research papers
A jumping crystal predicted with
and analysed with TLS against powder diffraction dataaDepartment of Pharmacy, University of Copenhagen, Copenhagen, Denmark, bInstitute for Inorganic and Analytical Chemistry, Goethe-University Frankfurt, Frankfurt am Main, Germany, cSchool of Chemistry/Centre for Science at Extreme Conditions, University of Edinburgh, Edinburgh, UK, and dChemistry Department, University of Malta, Msida, Malta
*Correspondence e-mail: jaccovandestreek@yahoo.co.uk
By running a temperature series of
(MD) simulations starting from the known low-temperature phase, the experimentally observed in a `jumping crystal' was captured, thereby providing a prediction of the unknown of the high-temperature phase and clarifying the phase-transition mechanism. The is accompanied by a discontinuity in two of the unit-cell parameters. The structure of the high-temperature phase is very similar to that of the low-temperature phase. The anisotropic displacement parameters calculated from the MD simulations readily identified libration as the driving force behind the Both the predicted and the phase-transition mechanism were verified experimentally using TLS (translation, libration, screw) against X-ray powder diffraction data.Keywords: jumping crystals; molecular dynamics; phase transitions; TLS refinement; XRPD.
CCDC reference: 1867396
1. Introduction
Occasionally, interesting crystal structures can be prepared by applying temperature or pressure to a crystalline phase in order to trigger a ), the dehydration of hydrates to prepare anhydrates (Fujii et al., 2012) or the heating of crystals to make them `jump' (Sahoo et al., 2013). These cases have in common that the structures of the starting phases and the preparation processes are known, but the structures of the end products are difficult to determine because the crystals tend to shatter in the process, leaving nothing but a sample in powder form to work with. Indeed, in these cases the of the resulting phase, which is often the phase of interest, must be determined from powder diffraction data (Trask et al., 2005; Bond et al., 2014; Panda et al., 2014).
and obtain a different solid phase. Examples are the preparation of co-crystals by solid-state grinding (Trask & Jones, 2005This led us to consider a computational approach to the determination of the structure of the new phase based on the two pieces of information that are known: the structure of the starting phase and the preparation process of the new phase. The known starting structure can be recreated computationally by means of molecular modelling (or, in the ideal case, by in silico with MD, we expect to observe a the product of which hopefully corresponds to the phase of interest.
prediction), whereas the process can be recreated computationally by means of (MD). In other words, when the known (or predicted) starting structure is heatedAs a model system, we used a jumping crystal for which the trans,trans,anti,trans,trans-perhydropyrene (I). The of the initial phase (P21/c, Z′ = 2 × ½) was reported in 1991, and the crystals are reported to jump and shatter at 344.5 K (Ding et al., 1991).
of the initial phase is known, whereas that of the post-salient phase is not:The success of our approach was subsequently verified by temperature-dependent powder diffraction experiments. A TLS (translation, libration, screw)
of the experimental data confirmed the phase-transition mechanism observed in the MD simulations.2. Methods
2.1. Molecular dynamics
The et al., 2016; reference code KOFHOC). To ensure that our calculations could have been performed even in the absence of any experimental data, to allow for the possibility of a true prediction, the experimental was first energy minimized with an unrestrained and dispersion-corrected density functional theory (DFT-D). This is the that would have been obtained from a modern-day prediction study (Reilly et al., 2016), even in the absence of experimental data, provided the study had been successful in locating the most experimentally stable polymorph. DFT-D is too computationally expensive to be used for MD, and the COMPASS force field (Sun, 1998) was used for classical MD simulations. The DFT-D was energy-optimized, with an unrestrained with the COMPASS force field to yield a structure corresponding to T = 0 K, the temperature of which was then slowly increased. The Forcite Plus module in Materials Studio was used, periodic boundary conditions were applied, the time step was 1 fs and the of all the simulation cells was P1. To allow the use of a relatively large cut-off distance for non-bonded interactions (electrostatic and van der Waals) and to reduce the self-interactions introduced by the periodic boundary conditions, relatively large supercells must be used. The cut-off distance for both types of non-bonded interaction was 25 Å and each side-to-side distance of the was not less than 50 Å (twice the cut-off distance). Furthermore, the number of molecules in each direction of the simulation cell must be commensurate with the number of molecules in the starting structure and the unknown target structure. This can be achieved by constructing a simulation box containing 12 × 12 × 12 = 1728 molecules, because 12 is a multiple of 1, 2, 3, 4 and 6, allowing a to virtually all of the 230 possible space groups.1 For the low-temperature phase of I, the 12 × 12 × 12 simulation box had dimensions a = 99.46 Å, b = 62.95 Å, c = 97.86 Å, α = 90.0°, β = 119.16°, γ = 90.0° and V = 535 123 Å3 at 0 K. Fig. 1 gives an impression of the dimensions of the simulation box.
of the low-temperature phase was taken from the Cambridge Structural Database (GroomThe COMPASS force field comes with its own atomic point charges, eliminating the ambiguity regarding the parameterization of the atomic charges associated with many force fields. The COMPASS force field is parameterized to be used without Ewald summation; instead, charge groups equating to whole molecules were used to ensure charge neutrality of the Coulomb summations.
