Spatial-adaptive active learning identifies ultra-durable and highly active catalysts for acidic oxygen evolution reaction

Oxygen evolution reaction (OER) catalysts face a major challenge in the practical implementation of acidic water electrolysis for hydrogen production, primarily due to limitations in catalytic activity and stability. Despite extensive research, the development of acidic OER catalysts still relies largely on trial-and-error experimentation rather than AI-driven, target-oriented approaches. In this work, we address these limitations by introducing a spatial-adaptive active learning strategy integrated with closed-loop experimentation for targeted catalyst optimization in two stages. In the first stage, Bayesian optimization identifies highly active catalysts and a conditional variational autoencoder generates an adaptive low-overpotential subspace of stability candidates, while the second stage active learning finds the most stable catalyst within this subspace. Using this strategy, we discover a novel Cu-RuO 2 catalyst that exhibits remarkable stability for 625 h and an overpotential of 177 mV at a current density of 10 mA cm −2 . We provide detailed characterization and mechanistic insights into the newly discovered catalyst. Our study presents a transformative method for accelerating the design of stable acidic OER catalysts, thereby advancing the feasibility of large-scale green hydrogen production via acidic water electrolysis.