Crystal Structure Prediction
Assessment of the exchange-hole
dispersion correction for the energy ranking stage of the seventh prediction blind testaDepartment of Chemistry, Dalhousie University, 6243 Alumni Crescent, Halifax, Nova Scotia, B3H 4R2, Canada, and bDepartamento de Química Física y Analítica and MALTA-Consolider Team, Facultad de Química, Universidad de Oviedo, 33006 Oviedo, Spain
*Correspondence e-mail: erin.johnson@dal.ca
This article is part of a collection of articles covering the seventh
prediction blind test.The seventh blind test of
prediction (CSP) methods substantially increased the level of complexity of the target compounds relative to the previous tests organized by the Cambridge Crystallographic Data Centre. In this work, the performance of density-functional methods is assessed using numerical atomic orbitals and the exchange-hole dispersion correction (XDM) for the energy-ranking phase of the seventh blind test. Overall, excellent performance was seen for the two rigid molecules (XXVII, XXVIII) and for the organic salt (XXXIII). However, for the agrochemical (XXXI) and pharmaceutical (XXXII) targets, the experimental polymorphs were ranked fairly high in energy amongst the provided candidate structures and inclusion of thermal free-energy corrections from the lattice vibrations was found to be essential for compound XXXI. Based on these results, it is proposed that the importance of vibrational free-energy corrections increases with the number of rotatable bonds.Keywords: crystal structure prediction; density-functional theory; London dispersion; free-energy correction.
1. Introduction
First-principles molecular et al., 2000; Motherwell et al., 2002; Day et al., 2005, 2009; Bardwell et al., 2011; Reilly et al., 2016), with the seventh held between 2020 and 2022 (Hunnisett et al., 2024a; Hunnisett et al., 2024b).
prediction (CSP) is a grand challenge in the field of computational chemistry and is fruitful ground for ongoing method development. To assess the performance of various methods for both the structure generation and energy ranking stages of CSP, the Cambridge Crystallographic Data Centre (CCDC) regularly holds blind test competitions. Solved but unpublished crystal structures for a small set of compounds are held in reserve by the CCDC and participants have a set period of time to submit predicted crystal structures given only chemical diagrams of the compounds. There have been six previous blind tests to date (LommerseIn the fourth (Day et al., 2009), fifth (Bardwell et al., 2011), and sixth (Reilly et al., 2016) blind tests, dispersion-corrected density-functional theory (DFT) was shown to provide large improvements in energy ranking of candidate crystal structures compared to classical force-field methods. The Neumann group was the first to use DFT in the blind tests with their own empirical dispersion correction (Neumann & Perrin, 2005; Neumann et al., 2008; Kendrick et al., 2011) and applied their methodology to the earlier blind tests retroactively (Asmadi et al., 2009). The more popular D3 (Grimme et al., 2010, 2011) and MBD (Tkatchenko et al., 2012; Ambrosetti et al., 2014) dispersion corrections have been used by other participants in submissions to the subsequent blind tests and also showed good performance (Brandenburg & Grimme, 2016; Reilly et al., 2016; Hunnisett et al., 2024a).
Dispersion corrections are routinely paired with generalized gradient approximation (GGA) functionals when modelling molecular crystals. However, GGAs suffer from delocalization error (Cohen et al., 2008b; Kim et al., 2013; Bryenton et al., 2023), which can lead to overstabilization of organic salts, halogen bonds, cooperative hydrogen-bonding networks, and extended π conjugation. Delocalization error can be reduced through subsequent single-point energy evaluations using hybrid functionals, for example with PBE0-MBD, as has been used in the sixth blind test and several other CSP studies (Marom et al., 2013; Shtukenberg et al., 2017; Hoja & Tkatchenko, 2018; Hoja et al., 2019; Mortazavi et al., 2019). Quite recently, we assessed the performance of our exchange-hole (XDM) dispersion correction (Johnson, 2017; Price et al., 2023b) in combination with hybrid functionals for the compounds appearing in all six previous blind tests (Price et al., 2023a). While delocalization error adversely affects the energy ranking provided with GGA functionals for organic salts in the fifth and sixth CSP blind tests (Whittleton et al., 2017a), and leads to errors in the intramolecular conformational energies of many conjugated molecules (Whittleton et al., 2017b; Greenwell & Beran, 2020; Greenwell et al., 2020; Beran et al., 2022), XDM-corrected hybrid functionals were shown to yield improved performance.
One of the key outcomes from discussion surrounding the sixth blind test was the idea of surpassing `zeroth-order' CSP (Price, 2018), using only the electronic energies, by inclusion of thermal free-energy corrections (Hoja et al., 2017). Such corrections have previously been used to convert experimental enthalpies to lattice energies in development of the C21 and X23 benchmarks of small-molecule crystals (Otero-de-la-Roza & Johnson, 2012; Reilly & Tkatchenko, 2013; Dolgonos et al., 2019), as well as for studies of aspirin (Reilly & Tkatchenko, 2014). However, the importance of these corrections for CSP has been a matter of some debate in the literature, with (Nyman & Day, 2015) showing they are generally small – less than 2 kJ mol−1 in 94% of cases taken from a set of 508 polymorphic compounds – using a distributed multipole force field. For a set of 17 polymorph pairs, DFT calculations confirmed these corrections to be ∼ 2 kJ mol−1 or less (Weatherby et al., 2022), although this study was limited to small, rigid molecules. DFT thermal free-energy corrections were also found to be of comparable magnitude for polymorphs of coumarin (Shtukenberg et al., 2017) and rotigotine (Mortazavi et al., 2019). Conversely, analogous calculations on the compounds from the sixth blind test occasionally showed significantly larger thermal free-energy corrections, up to 8.4 kJ mol−1 for the highly flexible compound XXIII (2-{[4-(3,4-dichlorophenethyl)phenyl]amino}benzoic acid; Hoja & Tkatchenko, 2018; Hoja et al., 2019).
