Physics and Chemistry of Minerals

, Volume 40, Issue 2, pp 107–113

Iron partitioning in pyrolitic lower mantle


    • Department of Earth and Planetary SciencesTokyo Institute of Technology
    • Bayerisches GeoinstitutUniversität Bayreuth
  • Kei Hirose
    • Department of Earth and Planetary SciencesTokyo Institute of Technology
Original Paper

DOI: 10.1007/s00269-012-0551-7

Cite this article as:
Sinmyo, R. & Hirose, K. Phys Chem Minerals (2013) 40: 107. doi:10.1007/s00269-012-0551-7


The partitioning of iron between Mg-rich perovskite (Pv) and ferropericlase (Fp) was investigated for a pyrolitic bulk composition over a wide range of simulated lower-mantle pressures and temperatures from 28 to 114 GPa and from 1,900 to 2,300 K, in a laser-heated diamond anvil cell (DAC). The recovered DAC samples are chemically homogeneous, indicating a relatively small temperature gradient during laser heating. The chemical compositions of coexisting Pv, Fp, and Ca-rich perovskite (CaPv) were determined by energy-dispersive X-ray spectroscopy (EDS) using an EDS instrument attached to a transmission electron microscope. Our results demonstrate that at pressures above 90 GPa, Pv becomes more Fe-rich with increasing pressure, which is likely due to the effects of high-spin to low-spin crossover of Fe3+ in Pv. We highlight that such a change in Fe–Mg partitioning between Pv and Fp should have a strong influence on the physical properties of the deep lower mantle.


Lower mantleIron partitioningPerovskiteFerropericlaseDiamond anvil cellTransmission electron microscope


The partitioning of iron between Pv and Fp, which are the major mineralogical constituents of the lower mantle, is thought to have a strong influence on a number of its various physical properties, including elasticity, electrical/thermal conductivity, viscosity, and density (Mao et al. 1997; Yamazaki and Karato 2002; Ohta et al. 2008; Ricolleau et al. 2009; Manthilake et al. 2011). Likewise, the pressure range at which iron spin crossover takes place within mantle minerals is dependent on iron content (e.g., Fei et al. 2007), and therefore this switch from high-spin to low-spin crossover of iron in the lower mantle is also controlled by iron partitioning between Pv and Fp. The iron contents in both Pv and the post-perovskite phase are also known to change with both absolute pressure and thickness of the post-perovskite transition (Catalli et al. 2009; Sinmyo et al. 2011). In addition, iron partitioning between Pv and Fp should affect certain rheological parameters such as grain size and dihedral angles within Pv + Fp aggregate, which in turn control the rheological behavior of the lower mantle (Yamazaki et al. 2009).

The Fe–Mg partition coefficient between (Mg, Fe)SiO3 Pv and (Mg, Fe)O Fp, in Al-free and thus Fe2+-dominant (or Fe3+-minimal) systems, has been examined in recent experimental studies (Kobayashi et al. 2005; Auzende et al. 2008; Sinmyo et al. 2008). However, in natural Al-bearing mantle systems (such as pyrolite), Pv is known to contain substantial amounts of Fe3+ in addition to Fe2+ (McCammon 1997; Frost and Langenhorst 2002; Frost et al. 2004; Irifune et al. 2010; Nakajima et al. 2012), which significantly alters the Fe–Mg partition coefficient between Pv and Fp by a factor of two or more. Recent multi-anvil experiments by Irifune et al. (2010) also examined changes in Fe–Mg partitioning between Pv and Fp for a pyrolite bulk composition at pressures up to 47 GPa. In addition, a preliminary laser-heated DAC experiments carried out by Kesson et al. (1998) and Murakami et al. (2005) have documented Fe–Mg partitioning data (between Pv and Fp) to deep lower-mantle pressures, but so far these data are somewhat limited. In particular, the effects of Soret diffusion, which occurs under the large temperature gradients imposed on experimental samples during laser heating, produces strong heterogeneities in iron content within the samples, and such experimental artifacts are therefore an additional concern in previous laser-heated DAC experiments (Sinmyo and Hirose 2010).

