Counteracting stagnation in genetic algorithm calculations by implementation of a micro genetic algorithm strategy

Abstract

A new strategy for implementing the concept of a “micro genetic algorithm” within a standard genetic algorithm (GA) procedure is proposed. The strategy operates by applying criteria to test for the occurrence of stagnation within the population of a standard GA calculation, and triggering the micro-GA procedure whenever stagnation is detected. The micro-GA is implemented in terms of the parallel evolution of a number of small sub-populations (comprising predominantly new randomly generated structures together with a few of the best structures from the stagnated population), and the sub-population of highest quality following the micro-GA procedure is used in the construction of the next population of the standard GA calculation. The micro-GA procedure is applied in the context of a GA for carrying out direct-space structure solution from powder X-ray diffraction data, and the results demonstrate that this strategy is an effective means of promoting structural diversity within a stagnated population, leading to significantly improved evolutionary progress. This strategy may prove to be more generally applicable as an approach for alleviating problems due to stagnation in GA calculations in other fields of application.