gc001503 1..21 - EILggg07.pdf

Oxygen-isotope and trace element constraints on the origins

of silica-rich melts in the subarc mantle

J. M. Eiler

Division of Geological and Planetary Sciences, California Institute of Technology, MC 100-23, Pasadena, California

91125, USA (eiler@gps.caltech.edu)

P. Schiano

Departement des Sciences de la Terra (UMR 6524 ‘‘Magmas et Volcans’’), Universite

́ Blaise Pascal, 5 Rue Kessler,

F-63038 Clermont-Ferrand Cedex, France

J. W. Valley and N. T. Kita

Department of Geology, University of Wisconsin, Madison, Wisconsin 53706, USA

E. M. Stolper

Division of Geological and Planetary Sciences, California Institute of Technology, MC 100-23, Pasadena, California

91125, USA

[

]

Peridotitic xenoliths in basaltic andesites from Batan island in the Luzon arc contain silica-rich (broadly

dacitic) hydrous melt inclusions that were likely trapped when these rocks were within the upper mantle

wedge underlying the arc. These melt inclusions have been previously interpreted to be slab-derived melts.

We tested this hypothesis by analyzing the oxygen isotope compositions of these inclusions with an ion

microprobe. The melt inclusions from Batan xenoliths have

VSMOW

values of 6.45 ± 0.51

. These

values are consistent with the melts having been in oxygen isotope exchange equilibrium with average

mantle peridotite at temperatures of



875

°

C. We suggest the

O values of Batan inclusions, as well as

their major and trace element compositions, can be explained if they are low-degree melts (or

differentiation products of such melts) of peridotites in the mantle wedge that had previously undergone

extensive melt extraction followed by metasomatism by small amounts (several percent or less) of slab-

derived components. A model based on the trace element contents of Batan inclusions suggests that this

metasomatic agent was an aqueous fluid extracted from subducted basalts and had many characteristics

similar to slab-derived components of the sources of arc-related basalts at Batan and elsewhere. Batan

inclusions bear similarities to ‘‘adakites,’’ a class of arc-related lava widely considered to be slab-derived

melts. Our results suggest the alternative interpretation that at least some adakite-like liquids might be

generated from low-degree melting of metasomatized peridotites.

Components:

14,199 words, 6 figures, 2 tables

Keywords:

subduction zone; slab melting; oxygen isotopes; adakites.

Index Terms:

1031 Geochemistry: Subduction zone processes (3060, 3613, 8170, 8413); 1037 Geochemistry: Magma

genesis and partial melting (3619); 1041 Geochemistry: Stable isotope geochemistry (0454, 4870).

Received

9 October 2006;

Revised

12 June 2007;

Accepted

19 June 2007;

Published

21 September 2007.

Geochemistry

Geophysics

Geosystems

Published by AGU and the Geochemical Society

AN ELECTRONIC JOURNAL OF THE EARTH SCIENCES

Geochemistry

Geophysics

Geosystems

Article

Volume 8

,Number9

21 September 2007

Q09012, doi:10.1029/2006GC001503

ISSN: 1525-2027

Click

Here

for

Full

rticl

1 of 21

Eiler, J. M., P. Schiano, J. W. Valley, N. T. Kita, and E. M. Stolper (2007), Oxygen-isotope and trace element constraints on

the origins of silica-rich melts in the subarc mantle,

Geochem. Geophys. Geosyst.

, Q09012, doi:10.1029/

2006GC001503.

1. Introduction

[

] Convergent margin magmatism is thought to be

dominated by partial melting of peridotitic mantle

wedge in response to influxes of aqueous fluids

and/or hydrous melts released from subducting

lithosphere [

Gill

, 1981;

Tatsumietal.

, 1983;

Davies and Bickle

, 1991;

McCulloch and Gamble

1991;

Davies and Stevenson

, 1992]. In this pro-

cess, slab-derived metasomatic agents control the

budgets of many minor and trace elements in the

mantle wedge [

Plank and Langmuir

, 1993;

Elliott

et al.

, 1997] and strongly influence its melting

properties and rheology [

Kushiro et al.

, 1968;

Stolper and Newman

, 1994;

Billen and Gurnis

2001;

Hirth and Kohlstedt

, 2003;

Baker et al.

2005], but they comprise only a small mass frac-

tion (



1–2% or less [

Eiler et al.

, 2000, 2005]) of

the sources of most arc magmas.

[

] Despite their dominantly peridotitic sources,

arc volcanoes erupt larger fractions of relatively

silica-rich lavas (e.g., basaltic andesites, andesites,

dacites and rhyolites) than volcanic centers in other

environments (e.g., ocean ridges, plateaus and

islands [

Gill

, 1981]). The major element composi-

tions of most silica-rich arc lavas are attributable to

crustal differentiation of more mafic magmas

[

Davidson

, 1996]. However, a variety of other

processes contribute to these silica-rich composi-

tions, including partial melting of basaltic material

in subducting slabs and/or reaction between

such melts and peridotites in the mantle wedge

[

Ringwood and Green

, 1966;

Condie and Swenson

1973;

Ringwood

, 1974;

Kay

, 1978;

Martin

, 1986,

1988;

Defant and Drummond

, 1990;

Rapp et

al.

, 1991;

Martin et al.

, 2005]; partial melting of

mafic rocks deep in the crust of the overriding

plate [

Atherton and Petford

, 1993]; differentiation

of basaltic melt in the upper mantle [

Kelemen

1990;

Macpherson et al.

, 2006]; or low-degree

(



1–2%) partial melting of peridotites in the

presence of water [

Baker et al.

, 1995;

Gaetani

and Grove

, 1998, 2003;

Hirschmann et al.

, 1998,

1999]. Resolving the relative importance of these

processes for explaining the silica-rich character of

arc lavas has implications for the thermal and

dynamical evolution of subducting slabs [

Peacock

2003;

Kelemen et al.

, 2003a], for the chemical

budgets of arc magmatism [

Plank and Langmuir

1993;

Elliott et al.

, 1997], for the mechanisms

responsible for growth of the continental crust

[

Rudnick and Fountain

, 1995;

Wang et al.

, 2003],

and for the composition of deeply subducted lith-

osphere [

Kelley et al.

, 2005].

[

] ‘‘Adakites,’’ magmatic rocks with unusually

low Y contents (generally less than 20 ppm) and

high Sr/Y ratios (generally greater than 50), have

been suggested to have a special significance for

our understanding of silica-rich arc magmas

[

Defant and Drummond

, 1990]. Archetypical ada-

kites are dacites from convergent margins where

relatively young ocean lithosphere has been sub-

ducted (although the rock name has been applied to

other rock types and settings). It has been sug-

gested that adakites are among the clearest exam-

ples of arc-related melts that form by partial

melting of basaltic components of subducted ocean

lithosphere [

Defant and Drummond

, 1990;

Rapp et

al.

, 1991]. This hypothesis is supported by the facts

that partial melts of eclogite are broadly dacitic

over a significant range in melt fraction [

Rapp et

al.

, 1991] and leave garnet-rich residues that retain

Y and heavy-rare earth elements, promoting low Y

and high Sr/Y. Alternatively, mirroring the debate

about silica-rich arc lavas generally, it has also

been suggested that adakites form by high-pressure

crystallization-differentiation of basalt [

Macpherson

et al.

, 2006] or by melting garnet- and/or amphibole-

rich crystal cumulates near the base of thick crust

[

Chung et al.

, 2003;

Wang et al.

, 2005]. Both

processes involve melt/garnet fractionation and

produce broadly dacitic melt and so can also

explain the critical features of adakites. This debate

also touches on the origin of high-Mg andesites,

many of which have the trace element character-

istics of adakites and have been suggested to form

by reaction of slab-derived melts with mantle

peridotites [

Green and Ringwood

, 1967;

Kay

1978;

Kelemen

, 1995;

Yogodzinski and Kelemen

1998]. Alternatively, high-Mg andesites may be

direct partial melts of hydrous peridotites [

Tatsumi

1981, 1982;

O’Nions and McKenzie

, 1988;

Grove

et al.

, 2002].