The seventh CSP blind test was separated into two phases, with the first focusing on structure generation and the second on energy ranking. In this second phase, sets of candidate crystal structures were provided by the CCDC for the five compounds shown in Fig. 1. These sets comprised 100 structures each for compounds XXVII and XXXI, and 500 structures each for compounds XXVIII, XXXII, and XXXIII. Within each set, one or more seed structure(s) (see Table 1) were provided that corresponded to the experimental crystal structure(s) after geometry optimization with a CSD-tailored force field (Hunnisett et al., 2024a). For compound XXVII, two equivalent instances (see below) of the low-temperature experimental form were included as seed structures (rotation of the isopropyl groups leads to a disordered form at high temperature). For the copper compound, XXVIII, only one polymorph was isolated experimentally and included as a seed structure. For the agrochemical compound XXXI, the provided set contained four experimental seed structures. One of these seeds corresponds to form B, which was found to be the most stable polymorph experimentally at low temperatures. At high temperature, a disordered form A became more stable; both the major and minor components of this form were included as seed structures. The final experimental seed structure for XXXI corresponds to the desolvate form C. For the large, flexible, drug-like compound XXXII, two experimental seed structures were provided, corresponding to the more stable form B and the major component of the less stable, disordered form A. Lastly, for the organic salt, XXXIII, two experimental seed structures were provided, with form B again more stable than form A, which was found to be a disappearing polymorph.
|
The present work reports on the performance of several XDM-corrected density functionals for the energy ranking phase of the seventh CSP blind test. Within this phase, our group was one of ten that employed dispersion-corrected DFT, with four of these using hybrid functionals; four groups also included thermal free-energy corrections (Hunnisett et al., 2024a). Due to time constraints, we were not able to include free-energy corrections in our blind test submission, but do include them here for selected low-energy structures of compounds XXXI and XXXII. In this article, we will focus only on presentation of our own group's results. Interested readers are directed to Hunnisett et al. (2024a) for a comparison of the results across groups. Our methodology was able to rank the most stable experimentally isolated polymorph as the lowest in energy for two of the five compounds considered (XXVIII and XXXIII), and second lowest for one additional compound (XXVII). However, for the other two compounds (XXXI and XXXII) the isolated polymorphs were ranked substantially higher energetically. For compound XXXI, this result is unsurprising for form C since it possesses large crystal voids, while form B is correctly predicted to be the most stable polymorph upon inclusion of thermal free-energy corrections arising from the lattice vibrations. Including thermal free-energy corrections also yields an improved ranking for compound XXXII, and we conjecture that thermal and kinetic effects become increasingly important as the number of rotatable bonds increases.
2. Computational methods
2.1. comparison
A key concern in assessing CSP methods is how to quantify whether a candidate structure is the same or different when compared to the experimental polymorph. By far the most common method for ), implemented in the CCDC's Mercury program (Macrae et al., 2020). COMPACK matches molecules within a given cluster size (commonly 20) by generating an optimal overlay that minimizes the root-mean-square deviation (RMSD) in the atomic positions. However, we have recently illustrated a problem with COMPACK that precludes its routine application to molecules with several highly branched functional groups (Mayo et al., 2022), such as compound XXVII in the seventh blind test. This problem likely arises from the application of a graph-matching algorithm (Ullmann, 1976) to identify matching molecules within the structures and align them.
comparison is the COMPACK algorithm (Motherwell & Chisholm, 2005COMPACK can also be quite sensitive to choices of cutoff values (Mayo & Johnson, 2021; Mayo et al., 2022) and cluster sizes (Mayo et al., 2022) that make its common use as a black-box method concerning. As a result, we have argued that distance-based methods like COMPACK should be used in conjunction with comparison methods based on powder X-ray diffraction (PXRD) to provide more certainty in the comparison results than can be achieved with either type of method alone.
PXRD-based methods quantify structure similarity using the overlap of the powder diffractograms of a pair of crystal structures. Among other ways, the overlaps can be computed using de Gelder's cross-correlation function (de Gelder et al., 2001). The dissimilarity metric is then the powder pattern difference (PWDF), defined as one minus the overlap, such that exactly matching structures have a PWDF score of zero. However, PXRD-based comparison methods are only effective for pairs of crystal structures obtained under the same experimental conditions, or optimized with the same level of theory for in silico generated structures. This is because the PXRD peak locations are highly sensitive to the unit-cell dimensions, which change depending on temperature, pressure, and the particular choice of computational method. To overcome this limitation, we developed the variable-cell powder difference (VC-PWDF) method (Mayo et al., 2022). VC-PWDF works by exploring a range of possible unit-cell definitions and distorting the cell vectors before evaluating the powder pattern difference. Overall, VC-PWDF was shown to agree with COMPACK in over 97% of almost 45 000 individual crystal-structure comparisons in the CSD (Mayo et al., 2022), although it can provide incorrect classification of and conformational phases because of the similarity in their unit cells.
In this work, we employ a mix of RMSD(20), PWDF, and VC-PWDF methods for structure comparison. The RMSD(20) values were obtained using the Python API interface to Mercury (Macrae et al., 2020), while PWDF and VC-PWDF scores were obtained using the critic2 program (Otero-de-la-Roza et al., 2014).
2.2. DFT calculations
The geometries of all provided candidate crystal structures were fully optimized using an in-house modified version of the FHI-aims program (version 210513; Blum et al., 2009). Geometry optimizations were performed using the B86bPBE functional (Becke, 1986; Perdew et al., 1996) with the XDM dispersion correction (Price et al., 2023b), with the `lightdenser' basis set and integration grid, and a relaxation convergence threshold of 0.005 eV Å−1. Subsequent single-point energy calculations were then performed on all optimized structures using either the hybrid B86bPBE-25X or B86bPBE-50X functionals with the same lightdenser basis set and grid settings, as well as with the B86bPBE functional with the tight basis settings, all with XDM dispersion. Price et al. (2023b) give the relevant damping parameters. Based on previous tests (Blum et al., 2009), and excellent agreement with planewave results (Lejaeghere et al., 2016; Price et al., 2023b), there should be negligible basis-set superposition error. For compound XXVIII, all calculations were run for a ferromagnetic configuration with one unpaired electron per copper atom. The single unpaired d electron is evident since CuII will have an of [Ar]3d9, and a ferromagnetic configuration is required in order to have a proper spin eigenstate that can be well described by Kohn–Sham DFT (Cohen et al., 2008a).
We note that there are differences in the relative energies and computational timings between this work and our blind test submission, due to the time constraints for the blind test and the fact that we were still making changes to the code when its second phase began. In this work, all geometry relaxations started from the crystallographic information files (CIF) provided by the CCDC with the finalized XDM implementation reported by Price et al. (2023b) and relaxed with a convergence threshold of 0.005 eV Å−1. However, in our blind test submission, we had relaxed all structures with a much looser threshold of 0.025 eV Å−1, and only completed relaxations with the 0.005 eV Å−1 threshold for candidates lying within 1.5 kcal mol−1 (6.28 kJ mol−1) of the minimum obtained with the looser convergence criterion. For the most part, the tighter convergence threshold did not substantially alter the relative energies. However, it did result in large geometry and energy changes for a handful of structures. One key example is structures 145 and 207 for compound XXVIII. Here, the initial structures differed from the experimental polymorph by a translation of some of the molecules in the (see Fig. 2) and they were (effectively degenerate) high-energy structures, lying > 30 kJ mol−1 above the minimum in our blind test submission. However, after full optimization with the tighter convergence threshold, the two structures became duplicate matches to the experimental form. Presumably, with the looser criteria, the optimization was deemed converged when the structure was occupying some nearly flat shoulder of a high-dimensional surface. With tighter criteria, the optimization continued and proceeded to follow a path with small negative gradients until it could move away from the flat shoulder and find a much more stable local minimum. This should serve as a warning to ensure sufficiently tight convergence criteria during geometry optimizations for CSP, despite the appeal of using looser criteria to reduce computational cost.
