A Mercury-like component of early Earth yields uranium in the core and high mantle 142 Nd

Abstract

Recent (142)Nd isotope data indicate that the silicate Earth (its crust plus the mantle) has a samarium to neodymium elemental ratio (Sm/Nd) that is greater than that of the supposed chondritic building blocks of the planet. This elevated Sm/Nd has been ascribed either to a 'hidden' reservoir in the Earth or to loss of an early-formed terrestrial crust by impact ablation. Since removal of crust by ablation would also remove the heat-producing elements--potassium, uranium and thorium--such removal would make it extremely difficult to balance terrestrial heat production with the observed heat flow. In the 'hidden' reservoir alternative, a complementary low-Sm/Nd layer is usually considered to reside unobserved in the silicate lower mantle. We have previously shown, however, that the core is a likely reservoir for some lithophile elements such as niobium. We therefore address the question of whether core formation could have fractionated Nd from Sm and also acted as a sink for heat-producing elements. We show here that addition of a reduced Mercury-like body (or, alternatively, an enstatite-chondrite-like body) rich in sulfur to the early Earth would generate a superchondritic Sm/Nd in the mantle and an (142)Nd/(144)Nd anomaly of approximately +14 parts per million relative to chondrite. In addition, the sulfur-rich core would partition uranium strongly and thorium slightly, supplying a substantial part of the 'missing' heat source for the geodynamo.

Full text

A Mercury-like component of early Earth yields
uranium in the core and high mantle
142
Nd
Anke Wohlers1& Bernard J. Wood1
Recent
142
Nd isotope data indicate that the silicate Earth (its
crust plus the mantle) has a samarium to neodymium elemental
ratio (Sm/Nd) that is greater than that of the supposed chon-
dritic building blocks of the planet. This elevated Sm/Nd has
been ascribed either to a ‘hidden’reservoir in the Earth
1,2
or to
loss of an early-formed terrestrial crust by impact ablation
3
.
Since removal of crust by ablation would also remove the heat-
producing elements—potassium, uranium and thorium—such
removal would make it extremely difficult to balance terrestrial
heat production with the observed heat flow
3
.Inthe‘hidden’
reservoir alternative, a complementary low-Sm/Nd layer is
usually considered to reside unobserved in the silicate lower
mantle.Wehavepreviouslyshown,however,thatthecoreisa
likely reservoir for some lithophile elements such as niobium
4
.
We therefore address the question of whether core formation
could have fractionated Nd from Sm and also acted as a sink for
heat-producing elements. We show here that addition of a
reduced Mercury-like body (or, alternatively, an enstatite-
chondrite-like body) rich in sulfur to the early Earth would
generate a superchondritic Sm/Nd in the mantle and an
142
Nd/
144
Nd anomaly of approximately +14 parts per million
relative to chondrite. In addition, the sulfur-rich core would
partition uranium strongly and thorium slightly, supplying a
substantial part of the ‘missing’heat source for the geodynamo.
Terrestrial rocks were recently found to have higher ratios of
radiogenic
142
Nd to nonradiogenic
144
Nd than do the chondritic
meteorites generally supposed to be representative of the material
from which Earth accreted
1,2
.
142
Nd was produced during the early
history of the Solar System from decay of the extinct radionuclide
146
Sm (half-life, t
1/2
= 68 million years
5
and the presence of a positive
142
Nd anomaly of ,20 parts per million (p.p.m.) calculated as
106142Nd
144Nd

