American Mineralogist, Volume 64, pages 274-287, 1979

Trace-element partitioning between garnet peridotite minerals and water-rich vapor:experimental data from 5 to 30 kbar

BldnN O. MvseN

Geophysical Laboratory, Carnegie Institution of WashingtonWashington, D. C.20008

Abstract

The values of individual rare earth element (REE) partition coefficients (weight ratio,Cfigg9"/CfiHit"r : l(rffi)r-crv"t'r) range between I and 200 lor garnet, clinopyroxene, ortho-pyroxene, and olivine in equilibrium with HrO-rich vapor under experimental conditionscorresponding to those of the upper mantle. The Kfift$o"-"'v"tur increases rapidly with increas-ing PH,O (e.9., 1qf,Po"-cp' = 0.01, 3, 12 and 31, respectively, at 5, 10, 20, and 30 kbar).Calculated bulk partition coefficients (as defined by Shaw, 1970) range from 5 to 250 for anaverage garnet peridotite assemblage in the pressure range 5-40 kbar, with D|$por-rock >D{gnor-rocx

The portions of the upper mantle sampled and brought to the surface by kimberlite magmacharacteristically display garnet contents ranging from near zero to about 20 percent. If a suiteof such garnet peridotites equil ibrated with HzO-rich vapor, a wide range of absolute REEcontents and degree of l ight REE enrichment would result. It is notable that the more depletedgarnet peridotite nodules in kimberlite (also the most garnet-poor) commonly show greatlight-REE enrichment (CelYb > l0), as would be expected if these rocks have sufferedmetasomatic alteration.

Andesites in island arcs commonly show considerable l ight-REE enrichment (Ce/Yb :2.5-

10). It is argued here that the peridotite wedge overlying a descending plate ofoceanic basalticcomposition with an eclogitic mineral assemblage would attain considerable light-REE en-richment as a result of equilibration with HrO-rich vapor released by dehydration of thematerial in the descending slab. If, for example, 0.5 weight percent HrO was released andinitially equilibrated with an eclogitic rock before migrating into the overlying peridotite, theperidotite (with an initially unfractionated REE pattern) could attain CelSm and CelTmratios near 3 and 5 respectively. Partial melting of this hydrated peridotite would result inmelts of andesitic bulk composition with CelSm : 3-4 and Ce/Tm - 10, closely resemblingthe REE patterns found in andesites in island arcs.

Introduction

The presence of volatile components in the uppermantle (H2O, CO2, S, etc.) has been documented by,for example, Oxburgh (1964), Anderson (1975), andMuenow et al. (1977). (For review of other data seealso Irving and Wyllie, 1975, and Boettcher et al.,1975.) As a result of these observations, extensiveexperimental work has been carried out on the effectsof volatiles on phase relations of upper mantle mate-rials (e.g., Yoder, 1969; Kushiro, 1969, 1972; Kushiroet al., 1968; Eggler, 1973, 1974, 1975, 1976;' Huangand Wyllie, 1976; Mysen and Boettcher, l975a,b).With a few notable exceptions (e.g., Rubey, 1955),

w03 404x / 7 9 /0304-0274$02.00

relatively little attention has been drawn to the sourceof volatiles in the upper mantle, to their migrationrates, and to the transport capacity of such volatilesfor silicate components.

Evidence from volcanic rocks suggests hetero-geneous distribution of volatiles in the source regionsof partial melts in the upper mantle. Mid-oceanicridge basalts appear to be essentially HrO-free, forexample (Delaney et al., 1977), whereas volcanicrocks in island arcs contain several percent HzO (e.9.,Anderson, 1975; Muenow et al., 1977). Kimberlitecontains large amounts of both COz and HzO.

A consequence of heterogeneous distribution of

274

MYSEN: TRACE-ELEMENT PARTITIONING 275

volatiles in the source region of melts in the uppermantle is the possibility of migration of these vola-tiles. Unfortunately, experimental data bearing oninfiltration of fluids in crystalline upper mantle arescarce. Marsh (1976a) suggested that HrO may beessentially immobile, whereas preliminary experi-mental data of Mysen et al. (1977) indicate that anHrO-rich front may infiltrate crystalline mantle mate-rials at the rate of several millimeters per hour atpressures above l5 kbar. Thus, an HrO-rich phase (ascontrasted with HzO bound in crystalline materials orsilicate liquid) may act as a transporting agent ofsilicate components in the upper mantle, providedsignificant amounts of silicate components can dis-solve in such fluids.

