Center for Advanced Radiation Sources, Argonne National Laboratory, University of Chicago, Argonne, Illinois, USA
Corresponding author: P. Dera, Center for Advanced Radiation Sources, Argonne National Laboratory, University of Chicago, Bldg. 434A, 9700 South Cass Ave., Argonne, IL 60439, USA. (email@example.com)
 Minerals containing silicon in four-fold coordination (IVSi4+) are common in crustal rocks, while those involving six-coordinated silicon (VISi4+) dominate the Earth's lower mantle and determine its properties. Here we show a new type of phase transition determined by single-crystal high pressure X-ray diffraction experiments in a diamond anvil cell (DAC) using natural diopside (CaMgSi2O6), the archetypic member of clinopyroxene family, and one of the most abundant minerals of the Earth's upper mantle. Above 50 GPa at ambient temperature diopside transforms to a previously unknown post-clinopyroxene phase,β-diopside, with half of the tetrahedralIVSi4+ layers converted to octahedral VISi4+coordination. This phase is most probably a metastable state that is kinetically accessible at room temperature and the transformation is fully reversible on decompression. This new type of phase transition provides important clues to the exact mechanisms of breakdown of clinopyroxene in the Earth's mantle and may be expected to take place in other pyroxenes at pressures higher than previously explored.
 The Ca-rich clinopyroxene (cpx) diopside with ideal end-member composition of CaMgSi2O6 is an important component of both mafic and ultramafic rocks, and is among the main constituents of the pyrolytic Earth upper mantle. Diopside is also abundant in meteoritic rocks and interplanetary dust particles [Christoffersen and Buseck, 1986].
 Diopside is believed to be thermodynamically unstable along the geotherm, above about 16 GPa in the range of the 660 km discontinuity, dissociating to MgSiO3 (garnet, akimotoite, or perovskite) and CaSiO3 [Irifune et al., 1989]. The diopside breakdown is also expected to occur during impact events of rocky asteroids. However, recent reports of meteoritic occurrences of diopside in heavily shocked L6 chondrites suggest that it can survive to well beyond the established phase boundaries [Zhang et al., 2006; Ozawa et al., 2009].
 A key concept in understanding the effects of high pressure on the crystal chemistry of minerals in planetary interiors is a tendency to increase the coordination number of cations with increasing pressure. This trend is particularly important for silicate mineral components of the rocky planets, where Si4+ prefers tetrahedral coordination in minerals at ambient pressure, but transforms to VISi4+ on compression. With the exception shown by Angel et al. transformation from the crustal rock-forming silicates withIVSi4+ to those containing VISi4+ does not pass through intermediate states. The change in Si4+coordination is usually associated with a reconstructive phase transition, and involves significant activation energies, which often results in the possibility to metastably quench the six-coordinated phase to ambient conditions (e.g., stishovite [afterOhtani et al., 2011]). In almost all pyroxene systems Si4+occupies the tetrahedral sites, whereas most of the products of high-pressure breakdown of pyroxenes, including akimotoite, as well as MgSiO3 and CaSiO3 perovskites involve a transformation of all IVSi4+ coordination to VISi4+. The inability to efficiently accommodate VISi4+ in cpx, without destroying the pyroxene structure, is probably one of the main reasons for the breakdown at high pressure and temperature.
 Synthetic high-pressure high-temperature experiments with Na-rich pyroxene compositions demonstrate that Si4+ can replace some of the Mg2+at the six-coordinated octahedral sites in the cpx structure with fractional occupancy as high as 50% [Angel et al., 1988; Yang et al., 2009]. To our knowledge, however, neither these synthetic high-pressure phases, nor any of the high-pressure structural transformations of cpx observedin-situ, indicated conversion of the Si4+tetrahedral sites to six-coordinated octahedral geometry.
