Background

There are regions at the base of the mantle where anomalous reductions of seismic wave velocities occur (e.g., Garnero and Helmberger 1995; Mori and Helmberger 1995). These are known as ultralow velocity zones (ULVZs). They vary in thickness from 5 to 40 km just above the core-mantle boundary (CMB) where there is a ~10–30 % seismic velocity reduction (e.g., Garnero et al. 1998). The possible presence of silicate melts resulting from partial melting of the lowermost mantle minerals (e.g., Williams and Garnero 1996) or remnants of the dense basal magma ocean (Labrosse et al. 2007) has been proposed to explain these anomalous seismic reductions at the CMB. Silicate melts under lower mantle conditions are not well understood because of the technical difficulties involved in conducting experiments on silicate melts under relevant high-pressure conditions. Although very recent in situ synchrotron X-ray diffraction measurements of basaltic melt were successful under lower mantle pressures (Sanloup et al. 2013), the pressure conditions of the measurements were limited to below ~60 GPa.

Silicate glasses have been used as possible analogs for silicate melts to understand the nature of dense magmas in the lower mantle (e.g., Murakami and Bass 2010, 2011; Petitgirard et al. 2015). Although the structural changes that silicate glasses undergo cannot be assumed to be a completely valid model for silicate melts, previous experimental works have shown that the Si-O coordination numbers of SiO2 glass and a molten basalt appear to change in generally similar ways (Meade et al. 1992; Lin et al. 2007; Sato and Funamori 2008, 2010; Benmore et al. 2010; Zeidler et al. 2014; Sanloup et al. 2013). This suggests that the analogy of the structural change that silicate glasses undergo to the changes in silicate melts appears to work well for Si-O coordination number and probably for most structural aspects at least above 10 GPa (Sanloup et al. 2013).

The behavior of silicate glasses under ultrahigh pressures above 100 GPa has recently been explored by acoustic wave velocity measurements using in situ high-pressure Brillouin scattering spectroscopy as a way to monitor structural changes in silicate melts under ultrahigh-pressure conditions (Murakami and Bass 2010, 2011). The results for SiO2 glass (Murakami and Bass 2010) showed that the changes in acoustic wave velocity occur at the same pressure as the changes in the Si-O coordination number of SiO2 glass from 4 to 6, inferred from in situ synchrotron X-ray diffraction measurements up to 100 GPa (Sato and Funamori 2010). This indicates that acoustic wave velocity can be a good indicator for tracking the pressure evolution of the Si-O coordination number of silicate glasses. Murakami and Bass (2010) also reported an abrupt increase of P-V S gradient (dV S /dP) in SiO2 glass above ~140 GPa, which was also observed in MgSiO3 glass at ~133 GPa (Murakami and Bass 2011) and proposed the possibility of a further structural change such as a gradual increase of the average Si-O coordination number from 6 to 6+. Such a structural change may increase the densities of silicate melts significantly, and therefore can constrain the dynamics of silicate melts that may exist at the base of the mantle.

Acoustic wave velocity measurements of silicate glasses under ultrahigh-pressure conditions have so far been conducted only in very simple systems such as SiO2 and MgSiO3 (Murakami and Bass 2010, 2011), and it is still unclear whether the densification mechanisms inferred from the velocity measurements of SiO2 or MgSiO3 glasses are directly applicable to the silicate glasses/melts in more realistic compositional systems in nature. To address this issue, we focus on the effect of Al2O3, the second-most abundant oxide in natural magma, on structural changes in silicate glasses. The SiO2-Al2O3 system is the simplest from among the aluminous silicate systems. A number of previous studies have reported structural features at ambient pressure, as inferred by Raman scattering (Kato 1976; McMillan and Piriou 1982), infrared absorption (Poe et al. 1992a; Okuno et al. 2005), X-ray diffraction (Morikawa et al. 1982; Okuno et al. 2005), nuclear magnetic resonance (NMR) spectroscopic techniques (Risbud et al. 1987; Sato et al. 1991; Poe et al. 1992a, b; Sen and Youngman 2004; Weber et al. 2008), and by molecular dynamics (MD) computer simulations (Poe et al. 1992a, b; Linh and Hoang 2007).

