The spin and valence states of iron in (Mg0.8Fe0.2)SiO3 perovskite were measured between 0 and 65 GPa using synchrotron Mössbauer spectroscopy. Samples were synthesized in situ in the laser-heated diamond cell under reducing conditions. The dominant spin state of iron in perovskite is high spin at pressures below 50 GPa. Above 50 GPa, the spectra shows severe changes which can be explained by appearance of two distinct iron sites with similar site weightings. One site has Mössbauer parameters consistent with high spin Fe2+, while the other has the parameters previously interpreted as intermediate spin. The latter intermediate-spin assignment is not unique, as similar Mössbauer parameters have been reported for high spin Fe2+ in almandine at ambient pressure. However, our data do not rule out the existence of low-spin iron, which may exist with a smaller fraction and explain the observation of lower spin moments in the X-ray emission spectroscopy of perovskite at high pressure. From these considerations, our preferred interpretation is that iron in perovskite is mixed or high spin to at least 2000 km depths in the mantle, consistent with computational results. Our study also reveals that reducing conditions do not inhibit the formation of Fe3+ in perovskite at deep-mantle pressures.
 Recent studies have identified an iron electronic configuration with a set of Mössbauer parameters (quadrupole splitting: QS = 3.5 ∼ 4.0 mm/s) that are outside the range normally observed for HS (QS = 2 ∼ 3 mm/s) or LS Fe2+ (QS = 0 ∼ 1 mm/s) [McCammon et al., 2008; Lin et al., 2008], with this electronic configuration becoming dominant above 50 GPa. Combined with X-ray emission spectroscopy (XES) studies that indicate a decrease in average spin moment with pressure [Li et al., 2004; Badro et al., 2004; Lin et al., 2008], this new site has been interpreted as Fe2+ in the intermediate spin (IS) state, with a stability field from 40 GPa to at least 135 GPa [McCammon et al., 2008; Lin et al., 2008]. However, computational results do not find this electronic configuration of iron to be stable [Zhang and Oganov, 2006; Stackhouse et al., 2007; Bengtson et al., 2008, 2009], and no mineralogical examples of IS Fe2+ at room pressure have been identified, while there are some reports on IS iron in molecular complexes [see Bengtson et al., 2009, and references therein]. We have conducted a set of experiments designed to better control the conditions under which Pv is synthesized and constrain the likely spin and valence states of iron in the lower-mantle Pv phase.
2. Experimental Method
 The starting material was a (Mg0.8Fe0.2)SiO3 pyroxene synthesized using the same technique as Lin et al.  under reducing conditions. The sample was 95% 57Fe enriched. The starting material was powdered and then pressed into platelets. Among total of five different samples, three platelets were compressed together with a thin (2–3 micron) iron foil with a natural 57Fe level to provide a reducing environment (Figure S1 of the auxiliary material), while the other two samples were loaded without the iron foil. Samples were loaded in diamond cells with 200 or 300 μm culets with argon for thermal insulation and to minimize deviatoric stresses. Small grains of pyroxene with the same composition were used for spacers to allow argon to penetrate underneath and increase thermal insulation. For pressure measurements, ruby grains were placed at the edge of the sample chamber and away from the sample to prevent reaction with the sample during laser heating [Mao et al., 1986].
 The Pv phase was synthesized at MIT using a Nd:YLF laser between 1500 and 2000 K to ensure the iron foils did not melt and react with the silicate. Each sample was pressurized to 37, 50, or 65 GPa and heated for 30 minute cycles 3 or 4 times to ensure full conversion to Pv. The synthesis of Pv was confirmed by X-ray diffraction on a pressure quenched, iron-foil free Pv sample at the GSECARS beamline at the Advanced Photon Source (APS). Laser annealing was performed after any pressure increase, but not after release of pressure from the diamond cells.
 Synchrotron Mössbauer spectra (SMS) were collected at Sectors 3 and 16 of APS. Measurements of SMS on our starting material confirms that it is Fe3+ free (Figure 1 and Table S1). SMS collection time was about 2 hours at Sector 3 and 6 hours at HPCAT, with separate spectra collected with a 10 μm stainless steel foil to obtain relative center shifts (CS). Details on the technique of SMS and the equivalency to conventional Mössbauer parameters are given by Sturhahn .
 In order to extract Mössbauer parameters, spectral fitting was perform using the CONUSS program [Sturhahn, 2000]. The results are shown in Figure 1 and Table S1. All spectra are fit with a three site model. Perovskite synthesized at 37 GPa required a site to account for iron metal that contaminated the spectrum, which has little effect on the Mössbauer parameters of the other two sites. Decompressed samples from 50 and 37 GPa have a magnetic site consistent with a small amount of elemental iron either from the foils or (possibly) due to charge disproportionation of iron during sample synthesis [Frost et al., 2004]. We do not use these fitting results due to relatively high χ2 and instability during spectral fitting.