To bracket the temperature of the anticipated et al., 1984) were used; in this step the cell parameters were fixed and the simulation time was 2.0 ps. Second, the NPT ensemble and the Berendsen thermostat and barostat (allowing for isotropic scaling of the unit-cell volume only) were used for 3.0 ps. In the third step, the NPT ensemble was used for 20.0 ps, with the Nosé–Hoover–Langevin (NHL) thermostat (Samoletov et al., 2007) to control the temperature and the Parrinello barostat (Parrinello & Rahman, 1981) to control the pressure. The equilibration was followed by a 25 ps NPT production run that was used for the calculation of the average unit-cell parameters. The temperature series was continued up to the melting point, which occurred in the simulation at 500 K, to exclude the possibility of a second in the MD simulations. So in total, 40 successive MD simulations were run, each simulation starting from the final positions and velocities of the previous one, with a total duration of 500 ps. At each temperature, the unit-cell parameters were sampled 432 times and averaged. The first temperature series put the phase-transition temperature between 250 and 300 K, which was refined to between 250 and 275 K with three additional MD runs at 225, 275 and 325 K.
MD simulations were carried out at intervals of 50 K, starting at 50 K. At each temperature, the atomic velocities were initialized with the final values from the previous temperature. The first MD simulation, at 50 K, was initialized with random atomic velocities. Because our systems are extremely small compared with real-life systems (of the order of 1000 molecules) and the time scales are extremely short (several ps), the sudden increase in temperature by 50 K represents a severe temperature shock for the system, and experimenting with different equilibration protocols showed that careful equilibration is needed to avoid artefacts. The equilibration at each temperature was split into three steps. First, the NVT ensemble and the Berendsen thermostat (BerendsenFrom the 12 points of the temperature series, it is largely the phases just before and just after the i.e. averages were calculated over 1001 snapshots (frames). The of the new phase was determined from the 12 × 12 × 12 P1 averaged over 1001 frames with the space-group perception tool in Materials Studio with a tolerance of 0.2 Å. Once the of the new phase had been established, which was determined to be P21/c, Z′ = ½, the supercells were collapsed onto the and all symmetry-equivalent atoms were averaged over space and time. In other words, for the calculated structure of the new phase, each atomic coordinate was the average of 1001 × 12 × 12 × 12 × 2 = 3 459 456 values. The unit-cell parameters of the new phase were obtained by averaging over the 1001 frames. From these two 1 ns NVT simulations, the time-averaged structures, the anisotropic displacement parameters and the X-ray powder diffraction patterns were calculated using in-house software.
that are of interest. These two phases were investigated with greater precision by first running a long (1 ns) NPT simulation from which the average unit-cell parameters were determined, followed by a long (1 ns) NVT simulation with the average unit-cell parameters from the NPT simulation imposed. The fixed unit-cell parameters in the second simulation are intended to mimic the effect of a much larger external crystal surrounding our simulation box and are necessary to yield correct anisotropic displacement parameters that are free from the effects of unit-cell fluctuations. The MD trajectory was sampled every 1 ps,On modern hardware, the calculations are relatively trivial in terms of resources. MD simulations do not require the huge amounts of memory (RAM) that are characteristic of quantum-mechanical calculations (such as DFT calculations). The algorithms have been parallelized, so by running the simulations on, 48 cores simultaneously for example, the wall-clock time of all the simulations described in the current work amounts to only a couple of weeks.
Minor rounding errors in the integration of Newton's equations of motion lead to a minor but noticeable drift of the centre of mass of the contents of the simulation cell. This drift is nonphysical and was removed by resetting the centre of mass of all frames to that of the first frame. The values of the drift corrections were manually inspected to ensure that they were very small and not indicative of any problems in the calculations.
2.2. Synthesis
A mixture of isomers of perhydropyrene (1 g) was purchased from Sigma–Aldrich with a purity of 95%. Purification of the mixture of isomers followed the procedure published elsewhere (Langer & Lehner, 1973). Benzene, n-hexane, dichloromethane and ethanol were obtained from Scharlau (reaction grade). AlCl3 was purchased from Alfa Aesar (anhydrous, 99.985% metal basis). All manipulations which included the use of dry AlCl3 were carried out under exclusion of air and moisture using a Schlenk-line.
A total of 1 g (4.58 mmol) of the mixture of isomers of I (MW 218.38 g mol−1) was dissolved in 20 ml of freshly distilled benzene and 670 mg of AlCl3 was added; the resulting solution was heated under reflux for 2 h.
The following steps were carried out in open air. After the solution was left to cool completely, 50 ml of 2 M HCl and 150 ml CH2Cl2 were added and stirred rapidly for 15 min. The organic phase was separated from the aqueous phase and washed three times with distilled water. Drying over NaSO4 and removal of the solvent resulted in a dark yellow/orange oil. In order to separate the dehydration products from I, was carried out using Al2O3 (column width 3 cm, length 10 cm) with n-hexane as Subsequent evaporation of the collected fractions and recrystallization in absolute ethanol yielded 296.5 mg of I (30% of the theoretical yield). The melting point of 103 °C confirmed the correct and pure product.
2.3. Powder diffraction
X-ray powder diffraction data were recorded at 350 K on a Stoe Stadi-P diffractometer equipped with a focusing Ge(111) monochromator and a linear position-sensitive detector using Cu Kα1 radiation. The sample was contained in a glass capillary with 1.0 mm diameter, which was spun during the measurement. Data were collected in a 2θ range from 2 to 100° with a step width of 0.01° and a total data collection time of about 18 h. The software WinXPOW (Stoe & Cie, 2009) was used for data acquisition.