2.3. Harmonic vibrational calculations
To apply thermal free-energy corrections in the harmonic approximation, vibrational frequency (phonon) calculations were performed for the 16 lowest-energy candidates of compound XXXI and the 26 lowest-energy candidates of compound XXXII, according to the GGA/light ranking. These numbers correspond to the ranks of the least stable of the experimental seed structures with the GGA/light level of theory (excluding the high-energy solvate for compound XXXI). The phonon calculations used the finite-difference approach as implemented in the phonopy package (Togo & Tanaka, 2015). The energies for all supercells were again evaluated with B86bPBE-XDM, lightdenser setting, using FHI-aims (Blum et al., 2009). For compound XXXI, the thermal free-energy corrections were converged to within 0.5 kJ mol−1 per molecule with respect to size. For compound XXXII, the thermal free-energy corrections were only converged to within 1.0 kJ mol−1 per molecule due to the large unit-cell sizes, which made calculations on larger supercells intractable. Lists of the candidate structures carried forward to phonon calculations, and the corresponding supercells used, are given in the supporting information.
3. Results and discussion
3.1. Structure changes
Because the candidate crystal structures provided by the CCDC were all optimized with a CSD-tailored force field, they may not correspond to stable minima with DFT. To assess how much structural change occurred during the DFT geometry optimizations, we evaluated the VC-PWDF score between the initial and final geometries for each candidate structure of each compound. The results are summarized by the box plot in Fig. 3, where the boxes span the interquartile range of the VC-PWDF values. The whiskers span 95% of the data, while outliers are shown as individual points. For context, a VC-PWDF score of < 0.03 is usually a good indicator of significant structural similarity for rigid molecules and < 0.05 for more flexible molecules (Mayo & Johnson, 2021; Mayo et al., 2022).
From Fig. 3, compound XXXIII shows the largest structural changes during optimization, followed by compound XXVIII, despite both systems comprising fairly rigid molecules. It is reasonable that the force field would be less accurate for the organic salt (XXXIII) due to the importance of electrostatic interactions. Similarly, the organometallic copper complex (XXVIII) may be problematic due to difficulties in parameterization for copper and the unpaired electron in the complex. Instances of large geometric changes with VC-PWDF > 0.1 also occurred for some candidate structures of compounds XXXI and XXXII, although the ranges spanned by 95% of the points are much narrower compared to those for XXVIII and XXXIII.
As it is fairly common for force fields to predict more energy minima than DFT (Price, 2013; Neumann & van de Streek, 2018; Francia et al., 2020; Butler & Day, 2023), we also assess how many of the candidate structures were duplicates (i.e. corresponding to the same polymorph) before and after geometry optimization. This provides insight into whether different force-field local minima correspond to a single DFT minimum. Since we are comparing structures that were all optimized with the same method (i.e. either the CSD-tailored force field or DFT), we do not need to consider variable cells and we can use the PWDF scores as our similarity metric. Here, two structures are deemed equivalent if their PWDF value is < 0.01.
The results in Table 2 show the number of duplicate structures before and after optimization. For the CCDC provided structures, there were only two sets of duplicates each for XXVII (structure IDs 28/61 and 63/77) and XXXII (IDs 58/66 and 169/488). These may have initially been deemed different by COMPACK due to problems with the graph-matching algorithm for highly branched, pseudosymmetric molecules (Mayo et al., 2022), such as XXVII, or due to the choice of distance and angle tolerances. After optimization, at least two pairs of duplicate structures are seen for all the compounds, with many duplicate structures occurring for XXVIII and XXXIII in particular. This finding is consistent with the larger structural changes seen for these two compounds in Fig. 3. An example was already discussed in Section 2.2 for compound XXVIII, where two structures (IDs 145 and 207) optimize to become equivalent to the seed structure for the experimental form (ID 144, see Fig. 2).
|
3.2. BT7-predicted GGA landscapes
We consider now the computed crystal energy landscapes obtained from the geometry relaxations with B86bPBE-XDM and the light NAO basis set; these results are shown in Fig. 4. For compound XXVII, two identical seed structures (IDs 28 and 61), chosen to correspond to the low-temperature experimental form, were included in the set of candidates provided by the CCDC. This seed structure was ranked second lowest on the computed crystal-energy landscape in Fig. 4(a). As it lies < 1.5 kJ mol−1 above the global minimum structure (ID 58), which is well within the accepted energy window for the stability of experimentally observable polymorphs (Nyman & Day, 2015), we consider this to be a successful prediction. Additionally, this energy difference is small enough that thermal free-energy corrections may reverse the ranking (Nyman & Day, 2016, 2015; Weatherby et al., 2022). Rotations of the isopropyl groups, which become disordered at room temperature, can also significantly affect the energy ranking (Hunnisett et al., 2024a).
For the copper compound, XXVIII, only a single polymorph was identified experimentally. The crystal-energy landscape in Fig. 4(b) shows that the experimental seed structure was ranked as the most stable candidate, again indicating a successful prediction. As noted above, two additional candidates became equivalent to the seed structure after geometry optimization.
For compound XXXI, Fig. 4(c) shows that the DFT calculations predict form C to lie quite high in energy, as expected due to the large voids in this that formed upon solvent evaporation. While this form lies 10.7 kJ mol−1 above the global minimum, significantly outside the usual `polymorph window' (< 7.2 kJ mol−1 above the global minimum in 95% of cases) (Nyman & Day, 2015), it has been shown that similar high-energy, low-density structures can be isolated from solvates (Yang & Day, 2021). Turning to the more stable experimental forms, our calculations predict the seed structures for form B and the major/minor components of form A to all lie within 5 kJ mol−1 of the global minimum. However, the predicted stability ordering is incorrect, with form B ranked as the least stable, lying 4.7 kJ mol−1 above the global minimum and 3.1 kJ mol−1 above the major component of form A.
As shown in Fig. 4(d), our calculations were quite unsuccessful in identifying the experimental forms of compound XXXII as being low in energy amongst the candidates. Here, both forms were found to lie 6–7 kJ mol−1 in energy above the global minimum. The more stable form B was ranked 22nd in energy, while form A was ranked 26th, recovering the correct experimental ordering, but with a predicted energy difference of only 0.4 kJ mol−1 separating the two forms.
Finally, for compound XXXIII, Fig. 4(e) shows that form B is correctly ranked as the most stable of the candidate structures, constituting another successful prediction. Form A is also found to be a fairly stable structure, ranked sixth in energy and lying < 5 kJ mol−1 above form A, which is still within the accepted polymorph window (Nyman & Day, 2015).
