Cs(Be2Li)Al2Si6O18, a cesium-stuffed host-guest structure, and its structure-property variations with temperature and pressure

The solid-state structure of (Cs,Na)[(Be2Li)Al2Si6O18], a beryl-type material, was studied in-situ on single crystal and powder samples at hydrostatic high pressure conditions (P 9.8 GPa) as well as under temperature variation (83 T 1173 K) by means of Raman spectroscopy, X-ray and synchrotron diffraction. The investigations show an inconspicuous, easily overlooked transformation from R3c to non-centrosymmetric R3c space-group symmetry on increasing pressure (Pc ~ 4 GPa) and decreasing temperature (Tc 1⁄4 312 5 K), as revealed by Raman band splitting as well as discontinuities in the evolution of Raman spectra, thermal expansivity coefficients and anomalous compression behavior. Distortions to acentric symmetry originate from the Cs and Na guest atoms inside the channels parallel the c-axis. The occurrence of diffuse scattering planes in all measured data sets indicates substitutional defects inside the channels following an approximate ~4.6 Å periodicity, without any remarkable ordering between adjacent channels. The determined thermal expansivity follows an inverse behavior compared to other beryl-type structures for the R3c phase. The bulk moduli for the two polymorphs compare, together with the observed compressional anisotropy, to other complex beryl-type framework structures. However, the symmetry reduction at high pressure leads to independent displacements of the framework oxygen positions, and as a consequence, to the increase of the coordination number from [6 þ 6] to [6 þ 9] for one of the


Introduction
The solid-state properties of microporous framework structures are in the focus of material science since the porosity on a molecular level provides possibilities for various applications, which include functions as microsieves, ion exchangers, catalytic solids, ion and molecule cages or being host materials for nuclear waste disposal e.g. Ref. [1][2][3][4][5]. Apart from physicochemical properties related to the mobility of the extra-framework components, it is the stability of the host framework itself and its resistance to changing conditions that determines the material capability. The beryl-type framework [4] T1 3 [6] M 2 [4] T2 6 O 18 (with e.g. T1 ¼ Be, Li; M ¼ Al, Fe, Mg, Sc; T2 ¼ Si, Al) is an outstanding structure because on one hand it does not represent a typical holotetrahedral framework, and on the other hand it has a pronounced one-dimensional microporosity. Representing a heterodesmic (strong, covalent bonds) framework, consisting of both tetrahedral [4] T and octahedral [6] M building units, this host structure reveals separated parallel channels, in a strictly one-dimensional channel system (Fig. S1, supplement), and therefore has high capabilities for the fixation of extra-framework guest molecules and suitably sized guest atoms, such as here of cesium ions.
The beryl-type framework structure is represented in nature by several (alumino)silicate compounds, for example by beryl (Be 3 Al 2-Si 6 O 18 , space group P6/mmc, [6]), bazzite (Be 3 Sc 2 Si 6 O 18 , [7]), stoppaniite (Be 3 Fe 2 Si 6 O 18 , [8]) and cordierite (Al 3 (Mg,Fe) 2 AlSi 5 O 18 , [9][10][11]). The compounds are known to host inside these pores considerable fractions of M þ alkali cations and various molecular volatiles (e.g. H 2 O, CO 2 , or N 2 ) [1,12]. Most recently, the series of recognized beryl-type □ 1-x M þ x [ [4] O 18 ], which is known even in nature as the gemstone mineral pezzottaite [13][14][15][16][17][18][19]. The crystallographic investigations [16,18,19] revealed a rhombohedral superstructure in the space group R3c, apparently originating from ordering between Li and Be cations on T1 sites. Moreover, the channels inside the stacked six-membered Si 6 O 18 ring units show the highest M þ content ever located in one of the isostructural compounds, which in case of (Cs,Na)[(Be 2 Li)Al 2 Si 6 O 18 ] is required for the charge balance as originating from the distribution of~1/3 Li þ and 2/3 Be 2þ on the individual T1 sites. Natural samples have been reported often to exhibit deficiency of M þ guest atoms, which is stoichiometrically compensated by the (Be, Li) substitution on the [4] T1 site, according to the formula □ 1-x M þ x [(Be 3-x Li x )Al 2 Si 6 O 18 ] [16]. As, in particular, the stuffed channels in Cs[(Be 2 Li)Al 2 Si 6 O 18 ] and the fractions of extra-framework guest atoms inside this open-structure is unique among the beryl-type compounds, the behavior of this material at non-ambient temperature (T) and pressure (P) conditions, and the role played by the channel constituent, was getting into the focus of our interest. Studies on the framework structure of Al 3 (Mg,Fe) 2 AlSi 5 O 18 revealed anomalous compression under hydrostatic pressure. In detail, an unusual linear volume compressibility was found, while the slope for the interval at low pressure was evolving in a (normal) positive fashion indicating slightly increasing stiffening, the evolution of bulk modulus for P > 0. 35 GPa followed a constant line. The same was found for the band c-axis, while the a-axis was the less compressible axis and showed conventional stiffening with pressure. For cordierite it was connected to an elastic softening phenomena, as a precursor effect ahead of a supposed structural transition around 6.9 GPa [10,20]. Al 3 (Mg,Fe) 2 AlSi 5 O 18 , a technically prominent ceramic component [21], reveals, among a low thermal expansivity and high resistance to thermal shocks, a pronounced structural stability at high-T conditions, although undergoing an order-disorder transition at high T. This structure type appears to be little influenced in terms of its stability by high degree of energy deposition, neither through high T nor through high irradiation doses [9,11], and is therefore of particular interest as a potential storage material for nuclear waste. Cs[(Be 2 Li)Al 2 Si 6 O 18 ] can contain up to 16 wt% of Cs 2 O. In continuation of investigations of structure-property relations, and associated structural phase transitions, on beryl-type analogue phases, we intended to study the elastic behavior of Cs[(Be 2 Li)Al 2 Si 6 O 18 ] under T and P and its deformation mechanisms at the atomic scale, by in-situ measurements. The results of in-situ investigations, by means of X-ray diffraction and Raman spectroscopy on T variations between 83 K and 1173 K (at 1 bar) and isothermal compression up to 9.8 GPa under hydrostatic P conditions at T ¼ 298 K, are presented in this study.

