Automated Optimization of Pseudopotentials for Faster and More Accurate Plane Wave Density Functional Theory Calculations in Materials

Abstract

In principle, the properties of materials are determined by the behavior of electrons. Therefore quantum mechanical electronic structure calculations play a critical role in materials modeling. These calculations owe much of their success in materials research to speed improvements enabled by the pseudopotential approximation to density functional theory (DFT). However, new pseudopotentials still need to be developed to meet the accuracy and speed demands of the materials design framework. The goal of this research is thus to improve on previous generations of pseudopotentials using automated searches for optimal pseudopotential parameters. In this dissertation, I show that optimizations using a multi-objective genetic algorithm can improve the accuracy and speed of pseudopotentials compared to those previously published in the literature. Additionally, I propose a new scattering property metric for pseudopotentials based on the arctangent of the logderivatives, which can cheaply identify ghost states and logderivative disagreements. This metric is key for enabling automated pseudopotential optimization, and facilitates reporting of the accuracy of published pseudopotentials. Finally, I demonstrate the automated design of a gallium pseudopotential using the arctangent metric, which results in a 40% speedup of defect calculations in gallium nitride.