Previous high-pressure experiments using NMR spectroscopic techniques showed that the proportion of five- and sixfold coordination of Al in aluminosilicate glasses quenched from high pressures and temperatures corresponding to the upper mantle is positively correlated with pressure (e.g., Yarger et al. 1995; Lee et al. 2004; Allwardt et al. 2005, 2007), suggesting that the increase in Al-O coordination number occurs at lower pressures than that in Si-O. In situ XRD measurement for CaAl2Si2O8 glass supports the conclusions inferred from those NMR studies (Drewitt et al. 2015). However, the role Al2O3 plays in the structure and physical properties of silicate glasses and melts under lower mantle pressures is still unknown.

To advance our understanding on the effect the presence of Al2O3 has on the structures of aluminosilicate glasses at lower mantle pressures, we performed in situ high-pressure Brillouin scattering measurements of SiO2-Al2O3 glasses up to pressures of 196.9 GPa. Based on the results, we discuss the effect of Al2O3 on the densification mechanism of aluminosilicate glasses at the lowermost mantle pressure conditions and consider possible implications for the dynamics of dense silicate melts at the base of the mantle.

Methods

Two aluminosilicate glass samples, SiO2 + 3.9 mol% Al2O3 glass (SA1) and SiO2 + 20.5 mol% Al2O3 glass (SA2) were synthesized by means of a containerless levitation heating method at the SPring-8 BL04B2 beamline at the SPring-8 synchrotron radiation facility in Hyogo, Japan (Kohara et al. 2007). By using this method, both contamination of the molten samples and heterogeneous nucleation by crucible surfaces were avoided. Blocks of silicate powders weighing 1–2 mg were levitated in air using an aerodynamic levitator and heated by a 100 W CO2 laser beam. After melting the samples, small, rapidly cooled vitreous spheroids (~0.5–1 mm in diameter) were obtained by blocking the CO2 laser.

The X-ray diffraction data for the SA1 and SA2 were corrected for polarization, absorption, and background, and the contribution of Compton scattering was subtracted using standard analysis procedures (see Kohara et al. 2007 for the technical details). The corrected data sets were normalized to give the Faber-Ziman total structure factor S(Q). The radial distribution function RDF(r) was obtained by a Fourier transformation of S(Q) (Fig. 1). The composition and homogeneity of those glasses were confirmed by scanning electron microscope with energy-dispersive X-ray spectrometer (SEM-EDS). The compositional errors in the Al2O3 mol% were ±0.4 % in SA1 and ±0.5 % in SA2, estimated from the standard deviation of 10 analyzed points in SA1 and SA2.

Fig. 1
figure 1

a Total structure factors S(Q) and b radial distribution functions of SA1 and SA2. The first neighbor T-O (T = Si, Al) lengths in SA1 and SA2 were 1.62 and 1.65 Å, respectively. The measurements were carried out in step-scanning mode with 0.05°/2θ in the range 0.3° < 2θ < 10.0°, 0.10°/2θ in the range 10.0° < 2θ < 19.9°, 0.15°/2θ in the range 19.9° < 2θ < 40.0°, and 0.20°/2θ in the range 40.0° < 2θ < 58.0° in transmission geometry. Incident X-ray energy was 61.55 keV and scattering vector (Q) changes were from 0.2 Å−1 to 30.2 Å−1

Full size image

In situ high-pressure acoustic wave velocity measurements were carried out at room temperature using a high-pressure Brillouin scattering spectroscopic system combined with a symmetric-type diamond anvil cell apparatus at Tohoku University (see also Mashino et al. 2016). The acoustic wave velocities, i.e., a longitudinal wave velocity (V P ) and a transverse wave velocity (V S ), in a symmetric geometry are calculated from the relation (Whitfield et al. 1976):

$$ V=\frac{\varDelta \omega \lambda }{2 \sin \left(\theta /2\right)} $$

where λ is the laser wavelength of the incident laser beam and θ is the external scattering angle. Δω indicates the Brillouin frequency shifts of longitudinal or transverse acoustic modes. An argon ion laser with a wavelength of 532 nm was used as the probe beam. The incident laser beam was focused to a spot size of ~20 μm. The scattered light was analyzed by a six-pass tandem Fabry-Pérot interferometer. In all Brillouin experiments, we used a symmetric scattering geometry with a 50° external scattering angle. A borosilicate crown optical glass (BK7) was used to calibrate the symmetric scattering geometry.

