Long range dependence of point processes, with queueing examples (Q1275964): Difference between revisions

The authors introduce two concepts of long range dependence for stationary point processes on the real line. Let \(\{Y_i:i=1,2,\ldots\}\) be the stationary sequence of distances of neighbour points (starting from a typical point) and \(N(0,x]\) the number of points of the process in the interval \((0,x]\). Then, by definition, the point process exhibits: 1. long range interval dependence (LRiD) when \(\limsup _{n\to \infty }n^{-1}\text{var}(\sum _{i=1}^nY_i)=\infty \), 2. long range count dependence (LRcD) when \(\limsup _{x\to \infty }x^{-1}\text{var} N(0,x]=\infty \). It is shown that a renewal process, which of course cannot be LRiD, can be LRcD, under an assumption on the tail behavior of the lifetime distribution. The authors further consider various queueing models and find conditions for the LRcD of the output process: e.g., the output of GI/M/1 as well as of M/G/1 can be LRcD under special assumptions, and the output of G/GI/\(\infty\) is LRcD if and only if the input process is LRcD.