Hierarchical Site‐to‐Composition Machine Learning for High‐Entropy Electrocatalysts Design

Abstract

High‐entropy alloys (HEAs), with their unique surface chemical disorder and rich active site distributions, hold great promise as electrocatalysts for hydrogen evolution. However, their vast compositional space poses a fundamental challenge for rational catalyst design. In this work, we propose a Hierarchical Site to Composition Machine Learning (HS2C‐ML) framework to accelerate the discovery of efficient HEA catalysts for alkaline hydrogen evolution reaction (HER). At the site level, a fine‐tuned, site‐resolved machine‐learning model rapidly predicts *H and *OH adsorption energies across complex HEA surfaces, enabling high‐throughput activity evaluation from adsorption‐energy distributions. At the composition level, a physics‐informed sparsifying operator algorithm establishes interpretable relationships between alloy composition and catalytic performance, allowing direct screening across thousands of compositions without explicit atomistic calculations. By screening ∼16,000 HEA compositions, we computationally prioritize a series of promising HEA candidates with favorable predicted alkaline HER activities, and the relevance of the predicted composition space is supported by consistency with selected literature reports and by the synthesis and electrochemical evaluation of one representative catalyst. This work provides a general and scalable strategy for data‐driven catalyst design in high‐dimensional HEA compositional spaces.