The SiO2-Al2O3 glass powders were loaded without a pressure-transmitting medium into a 50 or 70 μm hole drilled in a rhenium gasket. The samples were compressed with 100 and 130 μm beveled diamond anvils and a 300 μm non-beveled diamond anvil. Pressure was determined using the Raman T 2g mode of the diamond anvil (Akahama and Kawamura 2004) by measuring 3–5 points at the center of the incident laser beam and surrounding points, yielding uncertainties typically within 2 GPa (see Additional file 1: Table S1 and Additional file 2: Table S2). Representative Raman spectra of the surface of the diamond anvils directly in contact with the sample and their differential spectra dI∕dν (I intensity, ν: wave number) from the different points are shown in Fig. 2. The high-frequency edge of the Raman band was defined as a minimum of dI∕dν, which is determined by curve fitting using a Gaussian function. As shown in Fig. 2, we obtained very sharp peaks for the dI∕dν function at each pressure point, even above 100 GPa. Brillouin spectra were collected at pressure conditions from 11 to 169 GPa in 1–13 GPa increments for SA1 and from 10 to 197 GPa in 2–15 GPa increments for SA2. The collecting time for each Brillouin scattering measurement was from 3 to 96 h. At each pressure, the raw Brillouin spectra of Stokes and anti-Stokes peaks were fitted with Lorentzian functions to obtain the peak positions.

Fig. 2
figure 2

Representative Raman spectra of different pressure conditions. Raman spectra were obtained from SA1 (SiO2 + 3.9 mol% Al2O3) and SA2 (SiO2 + 20.5 mol% Al2O3), respectively. They were measured from reflected light of the incident laser beam three times (a1, a2, a3, b1, b2, b3). Their differential spectra dI∕dv are also shown (a4, a5, a6, b4, b5, b6). The high-frequency edge of the Raman band was defined as the minimum of the dI∕dv. The red lines represent the fitting curves using Gaussian functions to determine the minimum position. Numbers in parentheses represent uncertainties in the last digit for both wave number and pressure. Uncertainties for the wave number of the minimum position of dI∕dv in each spectrum were derived from the errors of the curve fitting. The total uncertainties for pressure were finally estimated from the standard deviation (1σ) of each determined pressure value at each point

Full size image

Results and discussion

The RDFs showed that the first peaks observed in SA1 and SA2 were at 1.62 Å and at 1.65 Å, respectively (Fig. 1). As the Al-O bond length in the tetrahedral site is 0.13 Å longer than the Si-O bond length in the tetrahedral site (Shannon 1976), the average first nearest-neighbor T-O (T = Si, Al) lengths in SiO2-Al2O3 glasses are longer than the Si-O length. These lengths for SA1 and SA2 were between that of SiO2 glass (1.59 Å; Meade et al. 1992) and those of SiO2 + 25.0 mol% Al2O3 glass (1.66 Å; Okuno et al. 2005) and SiO2 + 28.2 mol% Al2O3 glass (1.75 Å; Morikawa et al. 1982), showing that our results were in agreement with previous studies about T-O lengths at ambient conditions.

Brillouin peaks from the transverse acoustic modes of both SA1 and SA2 were observed over the entire pressure range up to 169.3 and 196.9 GPa, respectively (Figs. 3, 4, and 5, Additional file 1: Table S1, Additional file 2: Table S2). The peaks from the longitudinal acoustic modes of those glasses were also observed at pressures below 45 GPa (a and b in Fig. 3). The peaks from the longitudinal acoustic modes were masked by those of the diamond transverse acoustic modes above 45 GPa (c and d in Fig. 3). As shown in Fig. 3, no significant peak broadening was observed with increasing pressure, indicating that the hydrostaticity in the sample chamber did not change significantly as the pressure changed (see also full width at half maximum (FWHM) in Additional file 1: Table S1 and Additional file 2: Table S2). The velocity uncertainties based on the difference between the peak positions of the Brillouin spectra of Stokes and anti-Stokes peaks were 0.8 % in SA1 and 0.3 % in SA2 on average throughout the pressure range measured (Additional file 1: Table S1 and Additional file 2: Table S2).