Earth 142Nd
144Nd

chondrite
hi
=142Nd
144Nd

Earthor of ,9 p.p.m.
6
in the
silicate Earth would require an Sm/Nd ratio higher than chondritic
1,2
.
This high Sm/Nd ratio was established early in Earth’s history while
146
Sm was still ‘alive’(that is, undergoing radioactive decay).
A plausible mechanism for generating high Sm/Nd in Earth’s
mantle is partial melting and melt extraction to form a crust. Because
Nd is less compatible in mantle silicates than Sm
7
, partial melts have
relatively low Sm/Nd and the solid residue has high Sm/Nd. A low-
Sm/Nd crust could be completely removed from the mantle system by
subduction to an inaccessible region of the deep mantle
1
or removed
from Earth by impact ablation
3
. The problem with the former
hypothesis is the lack of evidence for a hidden silicate reservoir, while
the latter hypothesis suffers from the requirement that much of
Earth’s heat production, in the form of radioactive uranium (U),
thorium (Th) and potassium (K) would be removed together with the
low-Sm/Nd crust. Assuming chondritic abundances of U and Th and
a K/U ratio of ,12,000 for the silicate Earth
8
, the heat production in
the Earth is only about 0.6 times the current heat loss
9
. Reducing the
heat sources further by ablation loss would make it even more difficult
to reconcile heat production with heat loss.
An additional question in the context of heat production is that of
the energy source for the Earth’s magnetic field
10
. Arising from
convection in the core, Earth has had a magnetic field for at least
3.5 billion years. The crystallization of the inner core is an important
source of energy for the geodynamo
11
but most attempts to construct
histories of core cooling indicate that the inner core cannot be much
older than 11.5 billion years
10,11
unless a source of radioactive
heating is present. Numerous studies have focused on
40
Kasa
potential core heat source, because K, in common with all moderately
volatile elements
8
, is depleted in the silicate Earth relative to the
chondritic abundance.
Furthermore, high-pressure experiments
12,13
indicate that K enters
sulfide under oxidizing conditions and sulfur (S) is believed to be a
major component of the core’s complement of approximately 10% of
elements of low atomic number
14
. It appears, however that the
maximum possible K content of the core is insufficient to generate
more than a small fraction of the 25 TW required to generate
reasonable core thermal histories
11,13
. The alternative explanation—
that U and/or Th provide the energy for core convection—has some
support from early experiments on sulphidesilicate partitioning
15
but more recent results indicate very little partitioning of U into
S-bearing metals even under extreme conditions
16
.
We approached the problem of U, Th, Nd and Sm in Earth’s
mantle and core from the standpoint of recent work on partitioning
between sulfide melts and silicate melts
17
. Kiseeva and Wood
17
found
that the sulfidesilicate partition coefficient for any element i,
defined as D
i
=[i]
sulf
/[i]
sil
, is dependent on the FeO content of the
silicate melt, such that for FeS-rich sulfides:
logDi¼Anlog½FeOð1Þ
where Ais a constant and nis a constant dependent on the valency
of element i. Therefore, under strongly reducing conditions, where
the FeO content of the silicate melt is very low (,1% for example)
one would expect D
i
values to be much higher than under the
conditions of MORB crystallization where the FeO content of the
melt is about 8%10%. This hypothesis is consistent with the data of
Murrell and Burnett
15
, who observed strong partitioning of U into
sulfide liquid at low oxygen fugacity f
O2
. Given terrestrial accretion
models calling for prolonged periods of growth under reduced
conditions
18,19
, the demonstration that Mercury is a highly reduced
S-rich planet with a liquid core
20,21
and the association of the rare
earth elements (REEs), U and Th with sulfides in enstatite chondrite
meteorites
22
, we investigated partitioning of U, Th, Sm, Nd and
several other lithophile elements into liquid iron sulfide under
reducing conditions.
Experiments were performed at 1.5 GPa and temperatures
between 1,400 8C and 1,650 8C using starting materials that were
approximately 50:50 mixtures of silicate and FeS doped with a range
of lithophile trace elements including U, Th, La, Nd, Sm, Eu, Yb, Ce,
and Zr (see Methods). The silicate was a basalt-like composition in
the system CaOMgOAl
2
O
3
SiO
2
with variable FeO. Analysis
1
Department of Earth Sciences, University of Oxford, South Parks Road, Oxford OX1 3AN, UK.
16 APRIL 2015|VOL 52 0 |NATURE |337
G2015 Macmillan Publishers Limited. All rights reserved
LETTER doi:10.1038/nature14350

