Unraveling Scaling Relationships in Dual-Atom Catalysts with Electronic Descriptors: A Machine Learning Investigation for OER/ORR Activity

Abstract

Dual-atom catalysts (DACs) have emerged as a new frontier in heterogeneous catalysis, offering improved stability and superior performance in key electrocatalytic reactions. However, identifying optimal multimetallic DACs combination for a multistep reaction is challenging due to the vast chemical space. Herein, we develop a machine learning (ML) framework to expedite the screening of DACs, which consist of a heterometallic dimer embedded in the surface layer of a metal host, for improved oxygen evolution reaction (OER) and oxygen reduction reaction (ORR) performance. We encode the solid-state-derived d-band descriptors to accurately train the ML model and effectively capture the nonmonotonic bifunctional activity on DACs, without requiring expensive DFT calculations. Interestingly, we identify the nonscaling behavior of these DACs, with CoPd and CoCu dimer exhibiting superior OER and ORR activity. Furthermore, we employ the surface charging method to evaluate the potential-dependent activity and reveal the nonlinear relationship between catalytic activity and electrode potential. Overall, this study established the pivotal role of d-states in governing the catalytic performance and offers a practical pathway to accelerate the discovery of next-generation electrocatalysts for fuel cell applications. © 2026 American Chemical Society