Phase relations in harzburgite have been determined between 14 and 24 GPa and 1473 and 1673 K. At 1673 K, harzburgite transformed to wadsleyite + garnet + clinopyroxene below 19 GPa and decomposed into an assemblage of ringwoodite + garnet + stishovite above 20 GPa. Certain amounts of akimotoite were produced at still higher pressures (22–23 GPa). Finally, perovskite and magnesiowüstite were found to coexist with garnet at 24.2 GPa. Compositions of all the phases were analyzed and elemental partitioning coefficients were determined among coexisting phases. Combining our experimental data with available thermoelastic properties of major minerals in the earth's mantle, we modeled the velocity and density signatures of the stagnated oceanic slab in the mantle transition zone (MTZ) under eastern China, based on kinematic slab thermal structure analysis. We examined two end-member slab models: a conventional straight slab with deformation thickening and an undulate slab with an oscillating wavelength of 200 km. We found that an undulated (buckled) slab model yields velocity anomalies (about 1–2% for Vp) that are consistent with seismic tomography models, taking into account low-pass filtering effects in seismic tomography studies. On the other hand, straight slab models yield velocity anomalies that are too high compared with seismic tomography models. Our models provide important constraints on the thermal structure, mineralogy, composition, density, and velocities of slab materials in the MTZ.
 Subducted materials play an important role in affecting chemical composition and structure of the mantle transition zone (MTZ). Harzburgite is generally accepted as an important part of subducting slabs, overlain by a layer of basalt and underlain by a layer of depleted peridotite (lherzolite) [Ringwood, 1982; Ringwood and Irifune, 1988]. Seismic tomography studies have detected well-developed widespread fast anomalies in the MTZ and lower mantle around the circum-Pacific, southern Europe, southern Mexico, and southern America [e.g., van der Hilst et al., 1991; Fukao et al., 1992, 2001, 2009; Zhao, 2004]. These anomalies have been interpreted as stagnant oceanic lithosphere materials as these regions are closely associated with subduction zones, where the oceanic lithosphere plunges deeply into the mantle. Under eastern China, where the Pacific oceanic lithosphere has been subducting deep into the mantle since the Late Mesozoic [Li and van der Hilst, 2010], continuous fast anomalies (1-–2% faster than the surrounding mantle) are imaged at depths of ca. 500–660 km, extending from the Japan Wadati-Benioff zone to Ordos Block, over a longitude interval of nearly 30° [e.g., Huang and Zhao, 2006; Li and van der Hilst, 2010; Pei and Chen, 2010]. P-wave triplication studies have also detected a high-velocity and low-gradient layer between ~600 and 660 km, with maximum anomalies about 2–3% higher than the global average P-wave velocity at these depths [Tajima et al., 2009; Wang and Chen, 2009; Wang and Niu, 2010]. Receiver-function imaging studies show complex discontinuity structures in the lower part of MTZ, with transition zone thicknesses about 30–40 km greater than the global average [e.g., Ai et al., 2003; Chen and Ai, 2009]. These are also thought to be closely related to subducted Pacific lithosphere.
 Fast anomalies can be induced by low temperature, different composition/mineralogy, or a combination of both. Driven by gravitational forces, these thermal and/or compositional heterogeneities either float or sink through the mantle, facilitating thermal and chemical exchanges therein. Systematic experimental studies on slab materials at high pressure and high temperature (HPHT) are crucial in testing the stagnant slab hypothesis and providing quantitative constraints on thermal structure, mineralogy, composition, density, and velocities of slab materials in the deep mantle. However, previous HPHT experiments mainly focused on phase transitions of basalt [e.g., Irifune and Ringwood, 1987a, 1993; Hirose et al., 1999; Sanehira et al., 2008]; few reports on harzburgite are available [Irifune and Ringwood, 1987b]. By using a “harzburgite minus olivine” composition, Irifune and Ringwood [1987b] reported the phase relation of the residuum, but the interaction between “olivine normative component” and “pyroxene normative component” has been ignored in their study.
 In this study, we conducted HPHT experiments on a natural harzburgite at 1473–1673 K and 14.1–24.2 GPa. We first investigated mineralogy of subducted harzburgite under pressure and temperature conditions corresponding to the MTZ. Combining the experimental data with available thermoelastic properties of the high-pressure minerals and kinematic slab thermal structure analysis, we next examined velocity and density signatures of stagnated slabs. By comparing our results with seismic tomography studies, we evaluated thermodynamic and mechanical states of stagnated slabs in the MTZ under eastern China.
