Developing Transferrable Machine Learning Force Fields for Metal-Organic Frameworks

Metal-Organic Frameworks (MOFs) have garnered significant attention as promising materials for CO₂ capture. However, the development of accurate force fields to describe their potential interactions is challenging due to their large number of atoms and inherent complexity. Classical force fields using the point charge approach struggle to accurately model interactions involving CO₂ and H₂O. In this study, we introduce a transferable machine learning force field that enables simulating potentials across a diverse set of MOFs, rather than focusing on just a few types. Using this new force field, we screen an existing MOF database to identify candidates for CO₂ capture and evaluate CO₂/H₂O selectivity with ab initio levels of accuracy. Additionally, we chemically analyze the critical factors and characteristics necessary for a MOF to be a promising candidate for CO₂ capture, providing a comprehensive understanding of what makes certain MOFs more effective at capturing CO₂. This approach sets a new standard for the rapid and accurate screening and development of MOFs, accelerating progress towards effective solutions for carbon capture technologies.