, when the parabolic subgroups are exact, Ozawa [18] proved that the group $G$ is exact. Alternative proofs of the exactness of the group were given by Osin [16] who worked with parabolic subgroups with finite asymptotic dimension and by Dadarlat and Guentner [6] who worked with parabolic subgroups that are uniformly embeddable into a Hilbert space.

In [22], Zimmer introduced the notion of amenability of equivalence relations; see also the work of Connes, Feldman and Weiss [5]. By [2, Theorem 5.1], a measurable action of a countable group $G$ on a standard probability space $(X,\mu)$ is $\mu$-amenable if and only if $\mu$-almost all stabilizers are amenable and the orbit equivalence relation is $\mu$-amenable.

In this paper we generalize the result of Ozawa and work with relatively hyperbolic groups without any assumptions on the parabolic subgroups. In fact, we consider boundary actions from the Borel perspective. A countable Borel equivalence relation is called hyperfinite if it is a countable increasing union of finite Borel sub-equivalence relations. Dougherty, Jackson and Kechris showed in [8, Corollary 8.2] that the boundary action of any free group induces a hyperfinite orbit equivalence relation. The result of Dougherty, Jackson and Kechris was generalized to cubulated hyperbolic groups by Huang, Sabok and Shinko in [12], and later to all hyperbolic groups by Marquis and Sabok in [15]. In this paper, we prove the following:

Corollary B in particular strengthens the result of Ozawa [18] in case the parabolic subgroups are amenable. Indeed, hyperfiniteness implies $\mu$-amenability for every invariant Borel probability measure $\mu$ and by [3, Theorem 3.3.7], an action of a countable group on a locally compact space by homeomorphisms is topologically amenable if and only if it is $\mu$-amenable for every invariant Borel probability measure $\mu$.

We proceed by following a similar approach to [12] and [15], studying geodesic ray bundles $\text{Geo}(x,\eta)$ in relative Cayley graphs (Definition 2.2). For the case of a cubulating hyperbolic group $G$ studied in [12], the crucial property from which the hyperfinitess of the boundary action of $G$ follows is the finite symmetric difference of geodesic ray bundles: for any $x,y\in G$ and any $\eta\in\partial G$, $\text{Geo}(x,\eta)\triangle\text{Geo}(y,\eta)$ is finite (see [12, Theorem 1.4]). In [20], Touikan showed that this symmetric difference need not be finite in Cayley graphs of general hyperbolic groups, although in [14], Marquis provides many examples of groups acting geometrically on locally finite hyperbolic graphs where this finite symmetric difference property does hold. In [15], Marquis and Sabok define a modified version of the geodesic ray bundle, denoted $\text{Geo}_{1}(x,\eta)$ for $x\in G$ and $\eta\in\partial G$ (see [15, Definition 5.5] and Definition 2.6 in our paper) and show ([15, Theorem 5.9]) that these modified geodesic ray bundles satisfy a finite symmetric difference property: $|\text{Geo}_{1}(x,\eta)\triangle\text{Geo}_{1}(y,\eta)|<\infty$ for each $x,y\in G$ and for each $\eta\in\partial G$. Marquis and Sabok then deduce hyperfiniteness of the boundary action as a consequence of this finite symmetric difference property of the modified bundles (see [15, Section 6]).

Local finiteness of the Cayley graph plays a crucial role in establishing the finite symmetric difference property of the $\text{Geo}_{1}$ bundles in [15]. However, relative Cayley graphs of relatively hyperbolic groups are not locally finite. To make up for this loss of local finiteness, we rely on finiteness results about relative Cayley graphs of relatively hyperbolic groups from [17] (namely, [17, Theorem 3.26]).

We note also that the hyperfiniteness of boundary actions has been studied beyond relatively hyperbolic groups. Przytycki and Sabok have recently established the hyperfiniteness of the actions of a mapping class group of an oriented surface of finite type on the boundaries of the arc graph ([19, Theorem 1.1]) and the curve graph ([19, Corollary 1.2]) of the surface.

Acknowledgement: I owe great thanks to my advisor Marcin Sabok for his continuous support, patience and guidance throughout the production of this work.