2 Experiment Procedures
2.1 Starting Materials
 The harzburgite starting material was synthesized by mixing approximately 82 wt % olivine with 18 wt % enstatite. Both minerals were hand-picked from a natural peridotite sample (DMP-018) from Damaping (Hannuoba), Hebei Province, China. For detailed description of the geological setting and the sample, see Liu et al. [2005a] and Wang et al. . The olivine and enstatite crystals were first crushed and ground into fine powder (grain size 5–15 µm) separately, then ground together in alcohol with an agate mortar to produce a homogeneous mix. The chemical composition of the harzburgite (Table 1) thus prepared is similar to that from North Atlantic [Michael and Bonatti, 1985] and East Pacific Rise [Dick and Natland, 1996]. Compared with pyrolite, harzburgite is generally lower in alumina and calcia but higher in magnesia. The starting material used by Irifune and Ringwood [1987b], a “harzburgite minus 82 wt % olivine” composition consisting mainly of enstatite with minor amount of spinel (MgAl2O4), is also shown in Table 1 for comparison. Water content of the starting material was measured by FT-IR spectroscopy using the method of Bell et al. [1995, 2003], yielding about 60 ppm (by weight) H2O in enstatite and no water in olivine. Water contents in the run products could not be measured reliably as crystal sizes were too small.
Table 1. Compositions of Harzburgite Used in this Study and Previous Work
Harzburgite minus olivine
Pyrolite: McDonough and Sun . North Atlantic: harzburgite from North Atlantic, Michael and Bonatti . East Pacific: harzburgite from East Pacific, Dick and Natland . Harzburgite minus olivine: Irifune and Ringwood [1987b].
2.2 High-Pressure, High-Temperature Experiments
 HPHT experiments were carried out in a 1000-ton Kawai-type multi-anvil apparatus at the Laboratory for Study of the Earth's Deep Interior (SEDI Lab), China University of Geosciences (Wuhan). Cell assemblies, with 10 and 8 mm octahedral edge lengths and 5 and 3 mm truncation edge lengths (TEL), respectively (hereafter referred to as the 10/5 and 8/3 assemblies, respectively; Figure 1), were used on 25.4 mm edge-length WC cubes. The octahedral pressure media were made by injection-molded polycrystalline spinel, as developed by Leinenweber et al.  in a project supported by the Consortium for Materials Properties Research in Earth Sciences to develop standardized high-pressure assemblies for various conditions. A LaCrO3 sleeve was placed inside of the spinel octahedron for thermal insulation. A rhenium furnace was used for heating. The sample, contained in a Re capsule, was insulated from the furnace by an alumina inner capsule. Type C W-Re thermocouple was used to measure the temperature at the center of furnace. Typical temperature variation during the experiments was within ±10°C, without correcting for the pressure effect on electromotive force.
 Figure 2 shows the pressure calibration curves used in the present study. For the 10/5 assembly (Figure 1a), high-temperature pressure calibration was performed using the olivine-wadsleyite (14.6 GPa [Katsura and Ito, 1989; Fei and Bertka, 1999; Katsura et al., 2004]) and wadsleyite-ringwoodite transitions (20 GPa [Katsura and Ito, 1989; Suzuki et al., 2000]) in Mg2SiO4 at 1673 K, as well as the phase transition of coesite-stishovite in SiO2 at 1473 K (9.2 GPa [Zhang et al., 1996]). For details of the cell assembly and pressure calibration see Wu et al. . For the 8/3 assembly (Figure 1b), pressure calibration was made according to the olivine-wadsleyite transition (14.6 GPa [Katsura and Ito, 1989; Fei and Bertka, 1999; Katsura et al., 2004]) in Mg2SiO4 at 1673 K, wadsleyite-ringwoodite (21.3 GPa [Suzuki et al., 2000]) and the ringwoodite to perovskite+periclase transformations (23.1 GPa [Fei et al., 2004]) in Mg2SiO4 and akimotoite-perovskite in MgSiO3 (22.3 GPa [Hirose et al., 2001; Fei et al., 2004]), all at 1873 K. The calibration curves appear temperature independent between 1473 and 1873 K, and reproducibility of pressure in the present experiments was estimated to be about ±0.3 GPa and ±0.5 GPa, respectively, for the 10/5 and 8/3 assemblies.
