Introduction

A variety of evidence suggests that about 7% of the mantle may be eclogite. The question is, how is it distributed and where is it stored? Ringwood et al. (1994) and Hirose et al. (1999) argue that recycled oceanic crust eventually sinks into the lower mantle. Anderson (1979, 1987, 1989a,b) argued that it is trapped in the TZ, in the depth interval 410 – 650 km, which comprises about 11% of the mantle. The depth of trapping depends on the compositions and densities adopted for the eclogite and the adjacent mantle, and the composition of the lower mantle. If one adopts MORB as the protolith of eclogite it will be particularly dense because of its high SiO₂ (stishovite) content compared to other mafic protoliths, such as cumulates, restites, delaminated continental crust and the average composition of oceanic crust. On the other hand pyrolite is also not a completely satisfactory model for the TZ and the lower mantle (Cammarano et al., 2005; Duffy & Anderson, 1989, Lee et al., 2004). The density of eclogite relative to pyrolite may not be a pertinent guide to the fate of eclogite.

Long-lived reservoirs in the mantle must be either denser than the overlying layers, or highly viscous blobs, advected but resisting stretching and mixing. There are various options for bringing this material to the surface. If the material has a low melting point, such as eclogite, it need not be particularly hot, or embedded in a thermal boundary layer (TBL) in order to become buoyant. It just needs to heat up to near ambient temperatures (see Reheated slabs pages). Because of the effect of pressure on melting point, a shallow, e.g., upper mantle, location is favored for such a mechanism to operate. There is now general agreement there is a subduction barrier to eclogite between about 650 km (or even shallower) and 720 km (or even deeper). The window is even broader for Si-poor eclogites such as cumulates and many non-MORB protoliths. If the lower mantle is denser than pyrolite (e.g., Anderson, 1989b; Lee et al., 2004) then eclogites may be trapped in or above the TZ until they melt. Recycling and remelting of large blocks of eclogite is distinct from the thermal plume hypothesis and distinct from the pyroxenite vein hypothesis for increasing fertility. Thermal plumes originate in TBLs and require heating from below, and they are not an intrinsic part of plate tectonics.

The subduction of oceanic lithosphere into Earth's interior is thought to drive convection and create chemical heterogeneity in the mantle. The delamination of lower continental crust also affects the dynamics and composition of the mantle. The fates of basaltic crust, of the underlying peridotite layer, and of delaminated crust may differ because of differences in age, temperature, melting temperature, chemistry and density. It has been suggested that subducted crust (eclogite), although denser than ambient mantle at shallow depths, may in fact become buoyant somewhere between 560 km and the 650-km discontinuity, and therefore might be gravitationally trapped in the TZ to form an eclogite-rich or garnetite layer (Anderson, 1989b).

What might a chemically stratified mantle look like? The density and shear-wave seismic velocity of crustal and mantle minerals and rocks at standard temperature and pressure are arranged approximately according to increasing density in Figure 1. This approximates the situation in an ideally chemically stratified mantle. P and T effects may change the ordering and the velocity and density jumps. Eclogite can settle to various levels, depending on composition; the eclogite bodies that can sink to greater depths because of their density have low seismic velocities compared to other rocks with similar density. Note that MORB-eclogite contains stishovite at high pressure and may sink below about 500 km. Eclogite has a much lower melting point than peridotites and will eventually heat up and rise, even if it does not lie in a TBL, creating a sort of “yo-yo tectonics”. The ilmenite form of garnet and enstatite is stable at low temperature but will convert to more buoyant phases as it warms up. Velocity decreases do not necessarily imply hot mantle. Low-velocity zones (LVZs) have been found by seismology at various depths above 720 km. Depths of prominent seismic reflectors are noted.

Figure 1: Rocks and minerals arranged mostly in order of increasing density.

