Melting of subducted basalt at the core-mantle boundary

See allHide authors and affiliations

Science  23 May 2014:
Vol. 344, Issue 6186, pp. 892-895
DOI: 10.1126/science.1250466


The geological materials in Earth's lowermost mantle control the characteristics and interpretation of seismic ultra–low velocity zones at the base of the core-mantle boundary. Partial melting of the bulk lower mantle is often advocated as the cause, but this does not explain the nonubiquitous character of these regional seismic features. We explored the melting properties of mid-oceanic ridge basalt (MORB), which can reach the lowermost mantle after subduction of oceanic crust. At a pressure representative of the core-mantle boundary (135 gigapascals), the onset of melting occurs at ~3800 kelvin, which is ~350 kelvin below the mantle solidus. The SiO2-rich liquid generated either remains trapped in the MORB material or solidifies after reacting with the surrounding MgO-rich mantle, remixing subducted MORB with the lowermost mantle.

Delving deeper into the lower mantle

Earth's lower mantle is an enigmatic region, a transition zone between slowly churning solids and a liquid outer core. Large seismic structures and discontinuities in this region are probably due to sharp gradients in temperature, composition, or mineralogy. Teasing apart the precise effects of these factors requires experiments at lower mantle temperatures and pressures (see the Perspective by Williams). Zhang et al. found that the major mineral phase of the lower mantle decomposes into two minerals. Andrault et al. show how the melting of subducted basalt from the oceanic crust will form pile-like structures on top of the core/mantle boundary.

Science, this issue p. 877, p. 892; see also p. 800.

Numerous seismological studies have demonstrated the complexity of the lowest 150- to 300-km-thick mantle layer situated just above the core-mantle boundary (CMB). In many areas, there is an intermittent stratification, with 1.5 to 3% velocity discontinuities, as well as lateral shear-wave anisotropy (1). These anomalies could arise from mineralogical heterogeneities (2), magma ocean crystallization (3), the descent of subducted slabs deep into the lower mantle (4), and/or chemical reactions with the outer core. In addition, ultra–low velocity zones (ULVZs), with shear-wave velocity reduction of more than 10%, have been detected in specific mantle regions (5). Their size is limited to ~40 km thickness and ~100 km across, and they are ~10% denser than the surrounding mantle. These ULVZs could be due to partial melting occurring in a steep temperature gradient when approaching a very hot outer core. Between 5 and 30% partial melting could attenuate P and S seismic wave velocities to a similar amplitude as that reported for ULVZs (6). However, it was recently shown that partial melting of the bulk lower mantle cannot produce a residue with an equilibrium partial melt that is sufficiently dense to adequately represent the ULVZ mush (7).

Partial melting of pyrolitic or chondritic mantle would be possible if the CMB temperature were higher than ~4150 K (8, 9). Although such a high temperature is not precluded, it would require a very hot core. It is more than 1000 K above the most recent determination of the mantle adiabat extrapolated to the CMB (10). Moreover, the presence of a very hot core today makes it difficult to explain how a geodynamo could have been maintained for prolonged geological periods: An even hotter core would be required in the past, or an extremely high concentration [400 to 800 parts per million (ppm)] of K in the core (11). Alternatively, partial melting in the lowermost mantle could occur below 4150 K if the mantle composition were drastically altered. Chemical differences between peridotite, pyrolite, and chondritic-type mantle would produce only minor variations, because they contain similar mineralogical assemblages. However, a local concentration of fusible (such as alkalis) or volatile (H2O or CO2) elements, together with a change in mineralogy, could have a marked effect on the solidus temperature. In particular, a small water concentration can be enough to reach the saturation limit of the bulk mantle (12). Consequently, the melting curve of hydrous pyrolite has much lower melting temperatures (13).

We investigated the melting behavior of a natural mid-oceanic ridge basalt (MORB) collected at a 2800-m depth during the Searise-1 research cruise (table S1). When a MORB eventually reaches Earth's lower mantle, it is composed of four coexisting phases: (i) Fe-rich silicate perovskite (Mg-Pv); (ii) free silica, in contrast with free (Mg,Fe)O ferropericlase (Fp) in the bulk mantle (fig. S1); (iii) Ca-bearing silicate perovskite (Ca-Pv); and (iv) enough Al, Ca, Na, and K to generate cage structures such as calcium ferrite (CF), hollandite, or an aluminous phase (1416). Our starting material contained ~0.3 weight % H2O and traces of CO2 (fig. S9), which are common volatile concentrations for natural MORBs. At lower mantle pressures, the basaltic portion of the slab contains minerals that can carry water, such as the Al-bearing stishovite (17), δ-AlOOH (18), and phase H (19).

