Abstract
Organic solvent nanofiltration (OSN) stands out as an energy-efficient and low-carbon footprint technology, currently reliant on polymeric membranes. With their exceptional chemical stability and tunable sieving properties, two-dimensional (2D) nanolaminate membranes present distinct advantages over conventional polymer-based membranes, attracting tremendous interest in the OSN community. Computational approaches for designing innovative 2D nanolaminates exhibit significant potential for the future of OSN technology. Imitating the pressure gradient in filtration processes by applying an external force to atoms within a predefined slab, boundary-driven nonequilibrium molecular dynamics ((BD)-NEMD) is a state-of-the-art simulation method with a proven track record in investigating the water transport in nanopores. Nevertheless, implementation of (BD)-NEMD for a broad range of solvents poses a challenge in estimating the OSN performance of theoretical membranes. In this work, we developed a (BD)-NEMD protocol that elucidates the effects of several computational details often overlooked in water simulations but are crucial for bulky solvent systems. We employed a MXene (Ti(3)C(2)O(2)) nanochannel as a model membrane and examined the transport of nine solvents (methanol, ethanol, acetone, n-hexane, n-heptane, toluene, ethyl acetate, dichloromethane, and water) having different properties. First, the impact of ensemble type, thermostatting, channel wall model, and restraining force constant was elaborated. After optimizing the thermostatting approach, we demonstrated that the location of the force slab particularly affects the flux of bulky solvents by changing the density distribution in the feed and permeate sides. Similarly, the uniformity of intramolecular force distribution in bulky solvents and resulting flux are shown to be prone to manipulation by slab boundaries. Next, the magnitudes of the external force generating a linear relation between the pressure gradient and solvent flux were identified for each solvent to ensure that calculated fluxes could be extrapolated to experimentally related pressures. This linear relation was also validated for a mixture system containing 50% ethanol and 50% water. We then correlated the calculated solvent permeances with various solvent properties, such as viscosity, Hansen solubility parameters, kinetic diameter, and interaction energy. Remarkably, we observed a linear correlation with an R(2) value of 0.96 between permeance and the combined parameter of viscosity and interaction energy. Finally, the solvent permeances calculated with our proposed methodology closely align with the experimentally reported data. Overall, our work aims to serve as a practical guide and bridge the gap in established simulation methods that are suited for a broad range of solvents and membrane materials.