(218a) Discovery of O2-Selective Metal-Organic Frameworks Via Bayesian Optimization

Metal-organic frameworks (MOFs) present opportunities to tune chemical selectivity towards O2 by changing metal species and organic linkers. However, this also presents a combinatorial number of possible MOF structures, making a high-throughput screening study impossible with expensive quantum chemistry methods. We have shown previously that a machine learning technique called Bayesian optimization substantially reduces the number of calculations needed to find the top candidate MOF, thus enabling the screening of databases with time-consuming calculations. [1]

Here, we apply this technique to discover new O2-selective air separation MOFs using accurate density functional theory (DFT) calculations. Starting with the CoRE MOF database [2] and the enthalpy of adsorption of O2 in various MOFs in the experimental literature [3], we allow Bayesian optimization to suggest MOFs in the database to evaluate with DFT. We derive a new acquisition function, denoted as mean-squared-error expected improvement (MSE-EI) that allows us to target a binding enthalpy of 45 kJ/mol. This corresponds to roughly half the energy consumption needed for a state-of-the-art pressure swing adsorption system for air separation. We present the MOF candidates found in this procedure for further experimental validation.

[1] Taw, E.; Neaton, J.B.; Adv. Theory and Sim. 2022, 5, 3, 2100515
[2] Chung, Y.G. et al.; J. Chem. Eng. Data 2019, 64, 12, 5985-5998
[3] Jaramillo, D.; ... Taw, E. et al; in preparation