 Prior to each experiment, the cell assembly, with samples loaded, was dried at 393 K for 24 h in an oven. In all experiments, the samples were first compressed to the desired ram load, then heated at a rate of ~100 K/min and held at the constant target temperature for the desired duration, which varied from 5 to 24 h. Each experiment was terminated by switching off power to the furnace (quenching), followed by automatic decompression to atmospheric pressure at a rate of about -1.0 GPa/h. After recovery, the Re capsule containing the samples was mounted in a plastic ring filled with epoxy resin and polished for phase identification and compositional analysis.
2.3 Analysis of Run Products
 Microfocus X-ray diffraction (X'Pert PRO DY2198) was used to identify phases in the recovered run products. A field-emission scanning electron microscope (Quanta 2000-type) (FE-SEM), fitted with a LINK EDS detector, was used to study microstructure of the recovered samples. Quantitative composition analyses were conducted with an electron probe microanalyzer (EPMA-1600) at the University of Science and Technology of China, using an accelerating voltage of 15 kV, a beam size of 1.0 µm, and a beam current of 20 nA. General measurement errors are about 3% for elements of less than 1 wt %, 1 % for elements of less than 5 wt %, and 0.1 % for elements of greater than 10 wt %.
3 Experiment Results
 Figure 3 shows X-ray diffraction profiles of representative experiments. Details of the experimental conditions and run products are summarized in Table 2. Figure 4 shows representative back-scattered electron images of typical run products. Mineral proportions were calculated based on mass balance. Chemical compositions and densities of minerals, listed in Tables 3 and 4, were used to determine the volume fractions (vol%) of minerals in harzburgite (Table 2 and Figure 5). It can be seen that the effect of temperature on the phase proportion is negligible between 1473 and 1673 K.
Table 2. Summary of Experimental Conditions and Run Products
 For experiments below 16.1 GPa, granular and columnar garnet crystals formed with grain size about 1–5 µm, inside former clinopyroxene crystals (Figure 4a). Wadsleyite grains exhibited well-developedtriple-junctions and concordant relationships with garnet and clinopyroxene crystals (Figure 4b), suggesting that equilibrium was reached. The volumetric proportion of garnet increased significantly with pressure until 19 GPa (Figure 5). Between 20 and 22 GPa, small (<1 µm), needle-like stishovite crystals formed within large garnet grains, in association with ringwoodite (Figure 4c). The exsolution of stishovite from garnet resulted in an abrupt increase in ringwoodite volume fraction between 20 and 22 GPa (Figure 5). The small crystal size of stishovite in garnet matrix is a consequence of exsolution reaction in garnet and should not be considered as evidence of disequilibrium. Preferred orientations of the stishovite “needles” are a potential source of seismic anisotropy for a deformed slab, although we did not find systematic lattice preferred orientation of stishovite in our quasi-hydrostatic experiments. When pressure increased to 22–23 GPa, granular and short columnar akimotoite polycrystalline aggregates, similar to those found in the shocked-metamorphosed Tenham chondrite [Tomioka and Fujino, 1999], were observed to be in contact with ringwoodite (Figure 4d and e), accompanied by the disappearance of stishovite (Figure 5). The volume fraction of akimotoite decreased with increasing pressure. At 24.2 GPa, perovskite and magnesiowüstite were formed together with majoritic garnet (Figure 4f). The perovskite + magnesiowüstite assembly shows a typical texture after disproportionation reaction from ringwoodite and is virtually identical to that reported previously [Yamazaki et al., 2009]. Here again, the fine grain size should not be considered as evidence of disequilibrium. Because of the small grain size, however, reliable elemental partitioning information could be obtained only in areas where different minerals in contact were large enough to yield accurate chemical composition analysis. Results of the analyses of major minerals are given in Table 3.
 Figure 6a illustrates compositional changes in the olivine-normative component (olivine, wadsleyite, and ringwoodite) as a function of pressure at 1673 K. Fe preferentially partitions into wadsleyite at 14.1 GPa when olivine and wadsleyite coexist. Figure 6b shows compositional variation of garnet (oxygen number assumed to be 12) with pressure. Both Al and Fe contents decrease with pressure, while Si and Mg contents increase during the transformation from olivine to wadsleyite. Between 19 and 20 GPa, Si content drops and that of Al increases due to the wadsleyite-ringwoodite transformation and the formation of stishovite. With the formation of akimototite at 22–23 GPa, cation numbers of Mg, Fe, and Al decrease with an increase of Si.