2.4. Rietveld refinement
TOPAS (Coelho, 2018). To ensure that the was as robust and smooth as possible, it was carried out in stages, gradually releasing more and more parameters. The first two stages consisted of Pawley refinements to establish suitable starting values for the unit-cell parameters and the peak-shape functions, which were then transferred to the In the last cycle, all parameters were refined simultaneously, a requirement for the proper evaluation of the goodness of fit, which depends on the number of parameters included in the fit. Anisotropic peak broadening and peak asymmetry were included to allow the peak profiles to be described accurately. Since in both structures the molecules are situated on inversion centres, the asymmetric units in principle contain only five of the ten carbon atoms of the two central cyclohexane rings. However, the five symmetry-related carbon atoms missing from the list of atoms prevent the specification of several of the bond-length restraints and valence-angle restraints. Therefore, the entire molecule was specified in the input file but with appropriate symmetry constraints to ensure that the number of did not change. All restraints can then be expressed in natural variables such as C—C bond lengths and C—C—C valence angles without the need for dummy atoms.
was carried out withThree Rietveld refinements with three different models for the description of the thermal parameters were carried out. The first Uiso for all non-hydrogen atoms. The hydrogen atoms were assigned an isotropic displacement parameter equal to 1.2 times the isotropic displacement parameter for the non-hydrogen atoms. This is the model that we traditionally adopt when working with laboratory powder diffraction data. The second used the anisotropic displacement parameters (ADPs) calculated from the MD simulations, scaled by a linear scale factor. In this model, all atoms, including the hydrogen atoms, are described fully anisotropically, but only a single parameter is fitted, similar to the case of using a global Uiso. The third used the TLS model described in detail below.
used one global isotropic thermal parameter2.5. TLS refinement
It is generally not possible to refine anisotropic displacement parameters from powder diffraction data because the peak overlap reduces the information available, not leaving enough to refine six independent parameters per non-hydrogen atom. However, if it is assumed that the molecule behaves as a rigid body, then the thermal motion of the molecule as a whole can be described by means of a single set of TLS (Translation, Libration, Screw) parameters. For compound I, this reduces the number of parameters needed to describe the thermal motion of the non-hydrogen atoms in the harmonic approximation from 48, when individual ADPs are used, to 20 when ADPs from a rigid-body approximation are used. In addition, in both polymorphs the molecules are on an inversion centre, as a result of which the screw tensor vanishes, further reducing the number of parameters to just 12 (Schomaker & Trueblood, 1968; Dunitz et al., 1988; Downs, 2000).
The specification of the TLS TOPAS input file requires a conversion from fractional to Cartesian coordinates and from Cartesian coordinates to the ADPs in format (Grosse-Kunstleve & Adams, 2002), requiring many lines of equations for each atom.
in theThe DFT-D energy minimization from which the MD simulations were started provides us with near-perfect bond lengths and valence angles that can be used as restraints in the e.g. van de Streek, 2015). Librational thermal motion however, shortens the bond lengths calculated from the average atomic positions, while the DFT-D energy minimization is a T = 0 K calculation and therefore describes the molecule at rest. TLS describes the motion of a rigid body and the TLS parameters can be used to correct the intramolecular distances of the rigid body for the effects of the motion (Downs, 2000). In other words, the TLS parameters can be used to calculate the coordinates of the underlying rigid body as if it were at rest. We can then apply the bond-length restraints and the valence-angle restraints from the DFT-D energy minimization to the bond lengths and valence angles of the static rigid body recovered from the TLS The thermal motion correction was implemented in the TOPAS input file and the bond-length and valence-angle restraints from the DFT-D minimization were used in the Rietveld refinement.
(seeIn total, about 2000 lines of equations were required for the TLS I on an inversion centre with restraints corrected for thermal motion. An example TOPAS input file can be found in the supporting information.
of3. Results and discussion
3.1. Molecular dynamics
The results of the MD simulations were analysed by plotting the unit-cell parameters as a function of temperature, immediately revealing a discontinuity in the unit-cell parameters a and β between 250 and 275 K (Fig. 2). The b, c and unit-cell volume parameters appeared to change smoothly; α and γ remained at 90° (Fig. 2). Applying the symmetry-perception algorithm in Materials Studio to the time-averaged NVT structure at 275 K (after the phase transition) consisting of 12 × 12 × 12 molecules yielded a structure in P21/c, Z′ = ½, with the molecule occupying a centre of symmetry. Indexing of the powder diffraction pattern calculated from the MD trajectory gave the same result. The low-temperature and high-temperature phases are very similar and differ only in a minor reorientation of the molecules in the plane of the molecule [Fig. 3(a)]; when viewed along the plane of the molecule, the two structures are virtually identical [Fig. 3(b)]. A similar case of a crystal jumping upon heating while undergoing a transformation from a low-temperature phase to a structurally very similar high-temperature phase due to an anisotropic expansion (shown by three unit-cell parameters changing abruptly while the other unit-cell parameters showed no pronounced discontinuity) has been investigated with single-crystal X-ray diffraction (Lusi & Bernstein, 2013). In MD simulations of phase transitions upon heating in several other molecular compounds (van de Streek et al., unpublished results), all changes in the unit-cell dimensions were consistently smooth, suggesting that the jumping of crystals is associated with a discontinuity in at least one unit-cell parameter.
The new high-temperature phase is a local minimum in the DFT-D potential, i.e. at T = 0 K, with an energy that is only 0.1 kcal mol−1 higher than that of the low-temperature form.
The (a), with β = 122°, can be transformed to a with β closer to 90° (Fig. 4, β = 92°) by applying the unit-cell transformation [101, 010, −100]. The space-group setting then becomes P21/n, Z′ = ½.
shown in Fig. 3The ADPs calculated from the MD trajectory, which represent the thermal motion of the molecules, are shown in Fig. 6(a). To allow a direct quantitative comparison to the ADPs fitted directly to the powder data, the ADPs from the MD simulations at 275 K were fitted against the powder diffraction data at 350 K with a linear scale factor (Blessing, 1995). The scale factor refined to a value of 1.50 (2). The ADPs beautifully illustrate the mechanism behind the the molecules librate in the plane of the molecule until the orientations of the two symmetry-independent molecules line up and the molecules become translationally equivalent. This explanation is also in agreement with the similarity between the two phases when viewed along the plane of the molecules [Fig. 3(a)].
