Content uploaded by J. Zhangzhou
Author content
All content in this area was uploaded by J. Zhangzhou on Feb 09, 2018
Content may be subject to copyright.
American Mineralogist, Volume 101, pages 181–192, 2016
0003-004X/16/0001–181$05.00/DOI: http://dx.doi.org/10.2138/am-2016-5308 181
* E-mail: zhan1721@umn.edu
Experimental constraints on mantle sulde melting up to 8 GPa
Zhou Zhang1,* and Marc M. hirschMann1
1Department of Earth Sciences, University of Minnesota, Minneapolis, Minnesota 55414, U.S.A
abstract
We present high-pressure experiments up to 8 GPa that constrain the solidus and liquidus of a
composition, Fe0.69Ni0.23Cu0.01S1.00, typical of upper mantle sulde. Solidus and liquidus brackets of this
monosulde are parameterized according to a relation similar to the Simon-Glatzel equation, yielding,
respectively, T (°C) = 1015.1 [P(GPa)/1.88 + 1]0.206 and T (°C) = 1067.3 [P(GPa)/1.19 + 1]0.149 (1 ≤ P
≤ 8). The solidus t is accurate within ±15 °C over the pressure intervals 1–3.5 GPa and within ±30
°C over the pressure intervals 3.5–8.0 GPa. The solidus of the material examined is cooler than the
geotherm for convecting mantle, but hotter than typical continental geotherms, suggesting that sulde
is molten or partially molten through much of the convecting upper mantle, but potentially solid in the
continental mantle. However, the material examined is one of the more refractory among the spectrum
of natural mantle sulde compositions. This, together with the solidus-lowering effects of O and C
not constrained by the present experiments, indicates that the experimentally derived melting curves
are upper bounds on sulde melting in the Earth’s upper mantle and that the regions where sulde is
molten are likely extensive in both the convecting upper mantle and, potentially, the deeper parts of
the oceanic and continental lithosphere, including common source regions of many diamonds.
Keywords: Sulfide, mantle, solidus, melting, experimental constraint, calibration
introduction
Molten sulfides are important geochemical and geophysi-
cal agents in Earth’s interior. Sulfide mineral and melt are the
chief carriers of siderophile and chalcophile elements in the
upper mantle (Pearson et al. 2003) and mobilization of sulfide
melts may produce fractionated chalcophile and platinum group
element (PGE) patterns (Alard et al. 2000, 2002; Ballhaus et
al. 2006; Bockrath et al. 2004; Delpech et al. 2012; Hart and
Gaetani 2006; Li and Audétat 2012; Powell and O’Reilly 2007).
Furthermore, sulfides are key hosts of Os, Pb, and potentially He
and therefore play an important role in evolution of important
isotopic systems (Hart and Gaetani 2006; Huang et al. 2014;
Pearson et al. 2002; Roy-Barman et al. 1998). Consequently,
they are widely used targets for Re-Os and Pb-Pb geochronologic
studies (Pearson et al. 1998, 2003) but interpretation of result-
ing ages depends on sulfide parageneses. Sulfide melts are also
potentially responsible for mantle geophysical anomalies, as their
physical properties differ greatly from those of silicates. These
include higher density, surface tension, electrical conductivity,
and lower melting points (Bockrath et al. 2004; Helffrich et al.
2011; Mungall and Su 2005). For example, it has been speculated
that sulfide melts are responsible for seismic anomalies at ~100
km in continental cratons (Helffrich et al. 2011).
A key feature of natural sulfide is that it may be molten in
large parts of the mantle (e.g., Bockrath et al. 2004; Hart and
Gaetani 2006) and so constraining its geochemical and geophysi-
cal role requires defining the conditions of sulfide melt stability.
Although detailed one atmosphere studies have explored the
Fe-Ni-S phase diagrams at 900–1350 °C (e.g., Hsieh et al. 1987;
Waldner and Pelton 2004 and references therein), the majority
of high-pressure studies on sulfide melting to date have been
performed on simple stoichiometric or eutectic compositions
(Boehler 1992, 1996; Ryzhenko and Kennedy 1973; Sharp 1969;
Usselman 1975). Many of these have focused on the influence
of sulfide on core formation, and so have examined relations in
metal-rich compositions, including studies to very high pressures
(Boehler 1992; Fei et al. 1997; Morard et al. 2011; Stewart et
al. 2007). But fewer studies have considered melting relations
of compositions appropriate for the modern upper mantle, and
these have been limited to relatively low (≤3.5 GPa) pressures
(Ballhaus et al. 2006; Bockrath et al. 2004).
Comparison of experimental studies of stoichiometric sulfides
to those conducted in metal-rich sulfide-metal eutectics shows
that melting temperatures are strongly variable depending on
metal/sulfide ratios (Fig. 1). Furthermore, substitution of Ni
and Cu for Fe also influences melting temperatures (Hsieh et al.
1987; Urakawa et al. 1987). Consequently, understanding melt-
ing of upper mantle sulfides requires perspective on the range
of compositions likely to be present. Compositions of natural
mantle sulfides are quite variable, in part owing to their tendency
to exsolve on cooling (Pearson and Wittig 2014; Richardson et
al. 2001). The most reliable records derive from reintegrated
compositions from inclusions in olivine or diamond, studies of
which (Aulbach et al. 2009; Westerlund et al. 2006) indicate that
most upper mantle sulfides have compositions close to mono-
sulfide stoichiometry with metal/sulfide (M/S) ratios typically
between 0.9–1.2 (Fig. 2).
Previous experimental constraints on the high-pressure melt-
ing of monosulfide similar to natural mantle composition derive
ZHANG AND HIRSCHMANN: MANTLE SULFIDE MELTING UP TO 8 GPA182
chiefly from the studies of Bockrath et al. (2004) and Ballhaus
et al. (2006). Bockrath et al. (2004) documented the melting
interval of a bulk composition no. 1 with an M/S ratio of 0.93,
which is at the low end of the range present in natural mantle
compositions (Fig. 2a; Table 1). Ballhaus et al. (2006) reported
further compositional details about the phases produced by
melting at composition no. 1 and extended this work to include
two more metal-rich compositions, no. 2 with M/S of 1.06 and
no. 3 with M/S of 1.11. All three compositions had Ni contents
(15.5 wt%) appropriate for mantle sulfides as calculated Fe-Ni
exchange equilibrium with mantle olivine, and small amounts
of Cu. The solidus of the low M/S in composition no. 1 is near
1075 °C at 0.1 MPa and 1275 °C near 3 GPa, placing it below
the convecting mantle adiabat (Katsura et al. 2010), above typical
continental geotherms (Pollack and Chapman 1977), and similar
to temperatures for intermediate-age oceanic lithosphere (Tur-
cotte and Schubert 2002). More metal-rich compositions no. 2
and no. 3 have markedly lower solidi: that of composition no. 2
is 75–100 °C lower than for composition no. 1. The solidus of
compositon no. 3 appears to be as low as 800 °C and with very
little pressure dependence, perhaps because its composition falls
outside of the field for mss (Fig. 2b). It likely has more than one
phase below the solidus and so its melting behavior is similar
to that at the Fe-FeS or (Fe,Ni)S-(Fe,Ni)O eutectics (Fig. 1).
Together, these data indicate that mantle sulfide is partially
molten in the hotter parts of the upper mantle, but significant
questions remain. The experiments on natural monosulfide (mss)
compositions extend only to 3.5 GPa and so do not pertain to
the deeper parts of continental lithosphere or the oceanic low
velocity zone. Also, constraints on the liquidus are absent for
composition no. 2 and composition no. 3 and produce curious
results for composition no. 1, indicating a concave upward trend
that is contrary to expected melting behavior (e.g., Hart and
Gaetani 2006). Here we present partial melting experiments on
a bulk composition similar to compositon no. 1 up to 8 GPa.
Our purpose is to verify the solidus and liquidus determined
by Bockrath et al. (2004), and to determine the phase relations
of this relatively refractory composition to higher pressure.
An additional motivation is to refine pressure and temperature
calibrations in the piston-cylinder and multi-anvil devices, to
enhance the reliability of the sulfide melting results.
ExpEriMEntal MEthods
Starting materials and preparation of sample capsules
Experimental starting materials were prepared by mixing Alfa-Aesar reagents
including Fe (99.9% powder), FeS2 (99.9% powder), Ni (99.8% powder), and Cu
(99.9% powder). The mixture was homogenized by grinding under ethanol in an
agate mortar for 1 h. After mixing, samples were dried at 110 °C in a vacuum oven
for 5 min. Subsequently, starting mixtures were stored in a sealed glass container
in a sealed glass desiccator to avoid oxidation. Subsequent electron microprobe
analysis of post-experiment samples indicate <0.6 wt% oxygen, suggesting minimal
oxidation of the starting material.
Starting mixtures were loaded into silica glass capsules, which consisted of a
tube closed with a cap on both ends. For piston-cylinder experiments, tubes were
2 mm long with inner and outer diameters of 2 and 3 mm and the caps were each
1 mm long. For multi-anvil experiments, the tubes were 1 mm long, with inner
and outer diameters of 1 and 2 mm, and caps were 1 mm long.
