Earth and Planetary Sctence Letters, 113 (1992) 1-14 1

Elsevier Science Pubhshers B.V, Amsterdam

[uc]

Trace and minor element partitioning between garnet

and amphibole and carbonatitic melt

R.J. Sweeney a, D.H. Green a and S.H. Sie b

a Department of Geology, Unwerstty of Tasmama, G.P.O. Box 252C, Hobart 7001, Austraha

b CSIRO Dwtston of Exploration Geoscwnce, P O. Box 136, North Ryde, NSW 2113, Austraha
Recewed November 14, 1991; revision accepted July 3, 1992

ABSTRACT

The very low viscosity of carbonatlte melts potentially makes them a highly effectwe agent of mantle metasomat~sm and
It is important to quantify how the interaction of these melts with a lherzohtic mantle may affect trace element budgets. Our
experiments yield measurements of amph~bole-carbonat~te melt and garnet-carbonat~te melt partition coefficients for the
appropriate P,T cond,tlons (18-33 kbar) for a state of trace and minor elements (Ti, Nb, Zr, Ta, Ce, Lu, Y, Sr, Ba, Rb and
K). Measured partition coefficients permit prediction of fractlonation among trace element ratios in a carbonatite melt
relative to garnet and amphibole m the mantle source for th~s melt. For example, equilibration of a carbonat~te melt w~th
garnet wdl not fractlonate Z r / Y or N b / T a ratios m the melt but will decrease Zr/Nb, Lu/Y, Z r / L I L E and Y / L I L E ratios
in the melt. A carbonatite melt m equilibrium with amphibole will fractlonate Z r / N b w,th Nb enrichment In the melt, but to
a lesser extent than for garnet Metasomat~sm of the mantle wedge beneath island arcs through addition of carbonat~te melt
may explain the lower Z r / N b of island-arc basalts relative to mid-ocean ridge basalts Tl behaved compatibly with garnet
and amphibole m all experiments, amplymg relative depletion of T~ m coexisting carbonatlte melt and a decrease In
TI/LILE ratio in such a melt under equihbrlum conditions.

I. Introduction in a mantle environment. A carbonatitic melt (of

sodic dolomitic composition) may be in equilib-
In order to explain variations in incompatible rium with a fertile amphibole (pargasite) lherzo-
elements of several orders of magnitude in man- lite composition (approximating natural systems)
tle samples and mantle-derived melts, geo- under conditions from 21 to 30 kbar and 930 to
chemists and petrologists have commonly ap- 1080°C [6]. Further, Green and Wallace [7]
pealed to migration and crystallization of small showed that such melts may undergo decarbona-
melt fractions. Terms such as 'zone refining' [1], tion reactions at shallower levels and will then
'wall-rock reaction'[2], 'mantle metasomatism' [3], metasomatize the mantle. These authors [7] dealt
'enriched' or 'fertile' vs 'depleted' or 'refractory' specifically with the major element compositional
mantle, have been applied in models of how such and mineralogical changes that may be expected.
melt and mantle inhomogeneity may arise. Theo- Earlier work in synthetic peridotite and carbon-
retical modelling of melt migration through ate systems also documents the mineralogical
porous media [4] and experimental grain bound- changes which such interactions would cause [8-
ary studies examining the movement of carbonate 11]. Green and Wallace [7] also suggested that
melt through an olivine matrix [5] have led to the mantle metasomatism by a dolomitic carbonatite
view that carbonatite melts are extremely mobile melt would be an effective process to cause en-
richments in LILE, P and some HFSE within
lithospheric mantle, particularly beneath conti-
Correspondence to: R.J. Sweeney, Department of Geology, nental regions. This process has recently been
University of Tasmania, G.P.O. Box 252C, Hobart 7001, Aus- described from a suite of mantle xenoliths [12].
tralia. Specifically, the authors [7] noted the behaviour

0012-821X/92/$05.00 © 1992 - Elsevier Science Pubhshers B.V All rights reserved

2 R J SWEENEY ET AL

of Ti which has a low solubility in the carbonatite amphibole to fractionate particular incompatible
melt and is partitioned into amphibole, clinopy- element ratios, producing distinctive minor and
roxene and ilmenite, and predicted fractionation trace element pattern differences among melts
among HFSE elements and between HFSE ele- and between melts and residues.
ments, P and LILE due to sensitivity of carbon-
atite/amphibole element partitioning relation- 2. Experimental technique
ships.
In this paper we summarize the major element The Wallace and Green [6] 'primary' sodic
phase relations of the Wallace and Green car- dolomitic carbonatite composition (labelled CM1
bonatite compositions examined by Sweeney et here) was doped with trace and minor levels of
al. [13]. Chemical data on the silicate phases and Nb, Zr, Ta, Y, Ce, Lu, Ba and Sr (CM1 + 4%
co-existing carbonatite melts in these runs were H 2 0 given in Table 2) to correspond with the
obtained using both electron-probe (EPMA for range of observed abundances in natural carbon-
major and minor element oxides) and proton- atites [14] but with elevated Zr, Y and Lu to
probe (PIXE for Nb, Zr, Ta, Ce, Lu, Y, Sr, Ba facilitate measurement. This does not imply that
and Rb) microbeam methods. These data are such concentration levels are typical of a primary
then used to assess equilibrium in the experi- carbonatite magma, but the values are an approx-
ments and the merits of both analytical tech- imation. This approximation is considered rea-
niques in determining element abundances. This sonable with respect to the wide range of these
enables the calculation of certain minor and trace element abundances in natural carbonatites [14],
element partitioning behaviour between amphi- and therefore within the uncertainties of possible
bole and carbonatite melt and garnet and carbon- mantle carbonatite liquid compositions.
atite melt. The papers addresses the potential for The doped elements were added as oxides (of
a carbonatitic melt co-existing with garnet and Nb, Zr, Ta, Y, Ce and Lu) and carbonates (Ba

TABLE 1
Experimental run details for the bulk carbonaute composiuon CM1 (Table 2)

Experiment P (kbar) T (°C) t (h, rain) %melt liquidus assemblage

CM1 + 2% H 2° .
T3478 27 1000 40h00 40% dol + ga (+ titan)
T3351 27 1000 66h00 50% dol + ga + a m p h ( + t~tan)
T3349 22 1000 73h00 65% dol + a m p h + sp ( + titan + zircon + ilm)
T3509 18 960 47h00 55% dol + a m p h + titan ( + zircon)
T3177 15 1225 2h00 100% above liquidus
T3192 15 1125 2h30 100% above liquidus
T3438 15 1000 2h08 83% dol ( + i l m )
T3393 15 850 43h23 4% not equihbrium