The predicted phase-transition temperature of about 260 K is about 85 K lower than the experimental transition temperature of 344 K. The computationally modelled crystals are free from defects, impurities and interfaces which, combined with the short simulation times, systematically shift the transition temperatures in MD simulations to higher temperatures, so 260 K is an upper limit. The discrepancy of more than 85 K between the experimental and the simulated transition temperature can be attributed to the inaccuracy of the energy potential used in the MD simulations. Suggestions to improve the accuracy of the force field are given below.
3.2. Rietveld refinement
Fig. 5 shows the When only one isotropic thermal parameter for all non-hydrogen atoms was refined, it refined to the value Uiso = 8.94 (7) Å2 [Fig. 6(c)]. Table 1 shows various parameters characterizing the three Rietveld refinements.
‡Fig. 6(a). §Fig. 6(b). ¶With and without background subtraction, respectively. ††20 if the molecule is not on a special position. ‡‡March–Dollase correction with direction [010]. |
3.3. TLS refinement
The ADPs obtained from the TLS (b), confirming that the experimental and the simulated ADPs are very similar. The current structure must be one of very few structures determined from X-ray powder diffraction data – in this case even laboratory X-ray powder diffraction data – for which credible anisotropic displacement parameters for the hydrogen atoms were fitted directly against the X-ray data without the need for complementary experimental techniques. This is only possible of course because of the rigid-body approximation, such that the motion of the hydrogen atoms is effectively defined by the motion of the non-hydrogen atoms. This is a generally applicable method for the calculation of anisotropic displacement parameters for hydrogen atoms, and is well developed for single-crystal structures (Madsen & Hoser, 2014).
are shown in Fig. 6The experimental ADPs as a function of temperature can be analysed quantitatively to obtain information about the reaction pathway [Hummel, Hauser et al. (1990); Hummel, Raselli et al. (1990); see also footnote 73 in Skoko et al. (2010)], but this was beyond the scope of the current study.
3.4. Validity of the rigid-body approximation
The TLS I and to conclude, as we did, that this is a reasonable assumption. Indeed, in tests on a handful of other compounds for which the rigid-body approximation seemed unlikely to apply, the ADPs quickly became non-positive definite, the refinements did not converge properly and the ADPs of atoms far from the centre of the molecule refined to very extreme shapes. Dunitz et al. (1988) described a quantitative test to check if the thermal motion of the individual atoms supports an approximation as a rigid body: the average displacement along a vector connecting any two atoms must be identical for the two atoms. In practice, this is implemented by fitting individual ADPs for all atoms, i.e. without the rigid-body assumption, so that the average displacements can be calculated and compared. This approach, however, is not applicable in our case because the individual displacements are not accessible for powder diffraction data, so the test cannot be carried out.
against the XRDP data is based on the assumption that the molecule is rigid. To date, we have not presented any evidence to support this assumption other than to allow the reader to visually inspectTherefore, the most critical assessment of the reliability of the ADPs reported here is limited to the observation that two entirely independent methods (MD and TLS ), and that the rigid-body approximation is valid and hence our ADPs are probably correct.
against experimental data) give the same reasonable result, which furthermore suggests a phase-transition mechanism consistent with the similarities and differences between the two phases (Fig. 33.5. Extension to other systems: TLS refinement
The success of our TLS I is clearly described very well as a rigid body in the solid state. When trying to apply the TLS to several other Rietveld refinements of pigments and pharmaceuticals, it quickly became clear that in the general case of a molecular compound, indiscriminately treating the whole molecule as a rigid body invariably leads to non-positive definite atoms, convergence problems in the and extreme ADPs for atoms far from the centre of the molecule. In principle, this can be solved by partitioning the molecule into smaller fragments, each of which can be described as a rigid body (Painter & Merritt, 2006a,b). However, to start with, in the absence of individual ADPs for refinements against powder diffraction data, such a partitioning requires a substantial amount of trial and error with no quantitative measure for success available and, moreover, such a division requires 20 × Nfrag parameters, where Nfrag is the number of fragments. This counteracts the purpose of introducing TLS parameters in the first place, which was to reduce the number of parameters.
hinges crucially on two factors. First, compoundThe second factor that played a role in our successful TLS
is the fact that the molecule is conveniently situated on a centre of symmetry in our of interest, further reducing the number of parameters from 20 to 12. Biologically active molecules, a large group of compounds that are currently actively investigated with X-ray powder diffraction, rarely possess molecular symmetry and are therefore rarely able to occupy special positions. A TLS will be more complicated for this class of compounds, even if the rigid-body approximation is applicable.3.6. Extension to other systems: MD simulations
How transferable are the MD simulations presented here to other systems? The accuracy of force fields is limited, and there is no guarantee that starting an MD simulation with a force field reproduces phase transitions observed in experiments. As a case in point, initial calculations on 1,2,4,5-tetrabromobenzene showed that the crystal structures of the low-and the high-temperature phases, which both happen to be known for this jumping crystal and which (again) happen to be very similar, both converged to the structure of the high-temperature phase upon energy minimization at 0 K with the COMPASS force field. DFT-D recognizes the two structures as distinct minima, even at 0 K, so the neglect of thermal motion does not seem to explain this failure of the force field for the high-temperature phase; the fact that halogen atoms are notoriously anisotropic in the solid state and not described well by the spherical approximations in force fields, on the other hand, does. A more thorough comparison of the performance of four off-the-shelf force fields for the condensed phases of seven molecules in terms of the reproduction of unit-cell parameters and ADPs was published by Nemkevich et al. (2010), and their conclusions agree with ours that qualitative results are generally good, but quantitatively the results are off by significant margins. In fact, there is no guarantee that a molecular compound can be described by an off-the-shelf force field at all: the energy potential of the palladium-containing organometallic compound reported by Panda et al. (2014) to jump and shatter cannot be set up with the COMPASS force field. Our preference in this paper for a compound consisting solely of formally sp3 carbon atoms and hydrogen atoms, which can both be described very well by the approximations inherent in the COMPASS force field, is therefore by no means a coincidence, but is dictated by the failure of the computational methods for alternative systems. A (partial) solution is the introduction of tailor-made force fields (TMFFs): force fields parameterized for one specific chemical compound at a time, allowing for more rigour in the description of the details of the energy potential of that compound (Neumann, 2008). In a previous paper, we described how MD with a tailor-made force field for cocaine describes the room-temperature of cocaine very well (Li et al., 2017). The accuracy of these force fields can be further improved by, for example, introducing anisotropic van der Waals interactions for halogens (Day et al., 2005, see section 4.4.2) or replacing the atomic point charges by atomic multipoles (Pyzer-Knapp et al., 2016).
