Highly siderophile elements were stripped from Earth’s mantle by iron sulfide segregation

See allHide authors and affiliations

Science  09 Sep 2016:
Vol. 353, Issue 6304, pp. 1141-1144
DOI: 10.1126/science.aaf6919


Highly siderophile elements (HSEs) are strongly depleted in the bulk silicate Earth (BSE) but are present in near-chondritic relative abundances. The conventional explanation is that the HSEs were stripped from the mantle by the segregation of metal during core formation but were added back in near-chondritic proportions by late accretion, after core formation had ceased. Here we show that metal-silicate equilibration and segregation during Earth’s core formation actually increased HSE mantle concentrations because HSE partition coefficients are relatively low at the high pressures of core formation within Earth. The pervasive exsolution and segregation of iron sulfide liquid from silicate liquid (the “Hadean matte”) stripped magma oceans of HSEs during cooling and crystallization, before late accretion, and resulted in slightly suprachondritic palladium/iridium and ruthenium/iridium ratios.

The formation of Earth’s metallic core resulted from the segregation of liquid iron from silicates during accretion. This process partially removed siderophile (metal-loving) elements from the mantle by transporting them into the core. Moderately siderophile elements (MSEs), such as Ni, Co, W, and Mo, are variably depleted in the bulk silicate Earth (BSE) as a consequence of metal-silicate equilibration, because of their differing metal-silicate partition coefficients (1, 2). In contrast, the highly siderophile elements (HSEs; Re, Os, Ir, Ru, Rh, Pt, Pd, and Au) are present in near-chondritic relative abundances, even though their metal-silicate partition coefficients (measured over the pressure range 0 to 18 GPa) vary by orders of magnitude (3). This has led to the widely accepted hypothesis that the HSEs were stripped from the mantle by metal-silicate segregation and that the present concentrations were added by the late accretion of chondritic material after core formation had ceased (46). The mass of late-accreted material, as estimated from HSE concentrations, has also been used to determine the age of the Moon (6).

Unlike simple geochemical models of core formation, which unrealistically treat core-mantle equilibration as a single event (7), we modeled core formation as a multistage process because metal was delivered to Earth by accreting bodies throughout its accretion history. We followed the approach of (8), in which evolving MSE abundances in Earth’s mantle and core were modeled by integrating the dynamics of planetary accretion with the chemistry of core-mantle differentiation. In this approach, Earth’s growth history comes from state-of-the-art N-body accretion simulations based on the “Grand Tack” scenario (912), which start with a protoplanetary disk consisting of 80 to 220 roughly Mars-sized embryos and several thousand smaller planetesimals distributed initially over heliocentric distances of 0.7 to 10 astronomical units (AU). For the HSE modeling presented here, the results are not dependent on the choice of the Grand Tack scenario because planets grow through embryo-embryo and embryo-planetesimal collisions in all accretion scenarios. Each collision is an accretion event, which delivers mass and energy to the growing planets, resulting in melting, magma ocean formation, and an episode of core formation. Embryos and most planetesimals are assumed to have undergone early core-mantle differentiation. Unlike previous core formation models, the metal of projectile cores equilibrates with only a fraction of the target’s mantle, which is determined from a hydrodynamic model (12, 13). The compositions of metal and silicate produced in each core formation event are determined by a mass balance–element partitioning approach (8, 14). Five parameters are fit by least squares minimization so that the composition of the model Earth’s mantle matches that of the BSE (8). Metal-silicate equilibration pressures are fit assuming that they are a constant proportion (refined by least squares to be ~0.7) of the target’s core-mantle boundary pressure (PCMB) at the time of each impact, which on average is consistent with calculations of impact-induced melting during Earth’s accretion (15). A heliocentric oxidation gradient model defines the bulk compositions of all starting bodies and is defined by four of the five fitting parameters (8) (fig. S1A).

