Simulated annealing

Abstract

Simulated annealing belongs together with genetic algorithms and evolution strategies to major new optimization methods. It is used for optimization of highly multimodal functions, both for discrete as well as for continuous functions. Though origin of the method in physics has been nearly forgotten, it is useful to take into account also other terms connected with simulation of processes in statistical physics. These terms can be helpful for deeper understanding of the method. These terms can be also helpful in evaluation of newly proposed modifications of the method, like force and parallel simulated annealing in comparison with simple simulated annealing. The paper introduces statistical physics terms into evaluation of minimization process of simulated annealing. New modifications of simulated annealing like forced simulated annealing and parallel simulated annealing are described in Pascal-like algorithms, their performance is illustrated by examples accompanied by figures describing processes in simulated annealing in graph plots of time vs. statistical physics terms.