Active Learning-Driven Discovery of Sub-2 Nm High-Entropy Nanocatalysts for Alkaline Water Splitting

Abstract

High-entropy nanoparticles (HENPs) present a vast opportunity for the development of advanced electrocatalysts. The optimization of their chemical compositions, including the careful selection and combination of elements, is critical to tailoring HENPs for specific catalytic processes. To reduce the extensive experimental effort involved in composition optimization, active learning techniques can be utilized to predict and suggest materials with enhanced electrocatalytic activity. In this study, sub-2 nm high-entropy catalysts incorporating eight transition metal elements are developed through an active learning workflow aimed at identifying optimal compositions. Using initial experimental data, the approach successfully guided the discovery of a new octonary HENP catalyst with state-of-the-art performance in the hydrogen evolution reaction (HER). Catalyst performance is improved within the prediction uncertainty of the machine learning model. For the oxygen evolution reaction (OER), however, the initial model demonstrated limited predictive accuracy, leading to an assessment of the workflow's boundaries. These findings underscore how the integration of curated experimental data with active learning can accelerate electrocatalyst discovery, while also highlighting critical areas for further model refinement.