3.3. Dependence on functional and basis set
In previous benchmarking of XDM-corrected density functionals for the compounds comprising the first six blind tests, we found that hybrid functionals outperformed GGAs in some cases (Price et al., 2023a), specifically those with significant delocalization error (Bryenton et al., 2023). In particular, delocalization error was found to affect the ranking of the organic salts (compounds XIX and XXIV) (Price et al., 2023a; Whittleton et al., 2017a), as well as for compound X where the intramolecular conformational energy has a substantial contribution to the ranking (Price et al., 2023a; Greenwell & Beran, 2020; Whittleton et al., 2017b). Similarly, significant errors in GGA energy rankings have been found for other flexible molecules due to errors in the conformational energy (Greenwell & Beran, 2020; Greenwell et al., 2020; Beran et al., 2022). As a result, we performed single-point energy calculations with the B86bPBE-25X-XDM and B86bPBE-50X-XDM functionals, both with the light basis set, which was expected to lead to more accurate ranking relative to the B86bPBE-XDM results alone. We also performed single-point energy calculations with B86bPBE-XDM and the tight basis set to identify if the errors seen for compounds XXXI and XXXII were a result of basis-set incompleteness. This allowed us to approximate hybrid functional results with the tight basis using an additive scheme (Hoja & Tkatchenko, 2018; Hoja et al., 2019; Price et al., 2023b):
The identities of the minimum-energy structures obtained with each combination of functional and basis set are given in the supporting information. A comparison of the overall rankings and relative energies of the experimental seed structures, relative to the respective global minimums, are shown in Table 3. Note that the rankings were determined by taking some pairs of structures as equivalent: the experimental matches, structures 28 and 61, for compound XXVII (POWDIFF = 0.0004); the experimental matches, structures 144, 145 and 207, for compound XXVIII (POWDIFF ≤ 0.0005); and the DFT energy minima, structures 17 and 59, for compound XXXI (POWDIFF = 0.0039). Full crystal-energy landscapes are provided in the supporting information.
|
For the blind test submission, we selected results from the B86bPBE-25X functional for compounds XXVII and XXVIII, and results from B86bPBE-50X for the remaining three compounds, as we expected them to potentially be more susceptible to delocalization error due to being able to adopt a range of stable conformations (XXXI and XXXII) or being an organic salt (XXXIII). However, the results in Table 3 show that these were not the optimal choices. Overall, all three functionals give very similar ranking for compounds XXVII, XXVIII, and XXXIII. It would seem that the crystal energy landscapes of these three compounds are relatively easy to model accurately with dispersion-corrected DFT, since the ranking does not depend on the choice of functional in any consequential way. This consistence was expected for the first two compounds, as they are relatively rigid, non-polar molecules, and their crystals are therefore unlikely to have much inherent delocalization error.
In light of our previous results for the salt compounds XIX and XXIV, the lack of functional dependence was somewhat surprising for compound XXXIII. Therefore, we conclude that not all organic salts are equally affected by delocalization error. One likely explanation for the difference is that delocalization error frequently manifests in cases with extended conjugation (Greenwell & Beran, 2020; Greenwell et al., 2020; Beran et al., 2022; Whittleton et al., 2017b) and cooperative hydrogen bonding (Whittleton et al., 2017a), such as for COOH groups where the OH can act as a hydrogen-bond donor and the =O can act as a hydrogen-bond acceptor. As seen from their chemical structures, shown in Fig. 5, moieties in salts XIX and XXIV both exhibit extended conjugation, with all species being planar, and one ion in each possesses COOH groups that form cooperative hydrogen-bonds in the crystal (Whittleton et al., 2017a; Price et al., 2023a). Conversely, for XXXIII, the cation is not planar, the conjugation in the anion is broken by the SO2 group, and neither can participate in cooperative hydrogen bonding.
We now move to the two flexible molecules, compounds XXXI and XXXII. In sharp contrast to the other three members of this blind test, these compounds proved extremely challenging for DFT and also exhibit more functional dependence. From Table 3, the experimental forms of compounds XXXI and XXXII are ranked increasingly higher in energy (with respect to the global minimum) with increasing fractions of exact exchange. Increasing exact exchange does lower the energy of form B relative to the major component of form A for compound XXXI, but not enough to recover the correct ordering. However, for compound XXXII, form B was correctly predicted to be more stable than the major component of form A with the GGA functional and this ordering is reversed with both hybrids. Therefore, our choice of 50% exact exchange for the blind test submission was unfortunate here and this demonstrates how difficult it can be to know, a priori, whether delocalization error will affect a CSP landscape. Finally, we find that the tight basis set does not improve the rankings, widening the energy gap between forms A and B for compound XXXI and raising the experimental forms of compound XXXII to substantially higher energies. This allows us to rule out basis-set incompleteness for the very electronegative F atoms appearing in these two compounds as the source of the poor rankings.
3.4. Vibrational free-energy corrections
To summarize our results so far, the DFT energy rankings were quite successful for three of the blind test compounds (XXVII, XXVIII and XXXIII), but performed much less well for the landscapes of XXXI and XXXII. Notably, these are both fairly large, flexible molecules with several rotatable bonds. This is reminiscent of compound XX from the fifth CSP blind test, another large, flexible molecule for which our DFT energy rankings showed considerable errors (Price et al., 2023a). We previously conjectured that thermal free-energy corrections, neglected in zeroth-order CSP (Price, 2018), may be responsible for the poor energy ranking here, and this may also be the case for compounds XXXI and XXXII. To test this hypothesis, phonon calculations were performed for the 16 lowest-energy candidates for compound XXXI and the 26 lowest-energy candidates for compound XXXII at the B86bPBE-XDM/light level. These ranges of structures were selected to include all of the experimental seed structures, except for the high-energy solvate (see Table 3).
Fig. 6 shows the free-energy landscapes for compounds XXXI and XXXII, where the thermal free-energy corrections for a temperature of 298 K have been added to the B86bPBE-XDM/light electronic energies. Free-energy landscapes obtained using different base functionals and/or basis sets for the electronic energies are shown in the supporting information. Fig. 7 shows the change in ranking of the considered structures upon inclusion of the thermal free-energy corrections.
The results in Figs. 6(a) and 7(a) show that inclusion of thermal free-energy terms corrects the energy ranking for compound XXXI. Using the electronic energies alone, form B was found to lie 4.7 kJ mol−1 above the DFT minimum. However, using free energies, form B is successfully predicted to be the global minimum structure, while the major component of form A is ranked third, <0.5 kJ mol−1 above form B. Experimentally, form B is the more stable form up to a temperature of 48–55°C (Hunnisett et al., 2024b), after which the disordered form A is more stable. It should be noted, however, that this ranking only applies to the subset of structures for which free-energy calculations were performed. It is possible that some structures with even higher lattice energies would also be stabilized sufficiently by inclusion of free-energy corrections to affect these rankings.
