Seismic anomalies in subducted slabs, including low velocity zones and shear wave splitting, have often been related to hydrous regions. Phase D (MgSi2H2O6, 10–18 wt.% H2O) may be the ultimate water carrier in hydrous subducted peridotite and its seismic properties are thus essential for interpreting observed anomalies in terms of hydration. Here, we report the sound velocities and elasticity of Mg- and Al-Fe-bearing phase D single-crystals measured by Brillouin spectroscopy. The room conditions adiabatic bulk and shear modulus are: KS0 = 154.8(3.2) GPa and μ = 104.3(2.1) GPa for Mg-phase D and KS0 = 158.4(3.9) GPa and μ = 104.7(2.7) GPa for AlFe-phase D, suggesting minor effect of cation substitution on the elasticity of phase D. Based on the seismic velocity data, we found that 16 vol.% of AlFe-phase D in hydrous subducted peridotite with 1.2 wt.% H2O could provide a plausible explanation for the negative velocity anomalies of 3% observed in fragments of the Tonga slab below the transition zone.
 Tracking the degree of hydration of deep subducted slabs and the identification of the water storage sites is crucial to constrain the recycling and circulation of water in the Earth's interior. Seismic anomalies identified at various depths in subduction zones, including low velocity zones (LVZ) and high shear wave splitting [Abers, 2000, 2005; Gu et al., 2001; Chen and Brudzinski, 2003; Wookey et al., 2002] have been often related to the plausible presence of hydrous phases [Mainprice and Ildefonse, 2009; Hacker et al., 2003]. Among the candidate hydrous phases, phase D (ideal formula MgSi2H2O6, 10–18 wt.% H2O) has been proposed in petrological studies as the ultimate water carrier along cold slab geotherms [Kanzaki, 1991; Ohtani et al., 2000, 2001; Litasov and Ohtani, 2007] and could account for more than 50 vol.% of very hydrous subducted peridotites from 700 to 1250 km depth [Iwamori, 2004]. Additionally, water released during the breakdown of phase D in the uppermost lower mantle (1100–1400 km depth [Shieh et al., 1998]) may provide a mechanism for the rehydration of the overlaying mantle, triggering changes in the viscosity of the mantle with important consequences for mantle convection [Lawrence and Wysession, 2006]. The large volume fraction of phase D in hydrous peridotite together with its intrinsic anisotropy [Mainprice et al., 2007] could have a significant impact on the seismic velocity structure of deep slabs and may contribute to explain observed seismic anomalies. An accurate evaluation of the seismic properties of phase D is thus necessary to identify the seismic signature of water at depth and to constrain the degree of hydration of subducting slabs.
 Despite the potential role of phase D on the transport and recycling of water inside the Earth, significant uncertainties persist on its elastic properties. The isothermal bulk moduli obtained from compressional studies on powdered samples as well as from computational studies, span from KT0 = 130(1) GPa [with K′T = 7.3(1)] to KT0 = 166(3) GPa [with K′T = 4.1(3)] [Shieh et al., 1998; Frost and Fei, 1999; Tsuchiya et al., 2005; Litasov et al., 2007, 2008; Shinmei et al., 2008; Hushur et al., 2011] and large discrepancies are observed on the reported single-crystal and shear properties of phase D [Liu et al., 2004; Mainprice et al., 2007; Tsuchiya and Tsuchiya, 2008]. In order to identify the seismic signature of hydrous phases in deep subducted slabs, we have measured the sound velocities and elasticity of Mg-bearing and Al-Fe-bearing phase D at room conditions using Brillouin spectroscopy. The results are used to examine the possible contribution of phase D to observed seismic anomalies and allow discussing the degree of hydration of deep subducted lithosphere.
