Machine learning-enabled fast exploration of stable and active single-atom catalysts for oxygen evolution reaction

Abstract

Oxygen evolution reaction (OER) can convert renewable energy into hydrogen through water electrolysis. Identifying stable and active single-atom catalysts (SACs) for OER under acidic conditions holds great promise for developing cost-effective and efficient energy storage solutions, but challenging due to the vast number of potential material compositions and diverse surface morphologies. Here, to accelerate new discoveries, we present a high-throughput screening (HTS) framework that leverages the power of machine learning (ML) and density functional theory (DFT). The proposed framework includes an assessment of both the thermodynamic and electrochemical stability of support surfaces. In addition, the integration of ML and uncertainty quantification for predicting the binding energies dramatically reduces the computational cost (by over a factor of 10), facilitating the identification of catalytically active SACs. Following the proposed scheme, we suggest 14 new promising SACs for OER across the 795 binary oxide supports and 21 transition metal atom combinations. These catalysts are found to break the scaling relation due to the enhanced *OOH binding with the support, which arises from favorable hydrogen bonding interactions.