Benchmarking of GGA functionals for modeling structures of nanoporous, rigid and flexible MOFs

I'd like to bring to community attention a recent paper by an international group of authors (Wroclaw University; Prof. Lucyna Firlej, Montpellier; Dr Michael Fischer, Bremen) who looked into the accuracy of structural description of metal-organic frameworks (MOFs) with different exchange-correlation functionals and different dispersion correction schemes. MOFs are attracting a lot of attention right now thanks to their promise in areas of catalysis, sensing, gas storage and separation, etc.; there is a growing number of MOF suppliers for industrial applications.

The paper in question appeared in J. Chem. Phys., see https://doi.org/10.1063/1.5030493. This work is based on CASTEP calculations, and the main comparison is between PBE and PBEsol functionals, with either Grimme or Tkatchenko-Scheffler corrections. Earlier studies have shown that the PBEsol+TS scheme was the most accurate one for inorganic zeolite structures; this latest paper demonstrates that PBE+Grimme gives the best results for MOFs, be they rigid or flexible. This could be related to the different nature of bonding in organic systems where PBE is generally recognized as an accurate DFT functional. PBEsol has been designed with dense solids in mind, so it's not completely unexpected that it performs well but still comes behind PBE for very open nanoporous organic structures.

The conclusions are based on the analysis of interatomic bond lengths and angles and of lattice parameters for ten different MOFs. PBE-Grimme scheme impressively has RMSE error of about 0.7% for lattice parameters for rigid MOFs; flexible ones are harder to describe, 3% RMSE is a bit high but still better than any other combination of functionals/corrections. Overall the difference between different dispersion-corrected results is not large, the main and not surprising conclusion is that modeling of MOFs definitely requires a DFT-D approach.

One word of caution - these conclusions are based exclusively on the analysis of observed and simulation structures. There is not enough reliable data on energetics of such systems, so one cannot compare any properties based on e.g. lattice energy. Another missing piece of information is mechanical properties such as elastic constants - these are rather hard to measure for soft low symmetry structures, and expensive to calculate.

Finally, when the many-body dispersion correction scheme (Ambrosetti et al., J. Chem Phys., 140, 18A508 (2014)) becomes available in CASTEP, it will be most interesting to see whether it further improves agreement with experiment, at least for the structural parameters. Perhaps someone on this forum would be able to test this? Right now, however, a relatively simple PBE+Grimme DFT-D combination becomes a recommended setting for MOFs studies.

Victor Milman, BIOVIA Science Council