was by electron microprobe and laser ablation inductively coupled
plasma mass spectrometry (LA-ICP-MS) (Methods). Table 1 pre-
sents a summary of sulfidesilicate partitioning results (see
Extended Data Figs 13 and Extended Data Tables 14 for complete
analyses).
Figure 1a shows data from a series of experiments performed at
1,400 8C. As can be seen, the partition coefficients of U, Nd and Sm are
strong functions of the FeO content of the silicate melt, increasing
dramatically, as predicted, as the FeO content decreases below 1 wt%.
The negative slope of logD
i
as f(log[FeO
sil
]) reverses at high FeO
sil
,
however, because the sulfide dissolves progressively more oxygen as
the FeO content of silicate increases and these three lithophile
elements (Fig. 1a) follow oxygen into the sulfide. We found similar
behaviour in two more series of experiments at higher temperature
(Table 1 and Extended Data Fig. 1). Other lithophile elements, notably
Ti, Nb and Ta (B.J.W., unpublished data) behave similarly. Impor-
tantly, we find D
U
.D
Nd
.D
Sm
for partitioning into sulfide in all
experiments. At very low FeO contents all D
i
become .1 (Fig. 1a).
Furthermore (Fig. 1b) D
Nd
is always appreciably greater than D
Sm
,
with D
Nd
/D
Sm
approaching 1.5 in some cases.
The implications of Fig. 1 are that segregation of sulfide (or S-rich
metal) from reduced FeO-poor silicate will lead to enrichment of the
metallic phase in U and in Nd relative to Sm when compared to the
silicate. Addition of such material to the core and mantle respectively
of a growing planet would provide a core heat source and a mantle
with superchondritic Sm/Nd, Yb/Sm and Yb/La. Although poten-
tially detectable in terms of a mantle
142
Nd anomaly, the fractiona-
tion of heavy from light REEs (Table 1) in the primitive mantle of the
body (the bulk silicate Earth in this case) would have little effect on its
overall REE pattern (Extended Data Fig. 3). Similarly, there would be
no observable Eu anomaly despite the fact that Eu is probably in
the 2
+
oxidation state (unlike the other 3
+
lanthanides) under these
conditions (Extended Data Fig. 3). If such a body represented Earth
early in its history then the mantle would have a positive
142
Nd
anomaly relative to chondrite (as observed) and much of the energy
deficit identified for core convection
10
would be supplied by U (and
Th). We find that D
Th
/D
U
is about 0.1, indicating that U would be
accompanied by Th in the S-rich core. Addition of more-oxidized
material later in accretion would lead to the higher current FeO
content of the mantle (8.1%)
8
, but could not erase the super-
chondritic Sm/Nd ratio of the mantle and U content of the core
unless there were complete coremantle re-equilibration.
Figure 2 illustrates the impact of adding a highly reduced body rich
in sulfide to the growing Earth. The Th/U ratio of the silicate Earth
would be higher than chondritic (3.83.9
8,23
), which provides an
important constraint on how much U can be present in Earth’s core.
Based on the Pb-isotopic compositions of Archean galenas
24
and of
3.5-billion-year-old komatiites
25
the Th/U ratio of the Archean
mantle has been estimated to be $4.3. Tatsumoto
26
argued, on the
basis of the Pb isotopic compositions of basalts, for an early
differentiation of the mantle, which resulted in a Th/U of 4.24.5
in the mantle source regions. Since that time the Th/U ratio of the
mantle has decreased, probably owing to preferential recycling of the
more soluble U
27
.
Figure 2 shows four models of U content of the core and the
142
Nd
anomaly of the mantle (relative to the bulk Earth), based on our
partitioning data. We choose a reduced body of 0.15 mass fraction
sulphide, corresponding to the S content of primitive CI chondrites
8
and use values of D
Sm
(Sm
sulf
/Sm
sil
) that are close to the observed
maximum of 0.82.2, noting that D
Sm
values increase with
decreasing temperature and that segregation of sulfide from a
crystal-melt mush instead of melt alone would increase them further
because of the incompatibility of Sm, Nd and U in crystals. As can be
seen (Fig. 2a), adding 20% of such a body to Earth would lead to 45
parts per billion (p.p.b.) of U in the core, a Th/U of the silicate Earth
of 4.17 and a
142
Nd anomaly in the mantle relative to the bulk Earth
of ,7 p.p.m. Increasing the reduced body mass to 45% (Fig. 2b) leads
to about 8 p.p.b. U in the core, a Th/U of the silicate Earth of 4.5 and a
mantle
142
Nd anomaly of 13.9 p.p.m. relative to the bulk Earth.
We performed a sensitivity analysis (Extended Data Fig. 2) and
find that, if the Th/U of the silicate Earth is #4.5, the maximum U
content of the core is 8 p.p.b. with a Th content of ,8 p.p.b. These
figures increase to ,10 p.p.b. if the Th/U of the silicate Earth is #4.7.
The
142
Nd anomaly is 13.9 p.p.m. in the former case and ,17 p.p.m.
in the latter. The estimated U and Th contents of the core would lead
to 22.4 TW, sufficient to power the geodynamo
11
even without the
potential 0.40.8 TW from
40
K decay
13
. We can reduce the size of the
reduced body by increasing its S content (Fig. 2c, d) and increase
D
U
/D
Sm
but the overall effects on the core and mantle
142
Nd remain
U
Nd
Sm
–3
–2
–1
0
1
2
–1 –0.5 0 0.5 1 1.5
log[DsulDsil (wt%)]
log[FeO in silicate (wt%)]
0
0.5
1
1.5
2
–0.5 0 0.5 1 1.5
log[FeO in silicate (wt%)]
DNd/DSm
1,400 °C
1,500°C
1,650°C
1.5 GPa
1.5 GPa
1,400 ºC
b
a
Figure 1 | Sulphidesilicate partitioning data. a, Partition coefficients D
i
=
[i]
sulf
/[i]
sil
for U, Nd and Sm at 1.5 GPa and 1,400 8C plotted versus the log of
the FeO content of the silicate melt in weight per cent. b,TheratioofD
Nd
/D
Sm
plotted versus log[FeO]. Error bars in both cases are 62s.e.and,ifabsent,are
smaller than symbol size.
G2015 Macmillan Publishers Limited. All ri
g
hts reserved
RESEARCH LETTER
338 |NATURE |VOL 520 |16 APRIL 2015

