Bernoulli property of subadditive equilibrium states
Extracted source text · arXiv 2109.10721
\documentclass[11pt,oneside,english,reqno]{amsart}
\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc}
\setcounter{secnumdepth}{1}
\setcounter{tocdepth}{1}
\usepackage{color}
\usepackage{mathrsfs}
%\usepackage{babel}
\usepackage{amstext}
\usepackage{amsthm}
\usepackage{amssymb}
\usepackage{graphicx}
\usepackage[unicode=true,pdfusetitle,
bookmarks=true,bookmarksnumbered=false,bookmarksopen=false,
breaklinks=false,pdfborder={0 0 0},backref=false,colorlinks=true]
{hyperref}
\hypersetup{
citecolor=blue}
\makeatletter
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Textclass specific LaTeX commands.
\numberwithin{equation}{section}
\numberwithin{figure}{section}
\theoremstyle{plain}
\newtheorem*{cor*}{\protect\corollaryname}
\theoremstyle{plain}
\newtheorem{thm}{\protect\theoremname}[section]
\theoremstyle{definition}
\newtheorem{defn}[thm]{\protect\definitionname}
\theoremstyle{remark}
\newtheorem{rem}[thm]{\protect\remarkname}
\theoremstyle{plain}
\newtheorem{prop}[thm]{\protect\propositionname}
\theoremstyle{plain}
\newtheorem{lem}[thm]{\protect\lemmaname}
\theoremstyle{plain}
\newtheorem{cor}[thm]{\protect\corollaryname}
\theoremstyle{plain}
\newtheorem{example}[thm]{Example}
\input{pre}
\usepackage{float}
\address{Department of Mathematics,
The Ohio State University, Columbus, OH 43210, USA}
\email{call.119@osu.edu}
\address{School of Mathematics, KIAS, 85 Hoegiro, Dongdaemun-gu, Seoul, 02455, Korea}
\email{kihopark12@kias.re.kr}
\makeatother
\providecommand{\acknowledgementname}{Acknowledgement}
\providecommand{\corollaryname}{Corollary}
\providecommand{\definitionname}{Definition}
\providecommand{\lemmaname}{Lemma}
\providecommand{\propositionname}{Proposition}
\providecommand{\remarkname}{Remark}
\providecommand{\theoremname}{Theorem}
\providecommand{\theoremname}{Theorem}
\newcommand{\annotation}[1]{\marginpar{\tiny #1}}
\newcommand{\kp}[1]{\annotation{KP: #1}}
\newcommand{\bc}[1]{\annotation{BC: #1}}
\usepackage{ulem}
\newcommand\deleted[1]{{\tiny\color{red} \sout{#1}}}
\newcommand{\add}[1]{{\color{blue}{#1}}}
\newcommand{\comment}[1]{{}}
\normalem
\newcommand{\seenormal}{\renewcommand{\deleted}[1]{{}}}
\begin{document}
\title[Bernoulli property of subadditive equilibrium states]{Bernoulli property of subadditive equilibrium states}
\author{Benjamin Call, Kiho Park}
\date{\today}
%\subjclass[2010]{37D35}
%\thanks{B.C.\ is partially supported by NSF grant DMS-$1461163$.}
%\keywords{Equilibrium states, Thermodynamic Formalism, Kolmogorov property}
\begin{abstract}
Under mild assumptions, we show that the unique subadditive equilibrium states for fiber-bunched cocycles are Bernoulli. We achieve this by showing these equilibrium states are absolutely continuous with respect to a product measure, and then using the Kolmogorov property of these measures.
\end{abstract}
\maketitle
\section{Introduction}
In this paper, we study the Bernoulli property on a class of subadditive equilibrium states. We say a dynamical system $(X,f,\mu)$ is \textit{Bernoulli} if it is measurably isomorphic to a Bernoulli shift; see Definition \ref{defn: B}. Bernoullicity is the strongest ergodic property associated to complete randomness.
The study of Bernoulli systems was vastly enriched by the celebrated work of Ornstein \cite{ornstein1970isomorphic, ornstein1973KnotBernoulli, ornstein1973geodesic}. Among other things, he showed that Bernoulli shifts are classified by their entropy up to measurable isomorphism. He also helped introduce many new definitions that imply the Bernoulli property; including the Weak Bernoulli, Very Weak Bernoulli, and finitely determined properties \cite{FriedmanOrnstein,OrnsteinWeissFinDet,Ornstein1970}. Some of these, such as the Very Weak Bernoulli property, are easier to verify than others, and we will make use of it in establishing the Bernoullicity of the subadditive equilibrium states.
Thanks to these checkable conditions for Bernoullicity, Ornstein theory has successfully been implemented to show that many natural invariant measures of important dynamical systems are Bernoulli. Ornstein and Weiss \cite{ornstein1973geodesic} showed that geodesic flows on negatively curved surfaces are Bernoulli with respect to Liouville measure. This was extended by Pesin \cite{pesin1977characteristic} for geodesic flows on a more general class of manifolds. More recently, the first-named author and Thompson \cite{call2019equilibrium} have shown that the measure of maximal entropy for the geodesic flow on non-positively curved rank one manifolds is Bernoulli, and Carrasco and Rodriguez-Hertz \cite{carrasco2021equilibrium} have recently shown that some equilibrium states for center isometries are Bernoulli as well. All of these works proceed by first showing that a system is $K$, and then showing that it is Bernoulli. General arguments can be found to show that smooth $K$-systems with some amount of hyperbolic structure \cite{chernov1996nonuniformly, ornstein1998bernoulli} are Bernoulli.
The main content of this paper is of a similar flavor.
From here on, we will now focus on the case where the base dynamical system is a mixing subshift of finite type $(\Sig,\s)$.
In this setting, Bowen \cite{bowen1974some} showed uniqueness of the equilibrium states for \hol potentials.
These are important invariant measures for the underlying dynamical systems that often capture relevant statistical information, and they are also known to be Bernoulli \cite{bowen1974bernoulli, bowen1975ergodic}.
Recently, there have been great developments in the theory of subadditive thermodynamic formalism and its applications; see \cite{barreira2011dimension} and \cite{cao2019dimension}.
In particular, due to their connections to the dimension theory of fractals, subadditive potentials arising from matrix cocycles and their equilibrium states have been extensively studied; see \cite{MR3820437} and references therein. The norm potentials and singular value potentials are prime examples of such subadditive potentials (see Subsection \ref{subsec: cocycles}).
While the equilibrium states for the norm potentials are not necessarily unique in general, there are mild conditions on cocycles that guarantee the uniqueness of the equilibrium states.
In the case of locally constant cocycles, Feng \cite{feng2003lyapunov, feng2009lyapunov} showed that irreducibility implies uniqueness of equilibrium states.
Morris \cite{morris2018ergodic,morris2019necessary} then showed that total ergodicity of these equilibrium states implies mixing and that the failure of mixing can be characterized by certain structures of the cocycle. He recently improved his own result by showing that total ergodicity implies Bernoullicity \cite{morris2020totally}.
Under a different set of assumptions on the cocycle, Piraino \cite{piraino2018weak} showed that the equilibrium states are Bernoulli. While both results amount to establishing the Weak Bernoulli property, their methods of proof differ both from each other, and from the approach we take in this paper.
Similar questions on the ergodic properties of the subadditive equilibrium states were also studied beyond the realm of irreducible locally constant cocycles.
Within the class of fiber-bunched cocycles (see Subsection \ref{subsec: cocycles}), under suitable assumptions the second author \cite{park2019quasi} and Bochi and Garibaldi \cite{bochi2019extremal} established the uniqueness of the equilibrium states for the norm potentials. For such equilibrium states, the authors showed in \cite{call2020k} that total ergodicity implies the $K$-property (see Subsection \ref{subsec: ergodic properties} and Proposition \ref{prop: CP}). Our first result shows that these equilibrium states in fact have the stronger ergodic property of being Bernoulli. It extends our previous work and also generalizes existing results on locally constant cocycles to fiber-bunched cocycles.
\begin{thmx}\label{thm: A}
Let $\A \colon \Sig \to \glr$ be a fiber-bunched cocycle.
\begin{enumerate}
\item
Suppose that $\A$ is strongly bunched and irreducible. If the unique equilibrium state $\mu_\A$ for the norm potential $\Phi_\A$ is totally ergodic, then it is Bernoulli.
\item
Supposing instead that $\A$ is typical, the unique equilibrium state $\mu_{\A,s} $ for the singular value potential $\Phi^s_\A$ is Bernoulli.
\end{enumerate}
\end{thmx}
After collecting relevant previous results and establishing what we will later refer to as measurable local product structure (see Definition \ref{defn: lps}), Theorem \ref{thm: A} follows as a consequence of the following theorem; see Remark \ref{rem: thm A as cor}.
\begin{thmx}\label{thm: B}
Let $\mu $ be an ergodic measure on $\Sig$. If $\mu$ satisfies the $K$-property and furthermore, is absolutely continuous with respect to a non-atomic product measure on the local stable and unstable sets on an open neighborhood, then $\mu$ is Bernoulli.
\end{thmx}
A close variant of this result is possibly expected by experts; in particular, when the measure in question is equivalent to a non-atomic product measure. However, we believe that the proof has not previously been written down. Indeed, this result is similar in flavor to many existing results establishing Bernoullicity from the $K$-property under various additional assumptions (see also \cite{ledrappier2016ergodic, ponce2018bernoulli} for related results on Bernoullicity in different settings, as well as the excellent book \cite{ponce2019introduction} on settings in which $K$ implies Bernoulli). It is always necessary to have these assumptions, as there are systems that are $K$ but not Bernoulli \cite{ornstein1973KnotBernoulli, katok1980smooth, kanigowski2018non}. In our work, we will refer to our main additional assumption as \textit{measurable local product structure}. We define this to mean that on an open neighborhood, the measure is absolutely continuous with respect to a non-atomic product measure on local stable and unstable sets; see Definition \ref{defn: lps} for the precise definition.
\begin{rem}
We stress that in this paper, we only need our measure to be absolutely continuous, not equivalent, to a product measure for us to say it has measurable local product structure. Furthermore, we include the word \textit{measurable} to indicate that the Radon-Nikodym derivative need not be continuous.
\end{rem}
Our proof follows the classical approach of Chernov and Haskell \cite{chernov1996nonuniformly} and Ornstein and Weiss \cite{ornstein1973geodesic, ornstein1998bernoulli} in that we show the existence of Very Weak Bernoulli partitions. The major difference is that we do not require our measure to be smooth. While this approach is not new, applying it to subadditive equilibrium states provides a different perspective in studying ergodic properties of such measures beyond locally constant cocycles. Moreover, we have simplified the relevant parts of the proof according to the fact that the base dynamical system is uniformly hyperbolic, as opposed to non-uniformly hyperbolic in \cite{chernov1996nonuniformly}.
The paper is organized as follows. In $\S$\ref{sec: prelim}, we introduce relevant notations and survey preliminary results. In $\S$\ref{sec: LPS}, we establish measurable local product structure on subadditive equilibrium states with the Gibbs property, and in $\S$\ref{sec: Bernoulli} we prove an abstract theorem on Bernoullicity which implies Theorem \ref{thm: B}.
\\\\
\noindent\textbf{Acknowledgments:} We would like to thank Dan Thompson and Amie Wilkinson for helpful discussion and valuable comments on this paper. B.C. was partially supported NSF grant DMS-1954463, as well as a Presidential Fellowship from The Ohio State University.
\section{Preliminaries}\label{sec: prelim}
\subsection{Symbolic dynamical systems}
Let $T$ be a $q \times q$ square-matrix with entries in $\{0,1\}$. We define $\Sig \subset \{1,\cdots, q\}^\Z$ to be the set of all bi-infinite sequences $(x_i)_{i\in \Z}$ satisfying $T_{x_i,x_{i+1}}=1$ for all $i\in \Z$. The left shift map $\s\colon \Sig \to \Sig$ is defined as $(\s x)_i = x_{i+1}$ for $x = (x_i)_{i\in \Z}$ and $i\in \Z$.
