d Forward Reverse GPa 1,890 K e Forward GPa 1,860 K Intensity (arbitrary units) c Reverse GPa 2,860 K 131.

Transcription

1 Vol December 2009 doi: /nature08598 LETTERS Thickness and Clapeyron slope of the post-perovskite boundary Krystle Catalli 1, Sang-Heon Shim 1 & Vitali Prakapenka 2 The thicknesses and Clapeyron slopes of mantle phase boundaries strongly influence the seismic detectability of the boundaries and convection in the mantle. The unusually large positive Clapeyron slope found for the boundary between perovskite (Pv) and postperovskite (ppv) 1 3 (the ppv boundary ) would destabilize hightemperature anomalies in the lowermost mantle 4, in disagreement with the seismic observations 5. Here we report the thickness of the ppv boundary in (Mg 0.91 Fe0.09)SiO 21 3 and (Mg 0.9 Fe0.1)(Al Si 0.9 )O 3 as determined in a laser-heated diamond-anvil cell under in situ high-pressure (up to 145 GPa), high-temperature (up to 3,000 K) conditions. The measured Clapeyron slope is consistent with the D99 discontinuity 6. In both systems, however, the ppv boundary thickness increases to km, which is substantially greater than the thickness of the D99 discontinuity (,30 km) 7. Although the Fe 21 buffering effect of ferropericlase 8 10 could decrease the ppv boundary thickness, the boundary may remain thick in a pyrolitic composition because of the effects of Al and the rapid temperature increase in the D99 layer. The ppv boundary would be particularly thick in regions with an elevated Al content and/or a low Mg/Si ratio, reducing the effects of the large positive Clapeyron slope on the buoyancy of thermal anomalies and stabilizing compositional heterogeneities in the lowermost mantle. If the ppv transition is the source of the D99 discontinuity, regions with sharp discontinuities may require distinct compositions, such as a higher Mg/Si ratio or a lower Al content. The lowest km of the mantle (the D99 layer) is known to have seismic properties distinct from those of the overlying mantle, including a laterally varying discontinuity at the top of the layer (the D99 discontinuity) 11,12. The discovery of a phase transition in the dominant mantle silicate (the ppv transition) 1,13,14 has provided new opportunities to understand seismological observations better 2,15 and to constrain important geophysical parameters 16,17 in the D99 layer. Any transition in a material with variable composition should have a finite-depth interval in which both low- and high-pressure (P) phases coexist (we call this depth interval the boundary thickness). For seismic detection, the boundary thickness should be sufficiently small. A thickness of less than 30 km was estimated for the D99 discontinuity 7. Together with its Clapeyron slope (dp/dt; T, temperature), the thickness of the ppv boundary is critical to understanding mantle convection in the D99 region 4. The thickness of the ppv boundary has not been well constrained 18 21, owing to experimental difficulties. We have determined the thickness and the Clapeyron slope of the ppv boundary in (Mg 0.91 Fe )SiO 3 and (Mg 0.9 Fe )(Al 0.1 Si 0.9 )O 3 in situ at high pressures and temperatures in a double-sided, laser-heated diamond-anvil cell (DAC) using synchrotron X-ray diffraction under improved experimental conditions. After being compressed directly without heating, an amorphized (Mg 0.91 Fe 0.09 )SiO 3 pyroxene sample was heated for a total of 1.5 h at 1,500 2,700 K and pressures higher than 130 GPa. Under these conditions, the synthesis of a Pv1pPv mixture was observed and the phase assemblage remained stable throughout the heating (Fig. 1a). In the next heating run, at a slightly higher pressure, the sample transformed completely to pure ppv (Fig. 1b). A separate sample was heated to 2,000 3,000 K at 137 GPa, under which conditions we observed the synthesis of pure ppv within 10 min. When the pure ppv sample was decompressed by 5 GPa and heated, Pv diffraction lines appeared in 1 h (Fig. 1c), marking the ppv R Pv1pPv transition. These observations indicate that the boundary between Pv1pPv and ppv should be at GPa and 2,500 3,000 K. Further decompression of this sample, to 119 GPa, and heating to 2,800 K caused strong growth of Pv diffraction lines. In the other samples compressed to GPa, stability of the Pv1pPv mixture was observed during 1.5 h of heating. When the Pv1pPv mixture was decompressed and heated to 2,000 K Intensity (arbitrary units) f 3.0 Reverse GPa 1,890 K e Forward GPa 1,860 K d Forward GPa 2,860 K c Reverse GPa 2,300 K b Forward GPa 2,460 K a Forward GPa 2,470 K d spacing (Å) N N N N +Re Re +Re +Re Re +Re Au+Ar Au Ar +Re +Re Re Re sin!/" (Å 1 ) Ar Au Au Ar Figure 1 X-ray diffraction patterns at high pressure and temperature. a, b, c,pv1ppv (a), pure ppv (b) and ppv1pv (c) in (Mg 0.91 Fe 0.09 )SiO 3 ; d, e, f,pv1ppv (d), pure ppv (e) and ppv1pv (f) in (Mg 0.9 Fe 0.1 )(Al 0.1 Si 0.9 )O 3. The major diffraction lines are indicated for Pv (blue), ppv (red), Ar (pressure medium), Au (pressure standard), Re (gasket) and N (nitrogen), along with the direction of the transition path (forward or reverse), the pressure and the temperature. Backgrounds were subtracted. The diffraction intensities are plotted as a function of 1/d 5 2sinh/l (d, interplanar distance; h, diffraction angle; l, X-ray wavelength). Au Au Ar Ar 1 Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. 2 GeoSoilEnviroCARS, University of Chicago, Chicago, Illinois 60637, USA Macmillan Publishers Limited. All rights reserved

