A subset of the single-crystal elastic moduli of natural antigorite has been measured using Brillouin scattering at high-pressure up to 9 GPa. Aggregate properties and axial compressibilities are in good agreement with equation-of-state results from X-ray diffraction. Stiffness along the c-axis increases, becoming close to those within the silicate layer near 7 GPa. A slight discontinuity in the evolution of the elastic moduli near 7 GPa is associated with a phase transition. Raman spectroscopy shows that the transition does not occur in subducting slabs, except in the core of the coldest slabs where serpentine may be subducted to depths of about 200 km. Varying temperature and pressure has limited effects on the interpretation of seismic velocities, principally because of the limited depth range of serpentine stability (mostly above 100 km depth), and also due to the compensating effects of pressure and temperature that maintain velocity variations well within the uncertainties and statistical variability of seismological studies. Serpentinites are excellent candidates for explaining low velocities in the hydrated mantle wedge, inversion of the Moho, and thin anisotropic low-velocity layers at the plate interface. Serpentinite layers provide an alternative explanation to fluid-saturated oceanic crust for explaining these phenomena, and account for a transition to an aseismic, decoupled plate interface.
 The importance of serpentinites for the dynamics and seismicity of subduction zones has been examined in a number of studies, but to be correctly evaluated, precise knowledge of the phase relations, seismic velocities, and density variations of antigorite with pressure is required [Hyndman and Peacock, 2003; Reynard, 2012]. In subduction zones, serpentines form and become an important component of rocks in the mantle wedge where fluids from the dehydrating oceanic crust interact with the mantle [Bostock et al., 2002; Christensen, 2004; Hilairet et al., 2007; Hyndman and Peacock, 2003; Katayama et al., 2009]. Serpentine has several varieties, of which antigorite is the stable form up to 7–8 GPa in subduction zones [Ulmer and Trommsdorff, 1995; Wunder and Schreyer, 1997].
 The elastic modulus tensor of antigorite serpentine has been determined at ambient conditions using Brillouin spectroscopy and combined to Electron Back-Scattering Diffraction (EBSD) fabric measurements to determine the seismic properties and anisotropy of deformed serpentinites [Bezacier et al., 2010]. Ultrasonic measurements were performed on antigorite serpentinite at ambient conditions and at moderate pressures up to 600 MPa [Christensen, 1974, 2004; Kern et al., 1997; Watanabe et al., 2007] and to high temperatures [Kern et al., 1997]. Precise modeling of rock seismic properties can also be obtained from single-crystal elasticity data and textural measurements [Mainprice, 1990]. Results from both methods are comparable, but the combination of elastic moduli and texture or fabric measurements is more flexible because it allows modeling the rock properties over a wide range of textures and conditions, provided that the pressure and temperature dependence of elastic moduli is known.
 There exist no direct data on sound velocities of antigorite or serpentinites beyond 600 MPa. Density Functional Theory (DFT) calculations [Mookherjee and Capitani, 2011] are in general agreement with ambient pressure data but calibration against experimental data at higher pressures are lacking. DFT bulk and shear modulus are often higher than experimentally determined values at ambient conditions. This difference could be due to compositional effects that cannot be evaluated in the absence of data and to intrinsic uncertainties in the DFT calculations. Therefore, it remains desirable to obtain data on natural samples. In this paper, we present the results of high-pressure Brillouin measurements performed on the same sample used at ambient conditions [Bezacier et al., 2010]. The pressure dependence of elastic moduli is obtained to nearly 9 GPa, corresponding to depths of about 250 km, thus, encompassing the stability field of antigorite. Elastic moduli are then used to estimate variations of seismic velocities and anisotropy of serpentinite with pressure and with temperature effects estimated using a quasi-harmonic approximation. The consequences for interpretation of seismic velocities in the hydrated mantle wedge and lithosphere of subducting slabs are discussed.
