A machine learning scheme for the catalytic activity of alloys with intrinsic descriptors

Abstract

The application of density functional theory (DFT) has been accelerating the screening and design process of alloy catalysts for the carbon dioxide reduction reaction (CO₂RR), but the catalyst design principle still cannot be universally used to date because of the time-consuming DFT calculations and the unclear structure–property relationship of alloy catalysts. To address these issues, we combine machine learning methods and descriptors based on the intrinsic properties of substrates and adsorbates to develop a model, which allows rapid screening through a large phase space of alloys with the usual DFT accuracy. Our ML scheme sheds light on the size of active centers on transition metals and alloys, the effect of alloying on engineering adsorption energy, and the coupling mechanism of different adsorbates with substrates. These findings not only help us understand the structure–property relationship of alloy catalysts and the reaction mechanism of the CO₂RR, but also provide a basis for the design of catalysts. This universal design framework can be extended to other catalysts and other reactions towards efficient and cost-effective potential catalysts.