CM1 + 4 % H 2 0 '
T3506 33 1000 46h30 38% dol + ga ( + titan)
T3501 30 1000 47h40 43% dol + ga
T3372 27 1000 ca. 24h 66% dol + a m p h + sp ( + titan + zircon)
T3434 22 1120 2h05 100% above hquidus
T3404 22 1000 42h10 67% dol + amph + sp ( + t~tan)
T3474 18 960 44h53 51% dol + a m p h ( + sp + tim)
T3427 15 1050 3h00 98% ol ( + sp)
T3400 15 1000 24h00 94% dol + ol + sp ( + zircon)
T3403 15 850 31h40 < 2% not equilibrium
Experiments were conducted with 2% and 4% (by weight) added water. A n estimate is m a d e of the a m o u n t of melt present m each
run (mostly within + 10% relative error). Also indicated (unbracketed) are the major tiqmdus phases and minor (bracketed < 1
vol% of the liquldus assemblage) m each run (dol ~ dolomite, ga = garnet, a m p h = amphibole, ol = olivine, sp = spinel, ilm =
ilmemte, titan = Nb-Ta Utanate)
T R A C E AND MINOR ELEMENT PARTITIONING BETWEEN G A R N E T AND AMPHIBOLE AND CARBONATITIC MELT 3

and Sr), and Rb was present as an impurity in the age. Corrections were made for the interference
constituent carbonates. The composition used of Ti K/3 on Ba L a and Ba L/3 on Ce L a. The
contained chlorine, fluorine and sulphur added as use of the electron microprobe for the analysis of
CaC12, MgF 2 and CaSO 4 in amounts estimated these trace elements and the effectiveness of the
from abundances in natrocarbonatite form standardization and calibration with regard to
Oldoinyo Lengai [15] assuming ca. 70% fractional minimum detection limits (MDL typically 50-80
crystallization of olivine + spinel [6] from a pri- ppm), counting precisions (typically 5-10% rela-
mary carbonatite. Experiments were conducted in tive for 2 standard deviations) and accuracy (by
two bulk systems differentiated on water content: comparison with secondary standards) has been
water was added as AI(OH) 3 to give 2% (by described in detail elsewhere [17].
weight) bulk water content and 'free' water (an Run products were also analysed by proton-in-
extra 2% by weight) was added to another series duced X-ray emission (PIXE) techniques devel-
of experiments to give 4% H 2 0 (by weight) in the oped at CSIRO (North Ryde) and based on a 3
system. All runs were carried out in small capac- MeV Tandetron accelerator [18]. The proton
ity Ag75Pd25 capsules (outer diameter 2.2 mm) in beam was focussed to 2 0 - 4 0 / z m in diameter for
a high-pressure piston cylinder apparatus at the silicates and 150-200 /zm in diameter for the
University of Tasmania. A 0.5 inch diameter NaCI quenched carbonate matrix. The effective depth
or NaCl-pyrex assembly was used with a graphite of analysis is typically 20-30 /xm and care was
heater; this requires no pressure correction and taken to avoid data from silicates which had the
pressures are accurate to within 0.5 kbar. Tem- possibility of inclusion of the underlying matrix.
perature was monitored using Pt/Pt90Rhl0 ther- This had the effect of limiting analysis to larger
mocouples and controlled to within I°C of the set grains (> 40/xm in cross-section) of which there
point by Eurotherm 818P solid state controllers. were only a few in each run. The X-rays were
All experiments were furnace-buffered in terms detected using a Si(Li) detector positioned at an
of oxygen activity which, because of the presence angle of 135° with respect to the beam through an
of water in all runs, is 0.5-1.0 log units below Al filter chosen to optimize count rates for the
CCO [16]. Thus we assume a condition in primary detection of the trace elements. The energy dis-
carbonatite melts that is less oxidising than a persive spectra were analysed using techniques
carbonate (graphite absent) assemblage. Experi- developed at CSIRO [19] which correct for pile-up
mental techniques maintain carbonate stability effects, matrix absorption, and secondary fluores-
but may produce traces of graphite in the assem- cence effects. The accuracy of the analytical
blages. Individual run details are summarized in method has been tested [19] against a number of
Table 1. trace element standards (BCR-1, AGV-1, GSP-1),
Run products were analysed using a Cameca as well as a secondary standard [18] containing
SX50 wavelength dispersive 3-spectrometer elec- the trace elements of interest to this study.
tron microprobe (EPMA) at the University of Amongst the elements of interest, Rb, Sr, Y, Zr
Tasmania with major element precisions typically and Nb are well determined by PIXE through
< 3% relative and lower limits of detection in the their K-lines with MDL of < 10 ppm (at the 99%
range 0.02-0.08% for analyses undertaken at 15 confidence level) with counting precisions of bet-
kV and 20 nA (major element oxides). All EPMA ter than 3% relative (at 1 standard deviation). Ta
trace element determinations were made using and Lu were detected through their L-lines and
La-lines at 20 kV. In silicates the beam diameter MDL values are typically 30 and 50 ppm, respec-
was 1-3 /zm, and for trace element determina- tively, with counting precisions of about 10%
tions the filament current was 200 nA and peak relative. It is not possible to detect Ba and L RE E
and background counting times were 200 s. For (Ce) L-lines either because of interference of Fe
quenched carbonate melts multiple areas 50-150 K-lines, or high absorption by the filter. Conse-
/zm in diameter were analysed using a filament quently they are detected through their K-lines,
current of 50 nA for trace elements. In addition but with the much poorer excitation efficiencies
to reduced beam current, a maximum time of 200 (which occurs with increasing atomic number)
s was spent on each area to limit specimen dam- and decreasing efficiency of the Si(Li) detector
 4~

TABLE 2
Starting bulk carbonatite composition (CM1), selected mineral composttlons and corresponding melt compositions

