No label defined (Q4356503)

Let \((X_n, -\infty<n<\infty)\) be a stationary process with values in \(\mathbb R^d\), \(d\geq1\). Suppose that the process satisfies the recurrent equation \(X_{n+1}=f(X_{-\infty}^n)+W_{n+1}\), \(-\infty<n<\infty\), where \(f:{\mathbb R^d}^{Z_-}\to\mathbb R^l\) is a given function, \(Z_-=\{0,-1,-2,\ldots\}\), and \((W_n)\) is a sequence of independent and identically distributed random variables in \(\mathbb R^l\). The norm \(|{\mathbf x}|:=\sum_{i\geq 0}\lambda^i|x_{-i}|\) with some \(\lambda>0\) is defined on \({\mathbb R^d}^{Z_-}\). The following theorem is the main result of the paper. Let there exist \(\lambda>0\), \(\varepsilon_m\to 0\), \(m\varepsilon_m\to\infty\), and \(\gamma_k\to 0\), \(k\to\infty\), such that for any \({\mathbf x}=(x_0,x_{-1},\ldots)\), \({\mathbf y}=(y_0,y_{-1},\ldots)\), \(|f({\mathbf x})|+\lambda|{\mathbf x}|\leq(1-\varepsilon_m)|{\mathbf x}|\) for \(|{\mathbf x}|\leq m\), and \(|f({\mathbf x})-f({\mathbf y})|\leq\gamma_i\) for \(x_0=y_0,\ldots,x_{-i+1}=y_{-i+1}\), \({\mathbf E}\exp(a|W_1|)<\infty\) for any \(a>0\), and let the r.v. \(W_1\) have a density \(p\) which is everywhere positive, continuous, and satisfy the following Lipshitz type condition: \(p(z+v)/p(v)\leq 1+N_m|z|\), for any \(v\), \(m>0\), \(|z|\leq m\), and for some \(N_m>0\). Then the large deviation principle holds for the family \((n^{-1}\sum_{k=1}^ng(X_k),\;n\geq1)\), where \(g:\mathbb R^d\to\mathbb R^l\) is any bounded function. This theorem is deduced from another theorem on large deviations, proved by \textit{O. V. Gulinsky} and \textit{A. Yu. Veretennikov} [Large deviations for discrete-time processes with averaging. Utrecht: VSP (1993; Zbl 0838.60028)]. For the entire collection see [Zbl 0870.00038].