Electronic structure of iron in magnesium silicate glasses at high pressure



[1] A recent study attributed the source of an iron partitioning change between silicate melt and minerals at deep mantle conditions to a high-spin to low-spin change in iron, which was found in a Fe-diluted Mg-silicate glass at a similar pressure. We conducted X-ray emission spectroscopy and nuclear forward scattering on iron-rich Mg-silicate glasses at high pressure and 300 K in the diamond-anvil cell: Al-free glass up to 135 GPa and Al-bearing glass up to 93 GPa. In both glasses, the spin moment decreases gradually from 1 bar and does not reach a complete low-spin state even at the peak pressures of this study. The gradual change may be due to the existence of diverse coordination environments for iron in the glasses and continuous structural adjustment of the disordered system with pressure. If the result can be extrapolated to iron in mantle melts, the small, gradual changes in the spin state of iron may not be the dominant factor explaining the reported sudden change in the partitioning behavior of iron between silicate melt and minerals at lower-mantle pressures.

1. Introduction

[2] Recent studies have shown that iron undergoes high spin (HS) to low spin (LS) changes in crystalline lower-mantle phases at high pressures [Badro et al., 2003, 2004; Li et al., 2004; Bengtson et al., 2009; Catalli et al., 2010, 2011; Hsu et al., 2010, 2011]. Since iron is expected to partition more strongly into melts than magnesium, relative to coexisting solids [Nomura et al., 2011; Andrault et al., 2012], the electronic structure of iron in these melts is an important quantity to measure.

[3] Nomura et al. [2011]reported a change in iron partitioning between silicate perovskite and melt at 70 GPa in an Al-free system. They found that iron undergoes a spin transition at a similar pressure in Mg-silicate glass. They attributed the change in iron partitioning to a possible spin transition of iron in the melt. However, their melt contains much more iron (15–20%) than their glass (5%). The reported sudden change in spin state in the glass is surprising, as glass structures evolve gradually with pressure [Williams and Jeanloz, 1988; Shim and Catalli, 2009] and the spin transition is sensitive to the coordination environments around iron [Umemoto et al., 2008]. Furthermore, Sturhahn et al. [2005] showed that the spin transition should occur over a wide pressure interval at the high temperatures of the mantle even in crystalline solids. A more recent study by Andrault et al. [2012]found no sharp change in iron partitioning between silicate melt and minerals in an Al-bearing system up to 120 GPa.

[4] We performed both X-ray emission spectroscopy (XES) and nuclear forward scattering (NFS) on iron-bearing Mg-silicate glasses at high pressure. The iron content of our glasses is similar to that expected for mantle melts. The data allow us to understand the spin state changes of iron at high pressure in structurally disordered systems.

2. Experimental Methods

[5] The glass starting materials, 0.2FeSiO3 · 0.8MgSiO3 (Al free) and 0.2FeSiO3 · 0.05Al2O3 · 0.75MgSiO3 (Al bearing), were synthesized using the laser levitation method under a reducing atmosphere controlled by a 1:1 gas mixture of CO and CO2. The iron was 95% enriched 57Fe. The glasses were loaded in a symmetric-type diamond-anvil cell without a pressure medium. Diamond anvils with 300 or 200μm flat culets and with 150 μm beveled culets covered pressure ranges of 0–85 GPa and 80–135 GPa, respectively. Rhenium and beryllium gaskets were used for NFS and XES measurements, respectively. Pressure was measured from the ruby fluorescence pressure scale [Mao et al., 1986] at pressures lower than 85 GPa, and from the first-order Raman mode of diamond from the anvil tip [Akahama and Kawamura, 2004] at pressures greater than 85 GPa.

[6] XES was performed at the HPCAT sector of the Advanced Photon Source (APS) (Figure 1). An 11.35 keV X-ray beam was focused on an area of 30 × 40μm2on the sample. The emission spectra were collected for typically 5–7 hours for the Al-bearing glass and 2–4 hours for the Al-free glass. Following the procedure described inVankó et al. [2006], we normalized all the spectra with respect to the areas under the spectra and then shifted the spectra such that their centers of mass are the same as that of the reference spectrum. We calculated the integrals of the absolute values of the difference spectra (IAD) following the method in Vankó et al. [2006] (Figure 2):

display math

where SP denotes the spectrum at high pressure, and SPref denotes the reference spectrum. Vankó et al. [2006]showed a linear relationship between the IADs and the average spin numbers. Conversion of the IAD value into spin moment requires a reference spectrum with known spin state. The XES of LS iron is not known for Mg-silicate glass, so we use the 1-bar spectra for both compositions as references. Since it is unclear if iron is purely HS in the 1-bar spectra from our NFS results (see below), we applied the IAD method to measure relative spectral changes with pressure instead of converting it to spin moment at high pressure.

