Studying weak interactions in crystals at high pressures: when hardware matters

The quality of structural models for 1,2,4,5-tetrabromobenzene (TBB), based on data collected from a single crystal in a diamond anvil cell at 0.4 GPa in situ using two different diffractometers belonging to different generations have been compared, alongside the effects of applying different data-processing strategies.


Introduction
High-pressure data are widely used for the study of intermolecular interactions in crystals. In particular, high pressure can probe interactions and their role in stabilizing structures and their evolution across a variety of structural transformations: anisotropic structural distortion, polymorphic transitions and chemical reactions (Katrusiak, 1991;Boldyreva, 2008;Resnati et al., 2015;Yan et al., 2018;Parois et al., 2010). The quality of diffraction data [particularly completeness and the F 2 /(F 2 ) ratio] is critically important for obtaining reliable information on molecular conformations, intermolecular distances and even electron charge-density distribution (Veciana et al., 2018;Casati et al., 2017Casati et al., , 2016. Really impressive progress has been achieved over the last decade in obtaining more precise structural data from molecular crystal structures of increasing complexity. The improvements are related, first of all, to a new design of diamond anvil cells (DACs) with larger opening angles (Sowa & Ahsbahs, 2006;Ahsbahs, 2004;Boehler, 2006;Moggach et al., 2008). The improvements also include the use of 2D detectors instead of point detectors (Ahsbahs, 2004;Dubrovinsky et al., 2010;Kantor et al., 2012;Dawson et al., 2004), as well as applying new software for sample centering, absorption correction, recognizing and excluding unwanted reflections that do not belong to the sample, data reduction, and finding the orientation matrices for several crystallites in the same diamond anvil cell (Boldyreva et al., 2016;Katrusiak, 2008Katrusiak, , 2004Dera et al., 2013;Casati et al., 2007;Angel & Gonzalez-Platas, 2013). Special methods of data processing, in addition to precise experiments, now even make it possible to obtain data for charge-density studies (Veciana et al., 2018;Casati et al., 2017Casati et al., , 2016, and to follow related changes with pressure. This has been demonstrated for example by following the reduction in aromaticity of syn-1, 6:8,13-biscarbonyl[14]annulene on compression (Casati et al., 2016). Advances in the quality of high-pressure data for molecular crystals have often been related to the use of synchrotron radiation. However, with ISSN 2056-9890 limited access to synchrotrons, in-house experiments remain the most common type of high-pressure experiments for organic solids.
A new generation of laboratory diffractometers has been developed recently that makes it possible to collect data at high pressures from even small and weakly diffracting crystals. In this contribution, we present the results of a comparison of the data collected using two different diffractometers that were manufactured by the same company within a 10 year interval (Fig. 1). The first is an XtaLAB Synergy-S Dualflex diffractometer with Ag K radiation (PhotonJet-S source) and Pilatus3 X CdTe 300K hybrid photon-counting (HPC) detector from Dectris that was manufactured by Rigaku Oxford Diffraction in 2017, while the second is an Oxford Diffraction Gemini R Ultra diffractometer with Mo K radiation (Enhance X-ray source) and Ruby charge-coupled device (CCD) detector, manufactured by Oxford Diffraction in 2007. The main parameters characterizing the two instruments are compared in Table 1. We have collected data on the two different instruments from the same sample at the same pressure in the same DAC. We have also compared the results of applying different strategies for the data reduction.
As a sample we selected single crystals of 1,2,4,5-tetrabromobenzene (TBB). TBB is a well-known thermosalient compound, which exhibits large, spontaneous mechanical response across the phase transition on heating (Sahoo et al., 2013;Zakharov et al., 2018 and references therein). It has been shown recently that data on the high-pressure behaviour of such materials can be helpful in order to understand the origin of the thermosalient effect (Zakharov et al., 2017). TBB crystallizes in the monoclinic space group P2 1 /n. Being a thermosalient material, it shows a significant mechanical response, even though the phase transition on heating is accompanied by only minute rearrangements at the molecular level and only minimal changes in the intermolecular contacts (Sahoo et al., 2013;Zakharov et al., 2018). This makes it important to have high-quality structural data at multiple pressure and temperature (PT) conditions when studying the role of the intermolecular interactions in the thermosalient effect. High noise level, low data completeness, low F 2 /(F 2 ) and data-to-number of parameters ratios can lead to the loss of most of the information related to the electron-density distribution in the crystal. When using 'older-generation' inhouse diffractometers, low data quality can make it impossible to refine the crystal structure in even an isotropic approximation. Therefore, fine details in the orientation of anisotropic displacement parameters (ADPs) and precise values for the interatomic distance changes, which are of great impor- Cabinet view of diffactometers used: (a) XtaLAB Synergy-S Dualflex; (b) Oxford Diffraction Gemini R Ultra.