2 NATURE Vol December 2009 LETTERS at 107 GPa, sudden broadening of the ppv lines was observed within 10 min, indicating the instability of ppv. An amorphous (Mg 0.9 Fe 0.1 )(Al 0.1 Si 0.9 )O 3 sample was compressed to 132 GPa, where a Pv1pPv mixture was synthesized during heating. After a total of 55 min of heating at pressures greater than 139 GPa, we observed the transition from a Pv1pPv mixture to pure ppv (Fig. 1e), indicating the stability of the ppv phase. The pure ppv sample was decompressed and heated to 1,800 2,100 K at 130 GPa. After a total of 13 min of heating, a few diagnostic diffraction lines of the Pv phase appeared, indicating the ppv R Pv1pPv transition (Fig. 1f). For separate samples, the stability of a Pv1pPv mixture was observed during compression up to 137 GPa at high temperature. Therefore, the boundary between Pv1pPv and ppv should exist between 130 and 140 GPa at 2,000 K. In samples compressed to lower pressures, we observed the stability of a Pv1pPv mixture along both forward and reverse directions between 113 and 137 GPa at 2,000 3,000 K (Fig. 1d). In one of the samples compressed to a lower pressure, synthesis of pure Pv was observed for compression up to 109 GPa at 2,000 K. The much denser data coverage in (Mg 0.91 Fe 0.09 )SiO 3 tightly constrains the Clapeyron slope of the boundary between Pv1pPv and ppv to MPa K 21, which is in agreement with the seismological estimation of the D99 discontinuity 6 and an earlier computational prediction 2. A larger Clapeyron slope was reported in Mg endmember 3, but the discrepancy is probably due to the inconsistency among the different pressure scales used. As differences in pressure can be more reliably determined, the measured thickness should be less affected by this issue. Our data show that both 9 mol% Fe 21 and 10 mol% Fe 31 plus 10 mol% Al substantially increase the thickness of the ppv boundary, to GPa ( km) and GPa ( km), respectively. These values are much greater than the upper bound for the thickness of the D99 discontinuity, 30 km (ref. 7; Fig. 2). Some sources of uncertainty should be considered. Kinetic effects normally delay a phase transition, resulting in an overestimation of transition pressure along the forward path (from a low-pressure phase to a high-pressure phase) and an underestimation along the reverse path (vice versa). Between the forward and reverse paths there is a 2 5-GPa mismatch in the boundary between Pv1pPv and ppv and we therefore considered the boundary to lie at the mean value. Because the kinetic effects are opposite along these paths, measurement of the ppv boundary along both transition paths reduces errors from these sources. Heating to sufficiently high temperature for sufficient duration is important in reducing kinetic effects. However, in most previous ppv studies the samples were heated to less than 2,000 K and only a few data exist at higher temperature 3. In our study, about one-half of the heating runs were made between 2,500 and 3,000 K. Although each heating run was limited to min to prevent overheating of the DAC, heating was repeated at a given pressure for a total heating time of 1 2 h, whereas heating in many previous studies was limited to 3 30 min. As there are steep temperature gradients in laser heating, the colder spots suffer more from kinetic effects. In this study, in order to improve homogeneity in the heated spot, we used Ar as an insulation medium and the sample size was similar to the size of the heating spot. Our large boundary thicknesses in (Mg 0.91 Fe 0.09 )SiO 3 and (Mg 0.9 Fe 0.1 )(Al 0.1 Si 0.9 )O 3 are consistent with earlier estimations for similar compositions except for those in an in situ study 18 in which a negligible ppv boundary thickness in (Mg,Fe)SiO 3 was proposed. However, in these studies measurements were not made along both transitional paths and the data coverage was sparse. The in situ study 18 was based on diffraction intensity changes in the patterns of Pv1pPv mixtures, which can also be affected by recrystallization and preferred orientation during heating. In determining the seismic detectability of the ppv transition, it is important to consider the element partitioning among different phases, the shape of the ppv phase-fraction profile in the mixed-phase Temperature (K) 3,000 2,500 2,000 1,500 3,000 2,500 2,000 1, Perovskite Perovskite 90 Elevation from the core mantle boundary (km) Macmillan Publishers Limited. All rights reserved a b Pressure (GPa) +6.7 ± 0.5 MPa K 1 Post-perovskite Post-perovskite Figure 2 The pressure temperature conditions for the stability of Pv, Pv1pPv and ppv. a, Data for (Mg 0.91 Fe 0.09 )SiO 3 ; b, data for (Mg 0.9 Fe 0.1 )(Al 0.1 Si 0.9 )O 3. The right- and left-pointing triangles represent data points measured along the forward and reverse paths, respectively: Pv, blue; Pv1pPv, green; ppv, red. The solid black lines are the phase boundaries determined from our data, with Clapeyron slope shown in a. The black data points in a are from ref. 18: Pv or Pv1pPv (Pv growth), open squares; Pv1pPv (no growth), crossed squares; Pv1pPv (ppv growth), filled squares. Dashed line, ppv boundary. Error bars, 1s. region and the rapid temperature increase in the D99 layer. To investigate these factors, we calculated the ppv phase-fraction profiles using the ideal-solution model 22 in combination with our measured thicknesses and the results of recent element partitioning studies 8 10,23 (Fig. 3). Because some discrepancy exists among the measured partition coefficients, we considered all the available data and chose upper bounds for the nonlinear deviations in the ppv phase-fraction profile and the Fe 21 buffering effect of ferropericlase (Fp) for presentation in Fig. 3 (Supplementary Information). Existing partitioning studies are in agreement that Fe 21 content follows Fp. ppv. Pv when these phases coexist 8 10,19,23, and that Fe decreases the ppv transition pressure 10,19 (Supplementary Information). As shown in our calculation along the isotherm at 2,500 K, these factors make the ppv phase-fraction profile nonlinear (Fig. 3d) in such a way that the rate of increase of the ppv phase fraction is greatest near the bottom of the mixed-phase region. For the measured Fe partition coefficients 8,9,23, our calculation indicates that Fe 21 buffering by Fp can reduce the ppv boundary thickness to km in (Mg,Fe)SiO 3 with 30% Fp. To match the thickness of the D99 discontinuity (,30 km) 7, the Fp content needs to be greater than the 30% expected for a pyrolitic composition (Fig. 3d). The buffering effect of Fp also increases the Pv R Pv1pPv transition depth, but has very little effect on the Pv1pPv R ppv transition depth (Fig. 3d). Because the Fp content increases as the Mg/Si ratio increases, this result suggests that the depth of the ppv transition in (Mg,Fe)SiO 3 increases as the Mg/Si ratio increases. As shown in Fig. 3a, the deepening may result in no intersection between the ppv boundary and the steep geotherm in the D99 layer, and the ppv

3 LETTERS NATURE Vol December 2009 Elevation from the core mantle boundary (km) a d 4, % Fp 30% Fp % Fp 3, % Fp 2,000 0% Fp 30% Fp D discontinuity Temperature (K) 4,000 3,000 2,000 b ppv fraction e Isotherm Average Warm Cold D discontinuity 4,000 3,000 2,000 c f D discontinuity Isotherm Average Warm Cold 105 Figure 3 The Pv1pPv mixed-phase region with mantle geotherms and the ppv phase-fraction profiles in the mixed-phase region. a, d, The Fe 21 buffering effect of Fp on the ppv boundary thickness in (Mg 0.91 Fe 0.09 )SiO 3. Three cases with different Fp contents (0%, 30% and 50%) are shown, as indicated. In a, the thick and thin lines are the isotherm at 2,500 K and the mantle geotherm, respectively. d shows calculations along the isotherm at 2,500 K. b, e, (Mg 0.91 Fe 0.09 )SiO 3 with 30% Fp. The boundary is shifted by 28 GPa to generate the maximum overlap between the ppv boundary and transition therefore may not exist in regions with an Fp content of 30% or greater. Because the absolute pressure scale is uncertain (65 GPa) over 100 GPa, we do not rule out the possibility of an overlap between the thinned ppv boundary in (Mg,Fe)SiO 3 and the geotherm. In Fig. 3b, we intentionally shifted the ppv boundary by 28 GPa to produce the maximum overlap. Because the rate of temperature increase in the D99 layer is high and may be comparable to the Clapeyron slope of the ppv boundary, the ppv phase fraction along the mantle geotherm will not increase as rapidly as that along an isotherm (Fig. 3e). In other words, the steep mantle geotherm in the D99 region will suppress the nonlinearity of the ppv phasefraction profile and prevent completion of the ppv transition. Studies 24,25 have found that a significant fraction of iron in Pv and ppv is Fe 31 (Fe 31 /SFe < 60%), through charge-coupled substitution with Al. In the case of Al and Fe 31, there is no effect from partitioning with other phases in pyrolite, such as Fp, because Al solubility in Fp is very low and CaSiO 3 perovskite remains nearly pure 26. Because Fe 31 alone does not increase the thickness (Supplementary Information and Supplementary Fig. 2), we attribute the large thickness found in (Mg 0.9 Fe )(Al 0.1 Si 0.9 )O 3 to the effect of Al, which is consistent with previous studies 20,21. The result shows that the ppv phase fraction may increase rapidly at the top of the mixed-phase region (Fig. 3f). However, the steep temperature increase in the D99 layer reduces the rate of increase of the ppv phase fraction and prevents the ppv phase transition from completing in the lower mantle (Fig. 3f) Pressure (GPa) 2009 Macmillan Publishers Limited. All rights reserved the mantle geotherms (see text). c, f, (Mg 0.9 Fe 0.1 )(Al 0.1 Si 0.9 )O 3. The black lines in e and f are the ppv phase-fraction profiles along the isotherm at 2,500 K. The grey lines in d, e, and f represent a hypothetical linear phasefraction profile with the thickness of the D99 discontinuity 7. The mantle geotherms were calculated using the error function combined with temperature estimation for the region 17. Three different geotherms are shown, representing different temperatures: red, warm; green, average; blue, cold. Mg silicates in a pyrolitic composition are expected to contain 5 15 mol% of Fe and 7 12 mol% of Al (refs 19, 24, 25, 27), which are similar to the compositions of the samples studied here. Although the ppv boundary thickness in (Mg,Fe)SiO 3 may decrease significantly as a result of Fe 21 buffering by Fp, the thickness of the ppv boundary in a pyrolitic composition would remain much greater than the thickness of the D99 discontinuity 7 because of the large increase in the boundary thickness caused by Al and the large radial temperature gradient in the D99 layer. In other words, the D99 layer may consist of mixed phases of Pv1pPv1Fp instead of ppv1fp, if the bulk composition of the lowermost mantle is pyrolitic. The large positive Clapeyron slope of the ppv boundary would increase the density contrast between thermal heterogeneities and bulk mantle, intensifying thermal mantle flow as shown by computer simulations 4, which is not compatible with seismic observations of the large low-shear-velocity provinces in the lowermost mantle and the stable large-scale mantle plume model 5. However, the large thickness of the ppv boundary would spread the effect of the Clapeyron slope over a wide depth range, reducing the effects of the ppv boundary on the density contrast and therefore stabilizing high-temperature anomalies in the D99 region. This would be particularly strong in regions with a high Al content and/or a low Mg/Si ratio, influencing the dynamic stability of some chemical heterogeneities in the D99 layer. A sharp D99 discontinuity has been observed in some regions of the D99 layer 16,17. Therefore, the discrepancy in thickness between the ppv boundary in pyrolite-related compositions and the D99 discontinuity

