UC Davis

With advances in computing technology and data science approaches, computational material design is becoming an increasingly reliable and powerful tool for guiding experimental investigations. By constructing databases of known or hypothetical structures of interest, atomistic simulations can be performed in parallel to screen all structures and identify the most promising candidate materials for a given application. This dissertation will highlight some of the diverse applications where computational screening can be implemented to gain new insight about different classes of materials.To begin, a study leveraging density functional theory (DFT) calculations to screen a class of bimetallic porphyrin-based metal-organic frameworks (MOFs) for electrocatalytic reduction of oxygen in fuel cell devices is presented. The highly tunable 3-dimensional pore spaces of these MOFs are shown to provide ideal catalytic environments that surpass the performance of commonly used 2-dimensional surface-based electrocatalysts (e.g., platinum). Next, a systematic approach that combines theory and experiment to characterize active sites in supported atomically dispersed catalysts is discussed. By creating a comprehensive DFT library of possible catalytic sites and comparing simulations with several complementary experimental characterization techniques, an atomic-level understanding of atomically dispersed platinum on magnesium oxide is elucidated. Finally, acknowledging the limitations of quantum mechanics-based simulations due to computational expense, I highlight how machine learning interatomic potentials (MLPs) are revolutionizing atomistic simulations and how easy-to-use open-source software packages are increasing the throughput of the scientific community. Specifically, I discuss the workflows behind generating a large DFT data set of pure silica zeolite configurations, training an accurate and transferable MLP for these systems, and calculating various material properties. Taken together, the work herein will showcase the versatility of molecular modeling approaches, while emphasizing the central roles of experimental collaborations, machine learning, and software development.