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

2of21

[

] Melt inclusions (now glass) from metasomat-

ized xenoliths from Mount Iraya on the island of

Batan (northern-most Philippines; a part of the

Luzon-Taiwan volcanic arc) are a well-documented

example of silica-rich melts in a convergent margin

[

Richard et al.

, 1986;

Vidal et al.

, 1989;

Maury et

al.

, 1992;

Schiano et al.

, 1995]. These peridotitic

xenoliths are hosted by basaltic andesite and are

believed to sample the upper portions of the mantle

wedge beneath the Luzon arc. The melt inclusions

contained in these xenoliths have silica contents up



63% and Sr/Y ratios up to



250, within the

adakitic range. The melt inclusions correspond to

one end-member of the array of compositions of

basaltic andesites from Batan (Figure 1) [

Schiano

et al.

, 1995], and thus they could have bearing on

the petrogenesis of the lavas that carried the

xenoliths to the surface. These melt inclusions

clearly existed in the mantle and so cannot be

products of differentiation in or partial melting

of the arc crust (though they could be products of

differentiation of basalt in the mantle wedge

[

Macpherson et al.

, 2006]); and their low MgO

contents (0.5–1.7 wt.% [

Schiano et al.

, 1995])

(Table S3 in the auxiliary material

to this paper)

are within a range that has been previously sug-

gested to preclude reaction between slab-derived

melts and mantle peridotites [

Rapp et al.

, 1999]

(although they could also reflect modification of

peridotite-equilibrated melts by crystallization of

olivine and other phases). These inclusions have

been interpreted as partial melts of basalts in the

subducted South China Sea slab [

Schiano et al.

1995], and they have been cited as evidence that

adakites can be generated by melting of subducted

crust and can pass through the mantle wedge

without extensive reaction with peridotite [

Martin

et al.

, 2005, and references therein]. If these

interpretations are correct, Batan melt inclusions

could provide valuable information on the role of

slab melting in the petrogenesis of arc-derived

magmas, especially those that are relatively silica-

rich. Note, however, that the Batan melt inclusions

have major element characteristics (other than

MgO and Mg/Fe ratio, which can be disturbed by

crystallization) similar to those suggested for very

low degree (approximately



1–2%) partial melts

of peridotite at low pressure (<1.5 GPa [

Baker et

al.

, 1995;

Hirschmann et al.

, 1998]). Moreover,

McDermott et al.

[1993] suggested that the trace

element characteristics of Batan lavas (and, by

extension, compositionally similar Batan melt

inclusions) could be generated by melting perido-

tite that had been metasomatized by slab-derived

fluid. Here, we use oxygen isotopes to constrain

further the origins of Batan inclusions.

[

] Mantle peridotites typically have

VSMOW

values of approximately 5.5 ± 0.2

[

Mattey et al.

1994] (see

Eiler et al.

[1996] and

Ducea et al.

[2002] for exceptions). In contrast, the major

components of the upper



10 km of the ocean

lithosphere typically have

VSMOW

values in

the following ranges [

Kolodny and Epstein

, 1976;

Arthur et al.

, 1983;

Gregory and Taylor

, 1981;

Alt

et al.

, 1986;

Staudigel et al.

, 1995]: opaline oozes,

35–42

; carbonate oozes, 25–32

; pelagic

clays, 15–25

; terrigenous and volcaniclastic

sediments, and basalts and diabases subjected to

weathering and low-temperature hydrothermal

alteration (i.e., layer-2 ocean crust), 7–15

; and

gabbros subjected to high-temperature hydrother-

mal alteration (i.e., layer-3 ocean crust), 0–6

Importantly, all of these materials have broadly

similar oxygen concentrations, and so

O values

of mixtures between these components are approx-

imately linear functions of the mass fractions of the

end-members. It is also important to note that

magmatic differentiation processes (partial melting

and crystallization) also lead to variations in

[

Eiler

, 2001], and these variations must be consid-

ered before reaching conclusions regarding the

Figure 1.

Geochemical trend defined by Batan melt

inclusions (unfilled field) and host Mt. Iraya lavas (gray

field). These data are consistent with Batan inclusions

being a mixing end-member of magmas parental to

Mt. Iraya lavas or being products of a petrogenetic

process related to that responsible for generating Mt.

Iraya lavas. Data are from

Richard et al.

[1986],

McDermott et al.

[1993], and

Schiano et al.

[1995].

Auxiliary material data sets are available at ftp://ftp.agu.org/

apend/gc/2006gc001503. Other auxiliary material files are in the

HTML.

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

3of21

oxygen isotope compositions of the sources of

magmatic rocks.

2. Samples

[

] We examined inclusions and host olivine crys-

tals from xenolith B3N, collected from a high-K

basaltic andesite erupted



1.48 ka ago from Mount

Iraya, Batan Island [

Schiano et al.

, 1995]. The

peridotitic xenoliths in this flow are harzburgites

believed to come from a depth of



30 km, near the

top of the mantle wedge between the overriding

Philippine plate and the subducting South China

Sea plate [

Maury et al.

, 1992].

[

] Xenolith B3N is a deformed spinel harzburgite

consisting of primary porphyroclasts of olivine

(Fe

), orthopyroxene (En

)andspinel

(Mg

0.62

0.64

0.36

0.38

0.04

0.08

0.62

0.90

0.98

1.22

)[

Richard et al.

, 1986;

Maury et al.

1992;

Schiano et al.

, 1995]. Clinopyroxene is rare

in this sample, suggesting that it is a fragment of

the subarc mantle that is residual to high extents of

melting. The sample also contains hydrous meta-

somatic phases (phlogopite and amphibole) and

undeformed neoblasts of olivine (Fo

)and

orthopyroxene (En

). Petrographic observa-

tions suggest that neoblastic olivine formed from

recrystallization of primary olivine porphyroclasts,

accompanied by variable amounts of Mg loss and

Fe gain and growth of metasomatic phlogopite and

amphibole [

Richard et al.

, 1986;

Maury et al.

1992;

Schiano et al.

, 1995]. It also seems possible

that some neoblastic olivine and orthopyroxene

precipitated from the same infiltrating metasomatic

agent responsible for growth of hydrous metaso-

matic phases.

[

] Olivine porphyroclasts and neoblasts and meta-

somatic hydrous phases in sample B3N commonly

contain inclusions. Inclusions in olivine porphyr-

oclasts typically occur as linear trails of small

(approximately



m diameter) polyphase

assemblages of silica-rich (SiO

> 57 wt.%) glass

and H

O-rich bubbles of liquid and vapor [

Schiano

et al.

, 1995]. Polyphase inclusions are typically

50:50 mixtures of glass and fluid. Homogenization

temperatures of these polyphase inclusions (a mini-

mum estimate of their liquidus temperatures)

average 920 ± 10

°

C. Neoblasts and hydrous metaso-

matic phases contain larger (typically 30–150

diameter) inclusions of either a combination of silicate

glass and H

O-rich fluid (like the inclusions

in porphyroclasts) or H

O-rich fluid alone. This

suggests that both hydrous silicate melt and aqueous

fluid were present during metasomatism and recrys-

tallization [

Schiano et al.

, 1995]. These inclusions

typically define crystallographic planes in their host

minerals and were likely trapped from grain-boundary

melts and/or fluids during growth of the neoblasts and

hydrous metasomatic minerals [

Schiano et al.

, 1995].

There are no obvious compositional differences

between glass inclusions in porphyroclasts and inclu-

sions in neoblasts. We focused our analysis on

inclusions in neoblasts because their larger size

facilitated ion microprobe analyses of glass.

[

] Batan inclusions might have been trapped

from silicate melts or might instead be exsolved

from solute-rich aqueous fluids or supercritical

liquids. Glass and fluid in polyphase Batan inclu-

sions are typically subequal in volume within each

inclusion and glass typically contains approximately

5 wt.% H

Schiano et al.

, 1995]. Thus, prior to

phase separation, these inclusions were approxi-

mately 30 wt.% H

O and 70 wt% silicate (assum-

ing the trapped aqueous phase has a density of 1 g/cc

and glass has a density of 2.5 g/cc). This bulk

composition is too silicate rich to be an aqueous

phase in equilibrium with mafic or ultramafic rock

at 1 GPa; i.e., it lies on the ‘‘melt’’ side of the

solvus between hydrous, silicate melts and aqueous

fluids in these compositional systems [

Stalder et

al.