Sample material and sample environment
Single crystals and powder samples were obtained from the same gemstone material from the Sakavalana pegmatite (Madagascar) used by Ref. [18,19], which was structurally and chemically well characterized. The average chemical formula of this sample was therefore determined to Ref. [19] μm in thickness), which was prepared perpendicular to the hk0-plane, parallel to the crystallographic c-axis after orientation through X-ray diffraction. All temperature-dependent single-crystal investigations were carried out on crystal fragments with a size around 140 Â 120 Â 40 μm.
Diffuse scattering was demonstrated on an approximately 260 Â 140 Â  190 μm larger crystal fragment glued to a Kapton loop. High-pressure conditions were achieved in ETH-type diamond anvils cells [22], using anvils with 600 μm culets, conventional steel gaskets with pressure chambers being 250 μm in diameter and a height of 100 μm, and using proper hydrostatic pressure-transmitting media (4:1 methanol-ethanol mixtures for XRD, pure dense argon for Raman spectroscopy). Pressure on isothermal (T ¼ 298 K) compression was calibrated using the established quartz and ruby pressure gauges [23][24][25] to within uncertainties less than AE 0.01 GPa (for quartz) and AE0.07 GPa (for ruby), at 9.13 GPa and if only based on the volume fit uncertainty of around 0.01% [e.g. AE 0.009 Å 3 at 96.780 Å 3 ], and on the absolute band position error of around 0.005% [e.g. AE 0.5 cm À1 at 14404.24 cm À1 ]), respectively.

In-situ Raman spectroscopy experiments
Raman spectra at high-P and at T variation between 83 and 873 K were collected by means of a confocal Horiba Jobin Yvon LabRAM-HR 800 spectrometer using a He-Nd laser source with a wavelength of 632.8 nm for P. For T measurements, a 532 nm constant wave semiconductor laser was used, while the laser power was reduced to 25%. The spectrometer is equipped with an Olympus BX41 microscope and its Olympus LMPIanFL N 50 Â objective with a long working distance of 10.6 mm, which enables it to focus the laser beam onto the sample crystal inside the DAC or cooling/ heating stage. A total of 61 þ 18 Raman spectra at T ranging from 83 to 468 K (Fig. 1a) and 83-873 K (Fig. 1b) were collected, along with a series of 11 high-pressure spectra between 1 bar and 9.8 GPa (Fig. 2), covering a spectral range from 30 to 1300 cm À1 Raman shift, at exposure times between 2 Â 60 and 3 Â 60 s per step for T-and 3 Â 40 s per step for Pexperiments. For all measurements, a diffraction grating with 1800 lines per mm was used. Spectra were acquired using the LabSpec 6 software (HORIBA Scientific) and Peakfit v.4 (Systat Software Inc.) software was used for background subtraction as well as for the determination of the band position. All Raman spectra were fitted with the Gauss-Lorentz-area method.

In-situ X-ray diffraction experiments
XRD intensity data collections were performed using a Stoe StadiVari diffractometer (equipped with a Dectris Pilatus 300 K detector with a 450 μm silicon layer, an air-cooled Incoatec IμS molybdenum microfocus tube operated at 50 kV/1 mA, beam diameter of~110 μm FWHM) and a 120 mm sample-detector distance. In-situ measurements at 473 K, 373 K, 298 K, and 173 K were performed using an Oxford CryoSystems nitrogengas Cryostream cooler, with resistive heating at a temperature stability of the N 2 flow of around 0.1 K on two single crystal samples. In-situ highpressure data were collected from a~150 Â 120 Â 45 μm sample crystal ( Fig. S2, supplement). XRD intensity data were collected at 2θ 92 , performing 0.5 ω-scan at 50 s of exposure time per frame. The diffraction patterns were indexed and intensities integrated using the X-area 1.72 software (Stoe & Cie GmbH). Integrated intensities were corrected for absorption effects, through DAC components and sample, by using the ABSORB code [26]. Crystal structures were refined using neutral scattering curves from the International Tables for Crystallography [27] with  (a) Pressure dependence Δν/ΔP of selected prominent Raman bands as obtained from the series of high-P Raman spectra, inclusive the determined slope between 0 and 5 GPa. (b) Differences in Raman shifts between the modes at 686 and 883 cm À1 , compared to that at 111 cm À1 , provide evidence for a transition between 5 and 6 GPa. SHELXL [28] and the graphical user interface ShelXle [29]. Details of the individual data collections are summarized in Table 1.