We denote the corresponding one-sided subshift by $\Sig^+ \subset \{1,\cdots, q\}^{\N_0}$. The left shift map, again denoted by $\s$, is defined analogously with $i \in \N_0$. We denote the one-sided subshift on negative symbols as $\Sig^- \subset \{1,\cdots, q\}^{-\N_0} $ with the shift map given by $\s^{-1}$.
An \textit{admissible word of length $n$} is a word $i_0\ldots i_{n-1}$ with $i_j \in \{1,\ldots,q\}$ such that $T_{i_j,i_{j+1}} = 1$ for all $0 \leq j \leq n-2$.
Let $\L$ be the collection of all admissible words. For $\I\in \L$, we denote its length by $|\I|$.
For each $n \in \N$, let $\L(n) \subset \L$ be the set of all admissible words of length $n$. For any $\I = i_0\ldots i_{n-1} \in \L(n)$, we define the associated \textit{cylinder} by
$$[\I]=[ i_0\ldots i_{n-1}]:=\{y \in \Sig \colon y_j = i_j \text{ for all } 0 \leq j \leq n-1\}.$$
For any $x=(x_i)_{i \in \Z}\in \Sig$ and $n\in \N$ we denote by $[x]_n:=[x_0\ldots x_{n-1}]$ the $n$-cylinder containing $x$.
We endow $\Sig$ with the metric $d$ defined as follows: for $x = (x_i)_{i \in \Z},y = (y_i)_{i \in \Z} \in \Sig$, we have
$$d(x,y) = 2^{-k},$$
where $k$ is the largest integer such that $x_i = y_i$ for all $|i| < k$. This makes $\Sig$ into a compact metric space, and the left shift operator $\s$ becomes a hyperbolic homeomorphism on $\Sig$,
meaning that $\s$ has stable and unstable sets with uniform exponential rate, which satisfy local product structure.
Given $x\in\Sig$, the \textit{local stable set} of $x$ is
$$\Wloc^s(x) = \{y\in\Sig\mid x_i = y_i \text{ for } i\geq 0 \}$$
and analogously, the \textit{local unstable set} of $x$ is
$$\Wloc^u(x) = \{y\in\Sig\mid x_i = y_i \text{ for } i\leq 0\}.$$
%\deleted{These local stable and unstable sets extend to define global stable and unstable sets $\W^{s/u}(x)$, respectively, in the standard manner.}
Finally, we will always assume that the adjacency matrix $T$ is \textit{primitive}, meaning that there exists $N>0$ such that all entries of $T^N$ are positive. The primitivity of $T$ is equivalent to $(\Sig,\s)$ being topologically mixing.
\subsection{Subadditive thermodynamic formalism}\label{subsec: symbolic}
Let $\Phi = \{\log \vp_n\}_{n\in \N}$ be a sequence of continuous functions $\Sig$. We say $\Phi$ is \textit{subadditive} if
$$0 \leq \log \vp_n + \log \vp_m \circ \s^n \leq \log \vp_{m+n}$$
for every $m,n\in \N$. Often we will refer to $\Phi$ as a \textit{subadditive potential} on $\Sig$.
For every subadditive potential $\Phi$, we can define its pressure $P(\Phi)$, which satisfies the following
\textit{subadditive variational principle} \cite{cao2008thermodynamic}:
$$P(\Phi)=\sup\limits_{\mu \in \M(\s)} \Big(h_\mu(\s)+ \lim\limits_{n\to \infty} \frac{1}{n}\int \log \vp_n \,d\mu \Big)$$
Any invariant measure $\mu$ attaining the supremum is called an \textit{equilibrium state}.
We say that a probability measure $\mu$ on $\Sig$ satisfies the \textit{Gibbs property} with respect to $\Phi$ if there exists a constant $C\geq 1$ such that for any $x\in \Sig$ and $n\in \N$, we have
$$C^{-1} \leq \frac{\mu([x]_n)}{e^{-nP(\Phi)} \vp_n(x)} \leq C.$$
We now introduce properties that will be satisfied by all subadditive potentials considered in this paper. First, we say $\Phi$ has \textit{bounded distortion} if there exists $C \geq 1$ such that for any $x\in \Sig$, $n\in\N$, and $y \in [x]_n$, we have
$$C^{-1} \leq \frac{\vp_n(x)}{\vp_n(y)} \leq C,$$
which going forward we will write as $\frac{\vp_n(x)}{\vp_n(y)}\asymp C$.
\begin{rem}\label{rem: identification}
For each $\I\in \L(n)$, we define
$$\vp(\I):=\sup\limits_{x \in [\I]} \vp_n(x).$$
Using this, $\log \vp$ defines a subadditive potential on $\L$, which by an abuse of notation we denote again by $\Phi$.
Conversely, given a subadditive potential $\Phi:=\log \vp$ on $\L$, we can induce on $\Sig$ a subadditive potential, denoted again by $\Phi$, by setting $\vp_n(x) := \vp([x]_n)$.
We will identify such potentials (on $\Sig$ and on $\L$) with one another; this identification will simplify the presentation of the preliminary results described below.
Moreover, relevant objects such as the subadditive pressure and the set of equilibrium states remain unchanged under this identification if the original subadditive potential $\Phi = \{\log \vp_n\}$ on $\Sig$ has bounded distortion; see \cite{park2019quasi}.
\end{rem}
The last property we will make use of is called quasi-multiplicativity. We say a subadditive potential $\Phi=\log \vp$ on $\L$ is \textit{quasi-multiplicative} if there exist $c>0$ and $k\in \N$ such that for any $\I,\J \in \L$ there exists $\K \in \L$ with $|\K| \leq n$ so that
$$ \vp(\I\K\J) \geq c \vp(\I)\vp(\J).$$
The following proposition whose proof can be found in \cite[Theorem 5.5]{feng2011equilibrium} guarantees that quasi-multiplicativity is a sufficient condition for the uniqueness of the equilibrium state.
\begin{prop}\label{prop: qm}
Let $\Phi$ be a subadditive potential on $\L$. If $\Phi$ is quasi-multiplicative, then it has a unique equilibrium state $\mu \in \M(\s)$. Moreover, $\mu$ has the Gibbs property with respect to $\Phi$.
%qm then unique equilibrium state. Also Gibbs property.
\end{prop}
%We refer the readers to --- for a more comprehensive discussion of subadditive thermodynamic formalism.
\subsection{Ergodic properties of an invariant measure}\label{subsec: ergodic properties}
In this subsection, let $(X,f)$ be a homeomorphism on a compact metric space. We provide a brief introduction to the various mixing properties we use throughout this paper.
Given an invariant measure $\mu \in \M(f)$, we say $\mu$ is \textit{totally ergodic} if it is ergodic with respect to $f^n$ for all $n\in \N$. Total ergodicity is stronger than ergodicity, but weaker than mixing.
A stronger mixing property is the Kolmogorov property (the \textit{$K$-property}, for short), which is stronger than mixing of all orders but weaker than Bernoulli. It has many equivalent formulations, and the following definition (known as \emph{completely positive entropy}) is an equivalence established by Rokhlin and Sinai.
\begin{defn}\label{defn: K}
An invariant measure $\mu \in \M(f)$ has the \textit{$K$-property} if its Pinsker factor (i.e., maximal zero entropy factor) is trivial.
\end{defn}
The authors have shown in a previous paper that under the conditions below, total ergodicity implies the $K$-property:
\begin{prop}\cite{call2020k}\label{prop: CP}
Let $\Phi$ be a subadditive potential on $\Sig$, and suppose it is quasi-multiplicative and has bounded distortion. If its unique equilibrium state is totally ergodic, then it has the $K$-property.
\end{prop}
In this paper, the goal is to show that under these hypotheses, the underlying measure is Bernoulli.
In order to show this implication, we will make use of another one of the equivalent formulations for the $K$-property as described in the following proposition; see \cite[Section 2.8]{ponce2019introduction}. In order to state it, we first need an additional definition.
\begin{defn}\label{defn: eps almost everywhere}
Given a partition $\xi$ of $X$ and a probability measure $\mu$ on $X$, a property hold \textit{$\ep$-almost everywhere} on $\xi$ if the union of atoms on which the property does not hold has $\mu$-measure at most $\ep$.
\end{defn}
\begin{prop}\label{prop: K-partitions}
For every finite measurable partition $\xi$ of a $K$-system, for all $\ep > 0$ and finite collection of sets $A_i\subset X$, there exists $N\in \N$ such that for all $m_2 \geq m_1 \geq N$, for $\ep$-almost every $E\in \bigvee_{m_1}^{m_2}f^i\xi$ and each $A_i$,
$$\left|\frac{\mu(E\cap A_i)}{\mu(E)} - \mu(A_i)\right| \leq \ep.$$
\end{prop}
Finally, we give a formal definition of the Bernoulli property.
\begin{defn}\label{defn: B}
Given a finite set $S$ equipped with a measure $\nu$, we say the dynamical system $(S^{\N_0},\s,\nu^{\N_0})$ is a \textit{Bernoulli shift}. We say that an invariant measure $\mu \in \M(f)$ is \textit{Bernoulli} if $(X,f,\mu)$ is measurably isomorphic to a Bernoulli shift.
\end{defn}
\subsection{Matrix cocycles}\label{subsec: cocycles}
%\bc{Moved statement of H\"older up here, since it's used everywhere.}
Let $\A \colon \Sig \to \glr$ be an $\alpha$-\hol continuous cocycle. For $x\in \Sig$ and $n\in \N$, we define
$$\A^n(x):=\A(\s^{n-1}x) \ldots \A(x).$$
From the submultiplicativity of the operator norm $\|\cdot \|$, the sequence of functions $\Phi_\A:=\{\log \vp_{\A,n}\}$ given by $$\vp_{\A,n}(x):=\|\A^n(x)\|$$ defines a subadditive potential. We call this the \textit{norm potential} of $\A$.
We denote the singular values of $A \in \glr$ by $\s_1(A) \geq \ldots \geq \s_d(A)>0$. For $s \in [0,d]$, we define
$$\vp^s(A):= \s_1(A)\ldots \s_{\lfloor s \rfloor}(A)\s_{\lceil s \rceil}(A)^{\{s \}},$$
whereas for $s > d$ we set $\vp^s(A):=|\det(A)|^{s/d}$.
For any $\A \colon \Sig \to \glr$ and $s \geq 0$, we define the $s$-\textit{singular value potential} as $$\Phi_\A^s:=\{\log \vp_{\A,n}^s\}_{n\in \N} \text{ where } \vp_{\A,n}^s :=\vp^s(\A^n(x)).$$
Since $\|A\| = \s_1(A)$, the $1$-singular value potential $\Phi_\A^1$ coincides with the norm potential $\Phi_\A$. Therefore, singular value potentials are generalizations of norm potentials.
When $\A\ \colon \Sig \to \glr$ depends only on the zero-th symbol $x_0$ of $x=(x_i)_{i\in \N}$, we say that it is \textit{locally constant}.
When $\A$ is \textit{irreducible}, meaning that there does not exist a proper subspace $V \subset \R^d$ preserved under the action of $\A$, its norm potential $\Phi_\A$ has a unique equilibrium state. In this case, Morris has shown in \cite{morris2018ergodic, morris2019necessary, morris2020totally} that for such equilibrium states total ergodicity implies Bernoullicity.
In this paper we expand Morris' result on Bernoullicity to a class of cocycles that are not necessarily locally constant. Moreover, we apply a fundamentally different technique than Morris, relying on an approach pioneered by Ornstein and Weiss \cite{ornstein1973geodesic, ornstein1998bernoulli}; see Section \ref{sec: Bernoulli} for more details. We will focus on the class of fiber-bunched cocycles, which we define below.
We say an $\a$-\hol cocycle $\A \colon \Sig \to \glr$ is \textit{fiber-bunched} if
\begin{equation}\label{eq: fb}
\|\A(x)\|\cdot \|\A(x)^{-1}\| <2^\a
\end{equation}
for every $x\in \Sig$. It is clear from the definition that the set of fiber-bunched cocycles forms an open subset of the set of $\a$-\hol cocycles $C^\a(\Sig,\glr)$ and that perturbations of conformal cocycles are fiber-bunched cocycles. In fact, fiber-bunched cocycles may as well be thought of as nearly conformal cocycles.