Table 1 | Summary of sulfidesilicate partition coefficients
Run Pressure, P(GPa) Temperature, T(8C) log[FeO
sil
(wt%)] D
Sm
D
Nd
/D
Sm
D
U
/D
Sm
D
Th
/D
U
D
Eu
/D
Sm
D
La
/D
Sm
D
Yb
/D
Sm
421 1.5 1,400 0.50 0.005 1.42 2.47 0.036 5.85 1.38 0.16
σ0.001 0.29 0.37 0.008 0.93 0.23 0.07
428 1.5 1,400 0.08 0.013 1.35 1.56 0.038 5.50 1.35 0.16
σ0.001 0.19 0.17 0.005 1.06 0.20 0.02
427 1.5 1,400 20.25 0.062 1.30 1.81 0.028 2.36 1.23 0.13
σ0.006 0.19 0.24 0.004 0.40 0.22 0.02
426 1.5 1,400 20.30 2.247 1.25 6.81 0.048 0.14 1.03 0.16
σ0.333 0.30 1.43 0.010 0.05 0.15 0.09
429 1.5 1,400 1.21 0.005 1.04 3.68 0.200 2.04 1.10 0.67
σ0.0001 0.12 0.89 0.041 0.42 0.47 0.19
461 1.5 1,650 0.30 0.023 1.22 1.92 0.058 4.09 1.18 0.21
σ0.003 0.20 0.32 0.009 0.58 0.18 0.03
462 1.5 1,650 20.21 0.154 1.12 3.58 0.046 1.13 0.92 0.27
σ0.011 0.13 0.43 0.007 0.14 0.10 0.03
477 1.5 1,650 20.29 0.629 1.10 9.26 0.044 0.37 0.83 0.39
σ0.038 0.11 0.73 0.043 0.38 0.80 0.39
464 1.5 1,650 20.32 0.751 1.13 9.41 0.035 0.21 0.55 0.28
σ0.073 0.16 1.18 0.020 0.16 0.44 0.16
1,414 1.5 1,500 20.21 0.048 1.31 1.84 0.031 5.95 1.41 0.13
σ0.004 0.14 0.34 0.009 0.80 0.16 0.02
1,415 1.5 1,500 20.39 0.454 1.18 6.99 0.040 0.88 1.04 0.21
σ0.028 0.13 0.66 0.005 0.11 0.14 0.05
1,416 1.5 1,500 0.88 0.006 1.39 2.70 0.067 4.92 1.52 0.23
σ0.0005 0.18 0.25 0.007 0.56 0.19 0.02
1,417 1.5 1,500 1.06 0.007 1.20 3.11 0.150 2.71 1.19 0.50
σ0.001 0.24 0.52 0.042 0.48 0.21 0.30
Partition coefficients in weight ratio. σis calculated from error propagation.
0
5
10
15
U (p.p.b.)
3.2% sulfur
0
5
10
15
10 20 30 40 50
U (p.p.b.)
Reduced bod
y
mass (%)
4.6% sulfur
10 20 30 40 50
Reduced bod
y
mass (%)
8.1% sulfur
142Nd (p.p.m.)
DSm = 1
DSm = 2
DSm = 3
7.1% sulfur
142Nd = 5.7 Th/U = 4.10
142Nd = 6.6 Th/U = 4.17
142Nd = 3.8 Th/U = 4.04
142Nd = 13.9 Th/U = 4.54
142Nd = 16.7 Th/U = 4.74
142Nd = 9.0 Th/U = 4.25
142Nd = 6.9 Th/U = 4.17
142Nd = 5.1 Th/U = 4.06
142Nd = 7.1 Th/U = 4.23
142Nd = 12.6 Th/U = 4.51
142Nd = 9.3 Th/U = 4.27
142Nd = 13.7 Th/U = 4.64
ba
cd
Figure 2 | Core content of U (p.p.b.) and mantle
142
Nd anomaly
(p.p.m.). a, Calculated effect of adding to the growing Earth a reduced body
of 20% of Earth’s mass containing 0.15 mass fraction sulfide. The sulfide is
added to the core and the silicate to the mantle. Sulphidesilicate D
U
/D
Sm
is
fixed at 2, D
Nd
/D
Sm
at 1.4 and D
Th
/D
U
at 0.1 (Table 1). b, Same as aexcept the
mass of reduced body is 45% of the Earth’s mass. cand d, As for aand b
except the reduced body contains 0.22 mass fraction sulfide. The reduced
body and remainder of Earth each contain 14 p.p.b. U and 53.5 p.p.b. Th,
consistent with chondritic abundances. Sulfide extraction was assumed to take
place at the origin of the Solar System.
16 APRIL 2015|VOL 52 0 |NATURE |339
G2015 Macmillan Publishers Limited. All rights reserved
LETTER RESEARCH