We investigated the evolution of mantle Ir, Pt, Pd, and Ru concentrations during Earth’s accretion and differentiation using our accretion–core formation model (8). Metal-silicate partition coefficients for the HSEs decrease with increasing pressure and temperature (P-T) conditions (3). In addition, we have shown experimentally that increasing the sulfur content in the metal has a similar effect (12, 16). It is therefore essential to include sulfur in the initial bulk compositions of accreting bodies. Sulfur is a volatile element with a low 50% condensation temperature of 655 K at 10−4 bar (17). We therefore assumed that concentrations of S increased systematically with decreasing temperature and increasing heliocentric distance (18). We assumed that, before giant planet migration, fully oxidized bodies that formed beyond Jupiter and Saturn (>6 AU) contained the full complement of S [corresponding to 5.35 weight % (wt %) in a CI chondrite composition], with concentrations decreasing along a linear gradient toward the Sun (fig. S1B). We adjusted the heliocentric distance at which the S concentration becomes zero to 0.8 AU in order to obtain Earth’s bulk sulfur content (0.64 wt %). Although there are potential problems with this simple concentration-distance model, our main results do not depend on it (12). We used a partitioning model to determine the distribution of S between metallic and silicate liquids during each metal-silicate equilibration event (12, 19). Our ultimate objective was to obtain 1.7 to 2.0 wt % S in Earth’s core (20) and 200 to 250 parts per million (ppm) in the mantle (21) at the end of accretion.

Using high-pressure metal-silicate partition coefficients (3, 16), we considered the effects of metal-silicate equilibration and segregation on the evolution of Pt, Ru, Pd, and Ir concentrations by including the modeled S abundances in the metallic liquid (table S1). The final HSE concentrations were high and variably fractionated because of differing partition coefficients (3, 16), with the result that relative abundances in the mantle were strongly nonchondritic (Fig. 1A). Pd and Pt concentrations started to become especially high after the model Earth had accreted ~60% of its mass and increased to strongly exceed BSE values by the end of accretion. Contrary to the conclusions of previous studies, accreted metal in these growth models actually added HSEs to Earth’s mantle rather than removing them. In the case of differentiated planetesimals that underwent early [e.g., <3 million years (My)] core-mantle differentiation as a result of heating caused by the rapid decay of 26Al (22), HSE partition coefficients were extremely high (106 to 1011) (3) at the low P-T conditions of planetesimal differentiation (≤0.3 GPa and ≤1900 K). This means that the HSEs partitioned almost entirely into the metallic cores of planetesimals, and to a lesser extent into embryo cores, during differentiation. In contrast, at the high P-T conditions of metal-silicate equilibration after Earth had accreted ~60% of its mass, HSE partition coefficients were lower by two to five orders of magnitude than under the conditions of planetesimal differentiation (3). This resulted in HSEs in impactor cores being transferred to Earth’s mantle by metal-silicate equilibration so that mantle concentrations increased to exceed BSE values (Fig. 1A). High HSE abundances also resulted from the accretion of fully oxidized bodies and from the oxidation of accreted metal (delivered as small planetesimal cores and as dispersed metal in undifferentiated bodies) by water in the magma ocean (8, 12).

Fig. 1 Evolution of HSE and sulfur concentrations in the mantle during Earth’s accretion, based on metal-silicate equilibration and segregation.

Shown are mantle concentrations of (A) HSEs and (B) sulfur over time. Each symbol represents an impact, and “mass accreted” is the accumulated mass after each impact, normalized to Earth’s current mass (Me). The final giant impact, at 113 My, increases Earth’s mass from 0.872 to 0.997 Me. BSE abundances (21) are shown by dashed lines in (A) and the gray bar in (B). Error bars, based on the propagation of uncertainties in the partitioning parameters, are shown for the final Pt and Pd concentrations (A); propagated uncertainties for Ru and Ir are ±0.6 and ±1.7 parts per billion (ppb), respectively. Results obtained when only 50% of each batch of accreted metal equilibrates with silicate, as a result of incomplete emulsification of impacting cores (23), are presented in fig. S2. Although the oblique lines connecting the symbols show the general trends of mass and composition with time, they do not accurately represent the evolution paths, which in reality always involve a series of vertical steps.

The results shown in Fig. 1A are based on the assumption that 100% of the accreted metal equilibrated with the silicate liquid. If only a limited fraction of metal equilibrated because of incomplete emulsification (23), concentrations of all four HSEs would become even higher (fig. S2A); this is because higher equilibration pressures would then be required to reproduce the MSE concentrations of Earth’s mantle, causing further reductions in the HSE metal-silicate partition coefficients.

The final calculated mantle S content exceeded the BSE concentration by a factor of ∼30 (Fig. 1B) and, in addition, the S content of the core was very low (0.36 wt %). The high mantle concentrations that developed were not removed by metal-silicate equilibration and segregation (Fig. 1B) (12). To achieve a low BSE concentration (<200 ppm before late accretion) and a core concentration of 1.7 to 2.0 wt %, FeS liquid must be exsolved and segregated to the core, an event that has been termed the “Hadean matte” (24, 25).