, 2001;

Kessel et al.

, 2005a]. Thus these poly-

phase inclusions most likely quenched from H

O-

rich silicate melt, not aqueous fluid or a supercrit-

ical phase. Some glass inclusions contain crystals

of sulfide, indicating that this hydrous silicate melt

also contained significant amounts of sulfur.

3. Methods

[

] Major element compositions of minerals and

glasses in peridotite xenolith B3N were taken from

published analyses by

Schiano et al.

[1995] or were

measured using a JEOL 733 electron microprobe at

Caltech. These new measurements used an acceler-

ating voltage of 15 keV, a sample current of 10 nA, a

defocused beam (15–20

m) for glasses and a

focused beam for minerals. The H

O-rich character

of fluid associated with glass inclusions was estab-

lished by previously published laser-Raman analy-

ses; water contents of glasses were previously

determined by SIMS [

Schiano et al.

, 1995].

[

] Oxygen-isotope analyses were made using the

CAMECA IMS-1280 ion microprobe at the Uni-

versity of Wisconsin. Measurements were made

using a primary beam of

133

ions with a 20 keV

impact energy, focused to a diameter of approxi-

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

4of21

mately 5–10

m on the sample surface. Primary

ion intensities were between



1 nA and



5 nA but

were held constant for each analytical session. An

electron flood gun was used for charge compensa-

tion. Secondary O

ions were extracted, accelerat-

ed by 10 keV, and focused and collimated using a

secondary optic alignment described by

Kita et al.

[2004]. The mass resolving power was



2500,

enough to separate hydride interferences on

[

] The intensity of the

secondary ion beam

typically varied between 1 and 5



counts per

second (cps), depending on the intensity of the

primary ion beam (i.e.,



cps of secondary ions

were detected per nA of primary beam current).

Two Faraday cups were used to measure

(registered through a 10

resistor) and

(registered through a 10

resistor) simultaneously.

The baseline of each amplifier was calibrated once

each day; drift during the day was insignificant

compared to the noise level of the detectors

(



1000 cps for the circuit using the 10

resistor).

Immediately before each analysis, any small mis-

alignment of the secondary optics was corrected for

by scanning the position of the secondary ion beam

across the field aperture of the mass spectrometer

to maximize its detected intensity.

[

] Eachanalysisconsistedof20cyclesof4seconds

each; the internal precision of the

Oratio

was typically between 0.05 and 0.15

(varying

inversely with the secondary beam intensity).

The external precision, as measured by the repro-

ducibility of repeat analyses of nominally homo-

geneous standards of silicate minerals and glasses

(see the auxiliary material), averaged ±0.21

1 standard deviation, spot-to-spot. External preci-

sion was poorer than internal precision by a factor

of 2, on average. We suspect this reflects some

combination of imperfect charge neutralization of

the sample surface during analysis, variations in

analytical fractionation caused by moving the

sample in the instrument reference frame, and/or

isotopic heterogeneity of standards. In any event,

we suggest the external precision of our analyses of

unknowns, not considering any systematic errors in

standardization (below), is ±0.2

[

] Analyses of oxygen isotope composition by

SIMS typically involve an instrumental mass frac-

tionation that varies from material to material, the

so-called ‘‘matrix effect’’ [

Eiler et al.

,1997].

Matrix effects stem from a variety of known causes

but generally cannot be predicted on the basis of

physical principles and must be corrected for

empirically. Matrix effects in silicate minerals

previously have been corrected for by interpolation

based on major element composition; i.e., the

matrix effect for an unknown material is calculated

as the weighted average of those for two or more

standards that, when combined in the appropriate

proportions, approximate the major element com-

position of the unknown. The approach we follow

here is as follows: Batan glass inclusions have

normative compositions dominated by albite, an-

orthite, orthoclase, quartz and hypersthene, and

they contain approximately 5 wt.% H

Schiano

et al.

, 1995] (auxiliary material). We approximated

the instrumental fractionation for these glasses as

the weighted sum, by oxygen fraction, of those

measured for albite glass, anorthite glass, ortho-

clase glass, silica glass and crystalline orthopyrox-

ene (auxiliary material), in proportions equal to the

anhydrous norm of the unknown glass, and then

apply a correction equal to the difference between

anhydrous albite glass and albite glass containing

5 wt.% H

O. For olivines, there is no variation in

instrumental fractionation with major element com-

position over the range relevant to our samples, so

we standardized all measurements of unknown

olivines by assuming an instrumental fractionation

equal to that for San Carlos olivine standard. See

the auxiliary material for further details and data

for standards.

4. Results

[

] All oxygen isotope analyses of Batan inclu-

sions and host olivines made as part of this study

are summarized in Table 1. Seven analyses of five

separate inclusions (one of which was analyzed

three times; the others once each) yield an average

VSMOW

of 6.45 ± 0.51

(±0.19

1 standard

error, i.e., standard deviation divided by the square

root of the number of measurements). Twenty-five

analyses of three host olivine crystals yield an

average

VSMOW

of 5.34 ± 0.42

(±0.08

st. err.). Each of these two populations of analyses

has a standard deviation that is a factor of 2 worse

than the external precision for analyses of nomi-

nally homogeneous standards. We think this dif-

ference likely reflects some combination of errors

in corrections for matrix effects, analytical fractio-

nations caused by vertical relief between the sep-

arately mounted samples and standards, and real

isotopic heterogeneity of the samples. We can

examine the importance of fractionations caused

by sample relief by assuming that the three olivine

grains that host our analyzed inclusions are iden-

tical to each other in

O with a value equal to

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

5of21

their average measured value (a reasonable approx-

imation, given the small range in

O of most

mantle olivines [

Eiler

, 2001]). In this case, the

average value for glass changes little, to 6.59 ±

0.53 (±0.20

st. err) and the reproducibility for

olivines (i.e., based on reproducibility within each

grain) improves slightly to ±0.37

(±0.07

st.

err.). The important point, however, is that regard-

less of these small corrections, to first order, all

olivines have

O values similar to olivine from

typical mantle peridotites (



5.0–5.4

[

Mattey et

al.

, 1994;

Eiler

, 2001]), and all glasses have

values of



6.5

5. Discussion

[

] The oxygen isotope compositions of Batan

melt inclusions are consistent with those predicted

for melts of similar major element composition in

equilibrium with average mantle peridotite at tem-

peratures of 875

°

C or greater (Figure 2). That is,

while these inclusions are approximately 1–1.5

higher in

O than typical peridotitic olivines, this

difference is consistent with equilibrium oxygen-

isotope fractionation between quartzo-feldspathic

melts and typical mantle peridotite mineralogies at

magmatic temperatures [

Eiler

, 2001, and referen-

ces therein]. This result also suggests that Batan

melt inclusions contain little or no oxygen from

sources rich in weathered, hydrothermally altered

or authigenic phases, such as basaltic, gabbroic and

metasedimentary materials in a subducted slab.

Instead, our data are consistent with the interpre-

tation that Batan siliceous melt inclusions are

derived from or equilibrated with mantle perido-

tites having typical

O values, or are differenti-

ation products of more mafic melts that were

derived from or in equilibrium with mantle

peridotites.

[

] The oxygen isotope compositions of Batan

inclusions contrast with those of melt veins and

inclusions in a peridotite xenolith collected from

an alkali basalt from Simberi island, Papua New

Guinea [

Eiler et al.

, 1998] (Simberi is a volcano

associated with the Manus-Kilinailau subduction

zone). These latter glasses have

VSMOW

values up to 11.3 ± 1.3

, consistent with deri-

vation from a source rich in oxygen from sub-

ducted sediments and/or altered upper oceanic

crust. These previous results show that metaso-

matic melts and/or fluids can be extracted from a

subducted slab and transported through the mantle

wedge without undergoing extensive oxygen-

isotope exchange with the mantle peridotites

through which they move.

Eiler et al.

[1998]

suggested that this reflects either the fast rate of

Table 1.

