Chemical reaction between liquid iron and (Mg,Fe)SiO3 perovskite was investigated by a combination of laser-heated diamond anvil cell (LHDAC) and analytical transmission electron microscope (TEM). The compositions of quenched liquid iron in contact with perovskite were determined quantitatively using both energy-dispersive (EDS) and electron energy loss (EELS) spectra. Results demonstrate that the dissolution both of O and Si into molten iron is significantly enhanced with increasing pressure. The quenched liquid iron contains 5.3 wt% O and 2.8 wt% Si at 97 GPa and 3150 K that can account for the 7% core density deficit. Reaction products such as SiO2, FeSi, and FeO were not observed as stable phases in this study. The bottom of the mantle in contact with outer core liquid could be depleted in iron. O and Si are important light elements in the core, whose contents may have increased over the geological time.
 Seismic observations indicate that the Earth's outer core is less dense by 5–15 wt% than the pure liquid iron [e.g., Birch, 1964; Laio et al., 2000; Anderson and Isaak, 2002]. The core is therefore thought to contain significant amount of light element(s) such as S, O, Si, C, or H [Poirier, 1994]. The identification of light elements in the core, however, still remains uncertain. Since S, C, and H are highly volatile elements, they may have accreted only in minor amount [Ringwood, 1977]. Allegre et al.  estimated 1.21 wt% sulfur in the core that is much less than that required to account for the density deficit. In contrast, O and Si are the principal elements in the mantle silicates. O'Neill et al.  argued that a combination of O and Si are the most plausible light elements satisfying the cosmochemical and geochemical requirements if their solubilities in liquid iron are simultaneously high at core pressure and temperature.
 Several experiments were previously performed in order to investigate the chemical composition of liquid iron in contact with (Mg,Fe)SiO3 perovskite. O'Neill et al.  reported small amounts of O and Si in quenched liquid iron at 25 GPa using multi-anvil apparatus. The LHDAC experiments by Knittle and Jeanloz  demonstrated that large amount of O dissolved into liquid iron at ∼70 GPa based on scanning microprobe analyses. They also recognized the lesser amounts of Si and Mg in quenched molten iron. Goarant et al.  reported similar observations at ∼70 and ∼130 GPa using analytical TEM. Hillgren and Boehler  found that the dissolution of O and Si was enhanced with increasing pressure and temperature. In these earlier LHDAC studies, however, chemical compositions of liquid iron-alloys were not quantitatively determined, due to the low space resolution of the scanning microprobe [Knittle and Jeanloz, 1991; Hillgren and Boehler, 1998] and the difficulty in O analysis using EDS analytical system [Goarant et al., 1992]. Here we quantitatively determined the O and Si contents in liquid iron to be in equilibrium with (Mg,Fe)SiO3 perovskite up to 97 GPa and 3150 K by a combination of LHDAC and analytical TEM using both EDS and EELS spectra.
2. Experimental and Analytical Techniques
 We conducted a total of twenty-seven runs at pressures ranging from 25 to 97 GPa and temperatures from 2300 to 3150 K. Part of the results below 49 GPa have already been reported in Takafuji et al. . Similar experimental and analytical methods were used in this study. Two starting materials were prepared as powder mixtures of pure iron and gel with a composition of Mg0.9Fe0.1SiO3 or Mg0.8Fe0.2SiO3 (Table 1). The sample mixture was loaded into a 100-μm hole drilled in rhenium gasket, together with insulation layers of pure (Mg,Fe)SiO3 gel. The cell was placed at 400 K in Ar atmosphere before pressurizing in order to minimize the moisture on the sample. We used diamond anvils with 300-μm culet for compression. Samples were preheated at ∼2000 K to synthesize perovskite from gel before increasing temperature to melt iron. The uncertainty in temperature was ±150 K. Pressure was determined using the ruby-fluorescence technique at room temperature after heating the sample. Pressure at high temperature over 3000 K may be higher by 20% than that measured at room temperature after quenching, due to a contribution of thermal pressure. The oxygen fugacity was estimated to be between 1 and 2 log units below the iron-wüstite buffer [Takafuji et al., 2004].