tance for studying the mechanical response of the crystal to variations in PT conditions, will not be accessible. The newgeneration instruments are expected to improve the quality of the diffraction data and the structural models based on the refinement of these data. At the same time, using a newer instrument alone does not guarantee a high-quality structural model. The data-processing strategy is critically important for data collected from a sample in a DAC at high pressure (Boldyreva et al., 2016;Katrusiak, 2008Katrusiak, , 2004Dera et al., 2013;Casati et al., 2007;Angel & Gonzalez-Platas, 2013). These data are inevitably 'contaminated' by absorption of X-rays by the materials of the DAC (diamond, metal) and reflections originating from diffraction of the diamonds, gasket or the ruby calibrant. The presence of these reflections also corrupts the measured intensities of the sample reflections, either by direct overlap or because they may have an influence on the estimated background level. The aim of this study was to compare the data quality collected from the same sample in a DAC at high pressure in situ using diffractometers belonging to different generations. For data collected using both of the two instruments, we have used several different strategies for the data processing. The aim of this was to test the relative importance of applying different techniques for correction of the raw data for increasing the reliability and improving the quality of the structural model.
The sample was mounted in an Almax Boehler DAC (Boehler, 2006). A stainless steel sheet with an initial thickness of 200 mm was pre-indented to 100 mm and used as a gasket. The ruby fluorescence method was used for pressure calibration (Forman et al., 1972;Piermarini et al., 1975). A methanolethanol mixture (4:1) was used as hydrostatic pressure-transmitting medium (Piermarini et al., 1973;Angel et al., 2007).
Single-crystal X-ray diffraction data were collected on the same crystal in the same DAC at a hydrostatic pressure of 0.4 GPa. Data were collected using two different instruments: (1) an XtaLAB Synergy-S Dualflex diffractometer with Ag K radiation (PhotonJet-S source) and Pilatus3 X CdTe 300K HPC detector from Dectris (manufactured by Rigaku Oxford Diffraction in 2017), and (2) an Oxford Diffraction Gemini R Ultra diffractometer with Mo K radiation (Enhance X-ray source) and Ruby CCD detector (manufactured by Oxford Diffraction in 2007). Data collection, cell refinement and data reduction were performed using CrysAlis PRO software (Rigaku OD, 2016). Multiple strategies were tried on each instrument. Some of the strategies deliberately neglected good-practice techniques of introducing certain high-pressure data corrections in order to evaluate the extent to which this neglect can worsen the data quality.
For data collection (1), X-ray diffraction data were treated and attempts were made to refine the structure in three different ways: (a) Gaussian absorption correction using ABSORB-7 (Angel & Gonzalez-Platas, 2013) implemented in CrysAlis PRO software (Rigaku OD, 2016). Both crystal and DAC absorption were taken into account. The most disagreeable reflections from the sample that overlapped with diamond and gasket reflections were not excluded from the HKL file. All non-H atoms were refined anisotropically.
(b) Gaussian absorption correction using ABSORB-7 (Angel & Gonzalez-Platas, 2013) implemented in CrysAlis PRO software (Rigaku OD, 2016). Both crystal and DAC absorption were taken into account. The most disagreeable reflections from the sample that overlapped with diamond and gasket reflections were excluded manually from the HKL file. All non-H atoms were refined anisotropically.
(c) Spherical absorption correction as implemented in CrysAlis PRO software (Rigaku OD, 2016). Only crystal absorption was taken into account. The most disagreeable reflections from the sample that overlapped with diamond and gasket reflections were manually excluded from the HKL file. All non-H atoms were refined anisotropically.
For data collection (2), X-ray diffraction data were treated and attempts were made to refine in six different ways: (d) the same as for (a).
(e) the same as for (b).
(f) the same as for (c).
(g) the same as for (a), but carbon atoms were refined isotropically.
(h) the same as for (b), but carbon atoms were refined isotropically.
(i) the same as for (c), but carbon atoms were refined isotropically.
For all the refinements at high pressure, the initial crystal structure model was taken from single-crystal diffraction data at ambient conditions (Zakharov et al., 2018). Refinements were carried out with SHELXL2018/1 (Sheldrick, 2015) using Shelxle (Hü bschle et al., 2011) as the GUI without any restraints. Hydrogen-atom parameters were constrained using AFIX 43 with U iso (H) = 1.2U eq (C). Mercury (Macrae et al., 2008), checkCIF/PLATON (Spek, 2009) and publCIF (Westrip, 2010) were used for structure visualization, analysis and preparation of the CIF files for publication.

