Time inhomogeneous Markov chains with wave-like behavior (Q1958502)

In some their previous papers the authors considered the problem of obtaining quantitative results describing the ergodic behavior of time inhomogeneous finite Markov chains. In general. a, time inhomogeneous Markov chain, say on a finite set \(V\), is described by a sequence of Markov kernels \((K_i)^\infty_1\). They remind the property called the total variation merging when it is said that the chain driven by the sequence \((K_i)^\infty_1\) is merging. Now, in this paper, they discuss a stronger notion which they call relative-sup merging. By definition, the sequence \((K_i)^\infty_1\) is merging in relative-sup if \[ \max_{x,y,z\in V}\Biggl\{\Biggl| {K_{0,n}(x,z)\over K_{0,n}(y, z)}- 1\Biggr|\Biggr\}\to 0\quad\text{as }n\to\infty. \] And, in general, the relative-sup distance between two measures \(\mu\) and \(\nu\) (on a finite or countable state space) is defined by (note the asymmetry) \[ \max_{x\in V} \Biggl\{\Biggl|{\mu(x)\over \nu()}- 1\Biggr|\Biggr\}. \] In particular, for a time inhomogeneous chain driven by a sequence \((K_i)^\infty_1\) of Markov kernels, the authors consider quantities such as where \(\mu_n= \mu_0 K_{0,n}\) for some starting measure \(\mu_0\). Then, for any \(\varepsilon> 0\), they define the a relative-sup merging time \(T_\infty(\varepsilon)\) by \[ T_\infty(\varepsilon)= \min\Biggl\{n: \max_{x,y,z\in V} \Biggl\{\Biggl|{K_{0,n}(x, z)\over K_{0,n}(y,z)}- 1\Biggr|\Biggr\}< \varepsilon\Biggr\}. \] The authors observe that only a small set of examples have been treated in the literature mostly because proving anything about concrete time inhomogeneous Markov chains is difficult. Thus, in this paper a special class of examples whose structure is, in itself, quite interesting and for which some results can be obtained is described. The idea is as follows. It is given a Markov kernel \(K\) and a bijection \(g: V\to V\) on a finite or countable set \(V\). Then it is considered the time inhomogeneous Markov chain driven by the sequence of the kernels \[ K_i(x,y)= K(g^{i-1} x,g^{i-1} y),\qquad x,y\in V,\quad i= 1,2,\dots. \] Now the problem is to study this time inhomogeneous chain and its merging properties. Thus, it will be see that this covers some interesting examples and leads to interesting results as well as difficult open problems. After an Introduction, this study is organized in 5 sections. It is a good and useful paper.