Fig. 3
figure 3

In situ high-pressure Brillouin spectra. Spectra of a SA1 at 20 GPa, b SA2 at 16 GPa, c SA1 at 169 GPa, and d SA2 at 189 GPa. TA and LA indicate transverse and longitudinal acoustic modes of the Brillouin shift, respectively. Ghost peaks are artifacts of the interferometry method. LA peaks are masked above around 45 GPa because of the overlap of the diamond’s TA peaks

Full size image
Fig. 4
figure 4

P-V S profiles up to 200 GPa. SA1 (blue squares, this study), SA2 (light blue diamonds, this study), SiO2 glass (red circles, Murakami and Bass 2010), and MgSiO3 glass (green triangles, Murakami and Bass 2011). The curves show the fourth order polynomial fitting of SA1 (R 2 = 0.9960), SA2 (R 2 = 0.9978), SiO2 glass (R 2 = 0.9964), and MgSiO3 glass (R 2 = 0.9985). The pressure conditions at which the increase of dV S /dP occurs (the inflection pressure) are indicated by the semi-transparent dark-blue box (SA1) and the semi-transparent light-blue box (SA2)

Full size image
Fig. 5
figure 5

Pressure-transverse acoustic wave velocity (P-V S ) profiles up to 40 GPa. a SA1 (blue squares, this study), SA2 (light blue diamonds, this study), SiO2 glass (red circles, Murakami and Bass 2010), and MgSiO3 glass (green triangles, Murakami and Bass 2011) up to 40 GPa. b Regression curves of SA1 and SA2 fitted by polynomial functions. P-V S plots of SA1 from 11 to 60 GPa are fitted with a fourth order polynomial (R 2 = 0.9933) and P-V S plots of SA2 from 10 to 59 GPa are fitted with a quadratic function (R 2 = 0.9957)

Full size image

Variation in acoustic wave velocities with pressure was divided into three distinct pressure regions (Fig. 4): (i) the lowest pressure region where acoustic wave velocities increased rapidly up to 40 GPa; (ii) a high-pressure region from 40 to 100 GPa with a gentle gradient along a progressively flattening trend, and (iii) an ultrahigh-pressure region above 100 GPa, with an anomalously increasing V S , which trend is highly consistent with that of previous works on SiO2 and MgSiO3 glasses (Murakami and Bass 2010, 2011).

Figure 5a shows the P-V S profiles of SA1 and SA2 up to 40 GPa as functions of pressure together with those of SiO2 glass (Murakami and Bass 2010) and MgSiO3 glass (Murakami and Bass 2011). The values of V S of the SiO2-Al2O3 glasses increase rapidly from 10 to 40 GPa, with no discontinuous changes in slope. Figure 5b shows the regression curves of P-V S plots of SA1 fitted to a fourth degree polynomial function and SA2 fitted to a quadratic function. The P-V S profile of SA1 up to 40 GPa exhibited an increasing V S with an inflection point at ~26 GPa. Such a trend was also observed in the P-V S profile of SiO2 glass (Murakami and Bass 2010). In contrast, the P-V S profile of SA2 to 40 GPa can be fitted by a low-degree polynomial function such as a concave quadratic function (Fig. 5b), which is quite similar to that of MgSiO3 glass (Murakami and Bass 2011). The P-V S profile of SA2 below 40 GPa seems to be that of SA1 just shifted to lower pressures (Fig. 5). The difference in V S in the pressure range from 10 to 40 GPa cannot be explained simply by density differences caused by the differing Al2O3 contents in SA1 and SA2. Previous studies on SiO2-Al2O3 glasses at ambient conditions reported that the density of SiO2-Al2O3 glasses increases with increasing Al2O3 content (Okuno et al. 2005). If we apply this density relationship in the SiO2-Al2O3 system, the densities of SA1 and SA2 at ambient pressure can be estimated to be 2.24 and 2.40 g/cm3, respectively, which are denser than pure SiO2 glass (2.20 g/cm3) (e.g., Brückner 1970). If the shear moduli for the two glasses were equivalent, V S would be expected to decrease as Al2O3 content increases in SiO2-Al2O3 glasses because density is a term in the denominator in the formula of an acoustic wave velocity. However, the V S of the “Al2O3-rich” SA2 is higher than that of the “Al2O3-poor” SA1 at pressures between 10 and 40 GPa. This P-V S relation is thus likely to be due to differences in the elasticity/elastic moduli of SA1 and SA2, which appear in the numerator in the formula for acoustic wave velocity.

