A new molecular pore network model for the structure of nanoporous materials, and in particular membranes, has been developed. The construction of the model starts with a three-dimensional (3D) box in which the atoms that constitute the material are distributed, either in crystalline form, or as an amorphous material which is obtained by annealing. The box is then tessellated using the Voronoi algorithm that partitions the space into irregular 3D polyhedra. A fraction of the polyhedra is then designated as the pores of the material, and all the atoms inside such polyhedra, as well as the dangling (singly-connected) atoms are removed. The size distribution of pore polyhedra can be tuned to match experimental data for the pore size distribution (PSD) of a given nanoporous material with any correlation function. Since the pore polyhedra are interconnected, the model takes into account the effect of the pore connectivity.

Because the material is randomly tessellated and the dangling atoms are removed, the pores have rough internal surfaces, which is consistent with what is known experimentally. To test the model, we simulate adsorption isotherms for nitrogen, using equilibrium molecular dynamics simulations in three silicon carbide (SiC) membranes by adjusting the average pore size of the model to the experimental data. Good agreement was obtained between the simulated and measured isotherms. The experimentally-validated model was then used for modeling transport of gaseous mixtures in the SiC membrane under a variety of conditions (e.g., pore size, pressure, and temperature.)