Uniform CLT for Markov chains with a countable state space (Q5906861)

Let \(\{X(n);\;n=0,1,2, \dots\}\) be a positive recurrent Markov chain in \(S= 3D \{1,2,3, \dots\}\) with unique invariant probability measure \(\pi\), let \(N(k)\) be the \(k\)th hitting time of \(1,m(i,j): ={\mathbf E} \min \{n\geq 1:X(n) =j\mid X(0) =i\}\), and let \(F\) be a nonnegative function on \(S\), \(\Gamma: =\{f:| f| \leq F\}\). Then the uniform CLT holds over \(\Gamma\) if and only if \[ {\mathbf E} \bigl(N(2) -N(1)\bigr)^2 <\infty,\quad {\mathbf E} \sum_{N(1) <j\leq N(2)} F\bigl( X(j) \bigr)^2 <\infty,\quad {\mathbf E} \sum_{1\leq k< \infty} F(k) \pi(k) m(1,k)^{1/2} <\infty. \] This result generalizes \textit{S. Levental}'s theorem [Stochastic Processes Appl. 34, No. 2, 245-253 (1990; Zbl 0699.60013)]. It is established via first proving a uniform CLT for regenerative processes. The conditions are weaker than those involved when applying known mixing empirical process results.