Multidimensional renewal theory in the non-centered case: application to strongly ergodic Markov chains (Q1935439)

In this paper, the authors consider a sequence \((X_n,S_n)_{n \geq 0}\) of random variables taking values in \(E \times \mathbb{R}^d\) where \(E\) denotes an arbitrary measurable space, and \(d \geq 2\). \((S_n)_{n \geq 1}\) is assumed to be non-centered in the sense that \(\vec{m} = \lim_{n \to \infty} \mathbb{E}[S_n]/n\) exists in \(\mathbb{R}^d\) and is nonzero. Given some fixed measurable function \(f: E \to [0,\infty)\) such that \(\mathbb{E} [f(X_n)] < \infty\) for every \(n \geq 1\), the authors provide a sufficient condition under which the asymptotics of \(\sum_{n \geq 1} \mathbb{E} \big[f(X_n) g(S_n-a)\big]\) as \(a \in \mathbb{R}^d\) tends to infinity ``around the direction \(\vec{m}\)'' can be determined for all continuous and compactly supported functions \(g: \mathbb{R}^d \to \mathbb{R}\). To be more precise, the scaling needed to make the above series converge as \(a \approx \tau \vec{m}\) and \(\tau \to \infty\) is of order \(\tau^{\frac{d-1}{2}}\). The limit obtained is a constant multiple times the Lebesgue integral of \(g\). The result is obtained via Fourier techniques and applied to a situations when \((X_n,S_n)_{n \geq 0}\) is a special Markov random walk, namely, when \((X_n)_{n \geq 1}\) is a (strongly ergodic) Markov chain and \(S_n = \xi(X_1)+\ldots+\xi(X_n)\) for a measurable function \(\xi: E \to \mathbb{R}^d\).