Ab initio predictions of adsorption in flexible metal-organic frameworks for water harvesting applications.

Recently, metal-organic frameworks like MOF-303 and MOF-LA2-1 have demonstrated exceptional performance for water harvesting applications. To enable a reticular design of such materials, an accurate prediction of the adsorption properties with chemical accuracy and fully accounting for the flexibility is crucial. The computational prediction of water adsorption properties of metal-organic frameworks (MOFs) has become standard practice. However their predictive power to design new materials is hindered by the limited accuracy of the used interatomic potential and the limitations on how to account for the framework flexibility. In this work, we showcase a methodology to obtain chemically accurate adsorption isotherms that fully account for the framework flexibility. The method is founded on very accurate and efficiently trained machine learning potentials and transition matrix Monte Carlo simulations to account for framework flexibility. By first benchmarking the reference electronic level of theory used for the training, quantitatively accurate adsorption isotherms are obtained for MOF-303, a highly topical MOF being investigated for its potential use in water harvesting applications. We show that both an accurate level of theory and a proper inclusion of local and global framework flexibility is vital in the prediction of the adsorption properties of MOF-303. The broader applicability of our methodology is demonstrated through the study of related linker-exchanged materials, MOF-333 and MOF-LA2-1. Analyses of the density profiles of water adsorbed in these MOFs yields deeper insight into the origins and differences of the observed isotherms. An optimal water harvester should have initial seeding sites with intermediate adsorption strength, to prevent detrimental low-pressure water uptake. To increase the working capacity, linker extension strategies can be used while maintaining the initial seeding sites, as was done in the MOF-LA2-1. The proposed methodology is applicable to other guest molecules and MOFs, paving the way to future rational design of MOFs with specific adsorption properties for the application at hand.