In this paper, for a hyperbolic metric space $X$, $\partial X$ will denote the geodesic boundary of $X$. We will also denote $C_{hb}(X)$ the horoboundary of $X$ (see [15, Section 2.4] for a definition of the horoboundary).

In this section, we examine modified geodesic ray bundles in the relative Cayley graph $\hat{\Gamma}$ and prove that these modified bundles have finite symmetric difference for a fixed boundary point. This section generalizes [15, Theorem 5.9].

We begin by showing that $|\{\gamma(i):\gamma\in\text{CGR}(x,\eta)\}|$ is uniformly bounded for each $i$, each $x\in G$ and each $\eta\in\partial\hat{\Gamma}$, which is a well-known property in any uniformly locally finite hyperbolic graph.

We will make use of the following result, which states that geodesic triangles in the relative Cayley graph are slim with respect to the metric $d_{X}$ for some finite generating set $X$.

Here is the main result of this section.

As a corollary of Theorem 3.2, we obtain the following:

In this section, we establish the hyperfiniteness of the boundary actions of relatively hyperbolic groups as a consequence of Theorem 3.3. Our arguments follow [15, Section 6]. The main difference here is in our coding of labels of geodesics. In this section, we fix a finite generating set $X$ for $G$ as in Theorem 3.3 and let $\hat{\Gamma}$ denote the associated relative Cayley graph of $G$ with respect to $\{H_{1},...,H_{n}\}$ and $X$.

First, we give a binary coding to the symmetrized generating set $S:=(X\cup\mathcal{H})^{\pm}$. Using that $S$ is countably infinite, we fix a bijection $f:S\to 2^{<\mathbb{N}}:=\bigcup_{n\in\mathbb{N}}2^{n}$ from $S$ to the set $2^{<\mathbb{N}}$ of all finite binary sequences (which we can identify with the set of all finitely supported, infinite binary strings). The label of a geodesic ray is then coded as an element of $(2^{<\mathbb{N}})^{\mathbb{N}}$, the set of all infinite sequences of finite binary strings.

We will need to order elements of $(2^{n})^{n}$ (i.e. the set of length $n$ sequences of length $n$ binary strings) for each $n$. Following [8, Section 7], for each $m_{1},m_{2}\in\mathbb{N}$, each $w=(w_{0},w_{1},...,w_{n-1})\in(2^{m_{1}})^{m_{2}}$ and for each $n\in\mathbb{N}$ with $n\leq m_{1},m_{2}$, we put $w|_{n}=((w_{0})|_{n},(w_{1})|_{n},...,(w_{n-1})|_{n})$, where $(w_{j})|_{n}$ is the restriction of the length $m_{1}$ binary sequence $w_{j}$ to the first $n$ entries. Similarly, if $w\in(2^{\mathbb{N}})^{\mathbb{N}}$, we put $w|_{n}=((w_{0})|_{n},(w_{1})|_{n},...,(w_{n-1})|_{n})$. If we visualize $w\in(2^{n})^{n}$ as an $n\times n$ matrix, then $w|_{i}$ is an $i\times i$ submatrix of the $n\times n$ matrix $w$, starting at the top left corner of $w$.

For each $n\in\mathbb{N}$, we fix a total order $<_{n}$ on $(2^{n})^{n}$ as in [8, Section 7] such that for all $w,v\in(2^{n+1})^{n+1}$, $w|_{n}<_{n}v|_{n}\implies w<_{n+1}v$. Given $\gamma\in\text{CGR}(g,\eta)$, we define $\text{lab}(\gamma)\in(2^{<\mathbb{N}})^{\mathbb{N}}$ to be its coded label. Therefore, according to above, $\text{lab}(\gamma)|_{n}\in(2^{n})^{n}$ denotes the restricted label. Now, analogously to [15, Definition 6.1], we put:

