Toward accelerated discovery of solid catalysts using extrapolative machine learning approach

Abstract

Designing novel catalysts is pivotal for overcoming numerous energy and environmental challenges. Although data science approaches, particularly machine learning (ML) approaches, hold promise for accelerating catalyst development, discovering truly novel catalysts through ML remains rare. This is largely due to the perceived inability of the ML models to extrapolate and identify exceptional materials. In this Review, I present our approach taken to tackle this limitation. Specifically, we employed an advanced ML methodology that could make extrapolative predictions. This approach led to the discovery of multielemental solid catalysts for CO2 hydrogenation to CO. The results not only demonstrate the immense potential of ML in catalysis research but also set a new standard for the rapid development of high-performance catalysts.