No label defined (Q4258799)

A Markov chain \(X_n\) which may be interpreted as a random walk on a tree is considered. Let \(X\) be the set of all dyadic rational numbers of the open interval \((0,1).\) The homogeneous Markov chain \(X_n\) on \(X\) is defined in the following way. Let \(x \in X\) and \(x=0.\alpha_1 \alpha_2 \ldots \alpha_k\), \(k\geq 0\), be the dyadic expansion of \(x.\) Let \[ P\bigl( X_n=f_j(x)/ X_{n-1}=x\bigr) ={\mu_k/2},\qquad j=1,2, \] \[ P\bigl( X_n=f_3(x)/ X_{n-1}=x\bigr) =\lambda_k, \] where \(f_1(x)=0.\alpha_1 \alpha_2 \ldots \alpha_k01\), \(f_2(x)=0.\alpha_1 \alpha_2 \ldots \alpha_k11\), \(f_3(x)=0.\alpha_1 \alpha_2 \ldots \alpha_k1\), \(\mu_k +\lambda_k =1\), \(\mu_0=1\), \(\lambda_0 =1.\) Let \(\pi_n = \prod_{j=0}^{n-1}\bigl(\mu_j/ \lambda_{j+1}\bigr).\) The author proves under the assumption that \(\sum_{n=1}^\infty \pi_n < \infty\) the following assertions. 1) The Markov chain \(X_n\) has a stationary distribution of the form \[ p(k,j) ={\pi_k \over 2^k \sum_{m=0}^\infty \pi_m}. \] 2) This distribution satisfies the system of equations \[ p(k,j) =\lambda_{k+1}(p(k+1,2j-1) + p(k+1,2j) + \mu_{k-1}p(k-1,w(j))/2, \] where \[ w(j)=\begin{cases} j/2, & j \text{ even} \\ (j+1)/2, & j \text{ odd}. \end{cases} \] 3) For every other solution of this system there exist real numbers \(0 \leq \alpha < \beta \leq 1\) and \(r \neq 0\) such that \[ \lim_{k \to \infty}\sum_j' p(k,j)=r, \qquad \lim_{k \to \infty}\sum_j''p(k,j)=-r, \] where \(\sum_j'\) and \(\sum_j''\) denote the summation over \(2^k\alpha \leq j \leq 2^k\beta\) and \(\{ 1\leq j < 2^k \alpha \} \cup \{ 2^k\beta < j \leq 2^k \}.\) For the entire collection see [Zbl 0896.00028].