 Our synchrotron Mössbauer spectra of Pv consists of 1–2 irregularly spaced quantum beats up to 37 GPa and 2–3 beats at pressures above 50 GPa (Figure 1). This is in sharp contrast with the spectra reported by Lin et al.  and McCammon et al.  where evenly spaced 3–4 quantum beats were found (Figure 1). Quadrupole splitting (QS) values are shown in Figure 2 along with the ranges for different spin and valence states from previous high pressure studies. The combination of QS and CS allows us to identify 4–5 sites representing different electronic configurations of iron labeled in Figure 2. At pressures lower than 40 GPa, the Pv component can be fit with of two sites. Site 1 is compatible with HS Fe2+ in Pv [Fei et al., 1994] and reported in previous high pressure studies. Site 5 is a low QS site (0.2–0.6 mm/s) traditionally associated with the formation of HS Fe3+, but the value is also comparable to that expected Mössbauer parameters of LS Fe2+ [Li et al., 2006; Rouquette et al., 2008; Bengtson et al., 2009].
 Three distinct sites were found in the samples synthesized above 50 GPa. Site 3 is the high QS site which is previously interpreted as IS Fe2+ in Pv [McCammon et al., 2008; Lin et al., 2008]. In our experiment, this high QS site does not become the dominant feature of the spectra at high pressure, unlike McCammon et al.  and Lin et al. : the fraction of this site remaining at 20 – 30% both in the samples directly synthesized and laser annealed at 65 GPa, while McCammon et al.  reported 100% for this site at pressures higher than 60 GPa. A similar site with a slightly low QS value (3.3–3.5 mm/s) has been also reported by Jackson et al.  and Li et al. . They also found that the site has a smaller weighting (10–25%), similar to our results. McCammon et al.  argued that the lack of laser heating of Jackson et al.  and Li et al.  may cause the difference. However, in our study Pv is synthesized directly at high pressure with longer heating duration at higher temperature combined with the use of quasi-hydrostatic, thermally insulating argon pressure medium. The iron content in our sample (20%) is slightly higher than McCammon et al.  (12–14%) and may have a small effect on the relative sight weighting. However, computations indicate the spin transition pressure would not change between 0 and 30% Fe in Pv [Bengtson et al., 2008].
 Site 4 has a mid-range QS value (1.0–2.0 mm/s). This site was not given by Jackson et al.  and Li et al. , but by McCammon et al.  with a smaller site weighting (between 0 and 15%). We are uncertain as to the details of site 4, but it may be related to site 2 (HS Fe2+) with some degree of charge delocalization [Fei et al., 1994]. Site 1 contributes to the spectra to higher pressures in previous studies [Jackson et al., 2005; Li et al., 2006], while it disappears above 40 GPa in our study. The persistence of this low-pressure feature is likely due to lack of heating in those previous studies. This highlights the importance of sufficient heating to ensure iron is in the lowest energy configuration at given high pressure.
 We also conducted low-pressure SMS measurements on Pv recovered from high pressure. A total of 3 sites were identified: sites 1, 2, and 5. Site 2 has a similar QS value as site 1. These two sites are likely related to site 1 at high pressure, are better resolved due to the removal of deviatoric stresses at ambient conditions, and are consistent with HS Fe2+. Site 5 is consistent with HS Fe3+ and found to represent 20% (Figure 2), higher than the Fe3+ free Pv that Jeanloz et al.  measured, but consistent with other measurements of Pv decompressed from high pressure [Fei et al., 1994].
 From the comparison of the samples synthesized with and without iron foil, we found that adding the iron foil to ensure reducing conditions does not significantly change the Mössbauer parameters. The site weighting for the Fe3+-like sites (site 5) is lower for samples containing the iron foil, but within the error bars of the measurements. This behavior is similar for Pv synthesized in the multi-anvil press [Frost et al., 2004], and seems to indicate a surprising insensitivity to redox conditions during Pv crystallization.
 The fraction of site 5 ranges 25–43% at high pressure, which is higher than the site weighting for recovered samples. The site fraction measured at high pressure may contain much larger uncertainty than for the measurements on recovered samples, due to severe broadening of the QS value and lower spectral quality. Nevertheless, the higher fraction of site 5 appears to be systematic, as it persists over 6 data points at different pressures. The actual amount of Fe3+ in our samples may be better estimated from the spectra measured on pressure quenched samples (Fe3+/ΣFe ∼20%), as unloading the sample at room temperature is unlikely to change the valence state of iron. The Mössbauer parameters of LS Fe2+ are virtually indistinguishable to that of HS Fe3+ (Figure 2) and LS Fe2+ should transform back to HS during unloading [Rouquette et al., 2008]. If we assume our low pressure spectra give a more accurate accounting of Fe3+ in our sample, this means up to 5–20% of the additional low-QS Mössbauer signal at high pressure could be due to LS Fe2+.