High-pressure experiments
Experiments were performed using an end-loaded piston cylinder (PC) device
and a 1000 ton Walker-style multi-anvil (MA) apparatus using procedures described
in detail by Xirouchakis et al. (2001) and Dasgupta et al. (2004). The PC experi-
ments were conducted in 1.27 cm diameter pressure vessels from 0.8–3.3 GPa using
a hot piston in technique, with hydraulic pressure maintained constant during the run
period and with temperature controlled by a type B (Pt70Rh30/Pt94Rh6) thermocouple
located 1 mm above the top of the capsule. Assemblies consisted of BaCO3 cells
with MgO spacers and a graphite heater. MgO assemblies were dried at 1000 °C for
FigurE 1. Melting relations from selected previous sulde melting
experiments up to 15 GPa. Compositions and data sources are given in
Table 1. (Color online.)
FigurE 2. (a) Sulde compositions projected into the metal-sulfur-
oxygen atomic ternary. Metal is the sum of Fe, Ni, and Cu. Natural sulde
compositions are taken from bulk composition estimates of suldes trapped
in diamonds and olivine as inclusions (Aulbach et al. 2009; Bulanova et
al. 1996; Westerlund et al. 2006). (b) Sulde compositions projected into
the Fe-S-Ni ternary. Minor amounts of Cu and O have been projected to
the Ni and S apices, respectively. The yellow and pink areas correspond
to the elds of stability of mss and mss+melt at 0.1 MPa, 1000 °C (Hsieh
et al. 1987). Numbered circles refer to sulde compositions that have
been investigated experimentally, as listed in Table 1. Natural sulde
compositions are plotted as gray circles from the same references as a.
(Color online.)
ZHANG AND HIRSCHMANN: MANTLE SULFIDE MELTING UP TO 8 GPA 183
8–15 h and stored in a drying oven at 110 °C. The MA experiments were conducted
at 2.7–8.0 GPa using 18/12 OEL/TEL assemblies, including cast MgO-Al2O3-SiO2-
Cr2O3 octahedra and integrated gasket fins. Straight-walled graphite furnaces and
type C (W74Re26/W95Re5) thermocouples were used for all MA experiments and are
located ~1 mm from the sulfide charges. Samples were compressed to target pres-
sures, dwelling for 5 h to allow assembly stabilization and then heated to the target
temperature. Both PC and MA experiments were quenched by power termination.
Individual experimental run conditions are detailed in Table 2.
Pressure and temperature calibration experiments
To ensure that experimental temperatures and pressures were accurate for the
conditions of the sulfide melting experiments, we conducted several new calibra-
tions to refine previous documentation of PC and MA performance in the UMN
experimental petrology laboratory by Xirouchakis et al. (2001), Dasgupta et al.
(2004), Withers et al. (2011), and Tenner et al. (2012). Pressure calibrations were
conducted against the quartz-coesite transition (Bohlen and Boettcher 1982; Bose
and Ganguly 1995). Temperature calibration was conducted against the melting
temperature of Au (Mirwald and Kennedy 1979).
For quartz-coesite experiments in the PC device, a homogeneous mixture of the
two phases (50:50) was sealed in Pt capsules. The quartz was taken from a crushed
natural gem-quality crystal. The coesite was presynthesized from silica glass in a
graphite capsule at 900 °C and 4 GPa, based on previous syntheses conducted by
Luo et al. (2002), and verified by XRD. Each experiment lasted 2 to 12 h at the
target temperature and pressure and then the Pt capsule was cut and the portion of
the post-run product that had been closest to the thermocouple was collected for
XRD analysis. Because reaction was not always complete, the intensities of the
XRD peaks of experimental samples were compared to those of the starting mix-
ture. In the MA apparatus, the silica glass capsule sample lids from sulfide melting
experiments were examined post-experiment. The top lid of the silica capsule was
recovered and analyzed by Raman spectroscopy and ground to a powder for XRD.
For the Au melting experiments, a 2 mm diameter and 2 mm length hole was
drilled into the MgO spacers in the same geometry as the capsule would residue
during a normal experiment, and the hole was filled with NaCl powder into which an
Au wire was placed. The sample was heated to 30 °C below the target temperature
and held for 2 h to ensure pressure stabilization. After that, temperature increased
by 10 °C/min to the target temperature. The experiment was held at the target
temperature and pressure for 6 min and was then quenched to room temperature.
The post-run sample was placed in a beaker filled with water, thereby dissolving the
NaCl. Au spheroids indicated melting, whereas relict wires remained in subsolidus
melting experiments (Fig. 3).
Analytical methods
Following sulfide melting experiments, the assembly was gradually decom-
pressed to minimize sample fracturing and the recovered capsules were mounted
in epoxy and ground to expose the medial section of the charges. Owing to the
low hardness of sulfide, the sample was impregnated with epoxy prior to grinding.
Polishing was conducted with diamond polishing pads, starting from 9.0 mm grit
down to 0.5 mm. Run products and standards were carbon coated and analyzed
using a JEOL JXA8900R electron microprobe by WDS analyses with an ac-
celerating voltage of 15 kV and a probe current of 20 nA. Counting times were
20 s on peak centroid and 10 s on each background for all elements. Oxygen was
measured using a multilayer crystal (LDE2 with 2d = 9.7 nm). Primary standards
used for major elements analysis were pyrite (FeS2) for Fe and S, Ni metal for
(Ni), chalcocite (Cu2S) for Cu and S, and magnetite (Fe3O4) for O. Troilite was
used as a secondary standard for Fe and S. Pyrite was used as a blank for oxygen.
We employed a focused beam on crystalline sulfides and a defocused beam (1–20
mm diameter) on quenched melts.
rEsults and discussion
Pressure and temperature calibration
Comparison of the quartz-coesite reaction at 1100 °C
bracketed by the PC with the reaction determined by Bohlen
and Boettcher (1982) and Bose and Ganguly (1995) suggests
a pressure correction of –0.4 GPa (Table 3). For the MA, the
coesite-quartz bracket 1230–1250 °C in this study falls between
the previously determined force-pressure curves determined
at low temperature (1000–1200 °C, Dasgupta et al. 2004) and
high temperature (1440–1700 °C, unpublished data) (Fig. 4a),
and suggests a similar force-pressure relationship to that from
enstatite-pyrope experiments in the same assembly at 1300–1400
°C (Tenner et al. 2012). Because the high-temperature and
low-temperature force-pressure (F-P) curves converge at high
pressure, the intermediate-temperature quartz-coesite and en-
statite-pyrope brackets can be used to construct a force-pressure
calibration curve valid from 3 to 8 GPa (Fig. 4a).
Comparison of gold melting experiments to the Au fusion
curve of Mirwald et al. (1975) and Mirwald and Kennedy (1979)
indicate that sample temperatures are 10–15 and 15–20 °C hot-
ter than thermocouple readings for the PC and MA assemblies,
respectively. The former is in good agreement with the 12 °C
offset determined in the PC from multiple thermocouple mea-
surements (Xirouchakis et al. 2001) and the latter is similar to
that estimated by previous studies in the MA (Dasgupta et al.
2004; Withers and Hirschmann 2007; Tenner et al. 2012). All
sulfide melting temperatures and pressures reported in Table 1
have been adjusted for the pressure and temperature calibrations
reported here.
The pressure calibration determined for the BaCO3 assem-
blies in the PC at 1100 °C suggest a friction correction of –12%,
which is nearly twice to that determined at comparable pressures
in the same assembly in the UMN laboratory at 1250 °C based
on the anorthite breakdown reaction (Xirouchakis et al. 2001).
The difference is presumably owing to greater friction at lower
Table 1. Summary of high-pressures studies on sulfide melting up to 15 GPa
No. Descriptiona Chemical formula Chemical composition (wt%) Capsule Apparatusb References
Fe Ni Cu O S
MSS (0.93) Fe0.69Ni0.23Cu0.01 S1.00 45.3 15.8 1.0 – 37.9 silica glass PC + MA This study
1a MSS (0.93) solidus Fe0.69Ni0.23Cu0.01 S1.00 45.3 15.8 1.0 – 37.9 silica glass PC Bockrath et al. (2004)
1b MSS (0.93) liquidus Fe0.69Ni0.23Cu0.01 S1.00 45.3 15.8 1.0 – 37.9 silica glass PC Bockrath et al. (2004)
2 MSS (1.06) solidus Fe0.79Ni0.25Cu0.03 S1.00 47.9 15.6 1.9 – 34.7 silica glass PC Ballhaus et al. (2006)
3 MSS (1.11) solidus Fe0.83Ni0.25Cu0.03 S1.00 48.9 15.5 1.9 – 33.7 silica glass PC Ballhaus et al. (2006)
4a Pyrrhotite (0.92) Fe0.92 S1.00 62.0 – – – 38.0 alumina PC Ryzhenko and Kennedy (1979)
4b Pyrrhotite (0.92) Fe0.92 S1.00 62.0 – – – 38.0 soda glass BA Sharp (1969)
5 Troilite Fe1.00 S1.00 63.5 – – – 36.5 – DAC Boehler (1992)
6a Fe-FeS Eutecticc Fe1.18 S1.00 69.0 – – – 31.0 alumina PC Ryzhenko and Kennedy (1979)
6b Fe-FeS Eutecticc Fe 1.18 S1.00 69.0 – – – 31.0 – DAC Boehler (1996)
6c Fe-FeS Eutecticc Fe1.18 S1.00 69.0 – – – 31.0 boron nitride BA Usselman (1975)
7a FeS-FeO Fe5.28 O1.08 S1.00 85.7 – – 5.0 9.3 MgO MA Urakawa (1987)
7b (Fe,Ni)S-FeO Fe4.41Ni0.49 O0.50 S1.00 78.2 9.1 – 2.5 10.2 MgO MA Urakawa (1987)
Notes: No. refers to the curves labeled in Figure 2.
a Number in parentheses represents metal/sulfur atomic ratio.
b PC = piston cylinder; MA = multi-anvil; DAC = diamond-anvil cell; BA = belt apparatus.
c The eutectic composition at 1 atm.
