Benchmarking quantum chemical methods with X-ray structures via structure-specific restraints.

There is a need for fast, efficient and accurate solid-state structure optimization for imprecise crystal structures (`augmentation') for subsequent property prediction in the pharmaceutical industry. Crystal structures from single-crystal X-ray, 3D electron or powder diffraction are widely available but require augmentation to the same quality level for comparative studies. Properties can be best calculated when the level of theory is alike and the accuracy, as well as the precision, are high. Moreover, the size of molecules and the complexity of structures encountered in pharmaceutical research are increasing. Efficient procedures are thus required that can also treat structures with disorder and several molecules in the asymmetric unit of the unit cell. Hence, we investigated whether `molecule-in-cluster' (MIC) computations [Dittrich et al. (2020). CrystEngComm 22, 7420-7431] can reach the accuracy of full-periodic (FP) computations. Selected quantum mechanical methods are assessed. The evaluation criterion is how well the structures of 22 very low temperature high-quality structures are reproduced. Computational efficiency is also considered. A novel approach to evaluating the accuracy of quantum mechanical predictions is enforcing computed structure-specific restraints in crystallographic least-squares refinements. To complement this approach, root mean square Cartesian displacements of computed and experimental structures were also compared. Analysis shows that (a) MIC DFT-D computations in a quantum mechanics/molecular mechanics (QM:MM) framework provide improved restraints and coordinates over earlier MIC GFN2-xTB computations, (b) increasing QM basis-set size in MIC QM:MM does not systematically improve computations, and (c) the choice of DFT functional is less important than the choice of the basis set. Overall, MIC computations are an accurate and computationally efficient tool for solid-state structure optimization that can match FP computations to augment experimental structures.