The similarity of the structures before and after the
suggests that the energy barrier between them is small, which in all likelihood also played a major role in the straightforward reproduction of the Waiting for several seconds, minutes or longer until a has been initiated, as is possible in experiments, is not an option for atomistic simulations. Fortunately, in computational simulations, larger barriers may in principle be overcome by simply heating until the is sufficient to overcome the barrier within the short time scales of the simulation. Such an approach overshoots the actual transition temperature, possibly by a fair amount, but with a discrepancy of 85 K for this presumably best-case system the quantitative reproduction of the phase-transition temperature was never impressive.Jumping crystals are not the only materials where the starting material and the experimental conditions leading to the final product are known, while the structure of the product is not: an obvious class of interesting materials that require a similar approach involving MD to investigate the structure of the resulting phase at the molecular level are `dehydrated hydrates'. In the pharmaceutical industry, sometimes initially the ). Further processing may remove this water, resulting in a dehydrated hydrate, the structure of which is not necessarily the same as that of any known anhydrate of the compound (e.g. obtained by crystal growth under the exclusion of water). Again, the hydrate crystals tend to disintegrate during the dehydration process, hampering the determination of the relevant phase at the molecular level. For these dehydrated hydrates a modified MD protocol, allowing the water molecules to be removed, would potentially enable full insight into the dehydration process and product at the molecular level (Larsen, Rantanen et al., 2017; Larsen, Ruggiero et al., 2017).
of the solid form has water molecules incorporated into the crystal structures, forming a hydrate (Griesser, 20064. Conclusions
When an experimental starting point and procedure are known, computer simulations are, in principle, able to take us to the destination, even if that destination cannot be characterized or even accessed experimentally, providing in the process, a wealth of molecular-level information that might have been difficult or impossible to obtain directly through experiment. Whilst a simple model system was chosen to demonstrate this proof-of-principle, we are confident that improvements in computational methods will make these simulations useful for more complex systems, such as the elucidation of the structures of dehydrated hydrates in the pharmaceutical industry, perhaps, when combined with
prediction, even before the compounds have been synthesized.5. Software availability
The in-house software used to analyse the MD trajectory and to generate the input for the TLS TOPAS was written in C++ and is available from the corresponding author upon request.
withSupporting information
CCDC reference: 1867396
https://doi.org/10.1107/S205225251801686X/fc5028sup1.cif
contains datablocks global, I. DOI:Rietveld powder data: contains datablock I. DOI: https://doi.org/10.1107/S205225251801686X/fc5028Isup2.rtv
Example TOPAS input file for TLS https://doi.org/10.1107/S205225251801686X/fc5028sup3.txt
DOI:Data collection: WINXPOW (Stoe & Cie, 2004); cell
TOPAS-Academic (Coelho, 2018); data reduction: DASH 3.3 (David et al., 2006); program(s) used to solve structure: DASH 3.3 (David et al., 2006); program(s) used to refine structure: TOPAS-Academic (Coelho, 2018); molecular graphics: Mercury (Macrae et al., 2008).C16H26 | Z = 2 |