The fiber-bunching assumption guarantees the convergence of the canonical holonomies: for $y \in \Wloc^s(x)$, we define
$$H^s_{x,y}:=\lim\limits_{n\to \infty} \A^n(y)^{-1}\A^n(x).$$
The collection $\{H^s_{x,y}\}\subseteq \glr$ is called the \textit{canonical stable holonomy}. The holonomy satisfies the following properties: for any $y,z \in \Wloc^s(x)$,
\begin{enumerate}
\item $H^s_{x,x} = \id$ and $H^s_{y,z}H^s_{x,y} = H^s_{x,z}$,
\item $\A(y)H^s_{x,y} = H^s_{\s x,\s y }\A(x)$,
\item $(x,y) \mapsto H^s_{x,y}$ is continuous,
\item $\|H^s_{x,y} -\id \| \leq C \cdot d(x,y)^\a$.
\end{enumerate}
The second property says that the holonomy is $\A$-equivariant, and using this property we can define \textit{global stable holonomy} $H^{s}_{x,y}$ for $y \in \W^s(x)$ that does not necessarily lie on $\Wloc^s(x)$. The fourth property says that the stable holonomies are \hol continuous with the same exponent as the cocycle.
Using the inverse cocycle $\A^{-1}$ over $(\Sig,\s^{-1})$ defined by $\A^{-1}(x):=\A(\s^{-1}x)^{-1}$, we define the \textit{canonical unstable holonomy} $\{H^u_{x,y}\}$ as
$$H^u_{x,y} := \lim\limits_{n\to\infty} \A^{-n}(y)^{-1}\A^{-n}(x).$$
The unstable holonomy satisfies the analogous properties as above with $s$ and $\s$ replaced by $u$ and $\s^{-1}$, respectively. See \cite{kalinin2013cocycles} for a more detailed discussion on the canonical holonomies of fiber-bunched cocycles.
\begin{rem}\label{rem: bdd distortion}
For any fiber bunched cocycle $\A$, the \hol continuity of $H^{s/u}$ implies that the norm potential $\Phi_\A$ and the singular value potential $\Phi_\A^s$ for any $s \geq 0$ have bounded distortion; see \cite{park2019quasi}. This justifies the identification assumed in Remark \ref{rem: identification}.
\end{rem}
We now introduce the irreducibility and the strong bunching assumptions for fiber-bunched cocycles appearing in the first case of Theorem \ref{thm: A}. In the following definition, the \textit{bi-holonomy invariance} of a bundle $\V$ over $\Sig$ means that $\V_y = H^s_{x,y}\V_x$ for any $x \in \Wloc^s(x)$, and likewise for the unstable holonomy.
\begin{defn}
We say a fiber-bunched cocycle $\A \colon \Sig \to \glr$ is \textit{irreducible} if there does not exist an $\A$-invariant and bi-holonomy invariant subbundle.
\end{defn}
This definition of irreducibility for fiber-bunched cocycles is compatible with irreducibility for locally constant cocycles: when a fiber-bunched cocycle $\A$ is locally constant, then the canonical holonomies are trivial (i.e., $H^{s/u} \equiv \id$), and hence, the two notions of irreducibility coincide.
%We note that the introduced notion of irreducibility coincides with that of locally. If $\A$ is locally constant, the notions of irreducibilty coincides
\begin{defn}
We say $\A \colon \Sig \to \glr$ is \textit{strongly bunched} if
\begin{enumerate}
\item $d=2$ and $\A$ is fiber-bunched as in \eqref{eq: fb}, or
\item $d\geq 3$ and for every $x\in \Sig$,
$$\|\A(x)\|\cdot \|\A(x)^{-1}\| <2^{\a/3}.$$
\end{enumerate}
\end{defn}
Notice that when $d=2$, strong bunching coincides with fiber-bunching. It is only when $d\geq 3$ the strong bunching requires a more stringent bound compared to \eqref{eq: fb}.
\begin{rem}The strong bunching assumption was first introduced in Bochi and Garibaldi \cite{bochi2019extremal} in the more general setting of hyperbolic homeomorphisms. The general definition there requires that the product $\|\A(x)\|\|\A(x)^{-1}\|$ is bounded above by a constant depending only on the base dynamical system and the exponent of the cocycle. In the case of a subshift of finite type $(\Sig,\s)$ equipped with the standard metric defined in Subsection \ref{subsec: symbolic}, they showed that such a bound can be taken to be $2^{\a/3}$.
\end{rem}
We now introduce the typicality assumption appearing in the second case of Theorem \ref{thm: A}.
\begin{defn}Let $\A \colon \Sig \to \glr$ be a fiber-bunched cocycle. \\We say $\A$ is \textit{1-typical} if
\begin{enumerate}
\item there exists a periodic point $p\in \Sig$ such that $P:=\A^{\text{per}(p)}(p)$ has simple eigenvalues of distinct moduli, and
\item there exists a homoclinic point $z \in \W^s(p) \cap \W^u(p) \setminus \{p\}$ such that the holonomy loop $\psi:=H^s_{z,p}H^u_{p,z}$ maps the eigendirections $\{v_1,\ldots,v_d\}$ of $P$ into a general position.
\end{enumerate}
Moreover, we say $\A$ is \textit{typical} if the exterior product cocycle $\A^{\wedge t}$ is 1-typical for every $1 \leq t \leq d-1$.
\end{defn}
Typicality was first introduced by Bonatti and Viana \cite{bonatti2004lyapunov}. They showed that the Lyapunov exponents of typical cocycles with respect to the unique equilibrium states for \hol potentials are simple. Two assumptions appearing in the definition of typicality are called \textit{pinching} and \textit{twisting}, and their roles parallel those of proximality and strong irreducibility in the classical setting of i.i.d. product of random matrices.
We also note that typicality is a stronger assumption which implies irreducibility.
The following proposition summarizes the known results on the singular value potentials of the cocycles considered in Theorem \ref{thm: A}.
\begin{prop}\label{prop: cocycles}
Let $\A \colon \Sig \to \glr$ be a fiber-bunched cocycle.
\begin{enumerate}
\item If $\A$ is strongly bunched and irreducible, then its norm potential $\Phi_\A$ is quasi-multiplicative.
\item If $\A$ is typical, then its singular value potential $\Phi_\A^s$ is quasi-multiplicative for all $s \geq 0$.
\end{enumerate}
In both cases, Proposition \ref{prop: qm} gives a unique equilibrium state satisfying the Gibbs property.
\end{prop}
For the first statement, it was shown by Bochi and Garibaldi \cite{bochi2019extremal} that $\A$ is spannable. The second author and Butler \cite{butler2019thermodynamic} then showed that the norm potentials of spannable cocycles are quasi-multiplicative. The second statement is established by the second author in \cite{park2019quasi}. We also note that 1-typicality implies that the norm potential $\Phi_\A$ is quasi-multiplicative, and typicality ensures that the same is true for the $s$-singular value potentials $\Phi_\A^s$ for all $s \geq 0$.
Notice from its definition that if $\A \colon \Sig \to \glr$ is typical, then so are its powers $\A^n$ for all $n\in\N$. Hence, Proposition \ref{prop: cocycles} has an immediate corollary:
\begin{cor}\label{cor: typical then totally erg}
If $\A$ is typical, then for any $s\geq 0$ the unique equilibrium state $\mu_{\A,s}$ for $\Phi_\A^s$ is totally ergodic.
\end{cor}
This corollary was already observed in the previous paper of the authors \cite{call2020k}. This is the reason why we do not have to separately assume the total ergodicity on $\mu_{\A,s}$ from the second case of Theorem \ref{thm: A} because it is already implied by typicality.
\begin{rem}\label{rem: thm A as cor}
In view of Propositions \ref{prop: CP}, and \ref{prop: cocycles}, and Corollary \ref{cor: typical then totally erg}, we only need to show that the $K$ property implies Bernoullicity of our measures to prove Theorem \ref{thm: A}. This is the content of the next two sections, which contains the proof of Theorem \ref{thm: B}.
\end{rem}
\section{Measurable Local Product Structure}\label{sec: LPS}
\begin{defn}\label{defn: lps}
A measure $\mu$ on $\Sig$ has \textit{measurable local product structure} if almost every $x\in\Sig$ belongs to an open set $U$ of positive $\mu$-measure on which $\mu$ is absolutely continuous with respect to a non-atomic product measure on $\Wloc^s(x)\times \Wloc^u(x)$, where each component of the product measure is a standard probability measure.
\end{defn}
When working with an ergodic measure, this definition can be checked on a single open set.
\begin{prop}
If $\mu$ is ergodic, then $\mu$ has measurable local product structure if there exists at least one positive measure open set $U$ on which $\mu$ is absolutely continuous with respect to a non-atomic product measure on $\Wloc^s(x)\times \Wloc^u(x)$ for some $x\in U$.
\end{prop}
\begin{proof}
If $\lambda$ is a product measure on $\Wloc^s(x)\times \Wloc^u(x)$, then $\s^n_*\lambda$ is a product measure on $\s^n\Wloc^s(x)\times \s^n\Wloc^u(x)$ for all $n\geq 0$, and $\s^nx\in \s^nU$. Furthermore, we have
$$\mu = \s^n_*\mu \ll \s^n_*\lambda.$$
Finally, since $\mu$ is ergodic, we know that $\mu(\bigcup_{n\geq 0} \s^nU) = 1$. As such, for $\mu$-a.e. $x\in \Sig$, $x$ is contained in an open set $\s^nU$ on which $\mu$ is absolutely continuous with respect to a product measure, completing our proof.
\end{proof}
In what follows, we show that in the settings considered in this paper, our measures have measurable local product structure.
\begin{prop}\label{prop: LPS}
Let $\Phi = \log \vp$ be a subadditive potential on $\L$ and $\mu$ an ergodic measure on $\Sig$ satisfying the Gibbs property with respect to $\Phi$. Then $\mu$ has measurable local product structure.
% \deleted{Denoting the projection of $\mu$ onto $\Sig^+$ and $\Sig^-$ respectively by $\mu^+$ and $\mu^-$, we have $\mu \ll \mu^+ \times \mu^-$.}
% with $\psi:=d\mu/d(\mu^+\times \mu^-)$ nonzero $\mu$-almost everywhere.
\end{prop}
\begin{proof}
As $\mu$ is ergodic, we may suppose that we are working in the 1-cylinder $[i]$. Let $x\in [i]$ be arbitrary. We now define $\mu^+$ on $\Wloc^u(x)$ by $\displaystyle \mu^+(A) := \frac{\mu(\pi_u^{-1}(\pi_u A))}{\mu(\pi_u^{-1}(\pi_u \Wloc^u(x)))}$ where $\pi_u : \Sigma_T \to \Sigma_T^+$ is the projection, and define $\mu^-$ on $\Wloc^s(x)$ similarly but with $\Sigma_T^-$.
From its definition, $\mu^+$ also has the Gibbs property with respect to $\Phi$ and $\s$. Likewise, $\mu^-$ has the Gibbs property with respect to $\Phi$ and $\s^{-1}$, meaning that for any $x^- = \ldots j_{-3} j_{-2}j_{-1}i \in [i]$ we have $\displaystyle \frac{\mu^-([j_{-n} \ldots j_{-1}i])}{e^{-(n+1)P(\Phi)}\vp(j_{-n} \ldots j_{-1}i)} \asymp C$ for any $n\in \N$. Note here that the pressure $P(\Phi)$ is the same when $\Phi$ is considered as a subadditive potential over any of $(\Sig,\s)$, $(\Sig^+,\s)$, and $(\Sig^-,\s^{-1})$.
For any $\I,\J\in \L$ with $|\I| = n$ and $|\J| = m$, subadditivity of $\Phi$ gives
$$\mu([\I\J])\asymp Ce^{-(n+m)P(\Phi)}\vp(\I\J) \leq C^2e^{-nP(\Phi)} \vp(\I)\cdot e^{-mP(\Phi)} \vp(\J)\asymp \mu^+([\I])\mu^-([\J]).$$
From this, it follows that $\mu \ll \mu^+ \times \mu^-$.