close to those summarized above if the Th/U of the bulk silicate Earth
is constrained to be #4.5 or #4.7.
We note that the scenarios shown in Fig. 2 refer to a terrestrial core
containing between 3.2 wt% S and 8.1 wt% S. The concentration of
cosmochemically abundant volatile S in the core is unknown, but
recent suggestions range from a cosmochemical estimate of 1.7 wt%
(ref. 14) to ,6 wt% (ref. 28) from liquid-metal density measurements
and 14.7 wt% (ref. 29) from high-temperature, high-pressure
equation-of-state measurements. The range shown in Fig. 2 is,
therefore appropriate for the current state of knowledge.
We conclude that a period of growth of the accreting Earth
under reduced, S-rich conditions would generate a measureable
(,þ14 p.p.m.)
142
Nd anomaly in the silicate Earth, in agreement
with observations. This would also add sufficient U and Th to the core
to generate 22.4 TW of the energy required to drive the geodynamo.
Online Content Methods, along with any additional Extended Data display items
and SourceData, are available in theonline version of the paper; references unique
to these sectionsappear only in the online paper.
Received 17 December 2014; accepted 20 February 2015.
1. Boyet, M. & Carlson, R. W.
142
Nd evidence for early (>4.53 Ga) global
differentiation of the silicate Earth. Science 309, 576581 (2005).
2. Boyet, M. & Carlson, R. W. A new geochemical model for the Earth’s mantle inferred
from (SM)-S-146-Nd-142 systematics. Earth Planet. Sci. Lett. 250,
254268 (2006).
3. Campbell, I. H. & O’Neill, H. S. C. Evidence against a chondritic Earth. Nature 483,
553558 (2012).
4. Wade, J. & Wood, B. J. The Earth’s‘missing’niobium may be in the core. Nature
409, 7578 (2001).
5. Kinoshita, N. et al. A shorter Sm-146 half-life measured and implications for Sm-
146-Nd-142 chronology in the Solar System. Science 335, 16141617 (2012).
6. Sprung, P., Kleine, T. & Scherer, E. E. Isotopic evidence for chondritic Lu/Hf and
Sm/Nd of the Moon. Earth Planet. Sci. Lett. 380, 7787 (2013).
7. Blundy, J. D. & Wood, B. J. Prediction of crystalmelt partition coefficients from
elastic moduli. Nature 372, 452454 (1994).
8. Palme, H. & O’Neill, H. S. C. in The Mantle and Core Vol. 2 Treatise on Geochemistry
(ed. Carlson, R. W.) 138 (Elsevier, 2003).
9. Stein, C. A. in Global Earth Physics: A Handbook of Physical Constants (ed. Ahrens, T.
J.) 144158 (American Geophysical Union, 1995).
10. Labrosse, S., Poirier, J. P. & Le Mouel, J. L. The age of the inner core. Earth Planet. Sci.
Lett. 190, 111123 (2001).
11. Nimmo, F., Price, G. D., Brodholt, J. & Gubbins, D. The influence of potassium on
core and geodynamo evolution. Geophys. J. Int. 156, 363376 (2004).
12. Gessmann, C. K. & Wood, B. J. Potassium in the Earth’s core? Earth Planet. Sci. Lett.
200, 6378 (2002).
13. Murthy, V. M., vanWestrenen,W. & Fei, Y. W. Experimental evidence that potassium is a
substantial radioactive heat source in planetary cores. Nature 423, 163165 (2003).
14. Dreibus, G. & Palme, H. Cosmochemical constraints on the sulfur content in the
Earth’s core. Geochim. Cosmochim. Acta 60, 11251130 (1996).
15. Murrell, M. T. Partitioning of K, U, and Th between sulfide and silicate liquids—
implications for radioactive heating of planetary cores. J. Geophys. Res. 91,