One possibility for explaining such an elasticity change is that the proportion of Al ions with large oxygen coordination numbers could depend on the Al2O3 content. A number of studies have shown that the average Al-O coordination numbers of aluminosilicate glasses always reach five and six at lower pressures than the average Si-O coordination number of SiO2 glass and aluminosilicate glasses (e.g., Yarger et al. 1995; Lee et al. 2004; Allwardt et al. 2005; Drewitt et al. 2015; Benmore et al. 2010; Sato and Funamori 2010; Mysen and Richet 2005, and see also Additional file 3: Figure S1). In the present study, SA2 has more Al ion than SA1. In addition, some Al ions in SiO2-Al2O3 glasses may be in 5- and 6-fold coordination with oxygen even at ambient pressure. This can serve as a network modifier, at least if the Al2O3 content is lower than 20.5 mol% (the composition of SA2). The Al ions in SA2 may be mainly in fourfold coordination with oxygen and serve as a network former, but about 40 % of the Al ions may exist in five- and sixfold coordination with oxygen at ambient pressure, as inferred from MD simulations for liquid SiO2-Al2O3 (Poe et al. 1992a, b). Therefore, the proportion of Al ions with large oxygen coordination numbers (five- and sixfold Al) and the value of V S in SA2 are always greater than in SA1 at isobaric conditions below 40 GPa. The increase of average Al-O coordination number of SA2 with increasing pressure may reflect its V S increase. The V S difference between SA1 and SA2 up to 40 GPa implies that aluminosilicate glasses become elastically stiffer as the proportion of five- and sixfold Al ion increases. In the present study, SA2 is elastically stiffer than SA1 at isobaric conditions between 10 and 40 GPa, and the P-V S profile of SA2 seems, consequently, to be that of SA1 shifted to lower pressures.

The gradient of V S with pressure becomes less steep above 40 GPa in contrast to its trend at lower pressures, which appears similar to the behavior of SiO2 glass (Murakami and Bass 2010) and MgSiO3 glass (Murakami and Bass 2011) (Fig. 4). Previous high-pressure experimental studies on SiO2 glass showed that the Si-O coordination number of SiO2 glass most likely changes from four to six up to ~40 GPa (Meade et al. 1992; Lin et al. 2007; Sato and Funamori 2008, 2010; Benmore et al. 2010; Zeidler et al. 2014) and that the sixfold coordination state may be preserved up to at least ~100 GPa (Sato and Funamori 2010). This behavior offers a reasonable explanation for the high-pressure acoustic wave velocity profile of SiO2 glass (Murakami and Bass 2010). Because the trend of the P-V S profile of SA1 is nearly identical to that of SiO2 glass, SA1 also preserves the sixfold coordination state of Si, at least from 40 to 100 GPa. In contrast, the P-V S profile of SA2 is convex upward with a gentle gradient from 10 to around 100 GPa, which is similar to that observed in MgSiO3 glass (Murakami and Bass 2011). This gradually increasing V S for SA2, similar to that of MgSiO3 glass up to at least 100 GPa, also suggests that the average Si-O coordination number increases gradually from four to six in this pressure region.