Sulfide liquid exsolves from a magma ocean when a solubility saturation level, termed the sulfur concentration at sulfide saturation or SCSS, has been exceeded (26). We have experimentally determined the SCSS for peridotite liquid at 7 to 21 GPa and 2373 to 2673 K (16), thus enabling concentrations in S-saturated magma oceans to be estimated by extrapolation usingEmbedded Image (1)where SCSS is in parts per million, T is in kelvin, and P is in gigapascals. Average values along magma ocean adiabats are much higher than the concentrations shown in Fig. 1B (Fig. 2). However, SCSS decreases strongly with decreasing temperature and becomes much lower close to the peridotite melting curve (Fig. 2). Thus, droplets of immiscible FeS liquid exsolve from the silicate melt structure in a deep S-bearing magma ocean as it cools toward crystallization temperatures. This effect is enhanced as the melt fraction is reduced by crystallization; because of its high density, the exsolved FeS segregates by sinking to the core (12).

Fig. 2 Sulfur concentrations at sulfide saturation (SCSS) in peridotite liquid, as a function of pressure.

Equation 1 (16) has been used to calculate concentrations at temperatures about midway between the peridotite liquidus and solidus (8) and along adiabatic temperature profiles for magma oceans with basal pressures of 20 and 80 GPa.

We considered two end-member scenarios when modeling sulfide segregation, depending on magma oceans being either short-lived (27) or long-lived (28): (i) Sulfide segregation took place in multiple stages and occurred after each giant impact, provided SCSS was exceeded close to the peridotite melting curve, and (ii) FeS segregation took place in a single stage and occurred only after the final giant impact and before late accretion. The final results from these two scenarios are almost identical, although the evolutionary paths are different.

The strong P-T and depth dependences of SCSS in a deep convecting magma ocean (Fig. 2) require a simple modeling approach because of the challenges in determining the amount of FeS liquid that exsolves and equilibrates chemically in such conditions (29). We thus defined empirically an effective pressure Peq-S that describes the SCSS (using Eq. 1) and equilibration pressure for the entire magma ocean, assuming that the corresponding temperature lies between the liquidus and solidus of peridotite (8)Peq-S = kS × PCMB(2)Here PCMB is the core-mantle boundary pressure at the time of each FeS exsolution and segregation event, and kS is a constant so that Peq-S increases as Earth accretes. The amount of FeS that exsolves is the excess that is present above the SCSS value calculated using Eqs. 1 and 2.

Because HSEs dissolve in S-bearing silicate melts as HSE-S species, they are fractionated into sulfide liquid that exsolves from a magma ocean (30). We modeled the effect of segregating FeS liquid on mantle HSE concentrations, using our experimental data on the partitioning of Pt, Pd, Ru, and Ir between FeS and peridotite liquids [obtained at 7 to 21 GPa and 2373 to 2673 K (16)] (table S2). We assumed that equilibration between sulfide and silicate liquids occurs simultaneously with sulfide exsolution at pressure Peq-S (Eq. 2). Because of high convection velocities (several meters per second), combined with a time scale of at least 1000 years to cool to crystallization temperatures (27, 29), sulfur should be well mixed in the magma ocean before exsolution starts. We therefore assumed that droplets of FeS liquid exsolve pervasively and equilibrate with the entire mantle at pressure Peq-S, in contrast to the metallic liquid that segregated earlier, which equilibrates with only a limited fraction of the mantle at the base of the magma ocean (8).

The value of the adjustable parameter kS (Eq. 2) determines the final mantle S and HSE concentrations before late accretion (Fig. 3). Optimal results are obtained with kS ≈ 0.44. This fit requires the accretion of a late veneer to increase HSE concentrations to BSE values (12). Late accretion is modeled by terminating the segregation of sulfide liquid, which ends as a result of magma ocean solidification, because FeS liquid cannot percolate efficiently through crystalline mantle (12, 31).

Fig. 3 Final mantle concentrations after multistage sulfide segregation but before late accretion.

Shown are the final concentrations of (A) sulfur (with the BSE concentration shown by the horizontal dashed-outline bar) and (B) HSEs that result from multistage sulfide segregation, without late accretion, as a function of ks (Eq. 2). When late accretion is also modeled, BSE abundances of both sulfur and HSEs are best reproduced with ks ≈ 0.44 [vertical dashed-outline bar in (B)] (see Fig. 4). The concentrations of Pd and Ru are elevated because these elements are the least chalcophile at high P-T. A similar result is obtained for single-stage sulfide segregation.

