Modeling of CO2/CH4 Mixture Adsorption in Flexible Mg-MOF-74 via Machine-Learned Potentials

Abstract

Metal–organic frameworks (MOFs), particularly Mg-MOF-74 with open metal sites, offer a promising platform for the selective adsorption of CO2 over CH4. However, accurately modeling multicomponent gas adsorption in flexible frameworks remains challenging. In this work, we developed a fragment-based machine-learned potential (MLP) trained on high-level density functional theory data (PBE-D4/def2-TZVP) to describe all intramolecular and intermolecular interactions between the components of CO2/CH4 mixture and Mg-MOF-74. By integrating this MLP with a combined molecular dynamic-grand canonical Monte Carlo (MD-GCMC) hybrid scheme, we captured both framework flexibility and adsorption thermodynamics, enabling simulations of competitive adsorption and diffusion in multicomponent systems. Our results demonstrate that fragment-based MLPs can accurately represent binary gas mixtures in MOFs and reveal the critical role of framework flexibility in governing adsorption and transport behavior.