\end{proof}
\section{Bernoulli property}\label{sec: Bernoulli}
The goal of this section is to prove the following theorem on Bernoulli property, which is a more general version of Theorem \ref{thm: B} in view of Proposition \ref{prop: LPS}. As pointed out in Remark \ref{rem: thm A as cor}, Theorem \ref{thm: A} then also follows from it.
\begin{thm}\label{thm: abstract}
Suppose $X$ is a compact metric space and $f : X\to X$ is a hyperbolic homeomorphism. If $\mu$ is an invariant measure such that $\mu$ is everywhere absolutely continuous with respect to a product measure $\rho\x\nu$ on the local stable and unstable sets, then if $\mu$ has the $K$-property, it is Bernoulli.
\end{thm}
\subsection{Definitions}
For further details about these definitions, we refer the reader to the book by Ponce and Varao \cite{ponce2019introduction}, as well as the paper by Chernov and Haskell \cite{chernov1996nonuniformly}.
Consider two measure spaces $(X,\mu)$ and $(Y,\nu)$. A probability measure $\lambda$ on $X \times Y$ is a \textit{joining} if $(\pi_1)_*\lambda = \mu$ and $(\pi_2)_*\lambda = \nu$ where $\pi_1,\pi_2$ are the canonical projections onto $X$ and $Y$, respectively. We denote by $J(\mu,\nu)$ the set of all joinings of $\mu$ and $\nu$.
We now introduce the notion of $\ol{d}$-distance between partitions and sequences of partitions. Consider finite partitions $\xi=\{A_1,\ldots,A_k\}$ of $X$ and $\eta= \{B_1,\ldots,B_k\}$ of $Y$. For any $x\in X$, we define $\xi(x)$ as the unique index $j \in \{1,\ldots,k\}$ such that $x \in A_j$. Similarly, we define $\eta(y)$ for $y\in Y$. Then their $\ol{d}$-distance can be defined as follows:
$$\ol{d}(\xi,\eta):=\inf\limits_{\lambda \in J(\mu,\nu)} \lambda\{(x,y) \colon \xi(x) \neq \eta(y)\}.$$
This definition of the $\ol{d}$-distance can be extended to measure distance between two sequences of partitions $\{\xi_i\}_1^n$ on $X$ and $\{\eta_i\}_1^n$ on $Y$. First, for any $x\in X$ and $y\in Y$ we set
$$h(x,y): = \frac{1}{n}\sum_{i=1}^{n}\delta\{\xi_i(x) \neq \eta_i(y)\}$$
where $\delta$ is the indicator function. Using this function $h$, we define
$$
\ol{d}(\{\xi_i\}_1^n,\{\eta_i\}_1^n) := \inf\limits_{\lambda \in J(\mu,\nu)} \int h(x,y) \, d\lambda(x,y).
$$
Most of the time throughout the section, the space $Y$ will be a subset $E$ of $X$ equipped with the measure $\mu|_E$ conditioned on $E$ defined by
$$\mu|_E(F) := \mu(F)/\mu(E)$$
for any $F \subseteq E$. Moreover, the partition on $E \subset X$ will often be of the form $\xi \mid E$ where $\xi$ is a partition on $X$. The partition $\xi\mid E$ consists of elements $A \cap E$, $A \in \xi$.
Using these definitions, we introduce the notion of Very Weak Bernoulli partitions.
\begin{defn}
Let $f : X\to X$ be a $\mu$-measure-preserving transformation on a compact metric space, and let $\xi$ be a finite partition. We say that $\xi$ is \textit{Very Weak Bernoulli} (VWB) if for any $\ep > 0$, there exists $N_0\in \N$ such that for all $m_2\geq m_1 \geq N_0$, for all $n > 0$, and for $\ep$-almost every $E\in\bigvee_{m_1}^{m_2}f^k\xi$, we have
$$\bar{d}( \{f^{-i}\xi\mid E\}_1^n, \{f^{-i}\xi\}_1^n) \leq \ep$$
with respect to $(E,\mu|_E)$ and $(X,\mu)$.
\end{defn}
As the name suggests, a Bernoulli partition $\xi$ for $(X,\mu)$ is necessarily a Very Weak Bernoulli partition; see \cite[Theorem 3.1]{ponce2019introduction}. What makes Very Weak Bernoulli partitions important is the converse implication developed by Ornstein outlined in the next two propositions: VWB partitions provide a method for establishing the Bernoulli property, which is often difficult via a direct approach. In fact, many results in the literature establishing Bernoulli property follow this general outline \cite{pesin1977characteristic,ornstein1998bernoulli,chernov1996nonuniformly}.
\begin{prop}
If a partition $\xi$ of $X$ is VWB, then $(X,\mu,f,\bigvee\limits_{-\infty}^\infty f^{n}\xi)$ is Bernoulli.
\end{prop}
\begin{prop} Suppose there exists an increasing sequence of partitions $\{\xi_i\}_{i\in\N}$ such that $\displaystyle\bigvee_{i=1}^\infty\bigvee_{n=-\infty}^\infty f^{n}\xi_i$ generates the entire $\s$-algebra and that $(X,\mu,f,\bigvee_{-\infty}^\infty f^{n}\xi_i)$ is a Bernoulli shift for each $i\in \N$, then $(X,\mu,f)$ is a Bernoulli shift.
\end{prop}
We now describe a mechanism for verifying that a partition is VWB.
We will show this by means of the following lemma, which involves constructing maps which, informally, can be thought of as
\begin{enumerate}
\item Keeping \emph{most} points in the same set of partition elements
\item Being \emph{mostly} measure-preserving
\end{enumerate}
To formalize these two concepts we need to introduce one more definition.
\begin{defn}\label{defn: almost mp}
A map $\theta : (X,\mu)\to (Y,\nu)$ is \textit{$\ep$-measure-preserving} if there exists $G\subset X$ with $\mu(G) > 1-\ep$ such that for $A\subset G$,
$$\abs{\frac{\nu(\theta A)}{\mu(A)} - 1} < \ep.$$
\end{defn}
Using this, and recalling Definition \ref{defn: eps almost everywhere} of ``$\ep$-almost everywhere'', we state the following lemma from \cite{chernov1996nonuniformly}.
\begin{lem}\label{lem: dbar closeness}
Let $\{\xi_i\}_{1}^n$ and $\{\eta_i\}_{1}^{n}$ be finite sequences of partitions of $(X,\mu)$ and $(Y,\nu)$ respectively. Let $\theta : X\to Y$ be $\ep$-measure-preserving, and suppose that there exists $G\subset X$ with $\mu(G) > 1-\ep$ so that for all $x\in G$,
$$h(x,\theta(x)) < \ep.$$
Then $\bar{d}(\{\xi_i\}_1^n,\{\eta_i\}_1^n) < C\ep$ for a constant $C$ independent of both the partitions and $\ep$.
\end{lem}
Applying this to show that a partition is Very Weak Bernoulli, we have the following corollary which we will use to establish Theorem \ref{thm: abstract}.
\begin{cor}\label{cor: VWB}
Let $\xi$ be a finite partition of $(X,\mu)$, and $f : X\to X$ measure-preserving. Suppose that for all $\ep > 0$, there exists $N\in \N$ so that for any $m_2\geq m_1\geq N$, and for $\ep$-almost every $E\in\bigvee_{m_1}^{m_2} f^i\xi$, there exists a map $\theta : (E,\mu|_E)\to (X,\mu)$ satisfying:
\begin{itemize}
\item $\theta$ is $\ep$-measure-preserving;
\item There exists $D\subset E$ with $\mu|_E(D) > 1-\ep$ so that for all $n > 0$ and $x\in D$, with respect to the partition sequences $\{f^{-i}\xi \mid E\}_1^{n}$ and $\{f^{-i}\xi\}_1^n$, we have $$h(x,\theta(x)) < \ep.$$
\end{itemize}
Then $\xi$ is Very Weak Bernoulli.
\end{cor}
\subsection{Useful Facts from Measure Theory}\label{subsec: measure theory}
We will make repeated use of the following results. Throughout, we will assume that $\ep < \frac{1}{5}$. As we will eventually need our results to hold only for $\ep$ arbitrarily close to $0$, we can freely make this assumption.
\begin{lem}\label{lem: large intersections}
Suppose $\mu(B) > 1-\ep$ and $A_1,\cdots, A_n$ are disjoint sets. Then the union of all $A_i$ satisfying $\mu(A_i\cap B) < (1-\delta) \mu(A_i)$ has measure at most $\ep/\delta$.
\end{lem}
\begin{proof}
Let $\mathscr{A} = \{A_i\mid \mu(A_i\cap B) < (1-\delta)\mu(A_i)\}$. Now observe
\begin{align*}
\mu(\bigcup_{A_i\in\mathscr{A}} A_i) &= \sum \mu(A_i\cap B) + \mu(A_i\cap B^c)
\\
&\leq (1-\delta)\sum \mu(A_i) + \mu(B^c)
\\
&\leq (1-\delta)\mu(\bigcup_{\mathscr{A}} A_i) + \epsilon.
\end{align*}
Thus, $\displaystyle \mu(\bigcup_{A_i\in\mathscr{A}} A_i) < \frac{\ep}{\delta}$.
\end{proof}
We will often work with measures which we must later rescale to probability measures. This result simplifies our calculations in these cases.
\begin{prop}\label{prop: rescaling derivative}
Let $\ep > 0$ and $r\geq 2\ep + \ep^2$. Now suppose $(X,\mu)$ and $(X,\nu)$ are measure spaces, and $A\subset X$. If $\displaystyle \frac{d\mu}{d\nu}\Big|_A\in (r-\ep^2,r+\ep^2)$, then $\displaystyle \frac{d\mu|_A}{d\nu|_A}\in (1-\ep,1+\ep)$.
\end{prop}
\begin{proof}
Let $B\subset A$ be arbitrary. Then
$$\mu|_A(B) = \frac{\mu(B)}{\mu(A)} \leq \frac{\nu(B)(r+\ep^2)}{\nu(A)(r-\ep^2)} = \nu|_A(B)\frac{r+\ep^2}{r-\ep^2}.$$
Now, observe that
$$\frac{r+\ep^2}{r-\ep^2} = 1 + \frac{2\ep^2}{r-\ep^2} \leq 1 + \ep$$
so long as $r\geq 2\ep + \ep^2.$ Similarly, we see that
$$\mu|_A(B) \geq \frac{r-\ep^2}{r+\ep^2}\nu|_A(B) = (1 - \frac{2\ep^2}{r+\ep^2})\nu|_A(B).$$
As
$\displaystyle \frac{2\ep^2}{r+\ep^2}\leq \ep,$
we have the reverse inequality.
\end{proof}
The following proposition shows that if we have a measure-preserving function on most of the space, then we can change the measures slightly and the function will still be $\ep$-measure-preserving.
\begin{prop}\label{prop: perturbation of measure preserving}
Let $\ep>0$ be arbitrary. Suppose $X_1,X_2$ are metric spaces, and $\mu_1,\nu_1$ are probability measures on $X_1$, and $\mu_2,\nu_2$ are probability measures on $X_2$. Now, suppose that we have $Y_1\subset X_1$ and $Y_2\subset X_2$ satisfying the following conditions:
\begin{enumerate}
\item On $Y_1$, $\frac{d\mu_1}{d\nu_1}\in (1-\ep,1+\ep)$;
\item On $Y_2$, $\frac{d\mu_2}{d\nu_2}\in (1-\ep,1+\ep)$;
\item $\mu_1(Y_1)\geq 1-\ep$ and $\mu_2(Y_2) \geq 1-\ep$.
\end{enumerate}
Then, if $\theta : X_1\to X_2$ is injective and measure-preserving as a map from $(X_1,\nu_1)$ to $(X_2,\nu_2)$, it is $5\ep$-measure-preserving as a map from $(X_1,\mu_1)$ to $(X_2,\mu_2)$.
\end{prop}
\begin{proof}
First, let $A\subset Y_1\cap \theta^{-1}Y_2$. Then observe
\begin{align*}
\mu_2(\theta A) &= \int_{\theta A}\frac{d\mu_2}{d\nu_2}\,d\nu_2
\\
&\leq (1+\ep)\nu_2(\theta A)
\\
&= (1+\ep)\nu_1(A)
\\
&= (1+\ep)\int_A \frac{d\nu_1}{d\mu_1}\,d\mu_1
\\
&\leq \frac{1+\ep}{1-\ep}\mu_1(A)
\\
&\leq (1 + 5\ep)\mu_1(A).