Thermochemical convection

Two-dimensional simulations of simple mantle convection suggest that for whole-mantle convection the mantle should be homogenized in a time-scale of less than 1 Gyr, although unmixed patches may remain. The rate at which blobs of inserted material are stretched and assimilated into the flow depends on their relative viscosity; viscous blobs may survive intact for many mantle overturns. All such calculations assume vigorous convective stirring. If the mantle is layered or if pressure effects on physical properties are considered, convection is much less turbulent or chaotic. It is more probable that subducted and delaminated materials settle to equilibrium depths without being entrained into mantle flow. In fact, if mantle flow is mainly a result of sinking slabs and surface plates, then there will be little mixing.

Problems with, and alternatives to, efficient convective mixing are discussed by Meibom & Anderson (2003). They showed that a spectrum of basalt types can be generated simply by sampling, in different ways, a statistically heterogeneous mantle. Plate tectonic processes and recycling can give this kind of heterogeneity but can also create a chemically stratified mantle. Since the products of mantle melting and differentiation are so disparate in density, they are likely to settle to different depths. If eclogite, for example, is denser than upper mantle rocks, but intrinsically less dense than lower mantle lithologies, then it would settle into the TZ. However, because of its complex phase relations with temperature, low melting point, and possibly high radioactivity, it can change both its density and relative position over time. Although eclogite is usually considered to be a dense rock, Figure 1 shows that it can have similar densities to various peridotite minerals and assemblages that are believed to make up the mantle between 100 and 400 km. Cold eclogite may sink as far as the gamma-spinel phase change near 500 km depth. Quartz-rich protoliths such as unaltered MORB may sink deeper because of the high density of stishovite (Hirose et al., 1999). Eclogite is just one example; various peridotites and magmas differ in intrinsic density by amounts more than can be overcome by normal temperature variations, and should therefore concentrate at various levels. Even the crust and continental lithosphere appear to stratified by intrinsic density.

Eclogite in the mantle

Mantle eclogites have a variety of origins. Eclogites may be produced by high-pressure crystallization, they may be remnants of subducted oceanic crust, or the residue of partial melting of such crust. Some eclogites may be products of metamorphism of mafic lower continental crust, i.e., gabbroic to anorthositic protoliths and some may be high-pressure garnet-pyroxene cumulates or low-pressure plagioclase-pyroxene-olivine cumulates. High MgO eclogites may represent the lower parts of subducted oceanic crust or foundered mafic lower continental crust.

The main minerals in eclogite at low pressure are clinopyroxene and garnet. Accessory phases can be rutile, quartz, kyanite and orthopyroxene, depending on the starting composition. Below about 100 km depth excess silica converts to coesite, then stishovite, and the kyanite disappears. At TZ pressures the clinopyroxene dissolves in garnet, forming majorite; eclogite at the top of the TZ comprises garnetite+stishovite. At very high pressures calcium perovskite becomes stable. Hirose et al. (1999) report the phase relations and melting temperatures of a moderately differentiated MORB sample from the north Atlantic at pressures up to 64 GPa (equivalent to 1,500 km depth). This sample contains stishovite and Ca-perovskite at pressures greater than 24 GPa (equivalent to depths greater than 720 km), and is predicted to become denser than pyrolite at that depth. Stishovite has a density of 4.3 g/cm³ and shear seismic velocity of 7.1 km/s, much greater than even the in-situ values of the mantle down to depths of about 800 km. Hirose et al. (1999) predict that if subducted crust could penetrate the buoyancy barrier between 650 and 720 km depths (in a pyrolite mantle) it could sink further into the lower mantle. This assumes that the top part of the oceanic crust is representative of subducted crust, that it is unmodified as it sinks into the mantle (i.e. it stays unaltered and does not lose a silica-rich melt), and that some mechanism exists for dragging it through higher-density mantle. It also assumes that pyrolite is a good model for the TZ and lower mantle. Actually, Lee et al. (2004) found that the lower mantle is 2 – 4% denser that pyrolite, which would essentially reverse the conclusion of Hirose et al. (1999) regarding the ability of eclogite to sink into the lower mantle. Cammarano et al. (2005) determined that a pyrolite lithology has higher seismic velocities in the TZ and below, than are obtained from seismology. This suggests a more garnetitic mineralogy in the TZ than pyrolite and, perhaps, more FeO. The seismic data also suggest a gradient with depth in the TZ, with, possibly, more high-pressure olivine phases at the bottom. The TZ also appears to vary laterally (Ishii & Tromp, 2004).

