Butene (C4) isomers are key feedstocks to produce valuable chemicals, and their separation remains a major challenge due to the high energy demands of traditional distillation, driving interest in more energy-efficient separation techniques. As an alternative, adsorptive separation has gained significant attention and in this regard, metal-organic frameworks (MOFs) have emerged as promising adsorbents with easily tunable porous structures and chemical functionalities. However, accurately modeling flexibility and adsorption behavior in flexible MOFs is challenging, as it requires potentials capable of capturing structural transition upon external stimuli (e.g., in response to adsorption). In this study, we employ a hybrid potential to simulate the temperature-dependent structural transition of a flexible MOF (CALF-20) upon n-/iso-butene adsorption, and further evaluate its separation performance. Specifically, a machine learning potential (MLP) is developed for CALF-20, while conventional Lennard-Jones (LJ) potential is adopted C4 isomers. This hybrid approach is found to effectively capture framework-C4 interactions, elucidate both temperature-dependent and guest-induced structural transition of CALF-20, thus offering a streamlined solution to the rational design of new MOFs toward industrially important hydrocarbon separation.
