Seismic Evidence for Partial Melt at the Base of Earth's Mantle

The presence of an intermittent layer at the base of Earth's mantle with a maximum thickness near 40 kilometers and a compressional wave velocity depressed by ∼10 percent compared with that of the overlying mantle is most simply explained as the result of partial melt at this depth. Both the sharp upper boundary of this layer (<10 kilometers wide) and the apparent correlation with deep mantle upwelling are consistent with the presence of liquid in the lowermost mantle, implying that the bottom of the thermal boundary layer at the base of the mantle may lie above its eutectic temperature. Such a partially molten zone would be expected to have enhanced thermal and chemical transport properties and may provide constraints on the geotherm and lateral variations in lowermost mantle temperature or mineralogy.

The presence of an intermittent layer at the base of Earth's mantle with a maximum thickness near 40 kilometers and a compressional wave velocity depressed by -10 percent compared with that of the overlying mantle is most simply explained as the result of partial melt at this depth. Both the sharp upper boundary of this layer (<10 kilometers wide) and the apparent correlation with deep mantle upwelling are consistent with the presence of liquid in the lowermost mantle, implying that the bottom of the thermal boundary layer at the base of the mantle may lie above its eutectic temperature. Such a partially molten zone would be expected to have enhanced thermal and chemical transport properties and may provide constraints on the geotherm and lateral variations in lowermost mantle temperature or mineralogy.
Recent seismic observations show that the base of Earth's mantle has anomalously slow P-wave velocities (1,2). Evidence for a discrete layer is provided by two independent data types: (i) SKS waves that couple with short segments of core-diff-racted P waves (I) and (ii) precursors to the core-reflected PcP phase (2). This layer is within the thermal boundary between the mantle and the core; depending on its origin, it may strongly influence the geomagnetic field, and it likely affected the evolution and differentiation of the deep Earth. The P-wave velocity in this layer is reduced by at least 10% relative to that in the mantle, with its thickness varying between -5 and 40 km (1). Seismological data also imply that the transition between the basal layer and the overlying mantle is (at least locally) sharp [less than 10 km wide (2)]. The layer is thick beneath a region where the lower mantle has large-scale slower-than-average velocities (such as the central Pacific: Fig. 1) and is undetectable below average or faster-than-average lower mantle material (such as the circum-Pacific) (1), implying a relation with hot upwellings in the deep Earth.
Three phenomena could produce this anomalous basal layer: (i) partial melting, (ii) a chemical discontinuity, or (iii) a phase transition in mantle material near a pressure of 135 GPa, corresponding to depths just above the core-mantle boundary (CMB) (3). Although the first two possibilities are not exclusive, the last possibility is unlikely. Both Mg-rich silicate perovskite and Ca-SiO3-perovskite appear to be stable at pressures spanning those present within most of the mantle (4, 5); no significant phase transition is observed in (Mg,Fe)O in this pressure range (6), and the CaCl2 structure of SiO2 is apparently stable to pressures in excess of 130 GPa (7).
Thus, it appears that only a change in chemistry or the presence of melt can account for this reduction in seismic velocity. We first examined the possible origins of a change in melt fraction 40 km above the CMB. The amount of melt that can produce a 10% decrease in P-wave velocity may be estimated from the limited data set on elastic properties of silicate melts and their pressure dependence (8,9), coupled with treatments of the elastic behavior of two-phase aggregates (10,11 ). An upper bound on the effect of an Fe-rich fluid intercalated in the lowermost mantle may be derived if one assumes that the fluid has the properties of the outer core. If the fluid is instead an ultramafic silicate melt, its bulk modulus (K) under CMB conditions is uncertain but is likely to lie close to that of the solid mantle at these depths (8). Coincidentally, the K of core fluid is also close to that of the solid mantle at the CMB [644 GPa versus 653 GPa (12)]. We assumed that the K of the melt is equivalent to that of coexisting solid silicate mantle material. To estimate the size of the velocity change associated with the presence of partial melt, we also inferred the aggregate density. The density change associated with the melting of silicates at high pressures is SCIENCE * VOL. 273 * 13 SEPTEMBER 1996 Earth Sciences Board and Institute of Tectonics, University of California, Santa Cruz, CA 95064, USA. *To whom correspondence should be addressed. ko S * likely to be small [order, 1% (13)] and will thus produce only small variations in the inferred velocities.
The amount of liquid necessary to generate a velocity change depends on the fluid geometry in the solid-liquid aggregate. If the liquid is present in either needle-like or spherical inclusions (10), a 25 to 30% volume fraction of fluid is necessary to generate a 10% reduction in P-wave velocity (Fig. 2). For comparison, films of melt are more effective in reducing seismic velocities; film aspect ratios control the amount of velocity depression (I1). For aspect ratios of 0.01, melt fractions of as little as 5% could account for the observed shift in velocity (Fig. 2); smaller aspect ratios further reduce the necessary melt fraction. Thus, <5 to 30% melt can account for this layer: uncertainties in melt texture produce most of this variation.