The lower mantle

The controversy

There has been much controversy over the composition of the lower mantle. Ringwood (1975, 1994) and Kesson et al. (1994, 1998) advocated a chemically uniform pyrolite mantle with Mg/Si of about 1.5. This composition, however, has a density deficit with respect to geophysical mantle models. Others have shown that the geophysical data are consistent with a pyroxene-rich or an iron-rich lower mantle (e.g., Anderson, 1977; Butler & Anderson, 1978; Stixrude et al., 1992; Stacey & Isaak, 2000; Lee et al., 2004). In an attempt to resolve the density discrepancy, Ringwood (1975) suggested that enstatite might transform to MgSiO₃ (perovskite) with a density 3 – 7% greater than the isochemical mixed oxides. This transformation was subsequently found, but the density increase was only 2% relative to the mixed oxides at zero pressure and negligible at high pressure. The pyrolite density discrepancy is pertinent to the question of whether eclogite will sink into the lower mantle. Anderson (1989b) argued that it would not, using geophysical estimates of lower mantle density. Ringwood (1975,1994) argued that it would, using a theoretical pyrolite density for the lower mantle. Others have argued for a very cold lower mantle, assuming that this could raise the density enough for pyrolite to satisfy the geophysical data.

Constraints from 1D models

Cammarano et al. (2005) tested the compatibility of a constant pyrolite composition for the mantle, including the effects of phase changes. The pyrolite model has seismic velocities that are too low above 400 km, a velocity jump that is too large at 410 km, TZ gradients that are too low, a velocity jump at 650 km that is too small, and too strong a gradient below the discontinuity. It appears to be difficult to reconcile a pyrolitic mantle with global seismic data for the TZ, given current mineral physics constraints. Similarly, the geophysical data below 650 km are not consistent with a chemically uniform mantle; a pyrolite composition requires unreasonably low deep-mantle temperatures.

There is some trade-off between temperature and composition. If a simple mixture of perovskite and magnesiowüstite is assumed, the total iron content of the lower mantle is found to be much greater than the upper mantle, and Mg/(Fe + Mg) = 0.78 independent of temperature (Stacey & Isaak, 2000). Those authors considered this to be an implausibly high iron content and attributed it to neglect of the presence of Ca perovskite. There is no cosmochemical, petrological, geochemical or geophysical prohibition against such iron contents, except that this would preclude whole mantle convection. They concluded that Ca-perovskite, which they did not treat, must be an important constituent of the lower mantle and that it is seismologically conspicuous. Others have argued that CaSiO₃- perovskite is seismically invisible, having similar properties to Mg-perovskite or the lower mantle, or that there is little CaO or Al₂O₃ in the lower mantle because of accretional differentiation (Anderson, 1989a). Lee et al. (2004) used a more realistic composition, which included pyrolitic portions of calcium. They determined the high-pressure mineral assemblage of an undepleted natural peridotite – thought to be representative of the Earth’s upper-mantle – using high-resolution X-ray diffraction. The measured room-temperature bulk properties of this high-pressure assemblage, together with a range of estimates of thermal properties of the constituent minerals, appear to be inconsistent with seismological constraints on the density and bulk modulus of the lower mantle. Their results suggest that the lower mantle differs in bulk composition (e.g., is richer in iron) from current estimates for the upper mantle.

One way of satisfying the observed properties of the lower mantle with a pyrolite-like composition is to invoke a higher iron content than the preferred value for the upper mantle (Mg# = 0.90). The density deficit of ~2-4% obtained can then be explained by an increased abundance of iron, to an Mg# ~ 0.85. The result is non-unique; there are tradeoffs between temperature and composition (e.g., Si, Al and Ca abundances) for satisfying the observed properties of the lower mantle. Nevertheless, the intrinsic density difference between such an iron-enriched composition and pyrolite-like estimates of upper-mantle bulk composition is sufficient to stabilize layered convection. An increasing FeO content with depth in the lower mantle is also indicated.

