Large deviations for random processes with independent increments on infinite intervals (Q1290836)

The large-deviation principle for generalized Poisson processes defined on the positive half-line \([0,+\infty)\) is proved. Consider a space \({\mathcal X}\) of functions \(f:[0,\infty)\to {\mathbb{R}}^r\) with the following three properties: 1) The functions \(f\in {\mathcal X}\) are functions with locally finite variation, i.e., \(\text{Var}_{[0,t]} f<\infty\) for all \(0<t<\infty\). 2) The functions \(f\in {\mathcal X}\) are right-continuous at each point \(t\geq 0\). 3) The limits \(v(f)=\lim_{t\to \infty}\frac{f(t)}{1+t}\) exist and are finite. Consider a generalized Poisson measure \(P_{\pi}\) on Borel subsets of the space \({\mathcal X}\) (with a uniformly-weak topology) with the jump measure \(\pi\) such that \(\int_{{\mathbb{R}}^r}|y|\pi(dy)<\infty\), \(\int_{{\mathbb{R}}^r}(e^{a|y|}-1)\pi(dy)<\infty\) for some \(a>0\). Define a sequence of probability measures \( P_n=P_{\pi_n}\), where \(\pi_n(A)=n\pi(nA)\). The authors obtain a large-deviation principle for \(P_n\) with rate function \(I\), known in explicit form. With the help of the large-deviation principle proved in the paper, the authors give a new proof for the known results on the asymptotics of the logarithm of the probabilities of large delay in a queueing system with a single server and Poisson input flow.