ZHANG AND HIRSCHMANN: MANTLE SULFIDE MELTING UP TO 8 GPA184
Table 2. Experimental results and sulfide compositions by electron microprobe analysis
Chemical composition (wt%)
P (GPa) T (°C ) t (h) na Results Cu O S Fe Ni Total M/Sb FFec FNic
Subsolidus runs: Solid only
A1078 0.80 1050 15.0 22 mss 0.9 0.4 37.3 45.8 15.7 100.1 0.93 – –
A1080 1.30 1100 13.0 24 mss 1.1 0.5 37.1 45.6 16.1 100.4 0.93 – –
B107 1.80 1100 24.0 12 mss 0.9 0.4 37.4 45.5 16.1 100.2 0.92 – –
A1081 1.80 1125 20.0 18 mss 1.1 0.5 36.6 45.7 16.1 99.9 0.95 – –
A1076 1.80 1150 14.0 19 mss 0.9 0.3 36.6 46.1 16.3 100.2 0.96 – –
B554 2.30 1200 18.0 18 mss 1.1 0.3 37.2 45.0 16.2 99.8 0.94 – –
B641 2.80 1200 12.0 12 mss 1.0 0.4 36.4 45.5 16.4 99.6 0.96 – –
A1231 2.80 1225 12.0 20 mss 0.9 0.4 36.7 45.5 16.2 99.7 0.94 – –
M749 3.00 1250 4.0 14 mss 0.9 0.3 37.1 45.6 15.8 99.7 0.94 – –
B556 3.30 1250 18.0 14 mss 1.0 0.3 36.9 45.6 16.6 100.5 0.95 – –
M721 3.75 1250 8.0 10 mss 1.0 0.4 37.6 45.2 16.1 100.3 0.92 – –
M742 4.80 1300 4.0 25 mss 1.0 0.3 36.5 45.6 16.4 99.8 0.96 – –
M734 5.85 1300 8.0 16 mss 1.1 0.3 37.3 45.2 16.2 100.0 0.93 – –
M744 6.90 1350 1.0 36 mss 1.1 0.3 36.4 45.1 16.7 99.7 0.96 – –
M746 7.95 1400 1.0 8 mss 1.0 0.3 36.9 45.7 16.5 100.4 0.95 – –
Between solidus and liquidus: Melt-mss pairs
A1079 0.80 1075 14.0 14 melt 2.2 0.3 36.1 39.9 22.4 100.9 0.99 31 33
18 mss 0.6 0.3 38.1 48.0 13.2 100.1 0.91
A1077 0.80 1100 14.0 12 melt 1.7 0.3 36.4 42.5 18.3 99.3 0.95 55 60
8 mss 0.8 0.2 37.9 49.1 13.1 101.1 0.93
B633 0.80 1150 8.0 26 melt 1.2 0.5 36.1 44.4 17.1 99.3 0.95 62 62
15 mss 0.6 0.3 37.8 47.3 14.7 100.8 0.92
A1087 1.30 1125 17.0 25 melt 2.7 0.3 36.6 41.3 19.6 100.4 0.96 32 40
31 mss 0.7 0.4 37.9 47.5 13.9 100.3 0.91
B617 1.30 1150 6.0 22 melt 1.2 0.3 36.5 44.2 17.7 99.9 0.96 58 63
14 mss 0.7 0.3 37.4 47.3 13.7 99.5 0.92
B618 1.30 1175 12.0 17 melt 1.3 0.4 37.1 44.9 17.1 100.8 0.94 73 70
10 mss 0.7 0.3 38.1 47.1 14.1 100.2 0.91
A1128 1.80 1175 6.0 15 melt 1.9 0.3 34.9 43.7 18.9 99.8 1.02 67 53
14 mss 0.6 0.2 37.9 49.2 13.1 100.9 0.93
A1075 1.80 1200 16.0 16 melt 1.6 0.2 35.8 43.9 17.9 99.4 0.99 61 64
16 mss 0.6 0.3 38.1 48.0 13.2 100.1 0.91
B607 1.80 1225 6.0 18 melt 1.2 0.3 36.4 45.0 16.8 99.7 0.96 81 79
20 mss 0.7 0.5 37.7 47.6 13.9 100.4 0.91
A1114 2.30 1225 18.0 12 melt 2.4 0.3 35.8 39.2 21.8 99.5 0.98 39 42
10 mss 0.7 0.2 37.6 49.5 12.1 100.1 0.92
A1177 2.30 1250 12.0 17 melt 1.4 0.3 36.6 44.1 17.4 99.8 0.95 56 63
14 mss 0.7 0.3 37.8 47.3 14.2 100.3 0.92
M743 2.70 1250 3.0 24 melt 2.4 0.4 36.0 38.9 22.5 100.2 0.97 20 30
16 mss 0.6 0.3 37.5 47.1 13.5 99.1 0.91
B556 2.80 1250 18.0 24 melt 2.4 0.3 35.8 39.2 21.8 99.5 0.98 22 26
31 mss 0.6 0.3 37.7 47.3 14.2 100.1 0.92
B564 2.80 1275 12.0 16 melt 1.1 0.3 37.0 44.4 17.3 100.2 0.95 59 69
12 mss 0.7 0.4 38.1 47.1 13.8 100.0 0.90
A1157 3.30 1275 0.5 24 melt 1.2 0.4 36.0 44.2 17.8 99.7 0.97 55 63
33 mss 0.6 0.3 37.5 47.1 13.5 99.1 0.91
M716 3.75 1300 8.0 12 melt 1.3 0.2 36.2 44.2 18.4 100.3 0.99 62 59
12 mss 0.8 0.2 37.8 47.6 13.0 99.4 0.91
M724 5.85 1350 8.0 24 melt 1.6 0.3 36.4 43.5 18.3 100.2 0.97 64 62
8 mss 0.8 0.2 37.9 49.1 12.8 100.8 0.93
M794 6.90 1400 1.0 12 melt 1.1 0.3 36.6 43.6 17.8 99.5 0.95
M798 7.95 1425 1.0 10 melt 1.3 0.5 36.1 43.2 18.5 99.7 0.96 36 44
12 mss 0.6 0.2 37.6 46.8 14.4 99.6 0.92
Superliquidus runs: Melt only
B636 0.80 1175 8.0 18 melt 1.0 0.4 36.4 45.2 16.4 99.3 0.95 – –
B600 0.80 1375 0.5 18 melt 1.0 0.5 37.1 45.6 16.0 100.2 0.93 – –
A1074 0.80 1200 22.0 22 melt 1.0 0.4 37.3 45.8 15.7 100.2 0.93 – –
B596 0.80 1350 4.0 47 melt 1.0 0.4 37.1 45.6 16.2 100.3 0.94 – –
A1171 1.30 1200 8.0 20 melt 1.1 0.5 36.6 45.7 16.1 99.9 0.95 – –
B636 0.80 1175 8.0 18 melt 1.0 0.4 36.4 45.2 16.4 99.3 0.95 – –
B600 0.80 1375 0.5 18 melt 1.0 0.5 37.1 45.6 16.0 100.2 0.93 – –
A1074 0.80 1200 22.0 22 melt 1.0 0.4 37.3 45.8 15.7 100.2 0.93 – –
B596 0.80 1350 4.0 47 melt 1.0 0.4 37.1 45.6 16.2 100.3 0.94 – –
A1171 1.30 1200 8.0 20 melt 1.1 0.5 36.6 45.7 16.1 99.9 0.95 – –
B589 1.30 1350 6.0 14 melt 1.0 0.3 36.9 45.6 16.6 100.5 0.95 – –
B590 1.30 1375 12.0 16 melt 1.1 0.3 37.3 45.2 16.2 100.0 0.93 – –
B591 1.30 1400 4.0 18 melt 1.1 0.5 37.4 45.8 15.8 100.6 0.92 – –
B536 1.80 1300 24.0 9 melt 1.0 0.4 36.8 45.6 15.9 99.6 0.94 – –
A1127 1.80 1450 12.0 17 melt 0.8 0.2 37.7 45.4 16.5 100.6 0.93 – –
B585 1.80 1425 4.0 21 melt 0.8 0.3 36.9 45.5 15.9 99.5 0.94 – –
A1063 1.80 1400 4.0 49 melt 0.9 0.2 37.3 45.0 16.3 99.7 0.93 – –
A1067 1.80 1350 18.0 8 melt 1.0 0.3 36.8 45.5 16.3 99.8 0.95 – –
(Continued on next page)
ZHANG AND HIRSCHMANN: MANTLE SULFIDE MELTING UP TO 8 GPA 185
temperature. Previous studies for BaCO3 assemblies at compa-
rable pressures have found similar friction corrections (15% Fram
and Longhi 1992; 9% McDade et al. 2002). In contrast to results
in the UMN laboratory, McDade et al. (2002) found friction in
the BaCO3 pressure assembly to be independent of temperature
from 1000–1600 °C. It remains unclear whether differences
in calibration are owing to subtle differences in assemblies or
piston-cylinder performance, but comparison between different
calibrations suggest that interlaboratory pressure accuracies for
PC experiments are, in best cases, ±0.1 GPa.