 Samples were recovered from LHDAC and investigated with analytical TEM. We determined chemical compositions both of coexisting silicate perovskite and quenched liquid iron using EDS analytical system except for O (Table 1). Although the EDS system with an ultra-thin window can detect O, quantitative determination of O content is difficult because of a large absorption effect and peak overlapping with iron. The O analyses were therefore obtained from the EELS spectra using newly developed high-resolution energy-filtered field emission-TEM (JEOL-3100FEF omega filter). The cross-section ratio between oxygen and iron was determined experimentally by using a thin film standard Fe2O3 as EELS-k-factors in analogy to the thin film EDS analysis. The O content reported by Takafuji et al.  was recalculated using these EELS-k-factors. All the EDS and EELS analyses were made only for extremely thin parts where electron transparent rates were higher than 94%.
 Experiments were performed above the melting temperature of iron [Shen et al., 1998] except runs #5 and #6 (Table 1). Figure 1 shows TEM image and oxygen energy-filtered map of the quenched liquid iron and perovskite (run #2). Melting of iron was recognized from textures and selected area electron diffraction (SAED) patterns. A positive curvature inside the grains of perovskite is common when iron was molten (Figure 1a), but such texture is not observed when iron was solid. It is noted that dihedral angle at liquid iron-perovskite grain boundary is below 60° above ∼40 GPa [Takafuji et al., 2004]. The SAED patterns of the quenched molten iron show a Debye-ring feature, indicating that it is now composed of fine-grained aggregates of α-Fe with minor amounts of FeO-wüstite and SiO2-stishovite. These wüstite and stishovite grains are very small and are mixed with iron quench-crystals (Figure 1a). Both are not concentrated at the iron-perovskite interface but are homogeneously distributed in the melt pool. It is supported by the chemical uniformity within the melt pocket (Table 1 and Figure 1b). These observations suggest that wüstite and stishovite are quench-crystals formed from molten iron during temperature quenching. In addition, measurements of several different melt pools showed similar O and Si contents (run #2-1, #2-2). This also supports that O and Si were included in the liquid phase. Chemical segregation hardly occurs within the melt pocket during temperature quenching in the LHDAC experiments, owing to a rapid cooling rate. In the experiments conducted at subsolidus temperatures, compositions of perovskite were identical to the original one (runs #5 and #6), indicating that chemical reaction scarcely occurred.
 Both O and Si contents in quenched liquid iron significantly increased with pressure (Figure 2). This is clearly observed among experiments made at nearly constant temperature; 1.9 wt% O and 1.5 wt% Si at 55 GPa and 3000 K (run #3) and 4.7 wt% O and 3.1 wt% Si at 95 GPa and 3050 K (run #2). Mg was not detected at low-pressure runs but was measurable above 93 GPa. The FeSiO3 component in perovskite decreased concurrently with increasing O and Si contents in iron (Figure 3). Run #7 in which iron coexisted with FeO-rich perovskite shows that molten iron tends to have slightly higher O and lower Si contents. This is likely due to higher FeO/Fe ratio in the bulk sample, in other word higher oxygen fugacity.
 Previous experimental studies by Knittle and Jeanloz  reported that chemical reaction between liquid iron and (Mg,Fe)SiO3 perovskite produced SiO2 phase and metallic alloys of FeSi and FeO. In contrast, we observed no reaction products as stable phases other than iron-rich metal and perovskite under the TEM, consistently with Hillgren and Boehler . It suggests that FeSiO3 component in perovskite dissolved into molten iron without forming any other reaction products. This is also supported by the fact that the Si/O molar ratios of quenched liquid iron are close to 1/3 (Table 1).