2 Experimental Procedures
 The same natural single crystal of antigorite used for determination of the elastic moduli at ambient P-T conditions [Bezacier et al., 2010] was loaded in a diamond-anvil cell (DAC) designed for Brillouin measurements [Mao and Bell, 1980; Merrill and Bassett, 1974]. The chemical composition of the sample is (Mg2.62Fe0.16Al0.15)(Si1.96Al0.04)O5(OH)3.57 [Auzende et al., 2002]. Unit cell parameters at ambient conditions are a = 43.59(0.08) Å, b = 9.26(0.02) Å, c = 7.25(0.14) Å, and β = 91.16(0.02)° in the C2/c space group [Hilairet et al., 2006a]. The diameter of the diamond culets was 500 µm. A gasket was prepared from a 250-µm-thick foil of stainless steel that was pre-indented to a thickness of 80 µm and in which a 250 µm-diameter hole was machined with an electro-spark erosion system. The pressure-transmitting medium was a mixture of methanol-ethanol-water in proportions 16:3:1 (by volume). Two ruby chips were placed in the chamber for pressure measurements. Pressure was determined from the ruby fluorescence R1 line shift using the calibration of Mao et al. . In all experiments at pressures above 5 GPa, the diamond cell was externally heated (∼ 75 °C) to achieve stress relaxation in the pressure-transmitting medium and to maintain hydrostatic conditions. The pressure was measured before and after each Brillouin experiment, and the results were always within their mutual uncertainties (±0.03 GPa). No pressure gradient or non-hydrostaticity was detected in the investigated pressure range (<9 GPa).
 An argon ion laser (λ = 514.5 nm), a six-pass piezo-electrically scanned Fabry-Pérot interferometer, and a photon counting silicon avalanche photodiode were used in all experiments [Bass, 1989; Sandercock, 1982]. Two acquisition geometries were used: (1) a platelet/symmetric scattering geometry with an external angle between the incident and scattered beams of 80° and (2) a near-backscattering geometry with an angle of 171°. The DAC was mounted on a three-circle Eulerian cradle for Brillouin measurements. In symmetric geometry, the light scattered at 80° from the incident direction was collected by a lens and spatially filtered through a 150 or 200 µm pinhole. Measurements were taken every 30° by rotating the χ axis over an angular range of 180° at fixed φ and ω. Experiments were performed at eight different pressures. All Brillouin spectra had a high signal-to-noise ratio, especially for the backscattering geometry.
 The single-crystal was a thin flake with lateral dimension of about 100–150 µm in the (001) plane and a thickness of about 30 µm along the c*-axis. The flake was mounted with the (001) plane perpendicular to the DAC axis. Brillouin scattering spectra were collected for 7 to 13 distinct crystallographic directions at high pressures. A linear inversion technique was used to calculate a best-fit model of single-crystal elastic moduli, Cij, from the velocity data [Weidner and Carleton, 1977] using the Christoffel equation [Musgrave, 1970]. Because of the geometry imposed by the platy crystal shape which is determined by the perfect (001) cleavage, the elastic moduli along the c-axis (C33) was obtained independently by measuring velocities in a near-backscattering geometry with a scattering angle of 171°. The refractive index of antigorite at green wavelengths and room P-T conditions was measured to be 1.574 ± 0.004 using immersion fluids [Bass and Weidner, 1984]. C11, C22, C12, and C66 were determined by an iterative scheme from the platelet-symmetric geometry measurements [Sinogeikin and Bass, 2000]. Only one shear mode could be observed in the platelet-symmetric measurements (see supplementary online information). Thus, some of the moduli cannot be determined, and their values have been estimated in the following way: for C13, C23, C44, and C55, we started from the ambient condition values [Bezacier et al., 2010] and accounted for pressure variations of these elastic moduli using the first-principles calculations on serpentine [Mookherjee and Stixrude, 2009; Mookherjee and Capitani, 2011]. The assumed variations are dC13/dP = dC23/dP = 7, and dC44/dP = dC55/dP = 0.88. The elastic moduli C15, C25, C35, and C46 are relevant only to the monoclinic space group of antigorite, and have values that are extremely small compared to other Cij. We assumed these monoclinic-specific Cij to be constant with pressure. This assumption has a negligible effect on the inferred properties of antigorite rocks because of the small magnitude of these moduli.