Results and discussion
Crystal data, data collection and refinement parameters are summarized in Table 2. In comparison with the older Gemini R Ultra device, used for data collection (2), the Synergy-S    diffractometer, used for data collection (1), was superior for data collection. Compared to instrument (2), collection of single-crystal X-ray data on (1) was much faster (6 vs 32 h), with a higher F 2 /(F 2 ) ratio (18 vs. 10) and data completeness (68 vs 58%). A higher HKL range allowed us to increase the number of reflections used for cell-parameter refinement by a factor of 1.5. The resulting values of the lattice parameters appear to be almost the same in the two cases: the largest difference, 0.2%, was observed for lattice parameter b. Standard uncertainties for the cell parameters were slightly higher for (1) than for (2). This is presumably related to the smaller 2 values for stronger reflections owing to the use of the harder Ag K radiation. Shorter wavelengths are generally prefered for samples mounted in a DAC with a fixed window-opening size. From a data completeness point of view, this provides the same number of reflections in a narrower 2 range. Ag K radiation is therefore becoming popular for high-pressure X-ray diffraction studies (Saouane et al., 2013;Saouane & Fabbiani, 2015;Granero-García et al., 2017). The number of independent reflections for data collection (1) was 1.6 times greater than for (2) (893 vs 550), as a result of using a shorter wavelength. The more efficient HPC detector and the brighter X-ray source allowed us to measure reflection intensities with higher precision. This gave us a twofold lower R int value for data collection (1): 0.048 for data set (b) vs 0.105 for data sets (e) and (h). Displacement ellipsoid plots for the different methods of data treatment and refinement are shown in Fig. 2. Taking into account the refinement data presented in Table 2, one can conclude that the best results are provided by refinements (b) and (c), where the use of a modern device permitted a more precise and faster measurement of the intensities of the diffraction reflections. The quality of the diffraction data enabled a crystal-structure refinement in the anisotropic approximation for all non-H atoms, providing reasonable values and shapes of the displacement ellipsoids. For the refinement variant (a), for which the sample reflections that overlapped with diamond and gasket reflections were not excluded from the HKL file, the refinement did not converge, and when an anisotropic refinement was attempted a nonpositive-definite atomic displacement ellipsoid was obtained for one of the carbon atoms.
For data collection (2), the refinement results were of much lower quality than those for data collection (1). As expected, the worst results were provided by refinements (d)     Displacement ellipsoid plots for 1,2,4,5-tetrabromobenzene molecules obtained with different data-treatment procedures. Carbon atoms for structure refinements (g), (h) and (i) were refined using an isotropic approximation. Cubes show atoms with negative thermal parameters. Refinements marked V are preferable for publication; those marked W are publishable but not always preferable, and those marked X are not acceptable for publication (incorrect).
HKL file. The refinement did not converge, and two of the carbon atoms were characterized by non-positive-definite ellipsoids when attempting to use an anisotropic model. Removal of the corrupted reflections from the HKL file did not improve refinement results. The anisotropic thermal parameters were still not adequate for the (e) and (f) refinements. Publishable refinement results in this case of impossible anisotropic refinement could be obtained in two ways: viz. by applying SHELX restraints for the thermal parameters of carbon atoms, e.g. SIMU and DELU, with low standard uncertainty values, or by refining the carbon atoms in an isotropic approximation, as was done for the (h) and (i) refinements.
Different absorption correction types were tested for both data collection strategies. The refinement results provided by the Gaussian and spherical absorption corrections are defined as (b) and (c), (e) and (f), (h) and (i), respectively. One can see that the R-factors are comparable and acceptable for both absorption-correction strategies. A potential explanation for the similarity of the Gaussian and spherical absorption correction results for data collection (1) rests in the fact that TBB is a medium-absorbing sample ( is 10.33 mm À1 for Ag K). In the case of data collection (2), TBB is much more absorbing ( is 19.29 mm À1 for Mo K radiation) but the overall data quality is low (intensities are not measured precisely) and even the good-practice procedure of applying an absorption correction does not improve data quality. Generally, it is preferable to use a Gaussian absorption correction (both for the crystal and for the DAC), especially for strongly absorbing samples since it calculates the 'true' transmission factors using the actual crystal and DAC geometries. For example, data sets (b) and (h), and (e) in the case of reasonable anisotropic thermal displacement parameters, would be the most preferable for the experimental setup described.

