The electronic environment of the iron sites in post-perovskite (PPv) structured (57Fe,Mg)SiO3 has been measured in-situ at 1.12 and 1.19 Mbar at room temperature using 57Fe synchrotron Mössbauer spectroscopy. Evaluation of the time spectra reveals two distinct iron sites, which are well distinguished by their hyperfine fields. The dominant site is consistent with an Fe3+-like site in a high spin state. The second site is characterized by a small negative isomer shift with respect to α-iron and no quadrupole splitting, consistent with a metallic iron phase. Combined with SEM/EDS analyses of the quenched assemblage, our results are consistent with the presence of a metallic iron phase co-existing with a ferric-rich PPv. Such a reaction pathway may aid in our understanding of the chemical evolution of Earth's core-mantle-boundary region.
 A sintered platelet of polycrystalline orthoenstatite-structured (57Fe0.13Mg0.87)SiO3, characterized by >95% high-spin divalent iron [Jackson et al. 2009], with lateral dimensions of ∼30 × 30 μm2 and a thickness of <10 μm was loaded into the sample chamber of a pre-indented rhenium gasket in a sandwich configuration with dehydrated NaCl as a pressure medium and marker. Gold was placed ∼20 microns from the sample for use as an additional pressure marker. Two diamond anvil cells equipped with beveled anvils were prepared in this manner, then pressurized to over one megabar. The high temperature synthesis of PPv was performed with the CO2-laser heating system at the University of Nevada, Las Vegas. Angle-dispersive powder x-ray diffraction (λ = 0.34531 Å) at the High-Pressure Collaborative Access Team was used to confirm the synthesis of a post-perovskite structure consistent with CaIrO3-type (Cmcm) and 2 × 1 (P21/m) PPv symmetries (see Figure S1 of the auxiliary material) [Oganov et al., 2005; Tschauner et al., 2008]. The lattice parameters of NaCl-B2 and gold were determined from these measured spectra to determine the pressures in the two different DACs, namely P = 1.19 ± 0.04 Mbar and P = 1.12 ± 0.04 Mbar, respectively [Dewaele et al., 2004; Fei et al., 2007]. Including the effects of thermal pressure, the samples likely experienced an additional ∼10 GPa of pressure while heated. More details on sample preparation, synthesis, and pressure determination can be found in the auxiliary material.
3. Synchrotron Mössbauer Spectroscopy (SMS) Experiments and Data Evaluation
 SMS experiments were performed at beamline 3-ID-B of the Advanced Photon Source. The hyperfine parameters of iron are directly determined by analysis of the SMS time spectra. Additional constraints on the isomer shift (IS) and the quadrupole splitting (QS) of the sample and its thickness were provided by placing a natural stainless steel (SS) foil with a physical thickness of 3 μm in the x-ray beam path [Alp et al., 1995]. Therefore, spectra were collected with and without SS foil with collection times of ∼2 hours/spectrum (Figures 1a and 1b). Spectra without SS foil were evaluated first, to constrain the quadrupole splittings and weights of the sites. With this information, the isomer shifts of all sites were determined using the spectra containing the SS foil reference. The effective thickness, η, of the samples were determined to be ∼0.03, thus having negligible effects on the time spectra. All IS values are relative to α-iron; the IS of stainless steel relative to α-iron is −0.09 mm/s [Hawthorne, 1988]. Details of experimental set-up and data analysis using CONUSS as given by Jackson et al.  and Sturhahn , respectively.
 Two sites that are well distinguished by their hyperfine fields dominate the spectra. One site, representing 71% of the total iron (site #1), is characterized by a low QS around 0.74 mm/s and an IS of 0.45 mm/s (Figures 1a and 1b and Table 1). These QS and IS values are consistent with previous reports on the behavior of high-spin Fe3+ in similar coordination environments [McCammon, 1997; Jackson et al., 2005; Li et al., 2006] (see Table S1). The remaining 29% of the iron (site #2) is characterized by an IS of about −0.16 mm/s with no field gradient. We identify this site as a metallic iron phase [Pipkorn et al., 1964]. To our knowledge, no oxide or silicate exhibits such a small negative IS [Dyar et al., 2006]. Transmission spectra were calculated from CONUSS using these best-fit hyperfine parameters (Figure S2).
Table 1. Best Fit Hyperfine Parameters Obtained From Fits to the SMS Time Spectraa
Fe3+-like (Site #1) PPv
Metallic Iron Phase (Site #2)
Wt, weight fraction of the site; QS, quadrupole splitting; IS, isomer shift (relative to α-iron); FWHM, full width at half maximum of the IS. Parameters with error values were varied in the fits. The weights are normalized to 100%. Uncertainties are given in parenthesis at the 90% confidence level for the last reported significant digit. The normalized χ2 values for the fits are all around 1.3.
 We note that the addition of a minority site contributing not more than 5% with a high QS of around 4 mm/s could be incorporated into our above model without changing our normalized χ2 values. This minority site is suggestive of a Fe2+-like site in a non-zero spin state in an 8–12 coordinated site, which has been observed in all existing Mössbauer measurements on Pv at megabar pressures [Jackson et al., 2005; Li et al., 2006; Lin et al., 2008; McCammon et al., 2008]. One can see that an addition of greater than 5% of a high QS component distinctive to Pv produces an observable mis-match to the SMS spectra (see Figure S3).
