No label defined (Q4955240)

Let \(T\) be an ergodic measure-preserving automorphism on a probability space \((M, {\mathcal M}, \mu)\). Let \(p\) be a function on \(M\) such that a.s. \(0 < p(x) < 1\). The authors are concerned with the following ``nearest neighbour'' Markov chain on \(M\): starting from some \(x_0\in M\), it jumps from \(x\in M\) to \(Tx\) with probability \(p(x)\) and to \(T^{-1}x\) with probability \(1 - p(x)\). They call this Markov chain a simple random walk along the orbit of \(T\). They first show that if this random walk is nonsymmetric, i.e. if \(\int \log p d\mu\neq \int \log (1-p) d\mu\), then it admits a stationary measure which is absolutely continuous with respect to \(\mu\) (the proof is given only if \(p\) is uniformly bounded from \(0\) and \(1\)). Then they prove their main theorem: if \(P^n_x\) is the probability measure on the set \(\bigcup_{|k|\leq n} T^k x\) with \(P^n_x(k)\) the probability of the random walks starting from \(x= x_0\) such that \(x_n = T^k x\), then under additional (and natural) assumptions on \(\mu\) and \(p\), \(P^n_x\) converges in distribution to the stationary measure. Finally, they consider the example where \(M\) is the \(d\)-dimensional torus endowed with the Lebesgue measure and \(T\) is an irrational rotation. Their main theorem allows them to improve a previous result of the second author [J. Stat. Phys. 94, No. 3/4, 695--708 (1999; Zbl 0948.60030)]. For the entire collection see [Zbl 0937.00068].