Bridgmanite-Predominant Lower Mantle Evidenced from Velocity and Density Profiles across the 660-km Discontinuity

(Nature Communications style <150 words) Earth’s mantle composition is essential to our understanding of its physics and dynamics. Here we report single-crystal elasticity ( C ij ) of (Al,Fe)-bearing bridgmanite, Mg 0.88 Fe 0.1 Al 0.14 Si 0.90 O 3 with Fe 3+ /∑Fe=~0.65, up to ~82 GPa measured in diamond anvil cells. Together with heat capacity measurements on bridgmanite and ferropericlase, we develop a fully internally-consistent thermoelastic model to simultaneously evaluate lower-mantle mineralogy and geotherm via comparisons of P-wave, S-wave velocities, and density ( V P , V S , and ρ ) with one-dimensional seismic profiles. Our best-fit model demonstrates the lower mantle consists of ~89 vol% (Al,Fe)-bearing bridgmanite, ~4 vol% ferropericlase, and ~7 vol% CaSiO 3 perovskite. A chemically layered mantle with pyrolitic upper mantle and bridgmanite-predominant lower mantle would display ~3.2(±1.5)%, ~5.2(±1.5)%, and ~5.0(±1.0)% jumps in V P , V S , and ρ , respectively, across the 660-km discontinuity, which are well consistent with seismic reflection observations. The lower mantle could have become bridgmanite-predominant via accumulations of ancient silica-rich materials, which helps explain current deep-Earth seismic and geochemical signatures. mantle: (1) peridotite 50 with a harzburgite composition of Mg/Si=~1.5 (~80 vol% ringwoodite and ~20 vol% majorite); (2) pyrolite 8 with Mg/Si=~1.25 (~60 vol% ringwoodite and ~40 vol% majorite); (3) piclogite 9 with Mg/Si=~0.9 (~20 vol% ringwoodite and ~80 vol% majorite). Our calculations show that phase


Introduction
Strong seismic reflections at 660-km depth, marked by sharp velocity and density jumps 1-6 , have long been associated with a structural transition from ringwoodite to bridgmanite and ferropericlase 7 . Geochemical and petrological observations indicate that Earth's upper mantle likely consists of pyrolite 8 with approximately three portions of peridotite and one portion of basalt, although some studies also proposed a garnet-rich piclogite in the transition zone 9 . If one assumes the whole mantle is chemically homogenous in major elements, a pyrolitic lower mantle  10 . However, studies show that structural transition(s) alone in pyrolite cannot simply explain the aforementioned velocity and density jumps at 660-km depth 3,11 . A recent seismic study observed regional small-scale topography along the 660-km boundary, implying a discontinuous chemical layer across the region 12 . Moreover, such a pyrolitic model with Mg/Si=~1.25 has much less Si than the chondritic bulk Earth model 13 with Mg/Si=~1.0. To address the "missing Si" conundrum in the silicate Earth, Si as a light element in the core 14 and/or a Si-rich lower mantle 15 have been proposed previously.
Thus far, the first-order question on the mantle mineralogy remains as to whether or not the 660km discontinuity represents a structural transition as well as a chemical boundary. In other words, is the mantle chemically homogeneous or layered into upper and lower mantles?
Velocity and density comparisons between seismic observations 1,2 and mineral physics models 10, [16][17][18] have been widely used to provide insights into the mantle mineralogy. Of particular example is the study by Murakami et al. 17 , which used VS of polycrystalline Al-bearing bridgmanite up to 124 GPa and 2700 K to suggest a perovskitic lower mantle with greater than 93 vol% bridgmanite. However, it has been pointed out by Cottaar et al. 19 that the conclusion of a perovskitic lower mantle 17 was not supported because of using inconsistent thermoelastic modeling and inappropriate averaging schemes for mineral aggregates. A recent study 16 extended aggregate VP and VS of (Al,Fe)-bearing bridgmanite, Mg0.9Fe0.1Al0.1Si0.9O3 (Fe10-Al10-Bgm), up to 40 GPa at room temperature from single-crystal elasticity measurements and suggested a Fe 3+rich pyrolitic lower mantle above 1200-km depth. Nevertheless, further analyses 20 revealed different velocity profiles and large uncertainties of the reported elasticity 16 as a result of using insensitive crystal platelets with limited high-pressure VP data. To make matters worse, velocity and density profiles of candidate minerals in these studies 10,[16][17][18] were typically modeled along a previously reported geotherm [21][22][23][24] , a pressure-temperature (P-T) profile of the mantle. For instance, Katsura et al. 22 assume the whole mantle is dominated by heat convection and calculate an adiabatic geotherm for pyrolite with a temperature gradient: ( / ) = / , where T is the temperature at a certain depth/pressure, and α, ρ, and CP are thermal expansion coefficient, density, and isobaric heat capacity of constituent materials, respectively. Building a consistent mantle geotherm would thus require quantitative knowledge of its mineralogy and respective high P-T thermoelastic properties. In a nutshell, because of the lack of a complete dataset on extended VP, VS, ρ, and CP properties with small uncertainties and a fully internally-consistent thermoelastic model on geotherm, we could not unambiguously constrain the lower-mantle mineralogy and associated Mg/Si, and understand seismic and dynamic signatures across the 660-km boundary.