2. Experimental Procedure
 Single-crystals of magnesium phase D and Al- and Fe-bearing phase D (hereafter referred to as Mg-phD and AlFe-phD respectively) were synthesized at 24 GPa and 1200°C in a multianvil press. The starting materials were reagent grade SiO2 and Mg(OH)2in molar ratio 2:1 for Mg-phD and a stochiometric mixture of SiO2, Mg(OH)2, Al2O3 and Fe2SiO4for AlFe-phD. The composition of the samples determined by EMPA is Mg1.1Si1.9H2.4O6 (12.1 wt.% H2O) for Mg-phD and Mg1.0Fe0.11Al0.03Si1.9H2.5O6 (10.2 wt.% H2O) for AlFe-phD, where the H2O content is estimated from the loss in the analysis totals (±0.8 wt.%). The trigonal lattice parameters (space group P-31m), a = 4.762(3) Å, and c = 4.357(3) Å with V0 = 85.6(2) Å3for Mg-phD and a = 4.756(3) Å, and c = 4.342(3) Å with V0 = 85.1(2) Å3for AlFe-phD, are close to those reported byYang et al. . Calculated densities are 3.43(1) g/cm³ for Mg-phD and 3.56(2) g/cm3for AlFe-phD. For each composition, two single-crystals of high optical quality and sharp extinction in cross-polarized light microscopy were selected for the Brillouin measurements. One Mg-phD crystal was preoriented by X-ray diffraction and double-side polished into a plate with faces approximately parallel to the (100) plane. The other crystals were polished without preorientation in order to preserve the maximal surface. After polishing, the platelets were mounted on glass fibers and the orientation retraced by X-ray diffraction.
 Brillouin measurements were performed in 90 degrees symmetric scattering geometry using a solid state laser (λ0 = 532.1 nm) focused down to a 15 microns spot size. The scattered light was analyzed by a Fabry-Perot interferometer equipped with a photomultiplier (PMT) detector. Additional details of the experimental setup are given bySanchez-Valle et al. . Sound velocities Vi (i = P or S) were determined from the measured frequency shift (Δνi) using the relationship [Whitfield et al., 1976]: Vi = Δνi*λ0/2sin(θ*/2), where θ* (90 deg) is the angle between the incident and scattered beam outside the sample. The scattering angle was calibrated before the experiments using a MgO single-crystal standard and acoustic velocities were obtained with a precision of 0.5%. Spectra were collected with a laser power of about 50 mW at the sample surface to avoid dehydration due to overheating and typical collection times were of about 3 hours. Most spectra showed at least two acoustic modes: one compressional (VP) and one shear (VS) and were of excellent quality with high signal-to-noise ratio (Figure S1 in theauxiliary material). A total of 32 and 27 crystallographic directions were sampled in Mg-phD and AlFe-phD with a final dataset containing 77 and 51 velocity modes, respectively (Figure 1).
3. Results and Discussion
3.1. Elastic Properties of Phase D and Compositional Variations
 The orientations of phonon directions of the corresponding measured sound velocities were calculated using a least-square algorithm and the crystallographic orientation of the sample. The full dataset of acoustic velocities for each composition were inverted together with calculated phonon directions to determine the six independentCijusing a weighted least-square minimization procedure to solve the Christoffel equation [Musgrave, 1970]. The final fit model yields the elastic constants listed in Table 1with an average deviation of the velocities of 0.2% for Mg-phD and 1% for the AlFe-phD. An angular offset of 0.01% relative to the scattering angle was found for both compositions. The final errors on the longitudinal elastic constants are typically 0.5–1% because measurements were performed in directions close to the principal crystallographic axes (Figure 1), whereas those on the off-diagonal elastic constants are ∼2% due to higher covariance. Errors for the AlFe-phDCijare generally larger due to slight differences in composition between the two measured AlFe-phD crystals and the smaller number of sampled velocities used for the refinement. Errors include also uncertainties in density, orientation and small systematic errors on the determination of Brillouin shifts.
Table 1. Single-Crystal and Aggregate Elastic Moduli of Phase D at Ambient Conditionsa
RMS, root mean square; AHB1, asymmetric hydrogen bond structure.
RMS ( m/s)
 The longitudinal single-crystal moduliC11 and C33reveal highly anisotropic compressibility of both Mg-phD and AlFe-phD (Table 1), with the c-axis being 28(1) % more compressible than the a-axis. This behavior is related to the structure of phase D, which consists of SiO6 octahedron layers alternating with MgO6octahedron layers stacked along the c-axis [Kudoh et al., 1997; Yang et al., 1997]. The two dimensional framework of edge-sharing silicon octahedral makes the structure relatively more resistant to compression along the a-axis than along the c-axis. TheCijobtained for the two compositions agree well within the errors with the exception of the off-diagonal constant C13 (Table 1), which seem to be the most sensitive to compositional variations. Voigt-Reuss-Hill averages [Watt et al., 1976] of the aggregate moduli (KS and μ) and acoustic velocities (VP and VS) are reported in Table 1. Within uncertainties, compositional variations do not have a significant effect on the aggregate moduli and acoustic velocities of the investigated phase D samples. Compositional variations will thus be very difficult to detect between Mg and Al-Fe phase D compositions using seismic methods.