Conclusions
In order to obtain reliable information on intermolecular interactions in a crystal structure, one needs high-quality data. This is especially critical for data collected in a DAC at high pressure, when data completeness and the availability of reciprocal space are limited. A comparison of the results obtained using different instruments and different dataprocessing methods has illustrated that the data processing itself plays a crucial role in obtaining reliable results. At the same time, a modern instrument belonging to the new generation makes it possible to speed up data collection, increase the signal-to-noise intensity ratio and the number of observed reflections, and with shorter wavelength data completeness for a sample mounted in a DAC. Data collection for the 1,2,4,5-tetrabromobenzene crystal mounted in a DAC using a modern XtaLAB Synergy-S Dualflex diffractometer with Ag K radiation and a Pilatus3 X CdTe 300K HPC detector took six hours, and allowed us to obtain high-quality data for an anisotropic crystal-structure refinement without any restraints.
Using the older diffractometer from the previous generation, an Oxford Diffraction Gemini R Ultra with Mo K radiation and a Ruby CCD detector, did not allow us to obtain diffraction data of the same quality, even when using a higher exposure time, for which data collection took 32 h; the anisotropic refinement was possible only for the heavier bromine atoms. The carbon atoms could be refined reasonably only in an isotropic approximation, or by restraining their thermal parameters. Data completeness, HKL ranges and the F 2 /(F 2 ) ratio were lower, and the R-factors were higher compared to the values obtained when using the modern XtaLAB Synergy-S Dualflex diffractometer described above.
Crystal-structure refinement using the same primary data set, but different data-reduction strategies has revealed that eliminating the sample reflections with wrong intensities (affected by the presence of diamond, as well as powderdiffraction rings originating from the metal gasket) is the most important correction of primary data. The exact procedure for the absorption correction was less critical in the particular case considered in this work. However, generally and especially for strong absorbers, a Gaussian absorption correction both for the crystal and the DAC data can help to increase the quality of the refinement significantly, since it calculates the 'true' transmission factors using the actual crystal and DAC geometries.

1,2,4,5-Tetrabromobenzene (Ag-Absorb7-raw_a)
Crystal data where P = (F o 2 + 2F c 2 )/3 (Δ/σ) max = 0.014 Δρ max = 1.55 e Å −3 Δρ min = −1.48 e Å −3 Special details Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.

Fractional atomic coordinates and isotropic or equivalent isotropic displacement parameters (Å 2 )
x y z U iso */U eq where P = (F o 2 + 2F c 2 )/3 (Δ/σ) max = 0.001 Δρ max = 0.54 e Å −3 Δρ min = −0.54 e Å −3 Special details Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.

1,2,4,5-tetrabromobenzene (Ag-CAsphere_c)
Crystal data where P = (F o 2 + 2F c 2 )/3 (Δ/σ) max < 0.001 Δρ max = 0.53 e Å −3 Δρ min = −0.49 e Å −3 Special details Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.

Fractional atomic coordinates and isotropic or equivalent isotropic displacement parameters (Å 2 )
x y z U iso */U eq Br1 0.58346 (14) 0.79099 (6) 0.59430 (7) 0.0412 (2)   where P = (F o 2 + 2F c 2 )/3 (Δ/σ) max = 0.089 Δρ max = 2.65 e Å −3 Δρ min = −2.89 e Å −3 Special details Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.

1,2,4,5-tetrabromobenzene (Mo-Absorb7_e)
Crystal data Special details Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.
Fractional atomic coordinates and isotropic or equivalent isotropic displacement parameters (Å 2 ) where P = (F o 2 + 2F c 2 )/3 (Δ/σ) max = 0.523 Δρ max = 0.93 e Å −3 Δρ min = −0.83 e Å −3 Special details Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.

1,2,4,5-tetrabromobenzene (Mo-Absorb7-raw-Ciso_g)
Crystal data Special details Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.
Fractional atomic coordinates and isotropic or equivalent isotropic displacement parameters (Å 2 ) x y z U iso */U eq Br1 0.5856 (7)   where P = (F o 2 + 2F c 2 )/3 (Δ/σ) max < 0.001 Δρ max = 1.03 e Å −3 Δρ min = −0.88 e Å −3 Special details Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.

Fractional atomic coordinates and isotropic or equivalent isotropic displacement parameters (Å 2 )
x y z U iso */U eq Br1 0.5856 (4) 0.79134 (18)   Special details Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.