Results
Figure 1c-e shows representative BLS and ISLS spectra with high signal-to-noise ratios to derive VS1, VS2, and VP of single-crystal Fe10-Al14-Bgm at ~82 GPa. Furthermore, measured CP of Fe-bearing bridgmanite as a function of temperature are consistent with literature reports 30,31 of MgSiO3 bridgmanite within uncertainties ( Fig. 1f-g). This indicates that Fe substitution has a negligible effect on CP of bridgmanite. The modeled high P-T CP of bridgmanite as well as ferropericlase agree well with ab initio calculations 32,33 .
Collected velocities with rotations of the DAC about its compression axis over a range of 200° were fit using Christoffel's equations 34 to derive its full Cij at each experimental pressure . Results show that all Cij increase monotonically with pressure up to 82 GPa with uncertainties of 1-2%, except C55 and C23 with errors of ~3-4% (Supplementary Tables   3-4). In particular, the derived Cij of Fe10-Al14-Bgm at 25 and 35 GPa are comparable to those of literature single-crystal bridgmanite [35][36][37] . We note that the slight difference on pressure derivatives of Cij between this study and ab initio calculations for MgSiO3 bridgmanite [38][39][40]  Al preferentially in the B site and ~10% Fe mostly in the A site. Thus, the sample is not expected to undergo a spin transition and will display a monotonic increase in VS and VP with pressure.
Additionally, we note that the room-temperature VS and VP of bridgmanite are much higher than those of ferropericlase 27,43 and CaSiO3 perovskite 28 .

Discussion
Here, we have developed a fully internally-consistent thermoelastic model to simultaneously evaluate lower-mantle mineralogy and temperature profile. In this model, we assume that the lower mantle, to the first order, is chemically homogenous, adiabatic, and under gravitational selfcompression with Bullen's parameter close to one because of the consistency between 1D seismic profiles 1,2 and the Adams-Williamson equation 44 (see Methods for details).. Third-order Eulerian finite-strain equations 29 and Mie-Grüneisen EoS are adopted to model high P-T thermoelastic properties of bridgmanite, ferropericlase, and CaSiO3 perovskite, including VP, VS, ρ, α, and CP (Supplementary Tables 5-6, Supplementary Figs. 4-7). We have also considered effects of the spin crossover in ferropericlase 27,43 as well as Fe partitioning (KD) between bridgmanite and ferropericlase in the lower mantle 10 . To quantitatively evaluate Fe and Al substitution effects on the velocity and density of bridgmanite, experimental data from this study and the literature 16,17,[45][46][47] are fit together to derive the adiabatic bulk (shear) modulus, KS0 (μ0), and its pressure derivative, K′S0 (μ′0), as a function of Fe and Al contents. Moreover, considering the frequency dependence of seismic attenu ation 48 , corrections from experimental high-frequency data (~GHz) to seismic frequencies (~Hz) are also conducted for modeled velocity profiles (Methods, Supplementary Fig.   8). The amount of CaSiO3 perovskite is fixed as 7 vol% because its abundance in both pyrolitic and chondritic models 13 is about 5-10 vol%. With all these parameters taken into consideration, we initially use a pyrolitic lower-mantle mineralogy as a reference to evaluate VP, VS, and ρ profiles of the three major minerals along an adiabatic geotherm. Considering trade-offs between compositional and thermal parameters, an iterative procedure with uncertainty minimization is conducted until the best-fit model converges with PREM profiles 1 .