The unit-cell parameters and their variations with pressure were determined from measurements on a Stoe AED II diffractometer (Eulerian cradle, graphite-monochromatized Mo-radiation from 50 kV/30 mA sealed-tube source, point detector). Data acquisition and refinement was performed using the SINGLE software [30]. In addition to the sample crystal (~150 Â 120 Â 45 μm), a quartz single crystal (~160 Â 80 Â 40 μm) was added in the pressure-chamber of the DAC and its unit-cell parameters were measured for precise pressure determination. Applying the 8-position centring mode [31], both unconstrained and symmetry-constrained unit-cell parameters were determined for 12 pressure points in the range between 10 À4 GPa and 9.13 GPa at T ¼ 298 K (Table S1). Equation-of-state (EoS) parameters were fitted using the software EoSFit7GUI [32].

Synchrotron X-ray powder diffraction experiments
In-situ high-temperature powder X-ray diffraction data were collected at the MCX beamline at ELETTRA (Trieste, Italy) using a Huber 4-circle goniometer, focussed beam conditions, and double slit in front of a point detector. Powder samples were contained in a quartz-glass capillary. The use of a gas blower (Oxford Danfysik DGB-0002) allowed to control the temperature up to 1273 K (AE2 K). X-ray powder diffraction patterns were collected with a monochromatic beam of 0.7750 Å, and diffraction data were analysed by Le Bail method [33] to obtain precise unit-cell parameters in the temperature range between 298 and 1173 K. For the full-profile fit, the GSAS package [34,35] was used (profile function: pseudo-Voigt; background function: Chebyshev polynomial).  Table 2 Elastic properties and corresponding bulk moduli for the unit-cell volume V and the lattice parameters along directions a and c, as obtained from fitting thirdorder Birch-Murnaghan EoS and a thermal EoS after Berman (1988) to the respective T/P-V, T/P-a, and T/P-c data sets.  Table 2. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

In-situ Raman spectra under T and P variation
The vibrational properties of Cs[(Be 2 Li)Al 2 Si 6 O 18 ] have been investigated in detail by Refs. [19], who identified 44 A 1g þ 92 E g theoretically possible active Raman modes. The most characteristic Raman bands reported so far correspond to those observed within the wavenumber range 50-1300 cm À1 . In the following experiment we concentrated on 15 bands located at 112(s), 230, 265, 272, 293, 326(s), 403 þ 405, 416, 456, 485, 498, 685(s), 1058(s) and 1100(s) cm À1 at ambient conditions (sstrong Raman band, cf. Fig. 1). The modes below 300 cm À1 are attributed to rotation and breathing of the Si 6 O 18 ring [36,37]. The intense peaks at 403 þ 406 cm À1 may be ascribed to deformation of the ring units [38], while the strong peak at 685 cm À1 has been assigned to the stretching vibration of the ring unit, coupled to the Be-O stretching mode [19]. The Raman spectrum is dominated by a strong mode at 1100 cm À1 , which corresponds to the Si-O stretching mode, as pointed out by previous studies (cf. [19] and references therein).
The series of spectra under T variation ( Fig. 1) show no apparent changes with respect to the number, position and intensity of the observed peaks. It is evident that the Cs[(Be 2 Li)Al 2 Si 6 O 18 ] structure, in terms of arrangement of the building units and their local symmetries, remains substantially unchanged across the T-range investigated. The evolution of spectra does not provide any hint of a phase transition. The only clear changes concern a gradual broadening of the peaks (Fig. S3, Supplement) in accordance with increasing lattice dynamics, as can be seen from the increase in the FWHM's from~3 to 5 cm À1 at 83 K to~10 to 20 cm À1 at 873 K. If band positions are plotted (Fig. 1c), the typical redshift to be expected for increasing T can be observed. It is noteworthy, however, that two separate ranges of linear trends, with different slopes, can be traced for individual Raman bands, with a smooth and apparently continuous transition approaching room temperature (RT).
An additional series of spectra was collected on an isothermally compressed sample at RT under quasi-hydrostatic conditions using argon as the pressure-transmitting medium. The sequence of in total 11 highpressure Raman spectra, recorded up to 9.8 GPa, also shows no substantial changes, which would reflect structural modifications. The prominent Raman bands attributed to the vibrational modes of individual framework building units show no deviations from an almost linear blueshift trend. Deviations from linear trends are only discernible in the low-frequency range of the spectrum (Figs. 2 and 3), e.g. for the band at 111 cm À1 . This band shows a redshift up to~3 GPa, followed by a blueshift beyond~5.3 GPa, which is accompanied by a band splitting with a shoulder emerging on the high-energy side of the band. By evaluating the relative differences in Raman shifts, a critical pressure P c can be determined around 5.3 GPa. The changes at a pressure beyond P c involve even new bands, e.g. that one at~75 cm À1 (cf. Fig. 2). Since bands in this range of low energy vibrational and rotational modes can be attributed to the weak bonds of highly coordinated atoms, it can be assumed that these supposed changes might affect the arrangement of the (extra-framework) channel population.