CM1 Minerals Melts

ga3 ga2 amp* amp* amp* titan 5 spin 1 spm 1 ilml
T3506 T3351 T3372 T3404 T3474 T3351 T3372 T3474 T3474 T3506 T3351 T3349 T3509 T3372 T3404 T3474
33kbar 27kbar 27kbar 22kbar 18kbar 27kbar 27kbar 18kbar 18kbar 33kbar 2 7 k bar 22kbar 18k bar 27kbar 22kbar 18kbar
IO00°C IO00°C IO00°C IO00°C 960°C 1000°C 1000°C 960°C 960°C 1000°C 1000°C 1000°C 960°C 1000°C 1000°C 960°C
%
SiO 2 3.17 41.68 40.24 43.61 41.77 41.91 1.83 1.66 1.69 1.29 3.75 3.05 2.64
TIO 2 0.52 0.19 0.70 1.29 1.52 1.76 58.48 0.21 0.46 31.54 0 38 0.45 0 41 0.28 0.65 0.46 0.47
AI20 3 2.10 22.36 21.71 16.79 16.03 15.40 1.26 4089 48.15 1.39 1.29 0.89 0.65 0.59 2.47 1.81 1.16
Cr20 3 0.24 1.16 1.15 0.75 0.56 0.42 1.82 22.14 8.31 1.50 < 0.05 < 0.05 < 0 05 < 0.05 < 0.05 < 0.05 < 0.05
FeO 4.96 11.07 11.54 7.12 7.00 6.99 7.25 14.58 13.59 19.97 4.99 5.43 4.81 4.69 4.59 4.35 5.28
Fe20 3 0.00 4.18 11.59 38.05
MgO 15.17 17.31 16.09 14.83 15.14 16.57 0.64 15.17 17.06 4.71 9.98 10.17 11.85 13.16 11.22 10.63 11.94
CaO 22.59 6.15 7.65 7.73 7.97 9.28 0.57 0.08 040 0.82 16.25 15.54 18.75 17.23 15.23 16.40 17.03
Na20 5.40 0.12 0.18 5.07 4.82 4.41 9.71 12.68 6.55 9.58 6.09 9.12 9.27
K20 0.38 < 0.05 < 0.05 0.62 0.58 0.45 1.33 0.33 0.32 0.25 0.45 0.47 0.29
P205 0.52 0.93 0.94 0.87 0.74 0.86 0.75 0.88
Nb20 s 19~3
Ta205 8.25
Total 55 05 100.04 99.24 97.81 95.39 97.19 98.20 97.25 99.56 97.98 46.69 48.08 45.90 47.81 45.30 47.03 48.96

Mg# 84.5 73.6 71.3 78.8 79.4 80.9 13.6 65.0 69.1 29.6 78.1 76.9 81.5 83.3 81.3 81.3 80.1

ppm
PIXE
Rb 16.1 < 3.3 217 200 161 765 800 532 410 985 660 280
Ba 2321 88.7 < 68 282 344 393 3280 4070 2900 2580 4115 2925 2905
Nb 1316 33.4 9.9 224 179 312 1440 1640 1390 1280 1935 1585 1665
Ta 484 < 18 < 22 202 131 184 295 335 352 274 545 430 415
Ce 3941 184 < 105 < 300 461 449 6070 6970 4777 4537 7045 4980 5410
Sr 13360 285 44.6 1070 2370 1420 15300 16800 12220 11900 17170 13430 14070
Zr 1115 1120 2220 345 318 646 6920 700 640 539 663 1280 955 715
Y 381 824 1290 73.2 101 120 940 400 440 348 356 495 410 435
Lu 95 795 1310 < 37 < 52 < 50 1450 45 <44 60 80 60 61
CM1 also contains 0.65% CI, 0.32% F, 0.86% 503, 4% H 2 0 and 37.54% CO2 (as carbonate). Fe is given as FeO except in splnels and ilmemtes where F e 2 0 3 is calculated
by stoichiometry. Oxides were determined by EPMA and traces by EPMA m t~tan 5 and PIXE in all other minerals; ga = garnet, amp = amphibole, titan = titanate,
spin = spinel, ilm = ilmenlte). rn
Z
* Trace and major elements were determined on &fferent amphiboles with major elements representing an average of analysed grains. Blanks = not analysed.
rn
-]
>
t-
TRACE AND MINOR ELEMENT PARTITIONING BETWEEN GARNET AND AMPHIBOLE AND CARBONATITIC MELT 5

CM1+4% H20
for X-ray energies > 30 MeV, they are less accu- I I I I
rately determined with typical MDL of 70 ppm I 38%
olivine dolomite []
for Ba and 300 ppm for Ce and counting preci-
sions of 20% relative. amp arn ~...\
43% \
3. P h a s e relations 30
\
IT
L?
The phase relations [16] of the 'primary' car- 66% 11}
bonatite of Wallace and Green [6] from 5 to 30
kbar are summarized here. The sub-liquidus as- i [] '°
~"
,x 25
semblage of each run in the CM1 + 2% H 2 0 and v
CM1 + 4% H 2 0 systems, along with an estimate 2
of the melt content (volume%) is given in Table ffl []
=
1. a.
The major phase relations of the CM1 + 4%
2O
H 2 0 bulk composition (Table 2) between 15 and /
33 kbar are given in Fig. 1. Most data were
obtained for this more water-rich system and the
phase relations in the CM1 + 2% H 2 0 system
are very similar [12]. Dolomite was the major 15 [] []
sub-liquidus phase with amphibole stable at pres- 1% ~~98%
sures of 18-27 kbar and olivine at lower pres-
sures (< 18 kbar). Above 27 kbar amphibole was
replaced by garnet as the near liquidus silicate I I I
900 1000 1100
phase. One run at 27 kbar and 1000°C (T3351) in Temperature (oc)
the CM1 + 2% H 2 0 system had both amphibole Fig. 1. Solidus and liquldus for the CM1+4% H 2 0 bulk
and garnet, whereas only amphibole was present composition (Table 2) with boxes indicating the hquidus min-
in the CM1 + 4 % H 2 0 system at these P - T eral assemblage for experimental runs (Table 1) and labelled
conditions. It seems likely that an increase in the with the percent melt present in each case. The shaded box
water content of the system increases the melt represents an experiment which was too fine-grained to deter-
mine the liquidus mineralogy and it is doubtful whether
fraction and eliminates garnet but not amphibole
equilibrium was achieved. Given for reference IS the area of
at 27 kbar and 1000°C. The Wallace and Green carbonatite melt stability In the perldOtlte + H 2 0 + CO 2 sys-
[6] 'primary' carbonatite composition determined tem, bounded by the Hawaiian pyrohte (HP) solidus and the
at 22 kbar and 1000°C occurred with residual amphibole-out boundary in this system determined by Wal-
olivine, orthopyroxene, clinopyroxene, garnet, and lace and Green [6]
ilmenite. If H20, CO 2 and other volatile compo-
nents have been correctly estimated then this
carbonatite composition should have a liquidus clinopyroxene in the parental lherzolite. It is clear
temperature of 1000°C at 22 kbar and the lherzo- also that small changes in the silicate component
lite mineralogy should be reflected on the liq- in CM1 would lead to changes in liquidus roles of
uidus, either occurring in reaction relationship or olivine and garnet. However, neither the reversal
as liquidus phases. The reconnaissance experi- of the Wallace and Green [6] experimental deter-
mental study (Table 1, Fig. 1) shows that this mination of carbonatite melt composition nor the
'reversal' has not been achieved, CM1 having a understanding of the sensitivity of melt composi-
liquidus temperature with 4% H 2 0 of about tion to small changes in volatile components,
ll00°C and an important sub-liquidus field of were objectives in this study. In this study, the
dolomite. Further work including varying H 2 0 Wallace and Green [6] 'primary' composition is
and CO 2 contents of the melt, would be required used to approximate a carbonatite melt system
to evaluate the expected reaction relationships over a range of P - T conditions. Thus, implicit in
between carbonatite melt and orthopyroxene and this study is the assumption that minor changes in
6 R J SWEENEY ET AL