 The evolution of the diopside crystal structure with pressure has been a topic of interest of mineral physicists for several decades. All of the experimental studies that reported structure refinements utilized laboratory diffractometers and were constrained to pressures lower than 10 GPa [Levien and Prewitt, 1981; Zhang et al. 1997; Thompson and Downs, 2008]. The most recent study using synchrotron in-situ powder diffraction and a synthetic diopside sample extended the pressure range for determination of the equation of state to 40 GPa, but did not produce structure refinements, making impossible verification of the linearity of the polyhedral and bond compression trends [Tribaudino et al., 2000].
 An in-situ Raman spectroscopic study by Chopelas and Serghiou  reported several discontinuities in the pressure dependence of vibrational frequencies, and noted a remarkable metastability of the cpx structure at ambient temperature, up to at least 70 GPa. While no diffraction data were obtained, the most significant change observed in this study, which took place above 50 GPa, was interpreted in terms of possible change from IVSi4+ to VISi4+.
 Recent technical advances facilitating synchrotron single-crystal X-ray diffraction methods allow the collection of scattering data for structure determination at ultrahigh-pressures [Dera et al., 2011a, 2011b]. Given the importance of transformations in cpx to elaborating the crystal chemistry of the silicate minerals at high pressure, we decided to examine the 50 GPa transition and attempt to solve the structure of the high-pressure phases encountered.
 Before the experiment, the sample was analyzed with the electron microprobe which indicated a composition close to pure end-member: CaMg0.97Fe0.03Al0.01Si1.99O6.
 Four single crystals with different crystallographic orientations were loaded into symmetric piston-cylinder Princeton-type DAC, as shown inFigure 1a. The DAC was loaded with neon pressure medium using the GSECARS/COMPRES gas loading system [Rivers et al., 2008]. Pressure was estimated based on a ruby fluorescence spectrum collected at each pressure point [Mao et al., 1986]. After the gas loading, the starting pressure for the diffraction experiment was 1.6(1) GPa. The samples were compressed in steps of approximately 8 GPa, with single-crystal data collection for each crystal at each pressure point up to 63 GPa. Diffraction data were collected at the experimental station 13IDD of the GSECARS facility at Advanced Photon Source, Argonne National Laboratory using a monochromatic beam with an incident energy of 37 keV, focused to a spot of 0.003 by 0.005 mm. Diffraction images collected using a MAR165 charge coupled device (CCD) detector, were analyzed using the GSE_ADA/RSV software package (P. Dera, GSECARS, Chicago, Ill., personal communication, 2007).
 The structure of the high-pressure phase was solved by means of the computer program Endeavour [Putz et al., 1999], starting from random positions for all atoms. Different atomic configurations were tested using the simulated annealing approach, by minimizing the difference between the calculated and observed diffraction patterns, and at the same time minimizing the energy of interatomic interactions as approximated by means of simple repulsion potentials with cost function set at 0.8. The resulting model was then used as a starting point for a conventional least squares crystal structure refinement using the SHELXL program [Sheldrick, 2008]. Structure refinements were done using isotropic displacement parameters for all atoms. The occupancy of Mg2+as well as Si4+ sites were set to 100%. Crystallographic parameters for structure refinements of diopside at 38.6(1) GPa and β-diopside at 56.1(1) GPa are summarized in Table S1 in theauxiliary material.
 All data collected up to 45.6(1) GPa showed no indications of change in symmetry from the original C2/c, and the structure could be successfully refined starting with the ambient pressure model (Figure S1a in the auxiliary material).
 The experimental unit cell volume at 45.6(1) GPa was 355.8(3) Å3, which is in excellent agreement with volumes calculated from DFT simulations [Walker et al., 2008] and the 40 GPa powder experiment [Tribaudino et al., 2000] (357.4 Å3 and 354.9 Å3, respectively), and slightly larger than 343.82 Å3, suggested by extrapolation of the 10 GPa experiment [Thompson and Downs, 2008].