In our best-fit model, bridgmanite displays slightly higher velocities than those of PREM, while ferropericlase and CaSiO3 perovskite show lower velocities (Fig. 2c-d and Supplementary   Fig. 9). In particular, VS of these three minerals are well distinguishable from one another after taking uncertainties into account, that makes it the most sensitive elastic parameter to evaluate the lower-mantle mineralogy. VP softening across the spin crossover in ferropericlase is smeared out but still visible at mid-lower mantle P-T. Notably, the spin crossover in ferropericlase increases its Fe content in the low-spin state 10 , which in turn flattens VS toward the deeper lower mantle. We note that if KD is constant with depth, VS of ferropericlase increases monotonically in the lower mantle ( Supplementary Fig. 9g-i).
Our best fits to PREM 1 show a lower-mantle mineralogy of ~88.7(±2.0) vol% (Al,Fe)-bearing bridgmanite, ~4.3(±2.0) vol% ferropericlase, and 7 vol% (fixed) CaSiO3 perovskite (Fig. 3), thereafter called the bridgmanite-predominant lower-mantle model. In contrast, a pyrolitic model shows lower VP and VS than those of PREM, whereas ρ profiles between pyrolitic and bridgmanitepredominant models are indistinguishable. With an adiabatic lower-mantle convection 44 , the derived geotherm ranges from 1920 K at 660-km depth, an anchor potential temperature for the post-spinel phase transformation 7 , to 2560(±80) K at 2800-km depth. Our best-fit geotherm is comparable with that for a chondritic composition from ab initio calculations 24 . On the other hand, if we assume the lower mantle is superadiabatic 44,49 with Bullen's parameter of ~0.8, the superadiabatic geotherm will be ~300 K higher than our best-fit adiabat at ~2800-km depth. It should be noted that such a superadiabatic geotherm will consequently result in significant velocity mismatches with PREM in both bridgmanite-predominant and pyrolitic models. To further test the sensitivity of KD and the amount of CaSiO3 perovskite to our modeled results, we allow them to vary within a reasonable compositional range and find that variations in these two parameters have limited effects on the derived lower-mantle mineralogy (e.g., they would cause ~2 vol% variation in bridgmanite content) and <0.5% in geotherm (Supplementary Fig. 10 and Supplementary Table   7). We note that the anelastic relaxation correction from ~GHz to ~Hz slightly decreases modeled velocity profiles by ~0.3%; without such a correction, the modelled velocity profiles would be more similar to a pyrolitic composition (less bridgmanite) in the lower mantle.
It is important to analyze uncertainties of these modeled results before one can reach a firm conclusion on the mantle mineralogy and geotherm. Using standard error propagations, our modeled high P-T VP, VS, and ρ profiles typically have uncertainties of ±2.2%, ±1.8%, and ±1.1%, respectively, at ±1σ level (see Methods). A pyrolitic model 10 with +1σ upper bounds would marginally overlap with PREM profiles 1 and display an adiabatic geotherm ~100 K higher than our best-fit model at ~2800-km depth (Fig. 3). Alternatively, if the chosen anchor potential temperature at 660-km depth decreases by ~500 K to ~1420 K, a pyrolitic model could also match PREM well. We note that this temperature reduction might be too large to be realistic at the topmost lower mantle according to literature reports on post-spinel and/or post-garnet transformations 7 . Overall, with the most comprehensive thermoelastic study on major lowermantle minerals along an adiabatic geotherm to date, our results indicate bridgmanite is the most predominant mineral ranging from ~75 to ~100 vol% after taking ±1σ uncertainties into account. Mg/Si=~0.9 (~20 vol% ringwoodite and ~80 vol% majorite). Our calculations show that phase transformations in these models produce jumps that are not consistent with seismic reflection observations [3][4][5][6] . For instance, the peridotite model displays ~3.3(±0.5)% in ΔVP/VP, ~7.6(±0.5)% in ΔVS/VS, and ~9.0(±0.3)% in Δρ/ρ, while the piclogite model 9 would increase ΔVP/VP and ΔVS/VS to ~4.6(±1.0)% and ~8.6(±1.0)%, respectively, and decrease Δρ/ρ to ~5.8(±0.8)%. We note that early studies 51 suggest potential mantle differentiation could result in physical mixing of multiple lithologies. We have used peridotite and piclogite as end members to evaluate physical mixtures with different lithology percentages, but their velocity and density jumps are still not comparable with seismically-observed discontinuities at the 660-km depth [3][4][5][6] . In summary, these analyses provide supportive evidence for a chemically layered mantle with a bridgmanite-predominant lower mantle and pyrolitic upper mantle.