High pressures and temperatures were provided by a laser-heated diamond anvil cell [fig. S2 (8)]. The sample behavior was followed continuously using in situ x-ray diffraction (Fig. 1 and supplementary materials). Upon heating, we first observed the appearance of a continuous diffraction ring on the charge-coupled device detector, at dhkl distances typical of Mg-Pv. This ring occurred above 1800 and 2500 K, at 40 and 135 GPa, respectively. The relatively large peak profile is typical of a powder with very small grain size (Fig. 1A). At higher temperatures, Ca-Pv and silica appeared simultaneously (fig. S3 and table S2). The latter adopts either the structure of rutile (stishovite), CaCl2, or α-PbO2 (seifertite), as a function of pressure (fig. S4). Diffraction peaks of the post-perovskite phase of MgSiO3 also appeared above 115 GPa. Diffraction features typical of CF and/or the so-called new aluminous phase remained weak because of major overlaps of their diffraction peaks with those from the other phases present in the sample (15).

Fig. 1 Sequence of x-ray diffractions recorded with increasing temperatures from 1850 to 4050 K at 56 GPa.

Diffraction peaks of the KCl pressure medium are superimposed in blue. Upon heating, (A) we first observed the appearance of weak Mg-Pv diffraction peaks, (B) followed by crystallization of all other phases, including Ca-Pv and stishovite. Above 2930 K, (C) diffraction peaks become suddenly much sharper. Also, all diffraction peaks for a given sample drastically change position on images recorded successively at the same pressure and temperature conditions. Above 4050 K, (D) only a few diffraction peaks remain visible. The quench image (E) contains several new peaks, showing that the lack of diffraction peaks at the highest temperature was not due to a loss of sample, but rather to complete sample melting.

Less than 100 K above the temperature at which Ca-Pv and silica crystallized from the glass, the initially continuous diffraction rings evolved suddenly into a discontinuous juxtaposition of spots (Fig. 1B), and the peak shape sharpened drastically, showing a net discontinuity in the rate of grain growth. At slightly higher temperatures, diffraction images acquired repeatedly at the same temperature, with a time interval of 20 s (acquisition time), showed large diffraction spots in totally different positions, indicating fast grain rotation. These changes in the sample properties were associated with a flattened temperature profile, although laser power was increased continuously (fig. S5). This effect could result from a change in laser absorption due to the structural transformation (20). Our results show that the temperature gap between rapid grain growth and fast grain rotation is less than 200 K (fig. S3). In theory, the extensive grain rotation should take place close to the solidus temperature. The sudden change in sample behavior is most probably due to the appearance of a small amount of liquid at grain boundaries. Because the very first degree of melting can be difficult to detect, we bracketed the solidus using these two major criteria—rapid grain growth and fast grain rotation—where the solidus temperature profile should plot (fig. S3).

The solidus increases continuously from about 2100 (±150) to 3200 (±150) K with an increase in pressure from 20 to 80 GPa (Fig. 2). At the latter pressure, which corresponds to a mantle depth of 1900 km, the MORB solidus is identical to that of a chondritic-type mantle (8). With a further increase in pressure to 135 GPa, the MORB solidus temperature increased to 3800 (±150) K, whereas that of the bulk mantle increased to 4150 (±150) K. The flattening of the MORB solidus could be linked to a change in liquid composition with pressure. Extrapolation of our solidus melting curve to ambient pressure yields a melting temperature of ~1300 K, in agreement with previous studies (21). However, we could not reproduce the rapid increase of solidus temperature at the low pressures observed in previous studies using a large-volume press (21, 22). Because analyses were performed on quenched samples, for which it is difficult to observe low degrees of partial melting, we believe that the solidus temperature was possibly overestimated in these previous studies. Our solidus curve also plots at much lower temperatures than a previous study performed up to 65 GPa, using a laser-heated diamond anvil cell (22). In that study, the melting criterion based on the change of temperature with laser power, for a MORB sample sandwiched between two Re foils, may have hampered detection of the solidus. The maximum discrepancy between the different studies is ~400 K and tends to decrease with increasing pressure above 20 GPa. At 60 GPa, the discrepancy is down to ~250 K between the previous optical (22) and our in situ measurements.