SIMS Analysis of

VSMOW

Values of Batan Inclusions and Host Olivines

B3N1

B3N2

B3N3

Host

Olivine Inclusion

Normalized

Inclusion

Host

Olivine Inclusion

Normalized

Inclusion

Host

Olivine Inclusion

Normalized

Inclusion

5.45

5.79

5.63

5.31

6.31

6.64

5.14

6.84

6.83

5.35

7.06

6.90

4.58

5.84

6.17

4.75

5.04

5.12

7.10

7.43

5.38

5.47

5.05

6.18

6.51

6.19

5.38

4.93

5.60

5.11

5.14

5.27

5.22

4.96

5.16

5.10

5.13

5.33

6.30

6.12

6.10

5.88

Average

5.50

6.42

6.26

5.01

6.36

6.69

5.35

6.84

6.83

1 St.dev.

0.41

0.63

0.21

0.46

0.42

0.18

1 St.err.

0.11

0.45

0.08

0.23

0.16

0.18

Overall averages: All olivines, 5.34 ± 0.42 (±0.08 st. err.); all inclusions, 6.44 ± 0.51 (±0.19 st. err.); all normalized inclusions, 6.59 ± 0.53

(±0.20 st. err.).

Restandardized assuming host olivine has a

VSMOW

value of 5.34

Based on external precision of standard glasses.

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

6of21

movement of slab-derived melt and/or fluid

through the wedge (i.e., because diffusion-limited

oxygen isotope exchange is relatively slow, even

at mantle temperatures) or that slab-derived melt

can move through conduits that were previously

armored by metasomatic phases.

[

] The mantle-equilibrated oxygen isotopic ratios

of Batan inclusions argue against involvement of

several petrogenetic processes in their origins and

are consistent with several others:

[

] Batan inclusions are not partial melts of

weathered, hydrothermally altered or sedimentary

constituents of subducted oceanic lithosphere that

were transported through the mantle wedge with

little or no reaction or exchange with peridotites.

Similarly, Batan inclusions are not well explained

as partial melts of portions of the mantle wedge

that underwent extensive metasomatism (i.e.,

where metasomatic agent contributes tens of % of

the oxygen of the hybrid source) by fluids or melts

derived from

O-rich or

O-poor sections of

subducted ocean crust (this was the preferred

interpretation of high-

O inclusions from Sim-

beri island presented by

Eiler et al.

[1998]).

[

] On the other hand, Batan inclusions

could

partial melts of parts of subducted ocean crust that

fortuitously yield melts having

Ovaluesin

equilibrium with mantle peridotite (or, similarly, a

combination of melts from

O-rich and

O-poor

sources that fortuitously yields such

O values

[

Bindeman et al.

, 2005]). They also could be slab-

derived melts that underwent extensive oxygen

isotope exchange with mantle peridotites during

transport through the mantle wedge. Or, they

might be products of high-pressure crystallization-

differentiation of normal-

O basaltic partial melts

of mantle wedge. Finally, they could be low-degree

partial melts of normal-

O mantle wedge peri-

dotites. While it may not be possible to strictly rule

out any one of these possibilities, the following

sections examine their various strengths and weak-

nesses in explaining the full spectrum of major

element, trace element and isotopic constraints on

the origins of Batan inclusions.

5.1. Slab-Derived Melts That Lack

Anomalous Oxygen Isotope Signatures?

[

] Oxygen isotope data are consistent with Batan

inclusions being partial melts of basaltic or gab-

broic components of the subducted South China

Figure 2.

The heavy curve plots the

VSMOW

values of dacitic melts in oxygen-isotope exchange equilibrium

with average mantle peridotite as a function of the temperature of equilibration. This calculation assumes a major

element composition for silica-rich melt equal to the average measured for Batan glass inclusions [

Schiano et al.

1995] (Table S3 in the auxiliary material for this paper), an oxygen-isotope composition for mantle peridotite equal to

the sources of average, normal mid-ocean-ridge basalts [

Eiler et al.

, 1998;

Eiler

, 2001], and oxygen-isotope

fractionation factors summarized by

Eiler

[2001]. The horizontal and vertical gray bands indicate the oxygen-isotope

compositions (this study) and homogenization temperatures [

Schiano et al.

, 1995], respectively, of Batan inclusions

(in both cases the position and width of the band reflect the average(2 std. err.). Data presented in this study are

consistent with Batan inclusions being in oxygen-isotope exchange equilibrium with typical upper mantle peridotites.

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

7of21

Sea slab that are in oxygen isotope equilibrium

with the mantle either because their sources

escaped weathering or hydrothermal alteration

before subduction or because they underwent

oxygen isotope exchange with the mantle wedge

before or during entrapment. This hypothesis is

appealing because Batan inclusions have traits

often associated with slab-derived melts (e.g.,

elevated SiO

contents, low abundances of heavy

rare earth elements and Y, and high Sr/Y ratios).

On the other hand, normal-

O components of old

oceanic crust are unusual [

Gregory and Taylor

1981;

Alt et al.

, 1986;

Staudigel et al.

, 1995],

and previous data for Simberi island inclusions

[

Eiler et al.

, 1998] demonstrate that melts of

sources rich in slab components can retain rela-

tively dramatic oxygen-isotope anomalies, despite

later transport through the mantle. Similarly, sev-

eral suites of arc lavas are known to have oxygen

isotope anomalies inherited from slab-derived

components, despite transport through the mantle

wedge [e.g.,

Eiler et al.

, 2000, 2005;

Bindeman et

al.

, 2005]. Thus, if Batan inclusions are slab melts,

their oxygen isotope compositions contrast both

with simple expectations and with previous data

for subduction-zone melts.

[

] Figure 3a presents the average primitive-man-

tle-normalized trace element composition of Batan

inclusions, based on analyses of a subset of the

plotted elements in glass inclusions (circles) and

analyses of a larger subset of elements in inclusion-

rich olivines (diamonds; see the caption to Figure 3a

for details; both sets of data from

Schiano et al.

[1995]). Note that the inclusion-rich olivines in

Batan xenoliths contain sulfides, which doubtless

host a large proportion of the Pb and Cu added to

these xenoliths by metasomatic melt [

Schiano et

al.

, 1995]. Measurements of inclusion-rich olivines

will faithfully represent the bulk abundances of

these elements in metasomatic melt only if all

phases formed by quenching of metasomatic melt

(glass, fluid, sulfide, etc.) are sampled in their

correct relative proportions (i.e., preferential

enrichment or depletion in sulfide could enrich or

deplete olivine separates in Pb and Cu). However,

repeat measurements of inclusion-rich olivines

from Batan xenoliths are reproducible in their

relative abundances of trace elements, and essen-

tially indistinguishable in their Pb anomalies; so, it

seems unlikely to us that selective sampling of

sulfide (or any other quench phase) in the inclusion-

rich olivine separates has greatly biased the esti-

mated compositions of metasomatic melts.

[

] The most noteworthy features of the trace

element compositions of Batan inclusions are: a

approximately hundredfold enrichment of incom-

patible elements relative to compatible elements; a

pronounced (approximately fivefold) positive

anomaly in Pb; a large (approximately tenfold)

negative anomaly in Nb; and high concentrations

of Cu and Sn (despite uncertainty in the correct

location of these elements on the horizontal axis in

Figure 3, they likely constitute



10x positive

anomalies). Batan inclusions also have modest

(factor of 2 to three) positive anomalies of U and

Sr and negative anomalies of Eu and Ti.

[

] Figure 3b compares the estimated average

composition of Batan inclusions from Figure 3a

(filled circles) to predicted compositions of low-

degree (1 wt.%) partial melts of altered ocean crust

[

Bach et al.

, 2003] in the eclogite facies. These

calculated slab melt compositions are based on

estimated or experimentally measured garnet/melt,

clinopyroxene/melt and rutile/melt distribution

coefficients for the elements of interest, and an

assumed residual mineralogy of 60% garnet, 40%

clinopyroxene (mixed with 1% rutile for the mod-

els labeled ‘‘Kelemen-rutile’’ and ‘‘Xiong-rutile’’).