 Figure 6c shows iron partition coefficients between the olivine- and the pyroxene-normative components (clinopyroxene, garnet, and akimotoite) at 1673 K as a function of pressure. Here the partition coefficient of Mg and Fe between two phases α and β is defined by Kα/β = (XFe/XMg)α/(XFe/XMg) β, where XFe = Fetotal/(Mg2++Fetotal). Below 14.1 GPa, Fe partitions preferentially into garnet and olivine, relative to olivine and clinopyroxene, respectively. Between 16 and 19 GPa, Fe partitions roughly equally between wadsleyite and garnet, while clinopyroxene becomes Fe-poor relative to wadsleyite by a factor of ~2. These values are generally in accordance with earlier experimental results for a pyrolite system under similar conditions [Irifune and Isshiki, 1998]. After wadsleyite transforms to ringwoodite, KRw/Grt increases from ca. 1.0 to 1.5 at 20–23 GPa, while KRw/Ak decreases from ca. 2.9 to about 2.1 at 22–23 GPa. With increasing pressure, Fe content in akimotoite increases relative to ringwoodite, while volume fraction of akimotoite decreases (Figure 5), consistent with previous experimental studies [Ito and Yamada, 1982] and thermodynamic calculation [Fabrichnaya, 1995] on the (Mg,Fe)SiO3 system. These consistencies suggest that equilibrium was achieved in the present experiments. One previous study [Irifune and Ringwood, 1987b], using a harzburgite-minus-olivine composition, reported that volume fraction of akimotoite increases with increasing pressure. We attribute this discrepancy to the interactions between the olivine and non-olivine components: in the “residuum” composition of Irifune and Ringwood [1987b], such interactions are absent. Increasing Fe content may be a factor controlling volume fraction of akimotoite. With the formation of stishovite and akimotoite, CaSiO3 perovskite should appear according to mass balance and Ca partition data. The amount is small (<0.1 vol.%) and therefore not shown in Figure 5.
 Partition coefficients of Al between akimototite and garnet is defined by KAl = Ak(Al+Cr)/(Si+Ti)/Grt(Al+Cr)/(Si+Ti), and the results are also shown in Figure 6c (vertical axis to the right). KAl increases from ~0.3 to ~1.0. At low pressures, Al partitions almost exclusively into garnet. Partitioning measurements on Ca have large uncertainties because Ca content is low in harzburgite (<0.5 wt %). These data are therefore not presented.
 The interaction between the olivine- and pyroxene-normative components is primarily through partitioning of Fe on the Mg site in the phases involved [Weidner and Wang, 2000]. It can be seen from Figure 6c that the partition of Fe between wadsleyite, ringwoodite, and garnet changes very little, indicating that the interaction between olivine- and pyroxene-normative components has little effect on the transformation of wadsleyite to ringwoodite. Between 20 and 22 GPa, the formation of stishovite and akimotoite affects the ringwoodite-versus-garnet proportion significantly in harzburgite and will cause detectable changes in seismic velocities and density (to be discussed below).
4 Mineral Physics Analysis on Subducted Lithospheric Materials in the MTZ
4.1 Buoyancy Relationship of Harzburgite Relative to a Pyrolitic Mantle
 We used the third-order high-temperature Birch-Murnaghan equation of state to calculate densities of minerals at high-temperature and high-pressure conditions. Procedures used for the calculation along the subduction geotherm were the same as that of Irifune and Ringwood [1987b]. Thermoelastic parameters of major minerals listed in Table 4 are used to calculate densities of harzburgite and pyrolite (Figure 7a), as well as basalt at various high-pressure and high-temperature conditions. Phase relations and mineral variations for harzburgite are based on this study, and those for pyrolite and basalt are from Irifune and Ringwood [1987a]. The following Clapeyron slopes of the phase transitions involved were used in this study, olivine-wadsleyite: 0.0036 GPa/K [Morishima et al., 1994], wadsleyite-ringwoodite: 0.00691 GPa/K [Suzuki et al., 2000], ringwoodite-perovskite + magnesiowüstite: -0.0013 GPa/K [Fei et al., 2004], garnet-stishovite+ringwoodite: 0.006 GPa/K [Sawamoto, 1987], stishovite+ringwoodite-akimotoite: -0.002 GPa/K [Sawamoto, 1987], akimotoite-perovskite: -0.0029 GPa/K[Ito and Takahashi, 1989; Ono et al., 2001; Hirose et al., 2001; Fei et al., 2004], clinopyroxene-garnet: 0.002 GPa/K [Irifune and Ringwood, 1987b]. Effects of major element partitioning are taken into account whenever composition dependence on density and elasticity is available.