4 NATURE Vol December 2009 LETTERS raises important questions about the origin of the discontinuity. According to our results, the detectability of the ppv boundary could be enhanced in regions with a high Mg/Si ratio and a low Al content, requiring compositional changes for the observation of a sharp ppv boundary. In addition, it has been proposed that strong texturing of ppv might enhance the detectability of the ppv boundary 27. It is notable that multiples of laterally extending reflectors have been identified together with a pair of discontinuities inferred to be the double crossing of the ppv boundary within the bottom 400 km of the lower mantle 16,17,28.Inaddition,stronglytiltedsharpboundaries have been documented in seismic studies of the large low-shear-velocity provinces in the lowermost mantle 29.Theseobservationsaredifficultto explain in terms of a single isochemical phase transition. METHODS SUMMARY We mixed pyroxene and glass starting materials with 10 wt% gold, which serves as an internal pressure standard. The platelets of these mixtures were loaded in DACs with argon, which serves as a pressure-transmitting and insulating medium. We compressed the samples using bevelled diamond anvils with culets either 75 mm or 100 mm in diameter. In situ high-pressure, high-temperature measurements were conducted in the laser-heated DAC at the GeoSoilEnviroCARS sector of the Advanced Photon Source (Argonne National Laboratory) using double-sided laser heating and an angle-dispersive-diffraction set-up. The data sets for (Mg 0.91 Fe 0.09 )SiO 3 and (Mg 0.9 Fe 0.1 )(Al 0.1 Si 0.9 )O 3 were measured using the same experimental methods, including the pressure scale and pressure medium. This enabled us to measure the thickness and transition pressure among different compositions in an internally consistent fashion. Full Methods and any associated references are available in the online version of the paper at. Received 12 August; accepted 19 August Oganov, A. R. & Ono, S. Theoretical and experimental evidence for a postperovskite phase of MgSiO 3 in Earth s D99 layer. Nature 430, (2004). 2. Tsuchiya, T., Tsuchiya, J., Umemoto, K. & Wentzcovitch, R. M. Phase transition in MgSiO 3 perovskite in the Earth s lower mantle. Earth Planet. Sci. Lett. 224, (2004). 3. Tateno, S., Hirose, K., Sata, N. & Ohishi, Y. Determination of post-perovskite phase transition boundary up to 4400 K and implications for thermal structure in D99 layer. Earth Planet. Sci. Lett. 277, (2009). 4. Nakagawa, T. & Tackley, P. J. Effects of a perovskite-post perovskite phase change near core-mantle boundary in compressible mantle convection. Geophys. Res. Lett. 31, L16611 (2004). 5. Garnero, E. J., Lay, T. & McNamara, A. in Plates, Plumes, and Planetary Processes (eds Foulger, G. R. & Jurdy, D. M.) (Geological Society of America, 2007). 6. Sidorin, I., Gurnis, M. & Helmberger, D. V. Evidence for a ubiquitous seismic discontinuity at the base of the mantle. Science 286, (1999). 7. Lay, T. Sharpness of the D discontinuity beneath the Cocos Plate: implications for the perovskite to post-perovskite phase transition. Geophys. Res. Lett. 35, L03304 (2008). 8. Kobayashi, Y. et al. Fe-Mg partitioning between (Mg,Fe)SiO 3 post-perovskite, perovskite, and magnesiowüstite in the Earth s lower mantle. Geophys. Res. Lett. 32, L19301 (2005). 9. Auzende, A.-L. et al. Element partitioning between magnesium silicate perovskite and ferropericlase: new insights into bulk lower-mantle geochemistry. Earth Planet. Sci. Lett. 269, (2008). 10. Ono, S. & Oganov, A. R. In situ observations of phase transition between perovskite and CaIrO 3 -type phase in MgSiO 3 and pyrolitic mantle composition. Earth Planet. Sci. Lett. 236, (2005). 11. Lay, T., Williams, Q. & Garnero, E. J. The core mantle boundary layer and deep Earth dynamics. Nature 392, (1998). 12. Garnero, E. J. Heterogeneity of the lowermost mantle. Annu. Rev. Earth Planet. Sci. 28, (2000). 13. Murakami, M., Hirose, K., Kawamura, K., Sata, N. & Ohishi, Y. Post-perovskite phase transition in MgSiO 3. Science 304, (2004). 14. Shim, S.-H., Duffy, T. S., Jeanloz, R. & Shen, G. Stability and crystal structure of MgSiO 3 perovskite to the core-mantle boundary. Geophys. Res. Lett. 31, L10603 (2004). 15. Wookey, J., Stackhouse, S., Kendall, J.-M., Brodholt, J. & Price, G. D. Efficacy of the post-perovskite phase as an explanation for lowermost-mantle seismic properties. Nature 438, (2005). 16. Lay, T., Hernlund, J., Garnero, E. J. & Thorne, M. S. A post-perovskite lens and D heat flux beneath the central Pacific. Science 314, (2006). 17. van der Hilst, R. D. et al. Seismostratigraphy and thermal structure of Earth s coremantle boundary region. Science 315, (2007). 18. Hirose, K., Sinmyo, R., Sata, N. & Ohishi, Y. Determination of post-perovskite phase transition boundary in MgSiO 3 using Au and MgO pressure standards. Geophys. Res. Lett. 33, L01310 (2006). 19. Mao, W. L. et al. Ferromagnesian postperovskite silicates in the D layer of the Earth. Proc. Natl. Acad. Sci. USA 101, (2004). 20. Tateno, S., Hirose, K., Sata, N. & Ohishi, Y. Phase relations in Mg 3 Al 2 Si 3 O 12 to 180 GPa: effect of Al on post-perovskite phase transition. Geophys. Res. Lett. 32, L15306 (2005). 21. Nishio-Hamane, D., Fujino, K., Seto, Y. & Nagai, T. Effect of the incorporation of FeAlO 3 into MgSiO 3 perovskite on the post-perovskite transition. Geophys. Res. Lett. 34, L12307 (2007). 22. Stixrude, L. Structure and sharpness of phase transitions and mantle discontinuities. J. Geophys. Res. 102, (1997). 23. Sinmyo, R. et al. Partitioning of iron between perovskite/postperovskite and ferropericlase in the lower mantle. J. Geophys. Res. 113, B11204 (2008). 24. McCammon, C. Perovskite as a possible sink for ferric iron in the lower mantle. Nature 387, (1997). 25. Sinmyo, R., Hirose, K., O Neill, H. S. C. & Okunishi, E. Ferric iron in Al-bearing postperovskite. Geophys. Res. Lett. 33, L12S13 (2006). 26. Hirose, K., Fei, Y., Ma, Y. Z. & Mao, H.-K. The fate of subducted basaltic crust in the Earth s lower mantle. Nature 397, (1999). 27. Murakami, M., Hirose, K., Sata, N. & Ohishi, Y. Post-perovskite phase transition and mineral chemistry in the pyrolitic lowermost mantle. Geophys. Res. Lett. 32, L03304 (2005). 28. Hutko, A. R., Lay, T., Revenaugh, J. & Garnero, E. J. Anticorrelated seismic velocity anomalies from post-perovskite in the lowermost mantle. Science 320, (2008). 29. Ni, S., Tan, E., Gurnis, M. & Helmberger, D. Sharp sides to the African superplume. Science 296, (2002). Supplementary Information is linked to the online version of the paper at. Acknowledgements This work is supported by the US National Science Foundation (NSF) grant EAR (S.-H.S.) and a US Department of Energy (DOE) National Nuclear Security Administration Stewardship Science Graduate Fellowship (K.C.). A. Kubo and B. Grocholski assisted in X-ray measurements. Discussion with T. L. Grove and R. D. van der Hilst improved the paper. This work was performed in the GeoSoilEnviroCARS sector of the Advanced Light Source (APS), Argonne National Laboratory. GeoSoilEnviroCARS is supported by the NSF and the DOE. Use of the APS is supported by the DOE. Author Contributions K.C. and S.-H.S. prepared and made the measurements on (Mg 0.9 Fe 0.1 )(Al 0.1 Si 0.9 )O 3 and (Mg 0.91 Fe 0.09 )SiO 3, respectively. V.P. assisted in the synchrotron measurements. K.C. and S.-H.S. conducted the data analysis and calculations. S.-H.S. and K.C. wrote the paper. All authors discussed the results and commented on the manuscript. Author Information Reprints and permissions information is available at Correspondence and requests for materials should be addressed to S.-H.S. (sangshim@mit.edu) Macmillan Publishers Limited. All rights reserved 785