Figure 1.

X-ray emission spectra and spectral differences with respect to 1-bar spectra of the iron-rich (20%) (a) Al-free and (b) Al-bearing Mg-silicate glasses at high pressure. We digitally scanned the XES data reported byNomura et al. [2011]for an iron-dilute (5%) Al-free Mg-silicate glass and presented three of them at important pressures in the inset of Figure 1a. For comparison between our data and the data byNomura et al. [2011], we also include the spectra of iron metal (dotted curves) which has very weak satellite peak intensity comparable to that of the iron-bearing glass reported at pressures above 77 GPa inNomura et al. [2011] (orange curve in the inset).

Figure 2.

The integrals of the absolute values of the difference spectra (IAD) obtained relative to the 1-bar spectra as a function of pressure. Curves are guides for the eye.

[7] NFS was performed at sector 3 of APS. A 14.4-keV X-ray beam was focused on an area of 6 × 6μm2in the sample. The storage ring was operated in top-up mode with 24 bunches separated by 153 ns. Nuclear resonant scattering was measured in a time window of 15–130 ns with a typical data collection time of 2–3 hours (Figure 3). For the Al-bearing starting material at 1 bar, we conducted conventional Mössbauer spectroscopy at Argonne National Laboratory (Figure 3). The NFS data were fitted using the CONUSS package [Sturhahn, 2000], in order to obtain hyperfine parameters (Figure 4). Attempts were made to fit the spectra with one-site, two-site, and three-site models. The spectra of both starting glasses at 1 bar and the Al-free glass at 5 GPa require two-site models. Fitting for all other high-pressure spectra required a three site model. The quality of the spectral fitting achieved is comparable to that reported for crystalline silicates with similar compositions in the diamond-anvil cell [e.g.,Grocholski et al., 2009; Catalli et al., 2010, 2011]. The spectral fitting results are presented in the auxiliary material.

Figure 3.

The measured (circles) and fitted (solid curves) nuclear forward scattering spectra of the iron-rich (20%) (a) Al-free and (b) Al-bearing Mg-silicate glasses. The inset in Figure 3b shows the 1-bar spectrum of the Al-bearing glass obtained from conventional Mössbauer spectroscopy and the calculated 1-bar NFS spectrum is shown.

Figure 4.

Hyperfine parameters of the iron-rich (20%) Al-free (open circles) and Al-bearing (closed circles) Mg-silicate glasses obtained from spectral fitting of NFS: (a) site fractions, (b) quadrupole splitting, and (c) full width at half maxima of quadrupole splitting. The blue and red symbols represent high-QSand low-QS iron, respectively. The curves are guides for the eye.

3. Results and Discussion

[8] The intensity of the satellite peak near 7045 keV in the emission spectrum has been related to the spin moment and the disappearance of the satellite intensity has been interpreted as a transition to LS iron. Vankó et al. [2006] pointed out that the position of the main peak, or the center of mass of the spectrum, may shift to a lower energy as spin moment decreases. In 0.05FeSiO3 · 0.95MgSiO3 glass, Nomura et al. [2011] reported very little change in the spectra up to 59 GPa followed by a sudden drop of the satellite peak intensity between 59 and 77 GPa (inset in Figure 1).

[9] As shown in Figure 1, our XES data on iron-rich glasses (20% Fe) differ with those on iron-dilute, aluminum-free glass (5% Fe) reported byNomura et al. [2011]in that: (1) the spectra change very little up to the peak pressures of the measurements (135 GPa for Al free and 85 GPa for Al bearing), (2) the satellite peak intensity decreases gradually throughout the pressure range, and (3) the satellite peak intensity never decreases to the level reported in 5% Fe-bearing Mg-silicate glass and the satellite peak still remains well separated from the main peak even at the peak pressures of our study.

[10] The IAD values represent the degree of the spectral change with respect to a reference spectrum, in our case the 1-bar spectrum.Figure 2supports the qualitative analysis of raw spectra discussed above. In addition, the IAD plot reveals similarity in the behaviors of the Al-free and Al-bearing glasses. They both show steep changes in the lower pressure ranges (P ≤ 30 GPa) followed by less steep changes at higher pressures.

