Tight-Binding Approximation-Enhanced Global Optimization

Abstract

Solving and predicting atomic structures from first-principles methodologies is limited by the computational cost of exploring the search space, even when relatively inexpensive density functionals are used. Here, we present an efficient approach where the search is performed using density functional tight-binding, with an automatic adaptive parametrization scheme for the repulsive pair potentials. We successfully apply the method to the genetic algorithm optimization of bulk carbon, titanium dioxide, palladium oxide, and calcium hydroxide, and we assess the stability of the unknown crystal structure of palladium hydroxide.