 In-situ Raman spectroscopy was performed in an externally heated diamond anvil cell, using diamond culets of 500 µm. Pressure was monitored with a temperature-corrected ruby fluorescence scale. Raman spectra were obtained using a confocal LabRam HR800 spectrometer.
3.1 Elasticity of Antigorite at High Pressure
 Because of the alternation of rigid tetrahedral silicate layers and weak OH and octahedral layers, elastic moduli and the velocity of sound waves are high within the (001) plane and low perpendicular to it [Bezacier et al., 2010]. At high-pressure, the difference decreases with increasing pressure (Figure 1a) because of the high compressibility along the c-axis. C33 (corresponding to the stiffness perpendicular to the silicate layers) increases strongly with pressure whereas C11 and C22 (corresponding to longitudinal waves in a and b directions) slightly decrease (Figure 1b). Above 6.81 GPa, C33, C11, and C22 have similar values, approaching longitudinal velocity isotropy. The uncertainties given for the elastic moduli are one standard deviation calculated from root-mean-square (RMS) residual in velocity of the data set at each pressure. Likewise, axial compressibilities become similar at high pressure with βc decreasing with pressure and compression along c-axis, whereas βa and βb remain nearly constant (Figure 2), in agreement with X-ray diffraction data [Hilairet et al., 2006a; Nestola et al., 2010]. Axial compressibilities are calculated by inverting Cij tensor to obtain the Sij compliance tensor with monoclinic symmetry [Nye J. F.1957. Physical properties of crystals: Their representation by tensors and matrices].
 Our measurements show an anomaly in the sound velocities and elastic moduli around 7 GPa (Figures 1a and 1b). Single-crystal X-ray diffraction experiments [Nestola et al., 2010] showed evidence for a phase transition approximately at around 7 GPa. This is consistent with the small (less than 4%) decrease observed in compressional wave velocities perpendicular to (001) at about the same pressure. Within the sheets, compressional wave velocities are not significantly affected by the phase transition, nor are shear sound velocities.
 In addition to pressure effects, the effects of temperature should be accounted for and were estimated using the quasi-harmonic approximation (QHA). In the QHA, sound velocities or elastic moduli are assumed to depend only on the specific volume of the crystalline phase. Temperature effects can be calculated using available data on compressibility (this study) and thermal expansivity [Gregorkiewitz et al., 1996] using the following equations and linear regression: αχij = d(lnCij)/dT and χij = d(lnCij)/d(lnV), where α is the thermal expansion. We set α to a constant value of 3⋅10−5 K−1, and the volume V is taken from [Hilairet et al., 2006b]. Tests on olivine data [Kumazawa and Anderson, 1969] show that the QHA is valid up to 1000 K, within the stability field of antigorite [Ulmer and Trommsdorff, 1995]. Individual elastic moduli exhibit a variation from 2 to 17% from 298 K to 1000 K. In the stability field of antigorite, it will not greatly affect the elastic properties of antigorite. Finally, potential compositional effects on elasticity of antigorite will need to be evaluated, but we estimate that these are negligible with respect to state-of-the-art experimental and computational uncertainties because compositional variability of antigorite from various subduction zones is limited [Reynard et al., 2011].
 Table 1 presents the isotropic aggregate properties (density, KS, μ, compressional velocity (VP), and shear velocity (VS), for Reuss and Voigt approximation) and elastic moduli of the single-crystal. The bulk modulus increases with pressure and the best fit, obtained for a third order Birch-Murnaghan equation of state, gives K0 = 61.0(1.3) GPa and K′ = 6.7(0.9) (this study), in good agreement with K0 = 62.0 (2.2) GPa and K′ = 6.4(1.0) of [Hilairet et al., 2006a] and K0 = 62.9(0.4) GPa and K′ = 6.1(0.2) of [Nestola et al., 2010]. The densities at each pressure (Table 1) were taken from X-ray diffraction equation-of-state results [Hilairet et al., 2006a].
