Large deviation properties of a correlated arrival process (Q4951464)

The background sequence in the paper is given by \(U_n= U_0\) if \(n=0\), and \(U_n= \langle U_{n-1}+ V_n\rangle\) if \(n> 0\), where \(U_0\) is uniformly distributed on \([0,1)\), \(\{V_n\), \(n> 0\}\) is a sequence of independent identically distributed random variables with prescribed density function constituting the sequence of innovations, and \(\langle x\rangle\) designates the fractional part of \(x\). The target sequence \(X_n\) is given by \(X_n= D(U_n)\). Let \(g(\nu-u)\) be the conditional density function of inter-arrival time \(v\) of the background uniform variable given the previous inter-arrival time \(u\). An exact matrix representation of the operator \(T_\theta\): \[ T_\theta h(u)= \int_0^1 e^{-\theta D(v)} g(v-u) h(v) dv \] is obtained and an accurate numerical procedure is derived for the computation of the dominant eigenvalue \(\lambda_0(\theta)\). The peakedness function is studied in order to obtain a useful approximation to \(\lambda_0(\theta)\). Limit formulae useful in queueing theory and in large deviation results are also obtained.