Technical Comments

(Mg,Fe)SiO3-Perovskite Stability and Lower Mantle Conditions

See allHide authors and affiliations

Science  13 Aug 1999:
Vol. 285, Issue 5430, pp. 983
DOI: 10.1126/science.285.5430.983a

The stability of (Mg,Fe)SiO3-perovskite at lower mantle conditions has been the subject of three recent studies (1, 2, 3). While our work (2) and the study of Meade et al. (3) show that perovskite may not be stable in the lower mantle and dissociates into its component oxides, G. Serghiou et al. (1) come to the opposite conclusion. In that report (1), perovskite was synthesized from a mixture of oxides, and attempts were made to decompose a glass of MgSiO3 composition between pressures of 73 to 100 GPa and at temperatures to 3000 K. Their experiments (1) were performed in CO2-laser heated diamond anvil cells (DACs) in Ar or CsCl pressure mediums, and processes in the samples were characterised by Raman spectroscopy or by study of fluorescence Cr-doped samples. Serghiou et al. (1) state that the results of conflicting studies (2 or 3) are incorrect and result from significant pressure gradients in experiments done without pressure medium and under temperature gradients introduced by heating with an Nd–yttrium-aluminum-garnet (Nd-YAG) laser. The difference between experiments (1) and (2) appears to be mostly the result of different ways of characterizing the conditions of experiments and specimens; perovskite does indeed dissociate into its component oxides.

The temperature gradients introduced in the sample by laser heating are very large, in the range of 10 to 50 K/μm (4, 5), but temperature distributions within the hot spot are quite similar (10 to 15 K/μm) for both CO2 lasers and Nd-YAG lasers running in multimode (5). Moreover, the thermal stresses produced by a temperature gradient on the order of 10 K/μm (107 K/m) could increase to 10 GPa at a peak temperature of 3000 K at an average pressure of 100 GPa, which would affect results of the experiments (5, 6). Thus, the thermal gradient by itself is not likely to be responsible for dissociation of perovskite in experiments with Nd-YAG laser, in contrast with experiments with CO2 laser.

The pressure gradients also could be quite different in experiments in (1) as contrasted with those in (2, 3) because of the difference in sample preparation and pressure medium. Meadeet al. (3) conduct experiments in NaCl and detected decomposition of (Mg,Fe)SiO3- perovskite. But even compression of pure perovskite or perovskite in CsCl pressure medium (for example, at 100 GPa) would not make any significant difference; estimated bulk moduli of MgSiO3-perovskite at such pressure is ∼700 GPa (7) and for CsCl it is ∼600 GPa, which means that perovskite alone is not much harder than CsCl (8).

The methods of pressure measurement in (1) and (2,3) were different. While the pressure in (2, 3) was based mostly on in situ x-ray measurements of pressure on the heated spots (during heating or in quench samples still under pressure), in (1) pressure was determined from a secondary ruby scale, with the pieces of ruby placed outside of the heated spots (9). Serghiou et al. (1) state that “the pressure difference between ruby chips in the heated area did not exceed 1 GPa.” First, in a heated area, ruby could react with any or all of the compounds (MgO, SiO2, and MgSiO3), which would affect pressure determination and the results of experiments. There is theoretical and experimental evidence of phase transition in corundum at a pressure of 78 to 100 GPa, especially at high temperature (≳ 1000 K), which might make the ruby pressure scale at >78 GPa problematic (10). It was already discussed by Tshuchida and Yagi (11) that in the laser heated spots pressure changes greatly. For example, they found (11) that pressure dropped by 10 to 20 GPa after heating of stishovite over 1000°C at 70 to 100 GPa. For platinum with in situ recorded diffraction patterns, the splitting of reflections corresponding to 20 to 25 GPa on laser heating at 60 GPa (11). It was also shown (11) that the pressure measured before and after heating do not actually reflect pressure (or stress) variations during the laser heating. We observed the change of 5 to 7 GPa in ruby heated with CO2 laser at initial pressure 50 to 60 GPa in CsCl pressure medium. In transition of silica polymorph, with four-fold coordinated silicon to the polymorph with six-fold coordinated Si, or when enstatite (or MgSiO3glass) transforms to perovskite, the reduction of the volumes were almost 50%, which caused the pressure to drop as a result of the rheological properties of the sample discussed above. In other words, experiments (1) described as conducted in the pressure range of 70 to 80 GPa probably were conducted in a range between 50 to 60 GPa, which is below the pressure expected (above 75 GPa at ∼2250 K) for dissociation of perovskite (12).

