Improved real-space genetic algorithm for crystal structure and polymorph prediction

Abstract

Existing genetic algorithms for crystal structure and polymorph prediction can suffer from stagnation during evolution, with a consequent loss of efficiency and accuracy. An improved genetic algorithm is introduced herein which penalizes similar structures and so enhances structural diversity in the population at each generation. This is shown to improve the quality of results found for the theoretical prediction of simple model crystal structures. In particular, this method is demonstrated to find three new zero-temperature phases of the Dzugutov potential that have not been previously reported.