The values of V S for SA1 and SA2 increase more rapidly as pressure grows past ~100 GPa (Fig. 4), which was also observed in SiO2 glass (Murakami and Bass 2010) and MgSiO3 glass (Murakami and Bass 2011). Such dV S /dP changes observed above ~100 GPa have previously been interpreted to reflect the onset of the average Si-O coordination number changing from just 6 to 6+ (Murakami and Bass 2010, 2011). This conclusion subsequently found support in the results from MD computer simulations (Brazhkin et al. 2011; Wu et al. 2012). To determine the pressure conditions at which this abrupt change of dV S /dP occurs, we adopted the same criteria as Murakami and Bass (2010, 2011). Each P-V S profile was fitted to fourth and fifth polynomial functions of pressure with an adjusted R-square of above 0.995. The pressure conditions at the onset of the rapid V S increase in the P-V S profiles of SA1 and SA2 were determined by calculating the point at which the second derivative of the fitted polynomial functions with respect to pressure (d 2 V S ∕dP 2) is a maximum. The pressure conditions at which the increase of dV S /dP occur (hereafter referred to as “the inflection pressure”) are 130 ± 5 GPa in SA1 and 116 ± 9 GPa in SA2. The inflection pressure of SA1 is 10 GPa lower than that of SiO2 glass (Murakami and Bass 2010) and 3 GPa lower than that of MgSiO3 glass (Murakami and Bass 2011). The inflection pressure of SA2 is 24 GPa lower than that of SiO2 glass (Murakami and Bass 2010) and 17 GPa lower than that of MgSiO3 glass (Murakami and Bass 2011).

Such a change of dV S /dP may indicate the onset of further structural change in silicate glasses resulting from an average Si-O coordination number increase from 6 to 6+. Our results may suggest that the inflection pressure does not strongly depend on the amount of metal ions because the M/Si (M = Mg, Al) values for SA1 and SA2 are less than for MgSiO3 glass, which has much higher inflection pressure than SA1 and SA2 (Fig. 6). This indicates that Al2O3 is much more effective than MgO in reducing the pressure at which the average Si-O coordination number starts to increase from 6 to 6+. Another explanation of the lowered inflection pressure is that the average Al-O coordination numbers increase from 6 to 6+ at a lower pressure than do those of Si-O, similar to the transition of average Al-O coordination number from 4 to 6 that occurs at lower pressures than the transition of Si-O. However, no study has yet indicated the different pressures at which the average Al-O and Si-O coordination numbers exceed 6. In addition, the P-V S profiles of SA1 and SA2 have only one inflection point, at the onset of the average Si-O coordination number’s increase from 6 to 6+ (Brazhkin et al. 2011; Wu et al. 2012; Murakami and Bass 2010, 2011). Therefore, the inflection points under ultrahigh pressure may primarily reflect a change of the average Si-O coordination number from 6 to 6+.

Fig. 6
figure 6

Inflection pressures in the P-V S profiles of the silicate glasses. SA1 (blue square, this study), SA2 (light blue diamond, this study), SiO2 glass (red circle, Murakami and Bass 2010), and MgSiO3 glass (green triangle, Murakami and Bass 2011) are shown in the left graph. This is regarded as the onset of the average Si-O coordination number increasing from 6 to 6+ (“6 + -fold (Si)” in this figure) in each glass (see Murakami and Bass 2010, 2011). The solid black line shows the possible transitional boundary at which the average Si-O coordination number increases from 6 to 6+ in SiO2-Al2O3 glasses, calculated using a linear leasts squares method based on the plots of SiO2 glass, SA1, and SA2. The pressure-intercept is fixed at 140 GPa, derived from the inflection pressure of SiO2 glass reported by Murakami and Bass (2010). Assuming an Al2O3 content of 13 mol% in an SiO2-Al2O3 glass, the average Si-O coordination number is expected to exceed 6 at 124 GPa (purple lines), which corresponds to a depth of 2690 km, much shallower than the CMB. The dashed gray line shows the boundary at which the average Si-O coordination number increases from just 6 to 6+ in MgSiO3 glass. The picture on the right is a schematic image of the accumulation of “dense” partial melts having structures governed by 6+-fold Si, generated from MORBs

