Elucidating Poisson‐Like Anisotropic Deformation in Flexible MOFs for Precision C 2 Hydrocarbon/CO 2 Sieving via Multiscale Simulations and Interpretable Machine Learning

Precise kinetic separation of light hydrocarbons (C 2 H 2 , C 2 H 4 , C 2 H 6 ) and CO 2 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, –NH 2 , –NO 2 , –CH 3 , and –CF 3 ), 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‐CF 3 , the barrier reaches 176.43 kJ mol −1 for C 2 H 6 versus 110.81 kJ mol −1 for C 2 H 2 , yielding an exceptional theoretical barrier‐based selectivity of 3.18 × 10 11 and a dynamic diffusivity ratio ( D C2H2 / DC2H6 ) of 3.8 × 10 2 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 C 2 H n /CO 2 separations and introduces a general paradigm for programming guest‐responsive structural dynamics in flexible porous materials.