Each model corresponds to a different set of con-

straints on the relevant mineral-melt distribution

coefficients; see the auxiliary material for a com-

pilation and discussion of these various distribution

coefficients. Models plotted with solid lines are

based on direct experimental studies of eclogite

melting; those plotted with dashed lines use

inferred distribution coefficients based on experi-

ments in other systems. Note that the model

‘‘Kessel’’ is based on experimental partitioning

data between eclogite and an aqueous supercritical

phase that could be described as broadly ‘‘melt-

like’’ in its physical and chemical properties, but is

not, strictly speaking, melt [

Kessel et al.

, 2005a,

2005b]. Each of these models would result in lower

overall concentrations of incompatible elements

and somewhat weaker element-specific anomalies

for higher degrees of melting; otherwise, the pat-

terns of relative trace element concentrations are

insensitive to variations in degree of melting. For

this reason, our discussion focuses on the shape,

not position, of model results in Figure 3b. We do

not consider partial melts of subducted sediment

because the Sr isotope compositions of metasom-

atized Batan xenoliths (

Sr/

Sr = 0.7044 to

0.7048 [

Maury et al.

, 1992]) preclude a significant

component of pelagic sediment (

Sr/

Sr = 0.715

to 0.718 for Mesozoic and younger marine sedi-

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

8of21

ments [

Plank and Langmuir

, 1998]) in the meta-

somatic melts and/or fluids that infiltrated them.

[

] Trace element patterns of all seven model

eclogite melts plotted in Figure 3b resemble those

of Batan inclusions in several respects: they have

similar overall enrichments of highly incompatible

elements relative to moderately incompatible and

compatible elements, and all share a positive

anomaly in Sr and negative anomalies in Ba and

Nb. These are among the reasons why previous

study of Batan inclusions suggested they were well

explained as slab melts [

Schiano et al.

, 1995].

[

] On the other hand, there are also several

differences between these model slab melts and

Figure 3.

(a) Primitive-mantle-normalized trace element abundances in Batan inclusions and inclusion-rich

olivines. The dashed portions of the line for Batan inclusions indicate inferred abundances in inclusion glass of

elements that were analyzed in inclusion-rich olivines but not in inclusions alone. This estimate was made by

increasing the measured trace element concentrations for inclusion-rich olivines by a multiple equal to the average

measured concentration ratio for inclusions versus inclusion-rich olivines for elements of similar compatibility that

were analyzed in both (i.e., Cs, Rb, Ba, Th, and U in inclusions were calculated on the basis of measurements of Nb,

K, La, and Ce in inclusions and inclusion-rich olivines; Pb and Pr were calculated on the basis of data for Ce and Nd;

Eu, Gd, and Tb were calculated on the basis of data for Sm and Dy; and Er, Cu, and Sn were calculated on the basis of

data for Ti and the heavy-rare earths). (b) Primitive-mantle-normalized trace element abundances calculated for melts

in equilibrium with altered basalt having an eclogitic mineralogy compared to the average composition of Batan

inclusions (from Figure 3a). The names used to refer to each model correspond to the first authors of the publications

from which the relevant distribution coefficients were taken, including

Klemme et al.

[2002],

Xiong

[2006],

Kessel et

al.

[2005b],

Salters et al.

[2002], and

Kelemen et al.

[2003b]. The composition of altered basalt was taken from

Bach

et al.

[2003]. All data and models are normalized by the primitive mantle composition of

McDonough and Sun

[1995]. Data for Batan inclusions and inclusion-rich olivines are taken from

Schiano et al.

[1995]. See the auxiliary

material for the distribution coefficients and the altered basalt compositions used to calculate all model melt

compositions.

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

9of21

Batan inclusions: Ba/Th ratios of Batan inclusions

are



2 to 5-times higher than predicted by any of

these eclogite melting models; this could be

explained by slab melting only if those slab sources

were unusually Ba-rich. All models predict prim-

itive-mantle-normalized U/Th ratios (‘‘(U/Th)

’’)

lower than 1, whereas Batan inclusions have (U/

Th)

ratios of



1.5; thus slab-melting models of

Batan inclusions would also require unusually

U-rich sources. All slab-melting models that lack

rutile in the residue yield U/Nb ratios lower by

factors of



5 to 20 than those observed in Batan

inclusions. The models that contain 1% rutile in the

residue can explain this feature. However, it is not

clear whether Batan inclusions could be derived

from rutile-bearing residues. Batan inclusions con-

tain only 0.25 wt.% TiO

, lower by a factor of 3.5

than the most TiO

-poor rutile-saturated eclogite

melts produced by eclogite melting experiments at

shallow mantle pressures, and far lower than most

such melts (typically 2–6 wt.% TiO

[

Rapp and

Watson

, 1995;

Pertermann and Hirschmann

2003]). It is possible that some combination of

previously unexplored temperatures, pressures,

source compositions and degrees of melting will

yield appropriately low-Ti melts of rutile-bearing

eclogite sources. Also, one might imagine ways of

lowering the Ti contents of slab melts by reaction

with the mantle wedge [

Kelemen et al.

, 1990].

Nevertheless, the Ti contents of Batan inclusions

are clearly lower than expected for melts of rutile-

bearing eclogite based on published experimental

constraints, and thus the high U/Nb ratios of these

inclusions are poorly explained by slab melting.

[

] Batan inclusions are characterized by a large

positive Pb anomaly. Neither of the two experi-

mental studies of eclogite melting produced melts

with this feature. The models that use estimated

distribution coefficients based on data from non-

eclogitic systems (models ‘‘Kelemen’’ and ‘‘Salt-

ers’’ in Figure 3b) produce Pb anomalies nearly as

large as that in Batan inclusions, but the low solid/

melt Pb distribution coefficients they require have

not yet been confirmed by experiment.

Kamber et

al.

[2002] present an analysis of Pb abundances in

arc-related lavas that relates to this issue. Pb/Nd

ratios (an approximate measure of the size of the

Pb anomaly) in typical island arc basalts average

0.5, strongly elevated relative to values of primitive

or depleted mantle (0.12 and 0.04, respectively) or

altered MORB (0.09).

Kamber et al.

[2002] attri-

bute this characteristic of arc lavas to preferential

extraction of Pb from subducted slabs on release of

aqueous fluid. Adakites typically have lower Pb/

Nd ratios of approximately 0.35, more consistent

with ratios expected for slab-derived melts [

Kamber

et al.

, 2002]. Pb/Nd ratios of Batan inclusions

average 1.9, greatly in excess of adakites or even

normal arc basalts and, in the context of

Kamber et

al.

’s [2002] interpretation, strongly indicative of

enrichment in fluid-soluble elements relative to

‘‘normal’’ subducting crust.

[

] All model eclogite melts have negative Zr

anomalies and positive Sr anomalies, resulting in

(Sr/Zr)

ratios between



5 and



50. Measured

(Sr/Zr)

ratios of Batan inclusions are only mod-

estly elevated, averaging only 1.5, and these inclu-

sions do not have negative Zr anomalies.

[

] Finally, experimental eclogite melts are char-

acterized by pronounced depletions in heavy rare

earths and Y due to compatibility of these elements

in garnet, but relatively high contents of Ti. As a

result, all four models based on distribution coef-

ficients observed in eclogite melting experiments

yield high (Ti/Y)

ratios (from



20 for model

‘‘Xiong’’ to



4.6 for model ‘‘Kessel’’; Figure 3b).

Lower ratios are predicted by the ‘‘Kelemen’’ mod-

els (2.1, or 1.8 with residual rutile) because this

model assumes an exceptionally high solid/melt

distribution coefficient for Ti in the silicate phases

in eclogite. Nevertheless, even these ratios are far

higher than those in Batan inclusions (average (Ti/

ratio of



0.7). Similarly, Batan inclusions are

characterized by a negative Ti anomaly and none of

the model melts have such a feature.

[

] Trace element partitioning during eclogite

melting is relatively poorly understood, and it is

possible that future work will reconcile the above-

noted differences between Batan inclusions and

model eclogite melts. Moreover, our model esti-

mates are based on the altered ocean crust compo-

sition as given by

Bach et al.

