The Application of Genetic Algorithms For Density Functional Optimization and Development

Abstract

This work details the development of the density functional theory (DFT) implementation for nonadditive three-body dispersion using the exchange dipole moment (XDM) and the quantum chemistry functional KP16/B13 to calculate self-consistent field energy of molecular systems through the application of genetic algorithms. In the case of dispersion, current ab initio methods are accurate but computationally expensive. The development of three-body dispersion detailed here involves a density functional model approach resulting in comparable performance and advantages in computational efficiency. The KP16/B13 functional is developed to advance the implementation of a single reference functional that addresses nondynamic/strong correlation and for use as a general purpose functional. This work details the optimization of the two models with selected sets of atoms and molecules and benchmarking KP16/B13 with the Minnesota sets involving a variety of chemical properties. Both parts of this work compare results against contemporary methods demonstrating improved performance for some properties and comparable results in others. The calibration and optimization of the methods detailed above are my main contribution. Software development to achieve this goal resulted in a general purpose genetic algorithm code stack. The software developed also facilitated the organization of results and computation of the methods on computer clusters at MTSU and supercomputers at Oak Ridge National Laboratory. Through multiple iterations, refactoring, and design input from colleagues, advisors, and users, the final software stack is robust and will continue to be leveraged by members of Dr. Kong's group in future research.