Phase relations, solidus, and liquidus determinations
Phase relations were determined chiefly from textural and
compositional observations based on backscattered electron
(BSE) images (Fig. 5) and electron microprobe analyses (Table
2). Combining both textures and compositions, phase relations
are determined on melt and mss. On the one hand, textures of
mss and melt are different based on the observation of optical
images and backscattered electron (BSE) images. Mss in the
post-run charge is homogenous granular grains in texture; melt
was quenched into crystals with a wormlike intergrowth texture.
On the other hand, Cu, Ni, Fe, and S are fractionated between
melt and mss if both phases coexist and equilibrate at high
temperature before quenching, with Cu and Ni being incompat-
ible in the Fe-Ni-Cu-S monosulfide system (discussion in the
Table 2.—Continued
Chemical composition (wt%)
P (GPa) T (°C ) t (h) na Results Cu O S Fe Ni Total M/Sb FFec FNic
B601 1.80 1250 12.0 34 melt 1.0 0.4 37.9 45.4 15.8 100.3 0.91 – –
A1162 1.80 1275 12.0 16 melt 1.0 0.3 36.9 45.6 16.6 100.5 0.95 – –
B609 2.30 1275 2.0 12 melt 1.1 0.4 37.0 45.3 16.4 100.2 0.94 – –
B594 2.30 1425 4.0 28 melt 1.0 0.3 37.5 45.1 16.0 99.8 0.92 – –
B599 2.30 1400 4.0 19 melt 1.0 0.3 38.4 45.0 15.9 100.7 0.90 – –
A1167 2.30 1300 6.0 30 melt 0.8 0.3 37.4 46.0 15.5 99.9 0.93 – –
A1151 2.80 1450 4.0 10 melt 0.9 0.5 37.1 45.3 15.8 99.6 0.92 – –
B615 2.80 1325 6.0 14 melt 0.9 0.5 36.4 45.0 16.4 99.2 0.94 – –
A1122 2.80 1300 12.0 12 melt 1.0 0.4 36.0 45.3 16.7 99.4 0.97 – –
A1153 2.80 1425 4.0 40 melt 1.2 0.4 36.9 45.0 15.9 99.5 0.93 – –
A1163 2.80 1350 12.0 38 melt 1.0 0.6 36.7 45.1 16.2 99.7 0.93 – –
A1160 3.30 1425 6.0 28 melt 1.0 0.3 38.0 45.3 16.2 100.8 0.91 – –
A1152 3.30 1450 4.0 16 melt 1.0 0.4 37.6 45.2 16.3 100.5 0.92 – –
A1161 3.30 1350 12.0 53 melt 1.0 0.5 37.0 45.2 16.2 100.0 0.93 – –
A1159 3.30 1300 12.0 18 melt 1.1 0.3 37.3 45.2 16.2 100.0 0.93 – –
A1180 3.30 1325 12.0 8 melt 0.9 0.6 37.4 44.6 16.1 99.7 0.90 – –
M797 6.90 1425 2.0 10 melt 1.1 0.4 37.4 44.2 16.6 99.6 0.92 – –
M796 7.95 1425 2.0 10 melt 1.0 0.4 36.4 45.8 16.4 100.0 0.96 – –
a Number of electron probe spot analyses averaged to obtain the reported elemental concentrations.
b Atomic metal/sulfur ratio, oxygen is regarded as replacing sulfur in mss or melts.
c Melt fractions (F) are calculated by mass balance of Fe and Ni according to F = (Cbulk – Cmss)/(Cmelt – Cmss), where Cbulk is the starting composition (Cbulk = 45.5 wt% Fe
or 16.2 wt% Ni); Cmelt is the concentration of Ni or Fe in the melt, and Cmss is the respective concentration in the crystalline phase.
FigurE 3. Optical images of Au recovered from pressure and
temperature calibration experiments. Gold spheroids are produced from
experiments that exceeded the melting temperature of Au, whereas relict
wires are observed from experiments that remained below the Au fusion
curve. (Color online.)
Table 3. Pressure and temperature calibration experiments using
the quartz–coesite transition and gold fusion
Quartz–coesite transition experiments
PC Run no. Uncorrected T (°C) t (h) Run product
pressure (GPa)
A1199 3.50 1100 6 coesite
A1204 3.70 1100 6 coesite
A1209 3.30 1100 6 quartz
A1210 3.10 1100 6 quartz
A1218 3.45 1100 6 quartz
A1219 3.40 1100 6 quartz
MA Run no. Force T (°C) t (h) Run product
(metric tons)
M767 165 1250 2 coesite
M770 160 1220 2 coesite
M771 155 1230 2 coesite
M749 145 1250 3 quar tz
M743 131 1250 3 quar tz
M762 155 1250 2 quar tz
Gold melting experiments
PC Run no. Corrected Tc reading (°C) t (h) Run product
pressure (GPa)
A1141 0.80 1086 0.1 gold wire
A1137 1.80 1145 0.1 gold wire
A1144 2.80 1200 0.1 gold wire
A1135 1.80 1155 0.1 gold ball
A1142 1.80 1150 0.1 gold ball
A1139 0.80 1099 0.1 gold ball
A1156 2.80 1220 0.1 gold ball
A1147 2.80 1210 0.1 gold ball
MA Run no. Corrected Tc reading (°C) t (h) Run product
pressure (GPa)
M762 3.23 1230 0.1 gold ball
M767 3.43 1232 0.1 gold ball
M770 3.33 1200 0.1 gold wire
M771 3.23 1210 0.1 gold wire
ZHANG AND HIRSCHMANN: MANTLE SULFIDE MELTING UP TO 8 GPA186
next section below). In addition, mss phases are homogenous
by composition at micrometer scale while quenched melts are
highly heterogeneous by composition at micrometer scale. This
is the reason for the employment of a focused beam on crystalline
sulfides and a defocused beam (1–20 mm diameter) on quenched
melts. From the perspective of polishing, due to the hardness
contrast between quartz and sulfide, post-run products were
polished with prevalent cracks on the surface. During polishing,
subsolidus aggregates tend to disintegrate, whereas superliquidus
and partially melted experiments do not tend to disintegrate.
Exposed charges were unavoidably pervasively cracked dur-
ing polishing, owing to the hardness contrast between sulfide
charges and surrounding silica minerals. In some cases, this
prevented high-quality polishes across the entire charge, but
sufficient material was always exposed to allow textural and com-
positional analysis. Selected BSE images are shown in Figure 5
and electron microprobe analyses are reported in Table 2. The
subsolidus samples consist of a homogeneous single monosulfide
(mss) phase with granular texture and grain diameters typically
100–300 mm (Fig. 5a) with a composition that in all cases is
within analytical uncertainty of the starting material (Fig. 5a).
Experiments that underwent partial melting produced two phases
(Fig. 5b): a homogenous Fe-rich mss phase granular mss crystals
typically >100 mm and a Ni-Cu rich heterogeneous phase mate-
rial that, following quench, consists of smaller (50 mm) grains
interspersed with darker (on BSE) with wormlike intergrowths
between grains textures. Crystals and quenched melt are well-
segregated by gravity. Electron microprobe analyses indicate
compositions that are more heterogeneous than subsolidus mss,
but the bulk compositions are comparable to that of the start-
ing composition (Fig. 5b). Experiments inferred to have been
superliquidus quench to textures consisting of two sulfide phases
in a wormlike intergrowth, similar to those produce by partially
molten samples with the same composition as the starting material
within starting powder mixing uncertainty and analytical uncertainty
(Fig. 5c). Additional textural evidence that aided the interpretation of
solidus location was the behavior during polishing: subsolidus ag-
gregates tended to disintegrate, whereas superliquidus and partially
melted experiments were more cohesive.
Crystalline mss produced in subsolidus experiments are
compositionally homogeneous within analytical uncertainty
and are not distinguishable from the bulk composition (Table
2). When melt and crystals coexist at high temperature, granular
sulfides are compositionally homogeneous and enriched in Fe
and S and poor in Ni and Cu relative to the bulk composition.
In comparison, the quenched melt phase is heterogeneous on
a micrometer scale, necessitating analysis with an unfocused
electron beam. The quenched melt phase is depleted in Fe and
S and enriched in Ni and Cu relative to the crystalline solids or
the bulk composition. In one case (Experiment M794, 6.9 GPa,
1400 °C), quenched melt was observed but coexisting granular
crystals were not evident. However, sub-liquidus conditions
were inferred because the melt composition was enriched in
Cu and Ni relative to the bulk composition. Melts interpreted
to be quenched from superliquidus conditions based on textural
features are also heterogeneous, but have compositions within
analytical uncertainty of the starting composition.
Experimental determinations of the mss solidus and liquidus
up to 8 GPa can be fit with an empirical relation that minimizes
the disagreement between observations and the computed curve
according to a penalty function as described by Hirschmann
(2000), for which an objective variable, Y, given by Y = S(Ti)2,
is minimized, where Ti is given by
Ti = (Ti – Tmodel)2; if the assemblage observed at Ti disagrees with Tmodel
Ti = 0; if the assemblage observed at Ti agrees with Tmodel
FigurE 4. Revised pressure and temperature calibrations of multi-
anvil device at UMN using the 18/12 TEL/OEL assembly. (a) Pressure-
force relations of the best t functions are P = –1.15 × 10-5 F2 + 2.44 ×
10-2 F + 4.56 × 10-3 for 1000–1200 °C (Dasgupta et al. 2004), P = –1.29
× 10-5 F2 + 2.61 × 10-2 F – 5.17 × 10-3 for 1200–1400 °C (this study),
and P = –1.65 × 10-5 F2 + 3.03 × 10-2 F – 1.76 × 10-3 for 1400–1700 °C
(unpublished data), where P is pressure (GPa) and F is force (tons). The
blue square corresponds to 170 tons and 3.49–3.53 GPa (Tenner et al.
