MOFs to Enhance Green NH₃ Synthesis in Plasma Reactors: Hierarchical Computational Screening Enhanced by Iterative Machine Learning

Plasma reactors are promising to decarbonize the production of NH₃, but their NH₃ energy yields need to improve to facilitate their broad adoption. Two emerging strategies to reduce energy inefficiencies aim to protect the freshly formed NH₃ from destruction by the plasma by leveraging NH₃ adsorption properties of porous materials as either catalyst supports or as membranes. As metal–organic frameworks (MOFs) are promising porous materials for adsorption-based applications, we performed large-scale computational screening of 13,460 MOFs to study their potential for the above-mentioned uses. To reduce computational cost by ∼10-fold, we developed a generalizable hierarchical MOF screening strategy that starts with the selection of a 200-MOF set based on NH₃ adsorption Henry’s constants, for which the relevant performance metrics are calculated via molecular simulation. This set is used to “initialize” a machine learning (ML) model that predicts the relevant metrics in the whole MOF database, in turn guiding the selection of additional promising MOFs to be evaluated via molecular simulation. The ML model is then iteratively refined leveraging the emerging molecular simulation data from the MOFs selected at each iteration from the ML predictions themselves. From evaluation of only ∼10% of the database, for each use (catalyst support or membrane), 20 extant MOFs were holistically assessed and proposed for experimental testing based on desirable adsorption properties as well as complementary properties (e.g., high thermal decomposition temperature, constituted by earth abundant metals, etc.). Data-driven material design guidelines also emerged from the screening. For instance, a pore diameter of ∼10 Å and a heat of adsorption of ∼90 kJ/mol were found beneficial for the catalyst support use. On the other hand, for the membrane-based strategy, a pore diameter of ∼2.75 Å and a heat of adsorption of ∼80 kJ/mol were found beneficial. The presence of V was found beneficial for both uses.