Thermo-elastic behaviour
A series of 27 precise lattice parameters in hexagonal setting was determined from single-crystal X-ray diffraction between 83 and 500 K, in addition to 7 data points extracted from synchrotron powder data collected between 298 and 1173 K (Tables S2 and S3). The evolution of the unit-cell parameters with T confirmed that, within the T-range investigated, the rhombohedral cell, previously reported for pezzottaite at room conditions [18], is preserved. However, the variation of unit-cell edge lengths along [100] and [001] suggests a change at~308 K, where temperature dependencies exhibit different slopes (Fig. 4). Using a linear approximation for the data up to~300 K, or a simple extension after [39] to accommodate non-linear expansion for the high-T range, yield the coefficients as summarized in Table 2 (fits to the experimental data were done with EoSFit7GUI [32]). The negative thermal expansion coefficient along the a-axis direction (α ¼ À1.65 AE 0.47 Â 10 À6 K À1 at T ¼ 250 K) is remarkable for the low-T range, whereas for T > T c a clearly positive value for the thermal expansion coefficient (α ¼ 2.50 AE 0.12 Â 10 À6 K À1 at T ¼ 350 K) is obtained. A slight decrease in the values for α can be obtained along the c-axis direction at increasing T, though keeping pos- . The significant change observed along the direction parallel to (//) [100] also considerably affects the c/a ratio by increasing the slope from constantly around 7.7(8) Â 10 À7 K À1 between room temperature and 1000 K to about 1.0 Â 10 À5 K À1 between 100 K  and RT (Fig. 4). From the intersections of the respective empirically determined curves shown in Fig. 4, individual values of 311 AE 4 and 314 AE 5 K can be derived for the critical temperature T c . The discontinuity can also be seen in the evolution of the unit-cell volume, with a considerably smaller volume expansion below T c compared to that above T c ( Table 2). Beryl-type framework structures were found to have, in general, a low thermal expansivity (e.g. þ2.0 Â 10 À6 K À1 for polycrystalline cordierite ceramics [40]), usually with axial expansion along the c-axis showing small or even negative values {þ2.6 AE 0.1 Â 10 À6 K À1 //a and À2.9 AE 0.4 Â 10 À6 K À1 //c in Be 3 Al 2 Si 6 O 18 (beryl); 1.7 AE 0.1 Â 10 À6 K À1 //a and þ0.16 AE 0.60 Â 10 À6 K À1 //c Be 3 (Al,Cr 3þ ) 2 Si 6 O 18 (emerald) after [41]; þ3.3 Â 10 À6 K À1 //a, þ2.7 Â 10 À6 K À1 //b and À1.2 Â 10 À6 K À1 //c in Al 3 Mg 2 AlSi 5 O 18 (cordierite) after [42]}. The expansion of any given beryl-type framework depends on the T response of the framework itself involving T-O-T and O-T-O angular changes, the expansion coefficients of the non-tetrahedral bonds, and the interplay between these two factors [43]. The thermal expansivity of Cs[(Be 2 Li)Al 2 Si 6 O 18 ] above T c ( Table 2) is comparably low, being similar in isotypical framework materials, mentioned above. Nevertheless, the anisotropy of thermal expansion was found to be less pronounced as the two main components of the α ij tensor reveal positive values, at least for the T range between 300 and 973 K. A negative thermal expansion value has indeed been observed for the phase at T < 300 K, but it affects, the first time reported for any beryl-derived structures, the a-axis direction.

Isothermal compressibility and equation of state
The unit-cell parameters measured at different P are summarized in Table S1. The least-squares refinements of base vectors were performed in a first step with symmetry-unconstrained parameters in a triclinic setting, in order to determine possible pseudo-symmetries as due to symmetry breaking associated with any structural transition. The symmetry-unconstrained refinements confirmed the hexagonal setting of base vectors with a being equal to b within the experimental uncertainties, and the angles α, β, and γ showing values close to 90 and 120 , respectively.
The P dependency of unit-cell parameters (Fig. 5) reveals a compressional anisotropy similar to that reported for other beryl-type materials, with the structure being % 22% less compressible along the a-axis than along the c-axis [9][10][11]44,45]. A careful inspection of the P dependencies shows that both axial and volume compressibilities increase at elevated P, although at increasing density one can usually expect opposite trends. Fitting a Birch-Murnaghan EoS [46] to the experimental data ( Table 2, Fig. S4 in the supplement), the pressure derivatives of the moduli K (for volume) and M (axial) show negative values, i.e. M' ¼ dM/dP for a and c is À3.8 AE 0.8 and À2.8 AE 0.6, K' ¼ dK/dP for V is À1.17 AE 0.14. Axial negative values have been previously reported for irradiated and non-irradiated cordierite [9][10][11], while a study on beryl itself did not report any anomalous behaviour [44]. Structural instabilities have been reported for isostructural frameworks suggesting in Be 3 Al 2 Si 6 O 18 and Al 3 (Mg,Fe) 2 AlSi 5 O 18 transformations to intensively twinned or gently modulated polymorphs of lower symmetry [9,10,20,45]. Thus, the remarkable elastic-softening behaviour in Cs [(Be 2 Li)Al 2 Si 6 O 18 ] can be interpreted as a precursor of an impending polymorphic transition at P beyond those achieved in this study.
