Accelerated Discovery of Oxygen Evolution Electrocatalysts by Machine Learning

Iridium oxide (IrO₂) is the predominant electrocatalyst for the oxygen evolution reaction (OER) with high stability in both alkaline and harsh acidic media, but its low efficiency and high cost limit its applications. In this work, we have developed a high-throughput computational strategy based on density functional theory (DFT) and employ the machine-learning (ML) techniques to extensively explore the optimal dopants in IrO₂ and enhance its OER activity. Thirty-six kinds of metal dopants are considered to substitute for Ir to form binary and ternary metal oxides with different concentrations, and the most stable surface structures are selected from a total of 4,648 structures for OER activity evaluation. By using the neural network language model (NNLM), we are able to associate the atomic environment with the formation energies of crystals and adsorption free energies of OER intermediates, which is efficient for complex compounds and accurately evaluates the structural stability and electrochemical activity. A series of potential candidates have been identified as superior OER catalysts, and some of them are novel with the lowest calculated overpotential. Our strategy could efficiently explore the promising electrocatalysts, especially for evaluating complex multi-metallic compounds.