Computational Doping for Fuel Cell Material Design Based on Genetic Algorithms and Genetic Programming
Date of Award
Open Access Dissertation
Computer Science and Engineering
Developing new materials have historically been time-consuming. Computational material discovery can search large design space to identify promising candidates for experimental verification. Recently, Density Functional Theory (DFT) based first principle calculation has been able to calculate many electrical and physical properties of materials, making them suitable for computational doping based material discovery. In material doping, given a base material, one can change its properties by substituting some elements with new ones or adding additional elements. In computational doping, we have a grid of atoms in a supercell, some of which can be substituted with dopant atoms. There are many possible doping positions for the doped elements in the supercell, among which the most stable supercell with the lowest free electronic energy is the one that most likely appears in experiments. So finding the most stable doped supercell configuration is the first step for computational doping, which is usually done exhaustively nowadays. For each such substitution, the Vienna Ab-Initio Simulation Package is usually used to calculate its energy and higher level physicochemical properties. Free energy calculations take about 15-30 hours for a supercell of 75 atoms for substituting two positions out of 15 with a single dopant element, and it may take days to weeks for multiple dopant elements. This is a typical optimization problem with expensive evaluation functions. Here we first developed a genetic algorithm for finding the most stable structure of the doped material with the lowest free electronic energy for a single dopant element. It can reduce the running time for computational doping by up to 75%. We used SrTiO3 perovskite as the base material and Nb as the substitution element. We also developed another genetic algorithm for multiple dopant elements. Since the search space becomes larger, the genetic algorithm works better and saves up to 85% of calculations for finding the most stable structures. Finally, we developed a genetic programming (GP) algorithm for computational doping which can simultaneously determine multiple dopant elements with different doping ratios. The simultaneous search of dopant elements and their ratios can speed up the search process for large doping spaces.
Atilgan, E.(2016). Computational Doping for Fuel Cell Material Design Based on Genetic Algorithms and Genetic Programming. (Doctoral dissertation). Retrieved from https://scholarcommons.sc.edu/etd/3576