The fit with a third-order Birch-Murnaghan EoS to the volume data across the entire P range (Fig. S4, supplement) provides a bulk modulus Fig. 8. Arrangement of Na and Cs sites within a single channel along the c-axis (between z ¼ 0 and z ¼ 1). Electron densities correspond to the maps calculated from the data collected at 173 K. Na1, Na2A, Na2B, Cs1 and Cs2 positions correspond to the refined site. Cs1* and Cs2* are the theoretical central positions on the triad; Na2* is the mean position that splits into Na2A and Na2B. The presumable Ow positions (i.e. Ow1 and Ow2) are necessary for the completion of the Na coordination, but their positions superimpose with Cs sites and have not been resolved. It is remarkable that the atom distribution follows multiples n of 1/12 along the z-coordinate inside the channels.  Fig. S5, supplement), as calculated from the differences between the experimental unit-cell volume and the theoretical one derived from the EoS fit, gives a hint for deviations in the proximity of 3-4 GPa. Deviations ΔV reach almost~3 Å 3 , compared to 0.5 to 0.9 Å 3 at P < 3 GPa. While the residuals for the EoS fit across the entire pressure range (0-9 GPa) might be not that suspicious, the truncation of the pressure-range fit at the critical pressure between 3 and 4 GPa provides a much clearer picture (Fig. 5). Although the EoS fits for the 0-3 GPa range can not be performed with unconstrained pressure derivatives for the bulk moduli, the obtained results are realistic (Table 2), and provide a sufficient explanation for the earlier discussed negative vales for K 0 and M'. The occurrence of an apparently continuous structural change, which can be interpreted as a weak second-order transformation, is also supported by the axial compressional trends along [100] and [001] and in particular by the c/a ratio (Fig. 6).

Structure evolution at non-ambient conditions
All the X-ray diffraction patterns collected in this study at nonambient conditions (i.e. 173 AE 1 K, 298 AE 2 K, 373 AE 1 K, and 473 AE 1 K at 1 bar; 9.13 AE 0.07 GPa at 298 K) show reflection conditions conforming to Laue class 3m1, with the possible space groups R3c (no.167) and R3c (no. 161), in one of the two possible rhombohedral orientations following obverse and reverse settings.
Diffraction images recorded at different temperatures shows weak but recognizable diffuse scattering (Fig. 7), which is independent of the temperature (Fig. S6, supplement). The appearance of the diffuse scattering corresponds to disk-like intensity distributions in reciprocal space, with the disks arranged along the c* direction. Since, single detector frames show only a plane cut through the reciprocal space, these disks appear either imperceptible or as broad diffuse lines (Fig. 7a), which in contrast to diffuse planes parallel to (001) clearly lose intensity with rising 2θ angle. The scattering features occur with maximum intensities at multiples of l ¼ 6n (Fig. 7), in analogy to the~4.6 Å spacings (corresponding to the Cs⋯Cs distance within the channels). The scattering intensity within these disks is radially symmetrical within reciprocal layers parallel to hk0 (Fig. 7), with gradually decreasing intensity with increasing scattering vector s or q (commonly used in aperiodic crystallography as q-vector). Note that the q-vector is used here rather in terms of its length, which is defined using the wavelength (λ) and scattering angle (θ) as q ¼ 4π ⋅ sin(θ)/λ. In contrast, rational (1/4) or irrational (0.26789) fractions of hkl are commonly used for q (modulation wave vector) to describe commensurate or incommensurate modulated crystal structures. This occurrence suggests one-dimensional substitutional disorder (e.g. Cs1, Cs2 and Ow1, Ow2) of extra-framework population along the channel direction, which are quite ordered along the c-direction and which, in their lateral arrangement, should be arranged independently of one another.
The structure refinements based on the integrated Bragg peaks was attempted for all data sets using the structure model reported by Ref. [16,18], with fully occupied framework positions and corresponding partial site occupation for the extra-framework atoms Cs and Na. After a careful inspection of the electron-density maps, it was not possible to clearly identify (and refine) the extra-framework positions ascribed to H 2 O (Ow1 and Ow2), reported as O10 and O11 by Ref. [16], from superimposing Cs2 and Cs1 sites. Exceptionally large and systematically increased values for weighting factors, as well as of the several anisotropic displacements parameters (in particular those of Cs, Na, Li, Be and some oxygen positions), gave reason to carry out comparative symmetry evaluations, using the tools provided by the Bilbao Crystallographic Server [47][48][49]. In particular, the data sets at 173 K and 9.13 GPa revealed also small but significant oxygen atom displacements (with similar orientation) for the same type of atoms (cf. Fig. S7 in the supplement), pointing out the true symmetry could be non-centrosymmetric R3c rather than centrosymmetric R3c.