Figs. 6(b) and 7(b) show that inclusion of thermal free-energy corrections also improves the ranking for compound XXXII. Here, form B is now ranked seventh with a ΔG of 4.6 kJ mol−1, while form A is ranked fifth with a ΔG of 5.9 kJ mol−1, in contrast to the rankings of 22nd and 26th in Table 3 obtained with electronic energies alone. However, the two experimental forms do remain fairly high in free energy above the global minimum. This finding suggests that (i) there are errors in our DFT treatment, (ii) anharmonicity and/or is important in determining the vibrational free-energy correction and the quasi-harmonic approximation is needed, (iii) kinetic and/or surface-energy effects prevent formation of the lower-energy candidates identified with DFT, or (iv) some combination of these factors (Price, 2013).
4. Conclusions
In this work, we applied XDM-corrected DFT methods to the energy ranking phase of the seventh CSP blind test. Large changes in some of the provided candidate structures were observed after DFT geometry optimization, particularly for compounds XXVIII and XXXIII. Additionally, many candidates collapsed to the same local minimum, giving duplicate structures after optimization. This points to the importance of developing improved force fields for the initial structure generation and geometry optimization steps of CSP. In terms of the DFT energy ranking itself, our methodology gave successful predictions of the most stable experimentally isolated polymorphs as being the lowest or second-lowest energy structures for compounds XXVII, XXVIII and XXXIII. However, the most stable experimental polymorphs were ranked relatively high in energy for compounds XXXI and XXXII.
We conclude that our DFT methodology provides good reliability for CSP studies of fairly rigid molecules, but there remain challenges for large and highly flexible pharmaceutical targets. We suggest that thermal free-energy corrections may be much more important for such structures than previously demonstrated for smaller, more rigid molecules (Nyman & Day, 2016; Nyman & Day, 2015; Weatherby et al., 2022). While the thermal free-energy corrections are typically ≤ 2 kJ mol−1, they amounted to as much as 4.7 and 5.9 kJ mol−1 for compounds XXXI and XXXII, respectively. Previous work found even larger thermal free-energy corrections, up to 8.4 kJ mol−1, for compound XXIII in the sixth blind test (Hoja & Tkatchenko, 2018; Hoja et al., 2019). Here, the most stable isolated polymorph for compound XXXI was successfully predicted as the global minimum with DFT upon inclusion of a vibrational free-energy correction. This correction improved the ranking for compound XXXII as well, although the experimental structures were still ranked sixth and 15th in energy. It is possible that corrections beyond the harmonic approximation are required, or that kinetic and/or surface energy effects prevent crystallization of the lower-energy candidates identified with DFT in this case.
Supporting information
Energy rankings. DOI: https://doi.org/10.1107/S2052520624002774/aw5086sup1.txt
Tables I, II, III, and 19 extra figures. DOI: https://doi.org/10.1107/S2052520624002774/aw5086sup2.pdf
Acknowledgements
ERJ thanks the Natural Sciences and Engineering Research Council of Canada (NSERC) for financial support of this research. RAM thanks the International Centre for Diffraction Data for a Ludo Frevel Crystallography Scholarship. All authors are grateful to the Atlantic Computing Excellence Network (ACENET) and Compute Canada for providing computational resources.
Funding information
AOR thanks the Spanish Ministerio de Ciencia e Innovación and the Agencia Estatal de Investigación, projects PGC2021-125518NB-I00 and RED2022-134388-T cofinanced by EU FEDER funds; the Principality of Asturias (FICYT), project AYUD/2021/51036 cofinanced by EU FEDER; and the Spanish MCIN/AEI/10.13039/501100011033 and European Union NextGenerationEU/PRTR for grant TED2021-129457B-I00.
References
Ambrosetti, A., Reilly, A. M., DiStasio, R. A. Jr & Tkatchenko, A. (2014). J. Chem. Phys. 140, 18A508. Web of Science CrossRef PubMed Google Scholar
Asmadi, A., Neumann, M. A., Kendrick, J., Girard, P., Perrin, M.-A. & Leusen, F. J. J. (2009). J. Phys. Chem. B, 113, 16303–16313. CrossRef PubMed CAS Google Scholar
Bardwell, D. A., Adjiman, C. S., Arnautova, Y. A., Bartashevich, E., Boerrigter, S. X. M., Braun, D. E., Cruz-Cabeza, A. J., Day, G. M., Della Valle, R. G., Desiraju, G. R., van Eijck, B. P., Facelli, J. C., Ferraro, M. B., Grillo, D., Habgood, M., Hofmann, D. W. M., Hofmann, F., Jose, K. V. J., Karamertzanis, P. G., Kazantsev, A. V., Kendrick, J., Kuleshova, L. N., Leusen, F. J. J., Maleev, A. V., Misquitta, A. J., Mohamed, S., Needs, R. J., Neumann, M. A., Nikylov, D., Orendt, A. M., Pal, R., Pantelides, C. C., Pickard, C. J., Price, L. S., Price, S. L., Scheraga, H. A., van de Streek, J., Thakur, T. S., Tiwari, S., Venuti, E. & Zhitkov, I. K. (2011). Acta Cryst. B67, 535–551. Web of Science CrossRef IUCr Journals Google Scholar