\end{align*}
The reverse inequality follows similarly.
Now, we will now show that $\mu_1(Y_1\cap \theta^{-1}(Y_2)) \geq 1 - 5\ep$. First, observe that
$$\nu_1(Y_1) \geq \frac{1}{1+\ep}\mu_1(Y_1) \geq \frac{1-\ep}{1+\ep} \geq 1 - 2\ep$$
with the same proof showing $\nu_2(Y_2) \geq 1 - 2\ep$.
Observe that
\begin{align*}
\mu_1(Y_1\cap \theta^{-1}(Y_2)) &\geq (1-\ep)\nu_1(Y_1\cap \theta^{-1}(Y_2))
\\
&\geq (1-\ep)(\nu_1(\theta^{-1}(Y_2)) -2\ep)
\\
&= (1-\ep)(\nu_2(Y_2) - 2\ep)
\\
&\geq (1-\ep)(1 - 4\ep)
\\
&\geq 1 - 5\ep
\end{align*}
where the second inequality follows because $\nu_1(Y_1^c) \leq 2\ep$.
\end{proof}
\subsection{Rectangles}
From here on, we will write $\chi\in (0,1)$ for the hyperbolicity constant of $f$, that is, local stable (resp. unstable) sets contract by $\chi$ under $f$ (resp. $f^{-1}$). Furthermore, we will assume that $\xi$ is an arbitrary partition of our space $X$ with the property that there exists $C > 0$ such that for all $\ep > 0$,
\begin{equation}\label{eq: boundary measure}
\mu(B(\partial \xi, \ep)) < C\ep.
\end{equation}
The existence of such partitions with arbitrarily small diameter can be found in \cite[Lemma 4.1]{ornstein1998bernoulli}.
We now introduce the notion of rectangles, which will serve as the building blocks when we construct the map $\theta$ from Corollary \ref{cor: VWB}.
\begin{defn}
A \textit{rectangle} $R \subset X$ is a set such that for $x,y\in R$, $W^s(x)\cap W^u(y) \in R$.
\end{defn}
In order to define $\theta$, we need the notion of a layerwise intersection.
\begin{defn}
We say that $A$ intersects a rectangle $R$ \textit{layerwise} if given $x\in A$, the unstable set of $x$ stretches across the entirety of $R$, i.e.
$$W^u(x)\cap R \subset W^u(x)\cap A.$$
\end{defn}
We will now show that ``most'' $E$ of $\bigvee_{m_1}^{m_2} f^i\xi$ satisfy the necessary properties above.
\begin{prop}\label{prop: layerwise} Let $\mathcal{R} = \{R_i\}_{1}^n$ be a finite set of disjoint rectangles in $X$.
For all $\ep > 0$, there exists $N \in \N$ such that for all $m_2 \geq m_1 \geq N$, for $\ep$-almost every $E\in \bigvee_{m_1}^{m_2} f^i\xi$, there exists $A\subset E$ which intersects each $R_i$ layerwise and satisfies $\mu(A) \geq (1-\ep)\mu(E)$.
\end{prop}
\begin{proof}
For $ x\in \bigcup_{R\in\mathcal{R}}R$, let $R(x) \in \mathcal{R}$ be the rectangle which contains $x$. For any $F \in \xi$ and $k\in \N$, we denote the subset of $f^kF$ consisting of points that do not intersect the rectangles in $\mathcal{R}$ layerwise as
$$A^k_F := \Big\{x\in f^k F \cap \bigcup_{R\in\mathcal{R}}R \bigm| W^u(x)\cap R(x) \not\subset f^k F\cap R(x)\Big\}.$$
Observe that if $x\in f^{-k}A^k_F$, then $d(x,\partial F) <C \chi^{k}$ for some constant $C$, where $\chi$ is the hyperbolicity constant for $f$.
Consequently, defining $\displaystyle B_k = \bigcup_{F\in \xi}A^k_F $ to be the union of all such ``bad'' parts of our atoms, by our choice of partition \eqref{eq: boundary measure}, it follows that $\mu(B_k) \leq C\chi^{k}$ for some other constant $C$. Now choose $N \in \N$ large enough so that $\displaystyle B := \bigcup_{k=N}^{\infty} B_k$ satisfies $\mu(B) < \ep^2$.
Let $m_2 \geq m_1 \geq N$ be arbitrary. As $\mu(B) < \ep^2$, by Lemma \ref{lem: large intersections}, for $\ep$-almost every $E\in \bigvee_{m_1}^{m_2}f^i\xi$, we have $\mu(E\cap B^c) \geq (1-\ep)\mu(E)$. For such $E$, we set $A = E\cap B^c$ which intersects each $R_i$ layerwise, as required.
\end{proof}
\subsection{Creating the mapping Rectangles}
The following proposition essentially says that if we have two equivalent measures, we can divide most of our space up into ``nice'' subsets, on which the ratio of our two measures does not change much. Applying this result to dynamical rectangles formalizes the argument that we need only measurable local product structure to apply the classical argument that $K$ implies Bernoulli.
\begin{prop}\label{prop: creation of general covering}
Suppose that $\mu$ and $\lambda$ are probability measures on a compact metric space $X$ with $\mu\ll\lambda$, and $\mathscr{C}$ is a generating semi-ring. Then for all $\ep_0 > 0$, there exists a pairwise disjoint collection $\{R_i\}\subset\mathscr{C}$ equipped with the measure $\lambda_i := \lambda|_{R_i}$ satisfying the following:
\begin{enumerate}
\item $\mu(\bigcup R_i) > 1 - \ep_0$;
\item There exists $G_i\subset R_i$ such that $\mu|_{R_i}(G_i) > 1 - \ep_0$ and $\lambda_i(G_i) > 1 -\ep_0$;
\item For each $R_i$, $\frac{d\mu|_{R_i}}{d\lambda_i} \in (1-\ep_0,1+\ep_0)$ restricted to $G_i$;
\item $\mu(\bigcup_i G_i) \geq 1-\ep_0$.
\end{enumerate}
\end{prop}
\begin{proof}
Let $\ep=\ep_0/4$ which be the scale at which we will carry out the construction below. Now, because $\mu \ll\lambda$, $\frac{d\mu}{d\lambda}$ exists and is non-zero on a set $Y_0\subset X$ of full $\mu$-measure. Now, by regularity of $\mu$ and $\lambda$, take $Y_1\subset Y_0$ to be compact, such that $\mu(Y_1) > 1-\ep^4$ and $\lambda(Y_1) > (1-\ep^4)\lambda(Y_0)$. Then, applying Lusin's theorem, there exists a compact set $K\subset Y_1$ on which $\frac{d\mu}{d\lambda}\Big|_K$ is continuous, and we have $\mu(K) > 1 - \ep^3$, and $\lambda(K) > (1-\ep^3)\lambda(Y_0)$.
By continuity of the derivative, compactness of $K$, and the fact that the derivative is non-zero on $K$, we have definite upper and lower bounds
$$0 < m \leq \frac{d\mu}{d\lambda}\Big|_K \leq M < \infty.$$
Furthermore, take $\gamma < \ep$ small enough so that $m\geq 2\gamma + \gamma^2$ in order to apply Proposition \ref{prop: rescaling derivative}. Then, by continuity of the derivative, we can choose $\delta > 0$ such that on any set $R$ with diameter at most $\delta$, there exists $r \in [m,M]$ such that $\displaystyle \frac{d\mu}{d\lambda}\Big|_{K\cap R} \in (r - \gamma^2,r+\gamma^2)$. In particular, Proposition \ref{prop: rescaling derivative} and the choice of $\gamma$ gives $\displaystyle \frac{d\mu|_{K\cap R}}{d\lambda|_{K\cap R}} \in (1-\ep,1+\ep)$.
Moving forward, we will fix $0 < \beta < \ep$ so that if $\lambda(A) < \beta^3$, then $\mu(A) < \ep^3$ for all $A\subset X$. Such a $\beta$ is guaranteed because $\mu\ll\lambda$. Now, since $\mathscr{C}$ is a generating semi-ring, there exists a finite collection of pairwise disjoint sets $\mathscr{R}\subset\mathscr{C}$ of diameter at most $\delta$ for which we can approximate $K$ arbitrarily well. We choose $\mathscr{R}$ so that
\begin{equation}\label{eq: scribed R}
\lambda\Big(K\vartriangle \bigcup_{R\in\mathscr{R}}R\Big) < \beta^4.
\end{equation}
This implies that the union of $R\in \mathscr{R}$ with $\lambda(R\cap K) < (1-\ep)\lambda(R)$ has very small measure with respect to $\lambda$. In particular, letting $\mathscr{B}_{\lambda}$ denote this set, the analogous argument of Lemma \ref{lem: large intersections} applied using \eqref{eq: scribed R} gives
\begin{align*}
\lambda\left(\bigcup_{\mathscr{B}_{\lambda}}R\right) &\leq \frac{1}{\ep}\sum_{\mathscr{B}_{\lambda}}\lambda(R\cap K^c)
\\
&\leq \beta^3.
\end{align*}
Therefore, $\displaystyle \mu\Big(\bigcup_{\mathscr{B}_\lambda}R\Big) < \ep^3$ from the choice of $\beta$.
Furthermore, again by our choice of $\beta$ and \eqref{eq: scribed R}, we know that $\displaystyle \mu(K\vartriangle \bigcup_{R\in\mathscr{R}}R) < \ep^3$, and so, defining $\mathscr{B}_\mu$ analogously as the set of $R\in \mathscr{R}$ with $\mu(R\cap K) < (1-\ep)\mu(R)$, we have that
$$\mu\left(\bigcup_{\mathscr{B}_{\mu}}R\right) \leq \frac{1}{\ep}\sum_{\mathscr{B}_{\mu}}\mu(R\cap K^c)\leq \ep^2.$$
Consequently, $\displaystyle \mu(\bigcup_{\mathscr{R}\setminus \mathscr{B}_\lambda\cup \mathscr{B}_\mu} R) > 1 - 2\ep^2$. Henceforth, we will take our collection $\{R_i\}_{i=1}^n$ to be $\mathscr{R}\setminus (\mathscr{B}_\mu \cup \mathscr{B}_{\lambda})$, and we will take
$$G_i := R_i\cap K.$$
In other words, our collection $\{R_i\}_{i=1}^n$ consists of $R\in \mathscr{R}$ whose intersection with $K$ has large relative measure with respect to both $\mu$ and $\lambda$.
We can now proceed to check our desired conditions. First, observe that
$$\mu(K\setminus \bigcup R_i) \leq \mu(K\setminus \bigcup_{\mathscr{R}}R) + 2\ep^2 \leq \ep^3 + 2\ep^2 \leq 3\ep^2.$$
Therefore, since $\mu(K) > 1 - \ep^3$, we have that
$$\mu(\bigcup R_i) \geq \mu(K) - \mu(K\setminus \bigcup R_i) > 1 - \ep^3 - 3\ep^2 > 1 - \ep_0.$$
The second condition, that $\mu(G_i) \geq (1-\ep)\mu(R_i)$, follows because $R_i\notin \mathscr{B}_\mu$, and similarly for $\lambda$ using $\mathscr{B}_\lambda$. The fourth condition also readily follows from the first two:
$$\mu(G) = \sum_i \mu(G_i) \geq (1-\ep) \sum_i \mu(R_i) \geq (1-\ep)^2 \geq 1-2\ep.$$
We are left with the third condition. First, recall $G_i$ has relatively large $\mu$ and $\lambda$ measure in $R_i$, so we have
$$1-\ep < \frac{d\mu|_{R}}{d\mu|_{G}} \leq 1 \quad \text{ and }\quad 1 \leq \frac{d\lambda|_G}{d\lambda|_R} < \frac{1}{1-\ep} \leq 1 + 2\ep.$$
Additionally, recall from the construction that elements of $\mathscr{R}$ have small enough diameter so that $\frac{d\mu|_{G}}{d\lambda|_{G}} \in (1-\ep,1+\ep).$ Therefore, on $G$ we have
$$\frac{d\mu|_R}{d\lambda|_R} = \left(\frac{d\mu|_R}{d\mu|_G}\right)\left(\frac{d\mu|_G}{d\lambda|_G}\right)\left(\frac{d\lambda|_G}{d\lambda|_R}\right) \in ((1-\ep)^2,(1+\ep)(1+2\ep))\subset (1-2\ep,1+4\ep).$$
Recalling that $\ep_0=4\ep$, we are done.
