Physical measures for multivalued inverse iterates near hyperbolic repellors (Q977197)

The author studies new invariant probability measures, describing the distribution of multivalued inverse iterates (i.e., of different local inverse iterates) for a non-invertible smooth function \(f\) which is hyperbolic, but not necessarily expanding on a repellor \(\Lambda \). The methods for the higher dimensional non-expanding and non-invertible case are different than the ones for diffeomorphisms, due to the lack of a nice unstable foliation (local unstable manifolds depend on prehistories and may intersect each other, both in \(\Lambda \) and outside \(\Lambda \)), and the fact that Markov partitions may not exist on \(\Lambda \). The authour obtains that for Lebesgue almost all points \(z\) in a neighbourhood \(V\) of \(\Lambda \), the normalized averages of Dirac measures on the consecutive preimage sets of \(z\) converge weakly to an equilibrium measure \(\mu^-\) on \(\Lambda \); this implies that \(\mu^-\) is a physical measure for the local inverse iterates of \(f\). It turns out that \(\mu^-\) is an inverse SRB measure in the sense that it is the only invariant measure satisfying a Pesin type formula for the negative Lyapunov exponents. Also the authour shows that \(\mu^-\) has absolutely continuous conditional measures on local stable manifolds, by using the above convergence of measures. The authour proves then that \(f:(\Lambda ,{\mathcal B}(\Lambda ),\mu^-)\rightarrow (\Lambda ,{\mathcal B}(\Lambda ),\mu\mu^-)\) cannot be one-sided Bernoulli, although it is an exact endomorphism of Lebesgue spaces. Several classes of examples of hyperbolic non-invertible and non-expanding repellors, with their inverse SRB measures, are given in the end.