81268136 (1986).
16. Wheeler, K. T., Walker, D., Fei, Y. W., Minarik, W. G. & McDonough, W. F. Experimental
partitioning of uranium between liquid iron sulfide and liquid silicate: implications
for radioactivity in the Earth’s core. Geochim. Cosmochim. Acta 70,
15371547 (2006).
17. Kiseeva, E. S. & Wood, B. J. A simple model for chalcophile element partitioning
between sulphide and silicate liquids with geochemical applications. Earth Planet.
Sci. Lett. 383, 6881 (2013).
18. Wade, J. & Wood, B. J. Core formation and the oxidation state of the Earth. Earth
Planet. Sci. Lett. 236, 7895 (2005).
19. Rubie, D. C. et al. Heterogeneous accretion, composition and core-mantle
differentiation of the Earth. Earth Planet. Sci. Lett. 301, 3142 (2011).
20. Nittler, L. R. et al. The major-element composition of Mercury’s surface from
MESSENGER X-ray spectrometry. Science 333, 18471850 (2011).
21. Smith, D. E. et al. Gravity field and internal structure of Mercury from MESSENGER.
Science 336, 214217 (2012).
22. Gannoun, A., Boyet, M., El Goresy, A. & Devouard, B. REE and actinide
microdistribution in Sahara 97072 and ALHA77295 EH3 chondrites: a combined
cosmochemical and petrologic investigation. Geochim. Cosmochim. Acta 75,
32693289 (2011).
23. Blichert-Toft, J., Zanda, B., Ebel, D. S. & Albarede, F. The Solar System primordial
lead. Earth Planet. Sci. Lett. 300, 152163 (2010).
24. Hofmann, A. W. Reviving the layered mantle: plan D. American Geophysical Union
Fall Meeting Abstract V43E05 http://abstractsearch.agu.org/meetings/2011/
FM/V43E-05.html (2011).
25. Allègre, C. J., Dupre, B. & Lewin, E. Thorium uranium ratio of the Earth. Chem. Geol.
56, 219227 (1986).
26. Tatsumoto, M. Isotopic composition of lead in oceanic basalt and its implication to
mantle evolution. Earth Planet. Sci. Lett. 38, 6387 (1978).
27. Elliott, T., Zindler, A. & Bourdon, B. Exploring the kappa conundrum: the role of
recycling in the lead isotope evolution of the mantle. Earth Planet. Sci. Lett. 169,
129145 (1999).
28. Morard, G. et al. The Earth’s core composition from high pressure density
measurements of liquid iron alloys. Earth Planet. Sci. Lett. 373, 169178 (2013).
29. Seagle, C. T., Campbell, A. J., Heinz, D. L., Shen, G. Y. & Prakapenka, V. B. Thermal
equation of state of Fe3S and implications for sulfur in Earth’s core. J. Geophys. Res.
111, B06209 (2006).
Acknowledgements We acknowledge support from the European Research
Council grant number 267764. We thank J. Wade for his advice and comments.
A. Hofmann and T. Elliott provided advice and suggestions about Th/U of silicate
Earth.
Author Contributions Both authors performed experiments and microanalysis,
with about two-thirds done by A.W. Both authors contributed to writing the
manuscript, with about two-thirds done by B.J.W.
Author Information Reprints and permissions information is available at www.
nature.com/reprints. The authors declare no competing financial interests.
Readers are welcome to comment on the online version of the paper.
Correspondence and requests for materials should be addressed to A.W. (anke.
wohlers@earth.ox.ac.uk) or B.J.W. (berniew@earth.ox.ac.uk).
G2015 Macmillan Publishers Limited. All ri
g
hts reserved
RESEARCH LETTER
340 |NATURE |VOL 520 |16 APRIL 2015

