Toward data-driven predictive modeling of electrocatalyst stability and surface reconstruction

Abstract

Catalyst dissolution and surface restructuring are ubiquitous in electrocatalysis, often leading to formidable activity–stability trade-offs and obscure electrochemically induced surface species that severely hinder the understanding and optimization of electrocatalysts under diverse harsh operating conditions. As even state-of-the-art characterization techniques lack the resolution and efficiency for the unambiguous elucidation of decomposition kinetics and reconstruction dynamics at electrocatalytic interfaces, many atomistic modeling approaches—following the recent advances in physics-driven machine learning—have been widely used to facilitate the atom-by-atom understanding and rational engineering of electrocatalyst stability and dynamics. This Perspective systematically assesses classical and data-driven approaches in theoretical surface science and computational catalysis, recognizing their achievements and highlighting their limitations in throughput, efficiency, accuracy, bias, transferability, and scalability toward enabling realistic and predictive modeling of electrocatalyst degradation and reconstruction. By examining different methods spanning first-principle simulations, surface sampling, neural network interatomic potentials, and generative deep learning models, it is underscored how such data-driven computational techniques help elucidate the precise nature of various key interfacial atomistic processes to address existing technical challenges in surface modeling and provide a new paradigm to optimize dissolution kinetics and restructuring dynamics for electrocatalyst design.