Accelerating materials discovery for electrocatalytic water oxidation via center-environment deep learning in spinel oxides

Abstract

Identifying efficient electrocatalysts for the oxygen evolution reaction (OER) is vital for sustainable energy. This study focuses on efficient spinel OER electrocatalysts. We utilize a covalent competition-based interpretable machine learning process, combining DFT and ML, to evaluate 5329 spinel structures for identifying promising OER electrocatalysts. The key indicator in these structures is the covalent competition MAX(D_T, D_O) between cations and oxygen anions. Center-Environment (CE) features exhibit optimal predictive accuracy for MAX(D_T, D_O), compared with other feature models e.g., Magpie, ElemNet, and Voronoi. 14 structures with promising OER activity were predicted from 5329 spinel structures using a multi-level screening framework, and interpretability has been provided through SHAP analysis. MoAg₂O₄, synthesized experimentally, exhibited superior OER activity to MoNa₂O₄, ZnAl₂O₄, and commercial RuO₂, achieving a current density of 10 mA cm⁻² at 284 mV overpotential with long-term stability. The exceptional electrocatalytic performance of MoAg₂O₄ is attributed to highly efficient electron transfer coupled with optimized adsorption for OER intermediates. The synergistic effect of these two attributes improves the reaction kinetics and reduces the free energy of the rate-limiting step in the OER process. This study offers a strategy for creating better spinel electrocatalysts for the OER and advances machine learning in electrocatalyst discovery.