carbonatite melt composition to achieve multiple 4. Analytical results

saturation in lherzolite phases will not affect min-
eral-melt partitioning significantly. The products from runs containing small
The minor sub-liquidus phases include spinel amounts of melt ( < 5%, Table 1) still contained
in the pressure interval 15-33 kbar, Ti-Nb-Ta identifiable amounts of the starting mix oxides
oxide in the interval 22-33 kbar and ilmenite at and therefore these runs did not attain equilib-
P ~< 18 kbar (see examples in Table 2). Zircon is rium. Thus, only in runs containing appreciable
also present at P >/15 kbar where garnet is ab- amounts of melt (> 35%) and where run times
sent. It is significant that saturation of the melt in were sufficiently long (> 24 h) was it likely that
phases such as zircon and Ti-Nb-Ta oxide is equilibrium was achieved. Equilibrium is assessed
achieved with only trace amounts of Nb, Ta and by examining the homogeneity of silicate phases
Zr present in the melt (see equilibrium melt with respect to their major and minor element
compositions in Table 2). content. For example, sequential analysis across a

zoning in garnet 2 from T3351

12- original • • • • • •
CM1 I
10-
o ,
measured I
U. 8-
garnet
6- • --
e
4 I i
-1 O0 0 100 200 300
3000
EP~
I error
2000 '

E
n.
n.
1000 '
e

0 I
O0 0 100 200 300
1400
e
12oo EPkU~
I error

800 £
E
O.
O. 600 2
4O0 e

200 I I
°.
O0 0 100 200 300
Distance (microns)

Fig. 2. Zoning across garnet 2 in T3351 (Table 1) for FeO, Zr and Y. All measurements m the garnet are EPMA determinations as
this techmque enabled the analysis of 1-3 ~ m spot diameters permitting greater resolution than the PIXE technique (20-40/,~m
analysis diameters). The grain boundary is indicated by a vertical and the corresponding melt measurement (PIXE) is plotted at an
arbitrary 50/~m from the boundary. Also plotted are the original abundances in the bulk system (CM1, Table 2).
TRACE AND MINOR ELEMENT PARTITIONING BETWEEN GARNET AND AMPHIBOLE AND CARBONATITIC MELT 7

mineral data
large garnet (about 3 0 0 / z m in diameter) in T3351 2000
demonstrates good within-grain homogeneity in estimated 7 1 1 o (a)
major and minor elements (FeO shown in Fig. 2). error ~ _ ~ o

With respect to their trace element content, how- 1500

ever, there is greater apparent heterogeneity. This
is illustrated in Fig. 2 where the constancy of FeO
may be contrasted with the spread of Zr and Y
data. This spread in data may be due to (1) the
poor precisions of the E P M A trace element de-
t~
~

,,,°-
10o0

500'
/o/ o°
terminations (see error bars in Fig. 2) or (2) a
heterogeneous distribution (on the micron scale
of the electron beam) of these trace elements 0 . . . . , . . . . , . . . .
1000 2000 3000
within these minerals. If the latter explanation
for the spread in data is adopted, this distribution PIXE data ( p p m )

is not caused by disequilibrium conditions (with

melt data
respect to trace elements) as there are high and
low abundances present throughout the grain, i.e. 1500 (b)
this is not a diffusion profile. The better precision
of the P I X E technique will negate problem (1)
and the larger beam size (also giving a larger ~ 1000.
penetration depth) will effectively yield an inte-
grated analysis over most of the grain and negate
problem (2).
tl o i"'""
The determination of element abundances in ~ 500.
LIJ
silicate and carbonate matrices by E P M A and
PIXE is compared in Fig. 3 where abundances
measured by both techniques were above the
0 t
detection limit. Ideally, the data from both tech- 500 1000 1500 2000 2500

niques should fall on a 1:1 correlation line. PIXE data (ppm)

Clearly, with the exception of Ba and Ce (Fig.
melt data
3c), the differences in data obtained on silicate
150O0
minerals and carbonatite melts by the two tech- 1 '1 line (c)
niques are greater than the estimated analytical
error for both. In addition, E P M A determina-
tions are consistently lower than P I X E (or PIXE ~" 10000-
is consistently higher). Analysis of a secondary
glass standard by both techniques produced accu-
"o
rate data [17] and thus it is difficult to assess <
which analytical technique is correct. However, ~ 5000-
UA 4-
the better precision of P I X E and the larger grain
volume analysed (to negate the potential hetero-
geneities discussed above) makes the P I X E data 0
more useful and these are used in the calculation 0 5000 10000 15000 20000

of partition coefficients. PIXE data (ppm)

Fig. 3. A comparison of EPMA and PIXE determmatlons on
5. Mineral-melt partition coefficients minerals (a) and carbonate melts (b and c).

Partition coefficients (Table 4 and Fig. 4) are given in Table 2. The number of mineral determi-
calculated from P I X E mineral data for garnet nations by PIXE was limited by the availability of
and amphibole given in Table 3 and melt data mineral grains greater than 40 ~ m in analysed
8 R J SWEENEY ET AL

100 obtained for olivine as it was assumed that this

e- . . . . .e- - - 27kb (2%H20)

