We have performed in situ experiments on liquid immiscibility in Fe-O-S melts at 3 GPa and up to 2203 K using a synchrotron X-ray radiographic technique. The difference between immiscible melts and a miscible melt can be clearly observed in radiographs. The immiscibility gap of the Fe-O-S melt shrinks with increasing temperature at 3 GPa. Two separated phases appeared from a miscible melt during quenching. Without in situ observations, the two phases observed in quench textures would be interpreted as either quench products from primary immiscible melts at high temperature, or those exsolved from a homogeneous melt to immiscible melts passing through the stability field of immiscible melts during quenching. In situ measurements are required in order to determine the immiscibility gap of the liquid Fe with light element(s). Our results have important implications for the formation and chemical composition of the cores of Earth and Mars.
 In previous studies in the Fe-O-(S) system [Urakawa et al., 1987; Kato and Ringwood, 1989; Ringwood and Hibberson, 1990; Tsuno et al., 2007], investigations of the liquid immiscibility were based on the textural observation and chemical analysis of recovered high pressure run products. However, there is no consensus on textural interpretation of the immiscible melts in these studies. Based on the phase relations of Fe-binary and ternary systems at ambient pressure [e.g., Kubaschewski, 1982; Kress, 1997], a miscible melt at high temperature conditions passes through the stability field of immiscible melts during quenching. It is, therefore, necessary to determine directly the stability field of immiscible melts and a miscible melt at high-pressure and high-temperature conditions. Previously, in situ observations of liquid immiscibility between silicate melts and fluids at high pressure using the synchrotron X-ray radiographic technique were reported [e.g., Mibe et al., 2004]. In this study, we have developed a new method for in situ observations in liquid Fe with light elements, and have determined a precise liquid immiscibility gap in the Fe-O-S melt at 3 GPa.
 High-pressure experiments were carried out using a Kawai-type double-stage multianvil system driven by DIA-type cubic press. We used a 1500-ton multianvil press (SPEED-1500) on BL04B1 beamline of SPring-8, Japan, and a 700-ton multianvil press (MAX-III) at BL14C2 beamline on KEK-PF, Japan. For performing X-ray radiography measurements, we used a white X-ray beam at SPring-8, and a monochromatic X-ray with 35 keV at KEK-PF. Direct X-ray beams passed through the cell assembly at high pressure, and transmitted X-rays were observed as radiographic images using charge-coupled device (CCD) cameras with a YAG fluorescence screen. Real time radiographic images were recorded as a digital file.
 Starting compositions were Fe65O27S8, Fe62O23S15, Fe61O19S20, Fe57O16S27, and Fe54O13S33 (in atomic ratio) (Table 1), by mixing Fe, FeS, and Fe0.91O. We used a sintered Al2O3 sample container because it is less reactive with Fe-O-S melts compared to sintered MgO and hBN capsules that were also commonly used for liquid Fe with light element(s) samples. The pressure medium along the X-ray path was composed of MgO and boron-epoxy, and boron epoxy windows sandwiched by pyrophyllite were used as gasket in order to minimize X-ray absorption from the cell assembly. Temperature was monitored using a W97Re3-W75Re25 thermocouple without correction for the pressure effect on emf. Temperature uncertainty based on the fluctuation of EMF during the measurements was less than 10 K. Detailed cell assembly was described in elsewhere [Suzuki et al., 2002].
Table 1. Experimental Conditions and Results for X-Ray Radiographic Measurements at 3 GPa
Liq (m), Liq (i) (1723–2003 K)d One liquid (2023–2073 K)d
Liq (m), Liq (i) (1723–1873 K)e One liquid (1893–1973 K)e
One liquid (up to 1833 K)
One liquid (up to 1973 K)
 All experiments were performed at a constant pressure (3 GPa) which was calibrated in a separated experiment at an oil pressure of 350 ton and up to 1673 K (S1673), by measuring a lattice constant of hBN with an equation of state of hBN [Urakawa et al., 1993]. Pressure uncertainty based on high-temperature calibration was determined to be within 0.3 GPa.