 Present results show that the dissolution both of O and Si from silicate perovskite into molten iron is significantly enhanced with pressure (Figure 2). Positive pressure effect on the solubility of O in liquid iron is consistent with many of earlier experimental studies [e.g., Ohtani et al., 1984; Ito et al., 1995; Hillgren and Boehler, 1998] but contradicts the results by O'Neill et al.  and Rubie et al. . Pressure dependence of the oxygen solubility can be positive due to the higher compressibility of liquid iron than that of perovskite.
 Our data indicate that the outer core liquid has large simultaneous solubilities of O and Si. Birch  calculated the density deficit of the outer core to be 10 wt% using shock measurements of pure iron. More recently, Anderson and Isaak  found on the basis of static compression data that it ranges from 2.9 to 7.0 wt% when melting temperature of iron at 330 GPa is assumed to be 7500 to 4800 K. Theoretical calculations also support the 5 to 6 wt% core density deficit [Laio et al., 2000]. Present experiments demonstrate that the combined solubilities of O and Si in liquid iron in equilibrium with (Mg,Fe)SiO3 perovskite at 93-97 GPa and 3050-3150 K can already account for the 7 wt% density deficit (Figure 4a). Since the outer core liquid is most likely in chemically equilibrium, at least with the base of the silicate mantle, both O and Si are inevitably the predominant light elements in the core. Although minor presence of other light elements is not precluded, they would be much less important.
 The O and Si contents in liquid iron can be treated in terms of the equilibrium:
All our data from 25 to 97 GPa were fitted to the following equation as a function of pressure, temperature, and FeSiO3 component in coexisting perovskite:
where aFe, aSi, and aO are the activities of Fe, Si, and O in metal, respectively, that are assumed to be equal to their mole fractions. aFeSiO3 is the activity of FeSiO3 component in perovskite, calculated according to Fabrichnaya . ΔH, ΔS, and ΔV are the changes in enthalpy, entropy, and volume, respectively, for the reaction (1) and R is the gas constant. Fitting results give ΔH = 330888(±43267) Jmol−1, ΔS = −3.3(±17.8) JK−1mol−1 and ΔV = −1860(±145) JGPa−1.
 The simultaneous solubilities of O and Si in liquid iron in equilibrium with (Mg,Fe)SiO3 perovskite at the core-mantle boundary condition (135 GPa and 3500 K) are estimated from the equation (2). They are illustrated in Figure 4b as a function of molar fraction of FeSiO3 component (XFeSiO3) in coexisting perovskite. This diagram indicates that the bottom of the mantle in contact with the liquid core with <10% density deficit could be extremely depleted in iron (XFeSiO3 < 0.01). This extrapolation may include large errors due to the uncertainty in pressure effect on ΔV, but the base of the mantle is likely depleted in iron compared to the composition of bulk mantle. Such iron-depleted mantle is buoyant and will be replaced by the normal mantle with XFeSiO3 ≈ 0.1. Then, O and Si should dissolve more from the normal mantle by the chemical reaction with the liquid outer core. In this manner, the core becomes more enriched in O and Si with time.
 The Mg/Si atomic ratio of the Earth's upper mantle is 1.27 [Sun, 1982], that is much higher than 1.05 for C1 meteorites. The latter value is usually assumed to represent Mg/Si ratio of the primordial solar nebula, and hence the upper mantle appears to be relatively depleted in Si. Analysis of the meteorite data shows that the core needs to contain 5 to 7 wt% Si to reconcile the composition of bulk silicate Earth with that of carbonaceous chondrite [O'Neill, 1991; Allegre et al., 2001]. Present results do not directly predict the Si/O ratio in the core. However, when the 7% core density deficit is assumed, the outer core may include ∼3 wt% O and ∼6 wt% Si as light components to be consistent with the scenario that missing Si is in the core (Figure 4b).
 We thank M. Toriumi, K. Kurita, and S. Ohmori for discussions. Comments by anonymous reviewers improved the manuscript. This research was supported by grant-in-aid and Nano-technology Support Project from Monbukagakusho.