Another significant difference between data in (1) and (2, 3) arises from the methods of the phase analysis; as mentioned above, in (2, 3), x-ray was used, while Serghiouet al. (1) applied Raman spectroscopy to detect the products of heating of MgO + SiO2 mixture at high pressure. Periclase (or magnesiowustite) is Raman-inactive, and only perovskite or possible silica phases could be detected by Raman spectroscopy. Raman spectroscopy is widely used in high-pressure studies, and problems associated with broadening and decreasing intensities of the peaks, loss of the signal for crystallites of small size, and so forth, are well known (13). For example, Zerret al. (14), in studying the solidus of pyrolite-like composition after heating samples at pressure above 24 GPa, “were unable to detect any of major Raman lines of Mg-Si-perovskite” (15). Silica (amorphous and stishovite particularly) is known to produce weak Raman scattering at high pressure (above 20 GPa) or in samples quenched from DAC experiments. Spectrums of unheated mixtures of silica and periclase at high pressure are not presented in (1), and it is not possible to check if the absence of Raman reflections from silica is the result of a complete reaction of the MgO and SiO2 or a result of the absence of perovskite dissociation. Moreover, the quality of the Raman spectra presented in (1) allows the speculation that some extra Raman peaks might be presented in the spectrum collected, for example, after heating at 78 GPa.

Serghiou et al. (1) describe one more set of experiments with Cr-doped MgSiO3 glass. Cr-doped perovskite synthesised in situ in DAC, broken down at certain pressure and temperature conditions, purportedly produce Cr-doped MgO with easily detected fluorescence [figure 4a in (1)]. But the fluorescence spectrum of Cr-doped MgO not only varies with the concentration of Cr3+ content (16), but also changes significantly after compression or heating (17). After compression over 40 GPa, the intensity of fluorescence lines decrease rapidly and several features in the 710 to 715 nm region develop (the region of the fluorescence lines of Cr-doped Mg-Si-perovskite). Therefore, Cr-doped MgO is not a sensitive marker of possible decomposition of MgSiO3 perovskite. Moreover, there are no a priori reasons to believe that, on decomposition, Cr from the Cr-doped perovskite must go to MgO. The existence of Cr-doped silica is known (18).

There could be another reason why Serghiou et al. (1) observed the formation of Mg-Si-perovskite on heating of periclase and silica mixtures at pressure and temperature conditions at which decomposition of perovskite was detected earlier (2). Thermodynamics, lattice dynamics, and ab initio analysis (19) show that at pressure above 70 GPa, the Gibbs free energy of formation of Mg-Si-perovskite from MgO and silica phase with 4-coordinated Si (overpressurized quartz, silica gel, and so forth) is similar to that for the phase transition from silica into high-pressure modification. [Furthermore, we have found experimentally that if one uses stishovite or α-PbO2-like silica, MgO does not react to form perovskite at pressures above 80 GPa at T > 2000 K (20).] Therefore, if the low-pressure polymorphs of silica were chosen as a starting material for synthesis experiments, then on heating at high pressure with MgO, they could form perovskite metastably.


(Mg,Fe)SiO3-Perovskite Stability and Lower Mantle Conditions

Response: The suggestion by Saxena et al. (1) that silicate perovskite, considered to be the major component of the Earth's lower mantle, decomposes to its component oxides MgO and SiO2 above 65 GPa has important implications for the mineralogical constitution of the deep mantle. We showed in our report (2), however, that by significantly reducing the temperature and pressure gradients in comparison to those used by Saxena et al. (1), perovskite does not decompose; instead the reaction goes in the opposite direction. We conducted three types of experiments, the most important of which is the synthesis of MgSiO3-perovskite from these oxides at 100 GPa and 2800 K. We provided additional evidence for the stability of (Mg,Fe)SiO3-perovskite from Raman and fluorescence measurements, which yielded no indication for a breakdown to MgO and SiO2. The previously observed decomposition is most likely the result of large temperature (and possibly pressure) gradients present in those experiments, resulting from the use of an Nd-YAG laser and, more important, the lack of a thermally insulating, soft pressure medium.

Dubrovinsky et al. make, in sum, five points with which we disagree. We respond to each one in turn.

1) With regard to temperature gradients being similar in their and our experiments: Dubrovinsky et al. cite a laser study by Shenet al. (3) within the same reference as their own laser work (1, 4), as if the laser techniques were all the same, but they are not. We also disagree with their argument concerning solely different lasers. Even though important, the difference in the use of Nd-YAG lasers or CO2 lasers is not the main issue. An equally important factor in reducing temperature gradients in laser-heating experiments is the different cell geometry: In our study a defocused CO2 laser beam (power at least five times that of an Nd-YAG laser) was fully absorbed by a thin (10 μm) sample, which was thermally insulated from the highly conducting diamond anvils by a soft pressure medium of low thermal conductivity. This method largely eliminated axial temperature gradients and minimized radial gradients in the sample to about 10 K/μm (5). This laser-heating method is drastically different from the one used by Saxena et al. (1), where a Nd-YAG laser beam was focused to 20 μm heating the silicate sample indirectly via a metal foil. Moreover, the sample was in direct contact with the diamonds. The resulting axial temperature gradients were drastic, ranging from the peak temperatures of their measurements of 3000 K at the metal-silicate interface to about room temperature at the silicate-diamond interface (1, 4, 6–8). The radial temperature gradients were of order 50 K/μm (1). Additionally, thermal equilibrium was never achieved because the focused Nd-YAG laser had to be scanned across the sample.

