Evolutionary programming techniques for predicting inorganic crystal structures

Abstract

New techniques based on the implementation of genetic algorithms together with energy minimisation procedures are able to predict the crystal structures of complex inorganic solids. A crucial feature of our approach is the use in the initial stages of the simulation of a sophisticated cost function based on Pauling's rules, recently extended and quantified as the bond valence model. Using such functions, we are able to generate candidate structures whose energies may subsequently be minimised using standard lattice energy methods employing Born model potentials. We demonstrate the efficacy of the method in yielding accurate solutions of complex crystal structures by its application to the previously unsolved structure of the ternary oxide Li₃RuO₄.