Maximum asymptotic variance of sums of finite Markov chains (Q5953884)

The present note considers a reversible Markov chain \((X_n)\) defined on the finite state space \(E\) with irreducible and aperiodic transition matrix \(P=(p_{ij})\) and stationary distribution \(\pi\). Let \(f:E\to R\) satisfy \(\min f(E)=0\), \(\max f(E)=1\), and \(\mu=\int fd\pi\). The purpose of this note is to study the asymptotic variance of the empirical mean sequence \((f_n)={1\over n} \sum^n_{k=1} f(X_k)\), that is known to converge a.s. to \(\mu\). The main theorem of the article constructs an optimal bound for the asymptotic variance of the empirical mean sequence, and proves that this bound depends only on \(\mu\), on the second largest eigenvalue \(\lambda\) of the chain \((X_n)\), and on the endpoints of the support of \(f(E)\).