Fig. 2 The MORB solidus and liquidus compared with melting curves of the bulk mantle.

At the CMB boundary, the MORB solidus and liquidus reach 3800 (±150) K and 5050 (±300) K, respectively (fig. S3). The curvature of the MORB solidus is much more pronounced than that of chondritic-type mantle (8), which induces a lower melting point for the former at 135 GPa, the CMB pressure. The MORB solidus also plots 1000 to 1500 K below the melting curves of pure SiO2 and MgSiO3, as determined from shockwave experiments (21, 22, 2426).

To verify that the 350 K melting temperature reduction between MORB and the chondritic-type mantle was not an experimental artifact, we loaded both compositions in the same pressure chamber (fig. S2). Using the same melting criteria as in a previous study of mantle melting (8), we observed the solidus of chondritic mantle at temperatures of 2980 (±150), 3450 (±150), and 3680 (±150) K, for pressures of 58, 85, and 107 GPa, respectively, falling within ±60 K of our previous study (8). Given the higher amount of fusible elements in MORB, one could have expected a lower solidus temperature as compared to chondritic mantle. However, the role of SiO2 excess in MORB and Fp excess in the mantle cannot be neglected (fig. S1).

Upon further increase of the laser power, the sample temperature eventually rose above the plateau temperature, allowing us to record diffraction patterns up to ~5000 K. We rarely observed the total disappearance of all diffraction peaks from the sample. Instead, diffraction images often contained a couple of diffraction peaks appearing at random azimuthal positions on the images (Fig. 1D). Convection in the liquid sample could have induced some crystallization at the sample/KCl interface over a very short time scale. Also, the sample temperature was probably slightly below the liquidus temperature, because we observed a clear rim of solid phase just around the melt. We define the liquidus as the temperature at which complete loss of the continuous structure of the diffraction lines was achieved. This temperature corresponds to a clear loss of the three-dimensional solid structure of the sample, with free rotation of a couple of grains in a predominantly liquid fraction.

The temperature gap between the solidus and liquidus is found to increase progressively from 1500 to 2500 K with increasing pressure from 40 to 140 GPa. This gap is much larger than that reported for the chondritic-type mantle. It results in a liquid composition at the solidus, which is very different from the bulk MORB composition (23).

Three lines of evidence point to an increase of the SiO2 content in the liquid with increasing pressure during partial melting of MORB. First, the MORB solidus curve always plots 1000 to 1500 K below the lowest melting curve for pure CaSiO3, MgSiO3, or SiO2 phases (2426) (Fig. 2). Although SiO2 is the most refractory of these MORB phases at low pressures, it becomes the least refractory above 60 GPa. This suggests a displacement of the pseudo-eutectic composition from Mg-rich liquid (with silica phase on the liquidus) toward Si-rich liquid (with Ca-Pv and/or Mg-Pv on the liquidus) in the (Mg-Pv, Ca-Pv, SiO2) ternary diagram (fig. S1). A second argument is based on the last phase(s) to disappear from the diffraction patterns at temperatures approaching those of the liquidus. At pressures lower than ~70 GPa, clear peaks of SiO2 and CaSiO3 remained visible on the diffraction patterns at temperatures between the solidus and liquidus. At higher pressures, SiO2 peaks disappeared and instead those of Mg-Pv appeared (fig. S6). This indicates a sample undersaturated in SiO2, with Mg-Pv and Ca-Pv grains coexisting with the liquid at the highest pressures.

A third argument comes from mineralogical and chemical analyses of the recovered samples. After the laser was shut down, we maintained the pressure and made diffraction maps of the quenched molten region of the samples. This corresponds to regions of 30 × 30 or 40 × 40 μm2, with step resolutions of 2 or 3 μm. The diffraction maps can be treated in series to extract the phase content as a function of the position within the sample (fig. S7). This type of quenched sample generally shows a rim of liquidus phase(s) around the quenched liquid (13, 27). The mineralogical maps of our samples show clearly that the quenched liquid was depleted in the SiO2 phase and concentrated in Mg-Pv at pressures below ~70 GPa, and the opposite at higher pressures. This indicates an increasing SiO2 content in the liquid with increasing pressure (fig. S1).