5 doi: /nature08598 METHODS Starting materials. The starting materials were a natural pyroxene with a composition of (Mg 0.91 Fe 0.09 )SiO 3 and a synthetic glass with a composition of (Mg 0.9 Fe 0.1 )(Al 0.1 Si 0.9 )O 3. The pyroxene sample was also used in our recent X-ray diffraction study of the equation of state and crystal structure of ppv 30. The glass starting material was synthesized from a molar mixture of 0.9MgSiO Fe 2 O Al 2 O 3 by the containerless method under an O 2 atmosphere to ensure that all Fe remained Fe 31 (ref. 31). The purity of the glass starting material was examined by synchrotron X-ray diffraction and Raman spectroscopy. The valence state of Fe in Pv synthesized from the (Mg 0.9 Fe 0.1 )(Al 0.1 Si 0.9 )O 3 glass was confirmed to be 31 by synchrotron Mössbauer spectroscopy. Sample loading. The starting material was powdered and mixed with 8 10 wt% gold for the internal pressure scale. A thin platelet of the sample-plus-gold mixture (thickness,,5 mm) was loaded in either a 35-mm or 50-mm hole in a pre-indented Re gasket (Supplementary Fig. 1). Ar was cryogenically loaded as a pressure and insulation medium. The platelet was supported by grains with the same composition as the starting material, to prevent direct contact with the thermally conductive diamond anvils. Measurements. At GeoSoilEnviroCARS, a monochromatic X-ray beam (energy, 37 or 40 kev) was focused onto an area of mm 2 on the sample and coaxially aligned with two Nd:YLF laser beams focused on both sides of the sample in the DAC. The size of the laser beam focus (20 mm) is comparable to the size of the sample (20 30 mm). Diffraction images were measured using the MarCCD detector. The tilt of the charge-coupled-device detector and the sample-todetector distance were calibrated by measuring the diffraction images of CeO 2. The diffraction images were integrated into one-dimensional patterns using the Fit2D software 32. The unit-cell volume of gold, constrained by 2 5 diffraction lines, was used, in combination with its equation of state 33, to calculate the pressure. The temperature of the sample was estimated by fitting the thermal radiation from the sample to Planck s equation Shim, S.-H. et al. Crystal structure and thermoelastic properties of (Mg 0.91 Fe 0.09 )SiO 3 postperovskite up to 135 GPa and 2700 K. Proc. Natl. Acad. Sci. USA 105, (2008). 31. Tangeman, J. A. et al. Vitreous forsterite (Mg 2 SiO 4 ): synthesis, structure, and thermochemistry. Geophys. Res. Lett. 28, (2001). 32. Hammersley, A. P. Fit2d: An Introduction and Overview. ESRF Internal Report (European Synchrotron Radiation Facility, 1997). 33. Tsuchiya, T. First-principles prediction of the P V T equation of state of gold and the 660-km discontinuity in Earth s mantle. J. Geophys. Res. 108, 2462 (2003). 34. Jeanloz, R. & Heinz, D. L. Experiments at high temperature and pressure: laser heating through the diamond cell. J. Phys. (Paris) 45, C8 83 C8-92 (1984) Macmillan Publishers Limited. All rights reserved