 For the purpose of comparing the structures of diopside and β-diopside, we will discuss data collected at 38.6 GPa, which were of better quality than the 45.6 GPa dataset (Figure 1b). In the low-pressure diopside phase at 38.6(1) GPa the average Mg-O distance is 1.93 Å, while the polyhedral volume is 9.56 Å3 (DFT calculations by Walker et al.  predict 9.82 Å3). The octahedral angle variance, which is a measure of the distortion of bond lengths from the ideal polyhedron [Robinson et al., 1971], is 36.0°. The average Si-O distance is 1.60 Å, with polyhedral volume of 2.11 Å3 (DFT prediction is 2.07 Å3). The tetrahedral angle variance is 31.42°. The average Ca-O distance is 2.31 Å, and the polyhedral volume is 20.78 Å3 (DFT prediction is 20.72 Å3). The O3-O3-O3 angle, which is used as a measure of rotation of the individual tetrahedra along the chain, is 158.97°. The linear model of scaling tetrahedral chain rotation with pressure derived by extrapolation of the compression mechanism to 10 GPa byThompson and Downs significantly overestimates the rotation, and predicts an O3-O3-O3 angle of 150.04°. The experimental value obtained in our analysis is much closer to the non-linear trend revealed in the first principles simulations [Walker et al., 2008].
 The first indication of the phase transition was observed at 53.1(1) GPa. At this pressure, in addition to the set of diffraction peaks consistent with the C2/c cpx phase, we noticed a new set of peaks, which could be indexed on the basis of monoclinic unit cell showing systematic absences consistent with the P21/c space group (Figure S1b in the auxiliary material). This suggests co-existence of the two phases over a range of few GPa as separate domains within the same crystal. At 56.1(1) GPa the transformation was already complete, and only the peaks of the high-pressure phase remained (Figure S1c in theauxiliary material). The transition to the high-pressure phase is accompanied by about a 5% increase in density and the monoclinicβ-angle is reduced from 104.59(1)° in C2/c at 38.6(1) GPa to 96.34(2)° in the P21/c phase at 56.1(1) GPa.
 The structural parameters of the high-pressure phase are listed inTable 1. The main new feature of the β-phase is conversion of half of the layers of corner-sharing SiO4tetrahedral chains into layers of edge-sharing SiO6 octahedra (Figure S2 in the auxiliary material). This transformation is achieved by shifting the Si(1) silicon atoms from their original locations above the oxygen involved in the octahedral shared edge to a new site, almost directly above the Mg2+ (Figure 1c). The new configuration of Si(1) layers is topologically equivalent to the octahedral Si layers in ilmenite-type MgSiO3 akimotoite, as shown in Figure S3 in the auxiliary material. After the transition in the octahedral VISi(1) site the average Si-O bond length increases to 1.75 Å, which corresponds to an octahedral volume of 6.97 Å3. The Mg2+has average Mg-O bond length of 1.92 Å and polyhedral volume of 9.12 Å3, very similar to the corresponding values before the transition. The Mg(1)O6 octahedron is more distorted than in the low pressure phase, and more distorted than the Si(1)O6 octahedron, with octahedral angle variances of 82.3° and 57.2°, for MgO6 and SiO6 in the β-phase, respectively. The remaining half of the Si layers (Si2), retain the tetrahedral chain arrangement causing the lowering of symmetry to P21/c (Figure 1d). At the Si(2) tetrahedral site the average bond length is 1.60 Å, with polyhedral volume of 2.03 Å3. The tetrahedral angle variance is 75.2°, indicating a much stronger frustration of the coordination geometry than in the low pressure phase. At the Ca2+ site average bond length is 2.22 Å, with polyhedral volume reduced to 19.00 Å3.
Table 1. Fractional Atomic Coordinates and Isotropic Displacement Parameters for Diopside and β-Diopsidea
Ueq (Å2 × 103)
Diopside: C2/c, a = 9.234(1) Å, b = 8.255(1) Å, c = 4.959(1) Å, β = 104.592(10)o; β-diopside: P21/c, a = 9.053(4) A, b = 7.677(3) A, c = 4.751(1) A, β = 96.34(2)o.
