Lower Mantle Melting: Experiments and Thermodynamic Modeling in the System MgO‐SiO2

Characterizing and modeling melting relations in the system MgO‐SiO2 at lower mantle pressures rely on the location of the eutectic points for MgO‐MgSiO3 and MgSiO3‐SiO2. While at an uppermost lower mantle pressure there is general consensus on the eutectic composition in the former, large discrepancies exist for MgSiO3‐SiO2 from experiments in the diamond anvil cell, ab‐initio simulations and models built on them. In order to address this discrepancy, we have performed multi‐anvil press experiments at 24 GPa for Mg4Si6O16 and Mg3Si7O17 at temperatures of 2650±100 K and 2750±100 K. In the experiments at 2750±100 K, we observe the presence of partial melt. The recovered Mg4Si6O16 sample shows SiO2 stishovite as the liquidus phase, and electron microprobe analysis of the quenched melt determines XSiO2=0.53±0.03 as the eutectic composition. We fit a thermodynamic model to describe the melting relations in the MgO‐SiO2 system, and extrapolate to core‐mantle boundary pressure. At 136 GPa, we predict that the eutectic points have moved further away from enstatite composition, and solidus temperatures remain similar for MgO‐MgSiO3 and MgSiO3‐SiO2.


Introduction
Silicate melting has shaped the chemical and physical evolution of Earth's interior, from magma ocean formation during accretion (Elkins-Tanton et al., 2003;Elardo et al., 2011;Monteux et al., 2016;Wood & Halliday, 2005) to the continuous extraction of melt through plate tectonic processes. The latter is dominated by the ongoing formation of mid-ocean ridge basalt (MORB) from mantle peridotite (Kelemen et al., 1995). Melts may also form at the boundaries of the mantle transition zone (e.g., Revenaugh & Sipkin, 1994;Schmandt et al., 2014) and near the core-mantle boundary (CMB), where they serve as an explanation (e.g., Andrault et al., 2014;Labrosse et al., 2007) for the occurrence of seismic ultra-low velocity zones (ULVZ; Hernlund & Jellinek, 2010;Rost et al., 2005). Most deep melting processes are mitigated by the influence of volatile components (Dasgupta, 2018;Johnson et al., 2008), but despite volatile loss of slab material transported into Earth's deep interior (e.g., Chemia et al., 2015), dry melting may occur within the thermal boundary layer at the base of the mantle (Hirose et al., 1999). experiment constrains the eutectic composition in the MgSiO 3 -SiO 2 system. This is a significantly lower eutectic SiO

Model Formulation and Parameters
We extend the previously established thermodynamic model for melting phase relations in MgO-MgSiO 3 from Liebske and Frost (2012) to the MgO-SiO 2 system by using the location of both eutectic points as constraints. For a closed formulation with a minimum number of parameters, we represent all liquids through the oxide components MgO and SiO 2 , and use MgO periclase, MgSiO 3 bridgmanite and SiO 2 stishovite as solid phases. At both eutectics we equate the chemical potentials ( E ), are based on the Birch-Murnaghan Mie-Grüneisen model (cf. Supporting Information S1 for an extended presentation; e.g., Chust et al., 2017) for   Ito and Katsura (1992), DAC experiments by Zerr and Boehler (1993), the thermodynamic assessment of Liebske and Frost (2012) (2820 K), and molecular dynamics results by Di Paola and Brodholt (2016)

