The stability of magnesite in the transition zone and the lower mantle as function of oxygen fugacity



[1] The oxygen fugacity at which magnesite (MgCO3) is reduced to diamond in a typical mantle assemblage has been determined between 16 and 45 GPa and 1500–1700°C in experiments employing a multianvil device. This oxygen fugacity for carbonate stability, measured using a sliding redox sensor that employs IrFe alloy, was found to be greater than 2 log units above the iron-wüstite oxygen buffer (ΔIW+2). Reversal experiments employing FeNi alloy confirmed complete oxidation of Ni in the presence of magnesite and diamond even at 45 GPa. As the oxygen fugacity of the transition zone and lower mantle is most likely at or below the IW buffer, mantle carbon, if distributed relatively homogeneously, is unlikely to be hosted in carbonates throughout most of the mantle but is more likely present as diamond, methane, Fe-rich carbide or as a carbon-component dissolved in Fe-Ni metal. The existence of carbonate at these depths would imply the presence of unusually oxidized regions of the deeper mantle. Such regions could form in the deeper mantle from an influx of subduction related carbonate melt, which would reduce by causing oxidation of the surrounding silicates. Due to changes in the degree of oxidation of the surrounding mantle such melts could potentially travel further in the transition zone mantle than in the lower mantle. The results do not exclude the possibility that carbonate could coexist with Fe-Ni metal or carbide at the very base of the lower mantle.

1. Introduction

[2] The speciation, mobility and residence time of carbon in the mantle are strongly coupled to oxygen fugacity (fo2). At low fo2 the immobile carbon species diamond and graphite are stable, while at more oxidizing conditions species such as CO2 fluid and carbonate melt can form, migrate and potentially leave the mantle. Many studies have argued that the fo2 of the deep mantle may be more reduced than that of the upper mantle due to the increase in the ability of major mantle minerals to partition ferric Fe [Luth et al., 1990; Ballhaus et al., 1991; O'Neill et al., 1993; Ballhaus, 1995]. Garnet, wadsleyite, ringwoodite and magnesium silicate perovskite, for example, all contain significant quantities of ferric iron even in equilibrium with metallic iron. It has been proposed that this affinity will cause the mantle to become saturated in metallic Fe-Ni alloy at depths over 300 km [Ballhaus, 1995; Rohrbach et al., 2007; Frost and McCammon, 2008].

[3] Stagno and Frost [2010] compared the fo2 where carbonate minerals and melts reduce to diamond or graphite with a plausible fo2 profile for the upper mantle to ∼300 km in depth. They argued that the effect of increasing pressure on mineral ferric Fe components would exert a dominant control over mantle redox state, driving carbonate minerals and melts into the graphite or diamond stability field at depths >100 km. Such a scenario is likely for typical bulk mantle carbon concentrations [Dasgupta and Hirschmann, 2010], as they are relatively small and when homogeneously distributed will have a minor buffering capacity compared to Fe-bearing equilibria. Rohrbach and Schmidt [2011] performed a similar analysis for the transition zone and the very top of the lower mantle, arguing that carbonates entering peridotitic mantle from subducting lithosphere at these depths would undergo “redox freezing”, i.e. reduction of carbonate to diamond with concurrent oxidation of iron in the mantle. At higher pressures, although carbonate minerals have been shown to be thermally stable [Biellmann et al., 1993; Isshiki et al., 2004; Ono et al., 2007], there are no experimental data through which to assess the fo2 at which carbonate minerals would be reduced to diamond in a peridotitic assemblage. In addition, without information on the ferric Fe content of high-pressure mantle minerals coexisting with carbonate melt it is impossible to determine the extend of mantle oxidation expected from an influx of carbonate melt at depth. Recent studies on natural samples, on the other hand, have revealed the occurrence of carbonate phases as inclusions in diamond that are argued to have formed in the transition zone and lower mantle [Brenker et al., 2007; Walter et al., 2008; Kaminsky et al., 2009]. It is, therefore, crucial to not only determine the fo2 at which this process occurred but to understand the likely oxidation state expected for mantle minerals at these conditions.