 Density profiles of harzburgite and pyrolite were calculated along a chosen subduction zone geotherm [Litasov et al., 2006] and a “normal” mantle geotherm [Katsura et al., 2010], respectively (Figure 7a). Figure 7b shows the difference in density between the two profiles. Harzburgite is ~0.1 g/cm3 denser than pyrolite between ~420 and 650 km, and ~0.2 g/cm3 less dense between 650 and 680 km. Figure 7c shows vertically integrated density contrast of a detached harzburgite layer relative to the surrounding mantle as a function of mean layer depth. The buoyancy contrast was obtained by assuming a slab stagnant at different depths, with integrated density contrast over the entire vertical slab thickness. The integrated density contrast, denoted by D(h)integrate, is calculated using the following equation: , where h is the average depth of the stagnant slab, H_top and H_bottom are the depths of the top and bottom surfaces of the slab, respectively, ρharzburgite and ρpyrolite are the density of harzburite and pyrolite, respectively, and g is gravitational acceleration. Through integration, D(h)integrate represents density contrast-induced normal stress in the vertical direction, in units of (g/cm3)(N/kg)(km) = (103kg/m3)(N/kg)(103m) = 106 Pa = MPa.
 For various thicknesses from 20 to 80 km, the harzburgite layer is denser in the MTZ until ~660 km depth owing to combined effects of composition, thermal, and phase relations. Considering the upper basalt layer, however, the overall subducted oceanic lithosphere is denser than the surrounding mantle (Figure 8) and slabs will be driven to the bottom of the MTZ.
 Mantle dynamics is a complex process and cannot be determined by simple gravitation buoyancy alone. Karato  analyzed the mechanical behavior of subducted lithosphere and concluded that if garnets in the crustal component of the slab possess higher strength relative to the surrounding mantle, mechanical interaction may result in “peeling” of the crustal layer from harzburgite by folding. The decoupled harzburgite layer alone will be negatively buoyant and sink toward the bottom of the MTZ. On the other hand, slab bending moment, viscosity variations, and trench rollback dynamics [e.g., Bina and Kawakatsu, 2010] may provide positive buoyancy forces for the subducted slab as a whole. Beneath eastern China, numerous high-resolution seismic studies have shown that high P-wave velocity anomalies are located inside the MTZ; no clear fast anomalies exist below the 660 km discontinuity [e.g., Huang and Zhao, 2006; Zhao et al., 2009; Li and van der Hilst, 2010; Pei and Chen, 2010]. In the following, we attempt to model the thermal and physical properties of the anomalies beneath eastern China, by assuming that they are due to subducted oceanic lithosphere in the MTZ.
4.2 Thermal Modeling of the Subducting Slab for Eastern China: Two End-Member Models
 Two slab models were considered in this study. We examine a horizontal length of stagnated slab of ~1500 km (white boxes in Figure 9), corresponding to recent tomography studies beneath eastern China [e.g., Huang and Zhao, 2006; Li and van der Hilst, 2010]. Slab thermal models were calculated using the kinematic finite-difference heat-transfer algorithm TEMSPOL [Negredo et al., 2004], as modified by Bina and Kawakatsu  to allow prescribed changes in slab dip angle at depth, with further modification to allow for deformation of the stagnant slab in the MTZ. Thermal structure was calculated assuming uniform thermal conductivity throughout the region of interest; lateral temperature variations arise largely from the residence time of the stagnant slab.
 Model I deals with a standard flat slab, in which a slab subducts into the MTZ and bends to become horizontal just above the 660 km seismic discontinuity, similar to that modeled in Bina and Kawakatsu . We adopted the following thermal parameters appropriate for Japan Wadati-Benioff zone and east China: a 130 Ma lithosphere with a convergence rate of 8.5 cm/yr and a dip angle of 35o. We assumed a slab thickness of 120 km, consisting of a top basalt upper layer of 15 km, a harzburgite layer of 60 km, and a lherzolite layer of 45 km, to approximate thickening by pure-shear deformation of the slab within the MTZ, with a uniform thermal conductivity of 3.2 W/m/K throughout. Detailed temperature distribution in the slab is given in Figure 9a.