Here we determine the apparent Fe–Mg partition coefficient between Pv and Fp (K* = (FePv/MgPv)/(FeFp/MgFp)) for a pyrolitic bulk composition over a wide range of simulated lower-mantle pressures (28–114 GPa), using a combination of laser-heated DAC experimental techniques and TEM–EDS chemical analysis of recovered samples. We minimized the imposed temperature gradient and resulting Soret diffusion during laser-heated DAC experiments by following the method outlined in Sinmyo and Hirose (2010). The chemical homogeneity of the recovered laser-heated DAC experimental samples was carefully evaluated by X-ray mapping using a field-emission-type electron probe microanalyzer (FE-EPMA). We discuss the effects of high-spin to low-spin crossover of Fe3+ in Pv (Jackson et al. 2005; Li et al. 2005a; Zhang and Oganov 2006; Stackhouse et al. 2007; Lin and Tsuchiya 2008; Catalli et al. 2010, 2011; Hsu et al. 2011; Fujino et al. 2012)—a process that we invoke to explain the observed increase in iron content in Pv with increasing pressure above 90 GPa.

Experimental procedures

High-pressure and high-temperature syntheses

For this study of iron partitioning in pyrolitic lower mantle, we conducted four separate high-pressure and high-temperature experimental runs (#PYT01–03 and #PYP01). #PYP01 is conducted to compare samples heated by using coating technique and conventional technique. The sample #PYT04 is from an experimental run from a previous study (Sinmyo et al. 2011; see below for detail). The starting material for each experiment was a synthetic gel with a pyrolitic composition, identical to the starting gel used in our previous study (Sinmyo et al. 2011). Prior to experiment, the starting gel was dehydrated at 1,000 K for 1 h in a H2–CO2 gas-mixing furnace, in which the oxygen fugacity was controlled at slightly above the iron–wüstite buffer (i.e., 3 log units below the QFM buffer). The precise chemical composition of the gel was determined by electron microprobe. In our previous study, the Fe3+/ΣFe ratio of this starting material was determined to be 22 ± 13 % by electron energy-loss near-edge structure (ELNES) spectroscopy measurements (Sinmyo et al. 2011).

For the experiments, high-pressure and high-temperature conditions were generated in a laser-heated DAC. The starting material was coated with a layer of gold (~0.5–1.0 μm in thickness) on both sides to provide a laser absorber. This coating technique helps to minimize the temperature gradient during laser heating of the sample, thus avoiding internal variations in chemical composition caused by Soret diffusion (Sinmyo and Hirose 2010). For comparison, run #PYP01 was performed at 72 GPa and 1,900 K for 55 min using a powdered mixture of pyrolite gel and gold as the starting material. Each sample was loaded into a 50- or 100-μm-diameter hole that was drilled into a pre-indented rhenium gasket, together with a SiO2 glass pressure medium (in run #PYP01 we used NaCl instead of SiO2). The culet size of the diamond anvils was 300 μm or beveled 120 or 150 μm (inner diameter) sizes. Samples (20–80 μm diameter) were heated by double-side laser heating technique using a focused multi-mode Nd:YAG laser or a fiber laser. The laser spot diameter was around 20–80 μm. For each run, heating was carried out at a single pressure and temperature condition for 60 min, and scanning of the laser beam was not performed. Temperatures were measured using a spectroradiometric method, and pressure was determined in each run based on the Raman spectrum of the diamond after heating (Akahama and Kawamura 2004), correcting for thermal pressure. For the estimation of thermal pressure, we compiled the X-ray diffraction data taken for previous work (Fig. 1). Thermal pressure is almost independent from temperature at 1,800–2,500 K, and it ranges from +11 to +23 %. This value is well consistent with previous estimation (Ozawa et al. 2009; Fiquet et al. 2010). In this study, +15 % is practically applied for thermal pressure (#PYT01-03). Pressure of #PYT04 was determined under high-pressure and high-temperature condition by using EOS of gold (Hirose et al. 2008), since in situ X-ray diffraction data was available for #PYT04. The procedure of X-ray diffraction measurement was described elsewhere (Sinmyo et al. 2011).
Fig. 1