Our bridgmanite-predominant lower-mantle model is silica-rich with Mg/Si=~0.97. Assuming a pyrolitic upper mantle with Mg/Si=~1.25, the Mg/Si for bulk silicate Earth will be about 1.05.
The bulk Earth is generally believed to be chondritic 13 with Mg/Si=~1, despite some geochemical evidence challenging such a model 52 . The bridgmanite-predominant lower mantle is consistent with the chondritic bulk Earth model 15 by having more "missing Si" in the lower mantle.
Furthermore, the chemically layered mantle can have ramifications on dynamic processes of our planet. Seismic tomography shows that some subducting slabs can penetrate through the transition zone into the lower mantle to help homogenize mantle chemistry in a whole-mantle convection model. In contrast, stagnant slabs are found to be prevalent in the transition zone 53 , that might indicate a layered-mantle convection model. Bridgmanite is suggested to be ~1000 times rheologically stronger than ferropericlase 54 . The bulk viscosity for a bridgmanite-predominant lower mantle is estimated to be 1-2 orders higher than previously thought for a pyrolitic lower mantle 55 (Supplementary Fig. 13). A recent geodynamic study 55 proposed that ancient Si-rich materials in the lower mantle would behave sufficiently rigid to resist whole-mantle mixing, which helps maintain chemical layering over 4.6 Gyr of Earth's evolution. In the meanwhile, conduits and channels could exist between these rigid ancient blobs 55 to allow regional penetrations of subducting slabs into the lower mantle. Due to high viscosity of the bridgmanite-predominant lower mantle, some cold slabs could remain metastable and seismically visible with the slow mechanical mixing. Early studies indicate that such highly viscous convecting materials in the lower mantle might make it slightly superadiabatic 21 . Considering the mantle adiabaticity estimated from seismic profiles 1,44 and other potential contributions 49 , such as internal heating, future studies using combined high-quality mineral physics and high-resolution seismic data are needed to elucidate spatial and temporal signatures of the mantle geophysics and geochemistry.
conducted high-pressure BLS, ISLS, and XRD experiments. S.F. analyzed the data and performed the modelling. S.F. and J.F.L. wrote the manuscript draft, and all authors discussed and commented on the content of the manuscript.

Materials & Correspondence
Supplementary information is available. Correspondence and requests for materials should be addressed to J.F.L. (afu@jsg.utexas.edu) and S.F. (fsyxhy@gmail.com).

Competing interests
The authors declare no competing interests.
Selection of single-crystal platelets to derive its full C ij . Bridgmanite has an orthorhombic structure (Pbnm) in lower-mantle conditions with nine independent Cij to be constrained. National Laboratory (ANL) using an incident X-ray with a 0.3344-Å wavelength. ±15° rotations of each platelet about the vertical axis of the sample stage were employed to reliably determine its crystallographic orientation. Then, we calculated the sensitivity of synthetic velocities to Cij in each specific orientation so that we can select appropriate platelets that can be used to derive full Cij with small uncertainties. Based on these analyses (Fig. 1a- Re gaskets with initial thicknesses of 250 µm were pre-indented to 25 GPa or ~28-µm thick using a short symmetric 200-µm-culet DAC. Pre-indented areas were drilled with 130-µm diameter holes as sample chambers. Selected platelets were cut using focused ion beam into circular shapes, ~60-70 µm in diameter, and polished down to 10 µm thick to be loaded into sample chambers in two runs. Neon was used as pressure medium and a ruby sphere pressure calibrant was placed close to the sample to minimize pressure uncertainties (Fig. 1c insert). We note that, to achieve equilibrium and ensure pressure consistency between two runs, each DAC was kept at the target experimental pressure for 2-3 days before each measurement. Ruby fluorescence was taken before and after velocity measurements to evaluate pressure and pressure uncertainties 56 . We use the ruby pressure scale by Dewaele et al. 56  The pump laser is split into two excitation beams, which are then recombined at the sample position with a crossing angle of 20.3º. The focused probe laser has a beam size of 30-40 μm. To avoid potential geometrical errors, both BLS and ISLS systems were aligned precisely using a series of reference spots and iris diaphragms, and were calibrated monthly using distilled water and standard glass 26 . To avoid potential sample degradation due to the metastability of bridgmanite at low pressures, we intentionally started BLS and ISLS measurements from ~24 GPa. BLS and ISLS spectra were collected on two loaded samples in identical orientations as a function of azimuthal angles over 200° with a 10°-step rotation at each experimental pressure. Collected BLS spectra were used to derive VS, and time-domain ISLS spectra were Fourier-transformed into frequency-domain power spectra to derive VP of the sample at high pressure (Fig. 1c-e).
