Exponential mixing implies Bernoulli (Q7019343)

The Bernoulli property (that it, being isomorphic in the measurable sense to a Bernoulli shift) sits at the top of a hierarchy of notions of chaotic or random dynamical behaviour. It is known to hold for several classes of dynamical systems, including ergodic automorphisms of compact groups and many geodesic flows. During the rapid development of homogeneous dynamics, S. G. Dani conjectured that the \(K\) property (the absence of any non-trivial zero-entropy measurable factor) implies the Bernoulli property for left translations on quotients of semi-simple Lie groups (see, for example, the discussion of problems in dynamical systems collated by \textit{B. Hasselblatt} [Math. Sci. Res. Inst. Publ. 54, 273--324 (2007; Zbl 1153.37300)]). On the other hand estimates on the decay of matrix coefficients show that the \(K\) property implies an exponential rate of mixing in this homogeneous setting (these ideas go back to the fundamental work of \textit{R. E. Howe} and \textit{C. C. Moore} [J. Funct. Anal. 32, 72--96 (1979; Zbl 0404.22015)]). Dani's conjecture was shown by the second author [``Bernoulli property for homogeneous systems'', Preprint, \url{arXiv:1812.03209}], and the authors explain how this led to the far more general result obtained here. This main result requires the space to be compact (and so in the context of Dani's conjecture gives a proof under the additional hypothesis of a uniform lattice but with the modest starting hypothesis of exponential mixing) and applies in wide generality as follows: If \(f\) is a \(C^{1+\alpha}\) diffeomorphism of a compact manifold preserving a smooth measure that is exponentially mixing for smooth functions, then it is isomorphic to a Bernoulli shift. This important result is obtained using Pesin theory in part, which leaves open certain possible other results like rate of higher-order mixing and similar properties for invariant measures not assumed to be smooth. Some of the large number of settings where the main theorem applies are discussed, and for some of these the Bernoulli property was not really accessible via other methods. In some of the skew-product settings described here the result is also sharp in the sense that failure of exponential mixing precludes the Bernoulli property. There are several points of view from which this work is particularly welcome -- in particular, while there are many partially hyperbolic settings where the \(K\) property can be shown, the Bernoulli property is usually open for them.