Harnessing artificial intelligence for modeling amorphous and amorphous porous palladium: A deep neural network approach

Abstract Amorphous and amorphous porous palladium are key materials for catalysis, hydrogen storage, and functional applications. Conventional ab initio methods are computationally prohibitive for simulating large-scale systems and fail to reproduce mesoporous features accurately. To overcome this limitation, this study employs a deep neural network trained on 33,310 atomic configurations from ab initio simulations to model their interatomic potential. The AI-driven approach accurately predicts structural properties while significantly reducing computational costs. Validation against theory and experiment data confirms its reliability in reproducing structural distributions with deviations of only 2.22% in the pair distribution function and 1.53% in the plane-angle distribution. Graphical abstract