From Atomic Motif to Realistic Single Atom Catalysts through Machine Learning Interatomic Potentials
Seokhyun Choung, Miyeon Kim, Jinuk Moon, Jeong Woo Han
Abstract
High Resolution Image Download MS PowerPoint Slide Metal–nitrogen–carbon (M-N-C) catalysts demonstrate exceptional electrochemical performance, with density functional theory (DFT) simulations successfully guiding atomic-scale optimization of coordination environments. However, recent experiments reveal that catalyst performance depends on phenomena beyond DFT’s spatiotemporal limits. This Perspective examines how machine learning interatomic potentials (MLIPs) bridge this critical gap, achieving orders-of-magnitude acceleration while maintaining near-DFT accuracy. MLIPs capture previously inaccessible phenomena spanning atomic to mesoscopic scales, including structural complexity and electrolyte dynamics. These capabilities reveal how support architecture, collective site interactions, solvation, and reaction kinetics at the mesoscale determine rate-limiting steps in electrochemical reactions. By connecting atomic-level understanding to experimentally relevant scales, MLIPs transform catalyst design from isolated site optimization to comprehensive multiscale engineering.