2) With regard to pressure gradients in a hard silicate (Si-perovskite) being similar to those observed if a soft (CsCl) pressure medium is used: Shear properties of a material at high pressure are not simply described by the bulk modulus. Materials with relatively high bulk modulus may have low shear strength (for example MgO, a frequently used pressure medium in multi-anvil studies). Pressure gradients directly correlate with the yield strength. At low pressure the yield strength of CsCl is over one order of magnitude lower than that of perovskite, and at high pressure our observed pressure gradients in the diamond cell were about four times lower using CsCl in comparison to silicates without a pressure medium.

3) With regard to pressure in our experiments to 100 GPa being (as a result of heating) below those where they observed Si-perovskite decomposition (about 65 GPa): Dubrovinsky et al.'s statements concerning large pressure drops in our experiments seem to be based on incorrect interpretations of the literature. We stated (2) that pressures in the CsCl pressure medium were measured after heating from ruby chips 2 to 3 μm away from the sample. Relaxation of the pressure medium during heating of the sample reduced pressure gradients within this area to about 1 GPa. In fact, our ruby measurements of 100 GPa were consistent with the positions of the Raman frequencies of perovskite we measured (2). Pressure drops in silicates that undergo phase transitions have been documented by many authors including Tsuchida and Yagi (9). But this is only the case in the absence of a soft pressure medium, and therefore does not apply to our work.

4) With regard to the capability of Raman and fluorescence spectroscopy to detect SiO2 and Cr:MgO, respectively: First, we point out that the Raman spectrum of perovskite synthesized from MgO and SiO2 shown in our report is unambiguous. Second, the detection of small amounts of stishovite is equally straightforward, as demonstrated in our paper on the coesite-stishovite phase boundary (10). Third, we state in our paper (2) that Cr:MgO could be easily detected in a separate experiment after Cr-doped Mg2SiO4 broke down to perovskite and MgO. Cr:MgO always exhibits a sharp R-line [figure 4A in (2)]. Moreover, Cr3+ has also been shown to partition into MgO after heating chromium containing Mg2SiO4 at higher pressures (11). The crystal field stabilization energy in Cr3+-doped MgO is one of the highest for transition metal ions in oxide structures (about −233 kJ/g.ion) (12), which means that Cr3+ favorably partitions into MgO.

5) With regard to MgSiO3 forming metastably at 2800 K from MgO + SiO2 polymorphs with 4-coordinated Si: Dubrovinsky et al. do not mention the experimental findings indicating that the low pressure polymorphs of SiO2 (quartz, cristobalite) transform upon compression at room temperature (between 21 and 43 GPa for quartz and at 18 GPa for cristobalite) to rutile-based structures where silicon is coordinated to six oxygens (13, 14). The results are consistent with theoretical work (15). Therefore, between 70 to 100 GPa our 4-coordinated SiO2 starting materials had transformed to 6-coordinated structures, arguably with similar enthalpies and densities as SiO2 in the stishovite or the CaCl2 structure. None of these observations are considered in the references provided by Dubrovinsky et al. (16,17). It is therefore incorrect to state that, particularly at the high temperatures in our experiments, a dense SiO2phase would prefer to metastably react with MgO to form MgSiO3-perovskite and then re-dissociate to MgO and SiO2.

With regard to heating durations, Dubrovinsky et al. did not mention that our “comparably short heating durations” were the same as in their published experiments.

On a more general note, there has been a lot of disagreement in results coming from laser-heated diamond anvil cells. It has become clear that temperature gradients in earlier experiments were the main cause for these differences. Because these gradients are hard to characterize, as evident from numerous publications, experimental techniques had to be improved. One of the major improvements was thermal insulation of the samples, which reduced these large gradients by at least an order of magnitude. With respect to the study by Meade et al. (18), it seems fair to point out that a further study by Mao, Shen, and Hemley (19) supports the view that the previously observed decompositions of perovskite under deep mantle conditions (1, 18) were a result of temperature gradients. The experimental approach by Saxena et al. (1) in 1996 was not state of the art.


Navigate This Article