Machine-learning-guided tungsten single atoms promote oxyhydroxides for noble-metal-free water electrolysis

Lowering the overpotential of oxygen evolution reaction with electrocatalysts is essential for efficient renewable-electricity-driven electrolysis. Active noble-metal catalysts suffer from leaching and scarcity, while non-noble alternatives face limited intrinsic activity. Here we combine computational guidance with experimental validation to identify atomically dispersed tungsten within NiFe oxyhydroxide, namely W1-NiFeOOH, as a promising noble-metal-free oxygen evolution reaction catalyst. An equivariant transformer-based machine-learning interatomic potential accelerates out-of-domain adsorption energy predictions and nominates W1-NiFeOOH from 3,976 single-atom-incorporated metal oxyhydroxide configurations. Cyclic-electrodeposited W1-NiFeOOH achieves a high current density of 13.1 A cm-2 at 2.0 V and remains stable for 500 hours in alkaline exchange-membrane water electrolysis with commercial membranes. In situ spectroscopy and density functional theory calculations suggest that subsurface W promoter induces synergistic electron redistribution at neighboring Ni-O-Fe edge active sites, thereby lowering the proton-coupled electron-transfer barrier for the deprotonation step and facilitating transformation into the active γ-phase. This integrated computational-experimental workflow provides a blueprint for cost-effective catalyst design for sustainable energy systems. Sustainable hydrogen production requires efficient oxygen evolution catalysts. Here, the authors identify non-noble single-atom catalysts via machine learning-assisted computational screening and reveal their active-site mechanisms for high-performance anion exchange membrane water electrolysis.