Machine Learning-Driven Screening of Atomically Precise Pure Metal Nanoclusters for Oxygen Reduction

Abstract

Developing efficient catalysts for the oxygen reduction reaction (ORR) in proton-exchange membrane fuel cells is challenging due to high power density and durability requirements. Subnanometer clusters (SNCs) show promise, but their fluxional behavior and complex structure–activity relationships hinder catalyst design. We combine density functional theory (DFT) and machine learning (ML) to study transition metal-based subnanometer nanoclusters (TMSNCs) ranging from 3 to 30 atoms, aiming to establish structure activity relationship (SAR) for ORR. Subdividing the data set based on size and periodic groups significantly improves the accuracy of our ML models. Importantly, the ML model predicting the ORR catalytic performance is validated through DFT calculations, identifying 12 promising catalysts. Late group TMSNCs exhibit enhanced ORR activity, reflected in a noticeable shift toward Au/Ag metals on the volcano plot. This underscores the importance of investigating late group TMSNCs alongside Pt for the ORR. ML accelerates TMSNC design, surpassing computational screening and advancing catalyst development.