[2003]. This estimate is

based on a large body of evidence, but it is possible

that some of the discrepancies we have noted could

simply reflect an unexpected composition for the

crust subducted beneath Batan. Nevertheless, these

discrepancies pose a number of difficulties for slab

melting models of the origin of Batan inclusions,

and it is noteworthy that several of these discrep-

ancies involve elements that are relatively soluble in

aqueous fluids (Ba, U, Pb).

5.2. High-Pressure Differentiation of

Island Arc Basalt?

[

]

Macpherson et al.

[2006] presented a model

for the origins of adakites from Mindanau, the

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

10 of 21

Philippines, involving high-pressure (



1Gpa;

comparable to the depth from which Batan xen-

oliths come) differentiation of hydrous island arc

basalt in the upper levels of the mantle wedge. At

these pressures, precipitation of garnet and amphi-

bole can yield residual liquids with characteristic

adakitic trace element compositions (i.e., low abun-

dances of Y and heavy rare earths and high Sr/Y

ratios). Moreover, because many island arc basalts

are derived from mantle sources that were meta-

somatized by slab-derived aqueous fluids prior to

or during partial melting, this process could yield

broadly adakitic liquids having positive anomalies

in fluid-soluble trace elements and high ratios of

relatively fluid-soluble to relatively fluid-insoluble

elements (e.g., Pb/Ce, Ba/Th, U/Th and U/Nb), as

we observe for Batan inclusions (Figure 3a). Finally,

we estimate, on the basis of major element com-

positions of Batan island basalts [

Sajona et al.

2000], phase assemblages that crystallize from

such basalts at high pressure [

Muntener et al.

2001], and oxygen isotope fractionations among

relevant minerals and silicate melts [

Eiler

, 2001,

and references therein], that the required tens of %

crystallization should only lead to increases of



0.2 to 0.5

in the

O of the residual melt.

Given that island arc basalts, even those with

pronounced geochemical signatures of slab-derived

components, generally have

O values within



0.5

of typical mantle-derived basalts [

Eiler et

al.

, 2000], this process could generate liquids with

O values near equilibrium with mantle perido-

tite, as we observe for Batan inclusions.

[

]

Macpherson et al.

[2006] present a model for

high-pressure crystallization that can be compared

with the trace element compositions of Batan

inclusions and related lavas. Their description of

high-pressure differentiation yields an expected

trend of increasing Sr/Y ratio with decreasing Y

content that has its sharpest increase toward rela-

tively high Sr/Y (>50) at moderate Y concentration

(



10 ppm; Figure 4a). The trend in Figure 4a

defined by Batan basalts, basaltic andesites and

the Batan inclusions examined in this study broadly

resembles that predicted by the Macpherson et al.

model, although Batan inclusions increase toward

high Sr/Yat somewhat lower Y content (



5 ppm Y).

We examine this model further in Figure 4b by

comparing the Sr/Y ratio versus the MgO content

for model hydrous basalt undergoing high-pressure

crystallization with data for Batan inclusions

and related lavas. High-pressure crystallization-

differentiation is predicted to produce a nearly linear

trend of monotonically increasing Sr/Y ratio with

decreasing MgO content. In contrast, Batan lavas

vary over a wide range in MgO (from



10 to

2.5 wt.%) with no consistent trend in Sr/Y ratio,

and Batan inclusions depart markedly from that

trend toward high Sr/Y at



1 wt.% MgO. The few

Batan lavas that have Sr/Y ratios within the ‘‘ada-

kitic’’ range (



50 or higher) are distributed seem-

ingly randomly across the entire observed range in

MgO. It is possible that the data trend for Batan lavas

and inclusions plotted in Figure 4b reflects both

high- and low-pressure differentiation in some com-

bination that obscures the trend predicted by the

model of

Macpherson et al.

[2006]. Nevertheless,

there is no simple relationship between Sr/Y ratio

and differentiation in the Batan suites, despite the

fact that this suite defines simple, continuous geo-

chemical trends in other dimensions (e.g., Figure 1).

Thus, while it is possible that the adakitic trace

element compositions of Batan inclusions are a

result of high-pressure crystallization, there is no

clear evidence that this is the case.

5.3. Low-Degree Melting of

Metasomatized Mantle-Wedge Peridotite?

[

] In this section, we examine whether Batan

inclusions can be explained as low-degree melts

of metasomatized mantle wedge peridotite. This

hypothesis is motivated by two factors: (1) the

O values of Batan inclusions are indistinguish-

able from those expected of peridotite partial melts,

and (2) their trace element compositions are char-

acterized by several features that are expected for

melts of sources metasomatized by slab-derived

aqueous fluids, including high Ba/Th, U/Th, U/Nb,

and Pb/Ce [

Brenan et al.

, 1995a, 1995b;

Keppler

1996;

Stalder et al.

, 1998;

Kessel et al.

, 2005b].

Similarly, although the partitioning behavior of Cu

is not well-known, previous studies of island arc

and back-arc lavas have found that pronounced

enrichments in Cu, comparable to those seen in

Batan inclusions, are associated with enrichments

in other fluid-soluble elements [

Stolper and Newman

1994;

Eiler et al.

, 2005]. We are not aware of any

previous discussion of Sn anomalies in arc-related

magmas, but present these data for possible future

reference.

[

] Partial melting experiments conducted on peri-

dotites and synthetic peridotite analogues show

that alkali-rich melts formed at pressures less than

1.5 GPa can have silica contents approaching those

of Batan melt inclusions [

Hirschmann et al.

, 1998,

and references therein]. In the absence of water and

using natural starting materials, these experiments

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

11 of 21

fail to produce silica contents quite as high as those

of Batan inclusions (avg. 59.6 wt.% SiO

[

Schiano

et al.

, 1995]) at appropriate total alkali contents

(7 to 8 wt.% Na

O+K

O). However, Batan glass

inclusions contain



5wt.%H

O (and likely

contained even more before exsolving a water-rich

fluid), so the combined influences of high alkali

and water contents could be sufficient to produce

such high silica contents in low-degree, low-

pressure melts of peridotite [

Hirschmann et al.

1998, 1999;

Gaetani and Grove

, 1998, 2003].

1.2 GPa experiments by

Draper and Green

[1999] have shown that liquids with



60 wt.%

SiO



13 wt.% Na

O+K

O and



2.5 wt.%

O+CO

are saturated with olivine, orthopyr-

oxene and clinopyroxene at 1150

°

C. The MgO

contents and Mg/Fe ratios of Batan inclusions are

far lower than those of experimental peridotite

melts, including Si- and H-rich melts. Given the

evidence for formation of metasomatic hydrous

phases and neoblastic olivine and orthopyroxene

in Batan xenoliths, this could reflect modification

of the major element compositions of Batan inclu-

sions by crystallization-differentiation before, dur-

ing and/or after inclusion entrapment. We conclude

that the major element compositions of Batan

inclusions are not clearly distinctive of peridotite

melts, but could be peridotite melts that were

modified by crystallization-differentiation.

[

] We further examine this hypothesis by way of

the following trace element model: Batan xenoliths

and abyssal peridotites from the Bouvet fracture

zone [

Johnson et al.

, 1990] are similar to one

another in their modal abundances of clinopyrox-

ene [

Maury et al.

, 1992;

Johnson et al.

, 1990], and

Bouvet peridotites appear to be typical examples of

Figure 4.

Comparison of trends in (a) Sr/Y versus Y and (b) Sr/Y versus MgO for Batan inclusions (filled circles;

data from

Schiano et al.

[1995] and the auxiliary material of this paper) and related lavas (unfilled squares; data from

Sajona et al.

[2000]) with predicted trends for a model of high-pressure crystallization of hydrous basalt (solid

curves). The model trend assumes a fractionating assemblage and trace element distribution coefficients as given by

Macpherson et al.

[2006], mineral compositions from experiment B726 of

Muntener et al.

[2001], an initial MgO

content of 9 wt.%, and initial Sr and Y contents of 500 and 14 ppm, respectively, based on data for Batan island lavas

from

Sajona et al.