Becke, A. D. (1986). J. Chem. Phys. 85, 7184–7187. CrossRef CAS Web of Science Google Scholar
Beran, G. J. O., Sugden, I. J., Greenwell, C., Bowskill, D. H., Pantelides, C. C. & Adjiman, C. S. (2022). Chem. Sci. 13, 1288–1297. CrossRef CAS PubMed Google Scholar
Blum, V., Gehrke, R., Hanke, F., Havu, P., Havu, V., Ren, X., Reuter, K. & Scheffler, M. (2009). Comput. Phys. Commun. 180, 2175–2196. Web of Science CrossRef CAS Google Scholar
Brandenburg, J. G. & Grimme, S. (2016). Acta Cryst. B72, 502–513. Web of Science CrossRef IUCr Journals Google Scholar
Bryenton, K. R., Adeleke, A. A., Dale, S. G. & Johnson, E. R. (2023). WIREs Comput. Mol. Sci. 13, e1631. Google Scholar
Butler, P. W. V. & Day, G. M. (2023). Proc. Natl Acad. Sci. USA, 120, e2300516120. Web of Science CrossRef PubMed Google Scholar
Cohen, A. J., Mori-Sánchez, P. & Yang, W. (2008a). J. Chem. Phys. 129, 121104. CrossRef PubMed Google Scholar
Cohen, A. J., Mori-Sánchez, P. & Yang, W. (2008b). Science, 321, 792–794. CrossRef PubMed CAS Google Scholar
Day, G. M., Cooper, T. G., Cruz-Cabeza, A. J., Hejczyk, K. E., Ammon, H. L., Boerrigter, S. X. M., Tan, J. S., Della Valle, R. G., Venuti, E., Jose, J., Gadre, S. R., Desiraju, G. R., Thakur, T. S., van Eijck, B. P., Facelli, J. C., Bazterra, V. E., Ferraro, M. B., Hofmann, D. W. M., Neumann, M. A., Leusen, F. J. J., Kendrick, J., Price, S. L., Misquitta, A. J., Karamertzanis, P. G., Welch, G. W. A., Scheraga, H. A., Arnautova, Y. A., Schmidt, M. U., van de Streek, J., Wolf, A. K. & Schweizer, B. (2009). Acta Cryst. B65, 107–125. Web of Science CSD CrossRef IUCr Journals Google Scholar
Day, G. M., Motherwell, W. D. S., Ammon, H. L., Boerrigter, S. X. M., Della Valle, R. G., Venuti, E., Dzyabchenko, A., Dunitz, J. D., Schweizer, B., van Eijck, B. P., Erk, P., Facelli, J. C., Bazterra, V. E., Ferraro, M. B., Hofmann, D. W. M., Leusen, F. J. J., Liang, C., Pantelides, C. C., Karamertzanis, P. G., Price, S. L., Lewis, T. C., Nowell, H., Torrisi, A., Scheraga, H. A., Arnautova, Y. A., Schmidt, M. U. & Verwer, P. (2005). Acta Cryst. B61, 511–527. Web of Science CSD CrossRef CAS IUCr Journals Google Scholar
Dolgonos, G. A., Hoja, J. & Boese, A. D. (2019). Phys. Chem. Chem. Phys. 21, 24333–24344. Web of Science CrossRef CAS PubMed Google Scholar
Francia, N. F., Price, L. S., Nyman, J., Price, S. L. & Salvalaglio, M. (2020). Cryst. Growth Des. 20, 6847–6862. Web of Science CrossRef CAS Google Scholar
Gelder, R. de, Wehrens, R. & Hageman, J. A. (2001). J. Comput. Chem. 22, 273–289. Web of Science CrossRef Google Scholar
Greenwell, C. & Beran, G. J. O. (2020). Cryst. Growth Des. 20, 4875–4881. CrossRef CAS Google Scholar
Greenwell, C., McKinley, J. L., Zhang, P., Zeng, Q., Sun, G., Li, B., Wen, S. & Beran, G. J. O. (2020). Chem. Sci. 11, 2200–2214. Web of Science CrossRef CAS PubMed Google Scholar
Grimme, S., Antony, J., Ehrlich, S. & Krieg, H. (2010). J. Chem. Phys. 132, 154104. Web of Science CrossRef PubMed Google Scholar
Grimme, S., Ehrlich, S. & Goerigk, L. (2011). J. Comput. Chem. 32, 1456–1465. Web of Science CrossRef CAS PubMed Google Scholar
Hoja, J., Ko, H. Y., Neumann, M. A., Car, R., Distasio, R. A. & Tkatchenko, A. (2019). Sci. Adv. 5, eaau3338. CrossRef PubMed Google Scholar
Hoja, J., Reilly, A. M. & Tkatchenko, A. (2017). WIREs Comput. Mol. Sci. 7, e1294. Google Scholar
Hoja, J. & Tkatchenko, A. (2018). Faraday Discuss. 211, 253–274. CrossRef CAS PubMed Google Scholar
Hunnisett, L. M., Francia, N., Nyman, J., Abraham, N. S., Aitipamula, S., Alkhidir, T., Almehairbi, M., Anelli, A., Anstine, D. M., Anthony, J. E., Arnold, J. E., Bahrami, F., Bellucci, M. A., Beran, G. J. O., Bhardwaj, R. M., Bianco, R., Bis, J. A., Boese, A. D., Bramley, J., Braun, D. E., Butler, P. W. V., Cadden, J., Carino, S., Červinka, C., Chan, E. J., Chang, C., Clarke, S. M., Coles, S. J., Cook, C. J., Cooper, R. I., Darden, T., Day, G. M., Deng, W., Dietrich, H., DiPasquale, A., Dhokale, B., van Eijck, B. P., Elsegood, M. R. J., Firaha, D., Fu, W., Fukuzawa, K., Galanakis, N., Goto, H., Greenwell, C., Guo, R., Harter, J., Helfferich, J., Hoja, J., Hone, J., Hong, R., Hušák, M., Ikabata, Y., Isayev, O., Ishaque, O., Jain, V., Jin, Y., Jing, A., Johnson, E. R., Jones, I., Jose, K. V. J., Kabova, E. A., Keates, A., Kelly, P. F., Klimeš, J., Kostková, V., Li, H., Lin, X., List, A., Liu, C., Liu, Y. M., Liu, Z., Lončarić, I., Lubach, J. W., Ludík, J., Maryewski, A. A., Marom, N., Matsui, H., Mattei, A., Mayo, R. A., Melkumov, J. W., Mladineo, B., Mohamed, S., Momenzadeh Abardeh, Z., Muddana, H. S., Nakayama, N., Nayal, K. S., Neumann, M. A., Nikhar, R., Obata, S., O'Connor, D., Oganov, A. R., Okuwaki, K., Otero-de-la-Roza, A., Parkin, S., Parunov, A., Podeszwa, R., Price, A. J. A., Price, L. S., Price, S. L., Probert, M. R., Pulido, A., Ramteke, G. R., Rehman, A. U., Reutzel-Edens, S. M., Rogal, J., Ross, M. J., Rumson, A. F., Sadiq, G., Saeed, Z. M., Salimi, A., Sasikumar, K., Sekharan, S., Shankland, K., Shi, B., Shi, X., Shinohara, K., Skillman, A. G., Song, H., Strasser, N., van de Streek, J., Sugden, I. J., Sun, G., Szalewicz, K., Tan, L., Tang, K., Tarczynski, F., Taylor, C. R., Tkatchenko, A., Touš, P., Tuckerman, M. E., Unzueta, P. A., Utsumi, Y., Vogt-Maranto, L., Weatherston, J., Wilkinson, L. J., Willacy, R. D., Wojtas, L., Woollam, G. R., Yang, Y., Yang, Z., Yonemochi, E., Yue, X., Zeng, Q., Zhou, T., Zhou, Y., Zubatyuk, R. & Cole, J. C. (2024). Acta Cryst. B80, https://doi.org/10.1107/S2052520624008679. Google Scholar
