From prediction to synthesis: DFT-active learning-guided design of multimetallic catalysts for hydrogen evolution

Abstract

Multimetallic alloys have garnered significant attention in electrocatalysis due to their enhanced properties, such as superior electrical conductivity, exceptional shock stability, and abundant unique binding sites. Furthermore, substituting noble metals with cost-effective alternatives broadens their potential for scalable production. Despite these advantages, designing multimetallic alloy catalysts remains a formidable challenge due to their huge compositional space, making conventional trial-and-error approaches inefficient and almost impossible. Here, we present an active learning framework with density functional theory (DFT) to identify optimal compositions of multimetallic alloys with high hydrogen evolution reaction (HER) catalytic activity. A Gaussian process regressor (GPR) model serves as the learning agent, predicting adsorption energies at each binding site. Iteratively, the model evaluates prediction uncertainties and prioritizes binding sites for new DFT calculations, thereby maximizing informational gain. Remarkably, out of the full span of 390,625 available binding sites, our method successfully discovered high-performance alloy compositions through only 600 DFT-calculated surfaces. These compositions were experimentally validated using the carbothermal shock method, demonstrating excellent catalytic performance. This work underscores the transformative potential of active learning in catalyst discovery, offering a time-and cost-efficient approach to screening multimetallic alloys. The proposed framework can be extended to a variety of catalytic systems, paving the way for rational design in electrocatalysis.