Force-based optimization of pseudopotentials for non-equilibrium configurations

We have used a multi-objective genetic algorithm to optimize pseudopotentials for force accuracy and computational efficiency. Force accuracy is determined by comparing interatomic forces generated using the pseudopotentials and forces generated using the full-potential linearized augmented-plane wave method. This force-based optimization approach is motivated by applications where interatomic forces are important, including material interfaces, crystal defects, and molecular dynamics. Our method generates Pareto sets of optimized pseudopotentials containing various compromises between accuracy and efficiency. We have tested our method for LiF, Si0.5Ge0.5, and Mo and compared the performance of our pseudopotentials with pseudopotentials available from the ABINIT library. We show that the optimization can generate pseudopotentials with comparable accuracy (in terms of force matching and equation of state) to pseudopotentials in the literature while sometimes significantly improving computational efficiency. For example, we generated pseudopotentials for one system tested that reduced computational work by 71% without loss of accuracy. These results suggest our method can be used to generate pseudopotentials on demand that are tuned for a user's specific application, affording gains in computational efficiency.