[4] In the simplified Mg-Si-O-C system the fo2 at which magnesite (MgCO3) and diamond coexist in the wadsleyite or ringwoodite stability field of the transition zone (410–660 km) can be expressed by the equilibrium,

equation image

[5] In the lower mantle the same fo2 will be described by the equilibrium,

equation image

[6] The aim of this study is to determine the fo2 at which the carbonate mineral magnesite will be reduced to diamond in the transition zone and lower mantle. By employing sintered diamond anvils in the multianvil press we perform experiments at conditions corresponding to those of the deep lower mantle.

2. Experimental Methods

[7] Equilibrium (1) was studied at 16 GPa employing a carbonated harzburgite starting composition (equilibrium (1)) in the system Fe-Mg-Si-O-C assembled from a mineral mix of San Carlos olivine, San Carlos enstatite, synthetic magnesite and pure graphite in the molar ratio 3:2:1:1. In addition, in each experiment a layer of olivine mixed with enstatite and magnesite (2:1:1) was added sandwiched between two layers of the equilibrium (1) mix to facilitate Mössbauer measurements. Equilibrium (1) was studied at 23 GPa employing a Ca-bearing assemblage (equilibrium (1) + Ca) identical to starting composition (equilibrium (1)) but with CaSiO3 (glass) added in the same proportion as MgCO3. 3 wt % iridium metal powder (≤5 μm) was added to all experiments to act as a sliding redox sensor.

[8] Equilibrium (2) was studied at 25 and 45 GPa employing a mineral mixture in the system Fe-Mg-O-C (equilibrium (2)) comprising synthetic MgCO3, pre-synthesized ferropericlase (Fe/(Fe + Mg) = 0.17), pure graphite powder and 3 wt % pure iridium metal. At 45 GPa one run (M140) was also performed with a glass powder added to equilibrium (2) in order to crystallize (Mg,Fe)(Si,Al)O3 perovskite. This glass was prepared from high purity oxides at 1650 °C, and then reduced in H2-CO2-Ar atmosphere at 900°C at a fo2 of the Ni-NiO oxygen buffer. Further experiments at 25 and 45 GPa (H2981 and M131) employed the equilibrium (2) bulk composition but with 3 wt % Ni metal added instead of Ir metal.

[9] Experiments between 16 and 25 GPa were performed in 1000 and 1200 tons Kawai-type multianvil presses at the Bayerisches Geoinstitut. Tungsten carbide anvils of 8 and 4 mm truncation edge length (TEL) were used with 14 and 10 mm edge length Cr2O3-doped MgO octahedra and lanthanum chromite heaters. Starting powders were loaded into a graphite container that was enclosed in a rhenium foil capsule. The temperature was monitored with a W97Re3–W75Re25 (D type) thermocouple inserted within an alumina sleeve, with the junction in contact with the top of the capsule. Pressure calibrations are reported by Keppler and Frost [2005].

[10] Experiments at 45 GPa (M131 and M140) were carried out using the MADONNA D-DIA (1500 tons) press installed at the Geodynamics Research Center (Ehime University, Japan). In these runs the starting powder was placed in a graphite capsule and compressed inside an MgO pressure medium doped with 5 wt % Cr2O3 employing 14 mm edge length sintered diamond anvils. The initial width and thickness of the pyrophyllite gaskets were 1.2 and 1.8 mm. The cell assembly was similar to that described by Tange et al. [2008]. Technical details concerning the calibration are discussed in the auxiliary material. The experiments for this study were performed for approximately 1 hour during which the temperature was manually controlled within ±5°C.

[11] Recovered samples were mounted in epoxy resin, sectioned and polished under ethanol. Chemical analysis of the run products was performed using a Jeol JXA-8200 electron microprobe employing an accelerating voltage of 15 kV and a beam current of 5–20 nA with counting times between 20 and 10 sec on peak and background (30 and 15 for Ir-Fe metal). Standards were natural silicates and metals (Ir, Fe and Ni). The PRZ correction was applied. The carbon content of the IrFe alloy in the recovered samples was also analyzed employing vitrified carbon as a standard with synthetic C-bearing steel as a secondary standard (detection limit for C ∼0.3 wt %) and using with an accelerating voltage of 13 kV and 12 nA current beam [Nakajima et al., 2009].

[12] The ferric/ferrous ratios within recovered layers of wadsleyite and Al-bearing perovskite were determined at room temperature using 57Mössbauer spectroscopy. Ratios in one sample from 45 GPa (M140) were measured by electron-energy loss spectroscopy (EELS) in a transmission electron microscope (Philips CM20FEG) using a thin film extracted using a QUANTA 3D FEG focused ion beam. Details of ferric iron measurements by Mössbauer and EELS techniques are discussed in the supplemental material.

