ABSTRACT Predicting electrocatalyst performance under realistic conditions remains challenging due to the limitations of quantum mechanical (QM) methods. Recent advances in artificial intelligence (AI) have enabled data‐driven potentials that bridge quantum accuracy with realistic and nanoscale catalyst modeling. Here, a computational‐experimental approach is presented to investigate the oxygen reduction reaction (ORR) activity and durability of intermetallic PtCo nanoparticles by integrating QM calculations with neural network potential (NNP) based real‐sized simulation. The NNP, trained on Pt, Co, and PtCo systems, enabled accurate modeling of nanoparticles up to 5 nm containing several thousand atoms. ORR activity was elucidated through analysis of the d ‐band center, oxygen adsorption energies, free energy diagrams, and validated by rotating disk electrode (RDE) measurements. Dissolution potentials (U diss ) calculated via NNP revealed strong dependence on core configuration and particle size, with onion‐like ordered PtCo showing the highest U diss . These trends were experimentally validated through accelerated durability tests (ADT), and the consistency between NNP and QM results confirms the model's predictive reliability. This combined theoretical and experimental approach provides a reliable and scalable pathway for understanding degradation mechanisms and designing durable fuel cell electrocatalysts.
Lee et al. (Tue,) studied this question.