MgO-SiO 2 Melting Relations at Lower Mantle Pressure
As mentioned above, the melting curves for MgO and SiO 2 provide important anchor points for the phase relations (Equation 8) at higher E P , and that of MgSiO 3 serves as a test for the validity and consistency of the liquid model.
For MgO, our melting curve matches DAC data (Du & Lee, 2014;Fu et al., 2018;Kimura et al., 2017) for P throughout the lower mantle within their margins of error (Figure 3a), reaching  7510 CMB pressure, similar to de Koker et al. (2013) and ab-initio simulations by Soubiran and Militzer (2020) using thermodynamic integration. Previous assessments by de Koker and Stixrude (2009) and direct two-phase ab-initio simulations by Alfè (2005) show higher m E T , but still within experimental uncertainty.
With the changes in the thermodynamic parameters for SiO 2 liquid and stishovite, m E T of SiO 2 at  24 E P GPa is larger by  300 E K compared to de Koker et al. (2013), a difference that increases to  550 E K at CMB pressure (Figures 2 and 3b). However, our computed melting curve closely follows the experimental data on m E T by Andrault et al. (2020) up to 80 GPa; for  30 E P GPa our melting curve is within the margin of error of that computed by Usui and Tsuchiya (2010), and they approach the same value at CMB pressures. As has been pointed out by de Koker et al. (2013), the potential presence of CaCl 2 -structured SiO 2 (Tsuchida & Yagi, 1989) is expected to have a small effect on m E T . However, the reconstructive phase transition of SiO 2 to seifertite in the lowermost mantle (Das et al., 2020;Dubrovinsky et al., 2001) will expand the stability range of solid SiO 2 to higher T, an effect that can inherently not be captured by our model, and may account for the divergence of our model with the data by Andrault et al. (2020) for 100 E P GPa.
For MgSiO 3 , the melting curve predicted by our model shows a steep initial increase with E P (Figure 4), closely following the trend of the DAC experiments (Shen & Lazor, 1995;Zerr & Boehler, 1993) to 40 E GPa, and, with larger scatter in the data, to 62 E GPa, the highest P achieved in Zerr and Boehler (1993), and that of the shock wave experiments by Fei et al. (2021) (Ito & Katsura, 1992; IK92, blue triangles up and down) bracket melting, and melting temperatures in diamond anvil cell experiments are reported by Zerr and Boehler (1993) (ZB93, crosses) and Shen and Lazor (1995) (SL95, open circles), and a shock melting point by Luo et al. (2004)

