Chemical separations based on heat cost about 20% of the worlds energy. Finding ways to separate chemicals at ambient temperatures and pressures is key to the energy transition. Porous materials have proven very useful when it comes to separations based on non-thermal chemical characteristics. Matching outcomes from computational chemistry studies to experimentally measurable properties is the foundation of the materials design pipeline. IAST theory pre- dicts adsorption behavior of adsorbents in porous media such as zeolites and Metal-Organic Framework, omitting expensive computations of mixtures with Monte Carlo methods. Experimentally chromatographic methods have proven very useful for separation of adsorbents based on adsorbate affinity and binding stoichiometry, but is limited to studying mixtures. Herein we report the regres- sion of an Invertible Autoencoder (IAE) for the forward and backward mapping of pure and mixture isotherms. Pure isothermal binding curves are modelled as a 3-site Langmuir-Freundlich isotherm, with a broad range of equilibrium pressure and heterogeneity factors. A synthetic dataset is generated from the pure isotherms and mixture isotherms calculated with RUPTURA. The IAE predicts pure and mixture isotherms with high precision in both the Henry and high fugacity regime, for up to 6 components and 5-site isotherms. This work contributes to inverting the full design pipeline from physical gas separation to adsorbate design, enabling property-guided materials discovery.
Youri Ran