SPT: A stochastic tunneling algorithm for global optimization (Q5954666): Difference between revisions

The paper develops an algorithm for unconstrained continuous global optimization problems that combines a stochastic Pijavskij tunneling approach with the TRUST method of \textit{J. Barhen} and \textit{V. Protopopescu} [in E. A. Floudas (ed.) et al., Kluwer, Nonconvex Optim. Appl. 7, 163-180 (1996; Zbl 0871.90080) and with \textit{D. Reiter}, Science 276, 1094-1097 (1997)]. A series of local descents is coupled with stochastic searches: New local minima descent regions are located utilizing a rejection-based stochastic procedure, where, based on an estimate of the Lipschitz constant and previous function evaluations, certain exclusion regions are determined and discarded from further investigation. The algorithm works particularly well for low dimensional problems, while for higher dimensional problems the impact of the exclusion regions on the efficiency of the algorithm decreases. Computational results are presented which show that the algorithm is competitve with other global optimization techniques also on large problem instance, including a large-scale seismology problem.