Full size image

Density increases caused by structural transitions accompanied by Si-O coordination number growth occur continuously in glasses and melts and occur discontinuously in crystalline materials (e.g., Loerting et al. 2009). Although it is not obvious whether the analogy between silicate glasses and silicate melts works well as a model of their structural or density changes with pressure, experimental results on the evolution of average Si-O coordination number of SiO2 glass (Meade et al. 1992; Lin et al. 2007; Sato and Funamori 2008, 2010; Benmore et al. 2010; Zeidler et al. 2014) shows good agreement with that of SiO2 melt calculated by MD simulations (Karki et al. 2007). A recent experimental study suggested that the Si-O coordination state in molten basalt is roughly consistent with that in SiO2 glass and melt up to 60 GPa (Sanloup et al. 2013). If this is the case, the pressure-induced acoustic wave velocity changes of the aluminosilicate glasses explored in the present study may offer possible implications for the dynamics of dense magmas that potentially exist at the base of the mantle. The possible existence of dense silicate melts at or near the CMB has been proposed (e.g., Ohtani 1983; Ohtani and Maeda 2001; Williams and Garnero 1996; Labrosse et al. 2007) to explain the anomalous reduction of the seismic wave velocities just above the CMB (e.g., Garnero and Helmberger 1995; Mori and Helmberger 1995). Recent high-pressure melting experiments indicated that such dense silicate melts at the CMB can be generated by partial melting of mid-ocean ridge basalts (MORBs) (Andrault et al. 2014; Pradhan et al. 2015). The densification in SA1 and SA2, which probably reflects an average Si-O coordination number in excess of 6, occurs at lower pressures than in pure MgSiO3 glass (Fig. 6). Figure 6 shows the inflection pressure as a function of Al2O3 contents in the SiO2-Al2O3 system, indicating a linear relationship. If we assume that an Al2O3 content of ~13 mol% may possibly be included in partial melts of MORBs generated at around 100 GPa (Pradhan et al. 2015), the inflection pressure is expected to be 124 GPa corresponding to a depth of ~2690 km. The silicate melts generated by the partial melting of MORBs might thus undergo a structural transformation involving possible densification changes related to Si-O coordination states at depths of around ~2690 km, which is clearly shallower than the CMB.

Although the possibility of densification processes in the silicate melts occurring due to the addition of heavy elements such as Fe cannot be ruled out, the increases in density associated with structural changes in Si-O coordination states that possibly take place at pressures within the lower mantle might also play an important role in the dynamics of silicate melts at the base of the mantle.

Conclusions

The transverse acoustic wave velocities of SiO2-Al2O3 glasses were measured at pressures up to near 200 GPa. At pressures from 10 to 40 GPa, the V S of the Al2O3-rich SA2 is higher than that of the Al2O3-poor SA1, which is likely related to differences in the elasticity/elastic moduli between SA1 and SA2. Such a difference in the elasticity may be derived from the different proportions of five- and sixfold Al ions in those glasses. The dV S /dP gradients of both SA1 and SA2 became less steep at pressures from 40 to 100 GPa. In this pressure range, SA1 had a P-V S profile nearly identical to that of SiO2 glass, indicating that the sixfold coordinated structure is preserved. SA2, however, had a P-V S profile similar to that of MgSiO3 glass, suggesting a gradually increasing average Si-O coordination number from 4 to 6 up to around 100 GPa. dV S /dP also increases at 130 ± 5 GPa in SA1 and at 116 ± 9 GPa in SA2, which can be interpreted as the onset of the average Si-O coordination number change from 6 to 6+. The inflection pressures for SA1 and SA2 are much lower than in MgSiO3 glass, suggesting that Al2O3 promotes the reduction of pressure conditions of inflection caused by an increase of the average Si-O coordination number from 6 to 6+ much more strongly than does MgO. Such a densification mechanism associated with the Si-O coordination number change may play a key role in the gravitational stabilization of the partial melting of MORBs within the lowermost mantle deeper than 2690 km.

Abbreviations

CMB, core-mantle boundary; MD, molecular dynamics; MORBs, mid-ocean ridge basalts; NMR, nuclear magnetic resonance; RDF, radial distribution function; SA1, SiO2 + 3.9 mol% Al2O3 glass; SA2, SiO2 + 20.5 mol% Al2O3 glass; SEM-EDS, scanning electron microscope with energy-dispersive X-ray spectrometer; ULVZs, ultralow velocity zones; V P , longitudinal acoustic wave velocity; V S , transverse acoustic wave velocity.