(169ct) Diversity-Driven, Bayesian Optimization of MOF Designs to Optimize Performance for Environmental Applications Involving NH3 Adsorption

Metal-organic frameworks (MOFs) are promising materials to engender technology-enabling properties in numerous applications. However, one significant challenge in MOF development is their overwhelmingly large design space, which can be intractable to exhaustively explore experimentally (and even computationally) to find optimal designs. Assessed by its ability to optimize three NH3 sorption-based performance metrics that depend on different extent on material chemistry and structure, herein we demonstrate a diversity-driven optimization framework (Vendi Bayesian optimization, VBO) as an efficient way to find diverse MOF designs that optimize chemistry and structure-dependent properties. An advantage of this diversity-driven approach is preventing committing to a single MOF design (or similar ones) that for one reason or another may not be experimentally realized and/or tested. With ten simulated campaigns done for each case, we first statistically demonstrated that our VBO consistently does equal or better than random search to find high-performing designs within a 1,000-MOF subset for i) ammonia storage, ii) ammonia removal from membrane plasma reactors, and iii) ammonia capture from air. Then, with one campaign dedicated to ammonia storage on a “hybrid” 10,000-MOF database, we identify twelve extant and eight hypothesized designs that optimize ammonia working capacity ∆N_NH3 between 300 K and 400 K at 1 bar. The best MOF designs are predicted to achieve ∆N_NH3 values between 23.6 and 29.3 mmol/gm, potentially surpassing those that MOFs previously experimentally tested for NH₃ adsorption would have at the proposed operation conditions. The best hypothesized MOFs were predicted to be likely synthesizable (via free energy calculations), whereas the best MOF designs in general are predicted to be thermally stable (via machine learning prediction) and to require only ca. 10% of the energy content in NH₃ to release the stored molecule from the MOF. Finally, the analysis of the top MOFs indicates a pore size of around 10 Å, a heat of adsorption around 33 kJ/mol, and the presence of Ca as MOF design rules that could help optimize ammonia working capacity at the proposed operation conditions.