• bridgmanite
• lower mantle
• oxidation state
• spin transition
• bulk sound speed

The variable oxidation state of iron has a profound impact on a range of mantle properties, including the redox conditions and iron partitioning (1, 2). The strong partitioning of trivalent Al into bridgmanite in the lower mantle affects the stability of Fe³⁺ in the phase because the substitution requires defects or charge coupling (3). McCammon (4) reported a large amount of Fe³⁺ (Fe³⁺/∑Fe = 60%; fraction of Fe³⁺ with respect to total Fe in a phase) in Al-bearing bridgmanite synthesized in a multianvil press, which was subsequently confirmed under reducing conditions (1, 5). Stabilization of Fe³⁺–Al charge-coupled substitution and charge disproportionation where 3Fe²⁺ (bridgmanite) → 2Fe³⁺ (bridgmanite) + Fe⁰ (metal) both help explain the large amount of Fe³⁺ under reducing conditions. However, the multianvil studies are limited to ∼800-km depth (26 GPa and 2,300 K). High Fe³⁺/∑Fe has been reported in bridgmanite samples synthesized at higher pressures (2, 6⇓⇓–9). However, the existing data are sparse, and bridgmanite has the capacity for large amounts of Fe³⁺ under oxidizing conditions (5, 6, 10). Despite these issues, bridgmanite has been assumed to contain a large amount of Fe³⁺ throughout the lower mantle (1).

The energetics of Fe²⁺ and Fe³⁺ incorporation in bridgmanite, however, can change at higher pressures. Whereas oxygen-defect substitution is energetically competitive for the incorporation of trivalent cations at lower pressures, charge-coupled substitution may be increasingly stabilized at higher pressures (11). Also, the spin configuration of valence electrons in Fe in minerals undergoes changes in the lower mantle (12). Because in bridgmanite the spin behavior of Fe is different depending on the oxidation state and the coordination environment, site preferences of Fe³⁺ and Fe²⁺ change with pressure (13, 14). These factors affect the stable oxidation state and substitution mechanism at different pressures.

We examined the Fe³⁺/∑Fe of bridgmanite at a wide range of pressure–temperature conditions related to the lower mantle under reducing conditions. In our experiments, we mixed Fe²⁺-rich starting materials with approximately bridgmanite stoichiometry with 2–5 wt% Fe metal powder following the method used in multianvil syntheses (1). We enriched all starting materials with 67–95% Fe57. We loaded the sample mixture into diamond-anvil cells (DACs) together with a pressure medium (SI Appendix, Fig. S1). We heated 14 different samples to 2,000–2,300 K at different pressures with the synthesis monitored by using synchrotron X-ray diffraction (XRD) during and after heating (SI Appendix, Fig. S2). Energy-dispersive X-ray spectroscopy (EDS) in an aberration-corrected electron microscope (ACEM) showed that the bridgmanite samples synthesized at 47 and 56 GPa contain 0.19 ± 0.02 of Fe# and 0.10 ± 0.03 of Al# on the basis 3 oxygens, which agrees well with the composition of the starting materials within 10% uncertainties from the EDS measurements. Both XRD and ACEM (EDS and imaging) showed that our samples contain bridgmanite and metallic iron (with a small amount of silica only at P< 40 GPa), but found no other phases (see Materials and Methods and SI Appendix for details).

