(684c) Machine Learning Screening of Alloy Surfaces for Catalytic Stability in Reaction Conditions

Metal alloys are increasingly gaining traction in heterogeneous catalysis because they can have superior performance over single metal systems. A catalytic surface should be stable under reaction conditions to be effective. However, it takes significant effort to calculate stability of a surface, as this requires determining the most common surface intermediates from quantum chemical calculations, determining the energy of the surface with these intermediates and comparing to other possible arrangements of the same elements. As a result, we provide a data-efficient machine learning approach to accurately and efficiently predict the surface energies of bare metal alloy surfaces and combine it with a model for predicting the energetics of all intermediates in a reaction pathway.

We trained machine learning models for surface formation energies by screening multiple matrix-based feature sets and fitting to multiple datasets. We combined this with our previous, general model for predicting adsorption energies. This combination of models for predicting alloy surface energies with models for predicting adsorption energies of many species opens new avenues in efficient catalyst screening. We can predict the energy of an alloy surface in reaction conditions and compare it to surfaces with different arrangements of the same elements to identify which alloy surfaces are stable in reaction conditions. This significantly increases screening efficiency. We apply this approach to multiple reactions, including CO oxidation and methane steam reforming.