3. Results

[13] Experimental conditions and recovered phase assemblages are reported in Table 1 with chemical analyses in Tables S1a, S1b and S2 in auxiliary materials. In all experiments the graphite capsule transformed to polycrystalline diamond. At 16 GPa products from the study of equilibrium (1) contained wadsleyite, clinoenstatite, magnesite and diamond in addition to Ir-Fe alloy. At 23 GPa ringwoodite and stishovite coexisted with Mg- and Ca-perovskite. At 25 and 45 GPa products were ferropericlase, magnesite, diamond and Ir-Fe alloy (Figure S2). Microprobe analysis of run products found no detectable levels of carbon in the Ir-Fe alloy.

Table 1. Experimental Conditions and Run Products
Run NumberP (GPa)T (°C)Time (hr)Starting CompositionPhasesalogfO2 (ΔIW)b
  • a

    Phases: Wds = wadsleyite, Rwd = ringwoodite, Cen = clinoenstatite, dia = diamond, mst = magnesite, Mg/Ca-Prv = Mg, Ca-bearing perovskite, Sti = stishovite, Fe-per = ferropericlase.

  • b

    Uncertainties are those arising from the compositional variations in run products and thermodynamic data.

  • c

    Runs containing nickel.

S422616150012equilibrium (1)WdsCenmstdia IrFe2.88(2)
S427816155012equilibrium (1)WdsCenmstdia IrFe2.56(2)
H31022316001equilibrium (1) + CaRwdMg/Ca-PrvmstdiaStiIrFe2.98(47)
H294625150012equilibrium (2)Fe-per mstdia IrFe2.63(9)
H298225150016equilibrium (2)Fe-per mstdia IrFe2.30(14)
H288725155012equilibrium (2)Fe-per mstdia IrFe2.32(7)
S48072516001equilibrium (2)Fe-per mstdia IrFe2.46(15)
H2981c25150012equilibrium (2) + NiFe-per mstdia  0.22
M131c4517001equilibrium (2) + NiFe-per mstdia  0.03
M1404517001equilibrium (2) + AlFe-perMg, Al-Prvmstdia IrFe2.16(80)

[14] The TEM image in Figure 1 shows sample M140 from 45 GPa where coarse-grained ferropericlase is in contact with smaller well crystallized perovskite grains formed from an initial homogenous glass of perovskite composition. M140 shows evidence for textural equilibrium and no chemical zonation was found. In addition the determined KDFe/Mg between perovskite and ferropericlase (= {XFePv/XMgPv}/{XFeFp/XMgFp}) of 0.54 is in excellent agreement with values determined by Irifune et al. [2010] at similar conditions. The Fe-Mg partition coefficient between magnesite and ferropericlase remained constant between 25 and 45 GPa, which precludes effects of Fe spin state changes over this pressure range. In addition, the Fe-Mg partition coefficient between magnesite and silicates was observed to decrease with pressure (Figures S3a and S3b) in agreement with data by Rohrbach and Schmidt [2011].

Figure 1.

A bright-field TEM image of sample M140 (45 GPa/1700°C) taken along a [1equation image0] zone axis. Grains of aluminum and iron-bearing magnesium silicate perovskite (Mg-Pv) and ferropericlase (Fe-pc) are observed with small dark grains of iridium-iron alloy along the grain boundaries between the two minerals. In the upper left inset a SAED pattern is shown for a Mg-Pv grain that was used for EELS analysis.

[15] Mössbauer spectra for wadsleyite and aluminous silicate perovskite from 16 and 45 GPa are shown in Figure S4 and revealed Fe3+/∑Fe ratios of 0.04(±2) and 0.79(±10) respectively. Further EELS measurements made on coexisting perovskite and ferropericlase from 45 GPa, yielded Fe3+/∑Fe ratios of 0.68(±3) and 0.02(±5), respectively.