Some evidence suggests large solubilities (severaltens of a weight percent) of major elements in HrO-rich fluids under pressure conditions correspondingto those of the upper mantle (Burnham, 1967; Naka-mura and Kushiro, 1974; Eggler and Rosenhauer,1978). Cullers et al. (1973) and Zielinski and Frey(1974) measured partitioning of REE between HrO-rich vapors and some peridotite minerals at l-2 kbar ,and concluded that the REE were strongly parti-tioned into the crystals under the low-pressure condi-tions of their experiments. Some major-element dataof Kennedy et al. (1962), Burnham (1967), and Egg-ler and Rosenhauer (1978) indicate that the solubilityof silicate components in HrO-rich fluids dependsstrongly on pressure. The low-pressure data of Cul-lers et ql. (1973) and Zielinski and Frey (1974), there-fore, may not be applicable to upper mantle pres-sures.

In view of the importance attributed to knowledgeof the trace-element contents of potential parentalrocks of partial melts in the upper mantle, an under-standing of the processes governing their distribution

patterns is needed. Consequently, partition coeffi-cients involving a water-rich vapor and the constitu-ent minerals of garnet peridotite have been deter-mined for representative REE as a function ofpressure. These data were then used to evaluate therole of metasomatic alteration of the distribution pat-terns of these elements in the upper mantle.

Experimental technique

Starting materials were synthetic pyrope (Ga) anddiopside (Cpx) either with or without the trace ele-ment in solution. The starting material was ground toless than 5 pm grain size and run in sealed Pt capsulestogether with HrO containing a radioactive trace ele-ment. Reversal experiments were carried out by tak-ing a split of an experimental run product, washing itin alcohol, drying it, and running it together withtracer-free, double-distilled HrO in a sealed Pt cap-sule. The minerals grew to about l0 pm grain sizeduring a forward experiment, so the starting materialwas reground to less than 5 pm before the reversalexperiment was carried out. Reversal experiments aremarked with asterisks in the run table (Table l).

All experiments were conducted at I100"C in thepressure range 5-30 kbar. An internally-heated gas-media apparatus (Yoder, 1950) was used at P < l0kbar, whereas a solid-media high-pressure apparatus(Boyd and England, 1960) was used at higher pres-sures. Pressures in the solid-media high-pressure ap-paratus were corrected by -4 percent as calibratedagainst the quartz-coesite transition.

Vapor-crystal partition coefficients were deter-mined by analyzing the crystals and determining thetrace-element content of the coexisting vapor by massbalance. Cerium, samarium, and thulium, respec-tively, were chosen as representative light, inter-mediate, and heavy REE. Analyses were conducted

Table l . Experimental results at l l00.C

Run no.Pressure ,

kbarRun duraEion,

hourswt "a Hzo Crystal and

elenentConcentration, ppn

Kl{2O-crystalVapor Crys ta l

615616*620623*66L

959 692 . 5969 6

98L2L

v o . )

938 9

1 2 . 6 023 .L9L 2 . 8 415 . 00

6 . 6 6

4 . 0 84 . 3 69 . 4 69 . 6 L

18 . 83

Dl (Sm)ua (smJDi (Tn)Dr (rm)u l ( 5 m J

u a t S m JD i ( S n )

Py (sn)Py (rm)Py (ce)

6626646 7 3679690

2020202030

510202020

16011537!4L8!2

4 . 4 ! 0 . 4254!I8

0 . 25 !0 .25501364!6

0 .104 !0 .0078180r200

I4!L2 . 8 ! 0 . 21 . 0 i 0 . 1

0 . 2 3 1 0 . O 28 . 3 1 0 . 5

1 7 . 4 ! 0 . 914 . 610 . 5

32!30 . r 89 !0 . 011

/ r r o

1 1 1 1 . 3L3!2L8!2.5L9!2.4311 3

0 . 0110 .013 . 4 ! O . 22 . 0 r 0 . 3

0 . 5510 . 051151r0

*Reyersa-l experinent (trace eLffient in crgstal prior to expetiment).

276 MYSEN: TRACE-ELEM ENT PARTITIONING

with the beta-track mapping technique (Mysen andSeitz, 1975), using cer ium-141, samarium-151, andthulium-171 as sources of beta particles. Standardsfor the trace element analyses were NaAlSisOr glasseswith known amounts of trace element added.