Based on previous crystallographic discussions on twinning and on the presence of acentric twin domains in pezzottaite crystals [18], the distribution statistics of normalized structure factors were also carefully checked. The evaluation of the N(Z) distribution suggests, along with the Flack test [50], a high likeliness for centrosymmetry, but the occurrence of reticular merohedry and twinning (with obverse and reverse domains, cf. Fig. S8 in the supplement) raised doubts about the reliability of the results. For example, the ambiguous value for the Flack x parameter (i.e. 0.38 AE 0.04 at 173 K and 0.40 AE 0.11 at 9.13 GPa) is likely affected by the presence of merohedral twinning [51]. Accordingly, all structure refinements were carried out in both potential space groups (R3c and R3c), taking into account the possible twin laws (with obverse/reverse domains) and the twinning due to the loss of the inversion centre (i.e. two twin domains in R3c, four domains in R3c). Of course, lower residuals have been obtained for the T-series datasets and the high-P dataset at 9.13 GPa with refinements in the acentric space group R3c, but much more significant is that the b weighting parameter in SHELX for the high-P dataset was lowered dramatically (decreased by 60%). The residuals and further fit parameters also from using MERG0 and a fit with anisotropically refined Si and Al atoms (lowering the b value as well) are shown in the supplement (Table S4). The refinements of the series (taken with two crystals) under T variation yield reasonable scale factors for the four twin fractions, with approximately equal proportions of the oppositely acentric domains, i.e. with 43% and 46% at 9.13 GPa (cf. Table 1).
For the data collections at 298 K, 373 K and 473 K, the refinements in R3c were less stable, the weighting parameters were inconspicuous and the differences in atom position between R3c and R3c within the 3 sigma level (Fig. S5). Therefore, these data sets were refined in centrosymmetric R3c, with only two-individual twin fractions corresponding to the domains in obverse and reverse rhombohedral settings. The situation was slightly different at 173 K. The refinement using space group R3c produces a high weighting parameter b of 12.964 (a ¼ 0.026). As the violation of centrosymmetric symmetry seemed to affect only a few atoms (Fig. S5), even the data set collected at 173 K was finally refined in Fig. 10. Coordination polyhedra around Cs2 positions at 298 and 9.13 GPa illustrating the changed coordination of the cesium atom from approximately [6 þ 6] coordination at ambient conditions to [6 þ 9] at high-P. The high-P bonding configuration is not only the result of three further atoms (O6A), but a substitution of O9B atoms for O6A and O6B atoms. The O1 atoms where not influenced in that way, but clearly split up to O1B (3.88(2) Å Cs-O distance) and O1A (3.98(2) Å Cs-O distance).
R3c, using a structure model with disordered O3, O5 and Na2 positions. This already lowered the value of the b parameter in the weighting to 9.141 while the parameter a retains at 0.026. However, up to a value of around 5-6, for such a natural and huge structure with multiple site occupations, the b value is not considered to be crucial. On the other hand, it is a further sign that the chosen space group R3c is likely not the right choice to account for the symmetry of the channel content and nearby atom positions at 173 K. Changing from the R3c to the disordered R3c structure also lowered other parameters (R 1 [obs]: 2.86-2.73; R 1 [all]: 4.29-4.43; wR 2 : 7.15-6.68; highest difference peak: 1.536 e À Å À3 -1.416 e À Å À3 ; deepest hole: À0.793 e À Å À3 -À0.633 e À Å À3 ). It is noteworthy that the site-occupation factors (SOF's) of the split positions correspond in pairs to a ratio of 0.9 : 0.1, which in turn corresponds to the refined twin fractions of two oppositely acentric domains. Although weak electron densities (<1.4 e À Å À3 , 173 K) have been localized in the vicinity of the Cs atoms, and could not be unambiguously ascribed to the partial occupied Ow positions as proposed by Refs. [16], the final refinements of the data sets collected under T variation were carried out only with Na and Cs extra-framework population. However, the electron-density distribution around the average Cs positions located at the central sites 6a and 12c, and the ADP's for atoms refined at 0,0,z, suggest a small but considerable lateral displacement off the three-fold axis. Thus, the final refinements of the data sets collected under T variation were carried out accordingly with atoms at general positions with respective SOF's listed in Tables S5 to S8 and S10 to S13.
At high pressure, the Cs atoms were localized directly on the threefold axis, due to the reduced accessible reciprocal space. However, large isotropic displacement factors were observed for the Cs sites. In particular, this phenomenon is pronounced for the Cs2 site (U iso of 0.037 Å 2 ), if compared to the Si atoms (U iso of 0.006 Å 2 ), which results in a root-meansquare value for the atomic displacements of 0.19 Å (Cs2) and 0.08 Å (Si). As for the low-T structure, the refinement was first done using the R3c space group, resulting again in large values for the weighting factors and large displacement parameters for several atoms. Any attempt to test the non-centrosymmetry failed due to the lack of a sufficient number of Friedel pairs (Friedel coverage is 47.5%), which in turn is due to the limitation of the accessible reciprocal space for high-P measurements in a DAC. However, for a refinement in R3c, the findings suggested a plausibly large splitting of oxygen position, supporting the refinement of the high-P structure in the non-centrosymmetric space group R3c (cf .  Table S9).