Hunnisett, L. M., Nyman, J., Francia, N., Abraham, N. S., Adjiman, C. S., Aitipamula, S., Alkhidir, T., Almehairbi, M., Anelli, A., Anstine, D. M., Anthony, J. E., Arnold, J. E., Bahrami, F., Bellucci, M. A., Bhardwaj, R. M., Bier, I., Bis, J. A., Boese, A. D., Bowskill, D. H., Bramley, J., Brandenburg, J. G., Braun, D. E., Butler, P. W. V., Cadden, J., Carino, S., Chan, E. J., Chang, C., Cheng, B., Clarke, S. M., Coles, S. J., Cooper, R. I., Couch, R., Cuadrado, R., Darden, T., Day, G. M., Dietrich, H., Ding, Y., DiPasquale, A., Dhokale, B., van Eijck, B. P., Elsegood, M. R. J., Firaha, D., Fu, W., Fukuzawa, K., Glover, J., Goto, H., Greenwell, C., Guo, R., Harter, J., Helfferich, J., Hofmann, D. W. M., Hoja, J., Hone, J., Hong, R., Hutchison, G., Ikabata, Y., Isayev, O., Ishaque, O., Jain, V., Jin, Y., Jing, A., Johnson, E. R., Jones, I., Jose, K. V. J., Kabova, E. A., Keates, A., Kelly, P. F., Khakimov, D., Konstantinopoulos, S., Kuleshova, L. N., Li, H., Lin, X., List, A., Liu, C., Liu, Y. M., Liu, Z., Liu, Z.-P., Lubach, J. W., Marom, N., Maryewski, A. A., Matsui, H., Mattei, A., Mayo, R. A., Melkumov, J. W., Mohamed, S., Momenzadeh Abardeh, Z., Muddana, H. S., Nakayama, N., Nayal, K. S., Neumann, M. A., Nikhar, R., Obata, S., O'Connor, D., Oganov, A. R., Okuwaki, K., Otero-de-la-Roza, A., Pantelides, C. C., Parkin, S., Pickard, C. J., Pilia, L., Pivina, T., Podeszwa, R., Price, A. J. A., Price, L. S., Price, S. L., Probert, M. R., Pulido, A., Ramteke, G. R., Rehman, A. U., Reutzel-Edens, S. M., Rogal, J., Ross, M. J., Rumson, A. F., Sadiq, G., Saeed, Z. M., Salimi, A., Salvalaglio, M., Sanders de Almada, L., Sasikumar, K., Sekharan, S., Shang, C., Shankland, K., Shinohara, K., Shi, B., Shi, X., Skillman, A. G., Song, H., Strasser, N., van de Streek, J., Sugden, I. J., Sun, G., Szalewicz, K., Tan, B. I., Tan, L., Tarczynski, F., Taylor, C. R., Tkatchenko, A., Tom, R., Tuckerman, M. E., Utsumi, Y., Vogt-Maranto, L., Weatherston, J., Wilkinson, L. J., Willacy, R. D., Wojtas, L., Woollam, G. R., Yang, Z., Yonemochi, E., Yue, X., Zeng, Q., Zhang, Y., Zhou, T., Zhou, Y., Zubatyuk, R. & Cole, J. C. (2024). Acta Cryst. B80, https://doi.org/10.1107/S2052520624007492. Google Scholar
Johnson, E. R. (2017). In Non-covalent Interactions in Quantum Chemistry and Physics, edited by A. Otero-de-la-Roza & G. A. DiLabio, ch. 5, pp. 169–194. Elsevier. Google Scholar
Kendrick, J., Leusen, F. J. J., Neumann, M. A. & van de Streek, J. (2011). Chem. A Eur. J. 17, 10736–10744. Web of Science CrossRef CAS Google Scholar
Kim, M.-C., Sim, E. & Burke, K. (2013). Phys. Rev. Lett. 111, 073003. CrossRef PubMed Google Scholar
Lejaeghere, K., Bihlmayer, G., Björkman, T., Blaha, P., Blügel, S., Blum, V., Caliste, D., Castelli, I. E., Clark, S. J., Dal Corso, A., de Gironcoli, S., Deutsch, T., Dewhurst, J. K., di Marco, I., Draxl, C., Dułak, M., Eriksson, O., Flores-Livas, J. A., Garrity, K. F., Genovese, L., Giannozzi, P., Giantomassi, M., Goedecker, S., Gonze, X., Grånäs, O., Gross, E. K. U., Gulans, A., Gygi, F., Hamann, D. R., Hasnip, P. J., Holzwarth, N. A. W., Iuşan, D., Jochym, D. B., Jollet, F., Jones, D., Kresse, G., Koepernik, K., Küçükbenli, E., Kvashnin, Y. O., Locht, I. L. M., Lubeck, S., Marsman, M., Marzari, N., Nitzsche, U., Nordström, L., Ozaki, T., Paulatto, L., Pickard, C. J., Poelmans, W., Probert, M. I. J., Refson, K., Richter, M., Rignanese, G.-M., Saha, S., Scheffler, M., Schlipf, M., Schwarz, K., Sharma, S., Tavazza, F., Thunström, P., Tkatchenko, A., Torrent, M., Vanderbilt, D., van Setten, M. J., Speybroeck, V. V., Wills, J. M., Yates, J. R., Zhang, G.-X. & Cottenier, S. (2016). Science, 351, aad3000. CrossRef PubMed Google Scholar
Lommerse, J. P. M., Motherwell, W. D. S., Ammon, H. L., Dunitz, J. D., Gavezzotti, A., Hofmann, D. W. M., Leusen, F. J. J., Mooij, W. T. M., Price, S. L., Schweizer, B., Schmidt, M. U., van Eijck, B. P., Verwer, P. & Williams, D. E. (2000). Acta Cryst. B56, 697–714. Web of Science CSD CrossRef CAS IUCr Journals Google Scholar
Macrae, C. F., Sovago, I., Cottrell, S. J., Galek, P. T. A., McCabe, P., Pidcock, E., Platings, M., Shields, G. P., Stevens, J. S., Towler, M. & Wood, P. A. (2020). J. Appl. Cryst. 53, 226–235. Web of Science CrossRef CAS IUCr Journals Google Scholar
Marom, N., DiStasio, R. A. Jr, Atalla, V., Levchenko, S., Reilly, A. M., Chelikowsky, J. R., Leiserowitz, L. & Tkatchenko, A. (2013). Angew. Chem. Int. Ed. 52, 6629–6632. Web of Science CrossRef CAS Google Scholar
Mayo, R. A. & Johnson, E. R. (2021). CrystEngComm, 23, 7118–7131. Web of Science CrossRef CAS Google Scholar