 Model II kinematically takes into account an undulated (buckled) slab, a geometry that is increasingly recognized as a consequence of trench retreat and rheological contrast with the surrounding mantle (e.g., Christensen, 2001; Schmid et al., 2002; Ribe et al., 2007; Yoshioka and Naganoda, 2010; Lee and King, 2011; Li and Ribe, 2012; Čížková and Bina, 2012). Buckling deformation of the stagnant slab is represented by imposing, after the first 100 km of horizontal offset, a sinusoidal oscillation with both horizontal half-wavelength and vertical amplitude of 100 km. (Note that λ=200 km is chosen as a contrasting end member to the effective λ=∞ of the flat slab; oscillation periods in recent dynamic models [Lee and King, 2011; Čížková and Bina, 2012] suggest that λ=1000 km may be a more physically realistic intermediate value, although real slabs are unlikely to exhibit strictly harmonic buckling.) This model has an upper harzburgite layer of 45 km underlain by a depleted peridotite (lherzolite) layer of 50 km; the lherzolite layer in both models is approximated by pyrolite. The upper basalt layer is assumed to be separated from the subducted lithosphere due to folding of the slab and viscosity contrast between the crustal layer and surrounding mantle [Karato, 1997]. Figure 9b shows the thermal structure of the undulated slab model. Basic slab parameters were identical to the straight slab model (dip = 35o, velocity = 8.5 cm/yr, age = 130 Ma), except for a thermal conductivity of 4 W/m/K.
 The calculated temperatures in the straight slab range from 1100 to 1873 K throughout the slab, greater than the thermal variations estimated by joint mineral physics and seismic travel-time analysis (1400–1600 K) [Ritsema et al., 2009], but with a similar mean value. The buckled slab has relatively high temperatures, ca. 1273–1673 K, similar to those obtained by Ritsema et al. , as a result of the greater slab residence time and thermal equilibration within the MTZ in model II.
4.3 Velocity Signatures of the Slab Models
 The widely used Voigt-Reuss-Hill averaging method was adopted to approximate the aggregate velocities, assuming that harzburgite, pyrolite, and basalt are all isotropic mineral aggregates [Watt et al., 1976; Bina and Helffrich, 1992]. Thermoelastic parameters of major minerals listed in Table 4 and variations of mineral proportions based on this study and Irifune and Ringwood [1987a] are used to calculate velocities of harzburgite, pyrolite, and basalt at HPHT conditions. Because of compositional similarity between lherzolite and the surrounding mantle (e.g., pyrolite), the lherzolite layer in both models is approximated by pyrolite; all layers are assumed dry. Major phases in the two models are shown in Figures 10a and 11a, respectively. In the straight slab model, the harzburgite layer is mainly composed of ringwoodite, garnet, and stishovite. In the undulated slab model, ringwoodite, garnet, akimotoite, and stishovite are the main constituent minerals.
 Figure 10b–d shows velocity and density signatures in model I. The velocity and density anomalies are calculated relative to a pyrolitic mantle and are given in percent. P- and S-wave velocities of the basalt layer are about 2–3% slower than the surrounding mantle, consistent with recent ultrasonic measurements on oceanic crustal materials [Kono et al., 2007] and seismically observed anomalies under southern Africa [Shen and Blum, 2003], where a layer of accumulated oceanic crust was proposed to be stagnant. The harzburgite layer exhibits high velocities between ~550 and 620 km depths (Figure 10b and c) due to lower slab temperature and higher ringwoodite content than the surrounding mantle (Figure 5). Relative to the surrounding mantle, anomalies are +3–5% and +4.5–7.5%, respectively, for Vp and Vs. Figure 10d shows density anomalies caused by the stagnant slab. The upper basalt layer is about 5–8% denser than the surrounding pyrolitic mantle due to its high garnet content. The middle harzburgite layer and the lower depleted pyrolite layer are about 1–3% and 0–1% denser than pyrolitic mantle, respectively, due to the relatively low temperature in these layers.
 Figure 11b–d shows the velocity and density signatures in model II. Amplitudes of the calculated anomalies relative to a pyrolitic mantle are ~2–3.5% and 3–5% for Vp and Vs, respectively (Figure 11b and d). The relative low magnitudes of velocity anomalies in model II are primarily due to higher temperatures. Density anomalies (relative to pyrolite mantle) are about 4–6% in the upper part of MTZ, mainly caused by the relatively higher wadsleyite content in the harzburgite layer relative to pyrolite (Figure 11d). In the lower part of MTZ, density anomalies are about 0–2%, caused by relatively low temperature and high ringwoodite content.
