Developing Computational Screening Methods for Accelerated Catalyst Discovery

In the last 50 years, increasing human populations have resulted in three times more fossil fuels consumption to meet the demand for energy and fuels. Unfortunately, this has disrupted ecosystems and climate with excessive greenhouse gas emissions. We inevitably need to transition from fossil fuels to renewable energy technologies. Catalyst plays a key role in renewable technology by converting harvested energy to fuels and chemicals. However, today’s catalysts are inadequate for the widespread adoption of renewable technologies. Developing catalysts made from earth-abundant materials and have better performance is more urgent than ever.

This thesis is aimed to develop computational methods that search for new catalyst materials more efficient under time and budget constraints. Three areas of contributions in this thesis are: (i) developing open-source workflows and software tools, (ii) determining catalyst properties to screen catalyst more accurately and efficiently, and (iii) closer coupling of experiments and computational methods. We first present our informatics tool Generalized Adsorption Simulator for Python (GASpy) that uses dynamic dependency graphs to share, organize, and schedule adsorption energy calculations to achieve efficient descriptor-based catalysts surface screening. We then show a framework that predicts average surface energy using high-throughput DFT and a machine learning framework. This framework allows us to further down select catalyst surfaces based on their surface stability and the likelihood of their experimental occurrence. We also develop an algorithm that improves upon heuristic strategy and accelerates the search for more accurate descriptors (i.e. the lowestenergy configurations) in complex adsorbate-surface systems with machine learning potentials. Lastly, we implement and benchmark sequential learning agents that allow for the differentiation of data points of different fidelities with a sequential learning framework for closer coupling of theory and experiments.