The relationship between thermal pressure and heating temperature under 38–117 GPa. Pressure at high temperature and 300 K was determined by using EOS of gold (Hirose et al. 2008) and Raman spectrum of the diamond after heating (Akahama and Kawamura 2004), respectively. All samples were coated by gold, and pressure medium was SiO2 glass. The X-ray diffraction data of gold were originally prepared for previous work (Sinmyo et al. 2011)

Chemical analyses

We performed TEM–EDS chemical analyses on four samples including the products of three new experimental runs (#PYT01–03) and one sample (#PYT04) from an experimental run from a previous study (Table 1). Sample #PYT04 was originally reported as PY93 in Sinmyo et al. (2011) and was repolished and reexamined in the present study along a different area, because the “local bulk FeO” content reported in Sinmyo et al. (2011) for that sample was higher by 15 % than the original bulk composition of the starting pyrolitic gel. Local bulk FeO is the bulk FeO content at analyzed small portion, and it represents the degree of heterogeneity in iron content. Local bulk FeO was calculated from mineral proportions and FeO content of adjacent Pv + Fp. The mineral proportions were determined from mass-balance calculation. For example, if the ratio of phase is Pv:Fp = 80:20 in wt% and FeO content of Pv and Fp are 5 and 10 wt% respectively, local bulk FeO content is 6 wt%. The samples recovered from the DAC experimental runs in this study (along with sample #PYT04) were Ar ion-thinned using a JEOL EM-09100IS Ion Slicer. We obtained thin film samples for each run, cut parallel to the compression axis (Tateno et al. 2009), which were then examined with a JEOL JEM-2011 TEM instrument operating at an accelerating voltage of 200 kV, and equipped with an Oxford Inca EDS system (Fig. 2). Around several 10 μm × 10 μm area were continuously investigated. TEM observation revealed Pv, Fp, and CaPv, in addition to trace amounts of Fe-metal grains. CaPv was not found in #PYT03 and 04 (Table 1). The chemical compositions of coexisting Pv, Fp, and CaPv in these thin film samples were determined with the aforementioned EDS system, based on the k-factor method (Cliff and Lorimer 1975). The k-factor was previously determined for a natural alkali-basalt glass standard. For these quantitative chemical analyses by EDS, we adopted a 100 nm beam spot size during each measurement (cf. Fujino et al. 1998). In addition, to monitor chemical homogeneity in the products of experimental runs #PYT01 and #PYP01, we carried out X-ray mapping of element abundances (Fe) and backscattered electron imaging of polished sections of these two samples using FE-EPMA (Fig. 3).
Table 1

Results of chemical analysis







28 GPaa, 1,900 K

82 GPaa, 2,100 K

109 GPaa, 2,300 K

114 GPab, 2,300 K
























55.3 (23)

1.8 (2)

53.5 (5)

56.2 (8)

53.2 (9)

54.5 (7)

1.0 (0)

53.7 (27)

1.6 (8)


4.2 (6)

0.1 (2)

0.1 (0)

4.2 (8)

0.1 (0)

4.5 (8)

4.0 (10)


4.5 (7)

25.8 (37)

0.7 (9)

4.4 (9)

23.0 (18)

0.0 (0)

6.2 (10)

22.6 (31)

6.9 (16)

17.4 (24)


34.2 (9)

72.0 (36)

3.3 (25)

32.4 (11)

76.0 (13)

1.4 (5)

32.0 (15)

73.2 (26)

33.7 (24)

81.0 (23)


1.0 (10)

0.1 (12)

42.0 (32)

2.7 (14)

1.0 (10)

44.7 (0)

2.8 (11)

3.2 (10)

1.1 (8)

Local bulk FeO

8.2 (10)


7.5 (17)


8.7 (6)


8.4 (11)



0.35 (6) 


0.42 (9)


0.59 (11)


0.90 (14)