2009) and the purple circles represent coesite and red diamonds represent
quartz. (b) Temperature calibration results on PC and MA. Thermocouple
temperatures and corrected sample pressures are plotted together with gold
melting curve calibrated with previous calibration work by Mirwald and
Kennedy (1979). Lower temperatures read by thermocouples compared to
the fusion curve reect the offset in temperature between the thermocouple
location and the sample hotspot. (Color online.)
ZHANG AND HIRSCHMANN: MANTLE SULFIDE MELTING UP TO 8 GPA 187
and Tmodel is the temperature calculated from the solidus or
liquidus curve. We examined several types of functions to fit
to the solidus and liquidus brackets, including polynomial and
logarithmic forms, and though the differences among these fits
are small, ultimately chose to employ an equation similar to that
proposed by Simon and Glatzel (1929),
T
=Tref
P
a
+1
⎛
⎝
⎜⎞
⎠
⎟
c
where T is temperature (°C) and P is pressure (GPa). For the
solidus up to 8 GPa, we find Tref = 1015.1 °C, a = 1.88, and c =
0.206 (0 ≤ P ≤ 8) (Y = 768), and, for the liquidus, Tref = 1067.3
°C, a = 1.19, and c = 0.149 (Fig. 6).
In this study, the solidus fit T (°C) = 1015.1 [P(GPa)/1.88 + 1]0.206,
is thought to be accurate within ±15 °C over the pressure interval
from 1 to 3.5 GPa and within ±30 °C from 3.5 to 8.0 GPa, owing
to different temperature uncertainties in PC and MA devices and
different P-T densities of experimental brackets (typically 25 °C
for PC and 50 °C for MA experiments). The fit from 0.1 MPa to
0.8 GPa might be less accurate because brackets constraining
solidus at 0.1 MPa were taken from previous studies and the
present study included no experiments between 0.1 MPa to 0.8
GPa. Therefore, this solidus is constrained chiefly to mantle
depths of 30 to 250 km.
At high pressures, the solidus and liquidus curves tend to
converge, producing a narrowed melting interval. This may be an
artifact of imperfections in the regressed curves, with the Simons
equation predicting solidus temperatures that are near the high
limit of the experimental brackets, and the liquidus curve poorly
constrained owing to few high-pressure observations.
The melting interval for mss can be compared to that de-
termined by Bockrath et al. (2004) for an apparently identical
composition in similar capsules up to 3.5 GPa (Fig. 7). Solidus
temperatures are consistent, but the liquidus found by Bockrath
et al. (2004) extends to much higher temperature; e.g., at 3 GPa
Bockrath et al. (2004) observed mss coexisting with melt up to
1400 °C, but the new results place the liquidus near 1275 °C. As
noted by Hart and Gaetani (2006), the high melting temperatures
indicated by Bockrath et al.’s experiments suggest a concave
upward liquidus slope and, at pressures near 3 GPa, liquidus
temperatures hotter than the melting of pure Fe1–xS (Ryzhenko
and Kennedy 1973) (Fig. 1), both of which are unlikely. These
inconsistencies are absent from the new liquidus curve.
The melting interval for mss determined in this study is
intermediate between the high-temperature fusion curves of
pure troilite, FeS, or pyrrhotite, Fe1–xS, and eutectic melting in
the system FeS-Fe (Fig. 1). The pyrrhotite studied by Ryzhenko
and Kennedy (1973) had similar metal/sulfide stoichiometry to
the mss studied here (Table 1), meaning that the lower melting
temperature of the mss is owing chiefly to the effects of Fe-Ni
solid solution. This is consistent with many previous studies that
have found that Ni reduces the melting temperature of monosul-
fide (Hsieh et al. 1987; Urakawa et al. 1987).
Partial melting and fractionation of elements between
melt/mss
For experiments that produce coexisting melt and monosul-
fide crystals, melt fractions calculated independently from Fe and
Ni mass balance (Table 2) agree well with one another (Fig. 8).
Melt fractions calculated from Ni are considered most accurate
because of the strong mineral/melt partitioning and high con-
FigurE 5. Backscattered electron images of typical textures from
quenched experiments exemplifying (a) super-liquidus conditions (B536:
1.8 GPa, 1300 °C), (b) partially molten (A1075: 1.8 GPa, 1200 °C), and
(c) subsolidus (B534: 1.8 Ga, 1100 °C). (Color online.)
ZHANG AND HIRSCHMANN: MANTLE SULFIDE MELTING UP TO 8 GPA188
centrations (compared to Cu) produce phase compositions that
are most strongly separated compared to analytical uncertainties.
Therefore, Ni concentration in melts is chosen to represent melt
evolution. In all samples, the melts have slightly higher metal/
sulfur ratio than their coexisting residue mss (Table 2). Ni and
Cu preferentially partition into melts and Fe is concentrated in
residue mss (Fig. 9). The detection limit for oxygen established
by analyzing pyrite is ~0.05 wt%, with the analytical uncertainty
of ±0.1 wt%. Oxygen is detected in small amounts in both
mss and melt. Oxygen concentrations in the post-run products
vary from 0.2–0.6 wt%, with slightly greater concentrations in
quenched melts compared to crystals (Table 2). The concentra-
tions detected in our study are comparable to those (~0.37 wt%)
found previously in pyrrhotite annealed at 700 °C (Graham et
al. 1987; Graham and McKenzie 1987).
Partial melt compositions have higher M/S ratios than re-
sidual mss solids (Table 2), consistent with relations established
previously for melting of mss (Kullerud 1963; Bockrath et al.
2004; Ballhaus et al. 2006) and in simple systems (Ryzhenko
and Kennedy 1973). MSS compositions change comparatively
little, being restricted chiefly to M/S ratios between 0.90–0.93
owing chiefly to the constraints of M/S stoichiometry in mss
solid solutions (Ballhaus et al. 2006), whereas melt composi-
tions are more variable (M/S = 0.94–1.02). Partial melts are
also enriched in Ni and especially Cu compared to coexisting
mss, while Fe concentrations are similar in the two phases. This
is as expected owing to the respective radii of the cations (Cu
> Ni > Fe). Mineral/melt partition coefficients for Cu and Ni
become more extreme at low melt fractions, likely owing to the
increased Ni/(Fe+Ni) of the melts (Fig. 9). Based on the enrich-
ment or depletion of Cu and Ni, sulfides from mantle xenoliths
or diamond inclusions has been interpreted as either trapped
melts or residual mss crystals (Bulanova et al. 1996; Guo et al.
1999; Lorand and Alard 2001). Such interpretations are broadly
consistent with the experimental compositions observed here.
The variations in metal/silicate and Cu-Ni-Fe fractionations
with melt fraction and temperature are linked to the constraints
on mss stoichiometry and the influence of Ni on partitioning
behavior. These relationships vary only weakly with pressure, as
has been previously documented for M/S partitioning (Ballhaus
et al. 2006) and Ni partitioning (Jones and Walker 1991). The lack
of pressure influence on partitioning suggests that the melting
interval between solidus and liquidus curves should not become
significantly wider or narrower with pressure, and reinforces the
assertion above that the liquidus, which is under-constrained at
pressures above 3.5 GPa, likely is slightly hotter at high pressure
than indicated by the parameterization.
Sulfide composition in the mantle
The sulfide solidus is affected by composition, which in the
mantle, is controlled by reactions between sulfides and coexisting
silicate minerals (Eggler and Lorand 1993; Gaetani and Grove
1997). Solidus temperatures are lower when sulfide has higher
M/S ratios (e.g., Ballhaus et al. 2006), and when additional com-
ponents (such as C or O) are soluble (Ballhaus et al. 2006; Chi et
al. 2014; Dasgupta et al. 2009; Gunn and Luth 2006; Urakawa et
al. 1987). The sulfide M/S (0.93) investigated in this study is on
the low end of natural mss (0.9–1.1) (Fig. 2), with low content
of oxygen and nominally carbon-free.
Figu rE 6. Experimentally determined melting relations of
Fe0.69Ni0.23Cu0.02S1.00. Blue diamonds are sub-solidus mss, with lled
diamonds from piston-cylinder (PC) experiments and empty diamonds
from those conducted in the multi-anvil (MA), respectively. Red and
orange circles are melt-mss pairs by PC and MA, respectively. Green
and black triangles are superliquidus runs by PC and MA, respectively.
Solid curve is the parameterization of the solidus, given by T (°C) =
1015.1[P(GPa)/1.88 + 1]0.206 and the dashed curve is liquidus, given by
T (°C) = 1067.3[P(GPa)/1.19 + 1]0.149. Melting relations at 0.1 MPa are
taken from Bockrath et al. (2004). (Color online.)
FigurE 7. Comparison of mss melting relations calibrated from
experiments in this study (Fig. 6) with experiments done by Bockrath
et al. (2004). Diamonds are sub-solidus mss; circles are melt-mss pairs;
triangles are superliquidus runs. The square bracket with a red lling
and green triangle is the “problem bracket,” which is labeled as “melt
+ solid” by Bockrath et al. (2004), but is listed as “solid only” in their
supplementary material. Our results suggest that the experiment should
have been completely melted at the stated run conditions. The solidus
and liquidus derived from the parameterization of the present study (Fig.
6) are shown as solid and dashed curves, respectively. (Color online.)