METHODS
Experimental methods. Starting materials for high-pressure experiments
consisted of mixtures of ,50 wt% (Fe,Ni)S and ,50% of a synthetic silicate
approximating the 1.5 GPa eutectic composition in the anorthitediopside
forsterite system
30
. The sulfide component was analytical-grade FeS doped with
1%3% NiS. Trace elements were added as a mix consisting of Zr, La, Ce,Nd, Sm,
Eu, Yb, Th and U as oxides. After adding the trace-element mix such that each
element was present at 1,0002,000 p.p.m., the silicate and sulfide starting
materials were mixed in 50:50 proportions and ground under acetone for 20 min,
then dried at 110 8C before the experiment. Starting compositions were loaded
into 3 mm outer diameter and 1 mm inner diameter graphite capsules.
Experiments were conducted in a half-inch-diameter piston-cylinder appar-
atus using external cylinders eitherof BaCO
3
-silica glass (at 1,500 8C and 1,650 8C)
or CaF
2
(at 1,400 8C) and an 8 mm outer diameter graphite furnace with a
1-mm-thick wall. The unsealed capsule was separated from the graphite furnace
by an interior MgO sleeve, with a 0.5-mm-thick alumina disk on top to prevent
puncture by the thermocouple. Temperatures were controlled and monitored
using a tungstenrhenium thermocouple (W5%Re/W26%Re), and the temper-
ature was maintained within618C. Experimental conditions were 1,400 8C, 1,500
8C and 1,650 8C at 1.5 GPa and with experiment durations between 1 h and 4.5 h.
These times are sufficient to approach equilibrium in small graphite capsules
17
.
Experiments were quenchedby turning off the power supply. After quenching, the
capsule was extracted from the furnace, mounted in acrylic and polished for
further analyses with electron microprobe and LA-ICP-MS. All experimental
charges contained sulfide blebs embedded in a silicate glass matrix.
Microanalysis. Samples were analysed on the JEOL 8600 electron microprobe in
the Archaeology Department at the University of Oxford. Wavelength dispersive
analyses of the major-element compositions of silicate glasses and sulfides were
performed at 15 kV with a beam current of 20 nA and a 10 μm defocused beam
(Extended Data Tables 1 and 2). At least 20 analyses were taken of the silicate
and sulfide in each experiment. Count times for major elements (Si, Al, Ca, Mg,
Fe in silicate, Fe in sulfide) were 30 s on the peak and 15 s background. Minor
elements (S, Ni, O) were analysed for 60 s peak and 30 s background. We have
previously noted Ni loss from similar experiments
17
and the principal reason for
adding Ni was to provide an additional check on LA-ICP-MS analyses of the
trace elements of interest (see below). A range of natural and synthetic standards
was used for calibration. Standards for silicate were wollastonite (Si, Ca), jadeite
(Al), periclase (Mg) and haematite (Fe). Standards for sulfides were Ni metal
(Ni), galena (S) and haematite (Fe, O). Oxygen in the sulfides was determined
using the Kαpeak and LDE crystal.
We determined U and Sm contents of three product sulfides as a further check
on the LA-ICP-MS analyses. In this case we measured the Mαpeak for U and the
Lαpeak for Sm using standards of UO
2
and SmPO
4
respectively and a PET
crystal. Operating conditions were 15 kV, 40 nA and a 10 μm beam. The count
time for U was 120 s on peak and 60 s background. Sm was analysed for 150 s on
the peak and 75 s background.
Trace elements in silicates and sulfides were measured by LA-ICP-MS employ-
ing a NexION 300 quadropole mass spectrometer coupled to a New Wave
Research UP213 Nd:YAG laser at the University of Oxford. A laser repetition rate
of 10 Hz and spot size of 2550 μm were used for silicate glasses and sulfides
(ExtendedData Tables 3 and 4) with an energy density of ,12 J cm
−2
.Operatingin
time-resolved mode, we employed 20 s of background acquisition, followed by
ablation for 60 s. Between analyses we employed a 6090 s ‘wash-out’time. The
following masses were counted:
24
Mg,
27
Al,
29
Si,
57
Fe,
60
Ni,
43
Ca,
90
Zr,
139
La,
140
Ce,
142
Nd,
152
Sm,
153
Eu,
174
Yb,
232
Th,
239
U. Our external standard was NIST610 glass
and we typically collected three spectra of this at the beginning and end of each
sequence of 1015 unknowns. The BCR-2G standard was used as a secondary
standard to check the accuracy of the calibration. Ablation yields were corrected
by referencing to the known concentrations of Si and Ca (silicate glass) and
Fe (sulfides), which had been determined by microprobe. Data reduction was
performed off-line using the Glitter 4.4.3 software package (http://www.glitter-
gemoc.com/) which enabled us to identify occasional sulfide inclusions in the
silicate analyses. Since the Fe content of the NIST610 standard is only 460 p.p.m.,
the background is high and the matrices are very different, so cross-checks on the
sulfide analyses were required. Therefore, we measured Ni with the electron
microprobe and LA-ICP-MS. In agreement with Kiseeva and Wood
17
, we observed
no systematic offset between electron microprobe and LA-ICP-MS analyses for Ni
(Extended Data Tables 2 and 4). Additionally, as discussed above, the U and Sm
contents of the sulfides were measured by electron microprobe in three
experimental charges (numbers 1415, 464, and 477). Between 20 and 43 electron
probe analyses were performed on each sample. The highest U and Sm
concentrations were measured in experiment 1415 with LA-ICP-MS (U = 2,958
p.p.m., Sm = 719 p.p.m.). Comparative measurements with electron probe yielded
values of U = 3,280 6490 p.p.m. and Sm = 707 6110 p.p.m., (uncertainty is 2 s.e.)
and therefore show excellent agreement. Two samples with lower U and Sm
concentrations were also analysed. LA-ICP-MS measurements for experiment 464
yielded U = 952 p.p.m. and Sm = 327 p.p.m., while experiment 477 gave U = 927
p.p.m. and Sm = 300 p.p.m. Electron microprobe concentrations of U =1,164 6224
p.p.m. and Sm = 319 687 p.p.m. (experiment 464) and U = 991 669 p.p.m. and
Sm = 277 638 p.p.m. (experiment 477) are also in excellent agreement with the
LA-ICP-MS measurements. We conclude that our LA-ICP-MS results have no
detectable systematic offset due to matrix effects or calibration errors.
30. Presnall, D. C. et al. Liquidus phase relations on the join diopside-forsterite-
anorthite from 1 atm to 20 kbar: their bearing on the generation and crystallization
of basaltic magma. Contrib. Mineral. Petrol. 66, 203220 (1978).
31. Salters, V. J. M. & Stracke, A. Composition of the depleted mantle. Geochem.
Geophys. Geosyst. 5, http://dx.doi.org/10.1029/2003gc000597 (2004).
G2015 Macmillan Publishers Limited. All rights reserved
LETTER RESEARCH

