The diffraction pattern of phase I at 278 K is shown in Fig. 1. In the whole angular range, less than thirty Bragg peaks, with most often very low intensities, are observed. Therefore, we expect the existence of a high symmetry crystallographic system. To determine the lattice parameters, twenty Bragg peaks in the angular range 10–60° (2 q) are taken into account in the program N-Treor (Altomare et al., 2000). All the reflections are indexed with an hexagonal cell with the following parameters: a=14.9404, c=6.8997 Å, V=1333.7 Å3. The calculated figures of merit (de Wolff, 1968; Smith & Snyder, 1979) are: M(20)=76, F(20)=73(0.0064,43). Our lattice parameters are close to the values proposed by Edwards et al. (1997). The diffraction pattern from 10° to 70° (2q) was refined with the Le Bail method (Le Bail et al., 1988) of the program FullProf (Rodriguez-Carvajal, 2001; Roisnel & Rodriguez-Carvajal, 2002). A pseudo-Voigt function, linear combination of a Lorentzian and a Gaussian of the same FWHM, was used to fit the Bragg peaks. This FWHM has a ? dependence according to the Caglioti law (Caglioti et al., 1958). The asymmetry of the reflections was taken into account according to the Berar and Baldinozzi function (Bérar & Baldinozzi, 1993). The background was determined with a linear interpolation between 29 points regularly distributed from 10° to 70°. The 39 refined parameters are as follows: the lattice parameters a and c, the zero-shift, the Caglioti profile parameters U, V, W, the mixing parameter ?? of the pseudo-Voigt function and its 2q-dependence X, 2 parameters for the asymmetry of the Bragg peaks and 29 points to define the background. The best estimated space group was determined with the help of the program CHEKCELL (Laugier & Bochu, 2001). It corresponds to the space group having the maximum checked reflections and the minimum calculated ones. The best space groups were found to be R3c and R-3c. It is worthy to notice that any of them corresponds to the space group R3 which is proposed by Edwards et al. (1997). Consequently, Le Bail refinements were performed successively to test the two space groups R3 and R3c. The reliability factors obtained with these two groups are close. However, a careful visual inspection of the diffraction pattern and a detailed examination of the observed intensities provided by the refinements, show systematic extinctions due to the existence of a c-glide plane; i.e. the reflections (h0l), (0kl), (00 l) with l odd, are not observable. These considerations lead unambiguously to the possible space groups R3c and R-3c. At the end of the Le Bail refinements with our space groups, the conventional reliability factors are: Rp=0.0717, Rwp=0.0632, Rexp=0.0172 and the values of the lattice parameters are: a=14.9376 (3), c=6.8979 (1) Å, V=1332.93 (5) Å3. The widths of the Bragg peaks, close to the experimental resolution given by the NAC, do not allow a microstructural analysis (size of crystallites and micro-strain effects). The density of caffeine, close to 1.45 g/cm3 (Edwards et al., 1997) leads to an unit cell with 6 molecules. There are 18 equivalent positions by symmetry for the space group R3c and 36 for the group R-3c. This suggests the existence of an orientational disorder. The space group R-3c is discarded because of the presence of the inversion centre. This would mean the possibility of a turn over of the plane of the molecule. This eventuality appears unlikely and thus, the dynamical disorder consists of a rotation of the molecule in its own plane. Therefore, the Rietveld refinements were achieved with the space group R3c. On each molecular site, a molecule can adopt three preferential positions, each of them with an occupation factor equal to 1/3. The dynamical origin of the molecular disorder is confirmed by Dielectric Relaxation Spectroscopy. Fig. 2 shows the isochronal temperature dependence of the imaginary part of the dielectric permittivity when the sample is cooled from 425 K (phase I) to 220 K (metastable state, see below) for four representative frequencies. It is clearly evidenced the existence of relaxational motions which are slower as the temperature is lowered. This indicates the thermally activated nature of this relaxation phenomenon. The dipolar relaxation time, close to 10–4 s at 400 K is increased to 1 s at room temperature. It reaches 100 s at Tg ? 250 K (Descamps et al., 2004). Below Tg, the dynamical disorder associates to the molecular rotations is frozen out and caffeine is entered in a so called glassy crystal state. Given the weak number of reflections (hkl) (65 reflections in the range 10–70°) and the number of atomic coordinates to refine (14 non-H atoms x 3 coordinates), the molecule is taken as a rigid-body and its geometry is built from the atomic coordinates of monohydrate caffeine (Edwards et al., 1997), anhydrous theophylline (Ebisuzaki et al., 1997), 1,7,9-trimethylpurine-2,6-dione monohydrate (Parvez, 1994) and from the complex of sulphacetamide-caffeine (Leger et al., 1977). Fig. 3 shows a representation of the molecule drawn with ORTEP-3 (Farrugia, 1997) and the numbering of the atoms. The bond lengths are reported in Table 1. The bond angles have been chosen regular, the internal angles of the pyrimidine ring being equal to 120° and those of the imidazole ring equal to 108°. The position of the