A different lower mantle composition results if one assumes that the Earth differentiated during accretion, and that the lower mantle is the refractory residue after removal of crustal and upper mantle materials, including most of the basaltic elements (Anderson, 1989a). The depleted refractory residue is high in Si and low in Ca and Al. A lower mantle with a Mg/Si ratio, 1.07, or 20% lower than pyrolite, and a Fe/Mg ratio 20% higher than pyrolite satisfies the density and elastic properties of the lower mantle (Kanani Lee, personal communication, March, 2005). An FeO-rich lower mantle is also consistent with an enstatite chondrite protolith for the Earth.

Tomographic Constraints (Lateral Variations)

Ishii & Tromp (2004) determined that density variations have a negative or nearly zero correlation with shear- and compressional-wave seismic velocity variations in the TZ where the root-mean-square density amplitude is, however, high. They also showed that the TZ velocity and velocity-density correlations are completely different from the overlying and underlying regions. This is inconsistent with thermal variations or with whole-mantle convection. Subduction of cold eclogite into the TZ, however, will lower the shear-wave seismic velocity there (Figure 1), but will have little effect on the density (the eclogite displaces similar density material). An eclogite-rich TZ may also explain the velocity jumps at 410 km and 650 km which are too small and too large, respectively, to be entirely due to phase changes in pyrolite (Cammarano et al., 2005). Garnet does not undergo a phase change near 400 km so it dilutes the jumps created by phase transitions in olivine and orthopyroxene at depths of 400 and 500 km.

Gu et al. (2001) (see also Mantle convection page) showed that, near 650 km depth, there is a distinct change in seismic velocity variations, with large variations above and small variations below. The low shear-wave seismic velocity of eclogite and its low melting temperature are expected to create large tomographic variations, particularly since its distribution is not expected to be uniform.

Slab penetration?

What is one to make of the general consensus in the non-seismological community that both slabs and plumes have been imaged from the surface of the Earth to the core-mantle boundary? One way to approach this, for a non-specialist, is to look at a large number of tomographic maps at different depths and a large number of more-or-less randomly oriented cross-sections that have not been cropped or color-saturated to make a particular point. See, for example, Ritsema (2005) and the supplementary figures that accompany this paper. Many of these cross-sections cross slabs and proposed sites of deep mantle plumes but they usually show little continuity below 650 km. The impression one gets from these cross-sections is quite different from that imparted by inspection only of the relatively small number of widely reproduced ones that apparently show deep slab penetration.

It is useful to know how tomography works (see also Seismology: The hunt for plumes). In most body-wave studies the anomaly found at a given seismic station is initially distributed uniformly along the ray, with the expectation that, with the addition of enough data, the anomaly can be better localized. For a set of earthquakes in a given region, recorded at a local array of seismic stations, the ray bundle will be shaped like a banana, narrow at both ends and wide in the middle where the rays reach their deepest points. If the earthquakes are in a cold slab and the stations on a stable continent, then the banana will be fast – a blue banana. If the earthquakes are on midocean ridges, or in tectonically active continental regions, and recorded on stations in the latter areas – where most seismometers are – or on oceanic island stations, the banana will be slow – a red banana. If there are few rays that cross the banana from other directions, then the banana will retain the color imparted to it from the majority of the rays. The bananas show up best in cross-sections that include the bulk of the stations and sources. Other cross-sections will tend to show blobs and it is not so obvious that they are the result of smearing along ray bundles.

For example, there are many rays from earthquakes in South American subduction zones to stations on the Canadian Shield that sample the midmantle under the Americas and the western Atlantic but few rays from the Pacific or the Atlantic to help cancel out the upper mantle (both velocity and anisotropy) effects. Likewise, most of the data for ocean-island stations are from steeply incident rays from distant earthquakes. Since we know from surface-wave studies that most of the oceanic mantle down to 600 km is low-velocity, it will be no surprise that ocean-island stations appear to be associated with low-velocity anomalies. Randomly placed ocean-bottom seismometers would also be expected to yield anomalies that resemble low-velocity, nearly vertical bananas. However, it is islands where most of the stations are located. Finding ways to cancel out the effects at the ends of the bananas, and to localize the deep mantle parts of the anomalies is an ongoing challenge to seismologists. Using surface waves, and surface bounces such as PP and SS helps. Methods have yet to be developed for assessing artifacts and for judging which features of tomographic models are believable. Forward tests of complex synthetic models are useful, i.e. datasets are generated for known models, which are then inverted using current techniques.