[2000]. The model trends represent 0 to 50 wt.% crystallization, with arrows indicating the

direction of increasing extent of crystallization. High-pressure differentiation predicts trends of Sr/Y versus Y that

broadly resemble those of Batan inclusions and lavas, but does not resemble the distribution of data in dimensions of

Sr/Y versus MgO.

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

12 of 21

the residues of high degrees of fractional melting.

Consequentially, we describe a model peridotitic

source for Batan inclusions, prior to being meta-

somatized, as a depleted peridotite having a trace

element composition similar to harzburgites from

the Bouvet fracture zone.

Johnson et al.

[1990]

analyzed only select elements in clinopyroxene

from the Bouvet fracture zone, so we extend their

results to other elements of interest by assuming

clinopyroxenes have smooth chondrite-normalized

trace element patterns for elements as or more

compatible than Nb (i.e., interpolating between

Johnson et al.

’s [1990] measurements of Ce, Nd,

Zr, Sr, Sm, Ti and Yb), and contain no U, Th, Ba,

Rb or Cs (i.e., budgets of these elements in our

model metasomatized source are completely con-

trolled by the metasomatic agent). We calculated

the trace element contents of coexisting olivines,

orthopyroxenes, and spinels by multiplying clino-

pyroxene compositions by the ratio of the relevant

mineral-melt distribution coefficient to the clino-

pyroxene-melt distribution coefficient (using data

from

Green and Pearson

[1983],

Colson et al.

[1988],

Kelemen et al.

[1990],

Hart and Dunn

[1993],

Kennedy et al.

[1993],

Nielsen et al.

[1993, 1994], and

Green

[1994]). The resulting

model trace element composition of a depleted

peridotite source closely resembles measured trace

element compositions of whole-rock samples of

harzburgites from the Voykar ophiolite [

Sharma et

al.

, 1995]. We then added to this depleted perido-

tite 0.5 wt.% of a model metasomatic agent (i.e., so

little that it will not substantially change the

of the metasomatized source even if that agent is

slab-derived) and modeled 1% melting of the

hybrid source (i.e., a low enough degree that

product melt will contain tens of wt.% H

O). We

then solved for the composition of the metasomatic

agent needed to produce a model melt having the

composition of the average of Batan inclusions

(i.e., by solving the set of simultaneous linear

equations describing mineral-melt partitioning

and mass balance, assuming the constraints

described above; see Table S8 in the auxiliary

material for relevant distribution coefficients).

[

] This calculation is representative of a family

of broadly similar models that differ from one

another in the degree of depletion of the mantle

wedge prior to metasomatism, in the amount of

metasomatic agent, and in the degree of melting of

the hybrid source. Variations in these parameters

lead to variations in absolute concentrations of

trace elements in the model metasomatic agent

required to fit the data for Batan inclusions. How-

ever, all successful models of this sort require

initially depleted sources (i.e., prior to metasoma-

tism), small amounts of slab-derived component,

and enrichment in fluid-soluble elements in that

slab-derived component. Table 2 presents the com-

positions of model depleted peridotite, the solved-

for model metasomatic agent, and the model hybrid

source.

[

] Figure 5a compares the model metasomatic

fluid calculated using the above-described model to

slab-derived components previously inferred to

contribute to the sources of arc and back-arc lavas

from a variety of geographic locations [

McCulloch

and Gamble

, 1991;

Stolper and Newman

, 1994;

Eiler et al.

, 2000;

Grove et al.

, 2002] (and the

‘‘slab fluid’’ component from

Eiler et al.

[2005]).

Our model metasomatic agent has relative abun-

dances of Th, U, Nb, K, La and Ce that closely

resemble those for all of these previously estimated

slab-derived components (Figure 5a), but it has a

significantly lower Ba/Th ratio. The Pb anomaly in

our model metasomatic agent is similar in size to

that for the slab-derived component of

McCulloch

and Gamble

[1991], and greater than those for the

slab-derived components of

Stolper and Newman

[1994] and

Grove et al.

[2002] (

Eiler et al.

[2000]

did not estimate the Pb abundance in their slab-

derived component). Middle-rare earth and Sr

abundances in our model metasomatic agent

resemble those for all of these previously estimated

slab-derived components, although our model

component has a smaller positive Sr anomaly and

lacks the negative Zr anomaly shared by those

components. It is possible that the difference in

Zr content reflects our overestimation of the degree

of prior depletion of the peridotite source of Batan

inclusions prior to metasomatism. Such a change to

our model of Batan inclusions would slightly lower

all incompatible trace element concentrations in

our model fluid and exaggerate its negative Nb and

Ti anomalies. Finally, abundances of elements

more compatible than Sm in our model metaso-

matic agent are at the low end of the range

previously estimated for slab-derived components

and most resemble those from

McCulloch and

Gamble

[1991] and

Eiler et al.

[2005].

[

] Figure 5b compares the composition of our

model metasomatic agent to compositions of var-

ious aqueous fluids and aqueous supercritical

phases in equilibrium with altered oceanic crust

[

Bach et al.

, 2003] in the eclogite facies, based on

laboratory fluid/mineral or fluid/bulk eclogite par-

titioning experiments [

Brenan et al.

, 1995a, 1995b;

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

13 of 21

Stalder et al.

, 1998;

Kessel et al.

, 2005b]. Note that

the model labeled ‘‘Kessel 6 GPa’’ uses data for

partitioning between eclogite and a ‘‘melt-like’’

aqueous supercritical phase; this is the same as

model ‘‘Kessel’’ in Figure 3b. We also show the

expected composition of aqueous fluid in equilib-

rium with depleted mantle peridotite (

Ayers et al.

[1997] using the depleted mantle composition of

Workman and Hart

[2005]). In all cases, we

assume equilibration between 1% fluid and 99%

solid; a subset of these models assume 1% rutile in

the solid, as indicated in the legend. We do not

consider fluids in equilibrium with subducted ma-

rine sediment because, as mentioned above, meta-

somatized Batan xenoliths have

Sr/

Sr ratios far

too low for them to contain any appreciable

amount of Sr from subducted sediment.

[

] The expected trace element concentrations in

aqueous phases in equilibrium with subducted

crust vary widely both in absolute concentrations

(by two to three orders of magnitude for many

elements) and in abundance ratios (e.g., U/Nb

ratios of fluids in equilibrium with rutile-free

eclogite differ by a factor of 20 between

Brenan

et al.

[1995a, 1995b] and the 4 GPa experiment of

Kessel et al.

[2005b]). This diversity could reflect

any of a number of differences between these

various experiments, such as temperature, pressure,

solution chemistry (e.g., chloride content), the

presence or absence of rare but trace element-rich

phases, and/or experimental artifacts such as partial

equilibration or elemental gain to or loss from the

experimental charge. One factor that clearly

appears to be important is that concentrations of

most elements increase, and positive and negative

anomalies diminish, with increasing pressure.

These changes occur over a relatively narrow range

of pressure, from ‘‘fluid-like’’ low concentrations

and large element-specific anomalies at 4 GPa to

‘‘melt-like’’ high concentrations and modest

anomalies at 6 Gpa [

Kessel et al.

, 2005b].

[

] Our model metasomatic agent (filled circles in

Figure 5b) is characterized by abundances of rare

earth elements, Zr, Sr, and Nb that approach those

for the 6 GPa supercritical aqueous phase in

equilibrium with rutile-absent altered oceanic crust

basedontheresultsof

Kessel et al.

[2005b]

(unfilled circles). On the other hand, the positive

Pb anomaly, high U/Nb and U/Th ratios and high

Cs, Rb and Ba contents relative to other incompat-

ible elements characteristic of our model metaso-

matic agent more closely resemble aqueous fluid in

equilibrium with subducted crust at somewhat

lower pressures of 4 or 5 GPa [

Stalder et al.

1998;

Kessel et al.

, 2005b]. This apparent contra-

diction could indicate that our model metasomatic

agent last equilibrated with a subducted slab at

pressures between 4 and 6 GPa (and possibly at

temperatures different from those examined by

Kessel et al.

[2005b]), and thus had a trace element

composition intermediate to our ‘‘Kessel 4 GPa’’

and ‘‘Kessel-6 GPa’’ models. It is also possible that

the sources of Batan inclusions were metasomatized

by two or more components, one extracted from the

slab at relatively high pressure and one at relatively

low pressure, and our model metasomatic agent has

averaged the compositions of these components.