ZHANG AND HIRSCHMANN: MANTLE SULFIDE MELTING UP TO 8 GPA 189
In the shallow oxidized mantle, oxygen dissolved in sulfide
might be high (Gaetani and Grove 1999) and drive the sulfide
solidus downward (Gunn and Luth 2006; Urakawa et al. 1987).
As the mantle becomes more reduced with increasing depth
(Frost and McCammon 2008), the oxygen content should dimin-
ish, but this can be countered by increases in the M/S ratio owing
to an increase in metal activity, as the reactions
Fe2SiO4(ol) = FeSiO3(opx) + ½O2 + Fe(sulfide)
Ni2SiO4(ol) = NiSiO3(opx) + ½O2 + Ni(sulfide)
shift to the right. Equivalently, Eggler and Lorand (1993) suggest
that oxygen fugacity (fO2) and sulfur fugacity (fs2) are positively
correlated for peridotite-sulfide systems. Reactions similar to:
Fe2SiO4(ol) + ½S2 = FeSiO3(opx) + FeS(sulfide) + ½O2
also produce increases in M/S ratio of the sulfide as oxygen
fugacity decreases. Furthermore, as mantle enters the graphite/
diamond stability field (>4 GPa) (Frost and McCammon 2008),
dissolved carbon will likely further depress sulfide solidus
temperatures.
All of these considerations together suggest that the sulfide
solidus constrained in this study is near an upper bound. At low
pressures and more oxidized conditions, the solidus would be
lowered by greater dissolved oxygen, and at higher pressures
and more reduced conditions, it would be diminished by higher
M/S ratio and dissolved C.
Sulfide melts in the mantle
The solidus and liquidus of monosulfide constructed from
this study is below the calculated mantle adiabat (Katsura et al.
2010) up to the limit of the experimental data at 8 GPa (Fig. 10),
meaning that sulfide is molten in the convecting mantle at least
to depths of 250 km. We note that the composition investigated
has comparatively low M/S ratio, and that sulfides with higher
ratios melt at lower temperature (e.g., Ballhaus et al. 2006), so
that melting is expected for most or all mantle sulfide composi-
tions. The steeper temperature/pressure slope of the melting
curves compared to the adiabat may indicate that sulfide becomes
crystalline deeper in the mantle.
Increased M/S should in turn decrease the sulfide crystal-
lization temperature (e.g., Ballhaus et al. 2006). If conditions at
depth become sufficiently reducing such that a metal alloy phase
is stabilized (Frost et al. 2004; Rohrbach et al. 2007, 2011), then
sulfide melting is controlled by the (Fe,Ni)S-(Fe,Ni) eutectic,
which remains below the adiabat at least until pressures reached
deep in the lower mantle (>60 GPa; Campbell et al. 2007; Morard
et al. 2011; Stewart et al. 2007). Thus, sulfide melt may in fact
be stable throughout the convecting upper mantle and perhaps
into the transition zone.
The sulfide melting curves intersects geotherms for typical
oceanic lithosphere shallower than the intersection with the
adiabat, meaning sulfide could be partially molten in the thermal
boundary layer and deeper portions of the oceanic lithosphere
(Fig. 10). In the continental lithosphere, sulfides are less likely
molten in the colder settings such as the root of cratons. Of
course, sulfides with higher M/S ratios and dissolved carbon
could become molten at lower temperatures.
Sulfide inclusions in diamond
Sulfide inclusions are abundant in diamond, and are typically
Ni-rich (22–36 wt%) in peridotitic parageneses (age >3 Ga) and
Ni-poor (0–12 wt%) in eclogitic diamonds (age <3 Ga), which
corresponds to high bulk Ni contents (~3000 ppm) in peridotite
or and low bulk Ni (~300 ppm) concentrations typically found
in eclogite (Bulanova et al. 1996; Pearson et al. 2003; Shirey and
FigurE 8. Comparison of melt fractions calculated by Fe and Ni
mass balance. Calculations are in close agreement, but those from Ni
are believed to be more accurate because concentrations of Ni are more
distinct in mss and partial melts, giving more leverage to calculated mass
balances. (Color online.)
FigurE 9. Variations in mss/melt partition coefcient of Cu, Ni,
and Fe as a function of Ni content of the melt. Partitioning of Cu and
Ni between crystalline and solid sulde is more extreme for Ni-rich
smaller-degree partial melts, and less so for higher-degree Fe-rich melts.
(Color online.)
ZHANG AND HIRSCHMANN: MANTLE SULFIDE MELTING UP TO 8 GPA190
Richardson 2011; Stachel and Harris 2008). The overabundance
of sulfide in diamond suggests that sulfide acts either as a re-
ducing reagent or as a co-precipitating product during diamond
formation, more than likely as a liquid phase (Bulanova 1995;
Stachel and Harris 2008; Westerlund et al. 2006). Although the
solidus of monosulfide investigated here is hotter than typical
diamond P-T, molten sulfide could be trapped in diamond if
it originates from compositions with comparatively high M/S
ratios, or if additional components (such as C or O) lower the
solidus (Ballhaus et al. 2006; Gunn and Luth 2006). In addition,
for sulfides that were trapped as inclusions in large melt extrac-
tion process from peridotite, sulfides might be molten due to this
high-temperature process in the earlier geological history. This
is indicated by the mantle sulfides Re-Os modal age peaks of
1.2, 1.9, and 2.7 billion years that match similar periods of high
tectonic activity recorded in zircon populations (Pearson et al.
2007). If sulfides are trapped as liquids, then they are syngenetic
with their hosts and geochronologic interpretations of their Re-Os
or Pb-Pb isotope systematics can be related to the timing of the
formation of the enclosing minerals.
Mobility of sulfide melts
Having established that sulfide is molten or partially mol-
ten in significant portions of the mantle, a natural question is
whether such melts are mobile. Sulfide melt mobility is affected
by oxygen fugacity and pressure in the mantle (Gaetani and
Grove 1999; Shannon and Agee 1998; Shi et al. 2013). At con-
ditions similar to the fayalite-magnetite-quartz oxygen buffer,
molten sulfide may potentially dissolve up to 9 wt% oxygen
(Gaetani and Grove 1999) and sulfide melt can potentially form
an interconnected network along silicate grain edges, as the
olivine–melt dihedral angle is 52° (Gaetani and Grove 1999).
As conditions become more reduced and dissolved oxygen
diminishes, the dihedral angle increases (e.g., 90° at near the
iron-wüstite buffer, Gaetani and Grove 1999) and intercon-
nectivity is less likely. At lower mantle conditions, where the
principal silicate is perovskite rather than olivine, high dihedral
angles for Fe-Ni-S melts prevail (Shannon and Agee 1998; Shi
et al. 2013), inhibiting connectivity. Therefore, in the shallow
mantle (e.g., 30–60 km), sulfide melts, if stable, potentially
form an interconnected network in olivine-rich rocks. In the
deep lithosphere (e.g., >120 km), percolation may be limited
as conditions become more reduced (Frost and McCammon
2008). Additionally, sulfide melt transport may be coupled to
movement of associated silicate or carbonate melts (Delpech et
al. 2012; Lorand et al. 2004). In some cases, natural peridotite
textures indicate sulfide wetting silicate grain boundaries (Lo-
rand et al. 2013). Furthermore, geochemical variations observed
in xenoliths may not be easily explicable without sulfide mass
transport (Lorand 1989; Lorand et al. 2013).
Sulfide melt mobility may be a key factor determining their
geophysical and geochemical influence. For example, despite
molten sulfide’s extreme physical properties relative to silicate
rock, it may not impart significant geophysical anomalies simply
by melting in situ, as the melt fraction for rocks with typical
mantle S concentrations (200 ppm, McDonough and Sun 1995)
would be merely 0.05–0.10 vol%. If sulfide melts can migrate
and concentrate in isolated areas, they may feasibly produce
noticeable effects on shear wave velocities (e.g., Helffrich
et al. 2011) and other properties. Likewise, sulfide melts are
most likely to be responsible for geochemical variations in
chalcophile or PGE elements (e.g., Alard et al. 2002; Powell
and O’Reilly 2007) if they are mobile.
iMplications
We emphasize that we have investigated only a single com-
position of sulfide Fe0.69Ni0.23Cu0.01S1.00 (M/S = 0.93), which is
calculated as in equilibrium with mantle olivine at FMQ, with
trace amounts of oxygen (and likely carbon). As the upper
mantle becomes more reduced with depth (Frost and McCam-
mon 2008), the M/S ratio of sulfide should increase, leading to
further depression of the sulfide solidus (Ballhaus et al. 2006;
Rohrbach et al. 2007). Also, in the oxidized shallow mantle (e.g.,
<3 GPa), sulfides melts should have higher content of oxygen
and in the graphite/diamond stability field (e.g., >4 GPa), dis-
solved carbon should increase, and both of these should affect
melting temperatures. Therefore, the solidus constrained in this
work is likely an upper bound on sulfide melting in the Earth’s
upper mantle.
acknowlEdgMEnts
We are grateful for the constructive reviews from James Brenan and Glenn
Gaetani. We appreciate the aid and advice of UMN colleagues: Anthony Withers in
the experimental petrology laboratory, Anette von der Handt in the electron micro-
FigurE 10. Comparison of mss solidus (solid black curve) and
liquidus (dashed black curve) with mantle geotherms and domains of
potential silicate melting. The mantle adiabat is given in the solid red line,
with the shaded area representing the temperature uncertainty (Katsura
et al. 2010). A geotherm applicable to oceanic lithosphere of plate ages
ranging from 20 to 80 Ma is calculated from a cooling half-space model
(Turcotte and Schubert 2002) and representative continental geotherms
are shown in the shaded dark areas, bounded by calculations for heat
ows of 40 and 50 mW/m2 (Pollack and Chapman 1977). Regions of
diamond formation as inferred from inclusion thermobarometry are
from Stachel and Harris (2008) in green loop and Shirey et al. (2013) in
pink loop. The solidus of nominally anhydrous peridotite is shown as
the green solid curve (Hirschmann 2000); hydrous peridotite (0.1 wt%
bulk water) is the blue solid curve (Katz et al. 2003). (Color online.)