Values are in wt%. Numbers in parenthesis indicate errors in the last digits. Both the K* value and local bulk FeO content (all Fe as FeO) were calculated from the compositions of adjacent Pv and Fp. Sample #PYT04 is originally reported in Sinmyo et al. (2011) (see text for details)

a After correction for thermal pressure (+15 %)

bPressure was determined under high-pressure and high-temperature condition using EOS of gold by Hirose et al. (2008) (see text for details)

c Total iron as FeO
Fig. 2

TEM bright-field image of the sample recovered from the laser-heated DAC experimental run at 82 GPa and 2,100 K (#PYT02). Pv Perovskite, Fp ferropericlase
Fig. 3

X-ray maps of Fe (left) and backscattered electron images (right) of two samples recovered from the laser-heated DAC experiments, including a #PYP01 and b #PYT01, obtained by FE-EPMA

Results and discussion

Strong chemical heterogeneity was observed in the sample recovered from run #PYP01 (Fig. 3a), in which a starting mixture of both gold and gel powders was used. While the gold powder should have been homogeneously mixed within this starting material, at some point during the laser heating DAC experiment it became clustered to one side of the sample (Fig. 3a). Since gold is a laser absorber, it should cause heterogeneous temperature distributions to occur within a laser heating DAC experimental sample if it is incorporated into the sample as a powdered mixture (i.e., as is the case for sample #PYP01). Indeed, iron is observed to have migrated to the other side of the sample during run #PYP01 (Fig. 3a), where temperatures should have remained low during laser heating due to the absence of laser absorber in that region. In contrast, the recovered sample from run #PYT01 was observed to be chemically very homogeneous (Fig. 3b), which was expected because this gel sample was coated with gold on both sides in order to minimize the development of a temperature gradient. Notably, the other remaining samples (i.e., #PYT02–04), the results for which are described below, were synthesized in a similar manner.

The chemical compositions of coexisting phases found within samples #PYT01–04 are given in Table 1. Both the K* value and the local bulk FeO content (all Fe as FeO) in these samples were calculated from the compositions of adjacent Pv and Fp. No significant localized depletions or enrichments in iron or other major elements were observed in any of these samples. Please note that obtained local bulk FeO contents were similar with original bulk FeO content of the starting material (8.2 wt%). In contrast, when we use conventional technique, the local bulk FeO content can be <50 % higher/lower than starting material (e.g., Sinmyo et al. 2008). While the deviation of local bulk FeO content from starting material was improved from about +15 % (Sinmyo et al. 2011) to +2 % (this study) in #PYT04, the change in K* value was minor (from 1.1 to 0.9). The K* values obtained are plotted as a function of pressure in Fig. 4. While earlier experimental results on K* based on the multi-anvil apparatus are quite scattered below 30 GPa (Irifune 1994; Wood and Rubie 1996; Wood 2000; Nishiyama and Yagi 2003; Irifune et al. 2010), the present value of K* = 0.35 at 28 GPa determined here is quite close to the values determined by McCammon et al. (2004). Furthermore, the value of K* = 0.42 at 82 GPa determined here is in good agreement with the results of similar work by Murakami et al. (2005). The iron content in Pv in this study was observed to increase from ~4 wt% at 82 GPa to ~6–7 wt% at 109–114 GPa, resulting in substantially higher values of K* (0.59–0.90).
Fig. 4

Variations in K* = (FePv/MgPv)/(FeFp/MgFp) as a function of pressure in the pyrolitic lower mantle. The gray area indicates a plausible range of K* with including the results of previous studies (see text for details). Experimental temperatures are as follows: 1,873 K (Irifune 1994; Wood and Rubie 1996; Nishiyama and Yagi 2003), 1,900 K (Wood 2000), 1,923–2,123 K (McCammon et al. 2004), 2,000–2,300 K (Murakami et al. 2005), and 1,873–2,073 K (Irifune et al. 2010)

