A new mathematical framework is introduced for modeling diffusion in nanoporous materials or on surfaces exhibiting heterogeneity in properties over large length scales while retaining molecular scale information typically captured by molecular simulations only. This framework entails first the use of newly developed mesoscopic equations derived rigorously from underlying master equations by coarse-graining statistical mechanics techniques. Homogenization techniques are then employed to derive the leading-order effective mesoscopic models that are subsequently solved by spectral methods. These solutions are also compared to direct numerical simulations for selected two-dimensional model membranes with defects, when attractive adsorbate-adsorbate interactions affect particle diffusion. It has been found that not only the density but also the dispersion of defects significantly alters the macroscopic behavior in terms of fluxes and concentration patterns, especially when phase transitions can occur. It is also shown that homogenization techniques could potentially offer a promising alternative to direct numerical simulations, when complex, large-scale heterogeneities are present. Implications for various applications, including heterogeneous catalysts, materials growth, and separations using membranes, are also discussed.
* Corresponding author.
INTRODUCTIONThe introduction of Monte Carlo (MC) techniques 1 in conjunction with the rapidly increasing computational power have revolutionized our understanding about the mesoscopic structure, thermodynamic, and transport properties of a large spectrum of problems for which molecular interactions are important. Examples include bulk liquids, solids, surface reconstruction phenomena, protein folding, and crystallization 2-6 . Aside from equilibrium problems, MC methods, especially on a lattice, have been successfully employed for irreversible problems such as crystal growth and catalytic reactions [7][8][9][10][11][12] . MC simulations solve directly an underlying master equation, and given sufficient information about transition probabilities, they provide its exact solution.Despite the enormous progress achieved so far, MC simulations are computationally very intensive and limited to relatively small time and length scales. The issue of length scales is not restrictive when only microscopic inhomogeneities are present as a result, for example, of intermolecular forces. These can typically be captured adequately via periodic boundary conditions and sufficiently large simulation boxes. However, there are three broad classes of problems, which are currently intractable by MC simulations due to macroscopic inhomogeneities. The first class includes systems that exhibit mesoscopic or macroscopic patterns as a result of self-regulation mechanisms. Examples include Turing patterns 13 and patterns resulting from the competition of microphase separation, driven by attractive interactions, and chemical reaction 14,15 . The second class encompasses forced systems that opera...