\end{proof}
\begin{rem}\label{rem: delta scale}
Note that we could have chosen $\delta$, which served as the maximum diameter of $\{R_i\}$, as small as we desired. In particular, given any prescribed $\delta>0$, we could have ensured that the diameter of each element of $\{R_i\}$ is bounded above by $\delta$.
This fact will be used in Proposition \ref{prop: culmination}.
\end{rem}
This proposition can almost be interpreted as saying that there is an $\ep$-regular covering as defined by Chernov and Haskell \cite{chernov1996nonuniformly}. This follows because we can take our generating semi-ring to be the set of rectangles. The only missing component is that we do not make any claims regarding stable or unstable manifolds.
\subsection{Defining our map $\theta$}
For the rest of the section, we will suppose that for any scale $\ep_0>0$, the collection $\{R_i\}$ obtained from Proposition \ref{prop: creation of general covering} at scale $\ep_0$ satisfies the following assumptions:
\begin{itemize}
\item Each $R_i$ is a rectangle;
\item Each measure $\lambda_i:=\lambda|_{R_i}$ is a non-atomic product measure $\rho_i\x\nu_i$ on $R_i$, where $\rho_i$ is a measure on $W^s(z)$ and $\nu_i$ is a measure on $W^u(z)$ for some $z \in R_i$.
\end{itemize}
As rectangles are a generating semi-ring, the first assumption provides no additional restrictions in our setting. The second assumption is the content of measurable local product structure.
We now begin the proof of Theorem \ref{thm: abstract} by using Corollary \ref{cor: VWB}. Given a partition $\xi$ of $X$, recall that we want to build an $\ep_0$-measure preserving map $\theta : E \to X$ for $\ep_0$-almost every atom $E\in\bigvee_{m_1}^{m_2} f^i\xi$ satisfying extra properties listed in Corollary \ref{cor: VWB}. We do this by using the rectangles $\{R_i\}$ created in the previous subsection in the order of scale $\ep_0^2$. In particular, we will find a suitable subset $A\subset E$ and construct $\theta : A\cap R_i \to R_i$ for each $i$. Putting these maps together (over all $i$) and extending it to $E$ will be our desired map.
The following proposition is the most technical part of the proof. We will use an auxiliary measure $\lambda_0$ built out of $\lambda_i = \lambda|_{R_i}$ where $\{R_i\}_{i=1}^n$ is the collection of rectangles at certain scale.
\begin{prop}\label{prop: most E}
For any $\ep_0>0$, let $\{R_i\}_{i=1}^n$ and $G_i \subset R_i$ be the rectangles and their designated subsets created from Proposition \ref{prop: creation of general covering} at scale $(\ep_0/4)^2$.
Setting $G := \bigcup_{i=1}^n G_i$ and
\begin{equation}\label{eq: lambda_0}
\lambda_0 := (\sum_i \mu(R_i))^{-1}\Big( \sum_{i=1}^n\mu(R_i)\lambda_i\Big),
\end{equation}
there exists $N\in \N$ such that for all $m_2 \geq m_1 \geq N$, for $\ep_0$-almost every $E\in\bigvee_{m_1}^{m_2} f^i\xi$ with respect to $\mu$, there exists a subset $A\subset E$ such that
\begin{enumerate}[label=(\alph*)]
\item $A$ intersects each $R_i$ layerwise;
\item $\mu(A) \geq (1-\ep_0)\mu(E)$;
\item $\mu(A\cap G)\geq (1-\ep_0)\mu(A)$;
\item $\lambda_0(A\cap G)\geq (1-\ep_0)\lambda_0(A)$.
\end{enumerate}
Moreover, there exists a $10\ep_0$-measure-preserving map $\theta : (A,\mu|_A)\to (X,\mu)$ where $\theta(x) \in W^s(x)$ for all $x\in A$.
\end{prop}
We will split the proof into two parts: Lemma \ref{lem: construct A} and \ref{lem: second half}.
First, we will show that there exists $N\in \N$ such that (a)-(d) hold on $A$. Then, we will show that conditions (a)-(d) on $A$ imply the existence of the map $\theta$ with the claimed properties. In both proofs, we will need the following short lemma to compare $\lambda_0$ and $\mu$.
\begin{lem}\label{lem: B and G} Let $\ep = \ep_0/4$.
For any $B\subseteq X$, we have
$$(1-2\ep^2)\lambda_0(B \cap G) \leq \mu(B \cap G)
\leq (1+\ep^2)\lambda_0(B \cap G).$$
\end{lem}
\begin{proof}
Note that we have
$\mu(B \cap G) = \sum\limits_{i=1}^n \mu(B \cap G_i) = \sum\limits_{i=1}^n \mu(R_i)\cdot \mu |_{R_i}(B \cap G_i).$ Since $\ep^2= (\ep_0/4)^2$ is the scale of the rectangles $\{R_i\}_{1}^n$, from Proposition \ref{prop: creation of general covering} (3) we have $\displaystyle \frac{d\mu|_{R_i}}{d\lambda_i} \in (1-\ep^2,1+\ep^2)$ on $G_i$, and it follows from the definition \eqref{eq: lambda_0} of $\lambda_0$ that
$$(1-\ep^2)\left(\sum_i \mu(R_i)\right)\lambda_0(B \cap G) \leq \mu(B \cap G)
\leq (1+\ep^2)\left(\sum_i \mu(R_i)\right)\lambda_0(B \cap G).$$
As $\sum_i \mu(R_i) \in (1-\ep^2,1)$ from Proposition \ref{prop: creation of general covering} (1), we have the desired inequalities.
\end{proof}
Now, we will show that $\ep_0$-almost every $E\in \bigvee_{m_1}^{m_2} f^i\xi$ has a subset $A\subset E$ satisfying conditions $(a)$-$(d)$.
\begin{lem}\label{lem: construct A}
In the setting of Proposition \ref{prop: most E}, there exists $N\in \N$ such that for all $m_2 \geq m_1 \geq N$, for $\ep_0$-almost every $E\in\bigvee_{m_1}^{m_2} f^j\xi$, we have $A\subset E$ satisfying (a)-(d).
\end{lem}
\begin{proof}
Let $\ep = \ep_0/4$ as done in the lemma above.
Take $N \in \N$ large enough so that Proposition \ref{prop: layerwise} holds at scale $\ep/2$ with respect to $\{R_i\}_{i=1}^n$.
Because $\mu$ has the $K$-property, from Proposition \ref{prop: K-partitions} we may increase $N$ if necessary so that for all $m_2 \geq m_1 \geq N$, we can ensure that for $\ep/2$-almost every $E\in \bigvee_{m_1}^{m_2} f^j\xi$ and for each $i$, we have
\begin{equation}\label{eq: K apply}
\abs{\frac{\mu(E\cap G_i)}{\mu(E)} - \mu(G_i)} \leq \frac{\ep}{2} \cdot \frac{1}{n}
\end{equation}
where $n$ is the number of rectangles $R_i$.
Therefore, for $\ep$-almost every $E$ in our partition, both this property and the results of Proposition \ref{prop: layerwise} hold; namely,
there exists of a subset $A\subset E$ layerwise intersecting each $R_i$ and satisfying
\begin{equation}\label{eq: A and E}
\mu(A) \geq (1-\ep/2)\mu(E).
\end{equation}
This verifies (a) and (b).
Recall that we have chosen our rectangles $\{R_i\}$ to be built at scale
$(\ep_0/4)^2 = \ep^2$
from Proposition \ref{prop: creation of general covering}. Therefore, we have
\begin{equation}\label{eq: mu(G)}
\mu(G) \geq 1-\ep^2.
\end{equation}
In particular, for all such $E$, summing \eqref{eq: K apply} over all $i$ gives
\begin{equation}\label{eq: E n G}
\mu(E\cap G) = \sum\limits_{i=1}^n \mu(E \cap G_i) \geq \mu(E)(\mu(G) - \ep/2) \geq (1-\ep)\mu(E).
\end{equation}
Now, combined the fact \eqref{eq: A and E} that $A$ has large $\mu$-measure in $E$, we verify (c):
$$\mu(A\cap G) \geq \mu(E\cap G) - \mu(E\setminus A) \geq (1-3\ep/2)\mu(E) \geq (1-\ep_0)\mu(A).$$
The more difficult estimate to check is the fourth condition (d) on $\lambda_0$. While conditions (a)-(c) are relatively easy to check because $A\subset E$ has been constructed to satisfy relevant properties with respect to $\mu$, a priori we cannot compare $\lambda_0(A)$ to $\lambda_0(E)$ or $\lambda_0(A \cap G)$ to $\lambda_0(E \cap G)$ as we did for $\mu$. So, we make use of Lemma \ref{lem: B and G}. The other difficulty is that we must further restrict the elements of $\bigvee_{m_1}^{m_2}f^i\xi$ with which we work. Henceforth, we will assume that we are working with partition elements $E$ which satisfy properties (a)-(c) (which we have already shown to be $\ep$-almost every partition element), in addition to the following extra assumption:
$$\lambda_0(E\cap G) \geq (1-\ep)\lambda_0(E).$$
Now, Lemma \ref{lem: B and G} and \eqref{eq: E n G} give
$$\mu(E) \leq \frac{1}{1-\ep}\mu(E\cap G) \leq \frac{1+\ep^2}{1-\ep}\lambda_0(E\cap G) \leq (1+2\ep)\lambda_0(E).$$
In the reverse direction, Lemma \ref{lem: B and G} and \eqref{eq: A and E} show
$$\frac{\ep}{2} \mu(E) \geq \mu(E\setminus A) \geq \mu(G\cap (E\setminus A)) \geq (1-2\ep^2)\lambda_0(G \cap (E \setminus A)).$$
Combining the two gives $\displaystyle \lambda_0(G \cap (E \setminus A)) \leq \frac{\ep(1+2\ep)}{2(1-2\ep^2)} \lambda_0(E) \leq \ep \lambda_0(E)$.
Using the extra assumption on $\lambda_0(E\cap G)$, we establish (d):
\begin{align*}
\lambda_0(A\cap G) &\geq \lambda_0(E\cap G) - \lambda_0(G\cap (E\setminus A))
\\
&\geq (1-\ep)\lambda_0(E) - \ep\lambda_0(E)
\\
&\geq (1 - \ep_0)\lambda_0(A).
\end{align*}
In order to complete our proof, we need to justify our assumption that $\lambda_0(E\cap G) \geq (1-\ep)\lambda_0(E)$ by showing that the set of such partition elements has large $\mu$-measure.
Hence, recalling that we have already shown that $\ep$-almost every partition element $E \in\bigvee_{m_1}^{m_2}f^i\xi$ satisfies (a) through (c), we only need to show that our assumption of $\lambda_0(E\cap G) \geq (1-\ep)\lambda_0(E)$ is justified on most such $E$.
Observe that $\lambda_0(G^c) < \ep^2$, as we have
\begin{align*}
\lambda_0(G^c) &= \left(\sum_i \mu(R_i)\right)^{-1} \sum_i \mu(R_i)\lambda_i(R_i\setminus G_i)
\\
&\leq \left(\sum_i \mu(R_i)\right)^{-1} \sum_i \mu(R_i)\ep^2= \ep^2
\end{align*}
Therefore, Lemma \ref{lem: large intersections} applies and the $\lambda_0$-measure of the union of all partition elements $E \in \bigvee_{m_1}^{m_2} f^i\xi$ with $\lambda_0(E\cap G) < (1-\ep)\lambda_0(E)$ is at most $\ep$. Among such partition elements, we denote by $\mathscr{B}$ the set of all $E$ which also satisfy properties (a)-(c). In particular, $E \in \mathscr{B}$ satisfies $\mu(E \cap G^c) \leq \ep\mu(E)$ which corresponds to \eqref{eq: E n G}.
