Deep learning combined with IAST to screen thermodynamically feasible MOFs for adsorption-based separation of multiple binary mixtures

The structures of metal–organic frameworks (MOFs) can be tuned to reproducibly create adsorption properties that enable the use these materials in fixed-adsorption beds for non-thermal separations. However, with millions possible MOF structures, challenge is find best separate a given mixture. Thus, computational, rather than experimental, screening necessary identify promising merit further examination, process traditionally done using molecular simulation. even simulation become intractable when an expansive database their separation at more few composition, temperature, and pressure combinations. Here, we illustrate progress toward alternative computational framework efficiently highest-performing MOFs separating various gas mixtures variety conditions fraction cost This uses “multipurpose” multilayer perceptron (MLP) model predict single component small adsorbates, which, upon coupling ideal adsorbed solution theory (IAST), binary such as Xe/Kr, CH4/CH6, N2/CH4, Ar/Kr multiple compositions pressures. For this MLP+IAST work sufficient accuracy, found it critical MLP make accurate predictions low pressures (0.01–0.1 bar). After training capability, 95th 90th percentiles performance determined from calculations were 65% 87%, respectively, same simulation-predicted percentile across several diverse (on average). validating our framework, used clustering algorithm “privileged” are high performing separations conditions. As example, focused on industrially relevant 80/20 Xe/Kr 1 bar N2/CH4 5 bars. Finally, free energies (calculated entire database) privileged also likely synthetically accessible, least thermodynamic perspective.