Central limit theorem started at a point for stationary processes and additive functionals of reversible Markov chains (Q2428534)

Let \((\xi_n)_{n\in\mathbb Z}\) be a stationary ergodic Markov chain (MC) with values in a measurable space \((S,\mathcal A)\), with initial distribution \(\pi\), regular transition distribution \(Q(x,A)=\text{P}(\xi_1\in A|\xi_0=x)\) and associated Markov operator \(Qf(x)=\int_Sf(s)Q(x,ds)\). The authors give sufficient conditions for the quenched central limit theorem (CLT) \[ n^{-1/2}\sum_{i=1}^nf(\xi_i)\Rightarrow|\sigma_f|\,N\text{ under }\text{P}^x \] for \(\pi\)-almost every \(x\in S\), where \(N\) is standard normally distributed, \(f\in L^2(\pi)\) with \(\int_Sf\,d\pi=0\), and \(\sigma_f\) is a constant depending on the spectral measure of \(f\) with respect to \(Q\). Let \(V_n(f)=\sum_{i=0}^{n-1}Q^if\) and let \(Q^{\ast}\) denote the adjoint of \(Q\) in \(L^2(\pi)\). Then the quenched CLT holds if the MC is additionally reversible (\(Q=Q^\ast\)) and either \[ \sum_n\frac{(\log\log n)^2 \|V_n(f)\|_2^2}{n^2}\quad\text{ or }\quad\sum_n\frac{\|V_n(f)\|_2}{n^{3/2}} \] is finite. This extends previous CLT results under the stationary probability measure in case the MC is normal (\(QQ^\ast=Q^\ast Q\)). Reversibility is decisive, since the authors give an example of a stationary ergodic normal MC such that the first of the above sums is finite but the quenched CLT fails. The method of proof follows a classical martingale approximation: The authors show the existence of an adapted martingale \((M_n)_{n\geq1}\) with stationary and ergodic increments such that \[ n^{-1}\text{E}^x\left[\left(\sum_{i=1}^nf(\xi_i)-M_n\right)^2\right]\to0 \] for \(\pi\)-almost every \(x\in S\). To prove the main theorem, further results of independent interest on the quenched CLT for general stationary ergodic MC and on almost sure convergence for stationary reversible MC are given.