Then we see that
\begin{align*}
\mu(\bigcup_{E\in\mathscr{B}} E) &= \sum_{\mathscr{B}} \mu(E\cap G) + \mu(E\cap G^c)
\\
&\leq \sum (1+\ep^2)\lambda_0(E\cap G) + \ep \mu(E)
\\
&\leq (1+\ep^2)\lambda_0(\bigcup_{E\in\mathscr{B}} E) + \ep\mu(\bigcup_{E\in\mathscr{B}} E)
\\
&\leq (1+\ep^2)\ep + \ep\mu(\bigcup_{E\in\mathscr{B}} E),
\end{align*}
where the first inequality uses Lemma \ref{lem: B and G}. Consequently,
$$\mu(\bigcup_{E\in\mathscr{B}}E) < \frac{1+\ep^2}{1-\ep}\ep < 2\ep.$$
As we have already shown that $\ep$-almost every element of $\bigvee_{m_1}^{m_2}f^i\xi$ satisfies properties (a)-(c), by further excluding the elements of $\mathscr{B}$, it follows that $3\ep$-almost every element satisfies (a)-(d). As $\ep_0 = 4\ep$, our proof is complete.
\end{proof}
We now construct the desired map $\theta$ using the properties (a), (c), and (d) on $A$ and the fact that $\{R_i\}$ is constructed at scale $(\ep_0/4)^2$.
The standing assumption that $\lambda_i$ is a product measure allows us to first build $\theta$ in a \textit{measure-preserving} way with respect to the auxiliary measure $\lambda_0$. Using this, $\theta$ can be shown to be an $\ep_0$-measure preserving map with respect to $\mu$.
\begin{lem}\label{lem: second half} For each $A \subset E$ from Lemma \ref{lem: construct A}, there exists a $10\ep_0$-measure-preserving map $\theta : (A,\mu|_A)\to (X,\mu)$ where $\theta(x) \in W^s(x)$ for all $x\in A$.
\end{lem}
\begin{proof}
We will continue to work with the scale $\ep = \ep_0/4$ as in the above lemma. Recall also that each rectangle $R_i$ can be written as a product $W^s(z)\x W^u(z)$ for some $z\in R_i$ and is equipped with the product measure $\rho_i \times \nu_i$.
We begin by defining $\theta : A\cap R_i \to R_i$. As $\rho_i$ is non-atomic, there is a measure-preserving mapping $\theta_i$ from $(A\cap W^s(z),\rho_i|_A)\to (R\cap W^s(z),\rho_i)$. Because $A$ intersects each $R_i$ layerwise from (a), this mapping extends to $A \cap R_i$ along the $W^u$-direction according to $\nu_i$. In particular, such a mapping $$\theta : (A\cap R_i, \lambda_i|_A)\to (R_i,\lambda_i)$$ is still measure-preserving, and satisfies the property that $\theta(x) \in W^s(x)$. Moreover, the map $\theta : (A\cap \bigsqcup R_i, \lambda_0|_A)\to (\bigsqcup R_i, \lambda_0)$ is measure-preserving since $\lambda_i|_A$ is a scalar multiple of $\lambda_0|_A$ on each $R_i$. Therefore,
$$\theta : (A,\lambda_0|_A) \to (X,\lambda_0)$$ is measure-preserving as well, because $\lambda_0(X\setminus \bigsqcup R_i) = 0$ from its construction \eqref{eq: lambda_0}.
We will now proceed to show that Proposition \ref{prop: perturbation of measure preserving} can be applied at scale $2\ep_0$ to $X_1 = A$, $X_2 = X$, $Y_1 = A\cap G$, and $Y_2 = G$ where our auxiliary measures will be $\nu_1=\lambda_0|_A$ and $\nu_2=\lambda_0$.
In particular, we will show how properties (c) and (d) of $A$ and the scale $\ep^2$ of $\{R_i\}$ can be used to verify the three conditions in Proposition \ref{prop: perturbation of measure preserving}.
First, at scale $2\ep_0 = 8\ep$, condition (3) of Proposition \ref{prop: perturbation of measure preserving} corresponds to
$\mu|_A(A \cap G) \geq 1-2\ep_0$ and $\mu(G) \geq 1-2\ep_0$
directly follows from (c) and \eqref{eq: mu(G)}.
From Lemma \ref{lem: B and G}, on $G$ we have
$$\frac{d\mu}{d\lambda_0} \in (1-2\ep^2,1+\ep^2),$$
which shows condition (2). Using Proposition \ref{prop: rescaling derivative},
it also shows that $\frac{d\mu|_{A\cap G}}{d\lambda_0|_{A\cap G}}$ is between $1-2\ep$ and $1+2\ep$.
Hence, to check condition (1) of Proposition \ref{prop: perturbation of measure preserving}, we observe that on $A \cap G$
\begin{align*}
\frac{d\mu|_{A}}{d\lambda_0|_{A}} &= \frac{d\mu|_{A}}{d\mu|_{A\cap G}}\frac{d\lambda_0|_{A\cap G}}{d\lambda_0|_{A}}\frac{d\mu|_{A\cap G}}{d\lambda_0|_{A\cap G}}
\\
&= \frac{\mu(A)}{\mu(A\cap G)}\frac{\lambda_0(A\cap G)}{\lambda_0(A)}\frac{d\mu|_{A\cap G}}{d\lambda_0|_{A\cap G}}
\\
&\leq\frac{1}{1-4\ep} \cdot 1 \cdot ( 1+2\ep)
\\
&\leq 1 + 7\ep,
\end{align*}
where the third line uses property (c).
A similar argument using (d) instead yields the reverse inequality, verifying condition (1).
Consequently, we have shown the necessary conditions of Proposition \ref{prop: perturbation of measure preserving} with constant $8\ep = 2\ep_0$, and so $\theta : (A,\mu|_A)\to (X,\mu)$ is $10\ep_0$-measure-preserving. This completes the proof of the lemma, and also of Proposition \ref{prop: most E}.
\end{proof}
\subsection{Putting it all together}
Recall that $\chi \in (0,1)$ is the hyperbolicity constant for $f$ introduced in the beginning of Subsection \ref{subsec: measure theory}.
\begin{prop}\label{prop: culmination}
For an arbitrary suitably nice partition $\xi$, for any $\delta>0$, and for all $\ep_0> 0$, there exists $N$ such that for all $m_2 \geq m_1\geq N$, for $\ep_0$-almost every $E\in\bigvee_{m_1}^{m_2} f^i\xi$, there exists an $\ep_0$-measure-preserving map $\theta : (E,\mu|_E)\to (X,\mu)$ such that for all $n\geq 0$,
\begin{equation}\label{eq: distance}
d(f^nx,f^n\theta(x)) \leq \chi^{n}\delta.
\end{equation}
\end{prop}
\begin{proof}
Let $\ep = \ep_0/2$.
From Proposition \ref{prop: most E} at scale $\ep$, we have shown that for all $\ep > 0$, there exists $N$ such that for all $m_2 \geq m_1 \geq N$, for $\ep$-almost every $E\in\bigvee_{m_1}^{m_2}f^i\xi$, there exists $A\subset E$ intersecting each $R_i$ layerwise such that $\mu|_E(A) > 1-\ep$ and that the map $\theta : (A,\mu|_A)\to (X,\mu)$ is $\ep$-measure-preserving.
Furthermore, because $\theta(x)$ belongs to the local stable set of $x$ and we can choose our rectangles with arbitrarily small diameter (see Remark \ref{rem: delta scale}), we have the additional required condition on $\theta$. Now, extend $\theta : (E,\mu|_E)\to (X,\mu)$ by setting $\theta(x) = x$ for $x\notin A$. This clearly preserves the distance condition on $\theta$, so all that we need to check is that the map is $\ep$-measure-preserving.
This can be verified from the fact that original map $\theta : (A,\mu|_A)\to (X,\mu)$ is $\ep$-measure preserving. Denoting by $B$ the subset of $A$ on which $\theta$ is $\ep$-measure preserving (see Definition \ref{defn: almost mp}), the same set $B$ can be used for verifying the $\ep$-measure preserving property for the extended map $\theta : (E,\mu|_E)\to (X,\mu)$.
Indeed, from its definition we have
$$\abs{\frac{\mu(\theta C)}{\mu(C)} - 1} < \ep$$
for any $C\subseteq B$.
Moreover, the set $B$ still has large $\mu$-measure in $E$ because $A$ does:
$$\mu(B) \geq (1-\ep)\mu(A) \geq (1-\ep)(1-\ep)\mu(E) \geq (1-2\ep)\mu(E).$$
Consequently, we see that $\theta : (E,\mu|_E)\to (X,\mu)$ is $2\ep$-measure-preserving. Recalling that $\ep_0 = 2\ep$, we are done.
\end{proof}
\begin{prop}
The partition $\xi$ is Very Weak Bernoulli.
\end{prop}
\begin{proof}
Let $\ep > 0$. We will apply Proposition \ref{prop: culmination} to $\ep$ and $\delta \in (0, \ep^2/C)$, where $C$ is the constant from \eqref{eq: boundary measure} so that $\mu(B(\partial \xi,\delta)) \leq \ep^2$. This gives $N\in \N$ such that $\ep$-almost every $E\in\bigvee_{m_1}^{m_2}f^i\xi$ with $m_2 \geq m_1\geq N$ has an $\ep$-measure-preserving map $\theta : E \to X$ satisfying the distance condition \eqref{eq: distance}. For reference later, we denote by $\mathscr{D}$ the set of all such $E$.
For any such $E \in \mathscr{D}$, consider the partitions $\{f^{-i}\xi \mid E\}_1^n$ and $\{f^{-i}\xi\}_1^n$. If $$h(x,\theta(x)) > \ep$$ for some $x\in E$ (we are just using $h(x,\theta(x))\neq 0$ here), this means that
there exists $1 \leq m \leq n$ so that $f^{-m}\xi|_E (x)\neq f^{-m}\xi(\theta(x))$; that is, $x$ and $\theta(x)$ belong to different partition elements of $f^{-m}\xi$. Therefore, $f^mx\in B(\partial \xi,\chi^m\delta)\subset B(\partial \xi,\delta)$ from \eqref{eq: distance}. Hence
$$B_E := \{x \in E \mid h(x,\theta(x)) > \ep \}\subset B(\partial \xi,\delta).$$
Setting $B = \bigcup\limits_{E \in \mathscr{D}} B_E$, we then have $\mu(B) \leq \ep^2$.
Therefore, applying Lemma \ref{lem: large intersections} which amounts to further excluding some elements from $\mathscr{D}$ whose union has $\mu$-measure at most $\ep$,
we now have that for $2\ep$-almost every $E\in\bigvee_{m_1}^{m_2}f^i\xi$, the subset $D:=E\cap B^c$ satisfies $\mu|_E(D) > 1-\ep$ and $h(x,\theta(x)) < \ep$ for $x\in D$.
Thus, $\xi$ is Very Weak Bernoulli from Corollary \ref{cor: VWB}, and our proof is complete.
\end{proof}
\begin{cor}
The system $(X,f,\mu)$ is Bernoulli.
\end{cor}
\begin{proof}
We can find a generating sequence of Very Weak Bernoulli partitions, because we can choose $\xi$ above to have arbitrarily small diameter. From the above proposition and Corollary \ref{cor: VWB}, the system $(X,f,\mu)$ is Bernoulli, completing the proof of Theorem \ref{thm: abstract}.
\end{proof}
\bibliographystyle{amsalpha}
\bibliography{bernoulli}
\end{document}
%\usepackage[latin9]{inputenc}
\usepackage[bottom=1in]{geometry}
\setcounter{secnumdepth}{3}
\setcounter{tocdepth}{3}
\usepackage{amsmath}
\usepackage{amsthm}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Textclass specific LaTeX commands.