Mr = 218.38 | F(000) = 244 |
Monoclinic, P21/n | Dx = 1.067 Mg m−3 |
a = 14.3334 (3) Å | Cu Kα1 radiation, λ = 1.5406 Å |
b = 5.56223 (10) Å | T = 350 K |
c = 8.53021 (19) Å | Particle morphology: powder |
β = 92.0219 (15)° | colourless |
V = 679.65 (3) Å3 |
STOE-Stadi-P diffractometer | Scan method: continuous |
Ge(111) monochromator | 2θmin = 2.810°, 2θmax = 99.992°, 2θstep = 0.01° |
Least-squares matrix: full with fixed elements per cycle | 127 parameters |
Rp = 10.808 | 97 restraints |
Rwp = 10.005 | 0 constraints |
Rexp = 9.539 | Only H-atom coordinates refined |
RBragg = 0.799 | Weighting scheme based on measured s.u.'s w = 1/σ[Yobs]2 |
9719 data points | (Δ/σ)max = 0.001 |
Excluded region(s): none | Background function: Chebyshev function with 30 terms |
Profile function: fundamental parameters | Preferred orientation correction: March-Dollase Direction: [ 0 1 0 ] Refined parameter: 0.986(3) |
x | y | z | Biso*/Beq | ||
C1 | 0.0160 (10) | 0.0938 (9) | 0.0591 (9) | ||
C2 | −0.0292 (9) | 0.0605 (15) | 0.2157 (15) | ||
C7 | 0.1218 (9) | 0.1013 (19) | 0.0792 (15) | ||
H1 | −0.004 (4) | 0.245 (4) | 0.018 (6) | ||
C4 | 0.0025 (12) | 0.254 (2) | 0.3304 (16) | ||
C3 | −0.1344 (9) | 0.054 (2) | 0.1945 (12) | ||
H2 | −0.009 (4) | −0.090 (5) | 0.257 (6) | ||
C6 | 0.1530 (10) | 0.294 (2) | 0.1950 (19) | ||
C8 | 0.1664 (8) | 0.1333 (18) | −0.0772 (16) | ||
H7 | 0.141 (4) | −0.050 (5) | 0.121 (6) | ||
C5 | 0.1077 (12) | 0.267 (2) | 0.3512 (14) | ||
H4a | −0.021 (7) | 0.404 (8) | 0.293 (8) | ||
H4b | −0.025 (7) | 0.228 (14) | 0.429 (6) | ||
H3a | −0.156 (6) | 0.207 (7) | 0.160 (9) | ||
H3b | −0.162 (5) | 0.025 (17) | 0.292 (5) | ||
H6a | 0.2190 (13) | 0.290 (16) | 0.209 (14) | ||
H6b | 0.138 (6) | 0.447 (7) | 0.151 (9) | ||
H8a | 0.2323 (10) | 0.129 (14) | −0.063 (11) | ||
H8b | 0.151 (5) | 0.289 (6) | −0.117 (9) | ||
H5a | 0.125 (7) | 0.396 (9) | 0.419 (8) | ||
H5b | 0.130 (6) | 0.125 (7) | 0.402 (8) |
U11 | U22 | U33 | U12 | U13 | U23 | |
C1 | 0.073 (4) | 0.025 (5) | 0.055 (5) | 0.021 (5) | −0.004 (4) | 0.003 (4) |
C2 | 0.118 (6) | 0.065 (6) | 0.055 (5) | 0.037 (6) | 0.010 (4) | 0.002 (4) |
C7 | 0.077 (4) | 0.059 (6) | 0.097 (6) | 0.004 (6) | −0.023 (4) | 0.021 (5) |
H1 | 0.090 (5) | 0.022 (5) | 0.070 (5) | 0.023 (6) | −0.005 (4) | 0.003 (4) |
C4 | 0.243 (13) | 0.106 (9) | 0.072 (5) | 0.046 (9) | −0.010 (5) | −0.036 (5) |
C3 | 0.110 (6) | 0.118 (9) | 0.102 (7) | 0.057 (7) | 0.047 (6) | 0.029 (6) |
H2 | 0.117 (8) | 0.075 (7) | 0.055 (5) | 0.029 (7) | 0.005 (4) | 0.019 (4) |
C6 | 0.163 (8) | 0.092 (10) | 0.151 (10) | −0.034 (8) | −0.078 (7) | 0.008 (7) |
C8 | 0.071 (4) | 0.110 (9) | 0.140 (10) | 0.010 (6) | 0.011 (5) | 0.054 (6) |
H7 | 0.073 (4) | 0.074 (8) | 0.102 (8) | 0.018 (6) | −0.023 (5) | 0.028 (5) |
C5 | 0.260 (15) | 0.121 (12) | 0.109 (7) | −0.007 (11) | −0.083 (8) | −0.030 (7) |
H4a | 0.281 (17) | 0.094 (9) | 0.103 (7) | 0.062 (10) | −0.004 (9) | −0.053 (7) |
H4b | 0.32 (2) | 0.174 (15) | 0.069 (6) | 0.074 (14) | 0.014 (8) | −0.040 (7) |
H3a | 0.128 (7) | 0.125 (11) | 0.125 (11) | 0.083 (8) | 0.053 (7) | 0.021 (7) |
H3b | 0.155 (11) | 0.192 (15) | 0.126 (10) | 0.077 (10) | 0.082 (10) | 0.048 (9) |
H6a | 0.167 (12) | 0.150 (16) | 0.221 (17) | −0.063 (11) | −0.110 (11) | 0.038 (11) |
H6b | 0.199 (12) | 0.075 (9) | 0.182 (14) | −0.045 (10) | −0.081 (9) | 0.003 (8) |
H8a | 0.070 (4) | 0.179 (16) | 0.213 (16) | 0.004 (8) | 0.013 (7) | 0.089 (11) |
H8b | 0.090 (6) | 0.107 (10) | 0.155 (11) | −0.010 (8) | 0.006 (7) | 0.066 (7) |
