Interpretable machine learning on C3H6 and C3H8 diffusion in covalent organic frameworks: Incorporating the effects of framework flexibility

Abstract

The separation of propylene (C₃H₆) and propane (C₃H₈) is vital for producing high-purity C₃H₆. Molecular diffusion in porous materials governs equilibrium in adsorption-based separations and affects membrane efficiency by controlling permeation rates. In molecular dynamics (MD) simulations, the framework flexibility of porous materials significantly influences diffusion but is often overlooked. This study explores how the flexibility of covalent organic frameworks (COFs) impacts the diffusion of C₃H₆ and C₃H₈ using high-throughput simulations. A classification model with perfect accuracy was developed using the sure independence screening and sparsifying operator (SISSO) algorithm to predict the effect of flexibility on gas diffusion. Descriptor analysis identified PLD and LCD as key features. Reliable regression models were constructed to predict diffusion coefficients. The interpretable SISSO models show that COFs with carbonyl or hydrogenated groups facilitate gas diffusion, while ether bonds inhibit gas diffusion, offering valuable insights for understanding C₃H₆ and C₃H₈ diffusion in COFs.