5 Discussion: Stagnant Slab under Eastern China—Straight or Buckled?
5.1 Comparison with Seismic Tomography Data
 Before comparing mineral physics results with seismic tomography studies, effects of quality factor Q on seismic velocities need to be considered [e.g., Karato, 1993]. Anharmonicity contributions to temperature dependence of seismic velocities become especially important when Q is low. However, an accurate estimate of the effect of Q on seismic velocities is difficult because anelasticity of high-pressure minerals is poorly known. For a first-order approximation, we assume Q values for the slab and the surrounding mantle to be of the same magnitude (~400, see Dziewonski and Anderson, 1981). We then estimate the resultant -∆Vp/Vp caused by anelasticity using the ∂lnVp/∂T profile presented by Karato , assuming that average temperatures are approximately 1273 and 1773 K for the stagnant slab and the surrounding mantle, respectively. The corresponding -∆Vp/Vp values are very similar, about 0.01–0.02 and 0.01–0.03 for the stagnant slab and the surrounding mantle, respectively. Therefore, the effect of anelasticity is (to first order) to decrease the overall velocities by ~1–2%, without affecting velocity anomalies (i.e., relative velocity contrast) in our calculation results.
 Seismic tomography studies, relying primarily on P-wave travel times, report maximum P-wave velocity anomalies about +1–2% relative to average seismological mantle models (e.g., iasp91, Kennett and Engdahl, 1991) beneath eastern China [e.g., Huang and Zhao, 2006; Li and van der Hilst, 2010; Pei and Chen, 2010], much lower than that obtained by our mineral physics models. At least two important factors must be considered when comparing mineralogically derived velocity images with those obtained by seismic tomography studies. First, tomographic images generally underestimate magnitudes of velocity anomalies, because “regularization” of the tomographic inversion reduces the amplitudes of anomalies. Second, tomographic images are generally “smeared” both laterally and vertically (i.e., they are effectively low-pass filtered), so that sharp (i.e., high-frequency) features are damped, thereby spreading anomalies spatially while reducing their local magnitudes. Although we cannot fully reproduce the former in our mineralogical images, we attempt to approximate the latter by applying low-pass spatial (essentially moving-window) filtering to the mineralogical images. (In doing so, we assume spatially uniform tomographic resolution, as modeling the actual spatially varying resolution would require greater knowledge of details of the seismological data set used to constrain the velocity inversion).
 Here we assumed that the horizontal and vertical resolutions under eastern China are about 300 km and 100 km, respectively, roughly consistent with recent tomography studies [e.g., Li and van der Hilst, 2010]. We then convolved horizontal (6σ in 300 km) and vertical (6σ in 100 km) Gaussian filters with the mineralogical P-wave velocity structure, as most tomography studies were based on Vp. Figures 10e and 11e show results after convolving a 300 x 100 km Gaussian filter with two model structures of the MTZ (Figures 10b and 11b). For model I, the low-velocity region caused by the upper basalt layer disappear due to the reduction in spatial resolution. Amplitudes of the P-wave anomalies caused by the harzburgite layer are reduced by a factor of ~2 for the horizontal slab model to about 1.5–3.5%, still significantly higher than seismic tomography results. In addition, the fast anomalies persist to much greater distance than seismic tomography detects. For model II, the sharp, sinusoidal velocity anomalies are smeared to a smooth near-horizontal feature. The amplitude of P-wave velocity anomalies is reduced to about 1–2% throughout the slab except for where slab just enters the MTZ (Figure 11e). The horizontal extent of the fast anomaly is greatly reduced compared to model I. Overall, the buckled model (after filtering) better matches the P-wave anomalies observed in seismic tomography.
5.2 Comparison with Seismic Triplication Models
 Analyses of P- and S-wave triplication near the 660 km discontinuity provide more details on velocity variations with depth at the bottom of MTZ. Unfortunately, it is difficult to derive quantitative error bounds on discontinuity depths and amplitudes from triplication data [Kennett and Engdahl, 1991], due in part to the inherent trade-offs between the discontinuity properties (sharpness, topography, etc.) and velocities immediately below and above the discontinuities. An advantage of this type of modeling, however, is that it often provides a complete velocity versus depth profile extending from the MTZ to the top of lower mantle.
