Large-Scale Free Energy Calculations on a Computational Metal–Organic Frameworks Database: Toward Synthetic Likelihood Predictions

Metal–organic frameworks (MOFs) have captivated the research community because of their modular crystal structure that can be tailored to suit diverse applications. However, identifying ideal MOFs for an application of choice is difficult because of the millions (or more) of possible MOFs one could consider. Although computational screening of MOF databases has provided a fast way to evaluate MOF performance, experimental validation of the predicted “exceptional” MOFs is uncommon because of the uncertainty of the synthetic likelihood of computationally constructed MOFs, hence hindering material discovery. Aiming to leverage the perspective provided by large data sets, we calculated the free energies of each MOF in a topologically diverse database of 8500 frameworks and evaluated to what extent the descriptors of MOF thermodynamic stability “discriminate” previously synthesized MOFs. Upon defining a relative free energy, ΔLMFFL, that corrects for some force field artifacts specific to MOF nodes, we found that previously synthesized MOFs in our database clustered in a region below ΔLMFFL = 4.4 kJ/mol per atom. This suggests that a MOF below this ΔLMFFL threshold may have a higher probability of being synthesized, although other factors may ultimately impair synthetic accessibility. For instance, when isomorphism occurs, multiple isomorphs may reside under the ΔLMFFL threshold and relative stability among isomorphs comes into play. From 32 isomorphic MOF series we examined in detail, we found the synthesized isomorph was the one with the lowest free energy in 80% of cases, and in 20% of cases to be within 1 kJ/mol of the latter. These findings indicate that for a MOF to be synthetically accessible a “low” crystal free energy is necessary, albeit in some cases it may not be sufficient because of the role of other factors not considered here (e.g., solvents, modulators, and kinetics).