The models that include both sulfide segregation and late accretion predict Ir, Pt, and S concentrations that are close to BSE values, irrespective of whether FeS segregation occurred in a single stage or multiple stages (Fig. 4 and fig. S3). Furthermore, we reproduced suprachondritic Pd/Ir and Ru/Ir ratios for the BSE (4, 32) (Fig. 4D). Sulfide segregation is very efficient at depleting the mantle in Pt and Ir, but it is less efficient at depleting Pd and Ru because these elements are less chalcophile than Pt and Ir at high P-T (16). This contrasts with metal-silicate partitioning behavior, in which Pd and Pt are the least siderophile (3). However, a number of other explanations have been proposed for suprachondritic Pd/Ir and Ru/Ir ratios (4, 33) that cannot readily be dismissed, especially considering the uncertainties on our results (Fig. 4D).

Fig. 4 Final results based on metal-silicate segregation, sulfide segregation, and late accretion.

The evolution of (A) sulfur and (B and C) HSE concentrations with time is shown for ks = 0.44 (Eq. 2). The accretion history is shown in (A) to (C). Results are shown for multistage FeS segregation (m-s FeS seg.) and late single-stage segregation (s-s FeS seg.) at 118 My. Mass accreted is as in Fig. 1. The vertical dashed lines in (A) to (C) show the time of the final giant impact (GI) at 113 My and the start of late veneer accretion (LV) at 119 My. Horizontal bars show BSE concentrations. Error bars, based on the propagation of uncertainties in the partitioning parameters, are shown in (B) and (C) for the final Pd and Ru concentrations; propagated uncertainties for Pt and Ir are ±0.3 and ±0.6 ppb, respectively. As in Fig. 1, the oblique lines connecting the symbols show the general trends of mass and composition with time, but they do not accurately represent the stepwise progression of the evolution paths. (D) Final calculated HSE values for multistage FeS segregation, normalized to Ir and CI chondrite composition, compared with BSE values (32). The error bars on the calculated values are based on propagating the partitioning parameter uncertainties (tables S1 and S2).

Accretion of a late veneer depends on the mantle being largely crystalline, because otherwise sulfide segregation in a magma ocean would simply continue, and HSE concentrations would never increase to BSE levels (12). The mixing of HSEs into convecting crystalline mantle is expected to have been a slow process, perhaps consistent with a previously estimated mixing time of ~1.5 billion years (34), and also may not have been complete (35). The age of the Moon has recently been determined using correlations between the time of the final giant impact and the mass of late-accreted material, derived from a large number of accretion simulations (6). Our results indicate that this correlation provides only a lower limit on the age of the Moon because it actually dates Earth’s final magma ocean crystallization.

It has been argued that Earth accreted from the same reservoir before and after core formation because of correlated isotopic signatures of Mo and Ru; this argument is based on the assumption that Mo was added to the mantle mainly before late accretion, whereas Ru was added only with the late veneer (36). Here we show that this assumption is not valid because modeled Ru concentrations increased in the mantle at an early stage of Earth accretion and long before the addition of the late veneer, especially in the case of single-stage sulfide segregation (Fig. 4, B and C).

We conclude that the addition of sulfur to Earth occurred over the entire history of accretion (Fig. 4A), refuting the assumption that all sulfur was added during late accretion (37). Although our results are based on assumptions about the distribution of S in the early solar system (12), it is unlikely that a plausible distribution could be found that would change our conclusion. The addition of S throughout accretion affects core formation models that use the elements W and Mo because S strongly influences their partitioning behavior (38). Last, we speculate that a possible low-density layer at the top of Earth’s liquid outer core (39) could be the result of late FeS enrichment due to sulfide segregation.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S4

Tables S1 and S2

References (4062)

Databases S1 and S2

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: D.C.R., V.L., S.A.J., A.M., and H.P. were supported by the European Research Council Advanced Grant ACCRETE (Accretion and Early Differentiation of the Earth and Terrestrial Planets; contract number 290568). A.K.V. was supported by the German Science Foundation (DFG) Priority Programme SPP1385, “The first 10 million years of the solar system – a planetary materials approach” (grant Ru1323/2). We thank H. J. Melosh and R. J. Walker for discussions and three reviewers for their helpful and constructive comments. Data files with final results are available in the supplementary materials.
View Abstract

Stay Connected to Science

Navigate This Article