Both theory and experiments have demonstrated that Fe3+ in Pv undergoes a high-spin to low-spin crossover when pressure conditions exceed ~70 GPa at temperatures of 0–300 K (Li et al. 2005a; Zhang and Oganov 2006; Stackhouse et al. 2007; Catalli et al. 2010, 2011; Hsu et al. 2011; Fujino et al. 2012). Pressure of spin transition is reported to increase ~30 GPa with increasing temperature from 300 to 2,000 K (e.g., Lin et al. 2007); thus, the increase in K* value observed with increasing pressure above ~90 GPa in this study is possibly attributed to spin crossover of Fe3+ in Pv, while determined Fe3+/ΣFe ratio has relatively high uncertainty. The Fe3+/ΣFe ratio in Pv synthesized at 82 GPa (sample #PYT02, this study) is estimated to be 0.50 ± 0.09, as determined from its Al content according to the method of McCammon et al. (2004), or it is also estimated to be 0.57 ± 0.18 from its Al content and bulk Fe content according to the method of Nakajima et al. (2012). In comparison, Pv formed at 114 GPa (sample #PYT04) exhibits Fe3+/ΣFe ratio of 0.59 ± 0.12 by ELNES measurements (Sinmyo et al. 2011). These data suggest that the Fe3+ content in Pv increased from 2.2 to 2.5 wt% at 82 GPa to 4.1 wt% at 114 GPa, and the concentration of Fe2+ in Pv increased comparatively less across this same pressure change, from 1.9–2.2 to 2.8 wt% Fe2+. While the uncertainties in the aforementioned Fe3+/ΣFe ratios are relatively large, the apparent enrichment of Fe3+ in Pv is consistent with its spin crossover above ~90 GPa (Fig. 4). Alternatively, the increase in K* observed in this study can be simply attributed to the effect of temperature and/or pressure. For example, it is known that partitioning coefficient of Mg–Fe2+ between Pv and Fp (KD = (Fe2+Pv/MgPv)/(Fe2+Fp/MgFp)) significantly increases with increasing temperature and slightly decreases with increasing pressure (Sinmyo et al. 2008; Nakajima et al. 2012). Although thermodynamic parameter related to Fe3+ in Pv is not well determined so far, it is reasonable that temperature and/or pressure have effect on K* value.

A plausible range of K* in pyrolite as a function of pressure is illustrated in Fig. 4. While K* are scattered below 30 GPa, several studies reported increase in K* at the uppermost lower mantle (below 30 GPa), and it may be attributed to the enrichment of Al and resulting Fe3+ in Pv (Wood 2000; Nishiyama et al. 2004; Irifune et al. 2010). A recent multi-anvil study by Irifune et al. (2010) reported that the K* value becomes diminished on account of the iron spin crossover in Fp below 50 GPa. Both Murakami et al. (2005), and this study consistently yielded constant K* values (~0.4) in pyrolite at pressures between ~50 and ~90 GPa. Furthermore, at pressures of 109–114 GPa we have shown that the K* value of pyrolite increases to ~0.6–0.9, likely because of spin crossover of Fe3+ in Pv, as discussed above. This estimated variation of K* in the lower mantle should also affect various physical properties of Pv and Fp. For example, at pressures above 90 GPa (~2,100 km depth) iron-rich Pv should have lower bulk modulus relative to iron-poor Pv (Li et al. 2005b; Stackhouse et al. 2006). In contrast, the bulk modulus should not vary significantly between iron-rich Fp and iron-poor Fp (Jacobsen et al. 2002). Recently, Manthilake et al. (2011) have shown that the lattice thermal conductivity of Pv decreases substantially with increasing iron content, and that this same relationship is less pronounced in Fp. In addition, although the uncertainties in Fe3+ content are relatively large, possible high Fe3+ contents in Pv occurring at pressures above 90 GPa might have a strong effect on the elasticity and conductivity of the bulk rock at such deep mantle conditions (Xu et al. 1998; Li et al. 2005b).


R. S. was supported by a Research Fellowship for Young Scientists and a Postdoctoral Fellowship for Research Abroad, granted by JSPS. Review comments by two anonymous referees and editor helped significantly in improving the manuscript.

Copyright information

© Springer-Verlag Berlin Heidelberg 2012