Genetic Algorithm Parameter Analysis

Abstract

When a Genetic Algorithm (GA) is applied to solve any optimization problem, the questions arise; how to implement it; what structure to use; how much weight to give to the three main processes (mutation, recombination and reproduction). 1) The present work is a part of a systematic study of the last question, in particular, what population size and, how much mutation? The procedure used is as follows: a standard GA is applied to two physical problems, a) the determination of the most stable configuration of an atomic aggregate, by minimizing the interaction energy 2) and b) the calculation of the ground state of a two dimensional finite spin glass, also minimizing the energy. 3) As the two problems are quite different (the spin glass is a combinatorial problem and the atomic problem is a real variable problem), a coincidence in some result is significant, so here only common conclusions are reported. The first result shows the clear existence of an optimal value for the mutation rate, see Figs. 1 and 2, the point of minimum energy in this graph indicates the best value for the mutation rate for the corresponding population size. Another observation is that, when increasing the population size the mutation continues being important