3.2. Comparison With Previous Studies
 The single-crystal and aggregate elastic properties of phase D samples are compared inTable 1 to data from previous studies [Liu et al., 2004; Tsuchiya and Tsuchiya, 2008; Mainprice et al., 2007]. The single-crystal dataset ofTsuchiya and Tsuchiya  has been excluded from the comparison because the elastic tensor was refined for a triclinic distorted crystal structure. Although differences in the Cij components reported in this study and by Mainprice et al. exceed mutual uncertainties, they agree in the trend defined by the compressional and off-diagonal components ofCij, with C11 > C33 and C12 > C13. The results notably contrasts with those of the Brillouin study of Liu et al.  where they found C33 > C11 (Table 1), implying that the a-axis is more compressible than c-axis. This observation is not consistent with the structure of phase D described above [Kudoh et al., 1997; Yang et al., 1997], even though the authors reported good agreement between the derived axial compressibilities and the results of P-V studies [Frost and Fei, 1999]. The fundamental differences in the components of the elastic tensor reported by Liu et al. suggests the transposition of axis during data processing, probably due to the use of an aggregate of phase D composed of several single-crystals that were individually sampled. In the present study, theCijof phase D were obtained from measurements in high quality single-crystals for which the orientation was determined by X-ray diffraction to improve the precision of the data. The adiabatic bulk modulus obtained in this work is in fair agreement within uncertainties with values reported by the theoretical studies [Mainprice et al., 2007; Tsuchiya and Tsuchiya, 2008] although differences are within ±15% for the shear modulus (Table 1). Both the compressional and shear velocities of phase D are substantially lower than reported in the previous Brillouin study, indicating than the effect of phase D on the velocity structure of subducting slabs may be larger than previously suggested.
 The isothermal bulk modulus KTof Mg-phD and AlFe-phD calculated from the Brillouin data, KT0 = 149.1(3.5) GPa and KT0 = 152.7(4.0) GPa respectively, agree well with the values reported from first-principles byTsuchiya et al. and from powder X-ray diffraction byShinmei et al.  with K′T = 5.3 and by Hushur et al. when fitting the whole P-V data collected up to 56 GPa (Table S1). Discrepancies with the rest of P-V studies are unlikely explained by deviations in sample composition only (H2O content and Mg/Si ratios) and could also arise from impurities in the powdered samples (e.g., stishovite) that may interfere data analysis [Frost and Fei, 1999; Litasov et al., 2007], deviatoric stresses and uncertainties in pressure calibration that can affect compression experiments. Additionally, the large covariance between KT, K′T and V0when fitting the P-V data to an equation of state may contribute significantly to explain discrepancies [e.g.,Litasov et al., 2007; Shinmei et al., 2008] (Table S1). To reduce uncertainties on the EoS of phase D, we applied the formalism of Bass et al. to fit the quasi-hydrostatic PV dataset ofHushur et al. by a 3rd order Birch-Murnaghan equation of state with KT0 fixed to the value obtained in the present Brillouin study, 149.1(3.5) GPa. A fit to the dataset up to 56 GPa using a weighted least square refinement for both V0 and P, yields K′T = 5.51(7) with V0 = 85.45(4) Å3. Using only data to 30 GPa, we obtain K′T = 5.60(14) and V0 = 85.44(4) Å3, which are indistinguishable within errors. Based on these results, the best estimates for the EoS parameters of phase D are KT = 149.1(3.5) GPa, K′T = 5.5(1) and V0 = 85.45(4) Å3. Additional high-pressure Brillouin measurements are however necessary to independently constrain the pressure derivative of the bulk and shear modulus.
4. Geophysical Implications
 Hydrous regions at depth have been commonly associated to low-velocity zones, high shear wave splitting and seismic attenuation anomalies [e.g.,Hacker et al., 2003; Mainprice and Ildefonse, 2009]. In Tonga subduction, negative velocity anomalies of up to 3% in both VP and VS, and shear wave splitting between 0.3% and 0.9% with faster horizontally polarized shear waves (VSH > VSV) have been observed in a slab fragment which lies subhorizontally at the 660 km discontinuity [Chen and Brudzinski, 2003]. The fast rate of subduction in Tonga provides the conditions to preserve hydrous phases, including phase D, which could transport water beyond the transition zone.