 In order to determine the immiscibility gap in a wide compositional range, additional quench experiments at 3 GPa were performed using a 3000-ton Kawai-type multianvil press installed at Tohoku University, with starting compositions of Fe65O27S8, and Fe62O23S15 (Table 2). Two separated phases in the quench experiments were determined to be primary immiscible melts at high-temperature conditions (1883 and 2013 K) based on the in situ measurements at the same pressure and the same starting compositions (Tables 1 and 2). Texture and chemical compositions of recovered products were analyzed using a SEM/EDS (JEOL JSM-5410) and an electron microprobe (JEOL JXA-8800M). Analytical conditions of the electron microprobe are 20 keV of an accelerating voltage and 10 nA of a beam current. Chemical compositions of the primary immiscible melts, as well as the experimental conditions were listed in Table 2.
Table 2. Experimental Conditions and Chemical Compositions of Primary Two Immiscible Melts in Quench Experiments at 3 GPa
Experiments were performed at Tohoku University, except for S1692, performed at SPring-8.
Starting compositions are shown by atomic %.
Liq (m) and Liq (i) correspond to Fe-S metallic and Fe-O ionic melts, respectively.
Errors are shown in parentheses.
Analytical standards were pure iron for Fe, Fe2O3 for O, FeS for S, and Al2O3 for Al.
3. Results and Discussion
 We observed three types of melting and quench processes in the in situ measurements. Figure 1 shows the radiographic images of immiscible melts at 1973 K, and after quench, with a starting composition of Fe65O27S8 (S1692). Dark and light gray areas of radiographic images in Figures 1a and 1b correspond to Fe-S metallic and Fe-O ionic phases, respectively (Figure 1 and Table 2). This result shows that primary immiscible melts can be clearly observed in radiographs, and these two melts were quenched as two separated phases.
Figure 2 shows representative radiographic images (S1699) of immiscible melts and a miscible melt. We clearly observed the difference in images between immiscible melts and a miscible melt with starting compositions of Fe62O23S15 (S1699, Figure 2) and Fe61O19S20 (S1697). Primary immiscible melts were observed up to 2003 K (Figure 2a and Table 1). The dark and light gray regions correspond to the Fe-S metallic and Fe-O ionic melts. The boundary of these two melts became unclear in radiographs from 2003 to 2023 K due to severe expansion and move of the metallic melt (Figure 2b and Table 1), and a complete miscible melt was formed above 2023 K (Figure 2c and Table 1). The textural change has began abruptly. We increased the temperature to 2073 K, and the miscible melt was observed up to 2073 K. It was separated into two phases around 1883 K during quenching (Figures 2d, 2e, and 2f) because the miscible melt may pass through the stability field of immiscible melts at lower temperatures (1723–2003 K) (Table 1) during quenching. Therefore, if we did not perform the in situ measurement and the immiscibility gap was interpreted based only on the textures of the recovered product, we may overestimate the miscible/immiscible transition temperature by about 190 K.
 The samples with starting compositions of Fe57O16S27 (FeOSPF002) and Fe54O13S33 (S1695) showed a homogeneous melt to the target temperature (1833 and 1973 K, respectively), and also after quenching. Figure 3 shows the representative images of a homogeneous texture at 1833 K (Figure 3a) and after quenching (Figure 3b), with a starting composition of Fe57O16S27 (FeOSPF002). An immiscible two-liquid region does not exist in this compositional range. A back-scattered electron image of the recovered product showed a homogeneous texture (Figure 3c) and supports the miscible melt at high-temperature conditions. These results indicate that the miscible melt that was formed directly from the liquidus temperature can be quenched as a homogeneous texture of quenched crystals.
 In order to determine the liquid immiscibility gaps at 3 GPa and 1883–2013 K, we additionally performed quench experiments with starting compositions of Fe65O27S8 at 1883 and 2013 K, and of Fe62O23S15 at 2013 K (Table 2). These compositions were also used in radiography experiments to higher temperature (Table 1). We confirmed that the recovered products of two phases quenched from these conditions had been primary immiscible melts based on the radiography observations (Table 1). From the results of radiography measurements and additional quench experiments, the liquid immiscibility gap of the Fe-O-S system at 3 GPa shrinks with increasing temperature from 1883 to 2013 K (Figure 4), that is consistent with the previous result of Tsuno et al.  at 15 GPa and from 2303 to 2573 K. However, there is a possibility that the miscible regions determined by the electron microprobe analysis of the recovered samples is overestimated. Compared this study with the previous result of Kress  who estimated the liquid immiscibility gap at ambient pressure and 1673 K using the thermodynamic calculation, metallic melt compositions of this study is almost identical to his result (Figure 4). Given that pressure dependence between the ambient pressure and 3 GPa is negligible, the immiscibility gap of our study must be smaller due to the higher temperatures.