'G
Q)
- - "¢- - 33kb (4%H20) mineral would contain an insignificant budget of
10" basalt-ga / these elements. Where these elements are above
// detection limit, partition coefficients are also cal-
0
u
culated for Ti, Na and K using E P M A data.
t-
o Selected E P M A mineral data are presented in
Table 2; the complete data set is available from
.t the authors upon request.
Q.
Ti, Zr, Y and Lu were compatible in garnet
with the order of compatibility of these elements
E 01 generally Lu > Y = Zr > Ti (Fig. 4a). Ga3 in
r-
'~ \~ (a) T3506 is apparently an exception (Table 3), but
001 considerable variation in Ti content was observed
Ba K Na Nb Ta Ce Sr Zr Ti
' ' ' . . . . . . . Y Lu in garnets from this run (0.19-1.39%) and for an
average of 0.62% TiO 2 the DT, in garnet is then
1.6. Also plotted on Fig. 4a are the g a r n e t - b a s a l t
I 18kb (2%H20)
22kb (2%H20)
melt partition coefficients [20] which display simi-
10
Q) • 27kb (2%H20) lar crystal-liquid partitioning with the most no-
O .... "1---- 18kb (4"/*H20) /~i~ table difference being in Zr behaviour; Z r is
I ......... 22kb¢"%H20 I //Z~\ compatible in garnet in a carbonatite melt but
0
o incompatible in a garnet in a silicate melt.
E
] The amphibole partition coefficients from dif-
0
ferent experiments are quite similar (Fig. 4b).
The order of compatibility in amphibole is Ti > K
> Z r > Ta = Na > Rb > Nb > Y > Sr = Ba > Ce,
~ 1' with Ti and K being the only elements which are
E relatively enriched in amphibole equilibrated with
0
a carbonatite melt (Fig. 4b) (see also reference
.0
(b) [6]). A m p h i b o l e - c a r b o n a t i t e melt D ' s generally
E 01 show an inverse correlation with pressure (Fig.
Rb E~a K l~a 1,4b Ta (~_,e Sr 2~r T, Y 4b). This p r e s s u r e - D inverse correlation may be
Fig. 4. Partihon coefficients for garnet (a) and amphibole (b) explained in general terms by noting that crystal
gwen in Table 4. Where analytical data are below the detec- structures would become progressively more or-
tion limit this detection limit is used to calculate the D. These dered at greater pressures and therefore less able
minimum or maxamum D's are labelled with vectors Basalt-
garnet D's were determined at 25 kbar and 1000°C in a bulk to accommodate element substitution. T h e r e is
composition containing 10% H20 [20]. also a small but systematic change of D with bulk
water added for equwalent pressures: these ele-
ments become slightly less compatible in amphi-
bole with increasing water content in the bulk
cross-section. Except for one exceptional run system (Fig. 4b).
(T3351) where two garnets with diameters of
300-400 /~m crystallized, the n u m b e r of garnets 6. Potential for trace element fractionation in the
and amphiboles with a minimum diameter of mantle by carbonatite melts
> 40 /~m was typically limited to less than 3
grains per run. Excluded from the partition coef- A very large amount of geochemical literature
ficient calculations are the garnets analysed in is devoted to the documentation and explanation
T3478 and T3501 which contain such high Sr of minor and trace element abundances, and iso-
abundances by comparison with the large garnets topic geochemistry of mantle-derived magmas.
of T3351 (Table 4) that some analytical overlap Interpretation of data from magmas requires un-
with carbonatite melt is suggested. D a t a were not derstanding Of the nature of the melting process
 ~q

>

z
o
rn

:z
TABLE 3
>
EPMA (%) and PIXE (ppm) determinations on minerals (see text for analytical details; equilibrium melt compositions are given in Table 2)

Garnet Amphibole
z
CM1 + 2% water system CM + 4% water system CM1 + 2% water system CM1 + 4% water system ~3
T3351 T3478 T3501 T3506 T3509 T3349 T3351 T3474 T3404 T3372
27 kb 27 kbar 30 kbar 33 kbar 18 kbar 22 kbar 27 kbar 18 kbar 22 kbar 27 kbar
ga 1 ga 2 ga 4 ga 2 ga 3 amp 1 amp 4 amp 5 amp la amp lb amp * amp * amp *
% z

TiO 2 0 79 0.70 0.33 0 54 0 19 1 92 1.72 1.86 1 60 1.60 1.76 1.52 1.29

NazO 0 21 0.18 0.17 0.16 0.12 4.40 4.55 4.56 5 21 5.21 4 41 4.82 5.07
K20 < 0 05 < 0 05 < 0 05 < 0 05 < 0.05 0.42 0.55 0.64 0.76 0.76 0.45 0 58 0.62
ppm
Rb < 3.3 < 33 < 44 114 16.1 201 149 200 340 321 161 200 217
Ba < 66 < 68 132 669 88.7 591 184 344 482 289 393 344 282 z
Nb 28 5 9.9 65.8 340 33.4 445 390 460 339 223 312 179 224 o
Ta 31.9 < 22 26.4 74 5 < 18 247 235 262 237 111 184 131 202
Ce 117 < 105 294 1180 184 665 190 461 < 344 338 449 461 < 300
Sr 11.3 44.6 580 3170 285 2333 773 2370 919 1390 1420 2370 1070
Zr 2870 2220 1080 920 1120 604 783 564 615 407 646 318 345
Y 1400 1290 744 671 824 137 124 101 108 89 120 101 73 2 ©
Lu 1360 1310 829 633 795 28 < 23 < 52 < 47 < 37 < 50 < 52 < 37 z

* Trace and major elements were determined on different amphiboles with major elements representing an average of analysed grains. ,.4