6 NATURE Vol December 2009 NEWS & VIEWS USA. Stephen F. Traynelis is in the Department of Pharmacology, Emory University School of Medicine, Rollins Research Center, Atlanta, Georgia , USA. s: 1. Sobolevsky, A. I., Rosconi, M. P. & Gouaux, E. Nature 462, (2009). 2. Jin, R. et al. EMBO J. 28, (2009). 3. Karakas, E., Simorowski, N. & Furukawa, H. EMBO J. doi: /emboj (2009). 4. Kumar, J., Schuck, P., Jin, R. & Mayer, M. L. Nature Struct. EARTH SCIENCE The enigma of D Kanani K. M. Lee Mol. Biol. 16, (2009). 5. Armstrong, N., Sun, Y., Chen, G.-Q. & Gouaux, E. Nature 395, (1998). 6. Wo, Z. G. & Oswald, R. E. Trends Neurosci. 18, (1995). 7. Wood, M. W., VanDongen, H. M. & VanDongen, A. M. Proc. Natl Acad. Sci. USA 92, (1995). 8. Chen, G.-Q., Cui, C., Mayer, M. L. & Gouaux, E. Nature 402, (1999). 9. Qian, A. & Johnson, J. W. Physiol. Behav. 77, (2002). 10. Chang, H.-R. & Kuo, C.-C. J. Neurosci. 28, (2008). 11. Banke, T. G. & Traynelis, S. F. Nature Neurosci. 6, (2003). A phase transition of Earth s most abundant mineral occurs at pressures and temperatures corresponding to those thought to exist just above Earth s core. New experiments shed light on this enigmatic Dʹʹ region. At the half-way point in a journey to Earth s centre, at a depth of about 2,900 kilometres, an intrepid explorer would encounter the core mantle boundary. This is where Earth s rocky mantle meets the fluid, iron-rich outer core, and it is marked by a large change in density and chemistry. The boundary is also thermal, because Earth s core is more than 1,000 K hotter than the mantle 1. Thus, conduction is the characteristic form of heat transfer for some 100 kilometres or so immediately above the core, whereas above that, in the bulk of the mantle, convection dominates. The sharp change in temperature gradient creates a distinct seismological and mineralogical environment for this region, which is dubbed Dʹʹ (D double prime). There is, of course, no explorer to send back reports about the region, so the study of Dʹʹ depends on the analysis of seismic waves or on experiments on minerals at relevant temperatures and pressures. The latter is the approach taken by Catalli et al. (page 782 of this issue) 2, who have come up with a new estimate for the thickness of the Dʹʹ discontinuity and the nature of this region. The special mineralogy of Dʹʹ is produced by a transition in crystal structure from that of magnesium silicate perovskite, MgSiO 3, the most abundant mineral on Earth, to a form known as post-perovskite 3,4. Seismicwave speeds change abruptly in this region, often more than once 5, providing information for seismological investigation of Dʹʹ. The approach taken by Catalli et al. 2 instead involves experimental determination of a feature called the Clapeyron slope, which marks the coexistence curve between two mineral phases, and is defined by the change in temperature over the change in pressure. For the perovskite-to-post-perovskite transition, this slope is positive that is, as one goes deeper into the mantle to higher pressures, higher temperatures are required to produce postperovskite. In addition, for a material that varies in composition (for example, with iron and aluminium replacing some of the magnesium), there is a finite range of pressures and temperatures for which both the high-pressure and low-pressure phases coexist; thus a thickness of this phase transition will yield a boundary that is either sharp or broad. Catalli and co-authors have made precise measurements, at simultaneous high pressures and temperatures, using a laser-heated diamond-anvil cell, on compositions of magnesium silicate perovskites that include both iron and aluminium. Attaining pressures above 1 Mbar (a million times room pressure) is fairly routine. But heating a sample at these pressures to the temperatures of the phase transition, above about 2,000 K, requires special attention. This is because diamonds are great thermal conductors and can dissipate the heat produced by the infrared lasers used to create the high temperatures, causing large temperature gradients in the sample chamber. Great care is necessary to thermally insulate the samples, which have dimensions of only tens of micrometres. Besides the technical difficulties of reaching high pressures and high temperatures in these kinds of experiment, there is the issue of calibration. Previous results have shown a range of positive Clapeyron slopes for the transition that can be attributed, in part, to the different pressure-calibration standards used 6. To overcome this problem, Catalli et al. 2 used the differences in the phase-boundary pressures rather than the absolute pressures: the absolute values are not important when determining the thickness of the perovskite to post-perovskite boundary, making the results calibrationindependent. The upshot is that the authors conclude that the two phases, perovskite and post-perovskite, could coexist over a depth range of (±100) km a much larger range than the seismically estimated Dʹʹ discontinuity thickness of about 30 km (ref. 7). How can this discrepancy between a thin Dʹʹ discontinuity (that is, a seismically sharp boundary) and the results of Catalli et al. be explained? The authors contend that compositional differences, perhaps low aluminium content or high abundance of ferropericlase (Mg, Fe)O Earth s second most abundant mineral would sharpen the boundary. But this would run counter to the view that portions of the Dʹʹ layer contain some oceanic crust, subducted from the surface through plate tectonics. Although oceanic crust mineralogy would have a large proportion of magnesium silicate perovskite, it also has increased amounts of silica (SiO 2 ) and alumina (Al 2 O 3 ), which, given Catalli and colleagues results, suggests an even thicker Dʹʹ discontinuity. All in all, the Dʹʹ layer presents many puzzles. One, for instance, is that seismic data indicate that there must be directionality to the minerals present, possibly due to the planar crystal structure of post-perovskite. But there is no consensus as to how post-perovskite would provide a consistent mechanism for such directionality 8,9. Another puzzle concerns a region at the very bottom of Dʹʹ that is dubbed the ultra-low-velocity zone, where large decreases in the speeds of seismic waves are recorded. One explanation for this phenomenon could be the presence of a small amount of minerals in melted form; another explanation is the possible existence of iron-rich post-perovskite produced by reaction with the iron-rich core 10. Indeed, the Dʹʹ region is evidently chemically diverse whether owing to interactions with the core, old oceanic crust or other factors and this is likely to be the root cause of much of its unusual seismic behaviour. There is much left to pin down in understanding the causes of the seismic signatures in Dʹʹ. Catalli and colleagues results highlight the need for more experiments and computations on the behaviour of not just Earth-relevant compositions of magnesium silicate perovskite, but also of other mineral assemblages, to tease out the effects of chemical composition on the transition to post-perovskite. Quantifying the effects of melt in Dʹʹ on density and on seismic wave speeds will also be important to understanding the other puzzles presented by Dʹʹ. In addition to new lab results and sophisticated computations, fine tuning on the Dʹʹ features that can be resolved by seismological observations are also needed. Kanani K. M. Lee is in the Department of Geology and Geophysics, Yale University, New Haven, Connecticut 06511, USA. kanani.lee@yale.edu 1. Williams, Q., Jeanloz, R., Bass, J., Svendsen, B. & 731

7 NEWS & VIEWS NATURE Vol December 2009 Ahrens, T. J. Science 236, (1987). 2. Catalli, K., Shim, S.-H. & Prakapenka, V. Nature 462, (2009). 3. Murakami, M., Hirose, K., Kawamura, K., Sata, N. & Ohishi, Y. Science 304, (2004). 4. Tsuchiya, T., Tsuchiya, J., Umemoto, K. & Wentzcovitch, R. M. Earth Planet. Sci. Lett. 224, (2004). 5. Hernlund, J. W., Thomas, C. & Tackley, P. J. Nature 434, (2005). 6. Fei, Y. et al. Proc. Natl Acad. Sci. USA 104, (2007). 7. Lay, T. Geophys. Res. Lett. doi: /2007gl (2008). 8. Merkel, S. et al. Science 316, (2007). 9. Wookey, J. & Kendall, J. M. in Post-Perovskite: The Last Mantle Phase Transition (eds Hirose, K., Brodholt, J., Lay, T. & Yuen, D.) (Am. Geophys. Union, 2007). 10. Mao, W. L. et al. Science 312, (2006). IMMUNOLOGY Dendritic-cell genealogy Sophie Laffont and Fiona Powrie The differing origins of gut dendritic cells white blood cells that modulate immune responses may explain how the intestinal immune system manages to destroy harmful pathogens while tolerating beneficial bacteria. The immune system must protect the body from invading pathogens without mounting damaging responses to its own tissues. Dendritic cells, a rare population of white blood cells, have a crucial role in determining the nature of immune reactions and in fine-tuning the balance between tolerance (where the immune system ignores or tolerates an antigen) and the induction of inflammation to destroy pathogenic organisms. A long-standing question has been how dendritic cells drive these distinct immune outcomes. Two groups, Varol et al.1 and Bogunovic et al.2, report in Immunity that dendritic cells with distinct Bacteria Intestinal lumen Mucosa Bacteria Epithelial cell CD103 CX3CR1+ DC CD103+ CX3CR1 DC Tolerogenic responses Pro-inflammatory mediators Protective immune responses InflammatoryIncreased effector T-cell cell recruitment responses Inflammatory t signals? Lymph nodes Flt3 M-CSF Blood Pre-DC Bone marrow Pre-DC Monocyte MDP Monocyte Figure 1 Intestinal dendritic cells have different origins and different functions1,2. The two main subsets of intestinal dendritic cells (DCs) originate from distinct blood-cell precursors and depend on different growth factors for their development. Before this divergence, a common precursor, the macrophage and dendritic-cell precursor (MDP), gives rise to pre-dendritic cells (pre-dcs) and monocytes in the bone marrow. Pre-dendritic cells give rise to CD103+ CX3CR1 dendritic cells, and depend on the growth factor Flt3 for their development, whereas monocytes develop into CD103 CX3CR1+ dendritic cells, and depend on another growth factor, M-CSF. CD103+ dendritic cells transport microbial antigens to the lymph nodes, where they may initiate protective immune responses or promote the generation of regulatory T cells that help to maintain tolerance in the intestine. CX3CR1+ dendritic cells do not seem to migrate to lymph nodes, suggesting a more local role in promoting tissue inflammation by stimulating effector T-cell responses and producing pro-inflammatory mediators. 732 )''0 DXZd`ccXe GlYc`j_\ij C`d`k\[% 8cc i`^_kj i\j\im\[ functions have different developmental origins, providing a cellular framework for the diverse activities of these cells. Pioneering work by Steinman and colleagues3 in the early 1970s identified a minor population of immune cells that they named dendritic cells on the basis of their stellate shape and membranous processes. These cells were shown to be potent stimulators of another population of white blood cells, T cells. Dendritic cells are strategically placed within mucosal sites in the body, where they can detect infection and take up microbial antigens. On activation, these cells migrate to secondary lymphoid tissue, such as the lymph nodes, where they present the antigen to T cells. This activates the T cells, causing them to differentiate into effector cells that eradicate the pathogen. In the intestine, dendritic cells also promote regulatory T-cell responses that suppress immune reactions against beneficial commensal bacteria and food antigens, thereby preventing immune-related disease. Thus, intestinal dendritic cells are decision makers, ensuring selection of a T-cell response that is appropriate to the nature of the challenge to the immune system. It is now known that dendritic cells are a diverse population of cells, differing in their anatomical location, expression of surface proteins and function. The diversity of the dendritic-cell response may reflect the differential activities of hard-wired developmentally distinct populations or may be due to different maturation states induced by environmental signals. Gaps in our knowledge of dendriticcell developmental pathways have hindered finding answers to these questions. However, recently developed genetic techniques4 to ablate dendritic-cell populations, together with an improved ability to identify specific dendritic-cell precursors, have advanced this area of research. Dendritic cells are closely related to macrophages, which originate from white blood cells called monocytes. They are derived from a common precursor termed the macrophage and dendritic-cell precursor (MDP). Dendritic cells can also be generated5 from monocytes in cell culture using a growth factor called granulocyte macrophage colony-stimulating factor (GM-CSF), and much of what we know about the biology of dendritic cells has been based on the study of these laboratory-derived cells. However, recent elegant work6,7 has shown that monocytes and dendritic cells diverge in their developmental pathways downstream of the MDP conventional dendritic cells in lymphoid tissue arise from a precursor cell in the blood (the pre-dendritic cell), and their differentiation depends on the growth factor Flt3. Hence, under normal conditions, blood monocytes do not give rise to dendritic cells in lymphoid tissue, raising questions about the in vivo counterpart of monocyte-derived dendritic cells. Varol et al.1 and Bogunovic et al.2 studied the development of dendritic cells in the intestine.