4. Iron-Rich Clusters in Post-perovskite Assemblages
 In most iron-bearing oxide and silicate phases, oxygen fugacity controls the valence state of iron in the host phase. For example, the formation of Fe2O3, Fe3O4, FeO, and Fe metal at ambient pressure are typically controlled by the available oxygen. In contrast to these phenomena, the amount of Fe3+ in silicate perovskite is suggested to be independent of oxygen fugacity and correlated with Al3+ content [Lauterbach et al., 2000]. It has been suggested that the formation of Fe3+ in aluminous silicate perovskite is achieved by self-reduction of Fe2+ to form iron metal and Fe3+ via the following simplified reaction: 3Fe2+ → 2Fe3+ + Fe0metal [Frost et al., 2004]. Analytical transmission electron microscopy measurements performed on quenched samples from multi-anvil [Lauterbach et al., 2000; Frost et al., 2004] and diamond-anvil-cell experiments [Kobayashi et al., 2005; Auzende et al., 2008] have shown that such a reaction is possible in Al-bearing and in Al-free silicate Pv. Recent theoretical investigations have also suggested that the formation of metallic iron could be energetically favorable [Zhang and Oganov, 2006]. In Al-free systems, the reaction would proceed as follows:
with an enthalpy, ΔH, equal to −2.266 eV for Pv and −2.575 eV for PPv [Zhang and Oganov, 2006]. However, there has been no strong experimental evidence for such a reaction involving PPv. According to Kobayashi et al. , metallic iron was detected in an analytical transmission electron microscopy analysis of a quenched amorphous PPv-bearing assemblage from P ∼ 1.2 Mbar, but the grains were too small to be confidently associated with a particular phase. According to Auzende et al. , metallic iron was detected in similar quenched phase assemblies from P < 1 Mbar, but was not observed in their quenched assembly from P ∼ 1.2 Mbar. In both studies, chemical analysis was done at room-PT conditions, San Carlos olivine was used as starting material, and (Mg,Fe)O was always present. Only in the study by Kobayashi et al.  was there complementary in-situ x-ray diffraction confirmation of the synthesized PPv phase assemblage. Neither study performed in-situ analysis of iron's valence state.
 In the absence of (Mg,Fe)O, several studies suggest that Fe favors PPv compared to Pv [e.g., Caracas and Cohen, 2005; Tateno et al., 2007; Auzende et al., 2008]. Based on previous Mössbauer measurements on Al-free silicate Pv, approximately 10% of the iron is typically Fe3+ [McCammon, 1998]. Therefore, when PPv is formed, 10% Fe3+ is already present from the Pv phase. The remaining ∼90% Fe2+ in Pv would transform into ∼30% iron metal and 60% Fe3+ at the structural transition to PPv according to equations (1a) and (1b) [Frost et al., 2004; Zhang and Oganov, 2006], thus resulting in the 29±2% metallic iron phase and 71% Fe3+ observed in the SMS time spectra for the PPv phase. Note that if one considers an upper bound of 20% Pv (Figure 2) and equal partitioning of Fe between Pv and PPv, then the iron bound in Pv could contribute at most ∼7% metallic iron and ∼13% Fe3+ (according to equation (1b)) to the SMS signal. Therefore, at least ∼22% of the metallic iron phase observed is unequivocally associated with PPv.
 If the reaction proceeds according to equations (1a) and (1b) above, the amount of the metallic iron and α-PbO2 structured SiO2 phase [Dubrovinsky et al., 1997] in the diffracting volume would be 0.8% and 2.4%, respectively. The metallic iron phase and SiO2 formed in this reaction (equations (1a) and (1b)) could be poorly crystalline, nanocrystalline, and/or below the current detection limit of high-pressure powder diffraction. However, scanning electron microscopy (SEM) allows one to image features on the 100's of nanometer scale. A LEO 1550VP Field Emission SEM was therefore used to analyze the quenched assemblage from P = 1.19 Mbar. Analysis by SEM combined with semi-quantitative energy dispersive spectroscopy (EDS) shows a small (∼500 nm) iron-rich cluster within the (Mg,Fe)SiO3 quenched matrix (Figure 3). A line EDS analysis showed that the Fe content in the cluster-area is 50%, with a precision error of 0.3%. The Fe content in the surrounding area varies from 6% to 10.7%, with similar precision and indications iron migration towards the edge. Distinct areas of SiO2 enrichment could not be decoupled from the iron-rich and silicate areas, as the excitation volume is ∼5 μm3 below the surface.
 The PPv phase assemblage containing a metallic iron-rich phase presented here represents a likely reaction pathway to the higher PT assemblages suspected at the core-mantle boundary region. The co-existence of other likely components, namely (Mg,Fe)O, CaSiO3 perovskite, alumina, and partial melts should have an effect on this reaction process. The presence and composition of metallic iron-rich clusters in the reaction pathway from Pv to PPv should provide insights into the compositional evolution of this region.
 We thank J. Zhao for technical assistance, D. Stevenson for discussions, D. Adams and two anonymous reviewers for their comments. Support for this work was provided in part by the National Science Foundation (NSF) EAR 0711542 and Caltech (J.M.J.), NSF-EAR 0552010 and NNSA Cooperative Agreement DOE-FC52-06NA27684 (O.T.), the U.S. DOE, Office of Science, BES under DE-AC02-06CH11357, and COMPRES under NSF Cooperative Agreement EAR 06-49658. X-ray diffraction experiments were performed at HPCAT (Sector 16, APS). SEM and EDS analyses were carried out at the Caltech GPS Division Analytical Facility (MRSEC Program of the NSF under DMR-0080065).