Experiences with Global Optimization Techniques in Massively Parallel Processing Environments

摘要

The modeling of many physically relevant processes involves the location of global extrema, which are often surrounded by numerous local extrema. The effort involved in locating these extrema grows rapidly with increasing dimensionality of the problems being investigated. A method of analysis for algorithms from the perspective of a parallel processing environment is developed and applied to the family of Monte Carlo global optimization schemes. A new highly parallel Monte Carlo algorithm is presented. Chaotic iteration schemes resulting in nondeterministic algorithms are discussed, and an analysis of the optimal conditions for their application to this class of problems is presented. 11 refs., 4 figs. (ERA citation 13:058315)