We conducted synchrotron Mössbauer spectroscopy (SMS) on both the temperature-quenched samples at high pressure (hereafter “in situ” SMS) and the decompressed samples at 0–3 GPa and 300 K (hereafter “quench” SMS) (Fig. 1 and SI Appendix, Fig. S3). We used quench SMS for Fe³⁺/∑Fe, because the quench SMS allows for unambiguous determination of the oxidation state of Fe. The spin effect on isomer shift (IS) and quadrupole splitting (QS) generally overlaps with the oxidation effect at high pressures (SI Appendix, section S2 and Fig. S4). The interpretation of the parameters is still controversial at high pressure (13, 14) (SI Appendix, section S3). In contrast, from quench SMS, we can relate IS and QS to the oxidation states of Fe, because near 0 GPa, both Fe²⁺ and Fe³⁺ should be high spin in bridgmanite, and the effects from Fe in different crystallographic sites on the Mössbauer parameters are much less significant than the oxidation states (15) (SI Appendix, section S3). The IS and QS of bridgmanite are also well known at 1 bar with good agreement among existing studies (2, 4, 5) (SI Appendix, section S3 and Fig. S4). Because bridgmanite is metastable at ambient conditions, iron in the phase could be easily oxidized by air or moisture. Therefore, quench SMS were performed for the samples in an inert medium (Ar or Ne) in the DAC at 0–3 GPa.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Synchrotron Mössbauer spectra of bridgmanite. (A) Bridgmanite recovered to 0–3 GPa after synthesis at different pressures and 2,000–2,300 K (bottom four spectra). We also included synthetic spectra of bridgmanite at 1 bar with different amount of Fe³⁺ using IS and QS values from McCammon (4) (top two spectra). (B) Bridgmanite at in situ high pressure after laser heating. Synthesis pressures for the samples are shown on the right side (circles, measured intensities; curves, fitted spectra).

We found that the quench SMS spectra of bridgmanite synthesized at 47–63 GPa are distinct from the spectra at lower and higher pressures (Fig. 1A and SI Appendix, Fig. S3). The spectra in this pressure range have much more well-defined quantum beats compared with those in other pressure ranges. The synthetic SMS spectrum of Fe²⁺-rich bridgmanite agrees well with the measured SMS spectra of bridgmanite synthesized at the pressure range (Fig. 1A). Our spectral fitting showed that Fe³⁺/∑Fe drops to 13–22% at 47–63 GPa, whereas it is ∼60% at lower and higher pressures (Fig. 2 and SI Appendix, Fig. S4 and Table S1). Our glass-starting material contains 10% of Fe³⁺/∑Fe, similar to the Fe³⁺/∑Fe observed at 51–63 GPa. This similarity means that the system does not produce Fe³⁺ over this pressure range. The other Mössbauer parameters, QS and IS, are in agreement with the previous studies (SI Appendix, Fig. S4B), indicating that the changes found in the spectra originated from Fe³⁺ content.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Ferric iron content of aluminous bridgmanite. The parameters were measured for the samples quenched to 0–3 GPa. The gray scale for the symbols of the previous data points (1, 2, 4⇓⇓⇓–8) represents Al# for 3O basis bridgmanite formula (black for Al# ≥ 0.1). We plot only Al# ≥ 0.05, except for Piet et al. (9). The light red and gray curves are guides for the eye for the datasets from this study and other studies (7⇓–9), respectively.

Among the recovered samples for SMS, one synthesized at 47 GPa (Figs. 1 and 2) was sufficiently thinned for electron energy loss spectroscopy (EELS) (Fig. 3). EELS can measure Fe³⁺/∑Fe through energy changes during electronic transitions, which is a completely different physical process compared with SMS. Therefore, EELS provides an independent determination of Fe³⁺/∑Fe in bridgmanite, complementing the SMS data analysis which can be sensitive to starting models. However, the intense electron beam is known to alter Fe³⁺/∑Fe during EELS measurements, even for minerals stable at ambient conditions (16). Therefore, we optimized EELS parameters until we found no changes in Fe³⁺/∑Fe of the standard mineral specimens: olivine, andradite, and cronstedtite (Fig. 3). All of the EELS spectra measured for the sample showed negligible spectral signatures from Fe³⁺ in both L2 and L3, indicating that iron in the sample is predominantly Fe²⁺ (Fig. 3). Spectral fitting indicated that Fe³⁺/∑Fe should be <10%, which agrees with our quench SMS for the same sample.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

EELS of Fe in bridgmanite. The sample was recovered from 47 GPa and 2,000 K and compared with olivine (Fe³⁺/∑Fe = 0%), andradite (Fe³⁺/∑Fe = 100%), and cronstedtite (Fe³⁺/∑Fe = 50%). All of the spectra were measured under the same instrumental conditions (Materials and Methods). The structure at 708 keV in andradite is a prepeak of Fe³⁺.