To confirm the crystal quality of loaded platelets with neon pressure medium, we performed synchrotron single-crystal XRD at 13 ID-D of GSECARS at APS, ANL at several high-pressure points. Supplementary Fig. 1 shows representative XRD results of the platelet with an orientation of (-0.50, 0.05, -0.86) at ~76 GPa. These circular and round diffraction peaks with average FWHM of ∼0.04º-0.07º confirmed the high-quality of our crystals in DACs. Analyses of XRD also show consistent orientation information as determined at ambient conditions with small deviations (<0.2º).
A complimentary XRD run was conducted on single-crystal Fe10-Al14-Bgm up to 75 GPa at 300 K to evaluate its pressure-volume (P-V) relationship at 13 ID-D of GSECARS at APS, ANL (Supplementary Table 1). Following a literature procedure 26 , a piece of Fe10-Al14-Bgm platelet, ~20 µm in length and ~8 µm in thickness, was loaded into a symmetric DAC with 200-μm culets, together with Au as the pressure calibrant and neon as the pressure medium.
where Cijkl is the elastic tensor with full suffix notation, contracted to Cij in Voigt form in this study, v are measured velocities of VP, VS1, and VS2, ni are wave vector direction cosines, and δik is the Kronecker delta. We further used a finite-strain theory from the literature 29 to fit highpressure Cij to quantitatively derive their pressure derivatives ( Supplementary Fig. 3 and
However, Helmholtz free energy at higher orders increases with pressure and cannot be simply truncated at high P-T conditions 29 . Thus, equations used here are more applicable to candidate lower-mantle minerals.
Collected high-pressure XRD patterns of single-crystal Fe10-Al14-Bgm were analyzed to determine its unit cell parameters and volume. We fit P-V at 300 K using third-order Birch-Murnaghan EoS 59 :  Table 5).
We note that 68% residues of KS, μ, and ρ in the best fits are less than 1.1%, 1.0%, and 0.3%, respectively (1σ). where n is the LS fraction at high P-T. We note that studies 68 indicate that Fe effect on the onset pressure and width of the spin crossover is negligible for relatively Fe-poor ferropericlase (<25 mol% Fe), but becomes complex for the Fe-rich (Mg,Fe)O counterpart 68 . Our modeling is thus limited to ferropericlase containing less than 25 mol% Fe, which is most relevant to the lowermantle composition.
The fully internally-consistent thermoelastic model. VP, VS, ρ, and adiabatic temperature profiles of lower-mantle aggregates can be modeled from KS, μ, ρ, α, and CP using a fully internally-consistent thermoelastic model 29 : ( , ) = 300K + ∆℧ q / where P300K, KT_300K, and μ300K are pressure, isothermal bulk moduli, and shear moduli at 300 K, respectively, that can be calculated using eqs. (17)- (19), q0 is a volume-independent constant, γ is the Grüneisen parameter, ηS0 is the shear strain derivative of γ, and Δ℧q and Δ(CVT) are internal energy and heat differences between 300 K and high temperature, respectively. γ and α can be where R is the gas constant, n is the number of atoms in the mineral formula, and θD is Debye temperature, expressed as: where θ0 is the ambient Debye temperature. CP can be derived from CV with P = (1 + ) V .
In this fully internally-consistent thermoelastic model, ten parameters are involved to describe thermoelastic properties of each individual mineral, including F0, V0, K0, K'0, 0, '0, θ0, γ0, q0, and ηS0. We neglect the parameter F0, as it is only related to phase equilibrium and will involve large uncertainties due to limited experimental constraints. V0 is ambient unit cell volume from XRD.
K'0 is the pressure derivative of isothermal bulk modulus, K0. Here, we set K0 and K'0 from velocity measurements, because they provide direct velocity information. We note that our measured velocity data are used to relate the adiabatic bulk modulus, KS, therefore, proper conversions on KS0 and K'S0 to K0 and K'0 have been made using eq. (34). K0 and K'0 here are internally consistent.
