Molecular Geometry Optimization with a Genetic Algorithm

Abstract

We present a method for reliably determining the lowest energy structure of an atomic cluster in an arbitrary model potential. The method is based on a genetic algorithm, which operates on a population of candidate structures to produce new candidates with lower energies. Our method dramatically outperforms simulated annealing, which we demonstrate by applying the genetic algorithm to a tight-binding model potential for carbon. With this potential, the algorithm efficiently finds fullerene cluster structures up to ${\mathrm{C}}_{60}$ starting from random atomic coordinates.