(582j) Computational Screening to Identify Metal-Organic Frameworks for Water Harvesting

Freshwater scarcity is a crucial global threat. An alternative to water purification is atmospheric water harvesting, which can be applied in any geographical location. Metal–organic frameworks (MOFs) are a diverse class of porous materials, and some MOFs have shown promise for capturing water molecules from the atmosphere even at low relative humidity (RH ~10%). Only a few MOFs have been tested for this purpose, and they have not met all of the required properties, such as high water uptake at low RH, fast kinetics of uptake and release, low water adsorption energy, and adequate stability. Synthesizing thousands of MOFs and testing them to identify the best MOFs for water harvesting is unfeasible. In this study, we applied computational screening to accelerate this process. To that end, we generated 5000 MOFs on the computer using a diverse selection of metal nodes, organic linkers, and topologies. Important criteria such as MOF stability, hydrophilicity, synthetic feasibility, and relative cost have been considered in selecting building blocks. All structures were generated using the ToBaCCo 3.0 software and then minimized using the UFF4MOF force field. Many of the most stable MOFs in the literature include nodes based on Zr₆ clusters, and it has been suggested in prior work that standard force fields, such as UFF, do not perform adequately for interactions of water molecules with the Zr nodes. To test this, DFT calculations at the M06-L-D3/def2-TZVP level of theory were used to calculate the binding energy of water on representative Zr₆ and M₃(µ₃-O)(COO)₆ (M=Cr, Fe, and Al) nodes. By changing the Lennard-Jones parameters of UFF for µ₃-O, µ₃-OH, and terminal -OH groups to TraPPE, and terminal -H₂O groups to the TIP4P water model, the binding energy was closer to the DFT calculation results versus when using the default UFF parameters for the framework functional groups. This modification was thus considered for all simulations. The partial charges for the MOF atoms were calculated using a machine learning model, PACMOF, developed by our group previously. We applied the Widom insertion method to calculate the Henry’s constant (K_H) and heat of adsorption at infinite dilution (Q_st,0) at 298 K for all 5000 MOFs. Water uptakes at 0.05 and 0.1 relative pressures were calculated for all MOFs by conducting grand canonical Monte Carlo (GCMC) simulations. Around 100 MOFs with large pore size, high K_H, high water uptake, and low Q_st,0 were selected for comprehensive molecular simulations. For the selected MOFs, more accurate partial charges were calculated from DFT, and GCMC simulations were conducted at additional conditions. We also calculated water diffusion coefficients by carrying out molecular dynamics (MD) simulations. We will discuss the top-performing materials as well as structure/property insights from the simulations.