The textural distribution of any fluid at the CMB must depend on its composition, which is difficult to assess solely on the basis of seismic data and modeling of partially molten aggregates. The primary distinguishing factor between a silicate melt and an Fe-rich liquid is their relative densities; both have similar bulk moduli at CMB conditions. Because of the higher density of an Fe-rich liquid, only about 20% intercalated melt of core chemistry can produce this velocity anomaly if the liquid is present in tubules or spheres. However, there have been suggestions that, in contrast to its behavior at lower pressure, liquid Fe may wet grain boundaries of silicates at high pressures (14,15). If an Fe-rich fluid were present in grain-boundary wetting films with aspect ratios near 0.01, the small volume fraction of liquid (and thus a reduced effect on density) would produce an inferred liquid fraction similar to that of a silicate. However, the primary difficulty associated with an Fe-rich (or core-like) fluid may be dynamical in character: the high density, probable low viscosity, and likely wetting behavior (14) will probably cause such a liquid to descend into the outer core. If partial melting does produce this anomaly, the magnitude of S-wave velocity perturbations within this zone should be diagnostic. Anomalies in S-wave velocity are roughly independent of melt geometries and would correspond to -30% decreases (Fig.  3) for the inferred melt fractions of Fig. 2. Both seismic data sets used to image the low-velocity basal layer depend on P-wave velocity structure (1, 2); characterizing the magnitude of the shear velocity reduction depends on analysis of waveforms sensitive to S-wave velocity structure at these depths, such as the core-reflected ScP phase.
It is likely that chemical reactions such (1) occur at the CMB (16). Here, x is related to the amount of Fe present in reacting mantle material, and s is the amount of SiO2 generated by the reaction. By volume, we expect from Eq. 1 that the first 4% of intruding core fluid will react with perovskite to form Fedepleted silicate reaction products, with coexisting FeO and FeSi. It has been proposed that reaction products that are dominated by an Fe-depleted silicate perovskite have higher seismic velocities than the unreacted assemblage, with intruding (and reacting) Fe fluid initially augmenting seismic velocities (15). We estimated shifts in P-wave velocity for solid reaction products relative to reactants of differing Fe and Si contents (Fig. 4) by (i) using known elastic data for the phases in Eq. 1 (17), (ii) assuming that Poisson's ratio is constant from ambient conditions to those of the CMB, and (iii) calculating the relative velocity anomaly associated with CMB reac- tions for different amounts of added Fe (18). For plausible lower-mantle chemistries, the effect of such reactions on velocity (a maximum of 3 to 4%, with broad uncertainties) is smaller than the minimum 10% velocity depression needed to account for the seismic observations. A chemical effect that could generate this feature without the presence of liquid is the presence of solidified core or Fe-rich alloy in this region; such material should have a lower seismic velocity than normal mantle. The amount of such material required to produce this anomaly hinges critically on the value of the shear modulus of core alloy at the CMB. Using a probable upper bound of 170 GPa for the shear modulus of Fe under these conditions [derived from the difference between the estimated bulk sound velocity and observed sound velocity of solid Fe at 135 GPa (19)], we estimate that an upper bound of 70 to 75% solidified core alloy could produce this velocity anomaly. However, if core material just above the CMB is close to its liquidus, the shear modulus could be small and the actual amount of solid alloy that could generate this anomaly would lie between this upper bound and our inferred possible abundance of core fluid of -20%. Such entrainment of solid core material in the lowermost mantle would imply either (i) that a complex velocity structure is found in this layer, with entrained core material liquefying as the CMB is approached,  Fig. 4. Ratio of the P-wave velocity for assemblages that have undergone chemical reactions with variable amounts of Fe to that of the unreacted assemblage. The mole percent of Mg in perovskite (Pv) and magnesiowustite (Mw) in the different unreacted assemblages are given in parentheses, and we derived the Fe/Mg ratios of these phases using the Fe-Mg partition coefficients of (22). The differing lengths of the lines is an indication of the maximum amount of Fe that can chemically react with the different assemblages, as given by Eq. 1. Because perovskite volumetrically dominates both the reacted and unreacted assemblages and no compositional dependence of the elastic properties (other than density) of perovskite has been observed (4,5), the primary error in this ratio is produced by uncertainties in the elastic properties of FeO, FeSi, and SiO2; the error bar reflects uncertainties in the bulk moduli and densities of these phases under CMB conditions.  (20), as well as Fig. 1, document the region's complexity. The latter alternative is unlikely (unless there is either a significant thermal boundary at the top side of the core or suspended solids throughout the outer core) because the outer core adiabat is shallower than the liquidus of the core alloy (21). If this low-velocity layer is produced by melting, then the most plausible means for generating an abrupt onset of the partially molten zone is if the geotherm intersects the eutectic temperature of the mantle. To generate up to 30% partial melting, the eutectic could lie neither near the mantle composition at these depths nor near the composition of silicate perovskite (probably the most abundant phase under these conditions) because significantly higher melt fractions would result. With this amount of melting, magnesiowustite could lie near the eutectic and melt out of the solid assemblage over a narrow depth interval. Thus, the amplitude of this feature could produce a constraint on the mineralogy of the deep mantle; however, the topology of the multicomponent phase diagram of the mantle under CMB conditions is poorly constrained.
If this velocity anomaly is produced by the intersection of the geotherm with the solidus of silicate mantle material, then we would anticipate that this feature would be global in character but varying in its thickness, with lateral variations of the geotherm in the lowermost mantle. It is possible that, outside of the central Pacific and beneath part of the Atlantic (1), such a layer could be less than a few kilometers thick, rendering it difficult to image seismically. Because the outer surface of the core is essentially isothermal, it is possible that the topography of this basal layer could be produced by variable thermal gradients in the lowermost mantle, with thickening produced by local elevation of isotherms. The existence of such a partially molten layer should profoundly decrease the viscosity of the lowermost mantle, thus altering the heat transport out of the core, the stability of the thermal boundary layer at the base of the mantle, and potentially the flow regime in the lower mantle. Moreover, a partially molten layer is likely to have a higher electrical conductivity than unmelted mantle; this property could complicate the propagation of the geomagnetic field through this region of the planet.