Table 1. Isotropic Aggregate Properties (Density, Bulk and Shear Modulus, Compressional and Shear Velocities for Reuss and Voigt Approximation) of Antigorite Single-Crystal and Elastic Moduli C11, C22, C33, C66, C12 as a Function of Pressure and C44, C55, C13, and C23 in Italics (Assumed, See Text) Numbers in Parenthesis Represent 1 S.D. Uncertainties. Uncertainty in Pressure Is ≤0.03 GPa. Uncertainty Is Estimated to Be <1% in Velocitya
C15, C25, C35, and C46 are constant (2.4, −4.4, 2.5 and −13.1 GPa).
3.2 Raman Spectroscopy at High Pressure and High Temperature
 High-temperature high-pressure Raman spectra were acquired at 25, 60, 100, and 200 °C in order to follow this phase transition in PT space. In Raman spectra, the transition is marked by the progressive appearance of a high-frequency OH stretching mode above 3700 cm−1 that rapidly shifts with pressure [Auzende et al., 2004; Reynard and Wunder, 2006], but this change is too subtle to allow precise tracking of the onset of the transition. In general, changes in properties of antigorite associated with the transition are small, be they volume and cell parameters [Nestola et al., 2010], sound velocities (this study and Figure 1), or Raman frequencies (Figure 3a). We used a small (about 4%) and abrupt increase in frequency of a low-frequency mode (around 190 cm−1) as a marker of the transition. The relative frequency change at the transition is similar to that observed for compressional wave velocity along the c*-axis, but of opposite sign. The change was observed on both compression and decompression, consistent with the reversible character of the transition already observed from X-ray diffraction [Nestola et al., 2010], former Raman studies [Auzende et al., 2004; Reynard and Wunder, 2006], and from Brillouin measurements (Figure 1). Some main features of the spectral changes on compression and decompression are summarized in Figure 3b.
4.1 Elastic Properties of Antigorite and Serpentinite at Mantle Conditions
 The high-pressure high-temperature Raman spectra rule out the possibility of a high-pressure phase transition in natural settings (Figure 3a), even in the coldest subducting slabs [Syracuse et al., 2010]. Thus, in the absence of a displacive transition that may strongly affect the elastic properties, the present high-pressure elasticity data for the low-pressure form can be safely extrapolated to model the seismic velocities of hydrated mantle within the quasi-harmonic approximation described above. The present measurements at high pressure extend the modeling sound velocities of serpentinites to a deep subduction context.
 The sound velocities of isotropic aggregates can be calculated from the single-crystal elastic moduli under the Reuss and Voigt averaging schemes, giving lower and upper bounds appropriate to conditions of uniform stress and strain, respectively. A commonly used value for comparison with seismic data is the Hill average, which is the arithmetic mean of the Voigt and Reuss bounds. In practice, these bounds are similar within about 2% for moderately anisotropic minerals such as olivine and orthopyroxene, which makes the potential error in comparisons between laboratory and seismic data small, especially when considering uncertainties in velocities from seismic models. The situation is very different for serpentine because of its high elastic anisotropy that results in Voigt and Reuss bounds differing by 15–20% for VP, and 25–30% for VS (Figure 4). Static compression data in a hydrostatic medium are comparable to the Reuss bound of the bulk modulus, as appropriate to the boundary condition of uniform stress in these experiments. Ultrasonic data lie between the Hill average and the Reuss bound, perhaps suggesting that the Hill average overestimates the measured aggregate properties at MHz frequency. Quantitative evaluation of these effects is beyond the scope of this article, but would be of interest for mineralogical modeling of seismic properties [Mainprice and Ildefonse, 2009; Mattern et al., 2005; Stixrude and Lithgow-Bertelloni, 2005]. Nevertheless, recent ultrasonic data on polycrystalline samples of serpentines exhibit velocities close to Reuss bound, caused by the platy shape of antigorite grains [Watanabe et al., 2011].