Our Fe³⁺/∑Fe at the lowest pressure (35 GPa) agrees well with the multianvil results (1, 5) on aluminous bridgmanite under reducing conditions at 26 GPa (Fig. 2). There are three other datasets from laser-heated DAC (LHDAC) experiments (7⇓–9), which have a wide enough pressure range for comparison with our dataset, although these datasets have fewer data points, with larger gaps between the points (Fig. 2). None of these three datasets show a clear drop in Fe³⁺/∑Fe at 47–63 GPa. However, there are a few important differences in the experiments. The chemical compositions of these three datasets are all different from each other and also from our study (see SI Appendix for details). The three data points in Sinmyo et al. (7) all have different chemical compositions. A data point at 93 GPa in Piet et al. (9) was obtained from a different starting material with much more Al. Prescher et al. (8) used Fe³⁺-rich bridgmanite synthesized at 25 GPa as a starting material, which is different from our study and the other two (7, 9), where starting materials were amorphous. Unlike amorphous starting materials, which are highly metastable at high pressure and therefore affected less by kinetic effects during high-temperature synthesis of bridgmanite, conversion from Fe³⁺-rich to Fe²⁺-rich bridgmanite would have a higher kinetic barrier to overcome. We loaded metallic Fe to ensure the reducing conditions during synthesis. None of the three studies loaded metallic Fe. However, Sinmyo et al. (7) observed metallic iron after synthesis. Therefore, the redox conditions for the dataset from the study might be comparable to our dataset. Despite these differences and fewer data points, it is interesting that all three datasets suggest a possible decrease in Fe³⁺/∑FeFe (by 20–30%) in the midmantle (Fig. 2). The different experimental methods and composition may contribute to the different pressures in the Fe³⁺/∑FeFe decrease and its magnitude observed in the datasets (7⇓–9).

The drop of Fe³⁺/∑Fe found here (Fig. 2) requires changes in substitution mechanisms in bridgmanite with pressure (Fig. 4B). Existing data (1, 5) suggest that the dominant factors for high Fe³⁺/∑Fe in bridgmanite at 25–30 GPa are the Fe3+VIII–Al3+VI charge-coupled substitution [where VIII and VI represents the eight (or A) and six (or B) coordinated sites in a perovskite-type structure, respectively] and charge disproportionation (SI Appendix, Reaction S2). Previous studies also indicate smaller but important contributions from the Al3+VIII–Al3+VI charge-coupled substitution (SI Appendix, Reaction S5) and oxygen vacancy substitution (SI Appendix, Reaction S6) at 25–26 GPa, even when bridgmanite contains equal amounts of Al and Fe (SI Appendix, section S4). With an increase in pressure between 25 and 51 GPa, oxygen vacancies are expected to disappear (11) (SI Appendix, Reaction S10), and the [Al3+]VIII–[Al3+]VI configuration (SI Appendix, Reaction S11) may become energetically more favorable than the [Fe3+]VIII–[Al3+]VI configuration, leading to minimum Fe³⁺/∑Fe at 53–63 GPa. The proposed change may be driven by smaller volume of the [Al3+]VIII–[Al3+]VI configuration. Such a configuration will increase the content of Al in the A site, which decreases the volume of the A site where larger-sized Mg and Fe existed. It will also decrease the content of Al in the B site, which reduces the volume of the B site by increasing in the content of smaller-sized Si. However, conversion of Fe³⁺ to Fe²⁺ would increase the volume of the A site. Our in situ SMS indicates the appearance of a new Fe site with QS as high as 4 mm/s at the pressure where Fe³⁺/∑Fe rapidly decreases (42 GPa; Fig. 1B and SI Appendix, Fig. S5), suggesting either changes in spin state or local structure for Fe²⁺ in the A site (SI Appendix, section S4.5). If such changes can decrease the ionic radius of Fe²⁺, it will further stabilize the [Al3+]VIII–[Al3+]VI configuration and therefore decrease Fe³⁺/∑Fe. At 25–50 GPa, Fe³⁺ would remain in the A site (SI Appendix, section S4.5).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Seismic and mineralogical structures in the lower mantle. (A) Comparison of bulk sound speeds of seismic measurement [PREM (17); circles] and pyrolite models. Three different pyrolitic models are presented: depth-dependent changes in Fe³⁺/∑Fe of bridgmanite (red curve), constant Fe in bridgmanite (0.04Fe2+#) and ferropericlase (0.27Fe#) (solid blue curve), and constant Fe in bridgmanite (0.045Fe2+# and 0.045Fe3+#) and ferropericlase (0.15Fe#) (dashed blue curve). See Materials and Methods for details. The light red area represents estimated uncertainties, including those from thermoelastic parameters. (B) Changes in the crystal chemistry of bridgmanite and ferropericlase. For bridgmanite, the tilted squares and the diamond-shaped areas between them represent the B and A sites, respectively. For ferropericlase, the 45° rotated squares represent the MgO₆ octahedra. (C) Changes in viscosity (26). The black curve is for a density scaling of 0.2 and harmonic degrees of 2–7. The light gray area represents a range of different values found in all of the models in ref. 26. (D) Changes in radial correlations of the P and S wave tomography models (PMEAN and SMEAN) (27).