t-n
TABLE 4
Mmeral-carbonatlte melt part~tmn coefficmnts

Garnet Amphibole
CM1 + 2% water CM1 + 4% water CMI + 2% water CM1 + 4% water
27 kbar 33 kbar 18 kbar 22 kbar 27 kbar 18 kbar 22 kbar 27 kbar
(T3351, 1000*C) (T3506, 1000°C) (T3509, 960°C) (T3349, 1000°C) (T3351, 1000°C) (T3474, 960°C) (T3404, 1000~C) (T3372, 1000°C)
D (precision) D (precision) D (precision) D (precision) D (prects~on) D (precision) D (precision) D (precision)
T1 1.7 (1.6-1.8) 0.50 (0.47-0 53) 6 1 (5 7-6.5) 4 5 (4 2-4 8) 3.6 (3.4-3.8) 3 7 (3.5-3.9) 3.3 (3.1-3 5) 2.0 (1.9-2 1)
Na 0 016 (0 015-0.017) 0.012 (0.011-0 013) 0.47 (0 44-0 50) 0 70 (0 66-0.74) 0.41 (0 39-0.44) 0.48 (0 45-0 51) 0 53 (0 50-0 56) 0.83 (0 78-0.88)
K < 0.15 < 0.04 2.2 (2 1-2.3) 2 0 (1 9-2 1) 2.3 (2.2-2.4) 1 6 (1.5-1.7) 1 2 (1.1-1 3) 1 4 (1.3-1 5)
Nb 0 012 (0 006-0 017) 0.023 (0 021-0 025) 0.31 (0 29-0 32) 0 33 (0 31-0.35) 0.17 (0 16-0.18) 0 19 (0 17-0 20) 0 11 (0 10-0 12) 0 12 (0 11-0 13)
Zr 3.98 (3 7 - 4 2) 1.60 (1.5-1 7) 1 18 (1 11-1.25) 1 05 (0 99-1.11) 0.80 (0 75-0.85) 0 90 (0 85-0.96) 0 33 (0.31-0.35) 0 27 (0.25-0 29)
Ta 0.095 ( < 0 12) < 0 07 0.86 (0.7-1.1) 0.74 (0 61-0 90) 0.52 (0 43-0 64) 0 44 (0 36-0.54) 0 30 (0 25-0.37) 0.37 (0.30-0 45)
Ce 0.017(<0026) 0030(0.020-0.045) 0042(0.028-0.063)0097(0.06-0.15) 0048(<0.072) 0.083(0.06-012) 0093(0.06-014) <0.04
Lu > 30 17 7 (I4.5-21 6)
Y 3.06 (2 9-3.2) 2 06 (1 9-2.2) 0 348 (0 33-0.37) 0 29 (0 27-0 31) 0 22 (0 20-0 25) 0.28 (0.26-0.29) 0.28 (0.26-0.30) 0 15 (0 14-0.16)
Sr 0 0017 (0.0016-0.0018) 0.0t9 (0.018-0.020) 0 065 (0.061-0 069) 0 194 (0.18-0.21) 0 069 (0 065-0.073) 0.101 (0 09-0 11) 0.18 (0.17-0 19) 0 062 (0.06-0.08)
Ba" < 0 02 0 027 (0.018-0.041) 0 07 (0 04-0.11) 0.12 (0.08-0.18) 0.095 (0.063-0.140) 0.14 (0.09-0.21) 0.12 (0 08-0.18) 0.068 (0.05-0.10)
Rb < 0 004 0.021 (0.020-0 022) 0 36 (0 34-0 38) 0 38 (0.35-0 40) 0 41 (0.38-0 44) 0.58 (0.54-0 61) 0.30 (0.28-0.32) 0 22 (0 21-0.23)

D ~ concentratmn in melt/concentratmn m mineral calculated from Tables 2 (melt) and 3 (mineral). Quoted precislons are calculated from counting stattstlcs (quoted m the text).
Where the D t s a real number and only a maximum value is indicated for the precision means a second mineral determmaUon was below detection hmlt.
TRACE AND MINOR ELEMENT PARTITIONING BETWEEN GARNET AND AMPHIBOLE AND CARBONATITIC MELT 11

and particularly of the melt/residue characteris- atite melt metasomatism) and a depleted mantle
tics (nature and properties of phases, surface vs reservoir (region of carbonatite melt extraction).
bulk equilibrium, etc.). The inadequacy of single- In stable, intraplate regions, the nature of peri-
stage melting processes to produce the observed dotite-CO2-H20 solidi [6-11,16,21] are such that
fractionations among major, minor and trace ele- upward percolation of low-melt-fraction, carbon-
ments from a single well-mixed mantle source ated silicate melt (olivine-rich basanite to olivine
composition and, most importantly, the observed melilitite) may lead to lithospheric metasomatism
ranges of isotopic ratios (Sr, Pb, Nd, O, Be) in above ~ 90 km (30 kbar) [6,7,22,23]. In this pro-
mantle-derived magmas has led to the concept of cess,, the lower lithosphere ( ~ 90 to ~ 60 km)
mantle heterogeneity and attempts to correlate becomes enriched by access of extremely under-
particular geochemical signatures with tectonic saturated silicate melt, crystallization of titanian
settings and to models of magma genesis tapping pargasite _+ phlogopite, and access of sodic,
particular mantle reservoirs. dolomitic carbonatite melt. At and above ~ 60
This paper addresses one process which may km, the lithosphere may be enriched by access of
produce mantle inhomogeneity, i.e. the formation sodic, dolomitic carbonatite melt and decarbona-
and migration of a carbonatitic melt leading to tion reactions to yield apatite and pargasite-
mantle heterogeneity with possible formation of bearing lherzolite trending towards wehrlite
an 'enriched' mantle reservoir (region of carbon- [7,12]. Other scenarios for carbonatite melt for-

. . . . aoo . . . . . . -- I /17. / /',63co2+,~o . ~ - - - -

------ 900 . . . . . - - ~ I

- - - - 1200

/ ./I ~/" Solidus (alhcate melt)

~,zoO ~ ------ $ohdus (carbonate melt)

SCALE V-H

Fig. 5. Schematic model showing the potential influence of volatile fluxing ( H 2 0 + CO 2) off the slab on melting processes and
mineralogy of the wedge above a subducting slab. This volatile fluxing lowers the peridotlte solidus m the wedge significantly, which
then becomes a likely site for melting and diapir formation. The thick arrows represent movement of carbonatite melt from a
region where a carbonatite melt is stable, into regions of silicate melt formation. These metasomatlc carbonatlte melts may have
been equilibrated (above the p e r i d o t i t e + H 2 0 + C O E - S a t u r a t e d solidus [21]) with amphibole+garnet (cross-hatched region) or
garnet (hatched region). This is discussed fully in the text. The temperature contours are schematic but do reflect the availability of
several geophysical models, most specifically in the depiction of a temperature inversion above the slab [e g. 25].
12 R J S W E E N E Y E T AL