ZHANG AND HIRSCHMANN: MANTLE SULFIDE MELTING UP TO 8 GPA 191
probe laboratory, and Chris Crosby in the geobiology laboratory. We acknowledge
the support of grants NSF EAR1119295 and EAR1426772.
rEFErEncEs citEd
Alard, O., Griffin, W.L., Lorand, J.P., Jackson, S.E., and O’Reilly, S.Y. (2000)
Non-chondritic distribution of the highly siderophile elements in mantle
sulfides. Nature, 407, 891–894.
Alard, O., Griffin, W.L., Pearson, N.G., Lorand, J.P., and O’Reilly, S.Y. (2002) New
insights into the Re/Os systematics of sub-continental lithospheric mantle from
in-situ analysis of sulfides. Earth and Planetary Science Letters, 203, 651–663.
Aulbach, S., Griffin, W.L., Pearson, N.J., O’Reilly, S.Y., Kivi, K., and Doyle, B.J.
(2009) Mantle formation and evolution, Slave Craton: constraints from HSE
abundances and Re-Os isotope systematics of sulfide inclusions in mantle
xenocrysts. Chemical Geology, 208, 61–88.
Ballhaus, C., Bockrath, C., Wohlgemuth-Ueberwasser, C., Laurenz, V., and Berndt,
J. (2006) Fractionation of the noble metals by physical processes. Contributions
to Mineralogy and Petrology, 152, 667–684.
Bockrath, C., Ballhaus, C., and Holzheid, A. (2004) Fractionation of the platinum
group elements during mantle melting. Science, 305, 1951–1954.
Boehler, R. (1992) Melting of the Fe-FeO and the Fe-FeS systems at high pres-
sure: Constraints on core temperatures. Earth and Planetary Science Letters,
111, 217–227.
——— (1996) Melting and element partitioning Fe-FeS temperatures to 620 kbars.
Physics of the Earth and Planetary Interiors, 96, 181–186.
Bohlen, S., and Boettcher, A.L. (1982) The quartz-coesite transformation: a precise
determination and the effects of other components. Journal of Geophysical
Research, 87, 7073–7078.
Bose, K., and Ganguly, J. (1995) Quartz-coesite transition revisited: reversed
experimental determination at 500–1200 °C and retrieved thermochemical
properties, American Mineralogist, 80, 231–238.
Bulanova, G.P. (1995) The formation of diamond. Journal of Geochemical Explora-
tion, 53, 1–23, http://dx.doi.org/10.1016/0375-6742(94)00016-5.
Bulanova, G.P., Griffin, W.L., Ryan, C.G., Shestakova, O.Y., and Barnes, S.J. (1996)
Trace elements in sulfide inclusions from Yakutian diamonds. Contributions
to Mineralogy and Petrology, 124, 111–125.
Campbell, A.J., Seagleb, C.T., Heinzb, D.L., Shend, C.G., and Prakapenkae, V.B.
(2007) Partial melting in the iron–sulfur system at high pressure: A synchro-
tron X-ray diffraction study. Physics of the Earth and Planetary Interiors,
162, 119–128.
Chi, H., Dasgupta, R., Duncan, M.S., and Shimizu, N. (2014) Partitioning of carbon
between Fe-rich alloy melt and silicate melt in a magma ocean—implications
for the abundance and origin of volatiles in Earth, Mars, and the Moon. Geo-
chimica et Cosmochimica Acta, 139, 447–471.
Dasgupta, R., Hirschmann, M.M., and Withers, C.W. (2004) Deep global cycling
of carbon constrained by the solidus of anhydrous, carbonated eclogite under
upper mantle conditions. Earth and Planetary Science Letters, 227, 73–85.
Dasgupta, R., Buono, A., Whelan, G., and Walker, D. (2009) High-pressure melt-
ing relations in Fe–C–S systems: Implications for formation, evolution, and
structure of metallic cores in planetary bodies. Geochimica et Cosmochimica
Acta, 73, 6678–6691.
Delpech, G., Lorand, J.P., Grégoirec, M., Cottind, J.Y., and O’Reilly, S.Y. (2012)
In-situ geochemistry of sulfides in highly metasomatized mantle xenoliths
from Kerguelen, southern Indian Ocean. Lithos, 154, 296–314.
Eggler, D.H., and Lorand, J.P. (1993) Mantle sulfide geobarometry. Geochimica
et Cosmochimica Acta, 57, 2213–2222.
Fei, Y.W., Bertka, C.M., and Finger, L.W. (1997) High-pressure iron-sulfur
compound, Fe3S2, and melting relations in the Fe-FeS system. Science, 275,
1621–1623.
Fram, M., and Longhi, J. (1992) Phase equilibria of dikes associated with Protero-
zoic anorthosite complexes. American Mineralogist, 77, 605–616.
Frost, D.J., and McCammon, C.A. (2008) The redox state of Earth’s mantle. Annual
Review of Earth and Planetary Sciences, 36, 389–420.
Frost, D.J., Liebske, C., Langenhorst, F., McCammon, C.A., Trønnes, R.G., and
Rubie, D.C. (2004) Experimental evidence for the existence of iron-rich metal
in the Earth’s lower mantle. Nature, 428, 409–412.
Gaetani, G.A., and Grove, T.L. (1997) Partitioning of moderately siderophile ele-
ments among olivine, silicate melt, and sulfide melt: Constraints on core forma-
tion in the Earth and Mars. Geochimica et Cosmochimica Acta, 61, 1829–1846.
——— (1999) Wetting of mantle olivine by sulfide melt: implications for Re/Os
ratios in mantle peridotite and late-stage core formation. Earth and Planetary
Science Letters, 169, 147–163.
Graham, J., and McKenzie, C.D. (1987) Oxygen in pyrrhotite: 2. Determination of
oxygen in natural pyrrhotite. American Mineralogist, 72, 605–609.
Graham, J., Bennett, C.E.G., and Riessen, A. (1987) Oxygen in pyrrhotite: 1. Ther-
momagnetic behavior and annealing of pyrrhotites containing small quantities
of oxygen. American Mineralogist, 72, 599–605.
Gunn, S.C., and Luth, R.W. (2006) Carbonate reduction by Fe-S-O melts at high
pressure and high temperature. American Mineralogist, 91, 1110–1116.
Guo, J.F., Griffin, W.L., and O’Reilly, S.Y. (1999) Geochemistry and origin of
sulphide minerals in mantle xenoliths: Qilin, Southeastern China. Journal of
Petrology, 40, 1125–1149.
Hart, S.R., and Gaetani, G.A. (2006) Mantle Pb paradoxes: the sulfide solution.
Contributions to Mineralogy and Petrology, 152, 295–308.
Helffrich, G., Kendall, M., Hammond, J.O.S., and Carroll, M.R. (2011) Sulfide
melts and long-term low seismic wavespeeds in lithospheric and asthenospheric
mantle. Journal of Geophysical Research, 38, 11,301–11,305, http://dx.doi.
org/10.1029/2011GL047126.
Hirschmann, M.M. (2000) Mantle solidus: Experimental constraints and the effects
of peridotite composition. Geochemistry, Geophyics, Geosystems, 1, 1042.
Hsieh, K.C., Schmid, R., and Chang, Y.A. (1987) The Fe-Ni-S system II. A Ther-
modynamic model for the ternary monosulfide phase with the nickel arsenide
structure. High Temperature Science, 23, 39–52.
Huang, S., Lee, C.T., and Yin, Q.Z. (2014) Missing lead and high 3He/4He in ancient
sulfides associated with continental crust formation. Nature Scientific Reports,
4, 5314, http://dx.doi.org/10.1038/srep05314.
Jones, J.H., and Walker, D. (1991) Partitioning of siderophile elements in the Fe-Ni-
S system: 1 bar to 80 kbar. Earth and Planetary Science Letters, 105, 127–133.
Katsura, T., Yoneda, A., Yamazaki, D., Yoshino, T., and Ito, E. (2010) Adiabatic
temperature profile in the mantle. Physics of Earth and Planetary Interiors,
183, 212–218.
Katz, R.F., Spiegelman, M., and Langmuir, C.H. (2003) A new parameterization
of hydrous mantle melting. Geochemistry, Geophyics, Geosystems, 4, 1073.
Kullerud, G. (1963) Thermal stability of pentlandite. Canadian Mineralogist, 7,
353–366.
Li, Y., and Audétat, A. (2012) Partitioning of V, Mn, Co, Ni, Cu, Zn, As, Mo, Ag,
Sn, Sb, W, Au, Pb, and Bi between sulfide phases and hydrous basanite melt
at upper mantle conditions.
Lorand, J.P. (1989) Abundance and distribution of Cu-Fe-Ni sulfides, sulfur,
copper and platinum-group elements in orogenic-type spinel lherzolite mas-
sifs of Arirge (northeastern Pyrenees, France). Earth and Planetary Science
Letters, 93, 50–64.
