Isomorphism and embedding of Borel systems on full sets (Q368692): Difference between revisions

A \textit{Borel system} \((X,T)\) consists of a standard Borel space \(X\) and a bi-measurable bijection \(T:X\to X\). As in many branches of dynamics, a central concern is the classification of such systems up to isomorphism or embedding. An \textit{isomorphism (Borel embedding)} of such systems is an isomorphism (resp. measurable injection) of the underlying standard Borel spaces which intertwines the two transformations. A Borel system is \textit{free} if it has no periodic points. For a general Borel system \((X,T)\), the periodic points form a \(T\)-invariant Borel subset. Their removal leaves a free Borel system called the \textit{free part} of \((X,T)\). This allows one to decompose many classification questions into the free and periodic parts of systems. The periodic parts are generally simple to understand, and so most work goes into the study of free systems. The Gurevich entropy of a Borel system is the supremum of the (Kolmogorov-Sinai) entropies of all invariant ergodic probability measures supported on that system (or \(-\infty\) if there are no such measures). If there is a Borel embedding from one Borel system to another, then the Gurevich entropy of the target system must be at least that of a source. A measurable subset of a Borel system is \textit{universally null} if it is null for every invariant ergodic measure on the system. A subset is \textit{full} if its complement is universally null. A Borel system \((X,T)\) is \textit{\(t\)-universal} if any other free Borel system \((Y,S)\) of Gurevich entropy \(<t\) embeds into \((X,T)\), possibly after removing a universally null set from \(Y\). The system \((X,T)\) is \textit{strictly \(t\)-universal} if in addition it has no ergodic invariant measures of entropy \(\geq t\). Any two strictly \(t\)-universal systems are isomorphic on full sets. (See Proposition 1.4 and Section 4 of this paper.) For a general Borel system \((X,T)\), one can find an invariant Borel subset \(X_t\) which supports all the ergodic invariant measures which have entropy \(<t\), and no others. This \(X_t\) is unique up to a universally null set. It is called the \textit{\(t\)-slice} of \((X,T)\). The main result of this paper is the existence of \(t\)-universal and strictly \(t\)-universal systems. Moreover, they are given by familiar examples: Theorem 1.5 asserts that a mixing SFT \((X,T)\) is \(h(X)\)-universal, and that its free \(t\)-slice is strictly \(t\)-universal for every \(t \leq h(X)\). From this, the author deduces that many other natural examples have \(t\)-slices that are strictly \(t\)-universal for \(t\) up to the entropy of the system (Theorem 1.6). These include mixing sofic shifts, mixing Markov shifts on countable state spaces, mixing axiom-A diffeomorphisms, mixing synchronized shifts, and beta shifts. This can then be turned into classification results up to isomorphism for several of the systems in this list (Theorems 1.7 and 1.8). After some preparatory sections, the hard work of the paper is in Section 5. By a result of Weiss, it suffices to prove the main result for the shift on \(\mathbb N^\mathbb Z\). Then the starting point of the analysis is a folklore enhancement of Krieger's generator theorem (Theorem 5.1 in this paper) which allows the embedding of many invariant measures simultaneously. The remainder of the section is occupied by the delicate construction of the coding maps required for Theorem 1.5. The paper is clearly and pleasantly written. The author makes the construction of the required codings very accessible, without obscuring any important technical details.