[11] The NFS spectra between the Al-free and Al-bearing glasses are similar within the measured delay time range (Figure 3). They both undergo spectral changes from two broad beats to three or more beats at 5–32 GPa. No major spectral changes were found above 30 GPa except for the gradual migration of the beats to shorter delay times.

[12] While XES provides averaged electron spin moment of iron atoms which have diverse coordination environments and even possibly different valence states in the glasses, NFS distinguishes between iron atoms with different coordination and valence. Because glasses are disordered, and therefore the coordination environment should be diverse, the number of sites found in the spectral fitting should be regarded as a distinctive number of groups of sites rather than well defined sites.

[13] Our spectral fitting shows that both glasses have two groups of sites at 1 bar. In the Al-free glass, 90% of the iron has highQS (2.29 ± 0.16 mm/s) and 10% of the iron has low QS(1.46 ± 0.02 mm/s) at 1 bar. The Al-bearing glass has 64% of iron with highQS (2.16 ± 0.02 mm/s) and 36% with low QS(0.74 ± 0.02 mm/s). In both systems, the high-QS iron splits into two distinct groups of high QS at 5–30 GPa (Figure 4). The fractions of the low-QS iron increases steeply below 30 GPa followed by a gentle increase at higher pressures (Figure 4a).

[14] The QSvalues of all the site groups show similar increasing trends except for the low-QSiron in the Al-free glass: theQS appears to be almost insensitive to pressure (Figure 4b). The widths of the low-QSiron also increase steeply at 0–30 GPa with gentle or even no increase at higher pressure. The widths of the high-QS iron show sharp decreases at 0–30 GPa and become nearly pressure insensitive above this pressure range. NFS constrains differences in center shifts (ΔCS) among different sites without a reference spectrum. We do not find any sharp changes in ΔCS up to the peak pressure (see auxiliary material).

[15] The trends found in the hyperfine parameters obtained from NFS (Figure 4) are in general agreement with the trend found in XES (Figure 2). If iron in our glasses would have experienced a sharp spin transition as reported in Fe-diluted (5%), Al-free glass at 59–77 GPa [Nomura et al., 2011], it should have shown sharp changes in the hyperfine parameters presented in Figure 4. The trends we found in both XES and NSF suggest that there are no sharp changes in the electronic state of iron in Al-free and Al-bearing Fe-rich (20%) Mg-silicate glasses up to 135 GPa and 93 GPa, respectively. The pressure gradients in our glass samples may contribute to the gradual trend, as no pressure medium was used in our experiments. However, no pressure medium was used inNomura et al.'s [2011] XES measurements and therefore the different trends found in our study cannot be explained by sample setup.

[16] The hyperfine parameters can provide some insights on the valence and spin states of iron together with their fractions (Figure 5). Previous experimental and theoretical studies have shown that HS Fe2+ and HS Fe3+ have high and low QS and CS values, respectively. Complications arise when iron undergoes a spin transition: the QS and CS parameters of LS Fe2+ and LS Fe3+ overlap with those of HS Fe3+ and HS Fe2+, respectively.

Figure 5.

The center shift (CS, relative to stainless steel standard) and quadrupole splitting (QS) values measured for the iron-rich (20%) Al-bearing glass at 1 bar (small circles), 6 GPa (intermediate circles), and 58 GPa (large circles). For comparison, we plot theCS–QSvalues for high-spin Fe2+(filled blue polygon), low-spin Fe2+(filled red polygon), high-spin Fe3+(open blue polygon), and low-spin Fe3+(open red polygon) with high coordination numbers (6–12) from experiments on oxides, Mg-silicate perovskite, post-perovskite, and olivine [Bancroft et al., 1967; Rouquette et al., 2008; Grocholski et al., 2009; Catalli et al., 2010, 2011; Lin et al., 2012]. The CS–QSvalues for intermediate-spin iron in silicates are not well known in experiment. Because theoretical studies have reported only theQSof the intermediate spin for iron in Mg-silicate perovskite, we plot them at the right side of the figure (intermediate-spin Fe2+ (filled green polygons) and Fe3+ (open green polygons)): (a) Bengtson et al. [2009] and (b) Hsu et al. [2010, 2011].

[17] At low pressure, we found that the majority of iron has high QSand low-QSiron represents a smaller fraction. The low-QS iron can be interpreted as HS Fe3+. However, because of the overlap in hyperfine parameters and lack of complete HS standard for iron bearing glasses for XES, we cannot rule out the possibility that some low-QS iron is from LS Fe2+ at 1 bar.