Note that for each $n\in\mathbb{N}$, there exists $s\in(2^{n})^{n}$ which occurs infinitely often in $C^{\eta}$ because taking any $\gamma\in\text{CGR}(e,\eta)$, by [15, Proposition 5.8], $\gamma\setminus\text{Geo}_{1}(e,\eta)$ is finite, so there exists some $N$ such that for all $k\geq N$, $\gamma(k)\in\text{Geo}_{1}(e,\eta)$. Then $(\gamma(k),\text{lab}((\gamma(i))_{i\geq k})|_{n})\in C^{\eta}$ and $\text{lab}((\gamma(i))_{i\geq k})|_{n}\in(2^{n})^{n}$ for each $k\geq N$. Since $(2^{n})^{n}$ is finite, by the Pigeonhole Principle, some $s\in(2^{n})^{n}$ must repeat infinitely often in $C^{\eta}$, that is, $(\gamma(k),s)\in C^{\eta}$ for infinitely many $k\geq N$. For each $n\in\mathbb{N}$, we can therefore choose the minimal (in the order $<_{n}$ defined above) such $s\in(2^{n})^{n}$ occuring infinitely often in $C^{\eta}$. We shall denote this element by $s_{n}^{\eta}$.

We now fix a total order $\leq$ on the group $G$ such that $g\leq h\implies d(e,g)\leq d(e,h)$ (for instance, fixing a total order on $S$, we can define $\leq$ to be lexicographic order on elements of $G$ as words over $S$, where we choose for each element of $G$ the lexicographically least word over $S$ representing it). Using the same notation as in [15, Section 6], we put:

Note that $\min T_{n}^{\eta}$ exists because $T_{n}^{\eta}\subseteq\text{Geo}(e,\eta)$ and $\text{Geo}(e,\eta)$ is locally finite by Theorem 3.2. By definition of $\leq$ and since $s_{n}^{\eta}=(s_{n+1}^{\eta})|_{n}$ for each $n$, we have that $(T_{n}^{\eta})_{n}$ is a non-increasing sequence of sets and therefore the sequence $(k_{n}^{\eta})_{n\in\mathbb{N}}$ is a non-decreasing sequence of natural numbers.

We shall now generalize the results of [15, Section 6], which were stated for Cayley graphs of hyperbolic groups. Recall that we fix our hyperbolicity constant to be $\nu$ from Theorem 3.1. We recall that the topology on $G$ is the discrete topology induced by the relative metric $d$, the topology on $\partial\hat{\Gamma}$ is the canonical topology on the geodesic boundary, having countable neighbourhood base $V(\eta,m)^{g}=\{\mu\in\partial\hat{\Gamma}:\exists\gamma\in\text{CGR}(g,\mu)\text{ and }\lambda\in\text{CGR}(g,\eta)\text{ with }d(\gamma(t),\lambda(t))\leq 2\nu\text{ for each }t\leq m\}$ for each $m\in\mathbb{N}$, each $\eta\in\partial\hat{\Gamma}$ and each basepoint $g\in G$, $G^{\mathbb{N}}$ has the product topology and $C_{hb}(\hat{\Gamma})$ has the topology of pointwise convergence.

Let us establish a link between the topology of $\partial\hat{\Gamma}$ and sequences of CGRs in $\hat{\Gamma}$. The condition in the following proposition is often used as the definition of the topology on $\partial X$ when $X$ is a proper hyperbolic space, but in general does not give the same topology on $\partial X$ that we work with here.

We now generalize the claims of [15, Section 6] to relatively hyperbolic groups. We begin by generalizing Claim 1 of [15]. In Claim 1 in [15], the set $C$ below is proved to be compact, while here it is only closed.

The next claims are the exact relatively hyperbolic analogues of claims from [15] and their proofs are almost identical (most proofs are completely identical), however, we present all proofs for completeness.

By [15, Proposition 5.2], for each $\eta\in\partial\hat{\Gamma}$, we have that the section $M_{\eta}=\Xi(\eta)$ is finite, having cardinality bounded above by the constant $B$ from Theorem 3.2. Since $M$ is Borel and has finite sections of size at most $B$, by the Lusin-Novikov theorem we have Borel functions $\xi_{1},...,\xi_{B}:\partial\hat{\Gamma}\to C_{hb}(\hat{\Gamma})$ such that $M$ is the union of the graphs $G_{\xi_{i}}=\{(\eta,\xi_{i}(\eta)):\eta\in\partial\hat{\Gamma}\}$ of the $\xi_{i}$.