The theoretical uncertainty (exposure time and sta-tistical uncertainties) of an analysis of a mineral (lo)is approximately 4 percent (relative). Most analysesof minerals from experimental runs have uncer-tainties within this limit. It is concluded, therefore,that homogeneous grains were obtained with the runduration of 4 days used in these experiments. Theweighing error in calculating the trace-element con-tent of the vapor by mass balance (including loss offluid during welding) is less than 5 percent (relative).Consequently, the analytical uncertainty of a vapor-crystal partition coefficient determined by thismethod is of the order of 6 percent (1o).

The results (calculated as partition coefficients) ofthe reversal experiments with diopside were generallywithin l0 percent of the original value. A minimumvalue for the diffusion coefficient of Sm in diopside atI 100'C can be estimated from the run length andgrain size of the crystals (- 10 pm). The coefficient isequal to or greater than l0-15 cm2,/sec. This valuecompares with about 10 'n cm2/sec calculated by Zie-linski and Frey (1974) for Gd3+ in diopside at 800oC.Diffusion coefficients of the individual REE in garnetperidotite minerals are unlikely to vary by more thanan order of magnitude. It is concluded, therefore,that because 4 days run duration was sufficient toattain equilibrium with Sm in diopside, the same runduration is also sufficient to attain equilibrium withthe other REE in diopside and other garnet peridotiteminerals. This conclusion is further substantiated bythe observation that all mineruls were composition-ally homogeneous after an experiment.

Partition coefficients for olivine-vapor and ortho-pyroxene-vapor were calculated from REE clinopy-roxene-olivine and clinopyroxene-orthopyroxenepartition coefficients derived from the data of Mysen(1977a), which were obtained at 20 kbar in the tem-perature range 960o-1075"C. Details of this calcu-lation are described below. The temperature-depen-dence of these partition coefficients is small in thetemperature range considered, and no temperaturecorrection was applied to the values when the data ofMysen (1977a) were extrapolated to I100"C. Therelative uncertainty of the partition coefficients ofMysen (1977a) is about 5 percent (lo), resulting inan estimated relative uncertainty of the calculatedvapor-crystal partition coefficients, K$bo;ot andrfiio;oo', of about 8 percent.

Another important aspect of the experimentsreported here is the solubility behavior of the min-eral in HzO-rich vapor. The experimental part of thisstudy involved pyrope (MgrAlrSiO'r) and diopside(CaMgSLOo). Eggler and Rosenhauer (1978) re-ported that even though l0-15 weight percent CaMgSizO. was dissolved in HzO in the pressure nnge 20-30 kbar, no evidence was found for incongruent solu-tion of diopside in the vapor in this pressure range.Electron microprobe analysis of diopside after anexperiment in my study did not reveal non-stoichiometry, thus supporting the conclusion ofEggler and Rosenhauer (1978). Less information isavailable on the solubility behavior of pyrope inHrO-rich vapor at high pressure. The starting mate-rial (kindly supplied by R. C. Newton, University ofChicago) was crystallized under hydrothermal condi-tions at 1000'C and 28 kbar. Electron microprobeanalysis of this starting material as well as analysis ofthe pyrope after an experiment did not show non-stoichiometry within the uncertainty of the analysis(2-3 percent, relative). Furthermore, optical exami-nation of the run products indicated that pyrope wasthe only stable crystalline phase present. I conclude,therefore, that both diopside and pyrope dissolvecongruently in coexisting HzO-rich fluid under thepressure and temperature conditions of the presentexperiments.

Results

Experimental results are listed in Table l,in which reversal experiments are marked withasterisks. A comparison of values for K!'ro;"tv"t"t(:cH;Vcfitftar) for various garnet peridotite min-erals at 20 kbar and ll00"C is shown in Figure l.Two features of the data warrant comment. First,with the exception of K$go-c", the REE are parti-tioned into the HzO-rich vapor relative to the crys-talline phases. Second, the partition coefficientsvary with the atomic number of the REE.