Similarly, '0 and 0 can be derived from velocity results. θ0, γ0, q0, and ηS0 are four parameters related to high temperature extrapolations (see details for constraining in the followed section).
Constraints on thermoelastic parameters of major mantle minerals. Using the fully internallyconsistent thermoelastic model 29 67 . As to CaSiO3 perovskite, we modeled its high P-T elastic data from Gréaux et al. 28 and Sun et al. 75 , but excluded those from Thomson et al. 76 ( Supplementary Fig. 6). The density profile by Gréaux et al. 28 is consistent with that by Sun et al. 75 , while Thomson et al. 76 reported much lower density than these studies 28,75 . Our modelled profiles for CaSiO3 perovskite fall into the range of theoretical calculations 69,70 . indicate that as a result of the frequency dependence of seismic attenuation, the attenuation factor, q, with frequency, ω, can be represented as: where ΔE/Emax is the fraction of internal energy lost in one cycle of seismic waves, q00 is the reference factor with frequency ω00, and τ is a model-dependent and frequency-dependent parameter. Velocity variations 48 between V1(ω1) and V2(ω2) with frequencies of ω1 and ω2, respectively, can be expressed as: Early studies 78 show that τ generally ranges from 0.2 to 0.4. With τ as ~0.3, velocity differences with frequencies between GHz and Hz are less than 0.3% (Supplementary Fig. 8). We note that this equation is typically applied to anelastic dispersion within seismic frequency, ranging from Hz to several thousands of Hz. This equation is used for correcting frequency-dependent velocity to the first order.
Viscosity of the lower mantle. The viscosity of bridgmanite (10 21 -10 23 Pa·s) was inferred to be much higher than that of ferrpericlase(10 18 -10 20 Pa·s) 54 where a and b are density and viscosity contrasts between bridgmanite (ρBgm, ηBgm) and ferropericlase (ρFp, ηFp), calculated as a=ρBgm/ρFp and b=ηBgm/ηFp, respectively. We neglect the role of CaSiO3 perovskite due to the lack of reliable viscosity from the literature.  VP and VS at high pressure and 300 K. Solid red symbols are aggregate sound velocities of singlecrystal Fe10-Al14-Bgm in this study, and solid red lines are the best fits using a finite-strain theory 29 16 ; MgSiO3 (dashed gray lines) 46 and Al-bearing bridgmanite containing 5.1 wt% Al2O3 (dashed orange lines) 47 . Velocity profiles of CaSiO3 perovskite 28 (solid blue lines from modeling) and ferropericlase 27,43 (solid and open olive circles; olive lines show finite-strain fitting) at 300 K are plotted for comparisons. Uncertainties are smaller than symbols when not shown. c and d, VP and VS along an adiabatic geotherm derived in our best-fit model (refer to Fig. 3 and Supplementary Fig. 9). Red lines: (Al,Fe)-bearing bridgmanite; olive lines: ferropericlase; blue lines: CaSiO3 perovskite. KD between bridgmanite and ferropericlase with depth is taken from the literature 10 . The flattening in VS of ferropericlase above ~45 GPa results from Fe preferentially partitioning into ferropericlase across the spin crossover. Seismic profiles 1,2 are plotted as dashed black and pink lines. Vertical ticks show one standard deviation (±1σ) derived from error propagations.  ΔVS/VS versus ΔVP/VP; c, ΔVS/VS versus Δρ/ρ. Two sets of compositional models are considered here: chemically layered mantle and chemically homogeneous whole mantle. Solid red circles are results for a chemically layered mantle with a pyrolitic transition zone 8 and bridgmanitepredominant lower mantle in this study. Solid pink, orange, and blue circles are calculations for a chemically homogeneous whole mantle with Mg/Si of ~1.5 (peridotite 50 ), ~1.25 (pyrolite 8 ), and ~0.9 (piclogite 9 ), respectively. Literature results 11 for a pyrolitic whole mantle are plotted as solid gray circles (WW98). Errors of these models represent one standard deviation (±1σ). Open symbols are from seismic observations: EK96 (open purple stars) 4 ; LS06 (open olive stars) 6 ; CC00 (open olive squares) 5 ; SF99 (open black stars) 3 ; PREM (open black circles) 1 ; AK135 (open olive circles) 2 . Shaded dark and light gray areas are confidence ellipsoids within ±1σ and ±2σ limits, respectively, from Shearer and Flanagan 3 .