International Journal of Computational
Intelligence Research (IJCIR)
Volume 1, Number 1 (2005)
Analysis of a Non-Generational Mutationless Evolutionary Algorithm
for Separable Fitness Functions
University of Dortmund
Department of Computer Science
It is shown that the stochastic dynamics of non-generational evolutionary algorithms with binary tournament selection and gene pool recombination but without mutation is closely approximated by a stochastic process consisting of several de-coupled random walks, provided the fitness function is separable in a certain sense. This approach leads to a lower bound on the population size such that the evolutionary algorithm converges to a uniform population with globally optimal individuals for a given confidence level.