A benchmark study of force fields implemented in CSD software

We have performed a benchmarking study to assess the performance of the CLP, UNI, CSD-OPCS16, DreidingII and Momany force fields available in the CSD Software. Our evaluation focused on the reproduction of experimental sublimation enthalpies for a large dataset of 664 crystal structures from the Cambridge Structural Database (CSD) and on the analysis of their crystal structures after optimisation with a new force field based optimiser. An analysis of our results for subsets of different chemical compound classes allowed for further insight on the performance of each force field. We have also assessed the ability of the presented force fields to reproduce known relative stabilities of polymorph families by comparison with DFT data. Our results show that, while all force field methods are able to reproduce sublimation enthalpies within the expected accuracies, overall errors are smaller for the UNI (11%), CSD-OPCS16 (13%) and CLP (13%) force fields. DreidingII and Momany have overall larger errors (22% and 26%, respectively), which can be notably reduced with the aid of structural optimisations. When looking at different classes of compounds, we found that no one force field would perform better than the other, and that accuracy is strictly dependent on the chemistry of the system of interest. The limitations of general purpose force fields in reproducing the small energy differences typical of polymorphs are also discussed. In particular, we found these methods to be rarely suitable for this task, and only reliable when the energy differences between polymorphs are larger than the uncertainties associated with force field methods. We have outlined our key findings as a source of guidance to aid the selection of force fields for energetic studies, and have suggested potential applications of these methods for further exploration.

The online version contains supplementary material available at 10.1007/s11224-025-02715-z.