scientific article; zbMATH DE number 4092500

From MaRDI portal
Publication:3819798





[For the entire collection see Zbl 0635.00011.] In addition to the definition of a Markov chain being homogeneous in the forward or backward time directions, the author defines a third type of homogeneity, where the distribution of the process inside of an interval I, given the values on the endpoints, depends only on the length of I. In continuation of his earlier paper, Ann. Probab. 14, 459-479 (1986; Zbl 0594.60070), the author considers a two-sided Markov chain with discrete time parameter. For chains \(\{\) X(n)\(\}\) having all two-dimensional marginal probability distributions strictly positive, it is shown that there exists a positive matrix p, a family of measures \(\{\pi_ n\}\) and a family of strictly positive functions \(\{h_ n\}\) such that \(P\{X(n)=i\}=\pi_ n(i)h_ n(i)\), \[ P\{X(n+k)=i_ k,\quad 0\leq k\leq m\}=\pi_ n(i_ 0)[\prod^{m-1}_{k=0}p(i_ k,i_{k+1})]\cdot h_{n+m}(i_ m), \] m\(=1,2,...\); \(n=0,\pm 1,\pm 2,...\); \(i,i_ 0,i_ 1,...,i_ m\in S\), where S is the state space. The author mentions the connection of the above with Gibbs states [see \textit{F. Spitzer}, J. Functional Analysis 20, 240-255 (1975; Zbl 0333.60063)]. He also gives conditions under which the converse holds, which thus yield a method of constructing time-homogeneous two-sided Markov chains, when the quantities \(\{\pi_ n\}\) and \(\{h_ n\}\) are given. He also discusses special conditions under which a two-sided Markov chain becomes periodic. Some examples are discussed.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3819798)