Host framework and its variations at non-ambient P and T conditions
The [(Be 2 Li)Al 2 Si 6 O 18 ]host framework with its ideal symmetry R3c can be derived by Li and Be ordering at the [4] T1 position in the beryltype [ [4]  Al on [6] M and Si on [4] T2. The refinements of this study confirm, for any data set measured at non-ambient P,T conditions, the cation distribution as previously described for pezzottaite [16,18,19], with Li at the 18e and Be at the 36f Wyckoff positions in R3c. Symmetry lowering according to R3c→R3c leads to a distribution on three non-equivalent 18b sites in R3c (Li, BeA, BeB). Apart from the Be site, all Si and oxygen positions split into two independent sets (labeled A and B: Si1, Si2, Si3 in R3c → Si1A, Si1B, Si2A, Si2B, Si3A, Si3B in R3c), originating from the loss of the center of symmetry. Nevertheless, Raman spectra do not provide any clear evidence for the theoretically possible splitting of the relevant T-O vibration bands at low-T, and all bands in the relevant wavenumber range between 900 and 1200 cm À1 remain apparently unchanged across the entire T range (cf. Figs. 1 and 2), in terms of possible splitting, emerging of shoulders, or peak broadening. A plausible explanation for this would be that the individual tetrahedral units do not show such large differences at different T. Their vibration frequencies are, therefore, remarkably similar, and the actual changes due to the reduction in symmetry only manifest themselves in relative displacements of the TO 4 units within the framework. Accordingly, the only recognizable changes can be seen in the lowfrequency spectral range, in which the external vibrations and breathing modes can be observed.
Changes due to the applied pressure, on the other hand, show a broadening of these modes that can be assigned to the diversification of T-O vibrations, as well as emerging and splitting at P greater than 3 GPa (i.e. bands at 75 and 111 cm À1 , cf. Figs. 2 and 3). Especially this splitting could be due to the increase of Raman active modes attributed to Cs, which occupies two different Wyckoff positions, 12c and 6a at R3c (four possible Raman active modes) and three times the Wyckoff position 6a at R3c (nine possible Raman active modes, [52]). Moreover, the P-induced distortions of the structure are comparably larger than the T-related ones. In addition, the displacive framework distortions are subjected to other mechanisms owing to the interaction between framework and extra-framework population.
Nevertheless, the change in the P dependence of the band positions is an indication of the assumed change in symmetry. This is also manifested in the variation of T-dependent unit-cell parameters (cf. Fig. 4), which are only marginal but noticeable as exemplified by the c/a ratio (cf. Fig. 4). Since both structures in R3c and R3c remain substantially unchanged, most of the essential symmetry elements are preserved, and only the loss of the center of symmetry leads to the observable differences. Intersecting trends in Figs. 5 and 7 provide values of approximately 2.8 GPa for the critical transition pressure P c referring to the R3c→R3c transition on compression, and approximately 308 K for the R3c→R3c counterpart on T-increase T c . Regardless of the P or T being the tuneable property, the c/ a ratio decreases or increases for both structures in the same way, always with a steeper slope for the non-centrosymmetric This inverse behavior can only be attributed to the fact that this framework structure must accommodate extra-framework atoms intercalated inside the channels. With the placement of those guest atoms, volume optimization within the channels can be achieved only by stretching along the c-axis. That would also be plausible, especially since the channels at RT appear too wide for the Cs atoms (evidence of displacement off the triad were reported). The increasing vibration dynamics with T enables the Cs atoms to take up the more central position, which in turn reduces the lateral channel diameter, but takes up more space in the c-direction. The compressional anisotropy is apparently less dependent on the type and fraction of channel constituents, and the presence of Cs atoms does not seem to have a strong influence on the high-P properties of this framework type. The compressibility along the caxis is about 25-30% higher of all the other investigated isotypic material and the axial moduli M 0 //a and M 0 //c are in a relative ratio between 1.28 : 1 (this study and anhydrous Al 3 (Mg,Fe) 2 AlSi 5 O 18 [9], 1.22 : 1 in Be 3 Al 2 Si 6 O 18 [44], and 1.18 : 1 CO 2 -intercalated Al 3 (Mg,Fe) 2 AlSi 5 O 18 [11]).

Stereochemistry of extra-framework guest atom at non-ambient P and T conditions
The electron density maps determined in this study confirm the distribution of the extra-framework atoms Cs and Na as determined by Ref. [16]. In line with the higher-order centrosymmetric R3c symmetry, these atoms occupy sites along the central three-fold axis of the channel, i.e. Cs1 on the Wyckoff positions 12c (point symmetry 3.), Cs2 on 6a (32), Na1 on 6b (3), and Na2 on 12c (3.) (Fig. 8). A full occupation of these sites would provide 1 Cs þ 1 Na atom per formula unit. The actual refinements of the SOF's at 173 K, on the other hand, confirm that 73.0 AE 0.2% is occupied by Cs, and 13.5 AE 3.0% by Na.