 Recent studies have focused on the appearance the high QS site necessary to fit the spectra, inferring it to be IS Fe2+ due to Jahn-Teller distortion of the iron 3d orbitals [McCammon et al., 2008; Lin et al., 2008]. While IS iron has been documented in some molecular complexes [see Bengtson et al., 2009, and references therein], it is notable that there is no known example of IS iron in silicate and oxide minerals at ambient pressure to our knowledge. In addition, computational results do not find the IS state to be stable in Pv at lower-mantle pressures [Zhang and Oganov, 2006; Stackhouse et al., 2007; Bengtson et al., 2008]. Bengtson et al.  calculated the QS of IS Fe2+ in the A site to be much lower (∼0.7 mm/s) than 3.5–4.0 mm/s. On the other hand, some examples exist in the literature of HS Fe2+ with large QS (>3.5 mm/s), including synthetic almandine [Murad and Wagner, 1987] and naturally occurring garnet in eclogite [Li et al., 2005], which have (distorted) dodecahedral coordination environments most similar to the A site in Pv. Low spin Fe3+ also appears to have QS up to 3.5 mm/s in recent work by Catalli et al. .
 The interpretation of the high QS site as IS comes in part from X-ray emission spectroscopy (XES) of Pv [Li et al., 2004; Badro et al., 2004; Lin et al., 2008]. These spectra are much more sensitive to the spin state of iron in the sample, with coordination environment and valence state having small effects [Vanko et al., 2006]. However, it only provides information on average spin state in the sample.
 The decrease in satellite peak observed for Pv at high pressure may be due to spin pairing, with Fe going to IS or LS. On the other hand, the production of iron metal may also mimic the change in spectral features used to infer spin state [Rueff et al., 2008], which combined with recent observation of the charge disproportionation of iron in Pv (3Fe2+ → 2Fe3+ + Fe0) makes this a plausible explanation [Frost et al., 2004; Auzende et al., 2008]. The possibility remains other factors influence the spectral shape, as half of the intensity reduction from Li et al.  occurs between 0 at 27 GPa, which should reflect in the appearance of a significant amount of IS or LS iron by 30 GPa. This is inconsistent with current interpretation of Mössbauer spectra in that pressure range. Other probes have also failed to unambiguously detect spin transitions in Pv at high pressure [Narygina et al., 2009].
 Another alternative is to have LS Fe2+ and/or LS Fe3+. As discussed above, from the difference in the site fractions of site 5, we cannot rule out the possible existence of LS Fe2+ at high pressure. In addition, the Mössbauer parameters of LS Fe3+ are very similar to HS Fe2+, leaving the possibility of the existence of LS Fe3+ [Xu et al., 2001]. Indeed, LS Fe3+ may appear at much lower pressures than LS Fe2+ according to recent studies [Li et al., 2005; Stackhouse et al., 2007; Catalli et al., 2009].
 While the appearance of the high QS term is intriguing, it is important that experimental probes yield more consistent results as in the case for the (Mg,Fe)O system [Lin and Tsuchiya, 2008]. While our experiment does not give a definitive answer as to the nature of Fe2+ in Pv, we can rule out a predominantly intermediate spin iron in the mantle at least to 2000-km depth. We find that at high pressure Fe2+ exists in two different environments, but both are likely high spin along with octahedrally coordinated Fe3+ and possibly small amounts of LS Fe2+ (and LS Fe3+). Fe3+ is produced in the similar amounts under different redox conditions, consistent with results from lower pressure perovskite synthesis [Frost et al., 2004] and a recent study without oxygen fugacity control [Auzende et al., 2008; McCammon et al., 2008].
 For the discussion of the spin state of iron in the mantle it is important to know the effect of temperature and aluminum on the spin transition. Temperature will have an effect on the population of iron in different spin configurations [Hofmeister, 2006; Sturhahn et al., 2005]. Aluminum also appears to alter the valence state of iron in Pv and increases Fe3+/ΣFe to 60% [Frost et al., 2004]. As shown in recent studies [Stackhouse et al., 2007; Catalli et al., 2009], the spin state of iron could be valence-dependent: Fe3+ may undergo spin pairing at much lower pressure. The investigation of high temperature (in excess of 2000 K) and systems with realistic amount of Al for the mantle (5–10 mol%) would be important for determining the implication for the lower mantle.
 The authors would like to thank J. Barr and T. Grove for help synthesizing the starting material and acknowledge S. Speakman, K. Catalli, and V. Prakapenka for experimental assistance. We would like to thank the editor, anonymous reviewers, D. Morgan, A. Bengtson, and R. Jeanloz for helpful comments. Use of Sector 3 was partially supported by COMPRES. Portions of this work were performed at HPCAT (Sector 16), Advanced Photon Source (APS), Argonne National Laboratory. HPCAT is supported by DOE-BES, DOE-NNSA, NSF, and the W.M. Keck Foundation. APS is supported by DOE-BES, under contract DE-AC02-06CH11357. This work was supported by NSF to S.-H. S. (EAR0738655).