8 Crystal Structure of Postperovskite The measured di raction patterns of perovskite (Pv) and postperovskite (PPv) agree well with those calculated from the crystal structure models of these phases. In particular, the Rietveld refinements of the PPv di raction patterns 1 yielded structural parameters which are in excellent agreement with computational predictions 2,3. Thermal Insulation of Sample Controlling the thickness of the insulation layers is challenging in cryogenic Ar loading (Supplementary Fig. 1). Some samples show much broader di raction lines even at high temperature, suggesting thin or no insulation layers. These data were not included in our boundary determination. Some of the previous boundary studies did not use an insulating medium. Clapeyron Slope and Pressure Scale Issue We obtained the Clapeyron slope of the postperovskte boundary from the Pv PPv PPv boundary in (Mg 0 91 Fe 0 09 )SiO 3 because it provides tight constraints. The measured value ( MPa K) is in agreement with the seismologic estimation for the D discontinuity 4 and an earlier computational prediction 2. A larger magnitude Clapeyron slope ( MPa K) was reported in Mg end member 5. However, the pressure scale they used (MgO) is known to yield about a factor of two larger Clapeyron slope compared with the gold scale we used 6 and therefore the discrepancy is likely due to the inconsistency among the di erent pressure scales. Di erences in pressure can be more reliably determined and therefore the measured thickness should be less a ected by this issue. 1

9 E ect of Fe 3 on the Postperovskite Boundary Thickness A significant amount of Fe in mantle silicate Pv and PPv has been found to be Fe 3 (Fe 3 Fe 10 60%) 7 9. Using the same experimental methods, we determined the PPv boundary thickness in (Mg 0 91 Fe )(Fe Si 0 91)O 3 (Supplementary Fig. 2). For the starting material, we synthesized glasses from a MgSiO 3 Fe 2 O 3 mixture under an O 2 atmosphere by containerless method 10. After the high-pressure synthesis of Pv and PPv, the valence state of Fe is confirmed to be 3 by synchrotron Mössbauer spectroscopy. As shown in Supplementary Fig. 2, Fe 3 increases the thickness but to much smaller degree compared to Fe 2 and Fe 3 Al. We note that this sample contains a factor of two more Fe than the compositions presented in the article. When it is normalized for 9 mol% Fe 3 for comparison with our data presented in the article, the PPv boundary thickness is km. Considering the uncertainties, this means that the PPv boundary remains essentially sharp with Fe 3. This demonstrates that our method of constraining the boundary along both forward and reverse transitional paths is capable of detecting a sharp boundary. In addition, a recent study 5 using the same method also detected a sharp PPv boundary in MgSiO 3. The phase boundary in this system is located at GPa and 2500 K. Considering that the PPv transition pressure in MgSiO 3 has been reported to be between 120 and 135 GPa (refs 5,6,11,12), our result suggests that Fe 3 lowers the PPv transition pressure, which is consistent with a computational study 11. The PPv transition is also found in Fe 2 O 3 (refs 1,13) which is the Fe 3 end member of the MgSiO 3 Fe 2 O 3 binary system. In Fe 2 O 3, the PPv transition occurs at much lower pressures, GPa (refs 1,13). 2

10 Calculation of the Postperovskite Phase Fraction Profiles It is important to know the phase fraction profile in the mixed phase region in order to examine seismic detectability of a phase transition 14,15. The profile can deviate from a linear trend depending on the mixing behavior of cations. In other words, the magnitude of the nonlinear deviation (and the shape of the phase loop) can be calculated by knowing the partition coe cients and mixing behaviors of cations. The absolute pressure range over which the profile exists is constrained by boundary thickness measurements. A model based on an assumption of ideal mixing behavior 14 enables the simplest approach to compute the nonlinearity in the phase fraction profiles. The approach requires only partition coe cients. A consensus has emerged for the partition coe cients of Fe among Pv, PPv, and ferropericlase (Fp) in the literature. The existing studies agree that Fe 2 preferentially enters Fp over Mg-silicates (Pv and PPv) 11, Except for an earlier study 16, experimental and theoretical studies 11,17,18,20 22 are in agreement that Fe preferentially enters PPv over Pv, and decreases the PPv phase transition pressure. An earlier result 16, which proposed the opposite trend in Fe partitioning, appears to be contaminated by thermal gradients in the laser-heated diamond-anvil cell, according to a recent report from the same group 19. Because the spin transition in Fp might a ect the Fe partition coe cient 23, we used the partition coe cient of Fe measured above 100 GPa which is after completion of the spin transition in Fp. The partition coe cient of Fe between Fp and Pv (K Pv Fp D (Fe) [Fe Mg] Pv [Fe Mg] Fp ) ranges between 0.1 and 0.3 near the PPv transition pressure depending on studies (Supplementary Tab. 1). Studies also suggested the partition coe cient of Fe between Pv and PPv (K Pv PPv D (Fe)) ranges between 0.15 and 1 (Supplementary Tab. 1). As shown in Supplementary Fig. 3a, the largest deviation from a linear trend was found when the Fe partition coe cients by Auzende et al. 18 are used. This profile is chosen for presentation in Fig. 3. 3