4.2 Seismic Detection of Serpentinites
 In a previous study [Bezacier et al., 2010], we reevaluated the effect of serpentinization on sound velocities of the mantle from ambient pressure determination of the elastic moduli of antigorite, and compared them with available data for other serpentine varieties, and for “dry” peridotites [Browaeys and Chevrot, 2004; Christensen, 2004; Pera et al., 2003]. For that purpose, it is useful to plot seismic velocities as a function of density [Birch, 1961; Reynard et al., 2007], and this can be extended to the present high-pressure data (Figure 5 and Table 1). An isotropic reference value for the mantle at ambient condition has been plotted for a 70% olivine-30% enstatite peridotite [Browaeys and Chevrot, 2004]. High-pressure serpentinite velocities are lower than those of the surrounding mantle, by an amount that varies with the serpentine variety. The strongest effect is observed for chrysotile-bearing serpentinites [Christensen, 2004] and is attributed to effects of the specific texture associated with this low-pressure low-temperature nanotube variety [Reynard et al., 2007]. Chrysotile is metastable and its presence is limited to shallow levels of oceanic serpentinization associated with near-ridge extension and trench-bending faulting. Chrysotile is not observed in subduction-related serpentinization, where antigorite elastic properties are the pertinent phase to describe the effects of serpentinization on seismic properties [Bostock et al., 2002; Christensen, 2004; Hyndman and Peacock, 2003; Reynard et al., 2007]. This shows that an isotropic serpentine aggregate remains detectable at depth when compared to isotropic “dry” mantle peridotite. Velocities of mantle peridotite increase with pressure (VP = 8.5–8.9 km/s, VS = 4.9–5.0 km/s, San Carlos olivine, [Liu et al., 2005]), and VP/VS = 1.78 at 6.5 GPa. For the antigorite single-crystal VRH (Voigt-Reuss-Hill) calculations (Figure 5 and Table 1), VP shows a significant increase with pressure (from 6.7 to 7.5 km/s) whereas VS remains almost constant (3.8 to 3.9 km/s), thus leading to an increase of the VP/VS ratio from 1.82 (2.99 GPa) to 1.94 (8.83 GPa).
 Within the mantle wedge, serpentine is stable down to about 90 km, representing a pressure of about 3 GPa that has a greater effect on velocities than increasing temperature to 600 °C (close to the stability limits of antigorite serpentinite [Hilairet et al., 2006b; Ulmer and Trommsdorff, 1995; Wunder and Schreyer, 1997]). The effect of increasing pressure is to increase the amount of serpentine necessary to explain low mantle-wedge velocities in areas of hot subduction, whereas the effect of increasing temperature is to decrease the implied antigorite content. P-wave velocities as low as 6.4 km/s that are observed in different subduction zones cannot be explained by isotropic Hill averages, even if pressure and temperature effects are taken into account. For example, a low velocity layer with VP in the range 6.4–7.3 km/s is interpreted as serpentine in the Izu-Bonin subduction zone (10–12 km), and VP as low as 6.9 km/s [Kamimura et al., 2002] is observed in Central Japan at 20–40 km depth (~ 1 GPa). Most of these velocities require extensive serpentinization of at least 50% for VP around 7–7.3 km/s and up to 100% for VP lower than 6.7 km/s (Figure 5). Lower velocities may be explained by overestimation of velocities from Hill averaging, as discussed above, or by anisotropy effects [Bezacier et al., 2010]. Whatever the case, high levels of serpentinization in the range 50–100% are required to explain seismic velocities in hot subduction zones, for instance Cascadia [Bostock et al., 2002], Shikoku [Matsubara et al., 2008], and Costa Rica [DeShon and Schwartz, 2004], whereas small degrees of serpentinization are inferred for cold subduction zones like Tohoku-Hokkaido [Nakajima et al., 2009] or Chile [Dorbath et al., 2008; Wagner et al., 2005]. We note, however, that studies further south yield low mantle-wedge velocities that suggest large heterogeneities of mantle wedge hydration along the trench direction [Bohm et al., 2002].