mation and migration, in equilibrium with resid- 7. Geochemical signatures of depletion and en-
ual spinel or garnet, pargasite-bearing lherzolite, richment by carbonatite melts in the upper
can be modelled within the depth range of 60-90 mantle
km for appropriate wall-rock and mantle diapir
concepts or lithosphere stretching and upwelling The previous discussion presented intraplate
models. and convergent margin scenarios in which addi-
In Fig. 5, a schematic model of a subduction tion or extraction of carbonatite melt could deter-
environment with pronounced thermal inversion mine minor and trace element contents of differ-
above the down-going slab is depicted. The posi- ent mantle source regions, and thus of silicate
tions of the various solidi for peridotite + H 2 0 + magmas extracted from such source regions. We
CO 2 are taken from Falloon and Green [21]. For emphasize that carbonatite melts would not be
instance the silicate melt solidus for pyrolite + invariant in composition over the P,T ranges of
H 2 0 + CO 2 (fluid-undersaturated) is appropriate interest depicted in the scenarios above. How-
for regions away from the subducting slab, giving ever, the use of the sodic, dolomitic carbonatite
depths to the silicate solidus of about 90 km for a melt composition and the relatively large role
geotherm passing through ll00°C at this pressure played by pargasitic amphibole and garnet in
(Fig. 5). Closer to the slab where the volatile flux hosting and fractionating the trace elements of
ensures fluid saturation, the peridotite + H 2 0 + interest are appropriate to provide a first-order
CO 2 (fluM-saturated) silicate solidus at 970- evaluation of the potential of the process.
1000°C between depths of 70 and 100 km is more The garnet-carbonatite melt partition coeffi-
appropriate (Fig. 5). The shaded area at higher cients (Table 4, Fig. 4a) imply that separation of a
temperatures than the fluid-saturated silicate carbonatite melt from garnet lherzolite (27-35
solidus generation nearest the slab (dashed line kbar) would not fractionate Zr or Y relative to
in Fig. 5) depicts the region where a carbonatitic each other, i.e. Z r / Y in residual or enriched
melt may be in equilibrium with lherzolite con- regions would not change. This contrasts with the
taining garnet (lined) and amphibole (cross- results obtained for the equilibration of a
hatched). Because of the inverted temperature basanitic melt with garnet (25 kbar) [20] where
distribution there is a zone of carbonatite melt Dzr = 0.4-0.7 and D v = 2.5-9. In the intraplate
spatially beneath the region of silicate melt. This lithosphere scenario described above, a lower
melt would be formed by the reaction of CO 2 lithosphere enriched by silicate melt metasoma-
(and H20) from the slab with the overlying peri- tism tO titanian-pargasite bearing lherzolite would
dotite assemblage to produce a carbonate melt. have a higher Z r / Y ratio than its source
This carbonatite melt may coexist with a water- (asthenosphere) and particularly leave residual
rich fluid at temperatures between 930°C and lherzolite depleted (asthenosphere) with low
1000°C [6,7,21]. We suppose then that a carbon- Z r / Y ratio. Movement of a carbonatite melt
atitic melt moves from this region up-tempera- within the lithosphere from ~ 90 km to ~ 60 km
ture into the mantle wedge (heavy arrows), cross- would not change the Z r / Y ratio, but would
ing the silicate melt (peridotite + H20-rich fluid) decrease Zr/Nb, L u / Y , Z r / L I L E , T i / L I L E and
solidus at ~ 1000°C as the core of the wedge is Y / L I L E in the enriched or shallower lithosphere
approached. In this hotter region the degree of (trending to wehrlite) and increase these ratios in
melting is sufficiently large for diapir formation the lower lithosphere (garnet lherzolite with tita-
and this melt segregates and ascends (diapir in nian pargasite).
Fig. 5). Although the carbonatitic component in There is a substantial role by garnet/carbona-
the wedge-derived melts is very small, the incom- tite equilibria and a significant but lesser role for
patible trace element concentrations are high (e.g. amphibole/carbonatite partitioning in fractionat-
Nb, Ta, Rb, Sr, Ba, LREE) in this component ing Zr and Nb. Dzr is always greater than D ~
relative to the 'depleted lithosphere character' of with the effect being most pronounced at lower
the wedge. The effect of a volumetrically small pressures (18 kbar, Table 4, Fig. 4b). The process
carbonatite component on some trace element envisaged would also produce shallower litho-
ratios would be significant. sphere with lower Ti/Nb and higher LILE/Ti,
T R A C E AND MINOR ELEMENT P A R T I T I O N I N G BETWEEN G A R N E T AND AMPHIBOLE AND CARBONATITIC MELT 13

L I L E / Z r , L I L E / Y and Y / L u ratios. For these riched in LILE including LREE but Ti, Zr, Lu
effects to be quantified, modes of peridotite com- and HREE would remain in amphibole a n d / o r
positions and extent of metasomatic additions garnet lherzotite residue. Silicate melts produced
would be required, together with information on would probably be fluid-undersaturated at depth
clinopyroxene/carbonatite melt partitioning. but on moving to shallower depths (< 30 km) the
Nevertheless, garnet and amphibole, if they occur rapidly decreasing solubility of CO z with pressure
as modally significant phases (> 5%) in residual would cause exsolution of COe-rich fluid.
lherzolite, will exert the major control on frac- McCulloch and Gamble [26] noted that Z r / N b
tionation among the HFSE and LILE elements (for similar Nb values) was significantly greater in
and within the REE group. MORB than in IAB--this is consistent with en-
In the subduction zone scenario of Fig. 5, the richment of the IAB source by carbonatite melt
source of the primary silicate magmas, including containing significant Nb and a low Z r / N b ratio.
boninites, island-arc tholeiites and island-arc pi-
crites (broadly island-arc basalts) is envisaged as 8. Summary
the high-temperature 'core' of the mantle wedge
overlying the Benioff Zone. Primary magmas of Wallace and Green [6] demonstrated that it is
island arcs contain higher water contents and are possible to maintain a carbonatite melt of
more oxidized than MORB or OIB but they are dolomitic composition in equilibrium with a lher-
not water-saturated at high pressures, i.e. depths zolitic mantle. Using this estimate of primary
of magma segregation. The major element geo- carbonatite magma it has been possible to obtain
chemistry and particularly the mineralogy of liq- mineral/melt pairs for garnet/carbonatite and
uidus phases including olivine t o F094 , spinel to amphibole/carbonatite at pressures of 18-33
(100Cr/Cr + AI) = 90 and clinopyroxene of Mg93 kbar. Partition coefficients were determined from
composition, all indicate source peridotite of a analysis (using electron probe (EPMA) and pro-
very refractory nature [24]. However, Na/Ca, ton probe (PIXE) techniques) of co-existing car-
K / C a ratios and incompatible element abun- bonatite melt and the silicate minerals amphibole
dances are variable and not typical of residual, (18-27 kbar) and garnet (27-33 kbar) which crys-
refractory sources. This contrast is conventionally tallized at 960-1000°C. These data are given in
explained by addition of incompatible-element Table 4. Except for some subtle changes in am-
enriched components from the subducted slab, phibole-carbonatite melt partitioning with pres-
including H20-rich fluid or fluid-rich silicate melt sure, inter-element fractionations predicted by
(rhyodacitic melt) from the basaltic (eclogitic) the partition coefficients remain the same: the
component of the slab. However, primary island- spidergram patterns are similar for each mineral
arc basalts commonly contain 1.5 to 3.0 wt% (Fig. 4) but differ considerably between garnet
Na20 and up to 2.0 wt% K20, and if this were and amphibole. Thus we may compute averages
delivered to the source region in H20-rich fluid for partition coefficients. For garnet-carbonatite
then water contents in the melts in excess of 10% melt these are Ti (1.55), Nb (0.018), Zr (2.8), Ta
would be expected. Water-saturated melting (< 0.12), Ce (< 0.024), Lu(> 17), Y (2.6), Sr (.01),
would be required. Alternatively, two-stage en- Ba (< 0.03), Na (0.014), K (< 0.15) and Rb (<
richment by 'pre-conditioning' of mantle wedge 0.03). For amphibole-carbonatite melt these are
peridotite by low-temperature fluids would pro- Ti (3.9), Nb (0.21), Zr (0.76), Ta (0.54), Ce (<
duce talc + serpentine + chlorite + amphibole as- 0.07), Y (0.26), Sr (0.10), Ba (0.12), Rb (0.38), Na
semblages containing > 10% H20. Thus melting (0.57) and K (1.8).
of this pre-enriched source at deeper levels where These partition coefficients may be used to
these hydrous phases are still stable, would also predict inter-element fractionations between a
produce water-saturated melts. source region and a carbonatite melt. The mobil-
The scenario of Fig. 5 illustrates the potential ity of this melt in mantle environments [5] and
role of carbonatite melt in transporting incompat- the significant budget of some trace elements in a
ible elements into the silicate melting region of primary carbonatite (e.g. Nb, Ta, Rb, Sr, Ba and
the mantle wedge. Melts produced would be en- LREE) suggested by natural data [14], makes this
14 R J SWEENEYET AL