Mayo, R. A., Otero-de-la-Roza, A. & Johnson, E. R. (2022). CrystEngComm, 24, 8326–8338. Web of Science CrossRef CAS Google Scholar
Mortazavi, M., Hoja, J., Aerts, L., Quéré, L., van de Streek, J., Neumann, M. A. & Tkatchenko, A. (2019). Commun. Chem. 2, 1–7. CAS Google Scholar
Motherwell, S. & Chisholm, J. A. (2005). J. Appl. Cryst. 38, 228–231. CrossRef CAS IUCr Journals Google Scholar
Motherwell, W. D. S., Ammon, H. L., Dunitz, J. D., Dzyabchenko, A., Erk, P., Gavezzotti, A., Hofmann, D. W. M., Leusen, F. J. J., Lommerse, J. P. M., Mooij, W. T. M., Price, S. L., Scheraga, H., Schweizer, B., Schmidt, M. U., van Eijck, B. P., Verwer, P. & Williams, D. E. (2002). Acta Cryst. B58, 647–661. Web of Science CrossRef CAS IUCr Journals Google Scholar
Neumann, M. A., Leusen, F. J. J. & Kendrick, J. (2008). Angew. Chem. Int. Ed. 47, 2427–2430. Web of Science CrossRef CAS Google Scholar
Neumann, M. A. & Perrin, M.-A. (2005). J. Phys. Chem. B, 109, 15531–15541. Web of Science CrossRef PubMed CAS Google Scholar
Neumann, M. A. & van de Streek, J. (2018). Faraday Discuss. 211, 441–458. Web of Science CrossRef CAS PubMed Google Scholar
Nyman, J. & Day, G. M. (2015). CrystEngComm, 17, 5154–5165. Web of Science CrossRef CAS Google Scholar
Nyman, J. & Day, G. M. (2016). Phys. Chem. Chem. Phys. 18, 31132–31143. Web of Science CrossRef CAS PubMed Google Scholar
Otero-de-la-Roza, A. & Johnson, E. R. (2012). J. Chem. Phys. 137, 054103. Web of Science PubMed Google Scholar
Otero-de-la-Roza, A., Johnson, E. R. & Luaña, V. (2014). Comput. Phys. Commun. 185, 1007–1018. CAS Google Scholar
Perdew, J. P., Burke, K. & Ernzerhof, M. (1996). Phys. Rev. Lett. 77, 3865–3868. CrossRef PubMed CAS Web of Science Google Scholar
Price, A. J. A., Mayo, R. A., Otero-de-la-Roza, A. & Johnson, E. R. (2023a). CrystEngComm, 25, 953–960. CrossRef CAS Google Scholar
Price, A. J. A., Otero-de-la-Roza, A. & Johnson, E. R. (2023b). Chem. Sci. 14, 1252–1262. CrossRef CAS PubMed Google Scholar
Price, S. L. (2013). Acta Cryst. B69, 313–328. Web of Science CrossRef CAS IUCr Journals Google Scholar
Price, S. L. (2018). Faraday Discuss. 211, 9–30. Web of Science CrossRef CAS PubMed Google Scholar
Reilly, A. M., Cooper, R. I., Adjiman, C. S., Bhattacharya, S., Boese, A. D., Brandenburg, J. G., Bygrave, P. J., Bylsma, R., Campbell, J. E., Car, R., Case, D. H., Chadha, R., Cole, J. C., Cosburn, K., Cuppen, H. M., Curtis, F., Day, G. M., DiStasio, R. A. Jr, Dzyabchenko, A., van Eijck, B. P., Elking, D. M., van den Ende, J. A., Facelli, J. C., Ferraro, M. B., Fusti-Molnar, L., Gatsiou, C.-A., Gee, T. S., de Gelder, R., Ghiringhelli, L. M., Goto, H., Grimme, S., Guo, R., Hofmann, D. W. M., Hoja, J., Hylton, R. K., Iuzzolino, L., Jankiewicz, W., de Jong, D. T., Kendrick, J., de Klerk, N. J. J., Ko, H.-Y., Kuleshova, L. N., Li, X., Lohani, S., Leusen, F. J. J., Lund, A. M., Lv, J., Ma, Y., Marom, N., Masunov, A. E., McCabe, P., McMahon, D. P., Meekes, H., Metz, M. P., Misquitta, A. J., Mohamed, S., Monserrat, B., Needs, R. J., Neumann, M. A., Nyman, J., Obata, S., Oberhofer, H., Oganov, A. R., Orendt, A. M., Pagola, G. I., Pantelides, C. C., Pickard, C. J., Podeszwa, R., Price, L. S., Price, S. L., Pulido, A., Read, M. G., Reuter, K., Schneider, E., Schober, C., Shields, G. P., Singh, P., Sugden, I. J., Szalewicz, K., Taylor, C. R., Tkatchenko, A., Tuckerman, M. E., Vacarro, F., Vasileiadis, M., Vazquez-Mayagoitia, A., Vogt, L., Wang, Y., Watson, R. E., de Wijs, G. A., Yang, J., Zhu, Q. & Groom, C. R. (2016). Acta Cryst. B72, 439–459. Web of Science CrossRef IUCr Journals Google Scholar
Reilly, A. M. & Tkatchenko, A. (2013). J. Chem. Phys. 139, 024705. Web of Science CrossRef PubMed Google Scholar
Reilly, A. M. & Tkatchenko, A. (2014). Phys. Rev. Lett. 113, 055701. Web of Science CrossRef PubMed Google Scholar
Shtukenberg, A. G., Zhu, Q., Carter, D. J., Vogt, L., Hoja, J., Schneider, E., Song, H., Pokroy, B., Polishchuk, I., Tkatchenko, A., Oganov, A. R., Rohl, A. L., Tuckerman, M. E. & Kahr, B. (2017). Chem. Sci. 8, 4926–4940. Web of Science CSD CrossRef CAS PubMed Google Scholar
Tkatchenko, A., DiStasio, R. A., Car, R. & Scheffler, M. (2012). Phys. Rev. Lett. 108, 236402. Web of Science CrossRef PubMed Google Scholar
Togo, A. & Tanaka, I. (2015). Scr. Mater. 108, 1–5. Web of Science CrossRef CAS Google Scholar
Ullmann, J. R. (1976). JACM, 23, 31–42. CrossRef Google Scholar
Weatherby, J. A., Rumson, A. F., Price, A. J. A., Otero de la Roza, A. & Johnson, E. R. (2022). J. Chem. Phys. 156, 114108. CrossRef PubMed Google Scholar
Whittleton, S. R., Otero-de-la-Roza, A. & Johnson, E. R. (2017a). J. Chem. Theory Comput. 13, 441–450. Web of Science CrossRef CAS PubMed Google Scholar
Whittleton, S. R., Otero-de-la-Roza, A. & Johnson, E. R. (2017b). J. Chem. Theory Comput. 13, 5332–5342. Web of Science CrossRef CAS PubMed Google Scholar
Yang, S. & Day, G. M. (2021). J. Chem. Theory Comput. 17, 1988–1999. CrossRef CAS PubMed Google Scholar
This is an open-access article distributed under the terms of the Creative Commons Attribution (CC-BY) Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are cited.