Lorand, J.P., and Alard, O. (2001) Platinum-group element abundances in the up-
per mantle: New constraints from in situ and whole-rock analyses of Massif
Central xenoliths (France). Geochimica et Cosmochimica Acta, 65, 2789–2806.
Lorand, J.-P., Delpech, G., Grégoire, M., Moine, B., O’Reilly, S.Y., and Cottin,
J.-Y. (2004) Platinum-group elements and the multistage metasomatic history
of Kerguelen lithospheric mantle (South Indian Ocean). Chemical Geology,
208, 195–215.
Lorand, J.P., Luguet, A., and Alard, O. (2013) Platinum-group element systemat-
ics and petrogenetic processing of the continental upper mantle: A review.
Lithos, 164–167, 2–21.
Luo, S.N., Mosenfelder, J.L., Asimow, P.D., and Ahrens, T.J. (2002) Direct shock
wave loading of Stishovite to 235 GPa: Implications for perovskite stability
relative to an oxide assemblage at lower mantle conditions. Journal of Geo-
physical Research, 29, http://dx.doi.org/10.1029/2002GL015627.
McDade, P., Wood, B.J., Van Westrenen, W., Brooker, R., Gudmundsson, G., Sou-
lard, H., Najorka, J., and Blundy, J. (2002) Pressure corrections for a selection
of piston-cylinder cell assemblies. Mineralogical Magazine, 66, 1021–1028.
McDonough, W.F., and Sun, S.S. (1995) The composition of the Earth. Chemical
Geology, 120, 223–253.
Mirwald, P.W., and Kennedy, G.C. (1979) The melting curve of gold, silver,
and copper to 60 Kbar pressure: a reinvestigation. Journal of Geophysical
Research, 84, 6750–6756.
Mirwald, P.W., Getting, I.C., and Kennedym, G.C. (1975) Low-friction cell for
piston-cylinder high-pressure apparatus. Journal of Geophysical Research,
80, 1519–1525.
Morard, G., Andrault, D., Guignot, N., Siebert, J., Garbarino, G., and Antonangeli,
D. (2011) Melting of Fe–Ni–Si and Fe–Ni–S alloys at megabar pressures:
implications for the core–mantle boundary temperature. Physics and Chemistry
of Minerals, 38, 767–776.
Mungall, J.E., and Su, S.G. (2005) Interfacial tension between magmatic sulfide and
silicate liquids: Constraints on kinetics of sulfide liquation and sulfide migra-
tion through silicate rocks. Earth and Planetary Science Letters, 234, 135–149.
Pearson, D.G., and Wittig, N. (2014) The formation and evolution of cratonic
mantle lithosphere: Evidence from mantle xenoliths. Treatise on Geochem-
istry, 2, 255–292.
Pearson, D.G., Shirey, S.B., Harris, J.W., and Carlson, R.W. (1998) Sulphide inclu-
sions in diamonds from the Koffiefontein kimberlite, S Africa: constraints on
diamond ages and mantle Re–Os systematics. Earth and Planetary Science
Letters, 160, 311–326.
Pearson, N.J., Alard, O., Griffin, W.L., Jackson, S.E., and O’Reilly, S.Y. (2002) In
situ measurement of Re-Os isotopes in mantle sulfides by laser ablation mul-
ticollector-inductively coupled plasma mass spectrometry: analytical methods
and preliminary results. Geochimica et Cosmochimica Acta, 66, 1037–1050.
Pearson, D.G., Canil, D., and Shirey, S.B. (2003) Mantle samples included in
volcanic rocks: xenoliths and diamonds. Treatise on Geochemistry, 2, 171–275.
Pearson, D.G., Parman, S.W., and Nowell, G.M. (2007) A link between large
ZHANG AND HIRSCHMANN: MANTLE SULFIDE MELTING UP TO 8 GPA192
mantle melting events and continent growth seen in osmium isotopes. Nature,
449, 202–205.
Pollack, H.N., and Chapman, D.S. (1977) On the regional variation of heat flow,
geotherms, and lithospheric thickness. Tectonophysics, 38, 279–296.
Powell, W., and O’Reilly, S.Y. (2007) Metasomatism and sulfide mobility in
lithospheric mantle beneath eastern Australia: Implications for mantle Re–Os
chronology. Lithos, 94, 132–147.
Richardson, S.H., Shirey, S.B., Harris, J.W., and Carlson, R.W. (2001) Archean
subduction recorded by Re-Os isotopes in eclogitic sulfide inclusions in
Kimberley diamonds. Earth and Planetary Science Letters, 191, 257–266.
Rohrbach, A., Ballhaus, C., Gola-Schindler, U., Ulmer, P., Kamnetsky, V.S., and
Kuzmin, D.V. (2007) Metal saturation in the upper mantle. Nature, 449,
456–460.
Rohrbach, A., Ballhaus, C., Ulmer, P., Kamnetsky, V.S., and Gola-Schindler, U.
(2011) Experimental evidence for a reduced metal-saturated upper mantle.
Journal of Petrology, 52, 717–731.
Roy-Barman, M., Wasserburga, G.J., Papanastassioua, D.A., and Chaussidon, M.
(1998) Osmium isotopic compositions and Re–Os concentrations in sulfide
globules from basaltic glasses. Earth and Planetary Science Letters, 154,
331–347.
Ryzhenko, B., and Kennedy, G.C. (1973) The effect of pressure on the eutectic in
the system Fe-FeS. American Journal of Science, 273, 803–810.
Shannon, M.C., and Agee, C.B. (1998) Percolation of core melts at lower mantle
conditions. Science, 280, 1059–1061.
Sharp, W.E. (1969) Melting curves of sphalerite, galena, and pyrrhotite and the
decomposition curve of pyrite between 30 and 65 kilobars. Journal of Geo-
physical Research, 74, 1645–1652.
Shi, C.Y., Zhang, L., Yang., W, Liu, Y., Wang, J., Meng, Y., Andrews, J.C., and Mao,
W.L. (2013) Formation of an interconnected network of iron melt at Earth’s
lower mantle conditions. Nature Geoscience, 6, 971–975.
Shirey, S.B., and Richardson, S.H. (2011) Start of the Wilson cycle at 3 Ga shown
by diamonds from subcontinental mantle. Science, 333, 434–436.
Shirey, S.B., Cartigny, P., Frost, D.J., Keshav, S., Paolo Nimis, F.N., Pearson, D.G.,
Sobolev, N.V., and Walter, M.J. (2013) Diamonds and the geology of mantle
carbon. Reviews in Mineralogy and Geochemistry, 75, 355–421.
Simon, F., and Glatzel, G. (1929) Bemerkungen zur Schmelzdruckkurve. Zeitschrift
für anorganische und allgemeine Chemie, 178, 309–316.
Stachel, T., and Harris, J.W. (2008) The origin of cratonic diamonds—constraints
from mineral inclusions. Ore Geology Reviews, 34, 5–32.
Stewart, A.J., Schmidt, M.W., Westrenen, W.V., and Liebske, C. (2007) Mars: A
new core-crystallization regime. Science, 316, 1323–1325.
Tenner, T.J., Hirschmann, M.M., Withers, C.W., and Hervig, R.L. (2009) Hydrogen
partitioning between nominally anhydrous upper mantle minerals and melt
between 3 and 5 GPa and applications to hydrous peridotite partial melting.
Chemical Geology, 262, 42–56.
——— (2012) H2O storage capacity of olivine and low-Ca pyroxene from 10
to 13 GPa: consequences for dehydration melting above the transition zone.
Contributions to Mineralogy and Petrology, 163, 297–316.
Turcotte, D. L., and Schubert, G. (2002) Geodynamics, 2nd ed., pp. 456. Cambridge
University Press, U.K.
Urakawa, S., Kato, M., and Kumazawa, M. (1987) Experimental study on the
phase relations in the system Fe-Ni-O-S up to 15 GPa. In M. Manghnani and
Y. Syono, Eds., High-Pressure Research in Mineral Physics, p. 95–111. Terra,
Tokyo/AGU, Washington, D.C.
Usselman, T. (1975) Experimental approach to the state of the core: Part I. The
liquidus relations of the Fe-rich portion of the Fe-Ni-S system from 30 to 100
kbars. American Journal of Science, 275, 278–290.
Waldner, P., and Pelton, A. (2004) Critical thermodynamic assessment and
modeling of the Fe-Ni-S system. Metallurgical and Materials Transactions
B, 35, 897–907.
Westerlund, K.J., Shirey, S.B., Richardson, S.H., Carlson, R.W., Gurney, J.J., and
Harris, J.W. (2006) A subduction wedge origin from Paleoarchaean peridotitic
diamonds and harzburgites from the Panda kimberlite, Slave craton: evidence
from Re-Os isotope systematics. Contributions to Mineralogy and Petrology,
152, 275–294.
Withers, A.C., and Hirschmann, M.M. (2007) H2O storage capacity of MgSiO3
clinoenstatite at 8–13 GPa, 1,100–1,400 °C. Contributions to Mineralogy and
Petrology, 154, 663–674.
Withers, A.C., Hirschmann, M.M., and Tenner T.J. (2011) The effect of Fe on
olivine H2O storage capacity: Consequences for H2O in the martian mantle.
American Mineralogist, 96, 1039–1053.
Xirouchakis, D., Hirschmann, M.M., and Simpson, J.A. (2001) The effect of
titanium on the silica content and on mineral–liquid partitioning of mantle-
equilibrated melts. Geochimica et Cosmochimica Acta, 65, 2201–2217.
Manuscript received January 17, 2015
Manuscript accepted July 22, 2015
Manuscript handled by charles lesher