Thermal constraints

In a chemically stratified mantle, with periodic injections of cold slabs and warm delaminated continental crust, the temperature gradient will be complex; it will not be a simple adiabat. In particular, deep mantle temperatures will be higher than in a homogenous convecting mantle. Mattern et al. (2005) found that very low mantle temperatures are required for a uniform pyrolite composition mantle. Inferred temperatures are up to 600 K hotter for models based on cosmic abundances. They inferred a subadiatic temperature gradient from 660 to 1300 km that correlates with a decreasing iron content. They found no clear indication for a deeper, chemically distinct layer or compositional boundary as proposed by Kellogg et al. (1999) and van der Hilst & Karason (1999). In the region from 800 to 1300 km, the mantle is significantly heterogeneous. Between 1300 km and 2000 km depth the gradients of seismic velocity are consistent with a homogeneous and adiabatic mantle.

Discussion

Most continental flood basalts occur in sutures, old orogenic belts and adjacent to thick cratonic lithosphere, a. They form as continents either split up or are in the final stages of convergence. I suggest that these regions may be prone to deep crustal removal [e.g., Kay & Kay, 1993; see also Suzanne Kay's home page]. The upwelling of asthensosphere into the gap creates an initial pulse of magmatism, as has been widely discussed. What has not been discussed is the fate of the delaminated crustal material. What would be evidence for delaminated material in the mantle? First of all, it originated in the middle of a thermal boundary layer and is not particularly cold. Second, eclogite has higher densities but lower seismic velocities than other upper mantle rocks, so it would show up as a low-velocity feature in tomographic images, even while cold. Third, eclogite is fertile, and has a lower melting point than "normal" mantle. In summary, it does not look like a slab seismically, and will not act as a slab.

There is a variety of evidence supporting the view that the outer 1000 km of the mantle is heterogeneous, both radially and laterally. Some of this may be due to a non-uniform distribution of eclogite and peridotite. The idea that eclogite may be trapped in and above the TZ is attractive. Trapped slab components warm up by conduction from the surrounding mantle and their own radioactivity (see Reheated slabs pages). If temperatures reach the solidus of eclogite, melting and buoyant ascent can initiate and further buoyant decompression melting occurs as the material rises. Some eclogites may equilibrate at depths above the TZ (Figure 1). Wherever it exists, eclogite will create fertile patches with low melting point, and its presence will eliminate the need for excess temperature to explain melting anomalies and low-velocity zones. There is also no longer any need to involve the deep mantle to explain melting anomalies. The rationales for the plume hypothesis include the view that the upper mantle is entirely homogeneous and well stirred and that only high temperature can create melting and low-velocity zones; this need not be the case.

Delaminated lower lithosphere [Ed: Click here for explanation of lithospheric delamination] is expected to be refractory, chemically buoyant and only dense while cold. The lower crust, on the other hand, becomes intrinsically denser than the surrounding mantle and is also fertile. Since the liquidus of eclogite occurs at about about the same temperature as the solidus of peridotite, I expect that mantle eclogite blobs can become almost entirely molten as they warm up to ambient mantle temperatures. They become buoyant when about half the garnet is consumed, however. Deep low-velocity zones in the mantle may be due to sinking or neutrally buoyant eclogite rather than hot upwellings. Lower crustal delamination (e.g., Lustrino, 2005) may explain the low-velocity zones in the upper mantle in places such as the Ontong Java Plateau, Yellowstone-Snake River Plain, and the Parana and the deep low-velocity zones found atop the 410- and 650-km discontinuities. The low-velocity features found beneath some hotspots may be due to the re-emergence of these fertile blobs.