Diopside at 38.6(1)GPa
β-Diopside at 56.1(1)GPa
 Our attempt at increasing pressure above 60 GPa resulted in damage to the diamond anvils, which made impossible examination of the transition reversal behavior. The Raman spectroscopic results presented by Chopelas and Serghiou , however, strongly suggest that a full reversal is expected. Based on the above transition mechanism it seems likely that another transformation, converting the remaining half of tetrahedrally coordinated silicon atoms to octahedral configuration, occurs at pressures higher than 70 GPa, but further experiments are needed to verify this hypothesis.
 The crystalline cpx system was previously thought to be quite constrained with regard to accommodating VISi4+. However, in glasses and melts of diopsidic composition, where the long range order is lost, the situation is different. Spectroscopic evidence for a gradual and reversible change in coordination of silicon, accompanying compression of such silicate glasses was observed by Williams and Jeanloz . A similar trend was established for diopside liquid by first principles simulations [Sun et al., 2011]. A structural mechanism allowing for such gradual coordination change, which involves rotation of the individual tetrahedra along the silicate chain, accompanied by a change in the O3-O3-O3 angle was proposed byStolper and Ahrens . Our results demonstrate that the change from four- to six-coordinated silicon can also be achieved in crystalline state through a displacive mechanism, without breaking the integrity of the crystal lattice, and that, as in the case of cpx glass and liquid, it is reversible on pressure release.
 Based on the results of our experiments with diopside one can hypothesize that a similar kind of polymorphism appears in other pyroxenes at pressures higher than previously explored. The IVSi4+/VISi4+layer-by-layer intergrowth may constitute a general intermediate step in the transformation to the dense phases containing only six-coordinated silicon in Earth's lower mantle. The composition of the pyroxene may also have an important effect on the transition pressure, with Na-rich compositions, which were demonstrated to accommodateVISi4+ in the Mg2+ site, expected to transform to the P21/c β-phase at lower pressure.
 Laboratory high-pressure high-temperature quench experiments demonstrate that diopside decomposition to Ca-rich and Ca-poor products requires temperatures well above 1000°C, which are consistent with the geotherm conditions, however, the pressure dependence of the activation temperature for the breakdown process is not well constrained (from our results and those ofChopelas and Serghiou it is clear that at ambient temperature the breakdown does not take place even at 70 GPa). Seismic imaging data indicate that in geologic environments such as cold subduction zones the temperatures in the diopside-rich subducted oceanic slab, can be lower than the geotherm by as much as 1000°C [Abers et al., 2006]. Recent seismic travel time tomography evidence by Li and van der Hilst  suggests that some of the subduction zones, particularly in the Southeast Asia region, extend within the mantle well beyond the 660 km discontinuity, perhaps as deep as 1000 km, which corresponds to pressures above 40 GPa. The temperature of the slab at this depth is modeled to be approximately 1500°C [Bina et al., 2001], lower than required to transform diopside to garnet [Nishi et al., 2008]; conditions that could favor the diopside to the β-phase conversion.
 In addition to temperature, another factor that could promote the presence of the high-pressure phase of diopside in Earth's mantle is uniaxial stress. Recent experiments aimed at verification of the mantle superplasticity model byHiraga et al. demonstrate that polycrystalline aggregates of olivine and diopside can withstand homogeneous tensile elongation of up to 500% under subsolidus conditions. Such significant anisotropy of the stress state could potentially shift the cpx transformation to lower pressure conditions and cause the high-pressure phase to form above 800 km in deep cold subduction zones.
 This work was performed at GeoSoilEnviroCARS (Sector 13), Advanced Photon Source (APS), Argonne National Laboratory. GeoSoilEnviroCARS is supported by the National Science Foundation - Earth Sciences (EAR-1128799) and Department of Energy - Geosciences (DE-FG02-94ER14466). Use of the Advanced Photon Source was supported by the U. S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357. Work at Stony Brook, including sample preparation, data collection and analysis, and manuscript preparation was supported by DOE grant DE-FG02-09ER46650. The authors would like to thank Baosheng Li for sharing the diopside sample and Donald H. Lindsley for help in microprobe analysis.
 The Editor thanks Joseph Smyth and an anonymous reviewer for their assistance in evaluating this paper.