Ferric iron in the B site undergoes a high- to low-spin transition at 50–70 GPa (10, 14). If this spin transition results in a smaller ionic radius of Fe³⁺, likely comparable with that of Al³⁺, the charge disproportionation would become more stable, and Fe³⁺/∑Fe would increase by stabilization of the [HSFe3+,Al3+]VIII–[LSFe3+,Al3+]VI configuration (Fig. 4A and SI Appendix, section S4.7). We calculated the volume change in charge disproportionation of Fe to such a configuration (SI Appendix, Fig. S8C and section S4.7) and found that the volume of the [HSFe3+,Al3+]VIII–[LSFe3+,Al3+]VI configuration is significantly smaller, supporting the proposed change in the substitution mechanism at pressures >70 GPa.

Our results indicate that almost all of the iron in bridgmanite is ferrous (Fe²⁺), with little ferric (Fe³⁺) iron at 1,100- to 1,700-km depths (hereafter low ferric iron bridgmanite zone; LIBZ), whereas bridgmanite in the lower mantle regions above and below LIBZ contains >50% of iron in the ferric oxidation state (hereafter high ferric iron bridgmanite zone; HIBZ). Because Fe²⁺ partitions into ferropericlase, whereas almost all Fe³⁺ enters bridgmanite (2), bridgmanite will be depleted in Fe in LIBZ (also can be referred to as low-iron bridgmanite zone) compared with HIBZs (also can be referred to as high-iron bridgmanite zone), and ferropericlase will have an increased amount of Fe in LIBZ compared with HIBZs (Fig. 4B).

A number of studies have been conducted to compare the density and velocity profiles of compositional models (pyrolite) with seismic observations. However, these models have assumed that the compositions of individual minerals, in particular Fe content and oxidation state, do not change throughout the lower mantle (18⇓–20). Both experiments and computations have found that the bulk modulus (and therefore bulk sound speed) of ferropericlase decreases considerably within the pressure range where the phase is in a transitional state from high spin to low spin (therefore, a mixed-spin state), although the bulk modulus of ferropericlase with high-spin Fe is essentially the same as that with low spin (18⇓⇓–21). In contrast, the spin transition has no significant impact on density and shear velocity profiles (18⇓⇓–21).

The bulk sound speed profile of pyrolite along the mantle geotherm shows significant discrepancy with seismic values if it is calculated by assuming constant Fe³⁺/∑Fe in bridgmanite (therefore constant mineral compositions) throughout the lower mantle. As shown in Fig. 4A, such discrepancies are clear at 1,300- to 1,800-km and 2,100- to 2,400-km depths, if bridgmanite is rich in Fe³⁺ and Fe²⁺ (and therefore ferropericlase has low and high Fe contents), respectively, throughout the lower mantle (see Materials and Methods for details). Although the bulk sound speed depression should be detectable, such large-scale structures have not been found in the lower mantle (22). Decreasing the content of ferropericlase provides a substantially better fit. However, the required decrease increases the bulk sound speed of mineralogical models above the seismic values throughout the lower mantle and results in a much lower Mg/Si ratio for the lower mantle than the upper mantle. The change in Mg/Si ratio would make the lower mantle nonpyrolitic as a result, making the lower mantle richer in Si.