\numberwithin{equation}{section}
\numberwithin{figure}{section}
\usepackage{enumitem} % customizable list environments
\newlength{\lyxlabelwidth} % auxiliary length
\let\footnote=\endnote
\@ifundefined{lettrine}{\usepackage{lettrine}}{}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Theorem enviroment
\theoremstyle{definition}
\newtheorem{thmx}{Theorem}
\renewcommand{\thethmx}{\Alph{thmx}}
\def\a{\alpha}
\def\A{\mathcal{A}}
\def\b{\beta}
\def\B{\mathcal{B}}
\def\s{\sigma}
\def\Si{\Sigma}
\def\G{\Gamma}
\def\g{\gamma}
\def\la{\lambda}
\def\L{\Lambda}
\def\D{\Delta}
\def\R{\mathbb{R}}
\def\F{\mathcal{F}}
\def\C{\mathbb{C}}
\def\H{\mathbb{H}}
\def\HH{\mathcal{H}}
\def\L{\mathcal{L}}
\def\T{\mathcal{T}}
\def\CC{\mathcal{C}}
\def\N{\mathbb{N}}
\def\Z{\mathbb{Z}}
\def\pa{\partial}
\def\ep{\varepsilon}
\def\vps{\varphi^s}
\def\Md{M_{d \times d}(\R)}
\def\I{\mathsf{I}}
\def\J{\mathsf{J}}
\def\K{\mathsf{K}}
\def\T{\mathcal{T}}
\def\F{\mathcal{F}}
\def\V{\mathbb{V}}
\def\W{\mathcal{W}}
\def\WW{\mathbb{W}}
\def\ol{\overline}
\def\id{\text{id}}
\newcommand{\Wloc}{\mathcal{W}_{\text{loc}}}
\def\vp{\varphi}
\def\hol{H\"older }
\def\wt{\widetilde}
\def\Sig{\Sigma_T}
\def\M{\mathcal{M}}
\def\U{\mathcal{U}}
\def\glr{\text{GL}_d(\R)}
\def\gltwo{\text{GL}_2(\R)}
\newcommand{\eps}{\epsilon}
\renewcommand{\phi}{\varphi}
\newcommand{\es}{\O}
\newcommand{\excise}[1]{}
\providecommand{\abs}[1]{\left | #1 \right |}
\providecommand{\norm}[1]{\lVert #1 \rVert}
\providecommand{\x}{\times}
\renewcommand{\_}{\underline}
\providecommand{\ar}{\rightarrow}
\providecommand{\arr}{\longrightarrow}
\renewcommand{\check}[1]{{\color{red} #1}}
%\renewcommand{\comment}[1]{}
%\newcommand{\s}[1]{\mathscr{#1}}
\providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace}
\providecommand{\MR}{\relax\ifhmode\unskip\space\fi MR }
% \MRhref is called by the amsart/book/proc definition of \MR.
\providecommand{\MRhref}[2]{%
\href{http://www.ams.org/mathscinet-getitem?mr=#1}{#2}
}
\providecommand{\href}[2]{#2}
\begin{thebibliography}{Bow74b}
\bibitem[BG11]{barreira2011dimension}
Luis Barreira and Katrin Gelfert, \emph{Dimension estimates in smooth dynamics:
a survey of recent results}, Ergodic Theory and Dynamical Systems \textbf{31}
(2011), no.~3, 641--671.
\bibitem[BG19]{bochi2019extremal}
Jairo Bochi and Eduardo Garibaldi, \emph{Extremal norms for fiber-bunched
cocycles}, Journal de l’{\'E}cole polytechnique—Math{\'e}matiques
\textbf{6} (2019), 947--1004.
\bibitem[BM18]{MR3820437}
Jairo Bochi and Ian~D. Morris, \emph{Equilibrium states of generalised singular
value potentials and applications to affine iterated function systems}, Geom.
Funct. Anal. \textbf{28} (2018), no.~4, 995--1028. \MR{3820437}
\bibitem[Bow74a]{bowen1974bernoulli}
Rufus Bowen, \emph{Bernoulli equilibrium states for axiom a diffeomorphisms},
Mathematical systems theory \textbf{8} (1974), no.~4, 289--294.
\bibitem[Bow74b]{bowen1974some}
\bysame, \emph{Some systems with unique equilibrium states}, Theory of
computing systems \textbf{8} (1974), no.~3, 193--202.
\bibitem[Bow75]{bowen1975ergodic}
\bysame, \emph{Equilibrium states and the ergodic theory of anosov
diffeomorphisms}, Lecture Notes in Mathematics, vol. 470, Springer-Verlag,
1975.
\bibitem[BP19]{butler2019thermodynamic}
Clark Butler and Kiho Park, \emph{Thermodynamic formalism of
$\text{GL}_2(\mathbb{R})$-cocycles with canonical holonomies}, Discrete \&
Continuous Dynamical Systems-A (2019), 0.
\bibitem[BV04]{bonatti2004lyapunov}
Christian Bonatti and Marcelo Viana, \emph{Lyapunov exponents with multiplicity
1 for deterministic products of matrices}, Ergodic Theory and Dynamical
Systems \textbf{24} (2004), no.~5, 1295--1330.
\bibitem[CFH08]{cao2008thermodynamic}
Yongluo Cao, De-Jun Feng, and Wen Huang, \emph{The thermodynamic formalism for
sub-additive potentials}, Discrete \& Continuous Dynamical Systems-A
\textbf{20} (2008), no.~3, 639--657.
\bibitem[CH96]{chernov1996nonuniformly}
NI~Chernov and Cymra Haskell, \emph{Nonuniformly hyperbolic k-systems are
bernoulli}, Ergodic theory and dynamical systems \textbf{16} (1996), no.~1,
19--44.
\bibitem[CP20]{call2020k}
Benjamin Call and Kiho Park, \emph{The k-property for subadditive equilibrium
states}, Dynamical Systems (2020), 1--24.
\bibitem[CPZ19]{cao2019dimension}
Yongluo Cao, Yakov Pesin, and Yun Zhao, \emph{Dimension estimates for
non-conformal repellers and continuity of sub-additive topological pressure},
Geometric and Functional Analysis \textbf{29} (2019), no.~5, 1325--1368.
\bibitem[CRH21]{carrasco2021equilibrium}
Pablo~D Carrasco and Federico Rodriguez-Hertz, \emph{Equilibrium states for
center isometries}, arXiv preprint arXiv:2103.07323 (2021).
\bibitem[CT19]{call2019equilibrium}
Benjamin Call and Daniel~J Thompson, \emph{Equilibrium states for products of
flows and the mixing properties of rank 1 geodesic flows}, arXiv preprint
arXiv:1906.09315 (2019).
\bibitem[Fen03]{feng2003lyapunov}
De-Jun Feng, \emph{Lyapunov exponents for products of matrices and multifractal
analysis. part i: Positive matrices}, Israel Journal of Mathematics
\textbf{138} (2003), no.~1, 353--376.
\bibitem[Fen09]{feng2009lyapunov}
\bysame, \emph{Lyapunov exponents for products of matrices and multifractal
analysis. part ii: General matrices}, Israel Journal of Mathematics
\textbf{170} (2009), no.~1, 355--394.
\bibitem[Fen11]{feng2011equilibrium}
\bysame, \emph{Equilibrium states for factor maps between subshifts}, Advances
in Mathematics \textbf{226} (2011), no.~3, 2470--2502.
\bibitem[FO70]{FriedmanOrnstein}
N.~A. Friedman and D.~S. Ornstein, \emph{On isomorphism of weak {B}ernoulli
transformations}, Advances in Math. \textbf{5} (1970), 365--394 (1970).
\MR{274718}
\bibitem[Kat80]{katok1980smooth}
Anatole Katok, \emph{Smooth non-bernoullik-automorphisms}, Inventiones
mathematicae \textbf{61} (1980), no.~3, 291--299.
\bibitem[KHV18]{kanigowski2018non}
Adam Kanigowski, Federico~Rodriguez Hertz, and Kurt Vinhage, \emph{On the
non-equivalence of the bernoulli and$\backslash$begin $\{$document$\}$ $ k
$$\backslash$end $\{$document$\}$ properties in dimension four}, Journal of
Modern Dynamics \textbf{13} (2018), no.~1, 221--250.
\bibitem[KS13]{kalinin2013cocycles}
Boris Kalinin and Victoria Sadovskaya, \emph{Cocycles with one exponent over
partially hyperbolic systems}, Geometriae Dedicata \textbf{167} (2013),
no.~1, 167--188.
\bibitem[LLS16]{ledrappier2016ergodic}
Fran{\c{c}}ois Ledrappier, Yuri Lima, and Omri~M Sarig, \emph{Ergodic
properties of equilibrium measures for smooth three dimensional flows},
Commentarii Mathematici Helvetici \textbf{91} (2016), no.~1, 65--106.
\bibitem[Mor18]{morris2018ergodic}
Ian~D Morris, \emph{Ergodic properties of matrix equilibrium states}, Ergodic
Theory and Dynamical Systems \textbf{38} (2018), no.~6, 2295--2320.
\bibitem[Mor19]{morris2019necessary}
\bysame, \emph{A necessary and sufficient condition for a matrix equilibrium
state to be mixing}, Ergodic Theory and Dynamical Systems \textbf{39} (2019),
no.~8, 2223--2234.
\bibitem[Mor20]{morris2020totally}
\bysame, \emph{Totally ergodic generalised matrix equilibrium states have the
bernoulli property}, arXiv preprint arXiv:2011.11497 (2020).
\bibitem[Orn70a]{Ornstein1970}
D.~S. Ornstein, \emph{Imbedding {B}ernoulli shifts in flows}, Contributions to
{E}rgodic {T}heory and {P}robability ({P}roc. {C}onf., {O}hio {S}tate
{U}niv., {C}olumbus, {O}hio, 1970), Springer, Berlin, 1970, pp.~178--218.
\MR{0272985}
\bibitem[Orn70b]{ornstein1970isomorphic}
Donald Ornstein, \emph{Bernoulli shifts with the same entropy are isomorphic},
Advances in Math. \textbf{4} (1970), 337--352. \MR{257322}
\bibitem[Orn73]{ornstein1973KnotBernoulli}
Donald~S. Ornstein, \emph{An example of a {K}olmogorov automorphism that is not
a {B}ernoulli shift}, Advances in Math. \textbf{10} (1973), 49--62.
\MR{316682}
\bibitem[OW73]{ornstein1973geodesic}
Donald Ornstein and Benjamin Weiss, \emph{Geodesic flows are bernoullian},
Israel Journal of Mathematics \textbf{14} (1973), no.~2, 184--198.
\bibitem[OW74]{OrnsteinWeissFinDet}
Donald~S. Ornstein and Benjamin Weiss, \emph{Finitely determined implies very
weak {B}ernoulli}, Israel J. Math. \textbf{17} (1974), 94--104. \MR{346132}
\bibitem[OW98]{ornstein1998bernoulli}
Donald Ornstein and Benjamin Weiss, \emph{On the bernoulli nature of systems
with some hyperbolic structure}, Ergodic Theory and Dynamical Systems
\textbf{18} (1998), no.~2, 441--456.
\bibitem[Par20]{park2019quasi}
Kiho Park, \emph{Quasi-multiplicativity of typical cocycles}, Commun. Math.
Phys. (2020), https://doi.org/10.1007/s00220-020-03701-8.
\bibitem[Pes77]{pesin1977characteristic}
Yakov Pesin, \emph{Characteristic lyapunov exponents and smooth ergodic
theory}, Uspekhi Matematicheskikh Nauk \textbf{32} (1977), no.~4, 55--112.
\bibitem[Pir18]{piraino2018weak}
Mark Piraino, \emph{The weak bernoulli property for matrix gibbs states},
Ergodic Theory and Dynamical Systems (2018), 1--20.
\bibitem[PTV18]{ponce2018bernoulli}
Gabriel Ponce, Ali Tahzibi, and R{\'e}gis Var{\~a}o, \emph{On the bernoulli
property for certain partially hyperbolic diffeomorphisms}, Advances in
Mathematics \textbf{329} (2018), 329--360.
\bibitem[PV19]{ponce2019introduction}
Gabriel Ponce and R{\'e}gis Var{\~a}o, \emph{An introduction to the
kolmogorov-bernoulli equivalence}, Springer, 2019.
\end{thebibliography}