Gene Pool Recombination in Genetic Algorithms

Abstract

A new recombination operator, called Gene Pool Recombination (GPR) is introduced. In GPR, the genes are randomly picked from the gene pool defined by the selected parents. The mathematical analysis of GPR is easier than for two-parent recombination (TPR) normally used in genetic algorithms. There are n difference equations for the marginal gene frequencies that describe the evolution of a population for a fitness function of size n. For simple fitness functions TPR and GPR perform similarly, with a slight advantage for GPR. Furthermore the mathematical analysis shows that a genetic algorithm with only selection and recombination is not a global optimization method, in contrast to popular belief.