11 The ideal solution model yields the profile with respect to normalized pressure: P norm P (1) P tr (A) P tr (B) where P is pressure, and P tr (A) and P tr (B) are the transition pressures in the end members of a binary system. The profile shown in Fig. 3a is scaled to fit our measured thickness. However, it is important to further examine if P tr (MgSiO 3 ) and P tr (FeSiO 3 ) obtained by fitting the profile to our measured thickness are in reasonable agreement with direct measurements on these values. While an experimental attempt 20 to synthesize FeSiO 3 -PPv was unsuccessful, their first-principles calculation indicates the stability of the PPv phase in FeSiO 3 from 0 GPa. Synthesis of the PPv phase up to 80% FeSiO 3 was successful and their calculation showed that the PPv phase becomes stable at 63 GPa in (Mg 0 5 Fe 0 5 )SiO 3. The measured PPv transition pressure in Mg end member, P tr (MgSiO 3 ), ranges between 120 and 135 GPa (refs 5,6,11,12). The main source of uncertainty is the absolute pressure scales 5. What is robust in the literature is that Fe decreases the PPv transition pressure 11,20,24. Because our data indicates that the Pv PPv mixed phase region exists between 115 and 135 GPa, we choose 135 GPa for P tr (MgSiO 3 ). As shown in Supplementary Fig. 3b, for these two constraints, i.e., P tr (MgSiO 3 ) and P tr (Mg 0 5 Fe 0 5 SiO 3 ), the partition coe cients reported by Auzende et al. 18 produce the phase loop that is the most consistent with our thickness measurements in (Mg 0 91 Fe 0 09 )SiO 3. The comparison of P tr (MgSiO 3 ) and P tr (Mg 0 5 Fe 0 5 SiO 3 ) between our calculation and literature values provides a test. However, because the lower mantle silicates contain a relatively small amount of Fe (approximately 10%), only the shape of the phase loop near this bulk composition is important for the calculation of the PPv phase fraction profile. For the partition coe cients we used in Fig. 3, by Auzende et al. 18, we found that the PPv phase fraction profile depends only on the shape of the phase loop in 0 Fe (Fe Mg) 0.4. The reported partition coe cients were all 4

12 measured within this range and the partition coe cient may not change significantly within this range. Therefore, uncertainties in the binary loop away from 0 Fe (Fe Mg) 0.4 do not contribute to the calculated profile. In addition, an estimated phase loop between Pv and PPv in (Mg,Fe)SiO 3, which extends to 0 Fe (Fe Mg) 0.4, was presented in a previous temperature-quench study 24. This result is in qualitative agreement with the phase loop presented in Supplementary Fig. 3b. Although it is more realistic for the lower mantle, because our sample contains both Fe 3 and Al, it is di cult to separate the e ects from them in (Mg 0 9 Fe 0 1 )(Al 0 1 Si 0 9 )O 3. As discussed above, our data on the PPv transition in (Mg 0 91 Fe 0 09 )(Fe 0 09 Si 0 91 )O 3 suggest that Fe 3 alone does not increase the boundary thickness significantly but does decrease the PPv transition pressure. On the other hand, previous studies on (Mg,Al)(Al,Si)O 3 reported that Al increases the transition pressure and the boundary thickness 11,21, Therefore, the low PPv transition pressure found in (Mg 0 9 Fe 0 1 )(Al 0 1 Si 0 9 )O 3 would be due to the e ects of Fe 3. On the other hand, the large thickness found in (Mg 0 9 Fe 0 1 )(Al 0 1 Si 0 9 )O 3 should be due mainly to the e ect of Al. In addition, a study on the PPv boundary in (Mg 0 85 Fe 0 15 )(Al 0 15 Si 0 85 )O 3 (note the larger amounts of Fe 3 and Al) 28 reported that the mixed phase region exists from GPa to GPa, which is higher pressure than the system we study with lower Fe 3 Al content. Therefore, these observations indicate that Al controls the phase loop shape of (Mg,Fe)(Al,Si)O 3, and we only consider the partition coe cient of Al for the calculation of the PPv phase fraction profile for (Mg 0 9 Fe 0 1 )(Al 0 1 Si 0 9 )O 3. Existing studies are in agreement that Al preferentially enters Pv over PPv 11,21 and increases the PPv phase transition pressure 11,21,25,26. However, the partition coe cient of Al between Pv and PPv (K PPv Pv D (Al)) has not been measured. A computational study 11 estimated it to be 0.44 (Supplementary Tab. 1). In order to investigate the e ects of strong nonlinear behavior, we also PPv Pv 5

13 explore the case of K PPv Pv D (Al) 0 15 (Supplementary Fig. 4a). This value is obtained from Fe partitioning between Pv and PPv but a di erence is that Al preferentially enters Pv over PPv unlike Fe. This latter case is presented in Fig. 3. Unlike the case of (Mg 0 91 Fe 0 09 )SiO 3, because of the existence of Fe 3 it is di cult to compare the transition pressures in the end members between our prediction and existing measurements. In order to incorporate the e ect of the steep radial temperature increase at the D layer, the phase loop calculated from the ideal solution model was shifted with temperature using the measured Clapeyron slope of the PPv transition assuming that the thickness of the loop does not change with temperature. Our data supports no significant thickness change with temperature between 1500 and 3000 K (Fig. 2). We also assumed that the partition coe cients do not change significantly between 1500 and 4000 K, as the temperature dependence of the partition coe cients among Pv, PPv, and Fp are unknown. While the sense of nonlinear deviation (concave up or down) can be obtained just by knowing the e ects of the cations on the phase transition pressure for low-concentration cations 14, the magnitude of nonlinear deviation in the phase fraction profile is sensitive to partition coe cients and nonideality in mixing behavior. The former has been measured for PPv, Pv, and Fp, and we consider all the possible ranges reported in the literature. The latter is unknown for the PPv transition and therefore was not included in our calculation. In Fig. 3, we present the cases with the most nonlinear PPv phase fraction profiles and the largest Fe bu ering e ect of Fp. Even for these cases, we found that nonlinearity is not significant enough for the PPv transition to match the thickness of the D discontinuity 29. Furthermore, the large thickness of the PPv boundary allows the steep temperature increase in the D layer to severely a ect the PPv phase fraction profile, such that the steep radial temperature gradient decreases the rate of increase of the PPv phase fraction and prevents the completion of the PPv transition in the lower mantle. 6

14 A Test for the Ideal Solution Model Olivine-Wadsleyite Boundary In order to examine the uncertainty in the phase fraction profile calculated from the ideal solution model 14, we conducted calculations for the olivine-to-wadsleyite transition (Supplementary Fig. 5). The thickness of the olivine-to-wadsleyite (Ol-to-Wd) transition is significantly larger than that of the 410-km discontinuity and the Fe bu ering e ect of garnet (Gt) may decrease the thickness of the Ol-to-Wd boundary significantly 14,30,31. The Fe partition coe cients among Ol, Wd, and Gt are well known 14,30,31, which allows for the calculation of the Fe bu ering e ects of Gt using the ideal solution model. On the other hand, the nonideal mixing behavior of this system has been measured 31. Therefore, by comparing our ideal solution calculation results with the results from the nonideal mixing model, we can infer the uncertainties in the ideal solution model caused by ignoring the nonideal mixing e ects. The Fe Mg partition coe cients among di erent phases (K Ol Wd D (Fe) 0 5 and K Ol Gt D (Fe) 0 4) are obtained from a partitioning study 30. The transition pressure in Mg end member was obtained from experimental measurements 32,33. Because the phase transition does not exist in the Fe end member, we assigned 8.4 GPa, based on matching the experimentally measured thickness without the Fe bu ering e ect of Gt. The ideal solution calculation shows that the bu ering e ect of Gt decreases the thickness of the Ol-to-Wd boundary by 60% (from 0.35 GPa to 0.14 GPa) for Fe (Mg Fe) 0.1 (Supplementary Fig. 5b), which is in good agreement with results constrained by the nonideal symmetric solution model using a separate dataset 31. Therefore, the e ect is reasonably well reproduced by the ideal solution model. 7

