Rare events for stationary processes. (Q1877529)

Let \(X_t\) be a discrete Markov chain on a countable space, \(\alpha \) a fixed point (the origin) and \(F\) a rarely visited set, \(\alpha \notin F\). Let further \(R\geq 1\) and \(\tau (F)\) denote the first return time to \(\alpha \) after visiting \(F\) and the time to hit \(F\), respectively. Then the difference between both the times is asymptotically negligible. Let further \(\pi \) and \(K\) be the transition kernel and stationary probability measure, resp., and \(f(x), x \in F^c,\) the probability that \(X\) starting from \(x\) hits \(\alpha \) before \(F\). Denoting by \(E_x\) the law of the chain originating at point \(x\), then \[ \Lambda =1/E_{\alpha }R = \sum _{y \in F}\pi (y) \sum _{x \in F^c} K(y,x)f(x) \] has the meaning of the intensity of returns from \(F\) to \(\alpha \) and the asymptotic of rare events is described by \(\lim _{\pi (F) \to 0} \Lambda E_{\alpha }\tau (F) =1\) [cf. \textit{J. Keilson}, ``Markov chain models -- rarity and exponentiality'' (1979; Zbl 0411.60068) and \textit{D. Aldous}, ``Probability approximations via the Poisson clumping heuristic'' (1989; Zbl 0679.60013)]. These results are suitably extended to the stationary process \(\{ X_t\}\) defined on \((\Omega , \mathcal F, P)\) taking its values in some measurable space \((S, \mathcal S)\). In order to do this, the notions of asymptotic exponentiality in variance and asymptotic independence are introduced and several theorems and lemmas concerning rare events are proved.