 Phase D (AlFe-phD) displays VP and VS velocities that are 2.9% and 3.0% higher than Mg0.85Fe0.15O ferropericlase (Fpc) [Jacobsen et al., 2002] but 18.9% and 20.5% lower than MgSiO3perovskite (Mg-pv) [Sinogeikin et al., 2004] and 7.6% and 2.5% lower than CaSiO3perovskite (Ca-pv) [Li et al., 2006] respectively. To evaluate the effect of phase D on the seismic velocity structure of the slab and to discuss the seismic detection of hydrous regions, we calculate the velocity contrast between dry and hydrous peridotite assemblies with pyrolitic bulk composition [Litasov and Ohtani, 2007] and various degrees of hydration. The seismic velocities were calculated using the elasticity data for AlFe-phD obtained in this study together with room conditions literature data for the principal constituents in peridotite [Jacobsen et al., 2002; Sinogeikin et al., 2004; Li et al., 2006] and mineral proportions calculated using a linearization fitting procedure. The calculations show that the seismic velocity drop of 3% in both VP and VSdetected in fragments of the Tonga slab below the transition zone could be explained by the presence of 16 vol.% of AlFe-phD (10.2 wt.% H2O), implying water contents of about 1.2 wt.% in deep subducted peridotites. Such water content is plausible in cold subduction zones considering that after the breakdown of antigorite at 6 GPa, the assembly phase A + enstatite will retain ∼4.5 - 5 wt.% H2O [Schmidt and Poli, 1998; Ulmer and Trommsdorff, 1999] that could be subducted beyond the transition zone through solid-solid reactions associated to the formation of dense hydrous magnesium silicates [Ohtani et al., 2004]. Although the calculations were conducted at room conditions and the estimated amount of phase D in hydrous peridotite would most likely depend on compositional effects (e.g., Fe and Al substitutions) on the elasticity of perovskite and ferropericlase, they place first order constrains on the amount of water in subducted slabs if phase D is present in the deep mantle.
 Additionally, phase D may be a potential source of anisotropy in deep subducted slabs as its layered structure [Kudoh et al., 1997; Yang et al., 1997] displays substantial compressional and shear wave anisotropy, ΔVi =200 × [(ViMAX – ViMIN)/(ViMAX + ViMIN)], of 18.5% and 19.4% respectively (Figure 2). The layered structure of Phase D may also easily align in a non-hydrostatic stress field and may thus act as a low strength phase, which accommodates most of the strain in a peridotite. In the Tonga slab fragment, the deformation regime was found to be down-dip compressional [Vavryčuk, 2006], suggesting that phase D would most likely align along the stacking fault axis (c-axis) parallel to the principal compression axis.Figure 2shows, that in the case of a single-crystal, the seismic rays traveling through the basal plane will support the highest shear splitting and a polarization pattern with VSH > VSV. Consequently, the alignment of the crystal lattice proposed above could also explain the observed ray polarization geometry of the shear wave splitting in Tonga (VSH > VSV) for the detected seismic ray paths that travel nearly horizontal through the slab fragment. A better understanding of the potential contribution of phase D to seismic shear anisotropy would require however additional constrains on the deformation mechanism and texture development in phase D at relevant pressure-temperature conditions.
 In conclusion, hydration at depth may be detected seismically if water is stored in phase D and the discussed velocity contrast and anisotropy pattern identified in the present study persist at lower mantle conditions. If such, Phase D may contribute to low velocity zones and seismic shear anisotropy, features that have been seismically observed in deep subduction zones [Wookey et al., 2002; Chen and Brudzinski, 2003; Fukao et al., 2001; Lawrence and Wysession, 2006].
 We thank M. Wörle (ETH Zürich) for access to the X-ray facilities and assistance with the orientation of the samples, and P. Ulmer, A. Saikia and A. Rohrbach for advice on phase D synthesis. This work was supported by ETH Zurich (grant ETH-20 09-2) and SNF (grant 200020-130100/1). We thank two anonymous reviewers for comments that helped to improve this manuscript and M. Wysession for efficient editorial handling.
 The Editor thanks Hauke Marquard and an anonymous reviewer for their assistance in evaluating this paper.