Predicting Gas Loading Response in All-Silica MFI Zeolites Using ChIMES Machine-Learned Interatomic Potentials

Zeolite membranes, with their tunable nanoporous frameworks and exceptional thermochemical stability, are attractive materials for advanced gas separation technologies. Predictive design of these membranes requires atomic scale understanding of how gas loading affects framework dynamics, adsorption and diffusion, challenges that remain experimentally intractable. Here we present a machine-learned interatomic potential for all-silica MFI zeolite, based on the ChIMES physics-informed machine learning framework and trained on a diverse set of structural configurations to capture the material’s inherent flexibility. This approach enables large-scale molecular dynamics simulations that investigate the interplay between gas insertion and the lattice, revealing how different loading levels influence framework distortions, pore accessibility and transport pathways. Our simulations provide atomic-level insight into structural evolution and dynamic response under operating conditions, creating a critical link between gas uptake and macroscopic membrane performance. Preliminary studies also suggest this model is transferable to similar zeolite topologies, supporting extended applications in the design of advanced nanoporous separation materials.