Comparing Darwinian, Baldwinian, and Lamarckian Search in a Genetic Algorithm for the 4-Cycle Problem

Genetic algorithms abstract the fundamental processes of Darwinian evolution: natural selection and genetic variation through recombination and mutation. GAs can also implement non-Darwinian mechanisms such as Lamarckian search, in which local search is used to improve chromosomes, and Bald-winian search, in which local search improves chromosomes' tnesses but the chromosomes themselves are not modiied. This paper describes a comparison of Lamarckian, Bald-winian, and simple Darwinian strategies in a genetic algorithm for the 4-cycle problem, which seeks the shortest collection of disjoint 4-cycles on a set of n = 4k points in the plane. In tests on six instances of this problem , Baldwinian strategies perform poorly, while the most aggressive of three Lamarck-ian strategies yields the best results.