[18] The fraction of low-QSiron gradually increases with pressure, while the fraction of high-QS iron decreases. This can mean that HS Fe2+ changes to LS Fe2+with pressure, increasing the fraction of low-QSiron with pressure. It is interesting that the width of the distribution of low-QS iron drastically increases at the same pressure range, P ≤ 30 GPa (Figure 4). It is possible that the low-QS iron in the starting glasses may be HS Fe3+. Although pre-existing HS Fe3+in the glasses and pressure-generated LS Fe2+ may have similar QS and CS, they are likely slightly different and therefore the width of the QS distribution may increase with the HS → LS change in Fe2+ (Figure 4c).

[19] We measured the center shift of iron in the Al-bearing glass with respect to a stainless steel reference (CS) at 1 bar, 6 GPa, and 58 GPa (Figure 5). Most points lie within the known CS–QS pairs of HS and LS of Fe2+and Fe3+, supporting our interpretation above. However, the QS–CS at 58 GPa is located between HS Fe2+–LS Fe3+ and HS Fe3+–LS Fe2+ fields in Figure 5. First-principles computational studies have suggested that intermediate-spin (IS) Fe2+ has a QS between 0.5 and 1.5 mm/s [Bengtson et al., 2009; Hsu et al., 2010]. Interestingly, the intermediate QS–CS values found at 58 GPa overlap with those reported for IS Fe2+ by Hsu et al. [2010]. Therefore, we do not rule out the possibility of a gradual HS to IS change, instead of HS to LS change.

[20] We also measured CSat 5 GPa for the Al-free glass. The data point (CS = 0.59 mm/s and QS = 1.42 mm/s) overlaps with the 58 GPa data point for the Al-bearing glass (Figure 5). Unlike the low-QSiron in the Al-bearing glass, the low-QSiron in the Al-free glass shows no change inQS with pressure (Figure 4b), although other trends are very similar to those of the Al-bearing glass. This may indicate that the low-QSiron in the Al-free starting glass may not be HS Fe3+ and instead it is IS Fe2+. This low-QSsite represents only 10% of iron in the Al-free starting material. The Al-free starting material was synthesized under even more reducing conditions than the Al-bearing starting glass.

[21] The intervalence charge-transfer, Fe2.5+, has CS = 0.6–0.9 mm/s and a QS = 0.3–2.0 mm/sec [Dyar et al., 2006], which is similar to the intermediate CS–QS values found in this study. Electron hopping is considered to be a thermally activated process [Hawthorne, 1988] and Fe3+content is low (at most 10% for Al free and 40% for Al bearing), while the low-QSiron represents 50–60% at high pressure. Furthermore, intervalence charge-transfer cannot explain the spin moment decrease of iron in both of the glasses found in our XES data. All the possible assignments for the observed hyperfine parameters discussed above indicate that our NFS results do not support a sharp, complete HS to LS transition, which is also consistent with our XES results.

4. Conclusion

[22] Our XES data show that the satellite peak intensity decreases gradually with pressure from 1 bar, but never disappears up to 135 GPa in Al-free and 85 GPa in Al-bearing iron-rich (20%) Mg-silicate glasses. Our NFS data indicate that at least three groups of different electronic configurations of iron exist in the glasses. We found at least 35–50% of iron has hyperfine parameters consistent with HS Fe2+in Al-free glass at 70 GPa and in Al-bearing glass at 93 GPa, indicating that no complete HS to LS change exists up to these pressures, consistent with our XES data. The measured hyperfine parameters suggest that some Fe2+ undergoes a gradual spin moment decrease to either LS or IS with pressure. If our results on glass can provide some insights for iron in mantle melts at high pressure, the electronic change in iron should be gradual and smeared out at the high temperatures of mantle melts [Sturhahn et al., 2005], and therefore unlikely to induce a sharp change in iron partitioning in the deep mantle.


[23] We thank P. Asimow and D. Andrault for helpful comments that improved this paper. Use of Sector 3 was partially supported by COMPRES. Portions of this work were performed under the auspices of the DOE by the University of California, Lawrence Livermore National Laboratory (W-7405-Eng-48). Use of HPCAT was supported by DOE-BES, DOE-NNSA, NSF, DOD-TACOM, and the W.M. Keck Foundation. Use of the APS was supported by the DOE (DE- AC02-06CH11357). This work is supported by NSF to S.H.S. (EAR-0968685).

[24] The editor thanks Paul Asimow and Denis Andrault.