Accelerating hydrogen evolution catalyst discovery via data-driven strategy for high-performance single-atom catalysts embedded in h-BN

Abstract

Single-atom catalysts (SACs) have attracted considerable attention for electrochemical reactions due to their high atomic efficiency and tunable catalytic properties. Here, we systematically investigate the hydrogen evolution reaction (HER) activity of transition metal (TM) atoms embedded in hexagonal boron nitride (h-BN) with engineered vacancy defects, leveraging density functional theory (DFT) to examine 28 different TM@BN configurations at both single- and double-vacancy sites. Our calculations show that the TM atoms are strongly bound to the defect sites, ensuring robust structural integrity and resistance to aggregation and leaching under electrochemical conditions. The hydrogen adsorption free energies (DGH) indicate that Ir@SV-BN, Mo@SV-BN, and Pt@SV-BN exhibit near-optimal adsorption strengths, which was further supported by climbing-image nudged elastic band (CI-NEB) analyses revealing low activation barriers dominated by the Volmer-Heyrovsky mechanism. To expedite the discovery of high-performance catalysts, we employed machine learning (ML) models trained on the highthroughput DFT database, achieving an accuracy of R2 = 0.96 in predicting overpotentials and identifying key structural and electronic descriptors governing HER activity. Ab initio molecular dynamics (AIMD) simulations confirm the thermal and electrochemical stability of selected TM@BN systems under realistic operational conditions. Taken together, these findings highlight the potential of BN-supported SACs as next-generation electrocatalysts for sustainable hydrogen production and underscore the effectiveness of integrating computational screening, ML-driven optimization, and mechanistic insight to guide the rational design of acid-resistant, high-performance HER catalysts. (c) 2025 Science Press and Dalian Institute of Chemical Physics, Chinese Academy of Sciences. Published by Elsevier B.V. and Science Press. All rights are reserved, including those for text and data mining, AI training, and similar technologies.