Melting Near the Core-Mantle Boundary
This similarity in e E T , and therefore the solidus E T , for MgO-MgSiO 3 and MgSiO 3 -SiO 2 predicted here suggests that melting of basalt at lower E T than peridodite (Andrault et al., 2014;Kuwahara et al., 2018;Pradhan et al., 2015;Tateno et al., 2014Tateno et al., , 2018, invoked as an explanation for the occurrence of the ULVZ in the lowermost mantle (e.g., Thorne et al., 2019), is not simply tied to its higher SiO 2 E X value. Rather, further differences in chemical composition will influence m E T and melt relations either directly within a solid solution (e.g., a higher FeO or Al 2 O 3 content in bridgmanite) or through changes in the resulting phase assemblage. Compared to mantle peridotite (Workman & Hart, 2005), oceanic crust is strongly enriched in Al 2 O 3 , CaO, and Na 2 O, while the FeO content is similar in both lithologies (e.g., Chemia et al., 2015). For oceanic crust this leads to a phase assemblage that differs in various aspects from peridotite at LM pressure: (a) it contains significantly more CaSiO 3 perovskite (e.g., Chust et al., 2017) and (b) SiO 2 occurs as a free phase rather than the ferropericlase solid solution. Both CaSiO 3 perovskite and SiO 2 stishovite dissolve a few % of Al 2 O 3 and very little FeO (e.g., Kuwahara et al., 2018;Tateno et al., 2018), while ferropericlase in a peridotitic composition contains up to 20% FeO (e.g., Chust et al., 2017;Kuwahara et al., 2018). Further, (c) for basalt an Al 2 O 3 dominated mineral occurs in the LM, either as Ca-ferrite (Irifune & Ringwood, 1993;Tateno et al., 2018) or the NAL phase (Imada et al., 2012;Kato et al., 2013). As a consequence of these phase relations, bridgmanite in a basaltic composition can be expected to be significantly enriched in FeO compared to a mantle peridotite. A high FeO content in partial melts from basalt compositions at LM pressure (Kuwahara et al., 2018;Pradhan et al., 2015;Tateno et al., 2018) suggests that the solidus E T in the resulting bridgmanite is significantly lower than for the eutectic in the MgSiO 3 -SiO 2 system established here. This is further supported by the observation that while CaSiO 3 perovskite is the phase on the solidus in peridotite at LM pressure, it becomes the liquidus phase for MORB compositions, and bridgmanite simultaneously becomes the solidus phase (Kuwahara et al., 2018;Pradhan et al., 2015;Tateno et al., 2014Tateno et al., , 2018. However, the chemical complexity of natural or model systems investigated in these studies makes it challenging to disentangle the causes for these discrepancies. In order to analyze the influence of other chemical components in a systematic way by building up a thermodynamic model for mantle compositions, controlled melting experiments in the MA press like the ones performed here are required, ideally for MgO/SiO 2 ratios of the eutectic compositions. Not only dictated by chemical abundances, but also the crystal-chemical and phase relation arguments outlined in the previous paragraph, such a model should be built from the two-component system explored here to the FeO-MgO-SiO 2 and subsequently the FeO-MgO-Al 2 O 3 -SiO 2 systems. These will provide successive insights into the melting relations, for example, from the influence of FeO (and Al 2 O 3 ) in the bridgmanite solid solution and from an additional Al 2 O 3 -rich phase in the assemblage.
With bridgmanite the liquidus phase in a peridotite composition (Tateno et al., 2014), the shallow slope of its melting curve through much of the LM opens the possibility that crystallization of the mantle may have indeed initiated from the middle of the mantle, as proposed in a number of studies (Andrault, 2019;Stixrude et al., 2009). Most prominently, this would lead to the formation of a basal magma ocean (Herzberg et al., 2013;Labrosse et al., 2007), and a separation of two magma ocean reservoirs would strongly be aided if the crystallizing layer is neutrally buoyant. While bridgmanite is denser than the coexisting liquid with  SiO 2 0.5 E X , the strong partitioning of FeO into the melt (Nomura et al., 2011;Tateno et al., 2014) may lead to a situation in which the density contrast between magma ocean liquid and crystallizing bridgmanite is minimal.
With the eutectic in the MgSiO 3 -SiO 2 system exceeding  SiO 2 0.60 E X at  70 E P GPa, magma oceans in rocky exoplanets would likely crystallize bridgmanite rather than stishovite for most plausible scenarios, and therefore not create a very dense SiO 2 -dominated layer deep in their mantle that would impede the onset of convection and therefore efficient heat loss from the planet (e.g., Tosi et al., 2013).

Conclusions
In multi-anvil experiments at 24 E GPa we have measured the compositions of partial melts formed from a Mg 4 Si 6 O 16 starting composition at a temperature of  2750 100 E K. Electron microscope and microprobe analysis of the recovered experimental charges reveals that stishovite is the crystalline phase at the liquidus, and the coexisting liquid has a composition of    (2012) for MgO-MgSiO 3 and thermodynamic models for the liquid components SiO 2 and MgO as well as the solid phases SiO 2 stishovite, MgO periclase and MgSiO 3 bridgmanite, we have modeled the melting phase relations in the MgO-SiO 2 system using a symmetric solution model for the liquid components.
The extrapolation of our thermodynamic description for melting relations to higher pressures hinges on its temperature dependence and the melting curves of MgO and SiO 2 . Our model describes the melting curve of Mg-SiO 3 bridgmanite consistent with diamond anvil cell experiments up to 60 GPa, and recent shock wave experiment to 80 GPa. In our model, the eutectic compositions move to   0.66 0.02 E X at core-mantle boundary pressure, respectively. Eutectic temperatures of these two compositions remain similar, suggesting that lower solidus temperatures of basalt over peridotite in the deep mantle, which has been observed in a series of prior experiments on natural or model compositions, must be related to chemical components other than SiO 2 and MgO.

Data Availability Statement
In compliance with AGU's data availability requirements, the microprobe analysis of the melt in the experiment S7238 (Table 1)   GPa and at  136 E P GPa by blue symbols (dK13). For MgO-MgSiO 3 , eutectic temperatures of the model by Liebske and Frost (2012) are shown in black (LF12). For reference, we include the MgSiO 3 melting curve from our model (golden dashed curves) in both panels.