[16] In each experiment the fo2 was buffered by the coexistence of diamond and magnesite through the Mg and Fe-bearing equivalents of equilibrium (1) at 16 and 23 GPa or equilibrium (2) at 25 and 45 GPa. The fo2 in each experiment could be measured using a sliding redox sensor equilibrium based on the Fe content of IrFe alloy measured after the experiment [Stagno and Frost, 2010]. The redox sensor in the wadsleyite and ringwoodite stability field employs the equilibrium,

equation image

where the fo2 is therefore calculated from,

equation image

image and image are the activities of the Fe2SiO4 component in wadsleyite/ringwoodite, Fe in the IrFe alloy and silica respectively. ΔG0 is the standard state Gibbs free energy change of the equilibrium and R is the gas constant. In the wadsleyite and ringwoodite stability field the silica activity is determined as in the work by Stagno and Frost [2010]. In the calculation of image it is assumed that mixing occurs on two equivalent sites and the activity-composition relations are described using symmetric single-site Margules interaction parameters from Frost [2003]. The activity coefficient for Fe in the Fe-Ir alloy (γFemetal) was calculated from,

equation image

where P is pressure in bars [Schwerdtfeger and Zwell, 1968; Swartzendruber, 1984]. At 25 and 45 GPa the fo2 was measured by using the equilibrium,

equation image

with the expression,

equation image

aFeOFp is the activity of FeO in ferropericlase determined using a symmetric Margules interaction parameter [Frost, 2003]. Equilibrium (6) is the iron-wüstite oxygen buffer when pure phases are considered, for which the contraction ΔIW is employed to denote oxygen fugacities reported relative to this equilibrium. Data to calculate ΔG0 are given in Table S3, while fo2 determinations using equilibrium (3) and (6) are reported relative to IW in Table 1.

[17] Uncertainties propagated from those in the thermodynamic data are by far the largest source of error in the fo2 calculation but can be assessed by comparing thermodynamic data from various sources. For equilibrium (3) when two independent sources of end-member thermodynamic data are compared [Fabrichanaya et al., 2004; Frost, 2003] the difference in calculated fo2 is approximately 0.2 log units in the wadsleyite stability field and 0.47 in the ringwoodite field. For equilibrium (6) there is no uncertainty in the end-member thermodynamic data when fo2 is reported as ΔIW. Uncertainties in the activity composition relations for the IrFe alloy are significant, however, being 0.4 log units when propagated to 45 GPa, but this neglects possible pressure effects on the excess volume of mixing. If we make the default assumption that pressure causes FeIr alloy mixing to become ideal at 45 GPa, then the resulting fo2 uncertainty is 0.8 log units, which is the value we employ in Figure 2.

Figure 2.

The log fo2 (normalized to IW) buffered by a diamond and magnesite bearing mantle assemblage is shown as a function of pressure for experiments performed between 1500–1700°C (gray diamonds). The data points the higher fo2 region where magnesite is stable from the lower region where diamond forms. Previous measurements of the carbon-magnesite/carbonate melt buffer determined along a mantle adiabat between 3 and 11 GPa [Stagno and Frost, 2010] are also shown (open diamonds). White circles are fo2 measurements from Rohrbach and Schmidt [2011] using IrFe alloy as a redox sensor. The gray shaded regions indicate the fo2 of MORB mantle and the likely fo2 of the transition zone and lower mantle after Frost and McCammon [2008].

[18] Reversal experiments M131 and H2981 were performed with Ni metal replacing Ir in the starting powders. Although diamond and magnesite were found to exist in the run products, Ni metal was completely oxidized and partitioned strongly into ferropericlase. By considering thermodynamic properties of Ni-bearing ferropericlase and FeNi alloys the highest fo2 where FeNi metal can coexist with ferropericlase is calculated to be approximately IW at 25 and 45 GPa. This proves that magnesite can only be stable in the lower mantle at a fo2 above IW. Fe-Ir experiments approach equilibrium [2] from initially more oxidizing starting materials, while Ni-bearing runs are initially more reduced. Both run types, therefore, constitute loose reversal experiments.