We include the observed Fe³⁺/∑Fe change in the calculation (the red curve in Fig. 4A). From the iron partition coefficients for Fe²⁺- and Fe³⁺-rich bridgmanite (2, 23), we obtained 0.27 and 0.15 Fe# of ferropericlase (for 1 oxygen base chemical formula) in LIBZ and HIBZs for pyrolite composition in response to the Fe oxidation state change in bridgmanite, respectively (see Materials and Methods for details). The higher Fe content in ferropericlase in LIBZ (0.27 Fe#) will significantly increase the spin transition pressure for the Fe (18, 20). As shown in Fig. 4A (blue curves), because the spin transition does not occur until much greater depths, ferropericlase with higher Fe content in LIBZ will have only high-spin Fe²⁺. Ferropericlase (with lower Fe content) in HIBZs would not decrease bulk sound speed significantly, because the mixed-spin state, which causes the severe bulk modulus decrease for 0.15 Fe#, occurs at the depths of LIBZ. Instead, ferropericlase has almost all Fe²⁺ in high spin in shallow HIBZ at 660- to 1,100-km depths and low spin in the deep HIBZ at 1,700- to 2,900-km depths.

Unlike the models without the bridgmanite Fe³⁺/∑Fe change, we found no significant reduction in the bulk sound speed profile of pyrolite when we incorporate the observed Fe³⁺/∑Fe change in the calculation (Fig. 4A). The bulk sound profile of pyrolite with the bridgmanite Fe³⁺/∑Fe change agrees well with average [preliminary reference Earth model (PREM)] seismic values for the lower mantle (17). We also found that the density profile of pyrolite agrees well with seismic values (SI Appendix, Fig. S10). Fe³⁺/∑Fe in bridgmanite decreases and increases above and below LIBZ over a pressure interval of 10–20 GPa. Such a gradual change will not result in any sharp changes in the lower mantle seismic profiles, consistent with the absence of global seismic discontinuities in the 1,000- to 2,000-km depth interval.

We showed that the reduction of Fe³⁺ in bridgmanite at 1,100- to 1,700-km depths reconciles the discrepancy in bulk sound speed profile between pyrolitic lower-mantle composition and seismic observations. Therefore, there is no need for differences in major element chemistry between the upper and lower mantle, consistent with seismic observations of slabs penetrating to the lowermost mantle and therefore whole mantle convection (24).

An important unknown in the model above is possible compositional sensitivity of the depth interval for the Fe³⁺/∑FeFe drop in bridgmanite. A recent Brillouin spectroscopy study (25) found good agreement in velocities between Fe³⁺-rich bridgmanite in Mg_0.9Fe_0.1Al_0.1Si_0.9O₃ and seismic 1D model (PREM) at pressures expected for the shallower lower mantle. However, the velocities of Fe³⁺-rich bridgmanite showed significant mismatch with the seismic model at depths >1,300 km, while Fe²⁺-rich bridgmanite agrees better with the seismic model in the region. The observations led to a proposal that Fe³⁺/∑Fe in bridgmanite decreases at the depth range. The proposed depth is remarkably similar to the depth expected for the Fe³⁺/∑Fe drop in bridgmanite found in our study (1,100 km), despite the difference in the total Fe content.

Recently, elevated viscosity has been reported (26) near the depth of low-Fe³⁺ bridgmanite (LIBZ) (Fig. 4C). Most P-wave tomographic models (27) have also shown a transition from minimum radial correlation of seismic structures at 700–1,000 km to enhanced correlation at depths greater than ∼1,500 km (Fig. 4D). Recent tomographic models have found stagnation or broadening of some subducting slabs (28) and horizontal deflections of some mantle plumes (29) near this depth range. Numerical simulations demonstrated that the viscosity elevation at 1,000–1,500 km can produce such seismic observations (26).

Being the dominant phase in the region (70 vol%), bridgmanite controls the viscosity of the lower mantle (30). According to our study, bridgmanite in LIBZ at 1,100–1,700 km contains little or no Fe³⁺ and therefore a lower amount of Fe because of strong partitioning of Fe²⁺ to ferropericlase. Although the effect of Fe is still unknown for the viscosity of bridgmanite due to technical difficulties, iron can decrease viscosity as shown in other minerals (31). Because iron reduces the melting temperature of bridgmanite and viscosity can be scaled to homologous temperature (temperature normalized to material’s melting temperature) (32), low-Fe content bridgmanite at 1,100–1,700 km would have a high viscosity, providing a plausible explanation for the viscosity increase (26) without invoking any major changes in chemical composition over this depth range.