[

] Despite the difficulty in matching our model

metasomatic agent with a specific previous fluid-

solid partitioning experiment, the generally ‘‘fluid-

like’’ character of its trace element composition is

confirmed by its Pb/Sr ratio. We are not aware that

this ratio has been examined for such a purpose

before, but it is useful because it involves two

elements that are similarly incompatible during

melting for all common silicate and oxide minerals

at mantle pressures (i.e., excluding plagioclase),

but strongly fractionated from one another during

fluid/mineral partitioning. In this respect, it differs

Table 2.

Compositions of Model Components

Element

Depleted

Peridotite

Metasomatic

Agent

Hybrid

Source

0.0

0.06

0.0

250

1.25

0.0

1600

8.0

0.0

0.13

0.0

0.05

0.000066

0.099

0.024

51500

258

0.000065

112

0.56

0.00043

192

0.96

0.000042

130

0.65

0.000081

0.14

0.00044

0.25

0.021

648

3.3

0.046

1570

7.9

0.0016

7.7

0.040

0.0011

1.1

0.0066

0.0054

6.5

0.038

0.0020

0.9

0.0065

0.020

3.0

0.035

2000

0.13

0.18

0.031

1.0

0.036

0.046

1.0

0.051

3.0

8000

0.00052

120

0.60

All values are parts per million, by mass.

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

14 of 21

from the U/Nb ratio, which can be elevated due to

fluid/mineral partitioning or melting in the pres-

ence of residual rutile, and from the Sr/Sm and Pb/

Ce ratios, which can be elevated by both fluid/

mineral partitioning and by melting in the presence

of residual garnet.

[

] Figure 6 compares the primitive-mantle-

normalized Pb/Sr ratio of our model metasomatic

agent and average Batan inclusions (horizontal

bars) with the ratios expected of aqueous fluid,

melt and aqueous supercritical phase in equilibrium

with altered ocean crust. For comparison, we also

show Pb/Sr ratios expected for melts of peridotitic

depleted mantle. In each case, the modeled Pb/Sr

ratios are plotted versus the fraction of fluid or

melt, indicating the range of possible compositions

that might result from variations in degree of

dehydration or melting. Pb/Sr ratios of relatively

low-pressure aqueous fluids (i.e., 2 GPa from

Brenan et al.

[1995a, 1995b] and 4 GPa from

Figure 5.

Comparison of the trace element composition of the calculated model metasomatic agent (Table 2)

required to explain the compositions of Batan melt inclusions, assuming they formed by low degrees of melting of

metasomatized depleted peridotite (see text), with (a) model slab-derived components previously inferred to

contribute to the sources of arc lavas [

McCulloch and Gamble

, 1991;

Stolper and Newman

, 1994;

Eiler

, 2001;

Grove

et al.

, 2002] (the ‘‘aqueous fluid’’ component of

Eiler et al.

[2005]) and (b) aqueous fluids or melts calculated to be in

equilibrium with altered oceanic crust (based on partitioning experiments of

Brenan et al.

[1995a, 1995b] at 2 GPa;

Stalder et al.

[1998] at 5 GPa; and

Kessel et al.

[2005b] at 4–6 GPa) or with depleted mantle peridotite (based on

partitioning experiments of

Ayers et al.

[1997]). Note that the data based on 6 GPa data from

Kessel et al.

[2005b]

involve a broadly ‘‘melt-like’’ aqueous supercritical phase and were also included in Figure 3b. All models assume

1% fluid and 99% solid. Eclogites are assumed to be 60% garnet and 40% clinopyroxene (‘‘Kessel 4 GPa –rutile’’

and ‘‘Stalder 5 GPa – rutile’’ include 1% rutile) and to have the composition of altered basalt from

Bach et al.

[2003].

All models are normalized by the primitive mantle composition of

McDonough and Sun

[1995]. The depleted mantle

model (‘‘Ayers-peridotite’’) assumes the solid is a peridotite having the composition of depleted mantle from

Workman and Hart

[2005] (interpolated to yield concentrations of Cs, K, and Pb). Distribution coefficients and

source concentrations used to construct these models are provided in the auxiliary material.

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

15 of 21

Kessel et al.

[1995b, 2005b]) are an order of magnitude

higher than those of any melt or aqueous super-

critical phase at any degree of melting or dehydra-

tion; only 5 GPa aqueous fluid lies between these

extremes. Our model metasomatic agent has a Pb/

Sr ratio consistent only with those of relatively

low-pressure (



4 GPa) slab-derived aqueous flu-

ids. It is possible that this comparison is compro-

mised by some artifact in the measurement of Pb/Sr

ratios in Batan inclusions (e.g., selective enrich-

ment in sulfide; although we consider this unlikely

given the relative homogeneity in the size of the Pb

anomaly exhibited by those measurements). It

is also possible that the high Pb/Sr ratios seen in

2–4 GPa fluids are an experimental artifact caused

by Pb loss from the experimental charge (although

it is not obvious why this artifact would effect only

the low-pressure fluid experiments). Despite these

ambiguities, we suggest that the high Pb/Sr ratios

of Batan inclusions strengthen the argument that

fluid/mineral partitioning played an important role

in their origin.

[

] More generally, Figure 6 provides insight into

the possible roles of slab melts, fluids and aqueous

supercritical phases in arc magma genesis. The

primitive-mantle-normalized Pb/Sr ratios of island

arc basalts generally vary between 1 and 1.5

[

McCulloch and Gamble

, 1991] and slab-derived

components previously inferred to contribute to

arc-related lavas are similarly low (e.g., 1.27 for

McCulloch and Gamble

[1991], 0.48 for

Stolper

and Newman

[1994], 0.16 for

Grove et al.

[2002],

and 0.75 for

Eiler et al.

[2005]). Thus arc-related

magmas generally have Pb/Sr ratios far lower than

those of low-pressure aqueous fluids in equilibrium

with subducted slabs, and are better explained by

slab contributions from slab-derived melt or aque-

ous supercritical phase. Batan inclusions (and the

model metasomatic agent we infer controlled their

trace element compositions), despite having many

properties resembling slab-derived components

elsewhere (Figure 5a), are clearly unusual in Pb/

Sr ratio. One interpretation of this result is that a

relatively low-pressure slab-derived aqueous fluid

controls the compositions of Batan inclusions, but

that such fluids are relatively minor contributors to

the sources of most arc lavas (i.e., as compared to

slab-derived melts or aqueous supercritical phases).

[

] Finally, we note that no experimentally based

fluid compositions contain the negative Ti anomaly

characteristic of our model metasomatic agent (as

was also true for slab-derived melts; Figure 3b).

Residual rutile can retain Ti in subducted slabs. We

found this was not an acceptable explanation of the

Ti contents of Batan inclusions if they are slab melts

because rutile-saturated melts at relevant temper-

atures and pressures are too rich in TiO

[

Rapp and

Watson

, 1995;

Pertermann and Hirschmann

, 2003].

However, it is possible that rutile-saturated aque-

ous fluids could have negative Ti anomalies under

some conditions relevant to subducted slabs. This

is an attractive target for further experimental

Figure 6.

Comparison of the primitive-mantle-normalized Pb/Sr ratios (‘‘(Pb/Sr)

’’) of Batan inclusions and model

metasomatic agent (horizontal lines) with (Pb/Sr)

ratios predicted for aqueous fluids in equilibrium with altered

basalt (based on experiments of

Brenan et al.

[1995a, 1995b],

Stalder et al.

[1998], and

Kessel et al.

[2005b]), partial

melt of altered basalt in the eclogite facies (based on data from

Klemme et al.

[2002]), and partial melt of depleted

mantle peridotite (using distribution coefficients summarized by

Eiler et al.

[2005]). All models are plotted as

functions of the fraction of melt or fluid (F). Distribution coefficients and source compositions used to construct these

models are provided in the auxiliary material.

Geochemistry

Geophysics

Geosystems

eiler et al.: silica-rich melts in the subarc mantle

10.1029/2006GC001503

16 of 21