15 The phase fraction profiles calculated from our ideal solution model and the nonideal symmetric solution model 31 are shown in Supplementary Fig. 5a. Both include the Fe bu ering e ects of Gt. We found that the ideal solution model correctly predicts the shape of the phase fraction profile. The magnitude of the nonlinearity estimated from the ideal solution model is in reasonable agreement with the nonideal model constrained by experiments 31. References 1. Shim, S.-H. et al. Crystal structure and thermoelastic properties of (Mg 0 91 Fe 0 09 )SiO 3 postperovskite up to 135 GPa and 2700 K. P. Natl. Acad. Sci. 105, (2008). 2. Tsuchiya, T., Tsuchiya, J., Umemoto, K. & Wentzcovitch, R. M. Phase transition in MgSiO 3 perovskite in the Earth s lower mantle. Earth Planet. Sc. Lett. 224, (2004). 3. Oganov, A. R. & Ono, S. Theoretical and experimental evidence for a post-perovskite phase of MgSiO 3 in Earth s D layer. Nature 430, (2004). 4. Sidorin, I., Gurnis, M. & Helmberger, D. V. Evidence for a ubiquitous seismic discontinuity at the base of the mantle. Science 286, (1999). 5. Tateno, S., Hirose, K., Sata, N. & Ohishi, Y. Determination of post-perovskite phase transition boundary up to 4400 K and implications for thermal structure in D layer. Earth Planet. Sc. Lett. 277, (2009). 6. Hirose, K., Sinmyo, R., Sata, N. & Ohishi, Y. Determination of post-perovskite phase transition boundary in MgSiO 3 using Au and MgO pressure standards. Geophys. Res. Lett. 33, L01310 (2006). 7. McCammon, C. Perovskite as a possible sink for ferric iron in the lower mantle. Nature 387, (1997). 8

16 8. Frost, D. J. et al. Experimental evidence for the existence of iron-rich metal in the Earth s lower mantle. Nature 428, (2004). 9. Sinmyo, R., Hirose, K., O Neill, H. S. C. & Okunishi, E. Ferric iron in Al-bearing postperovskite. Geophys. Res. Lett. 33, L12S13 (2006). 10. Tangeman, J. A. et al. Vitreous forsterite (Mg 2 SiO 4 ): Synthesis, structure, and thermochemistry. Geophys. Res. Lett. 28, (2001). 11. Ono, S. & Oganov, A. R. In situ observations of phase transition between perovskite and CaIrO 3 -type phase in MgSiO 3 and pyrolitic mantle composition. Earth Planet. Sc. Lett. 236, (2005). 12. Murakami, M., Hirose, K., Kawamura, K., Sata, N. & Ohishi, Y. Post-perovskite phase transition in MgSiO 3. Science 304, (2004). 13. Ono, S., Kikegawa, T. & Ohishi, Y. High-pressure phase transition of hematite, Fe 2 O 3. J. Phys. Chem. Solids 65, (2004). 14. Stixrude, L. Structure and sharpness of phase transitions and mantle discontinuities. J. Geophys. Res. 102, (1997). 15. Weidner, D. J. & Wang, Y. Chemical- and Clapeyron-induced buoyancy at the 660 km discontinuity. J. Geophys. Res. 103, (1998). 16. Murakami, M., Hirose, K., Sata, N. & Ohishi, Y. Post-perovskite phase transition and mineral chemistry in the pyrolitic lowermost mantle. Geophys. Res. Lett. 32, L03304 (2005). 17. Kobayashi, Y. et al. Fe-Mg partitioning between (Mg,Fe)SiO 3 post-perovskite, perovskite, and magnesiowüstite in the Earth s lower mantle. Geophys. Res. Lett. 32, L19301 (2005). 18. Auzende, A.-L. et al. Element partitioning between magnesium silicate perovskite and ferropericlase: New insights into bulk lower-mantle geochemistry. Earth Planet. Sc. Lett. 269, (2008). 9

17 19. Sinmyo, R. et al. Partitioning of iron between perovskite postperovskite and ferropericlase in the lower mantle. J. Geophys. Res. 113, B11204 (2008). 20. Mao, W. L. et al. Iron-rich silicates in the Earth s D layer. P. Natl. Acad. Sci. 102, (2005). 21. Caracas, R. & Cohen, R. E. E ect of chemistry on the stability and elasticity of the perovskite and post-perovskite phases in the MgSiO 3 FeSiO 3 Al 2 O 3 system and implications for the lowermost mantle. Geophys. Res. Lett. 32, L16310 (2005). 22. Caracas, R. & Cohen, R. E. Ferrous iron in post-perovskite from first-principles calculations. Phys. Earth Planet. Inter. 168, (2008). 23. Badro, J. et al. Iron partitioning in Earth s mantle: Toward a deep lower mantle discontinuity. Science 300, (2003). 24. Mao, W. L. et al. Ferromagnesian postperovskite silicates in the D layer of the Earth. P. Natl. Acad. Sci. 101, (2004). 25. Caracas, R. & Cohen, R. E. Prediction of a new phase transition in Al 2 O 3 at high pressures. Geophys. Res. Lett. 32, L06303 (2005). 26. Akber-Knutson, S., Steinle-Neumann, G. & Asimow, P. D. E ect of Al on the sharpness of the MgSiO 3 perovskite to post-perovskite phase transition. Geophys. Res. Lett. 32, L14303 (2005). 27. Tateno, S., Hirose, K., Sata, N. & Ohishi, Y. Phase relations in Mg 3 Al 2 Si 3 O 12 to 180 GPa: E ect of Al on post-perovskite phase transition. Geophys. Res. Lett. 32, L15306 (2005). 28. Nishio-Hamane, D., Fujino, K., Seto, Y. & Nagai, T. E ect of the incorporation of FeAlO 3 into MgSiO 3 perovskite on the post-perovskite transition. Geophys. Res. Lett. 34, L12307 (2007). 10

18 29. Lay, T. Sharpness of the D discontinuity beneath the Cocos Plate: Implications for the perovskite to post-perovskite phase transition. Geophys. Res. Lett. 35, L03304 (2008). 30. Irifune, T. & Isshiki, M. Iron partitioning in a pyrolite mantle and the nature of the 410 km seismic discontinuity. Nature 392, (1998). 31. Frost, D. J. The structure and sharpness of (Mg,Fe) 2 SiO 4 phase transformations in the transition zone. Earth Planet. Sc. Lett. 216, (2003). 32. Morishima, H., Ohtani, E. & Kato, T. Thermal expansion of MgSiO 3 perovskite at 20.5 GPa. Geophys. Res. Lett. 21, (1994). 33. Suzuki, A. et al. In situ determination of the phase boundary between wadsleyite and ringwoodite in Mg 2 SiO 4. Geophys. Res. Lett. 27, (2000). 11

19 Figure 1: Schematic diagram (left) and micro-photograph (right) of the samples in the diamondanvil cell under high pressure. The micro-photograph was taken at 140 GPa. Figure 2: Pressure temperature conditions for the stability of Pv (blue triangles), Pv PPv mixture (green triangles), and PPv (red triangles) in (Mg 0 91 Fe )(Fe Si 0 91)O 3. The right and left triangles represent data points measured along the forward and reverse paths, respectively. The solid lines are the phase boundaries determined from our data. 12