Extended Data Table 1 | Major-element composition of silicate
glass.
Values are in weight per cent. σcalculated from error propagation. nis the number of
measurements. Σtrace is the sum of trace elements measured using LA-ICP-MS.
G2015 Macmillan Publishers Limited. All rights reserved
RESEARCH LETTER

Extended Data Table 2 | Major-element composition of
sulfides.
Values are in weight per cent. σcalculated from error propagation. n.m., not measured. nis the
number of measurements. Σtrace is the sum of trace elements measured using LA-ICP-MS.
G2015 Macmillan Publishers Limited. All rights reserved
LETTER RESEARCH

Extended Data Table 3 | Trace-element concentration in silicates.
Values are in parts per million. nis the number of measurements. n.m., not measured.
G2015 Macmillan Publishers Limited. All rights reserved
RESEARCH LETTER

Extended Data Table 4 | Trace-element concentration in sulfides.
Values are in parts per million. nis the number of measurements.
G2015 Macmillan Publishers Limited. All rights reserved
LETTER RESEARCH

Extended Data Figure 1 | Partition coefficients for U, Nd and Sm with
changing log[FeO] content in silicate melt (wt%). a, Results for Dvalues of
experiments performed at 1.5 GPa and 1,500 8C. b,Dvalue results at 1.5 GPa
and 1,650 8C.
G2015 Macmillan Publishers Limited. All rights reserved
RESEARCH LETTER

Extended Data Figure 2 | Core content of U (p.p.b.) and mantle
142
Nd
anomaly (p.p.m.) at D
U
/D
Sm
=3. The effect on the Nd and U content using
the same parameters (D
Nd
/D
Sm
at 1.4 and D
Th
/D
U
at 0.1) as in Fig. 2 but with
a higher D
U
/D
Sm
ratio. aand bshow the calculated effect of adding to Earth a
reduced body of 20% of the Earth’s mass or 45% of the Earth’s mass,
containing 0.15 mass fraction sulfide. cand dillustrate the same scenario
except that the reduced body contains 0.22 mass fraction sulfide.
G2015 Macmillan Publishers Limited. All rights reserved
LETTER RESEARCH

Extended Data Figure 3 | REE fractionation at 3.2% and 8.1% S in the
core. The calculated REE pattern in the bulk silicate Earth (BSE) for the two
extreme cases of Fig. 1a (3.2% S) and Extended Data Fig. 2d (8.1% S). Black
diamonds represent REE concentrations relative to chondritic abundances
and normalized to Yb = 1, at 3.2% S in the core (20% reduced mass impactor
containing 0.15 mass fraction sulfide). White diamonds illustrate the REE
fractionation at elevated S content (8.1% S in the core, 35% reduced mass
impactor containing 0.22 mass fraction sulfide). We assumed D
Sm
= [(Sm in
sulfide)/(Sm in silicate)]=1 and D
i
/D
Sm
ratios for other elements from
experiment 464. Both scenarios result in very small depletions of light REE
relative to heavy REE in the BSE. The trend is broadly consistent with that
seen in the depleted mid-ocean-ridge basalt (MORB)mantle composition
(blue diamonds) but much smaller. The effect on the REE pattern of the BSE
would, as can be seen, be undetectable. Blue diamonds illustrate the measured
ratio of depleted MORB mantle (from Salters and Stracke
31
) to the BSE
(Palme and O’Neill
8
, assuming chondritic abundances of refractory lithophile
elements in the latter. Error bars are from propagated error calculation and
correspond to 1 s.d.
G2015 Macmillan Publishers Limited. All rights reserved
RESEARCH LETTER