Global optimization by center-concentrated sampling
Toru Kambayashi · 1992
The simulated annealing method is one of the probabilistic methods for optimization which attract much attention.The simulated annealing process can be analyzed as a time inh~ mogeneous Markov chain (or process), Several estimations of the convergence rate of the Markov chain (or process) are known.The estimations are, however, very small when the temperature becomes low.The author introduces a novel Markov process for continuous optimization.This Markov process is proved to converge rapidly when the temperature is low, provided some condition holds for the sampling density, a structural factor of the Markov process.The author offers an algorithm for simulation of the process, and some numerical experiments are carried out.The theoretical arguments and the results of the experiments show that annealing is not essential to such a probabilistic approach to continuous optimization.