However, a detailed examination of the electron-density distribution around these positions shows that the electron density around the Cs atoms is clearly displaced laterally, as already reported by Ref. [16], who interpreted the Cs atoms to be located directly at 0,0,z (z % 1/12 for Cs1, ¼ for Cs2) and Ow1 and Ow2 of potential guest H 2 O in the immediate vicinity off the tri-fold axis. . Ideally, the Cs sites lie on the central position of the channel, and thus on the triad (labeled Cs1* and Cs2* in Fig. 8). Since the central position Cs* always leads to too long Cs-O bond distances, the lateral displacement off the channel center line also seems to be justified by stereochemistry. Accordingly, Cs1 and Cs2 position were refined on general positions 36f, which are displaced at 173 K bỹ 0.007(2) Å (Cs1) to 0.095(8) Å (Cs2) from the ideal Cs* positions located at 12c and 6a, respectively. The resulting coordination polyhedra correspond to a [3 þ 6þ3] coordination with Cs1-O distances ranging from 3.28 to 3.50 Å, and a [6 þ 6] coordination with Cs2-O distances from 3.28 to 3.75 Å (cf. Fig. 10).
Displacive shifts can also be observed for part of the Na sites, i.e. for the Na2 position (as located on 12c in R3c, now on two sites Na2A and Na2B, each located at 6a). In contrast to the Cs atoms, the Na atoms get displaced along the threefold axis while keeping coordinates 0,0,z. The bonding scheme is not changed significantly if compared to the findings of [16], who described it as [2 þ 6] coordination, i.e. two bonds to the Ow atoms (Ow1, Ow2) located at the Cs* sites, completed by six Na-O bonds to the framework oxygen atoms O7, O8, and 09 (Fig. 8). Again, under pressure, the positions are shifted, but the coordination of Na is not changed if compared to the pressure-induced stereochemical changes of the Cs atoms.
The arrangement of the channel sites follows a sequence of alternating Na and Cs atoms, in an approximate sequence at z ¼ 0, 1/12, 2/12, 3/12, …. Looking at the stereochemical possibilities and considering which positions can be occupied at the same time, there are only a few ordering schemes of the channel sites. This includes the sequence of [Cs1⋯Cs2⋯Cs1] n , with CsO 12 face-sharing polyhedra along the c-axis (Fig. 9). In this case, it would be possible to achieve a fully stuffed channel according to SOF's of 1.0 at each of the three Cs sites in R3c. Na could also have full occupancy, requiring the additional Ow1 and Ow2 positions, which implies a given fraction of H 2 O (Fig. 9). If the channels were exclusively filled with Na or Cs, there would be a full occupation of one alkali atom per formula unit. However, if one considers a mixed occupation by Na and Cs within a single channel, only a partial occupation of the Cs and Na positions is possible due to the limited possibilities for stereochemically meaningful bonds (cf. Fig. 8).

Conclusions
The present study shows that variation of P (at least up to 10 GPa) and T (at least up to 1200 K) leads to an apparently second-order displacive phase transition of (Cs,Na)[(Be 2 Li)Al 2 Si 6 O 18 ] from R3c→R3c, which does not influence the structural stability, despite its open-framework (beryltype) structure.
Both the framework topology and the R3c superstructure, due to (Li, Be) ordering scheme on [4] T1 sites, remain unaffected throughout the variations in P and T space. While the well-ordered framework is responsible for sharp Bragg reflections in the diffraction pattern corresponding to Laue class 3m1, disorder between adjacent onedimensionally ordered arrangements of the channel population (i.e. Cs, Na and H 2 O) is responsible for the appearance of detectable diffuse scattering. Since the diffuse scattering does not change in response to the applied P and T, it can be assumed that the order within the individual channels remains unaffected. Nevertheless, the investigations at low T and at the highest P show that deviations from the superordinate framework symmetry originate from displaced atom positions inside the channels, as required for adaption of individual bonds. Therefore, the centrosymmetry is broken and the structure is slightly distorted by cooperative atomic displacements in accordance with the acentric space group R3c. The resulting R3c-to-R3c phase transformation is relatively inconspicuous, recognizable in the measurements towards low-T conditions but most pronounced in the compressed structure. Hints for the subtle transformation can be found in: individual Raman modes in the region of the external vibrations, in lower residuals of refinements of the structures in the non-centrosymmetric space group R3c, but also in the behavior of the unit-cell parameters in connection with the thermal expansion or with isothermal compression. The critical transition point for the T space was found to be at T c ¼ 312 AE 5 K, while the P c seems to be in a region between 2.8 and 5.3 GPa. The description of the crystal structure in a correspondingly twinned acentric polymorph is a new aspect compared to previous findings on this stuffed beryl-type structure. Both the aspects of structural stability under the P, T variations carried out in this study and the thermomechanical properties* confirm that this class of material has a high potential for technical application [*low thermal expansion coefficient, high resistance to thermal shocks, a pronounced structural stability at high-T and high-P conditions (all typical for beryl type framework structures), less pronounced anisotropy of thermal expansion (compared to other beryl type framework structures), a negative thermal expansion value in the a-axis direction at T < 300 K (first time observed for a beryl type framework structure)]. If a simple synthesis route of this material will be developed, for example under hydrothermal conditions, the anhydrous end-member Cs[(Be 2 Li) Al 2 Si 6 O 18 ] would serve as a potential solid hosts for 137 Cs γ-radiation source in sterilization applications. The low amount of potential contained Cs (up to 19.9 wt% for the anhydrous endmember) likely makes this material not an ideal candidate for use in the immobilization of radioactive Cs-isotopes, though the Cs-retention under extreme conditions, especially at high-T, in a crystalline structure makes it suitable for this type of utilizations.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.