We then made chemical maps of our recovered samples, using a scanning electron microscope (SEM) (supplementary materials, Fig. 3, and fig. S8). At pressures below ~70 GPa, we observed a large amount of MgO in the center of the sample where the liquid had solidified, and the liquid was surrounded by a SiO2-rich rim. At intermediate pressures, the maps showed similar MgO and SiO2 content in the liquid as compared to the unheated part of the sample. At the highest pressures, the quenched liquid was enriched in SiO2 whereas the outer rim was enriched in MgO (fig. S8). MORB partial melting in the lower mantle should always produce liquids with (Mg + Fe + Ca)/Si ratios less than 1, because MORB contains free SiO2 in addition to Mg-Pv and Ca-Pv refractory phases.

Fig. 3 Phase relations in the sample recovered after partial melting at 120 GPa.

The chemical maps (Mg and Si) and mineralogical maps [Mg-Pv and seifertite (Seif)] measured by SEM and x-ray diffraction, respectively, show higher SiO2 and seifertite contents and lower MgO and Mg-Pv contents in the sample region corresponding to the quenched liquid.

Following the hypothesis that the ULVZ at the D′′ layer is associated with partial melting, our experimental results provide explanations for the seismological features. First of all, the nonubiquitous nature of the ULVZ could be due to specific regions where MORB slabs have reached the CMB. The thicknesses of the ULVZ (~40 km) and MORB portion of the subducted slabs [~6 km (28)] are of the same order of magnitude. Mantle resistance to the slab penetration, or folding and piling-up at the CMB, could certainly induce thickening of the MORB layer. If temperature in the D′′ region is between 3800 and 4150 K, the MORB should be the only material to undergo partial melting (the harzburgitic layer of the subducted slab being more refractory than the pyrolitic or chondritic-type mantle). Because of the large temperature gap found between the MORB solidus and liquidus, low degrees of partial melting are expected in the D′′ temperature range. If such liquid eventually percolates and diffuses out of the MORB pile, it will react with the excess (MgFe)O Fp present in the surrounding lower mantle to form (Mg,Fe)SiO3 Mg-Pv. For this reason, the mobility of the liquids extracted from the MORB piles must be extremely limited. Thus, it is not likely that large-scale pockets of liquid will form. It would also explain why seismic shear waves (Vs) can propagate through the partially molten ULVZ (5).

Also, the decrease of Fp content in the lowermost mantle, resulting from reaction between the bulk mantle and SiO2-rich liquid, would drive mantle composition toward the perovskitic end-members, in agreement with a recent report (29). The loss of Si from the MORB would drive it toward perovskitic end-members (fig. S1). The reaction would eventually stop when the SiO2 excess in the MORB was exhausted. This would result in the disappearance of the MORB signature, except for minor and trace elements, which could remain concentrated around the ULVZ. This scenario would lead to lower mantle homogenization, in contrast to the chemical segregation generally induced by mantle partial melting. Alternatively, if the MORB proportion is high, Fp could become reactively exhausted, which would imply a lowermost mantle saturated in SiO2. The answer to these two alternative scenarios depends on how well the underlying harzburgite remains attached to the MORB layer and is entrained into the lowermost mantle. If it is sufficiently abundant, then its reaction with the liquid originating from MORB melting should produce a typical mantle composition such as pyrolite.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S9

Tables S1 and S2

References (3034)

References and Notes

  1. Acknowledgments: We thank M. Benbakkar, J. L. Devidal, A. Fabbrizio, G. Garbarino, P. Parisiadis, V. Svitlyk, and F. van Wyk de Vries for their help and three anonymous reviewers for fruitful comments. The MORB starting material was kindly provided by P. Schiano. This work is supported by Institut National des Sciences de l’Univers, l’European Synchrotron Radiation Facility, and the project Oxydeep of the Agence Nationale de la Recherche. This is Laboratory of Excellence ClerVolc contribution no. 99. All experimental data are presented in the supplementary materials.
View Abstract

Stay Connected to Science

Navigate This Article