 Our results indicate that the two phases of the quench products can be interpreted as either from primary immiscible melts at high temperature conditions or a homogeneous melt which was converted from immiscible two-liquid regions at lower temperature. It is, therefore, necessary to perform in situ observations of the liquid Fe with light elements to determine the precise liquid immiscibility gap. In the case of the Fe-O-S melts, Urakawa et al.  and Tsuno et al.  investigated the liquid immiscibility gaps at 6–15 GPa and 15–21 GPa, respectively, based on textural observation and chemical analyses of the recovered products. However, these immiscibility gaps may be overestimated, since the two phases observed by Urakawa et al.  and Tsuno et al.  could be recovered products separated from the homogenous melt during quenching. On the other hand, homogeneous dendritic textures composed of FeS and FeO dendrites (Figure 3c), and Fe and FeO dendrites [Urakawa et al., 1987; Tsuno et al., 2007] show complete miscibility of the FeS-FeO and Fe-FeO melts at high temperature.
 Closure of a liquid immiscibility gap at 21 GPa and 2573 K shown in Figure 4 [Tsuno et al., 2007] is supported by the recovered products showing the homogeneous textures. Therefore, a homogenous Fe-O-S melt with a significant amount of oxygen can be formed in a terrestrial magma ocean exceeding 21 GPa, and significant amount of oxygen may be supplied to the Earth's outer core. McDonough and Sun  proposed a geochemical model with a large amount of oxygen (5.8 wt.% O) and sulfur (1.9 wt.% S) in the outer core, and both the present experimental results and those by Tsuno et al.  support this model.
 A liquid immiscibility gap at 3 GPa of this study (Figure 4) has an important implication for the composition and formation of the Martian core, since a low-pressure magma ocean scenario has been proposed for Mars [e.g., Righter et al., 1998]. A composition of 3.5–16 wt.% S is estimated in the Martian core based on the geochemical constraints [Morgan and Anders, 1979; Dreibus and Wänke, 1985; Sanloup et al., 1999], whereas O content in the core was not mentioned. If the core-forming melt has approximately 3.5 wt.% S, little amount of oxygen may be included by equilibrium in the magma ocean because of the low solubility of O in the metallic melt even at high temperatures. Therefore, our results are consistent with the results of Rubie et al.  and Asahara et al.  who have investigated the reaction between S-free metallic Fe melts and magnesiowüstite. On the other hand, if the core-forming melt has approximately 10 wt.% S or more, a large amount of oxygen may be included because of existence of a wide miscible Fe-O-S liquid region. The high oxygen solubility is consistent with the results of Gessmann and Wood  and Bouhifd et al. . They have investigated the reaction between Fe-S melts and silicates, and maximum of 8.7 wt.% O was incorporated into the Fe-S melts at pressures of 2.5 and 5 GPa. The actual oxygen content in the Martian core depends on the sulfur content due to the likelihood that the oxygen and sulfur may be incorporated into the Fe-metal simultaneously. In addition, oxygen solubility in the core-forming melt may change the interfacial energy between the metallic melt and silicate minerals at the bottom of the magma ocean, and may have a large effect on the percolation mechanism of the core formation [Terasaki et al., 2005].
 The authors thank K. Litasov, S. Ghosh, S. Shindo, H. Asanuma, A. Saikia, R. Shiraishi, and Y. Tange for technical assistance at SPring-8 and KEK-PF, and Y. Ito and H. Kawanobe for sample preparation and electron microprobe analysis. This work was partly supported by the grants-in-aid for scientific research from the Japanese Ministry of Education, Culture, Sports, Science, and Technology to E.O. (14102009 and 18104009), and by the 21st COE program Advanced Science and Technology Center for the Dynamic Earth at Tohoku University. These experiments were performed under a contract of SPring-8 (2006A1544 and 2006B1313) and KEK-PF (2006G044, 2007S2-002).