4.3 Anisotropy and Ultra-low Velocity Layers
 Seismic anisotropy can be caused by the intrinsic anisotropy of the single-crystals associated with Lattice-Preferred Orientation (LPO) and also by the alignment of heterogeneities of specific shape (SPO or Shape Preferred Orientation) [Park and Levin, 2002; Savage, 1999]. Previous studies of the distribution of anisotropy in the mantle wedge suggest that LPO is the dominant factor [Fouch and Fischer, 1998]. Trench-parallel shear-wave splitting in the mantle wedge was tentatively attributed to serpentine LPO [Katayama et al., 2009]. The P-wave anisotropy (AVP) defined as 100 (VPmax − Vpmin)/((VPmax + VPmin)/2) is 47% and AVS = 75% for single-crystal of antigorite at ambient conditions [Bezacier et al., 2010]. Anisotropy decreases with increasing pressure (at 7 GPa: AVP = 22%, AVS = 63%, Figure 6a).
 Antigorite aggregate anisotropy was calculated at high pressure by combining measured LPO [Bezacier et al., 2010] and the present high-pressure elastic moduli dataset. The maximum anisotropy decreases with increasing pressure owing to the large compressibility along the  axis (Figure 6a) from 37% at 0 GPa to 12% at 8 GPa (for VP). Maximum anisotropy of VS decreases only slightly and remains extremely high when compared to that of other mantle minerals, such as olivine, at the stability limits of serpentine near 7 GPa (AVS = 43%). Serpentine may be preserved up to 7 GPa only in the slab lithospheric mantle of the coldest subduction zones [van Keken et al., 2011] before transforming to phase A and enstatite [Stalder and Ulmer, 2001]. In the mantle wedge, serpentine will fully dehydrate near 3 GPa due to increasing temperature, and its anisotropy will remain very high even for P-waves. High anisotropy will remain a signature of serpentinites [Bezacier et al., 2010; Katayama et al., 2009; Kern et al., 1997]. Thus, interpretation of shear wave anisotropy in the mantle wedge is not significantly affected by varying pressure and temperature. Although foliated serpentinites can contribute to the observed shear wave splitting [Katayama et al., 2009], their exact contribution and geometry is difficult to reconcile simply with details of the ray path geometry [Bezacier et al., 2010].
 The variations of VP and VS as a function of the incidence angle of a seismic ray to the foliation (Figures 6b and 6c) show that increasing pressure up to 3 GPa can dramatically change the pattern of anisotropy. Increasing temperature with depth will partly compensate the effect of increasing pressure. Velocities at 90° to the foliation are remarkably low, in the range 5.9–6.7 and 2.8–3.1 km/s for P and S waves, respectively. Even lower values may be reached if the Hill average, used in our calculations here, is an upper bound as discussed previously. Such values are similar to those inferred for thin (1–2 km) anisotropic layers at the plate interface from receiver function imaging in Cascadia [Abers et al., 2009; Bostock et al., 2002; Nikulin et al., 2009; Park et al., 2004], or from waveform modeling in Mexico [Song et al., 2009]. These ultra-low velocity layers are interpreted as water-saturated oceanic crust [Audet et al., 2009; Peacock et al., 2011; Song et al., 2009], but the present elasticity results, indicating deformed serpentinites at the plate interface, are a valid alternative explanation. Highly deformed serpentinites with low viscosity or yield strength [Amiguet et al., 2012; Hilairet et al., 2007; Hilairet and Reynard, 2009] also account for plate decoupling down to 80 km, which matches the cold nose inferred from heat flux and seismic attenuation data [Abers et al., 2006; Wada et al., 2008].
 Bin Chen helped with the diamond cell experiment at UIUC. This study was supported by ANR project SUBDEF grant n° ANR-08-BLAN-0192 to B.R., CNRS-UIUC international exchange program, and by NSF grants EAR-0738871 and EAR 10–43050 (COMPRES) to J.D.B. The Raman facility in Lyon is supported by the Institut National des Sciences de l'Univers.