State space collapse for queueing networks (Q1126814)

Let \(Z(t)\) be the \(K\)-vector of numbers of customers at time \(t\) in a queueing network with \(J\) stations and \(K\) customer classes and \(W(t)\) the \(J\)-vector of workloads at the \(J\) stations. Typically, \(K\) is much smaller than \(J\), and state space collapse means that \(Z(\cdot)\) can be approximated by \(\Delta W(\cdot)\) where \(\Delta\) is a linear map from \(R^J\) to \(R^K\). This paper is a survey of how state space collapse can be obtained from fluid limits of the network, with particular emphasis on multiplicative state space collapse (meaning that \(Z(\cdot)-\Delta W(\cdot)\) can be bounded by \(\|W(\cdot)\|\)) and heavy traffic limit theory (convergence to multidimensional reflected Brownian motion in an orthant).