 For model I, the (unfiltered) P-wave velocity anomaly amplitudes (3–5%) are significantly greater than those in the triplication studies for east China (Figure 12). These velocity profiles of mineral physics models have been corrected for anelasticity effects as we discussed earlier. On the other hand, simulating numerically effects of slab buckling on seismic triplication based on model II is beyond the scope of this study. A main effect of lateral variation in seismic structure would be to introduce greater uncertainties into the depths of seismic velocity contrasts and their adjacent velocity gradients, thereby reducing the apparent anomaly magnitudes. Despite the uncertainties in velocity magnitude comparison, it is informative to examine the nature of velocity variation with depth. All P-wave triplication models for eastern China exhibit anomalously high gradients above ~600 km and much lower, or even negative, gradients below (Figure 12); both of these features are observed in our kinematic models. Of course, the undulating shape of the slab simulated here is schematic and highly simplified. For seismic signals dominated by wavelengths comparable to or shorter than slab undulation wavelengths, seismic waveforms will be most distorted. The lack of direct knowledge of the shape of the slab in this region makes direct comparison with triplication data impossible. However, recent analysis [Stähler et al., 2012] incorporating finite-frequency effects, by which “rays” sample a broader region of the MTZ, suggests that triplication signatures can still be recovered in the presence of meaningful topography.
 Buckling of subducted oceanic lithosphere has been observed in both numerical studies with complex geometries and rheologies and in laboratory simulation experiments [e.g., Christensen, 2001; Schmid et al., 2002; Bellahsen et al., 2005; Ribe et al., 2007]. Numerical models have shown that slab thickening observed in seismic tomographic images cannot be explained by “pure shear” deformation, which can only thicken the slab by at most a factor of two, but can be easily interpreted by the buckling mechanism [Ribe et al., 2007]. Recent numerical results also show that slab buckling may be a universal feature in many subduction zones, and can reconcile a variety of puzzling geological and geophysical observations, especially when the slab has a relatively slow trench migration and a relatively large viscosity increase across the 660 km discontinuity [Lee and King, 2011; Čížková and Bina, 2012]. In the present study, we provide additional evidence for the slab buckling model through kinematic thermal structure analysis and mineral physics calculations. Although only two end-member models are analyzed, the consistency of velocity signatures between our bucked slab model and seismic observations indicates that a certain degree of buckling should represent the real behavior of the stagnated slabs beneath eastern China.
 We present HPHT experimental results on a natural harzburgite under MTZ conditions and provide important constraints on the mineralogical constituents of subducted harzburgite materials. The buoyancy relationships between subducting oceanic lithosphere and surrounding mantle provide supporting evidence for the stagnation of subducted slabs. Velocity and density signatures of stagnated slabs were calculated based on two end-member slab geometries and composition, based on kinematic thermal structure analysis. These are forward model predictions, not fits to seismic observations. Examination of these predictions in light of seismic analysis suggests several important conclusions. (1) The fast anomalies and velocity structures observed under eastern China can be explained by subducted materials stagnant in the MTZ. (2) The harzburgite layer, subjected to some degree of undulation (buckling), is primarily responsible for the fast anomalies, whose magnitudes are in good accordance with seismic tomography studies. (3) The absence of an apparent low-velocity basalt layer in seismic models suggests that the basalt layer is either very thin (therefore nonresolvable seismically) or has been delaminated during subduction. (4) Mineral physics experiments are now capable of providing phase relations and thermoelastic databases for quantitative modeling of the subducted lithosphere. With further improvement in our understanding of slab thermal structure and mechanical coupling with the surrounding mantle, important constraints on the thermal structure, mineralogy, composition, density, and velocities of slab materials in MTZ may be obtained.
 We thank Chao Wang for useful discussions, Prof. Jianhua Rao for technical developments during the experiments, Prof. Jishun Yu and Dr. Jinhua Zhang for XRD analysis, associate Prof. Tao Chen and Dr. Yungui Liu for micro-Raman analysis, associate Prof. Haijun Xu and Dr. Siyu Feng for FE-SEM analysis, and Dr. Feng Shi for FT-IR analysis. We are grateful to two anonymous reviewers for their critical and constructive comments. This work was supported by the National Natural Science Foundation of China (41174076) and Sinoprobe-0801, and partially by the US NSF (EAR-0968456).