Parallel and Serial Successive Overrelaxation for Multicommodity Spatial Price Equilibrium Problems

Abstract

Computational experience is reported for various serial implementations of successive overrelaxation (SOR) applied to a linear multicommodity spatial price equilibrium problem posed as linear complementarity problem. Computational schemes that neglect the vast majority of the variables during most iterations are shown to be relatively efficient. A parallel implementation based on the same neglect is shown to exhibit encouraging average speedup over the single processor case. The SOR approach is shown empirically to converge for nonsymmetric problems. Dense network problems with up to 60 regions (each a potential supply or demand region) and 10 commodities, representing on the order of 36,000 variables, as well as sparse network problems with up to 140 regions, are typically solved in a few seconds of CPU time.