The pressure-dependence of the vapor-crystalpartition coefficients has been evaluated for diop-side using Sm at I100'C (Fig. 2). The uncertaintyof the results derived with the mass-balancingmethod used to determine the trace-element con-tents of the vapors increases as the trace-elementconcentration in the vapor decreases. For the ex'perimental technique used here, the sensitivity cor-responds to K$go;crvstat 0.01 in the pressurerange of the internally-heated gas-media apparatus.Because of the smaller volume used in the solid-media high-pressure apparatus, experiments withthat apparatus result in even lower sensitivity (cor-

MYSEN: TRACE.ELEMENT PARTITIoNING

1 0

o

olr:.t-(r:<

0 1 0 2 0 3 0P r e s s u r e k b o r

Fig. 2. Vapor-diopside partition coefficients for samarrum as afunct ion of pressure at I 100'C. Error bars, * lo.

pressure-dependence of crystal-crystal partitioningis also necessary for extrapolation of KfiS-".""r't tosuitable pressure and temperature conditions. Theonly data on the pressure-dependence of REE crys-tal-crystal partitioning in the pressure range consid-ered here (corresponding to the upper mantle) arethose of Mysen (1977a) for coexisting olivine andaluminous enstatite at l0 and 20 kbar. No depen-dence was seen. The following calculations assumethat other crystal-crystal partition coefficients(K€Bb t', K€ffi oo', and Kfifi;cnx) do not vary in thepressure range 5-30 kbar. This assumption may notbe completely justified, in view of the large volumechanges that probably accompany solution of REEin garnet peridotite minerals relative to the REE-free analogs. It is probable, however, that whateverthe relative pressure-dependence of Kft';vE"t"r - crvstarmay be, it is the same for all REE because of theirsimilar size and valence. Consequently, if an error

Table 2. Maximum values of Ce, Sm, and Tm partitioncoefficients at I l00oC and 5 kbar

0 l iv ineOrthopyroxeneClinopyroxene

L o P r P m E u T b H o T m L uC e N d S m G d D y E r Y b

Fig. l. Vapor-crystal partition coefficients for garner,c l inopyroxene, or thopyroxene, and ol iv ine at 20 kbar and I 100.C.Symbols in parentheses represent calculated partition coefficients.Error bars, 1 lo.

responding to Kf;6$-"'v"tur - 0.1). The data in TableI show that no Sm was detected in the vapor at 5kbar. Maximum values for other REE partition co-efficients at 5 kbar are listed together withKf,-dionsiau in Table 2. A very strong pressure-de-pendence of the Sm vapor-crystal partition coeffi-cients is evident from the data in Figure 2. Thelow-pressure data are comparable to those of Cul-lers e/ al. (1973) and Zielinski and Frey (1974) atpressures below 5 kbar. There are no other pub-lished experimental data with which to compare thepresent results at pressures corresponding to thoseof the lower crust and upper mantle.

The data shown in Figures I and 2 may be usedto calculate vapor-crystal partition coefficients forother REE and other garnet peridotite minerals,provided the individual REE are not fractionatedby the vapor and REE crystal-crystal partition co-efficients are available for the minerals and REE inquestion. Information on the temperature- and

0 . 20 . 0 5o . 0 2

0 . 10 . 0 40 . 0 1

0 . 0 80 . 0 6o . 0 2

278

is introduced because of this assumption, it affectsonly the absolute abundances and not frationationof REE. It is likely, therefore, that the conclu-sions reached here (see below) will not be signifi-cantly altered after the pressure-dependence ofK6Y;tat-c'""t'r becomes known,

If there is no fractionation of the individual REEin an HrO-rich vapor, data on one Kf;fio;"'v"t"r andthe REE crystal-crystal partition coefficients for theother minerals can be used to calculate otherKfr;$-"'v"t*t. Consequently, if two sets of vapor-crystal partition coefficients are known and one ofthese sets can also be reproduced by calculationbased on the other set, it can be concluded that thevapor does not fractionate between the individualREE. The results of such an exercise are shown inTable 3, using KSf-c" and the data of Mysen(1977a) for KEBb-G". The comparison of these re-sults suggests that the individual REE are not frac-tionated by HrO-rich vapor.

