Elucidating Poisson-Like Anisotropic Deformation in Flexible MOFs for Precision C2 Hydrocarbon/CO2 Sieving via Multiscale Simulations and Interpretable Machine Learning.

Precise kinetic separation of light hydrocarbons (C2H2, C2H4, C2H6) and CO2 remains challenging due to their overlapping kinetic diameters and subtle host-guest interactions. Here, we rationally functionalized a series of flexible metal-organic frameworks, TYUT-18-X (X = -H, -OH, -NH2, -NO2, -CH3, and -CF3), establishing an efficient platform for dynamic molecular sieving. Combining multiscale simulations with machine-learning interpretability analysis, we unveil an anisotropic "Poisson-like" framework deformation during guest diffusion. This synchronized expansion-contraction along distinct crystallographic axes functions as a dynamic molecular gate, dramatically amplifying kinetic selectivity. Specifically, ligand functionalization effectively tailors diffusion barriers: in TYUT-18-CF3, the barrier reaches 176.43 kJ mol-1 for C2H6 versus 110.81 kJ mol-1 for C2H2, yielding an exceptional theoretical barrier-based selectivity of 3.18 × 1011 and a dynamic diffusivity ratio (DC2H2/DC2H6) of 3.8 × 102 from finite-temperature MD. Furthermore, SHAP analysis identifies the localized pore aperture and functional moiety as the governing descriptors dictating this dynamic sieving effect. This work highlights high-performance candidates for C2Hn/CO2 separations and introduces a general paradigm for programming guest-responsive structural dynamics in flexible porous materials.