Non-equilibrium molecular simulations reveal a pore-flow-dominated transport mechanism in pervaporation membranes

Abstract

Pervaporation (PV) is a membrane-based separation process particularly suited for niche applications such as dehydration of organic solvents, separation of azeotropic or heat-sensitive mixtures, and desalination of high-salinity waters. Understanding how solvents permeate through polymer membranes under non-equilibrium conditions is critical for advancing PV technology, yet the underlying transport mechanisms remain poorly understood. In this study, we use non-equilibrium molecular dynamics (NEMD) simulations to investigate single- and mixed-solvent permeation through polydimethylsiloxane (PDMS) membranes, with a particular focus on whether and how a phase transition occurs within the membrane matrix. Our results reveal spatial gradients in pore size and porosity, non-uniform solvent distribution, and directional solvent transport pathways. Notably, a distinct transition zone is observed where solvent molecules shift from clustered, liquid-like viscous flow to individual, gas-like diffusion. This transition coincides with the internal pressure reaching the saturated vapor pressure of the solvent, indicative of a liquid–vapor phase change. Solvent trajectories and coordination number analyses support a pore–flow transport mechanism involving both pressure–driven viscous flow and gaseous surface diffusion in series. We further analyzed a water-ethanol solvent mixture, revealing that water retains a cluster-to-molecule transition behavior, whereas ethanol, present at low concentration, can exhibit solute-like characteristics. Simulations of crosslinked polyvinyl alcohol (PVA) membranes for PV desalination likewise revealed a liquid-vapor phase transition of water. These findings suggest that solvent transport during PV is better described by a pore–flow model and underscore the importance of incorporating pore structural characteristics—such as pore size distribution, connectivity, and porosity—into transport models. Overall, our work provides new mechanistic insights that could guide the design of PV membranes and improve process predictability under realistic conditions.