Materials Informatics for Catalyst Stability & Functionality

Abstract

Accelerating materials design & discovery will be critical to reducing society's dependence on fossil fuel-based energy sources. Computational methods such as density-functional theory, high-throughput workflow automation, and machine learning help to enable the understanding, discovery, and optimization of functional energy materials. I will demonstrate how these broad classes of tools may be applied to various important problems in the computational study of catalysts in three case studies. First, I will explain how computational probes of mechanical stability and catalytic suitability can be used to discover new forms of two-dimensional CO2 reduction photoelectrocatalysts. Second, I will show how random forests combined with feature engineering can be used to automate materials characterization efforts from X-ray absorption spectroscopy and discover new interpretable fingerprints in transition metal oxide K-edge spectra. Finally, I will detail ongoing efforts in developing a benchmarking dataset which can be used to probe the sensitivity of varying levels of theory for machine learning based interatomic potentials, which may then be used to study the stability of transition metal catalysts on experimentally relevant timescales.