4. Discussion and Conclusions

[19] Figure 2 shows the experimentally determined fo2 at which diamond and magnesite coexist in a peridotite mantle assemblage to 45 GPa. Results of Stagno and Frost [2010] and Rohrbach and Schmidt [2011] are also shown. At oxygen fugacities above the field of data points magnesite will be stable, while diamond is stable at lower oxygen fugacities. Grey bands indicate the range of fo2 reported for MORB and the fo2 of the transition zone and lower mantle, predicted assuming a fixed BSE composition similar to that observed in the upper mantle [Frost and McCammon, 2008]. The average mantle fo2 at depths >300 km is likely to be below the IW buffer [Frost and McCammon, 2008; Rohrbach and Schmidt, 2011]. In the lower mantle this low fo2 arises from the fact that aluminous silicate perovskite has a Fe3+/∑Fe ratio of over 0.5 even when coexisting with metallic Fe [Frost et al., 2004]. As estimates for the bulk Fe3+/∑Fe ratio of the upper mantle are of the order of 0.02 [O'Neill et al., 1993], this means that assuming a whole mantle convection with a total oxygen content of the upper mantle being similar to the lower mantle, metallic FeNi alloy must precipitate in the lower mantle to provide sufficient perovskite Fe3+. Fe3+/∑Fe measurements on uppermost mantle and transition zone minerals also imply that the fo2 of the mantle must be similarly reduced at depths >300 km [O'Neill et al., 1993; Ballhaus, 1995; Rohrbach et al., 2007]. Figure 2 indicates that plausible lower mantle oxygen fugacities are at least 2 log units below the stability field of magnesite between 15 and 45 GPa. The lower mantle and much of the transition zone should, therefore, typically be in the stability field of diamond or C-bearing FeNi alloy or carbide phases.

[20] Figure 2 indicates a small negative gradient of the diamond-magnesite transition with pressure. Although a positive gradient would also be consistent with the experimental uncertainties, we cannot exclude the possibility that magnesite would become the dominant host for carbon towards the base of the lower mantle, as the carbonate stability field is predicted to approach that of the IW buffer. If this occurred carbonate would likely coexist with Fe-rich carbides at the base of the mantle. The phase transformation in MgCO3 magnesite noted by Isshiki et al. [2004] at approximately 115 GPa may further help to stabilize MgCO3 with respect to diamond at the base of the mantle.

[21] While the effect of pressure on mineral ferric/ferrous equilibria will drive the fo2 down in the deep mantle to conditions where MgCO3 should be reduced, the presence or influx of sufficient carbonate could lead to local oxidation of the mantle to levels where carbonate minerals and melts would be stable [Rohrbach and Schmidt, 2011]. The dominant transition zone mineral wadsleyite synthesize at IW has a Fe3+/∑Fe ratio of approximately 0.02, which is similar to the bulk upper mantle [O'Neill et al., 1993]. If the bulk oxygen contents of uppermost mantle and transition zone are similar then the transition zone fo2 should be close to IW. Our results indicate that in equilibrium with carbonate and diamond, wadsleyite has a higher Fe3+/∑Fe ratio of 0.04. In a BSE composition an influx and reduction of approximately 350 ppm of MgCO3, from subducting lithosphere, would then be required to raise the Fe3+/∑Fe and fo2 locally to the level where carbonate minerals or melts are stable. Wadsleyite inclusions in diamonds would be expected to contain this level of Fe3+ if the diamonds formed from pure carbonate melts.

[22] In the lower mantle Al-bearing perovskite and ferropericlase have Fe3+/∑Fe ratios of approximately 0.6 and 0.02 in the presence of metallic Fe. Assuming similar bulk oxygen content for the lower mantle as the upper, this would require the formation of approximately 1 wt % FeNi alloy in the lower mantle to stabilize sufficient perovskite Fe3+ [Frost et al., 2004]. Oxidation of this amount of metal would require an influx of approximately 8000 ppm of MgCO3 [Rohrbach and Schmidt, 2011]. Our results show that to raise the fo2 to levels where carbonate can be stable requires perovskite and ferropericlase Fe3+/∑Fe ratios of approximately 0.68 and 0.02, which requires an additional 1000 ppm of MgCO3. This means the reduction of approximately 30 times more carbonate would be required before carbonate melt could exist in the lower mantle compared to the transition zone. Carbonate released into the transition zone would cause moderate oxidation that could affect a much larger volume of mantle than in the lower mantle, where carbonate reduction would be more efficient and carbonate transport extremely limited.


[23] V.S. gratefully acknowledges financial support under the EU funded Marie Curie, “project 019700” PhD fellowship and the support of the DFG through grant “FR1555/5-1”. Access to FIB was funded by the Leibniz Program “LA830/14-1”. We thank H. Schülze for his technical assistance in polishing accurately diamond capsules of recovered samples and T. Ohuchi for providing assistance during experiments at GRC. Finally, we thank D. Walker and M. Schmidt for providing a constructive review of the manuscript.