H5a | 0.38 (2) | 0.162 (17) | 0.152 (11) | −0.023 (16) | −0.125 (13) | −0.061 (11) |
H5b | 0.247 (18) | 0.155 (16) | 0.101 (8) | −0.005 (13) | −0.091 (10) | −0.008 (7) |
C1—C2 | 1.52 (2) | C3—H3a | 0.95 (5) |
C1—C7 | 1.52 (2) | C3—H3b | 0.95 (5) |
C1—H1 | 0.95 (3) | C3—C8 | 1.50 (2) |
C2—C4 | 1.51 (2) | C6—C5 | 1.51 (2) |
C2—C3 | 1.51 (2) | C6—H6a | 0.95 (3) |
C2—H2 | 0.95 (4) | C6—H6b | 0.95 (5) |
C7—C6 | 1.51 (2) | C8—H8a | 0.95 (2) |
C7—C8 | 1.51 (2) | C8—H8b | 0.95 (4) |
C7—H7 | 0.95 (4) | C8—C3 | 1.50 (2) |
C4—C5 | 1.51 (2) | C5—H5a | 0.95 (6) |
C4—H4a | 0.95 (6) | C5—H5b | 0.95 (5) |
C4—H4b | 0.95 (7) | ||
C2—C1—C7 | 111.1 (8) | C2—C3—C8 | 112.0 (9) |
C2—C1—H1 | 107 (3) | H3a—C3—H3b | 106 (6) |
C2—C1—C1 | 112.0 (8) | H3a—C3—C8 | 109 (4) |
C7—C1—H1 | 108 (3) | H3b—C3—C8 | 110 (4) |
C7—C1—C1 | 111.9 (8) | C7—C6—C5 | 112 (1) |
H1—C1—C1 | 107 (3) | C7—C6—H6a | 110 (5) |
C1—C2—C4 | 110.8 (9) | C7—C6—H6b | 109 (4) |
C1—C2—C3 | 110.6 (9) | C5—C6—H6a | 110 (5) |
C1—C2—H2 | 107 (3) | C5—C6—H6b | 110 (4) |
C4—C2—C3 | 112 (1) | H6a—C6—H6b | 106 (6) |
C4—C2—H2 | 108 (3) | C7—C8—H8a | 110 (4) |
C3—C2—H2 | 108 (3) | C7—C8—H8b | 109 (4) |
C1—C7—C6 | 111 (1) | C7—C8—C3 | 112 (1) |
C1—C7—C8 | 110.8 (9) | H8a—C8—H8b | 107 (5) |
C1—C7—H7 | 107 (3) | H8a—C8—C3 | 110 (4) |
C6—C7—C8 | 112 (1) | H8b—C8—C3 | 109 (4) |
C6—C7—H7 | 108 (3) | C4—C5—C6 | 111 (1) |
C8—C7—H7 | 108 (3) | C4—C5—H5a | 110 (5) |
C2—C4—C5 | 113 (1) | C4—C5—H5b | 109 (4) |
C2—C4—H4a | 108 (4) | C6—C5—H5a | 111 (5) |
C2—C4—H4b | 110 (5) | C6—C5—H5b | 110 (4) |
C5—C4—H4a | 110 (4) | H5a—C5—H5b | 106 (6) |
C5—C4—H4b | 110 (5) | C1—C1—C2 | 112.0 (8) |
H4a—C4—H4b | 106 (6) | C1—C1—C7 | 111.9 (8) |
C2—C3—H3a | 109 (4) | C1—C1—H1 | 107 (3) |
C2—C3—H3b | 110 (4) |
Footnotes
1In some space groups, unit cells with a 1 × 1 × 8 distribution of the molecules are possible, for example in P43212, Z′ = 1 (see e.g. Beurskens et al., 1980; CSD reference code MTKPIN) or a less exotic case, P212121, Z′ = 2 (see e.g. Birnbaum, 1977; CSD reference code HXUBIM10); P212121, Z′ = 2 is fairly common for biologically active compounds such as pharmaceuticals (Neumann & van de Streek, 2018). A 1 × 1 × 8 distribution is not commensurate with a 12 × 12 × 12 simulation box and hence, to capture all possible phase transitions, either a simulation box containing 24 × 24 × 24 = 13 824 molecules must be used, which would be prohibitively expensive in terms of the required hardware resources, or, as a more reasonable alternative, two separate simulations must be run, one with 12 × 12 × 12 = 1728 molecules and another with 8 × 8 × 8 = 512 molecules (or variations thereof, such as 16 × 8 × 8 = 1024 molecules or 16 × 16 × 8 = 2048 molecules to ensure that the simulation box is large enough in each direction to avoid self-interactions). It is not guaranteed however, that the initial phase allows the construction of such a simulation box. And even a 24 × 24 × 24 simulation box would not allow a to, for example, a Z′ = 5 phase, as 13 824 is not divisible by 5.
Acknowledgements
Professor Martin U. Schmidt (Goethe University, Frankfurt, Germany) is gratefully acknowledged for arranging all experimental work.
Funding information
The Materials Studio software was funded by the Lundbeck Foundation (Denmark) (grant No. R49-A5604). The Villum Foundation (Denmark) is gratefully acknowledged for financial support (project No. VKR023111). The Goethe-University Frankfurt is gratefully acknowledged for covering the open-access fee. Crystal structures were visualized using Mercury (Macrae et al., 2008).