a potentially important agent of mantle metaso- 11 W.-L. Huang and P.J. Wyllie, Carbonation reactions for
matism. A region where such processes may be mantle lherzolite and harzburgite, in: Proc. 27th Int. Geol.
Congr., Moscow, 9, pp. 455-473, 1984.
significant is in the mantle wedge above subduc-
12 Yaxley, G.M., A J. Crawford and D.H. Green, Evidence
tion zones as the source region in which different for carbonatite metasomaUsm in spinel peridotite xeno-
convergent margin (island arc) primary magmas liths from western Victoria, Australia, Earth Planet ScL
are formed. Lett. 107, 305-317, 1991.
13 R.J. Sweeney, T.J. Falloon and D.H. Green, Experimental
constraints on the possible mantle origins of natrocarbon-
Acknowledgements atlte, IAVCEI Spec. Publ., submitted
14 A.R. Woolley and D.R.C Kempe, Carbonatites: nomencla-
We thank Simon Stephens for careful work ture, average chemical composittons and element d~stribu-
tion, in: Carbonatltes, K. Bell, ed., pp. 1-14, Unwin Hy-
preparing polished sections and Keith Harris and
man, 1989.
Wislaw Jablonski for technical assistance. Trevor 15 J. Keller and M. Krafft, Effusive natrocarbonatite activity
Green and Simon Peacock are thanked for useful of Oldomyo Lengai, June 1988, Bull. Volcanol. 52, 629-
comment and discussion. Gerhard Brey and 645
Stephen Foley are thanked for constructive re- 16 W.R. Taylor and D.H. Green, The petrogenetic role of
methane: effects on liqmdus phase relations and the solu-
views. The study was supported by the Australian
blhty of reduced C-H-O volatdes, m: Magmatlc Processes:
Research Council and a South African FRD fel- Physiochemical Principles, B.O. Mysen, ed., Geochem.
lowship (RJS). Soc. Spec. Publ. 1, 121-138, 1987.
17 R.J. Sweeney, W. Jablonskl and S. Sic, DetermmaUon of
selected trace elements in silicates by electron probe mi-
References croanalysis, in prep.
18 S.H. Sle, An accelerator facility within a mineral research
1 P.G. Harris, Zone refining and the origin of potasslc facdlty, Nucl. Instr. Methods Phys. Res. B 1 0 / l l , 664-670,
basalts, Geochim Cosmochim. Acta 12, 195-205, 1957. 1985.
2 D.H. Green and A.E. Ringwood, The genesis of basaltic 19 C G Ryan, D.R. Cousens, S.H Sic, W.L. Griffin, G F
magmas, Contrib. Mineral. Petrol. 15, 103-190, 1967. Suter and E. Clayton, Quantltatwe PIXE microanalysis of
3 F.E. Lloyd and D.IC Bailey, Light element metasomatism geological materials using the CSIRO proton microprobe,
of the continental mantle: the evidence and consequences, Nucl. Instr. Methods Phys. Res. B47, 55-71, 1990.
Phys. Chem Earth 9, 389-416, 1975. 20 T.H. Green, S.H. Sic, C.G. Ryan and D.R. Cousens,
4 D. McKenzie, Some remarks on the movement of small Proton-microprobe determined partitioning of Nb, Ta, Zr,
melt fractions m the mantle, Earth Planet Sci. Lett. 95, Sr and Y between garnet, chnopyroxene and basaltic melt
53-72, 1989 at high pressure and temperature, Chem. Geol. 74, 201-
5 E.B. Watson, J.M. Brennan and D.R. Baker, Distribution 216, '1989.
of fluids m the continental mantle, in: Continental Mantle, 21 T.J. Falloon and D.H. Green, Solidus of carbonated fertile
M.A. Menzies, ed., pp. 111-125, Clarendon Press, 1990. pendotite under fired saturated conditions, Geology 18,
6 M.E. Wallace and D.H. Green, An experimental determi- 195-199, 1990.
nation of primary carbonatite magma composition, Nature 22 Green, D.H., ComposiUon of basaltic magmas as indica-
335, 343-346, 1988. tors of conditions of origin' apphcatlon to oceanic volcan-
7 D.H. Green and M.E. Wallace, Mantle metasomatism by ism, Phiios Trans. R. Soc. London Ser. A 268, 707-725,
ephemeral carbonatite melts, Nature 336, 459-462, 1988. 1971.
8 P.1. Wyllie and W.-L Huang, Carbonation and melting 23 Green, D.H. and R.C. Liebermann, Phase equihbna and
reactions in the system CaO-MgO-SIO2-CO2, Geology elastic properties of a pyrohte model for the oceanic upper
3, 621-624, 1975 mantle, Tectonophysics 32, 61-92, 1976
9 P.J Wylhe and W.-L. Huang, Carbonation and melting 24 Eggins, S.M., Origin and differentiation of picritic arc
reactions in the system CaO-MgO-S~O2-CO 2 at mantle magmas Ambae, Vanu Atu. Contrib. Mineral. Petrol., in
pressures with geophysical and petrological apphcations, press, 1992
Contrib. Mineral. Petrol. 54, 79-107, 1976. 25 S.M. Peacock, Flmd processes in subduction zones, Sci-
10 P.J. Wylhe, W.-L. Huang, J. Otto and A.P. Byrnes, Car- ence 248, 328-337, 1990.
bonation of peridotites and decarbonation of sihceous 26 M T. McCulloch and J.A. Gamble, Geochemical and geo-
dolomites represented in the system C a O - M g O - S i O 2 - dynamical constraints on subduction zone magmatism,
CO 2 to 30 kbar, Tectonophysics 100, 359-388, 1983. Earth Planet. Sci. Lett. 102, 358-374, 1991.