molecule in the asymmetric unit is thus defined by the determination of 3 Euler angles ?, ?, ? which characterize the orientation of a orthonormal molecular system with respect to the orthonormal crystallographic system. Three parameters x0, y0, z0 represent the reduced coordinates of the centre of the rigid-body (C4 atom) in the crystallographic system. The space group R3c imposes the coordinate z0 to be fixed. When plastic crystals are made up of nearly spherical molecules (CBr4, C60, adamantane, for example), the dynamical disorder consists in quasi-free rotations of the molecule around their centre of mass while the translational order persists. For globular molecules, the plastic phase is often of cubic symmetry. When the molecules are flat, like for benzene derivatives, only axial rotations are released and the rotator phase is hexagonal (Brand et al., 2002, and references therein). In that case of molecules with a quasi-circular disk structure, the rotation is easier around the axes perpendicular to the disk and the molecules preferentially lie in the hexagonal plane. Therefore, for the first Rietveld refinements, the plane of the molecule was kept parallel to the hexagonal plane by fixing the angles ? and ? at right values and its centre of mass was placed on the three fold axis. In the final refinements, there are 47 ajustable parameters: − 10 profile parameters: the lattice parameters a and c, the zero-point, U, V, W, ?0 and X defined above, and two asymmetry parameters. − 8 structural parameters: the scale factor, the parameters which define the molecular orientation ?, ?, ?, x0 and y0, a parameter G1 linked to preferred orientations (March, 1932; Dollase, 1986) which were found along [0,0,1], and a global isotropic temperature factor Bov. − 29 points to define the background. The final conventional Rietveld factors are: Rp=0.0911, Rwp=0.0862, Rexp=0.0178. The experimental and calculated diffraction patterns are shown in Fig. 1. Crystallographic data, profile and structural parameters are given in Table 2 and reduced coordinates for non-H atoms in Table 3. The parameter G1=0.836 (2) for the [001] preferred direction corresponds to a slightly needle-like habit of the crystallites. The overall temperature factor Bov=3.25 (18) Å2 is consistent with the existence of a disordered phase. The plane of the molecules is practically parallel to the hexagonal plane, as shown in Fig. 4a. The angle (C5—C4, c) is found to be equal to 91.28 (1)° and the angle (C2—C4, c) equal to 96.56 (1)°. The superposition of the three preferential orientations of the molecule is shown in Fig. 4 b and an example of equilibrium positions in this hexagonal plane is given in Fig. 5. Given the dynamical disorder, the existence of an ordered network of H-bonds is not conceivable. A further attempt to refine the individual coordinates of the non-H atoms with soft-restraints failed, as expected and explained above. The molecule becomes unrealistically distorted and non-plane. The two following structural models: molecular conformation from monohydrate caffeine (Edwards et al., 1997) and from the complex of sulfacetamide-caffeine (Leger et al., 1977), were also tested independently. The Rietveld refinements give similar results (as an indication, Rp = 0.0945, Rwp = 0.0879 in the former case and Rp = 0.0929, Rwp = 0.0893 in the latter case) which can be compared to those obtained with our average molecule (Rp = 0.0911, Rwp = 0.0862). It appears that these close results are a consequence of the weak dispersion of the bond lengths and bond angles in these substances.
Caffeine (1,3,7-trimethylpurine-2,6-dione, C8H10N4O2) is a common agrochemical and therapeutic agent. It is known to occur in a hydrated form and two anhydrous polymorphic varieties. The crystal structure of the hydrated form has been determined a long time ago (Sutor, 1958) and confirmed recently (Edwards et al., 1997). However, it is discussed of the existence of a fully ordered crystal phase for anhydrous caffeine (Griesser et al., 1999; Carlucci & Gavezzotti, 2004). The physical characterization of the anhydrous states is really challenging because of the apparent impossibility to produce monophased caffeine. Depending on the preparation and the history of the sample, it is found a mixing of different forms (Müller & Griesser, 2003). The commercial form (designed by II in the following) transforms upon heating to the form I. Cesàro & Strarec (1980) have evidenced a phase transformation at before the melting at and have suggested, on the base of calorimetric studies, that the high temperature phase I is structurally disordered. The crystal structure of the phase I has not yet been solved. From powder X-ray diffraction measurements, different propositions of rhombohedral space groups have been proposed for which the molecules should be linked by very weak H-bonds to form an ordered network (Edwards et al., 1997). Stowasser & Lehmann (2002) suggest a monoclinic cell containing 20 molecules in a volume of 4450 Å3. In this paper, we report the ab initio structure determination of the high temperature phase I of caffeine from powder X-ray diffraction experiments. We show that this phase can be maintained at low temperature in a metastable state. A deep undercooling of the phase I, below 250 K, transforms the metastable state into a glassy crystal state.