Functional limit theorems for occupation times of Lamperti's stochastic processes in discrete time (Q2475059): Difference between revisions

Let \(X= \{X_n\}_{n\in\mathbb{Z}_+}\) be a discrete time process with state space \(S\), which is devided into sets \(S_+\), \(S_-\), and \(\{0\}\). \(X\) is called a Lamperti process if the following conditions are satisfied: (i) \(P\,(X_0= 1)= 1\), (ii) if \(X_{n-1}\in S_+\) and \(X_{n+1}\in S_-\) or if \(X_{n-1}\in S_-\) and \(X_{n+1}\in S_+\), then \(X_n= 0\), (iii) \(P\,(\tau<\infty)= 1\) where \(\tau\) is the first hitting time at the state\(0\) of \(X\), (iv) it starts afresh when it recurs the state \(0\). Let \(A_+(n)\) denote the occupation time up to time \(n\) of the set \(S_+\) of \(X\). Occupation of the state \(0\) is counted or not according to whether the last other state occupied was in \(S_+\). We extend \(A_+(n)\) of the Lamperti process \(X\) to a process \([0,\infty\ni t\mapsto A_+(t)\) by the linear interpolation. Put \[ g(x)= \sum^\infty_{n=1} x^n P\,(X_n= 0,\, X_i\neq 0,\, 1< i< n). \] We assume that \[ \begin{gathered} \lim_{n\to\infty}\, E(A_+(n)/n)= p\qquad (0\leq p\leq 1)\quad\text{and}\\ \lim_{x\to 1-} (1- x)g'(x)/(1- g(x))= \alpha\qquad(0\leq \alpha\leq 1).\end{gathered} \] Then, by a \textit{J. Lamperti's} result [Trans. Am. Math. Soc. 88, 380--387 (1958; Zbl 0228.60046)] \[ (1/n)A_+(n)@>d>> A^{(\alpha, p)}_+(1)\overset {d}= (1/t) A^{(\alpha, p)}_+(t) \] for every \(t> 0\) where \(A^{(\alpha, p)}_+(t)\) denotes the occupation time on the positive side up time \(t\) with parameter \(p\) and \(\alpha\). In this paper, the authors prove that \[ (1/\lambda)A_+(\lambda t)@>{\mathcal L}>> A^{(\alpha, p)}_+(t)\quad\text{in }C([0,\infty): [0,\infty))\qquad (\lambda\to\infty), \] when \(0<\alpha< 1\) and obtain that the limit distributions of the finite-dimensional distributions of \((1/\varphi^{-1})(\lambda)(A_+(\lambda t)- p\lambda t)\) with some regularly varying function \(\varphi(\cdot)\) at \(\infty\) wen \(\alpha=1\) and \(0< p< 1\).