References
Berendsen, H. J. C., Postma, J. P. M., van Gunsteren, W. F., DiNola, A. & Haak, J. R. (1984). J. Chem. Phys. 81, 3684–3690. CrossRef CAS Web of Science Google Scholar
Beurskens, G., Noordik, J. H. & Beurskens, P. T. (1980). Cryst. Struct. Commun. 9, 23–28. CAS Google Scholar
Birnbaum, G. I. (1977). Can. J. Chem. 55, 1619–1623. CrossRef CAS Google Scholar
Blessing, R. H. (1995). Acta Cryst. B51, 816–823. CrossRef CAS Web of Science IUCr Journals Google Scholar
Bond, A. D., Cornett, C., Larsen, F. H., Qu, H., Raijada, D. & Rantanen, J. (2014). IUCrJ, 1, 328–337. CrossRef CAS PubMed IUCr Journals Google Scholar
Coelho, A. A. (2018). J. Appl. Cryst. 51, 210–218. Web of Science CrossRef CAS 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
Ding, J., Herbst, R., Praefcke, K., Kohne, B. & Saenger, W. (1991). Acta Cryst. B47, 739–742. CSD CrossRef CAS Web of Science IUCr Journals Google Scholar
Downs, R. T. (2000). Rev. Mineral. Geochem. 41, 61–87. CrossRef Google Scholar
Dunitz, J. D., Schomaker, V. & Trueblood, K. N. (1988). J. Phys. Chem. 92, 856–867. CrossRef CAS Web of Science Google Scholar
Fujii, K., Uekusa, H., Itoda, N., Yonemochi, E. & Terada, K. (2012). Cryst. Growth Des. 12, 6165–6172. Web of Science CrossRef CAS Google Scholar
Griesser, U. J. (2006). Polymorphism in the Pharmaceutical Industry, edited by R. Hilfiker, pp. 211–234. Weinheim: Wiley–VCH. Google Scholar
Groom, C. R., Bruno, I. J., Lightfoot, M. P. & Ward, S. C. (2016). Acta Cryst. B72, 171–179. Web of Science CrossRef IUCr Journals Google Scholar
Grosse-Kunstleve, R. W. & Adams, P. D. (2002). J. Appl. Cryst. 35, 477–480. Web of Science CrossRef CAS IUCr Journals Google Scholar
Hummel, W., Hauser, J. & Bürgi, H.-B. (1990). J. Mol. Graph. 8, 214–220. CrossRef CAS PubMed Web of Science Google Scholar
Hummel, W., Raselli, A. & Bürgi, H.-B. (1990). Acta Cryst. B46, 683–692. CrossRef CAS Web of Science IUCr Journals Google Scholar
Langer, E. & Lehner, H. (1973). Monatsh. Chem. 104, 1154–1163. CrossRef CAS Google Scholar
Larsen, A. S., Rantanen, J. & Johansson, K. E. (2017). J. Pharm. Sci. 106, 348–355. Web of Science CrossRef CAS PubMed Google Scholar
Larsen, A. S., Ruggiero, M. T., Johansson, K. E., Zeitler, J. A. & Rantanen, J. (2017). Cryst. Growth Des. 17, 5017–5022. CrossRef CAS Google Scholar
Li, X., Neumann, M. A. & van de Streek, J. (2017). IUCrJ, 4, 175–184. Web of Science CrossRef PubMed IUCr Journals Google Scholar
Lusi, M. & Bernstein, J. (2013). Chem. Commun. 49, 9293–9295. Web of Science CrossRef CAS Google Scholar
Macrae, C. F., Bruno, I. J., Chisholm, J. A., Edgington, P. R., McCabe, P., Pidcock, E., Rodriguez-Monge, L., Taylor, R., van de Streek, J. & Wood, P. A. (2008). J. Appl. Cryst. 41, 466–470. Web of Science CrossRef CAS IUCr Journals Google Scholar
Madsen, A. Ø. & Hoser, A. A. (2014). J. Appl. Cryst. 47, 2100–2104. Web of Science CrossRef CAS IUCr Journals Google Scholar
Nemkevich, A., Bürgi, H.-B., Spackman, M. A. & Corry, B. (2010). Phys. Chem. Chem. Phys. 12, 14916–14929. Web of Science CrossRef CAS PubMed Google Scholar
Neumann, M. A. (2008). J. Phys. Chem. B, 112, 9810–9829. Web of Science CrossRef PubMed CAS Google Scholar
Neumann, M. A. & van de Streek, J. (2018). Faraday Discuss. 211, 441–458. CrossRef CAS Google Scholar
Painter, J. & Merritt, E. A. (2006a). J. Appl. Cryst. 39, 109–111. Web of Science CrossRef CAS IUCr Journals Google Scholar
Painter, J. & Merritt, E. A. (2006b). Acta Cryst. D62, 439–450. Web of Science CrossRef CAS IUCr Journals Google Scholar
Panda, M. K., Runčevski, T., Chandra Sahoo, S., Belik, A. A., Nath, N. K., Dinnebier, R. E. & Naumov, P. (2014). Nat. Commun. 5, 4811. Web of Science CrossRef PubMed Google Scholar
Parrinello, M. & Rahman, A. (1981). J. Phys. 42, 511–515. Google Scholar
Pyzer-Knapp, E. O., Thompson, H. P. G. & Day, G. M. (2016). Acta Cryst. B72, 477–487. CrossRef IUCr Journals 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 Jr, R. A., 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
Sahoo, S. C., Panda, M. K., Nath, N. K. & Naumov, P. (2013). J. Am. Chem. Soc. 135, 12241–12251. Web of Science CrossRef CAS PubMed Google Scholar
Samoletov, A. A., Dettmann, C. P. & Chaplain, M. A. J. (2007). J. Stat. Phys. 128, 1321–1336. CrossRef Google Scholar
Schomaker, V. & Trueblood, K. N. (1968). Acta Cryst. B24, 63–76. CrossRef CAS IUCr Journals Web of Science Google Scholar
Skoko, Ž., Zamir, S., Naumov, P. & Bernstein, J. (2010). J. Am. Chem. Soc. 132, 14191–14202. Web of Science CSD CrossRef CAS PubMed Google Scholar
Stoe & Cie GmbH (2009). WinXPOW, Version 2.23. Darmstadt, Germany. Google Scholar
van de Streek, J. (2015). Acta Cryst. B71, 89–94. CrossRef IUCr Journals Google Scholar
Sun, H. (1998). J. Phys. Chem. B, 102, 7338–7364. Web of Science CrossRef CAS Google Scholar
Trask, A. V. & Jones, W. (2005). Organic Solid State Reactions. Topics in Current Chemistry: Crystal Engineering of Organic Cocrystals by the Solid-State Grinding Approach, edited by F. Toda, pp. 254. Berlin/Heidelberg: Springer. Google Scholar
Trask, A. V., van de Streek, J., Motherwell, D. W. S. & Jones, W. (2005). Cryst. Growth Des. 5, 2233–2241. CrossRef CAS 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.