From the above considerations, it is tentativelyconcluded that the strong positive pressure-depen-dence measured for K!;o-aton"ra" is caused by theincreasing solubility of REE in HzO-rich vapor,and that the same pressure-dependence is to be ex-pected for other REE when other minerals are con-sidered. It may be argued that the REE in the va-por may form complexes with cations from themajor-element silicates. No data are available insupport of this suggestion. This possibility, there-fore, is not taken into account in the following dis-cussion. It is then possible to calculate all REE va-por-crystal partition coefficients as a function ofpressure, provided that the different solubilities ofmajor element components in the vapor are ne-glected. The neglect of major elements dissolved inthe vapor is justified in view of the fact that themole fraction of HrO in the fluids is of the order of0.94-0.98, depending on whether the dissolved sili-cate components ate calculated as oxides orCaMgSirOu, respectively. It is assumed that thissmall silicate impurity in the vapor does not signifi-cantly affect the REE solubility in the fluid. The

Table 3. Comparison of measured and calculated K.nzo-diorstoeat I 100'C and 20 kbar

MeasuredCalculated

MYSEN: TRACE.ELEMENT PARTITIONING

L o P rCe

c P x

io Pr Pm Eu lbCe Nd Sm Gd 0y

Ho Tm LuEr Yb

1 0

P m E u T b H o T m L uNd Sm Gd Dy Er Yb

T bG d

O L

' \ 99''l

L o P . P m E u T b H o T m L uCe Nd Sm Gd Dv Er Yb

C e N d

H o T m L u

0 y E r Y b

Fig. 3. Pressure dependence of 6fr6o-"'v"t"t at l000oC forolivine, orthopyroxene, clinopyroxene, and garnet. Error bars,* l c .

results of such calculations are shown in Figure 3.The 4O-kbar curves are based on an extrapolation of16ff-aronstae to 40 kbar (see Fig. 2). The results ofthese calculations show that a water-rich fluid inthe upper mantle constitutes a major sink for REE.This observation is further substantiated by the cal-culated bulk partition coefficients for two modelgarnet peridotites (Table 4) shown in Figure 4. Theresults in this figure show that the water-rich fluidis strongly enriched in light REE, and that the de-gree of light-REE enrichment increases with in-creasing modal garnet content of the rock. The ab-solute REE abundance of the fluid will alsoincrease rapidly with pressure if the assumptionsbased on the vapor-diopside partitioning data areappropriate. The bulk partition coefficient involvinggarnet-free peridotite and water-rich vapor does notvary significantly with atomic number because none

Sm

I z t L . 51313

18 . 512 . 52L!3

MYSEN: TRACE-ELEMENT PARTITIONING 2't9

SiO2A1z0 stIgOCaO

GacPxoPxOI

Ce5mTu

Table 4. Compositions of garnet peridotites (weight percent)

cPl cP l (s ) cP l (10) cP l ( r6 ) cP2

sampled by kimberlite beneath the South Africancraton. The granular nodules show evidence of ma-jor-element depletion perhaps resulting from partialmelting. The sheared nodules are not depleted inmajor elements (these two types of peridotite nodulesare referred to as depleted and fertile, respectively).This distinction is not seen in the REE patterns (Shi-mizu, l975a,b; Ridley and Dawson, 1975; Mitchelland Carswell, 1976). The contrast between major andtrace elements is perhaps best illustrated by the REEpatterns of a sheared and a granular garnet peridotitenodule (Fig. 5). These data show a considerable rela-tive light-REE enrichment of the depleted granularnodule (352), whereas the sheared fertile nodule(l6l l) shows an essentially unfractionated REE pat-tern. The light-REE enrichment is also seen in otherdepleted garnet peridotite nodules from kimberliteand related rocks (Fig. 5). The patterns of nodules352, BD 738, and BD 776 are not the result ofinter-action between the nodules and the host magma dur-ing ascent, according to evidence and discussion bythe original authors (Shimizu, 1975a; Ridley andDawson, 1975).

The model used to explain the contrasted major-and trace-element contents of depleted and fertilegarnet peridotite is as follows. Consider an uppermantle consisting of fertile peridotite. As the result ofpartial melting and extraction of the melt from thesource region, the residual peridotite after the meltingevent will acquire a depleted major- and trace-ele-ment character. Metasomatic alteration of the de-pleted material is accomplished by dehydration ofminor minerals in neighboring fertile peridotite. The

1 0 0 0

1100 0c

Lo Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb Lu

Fig. 4. Bulk partition coefficients, Df,6$-"'v"t"t, for garnetperidoti tes GPI and GP2 as a function of pressure as l l00oC.Error bars f lo.

4 7 . 6

4 7 , 32 . 6

4 7 . 4L . 9

4 8 . 5

1 . 38 , 3

20,66 3 . 8

4 7 , 31 . 1

4 9 . 9

6 . 5z t . 358 .0

4 7 . 0{ . 15 2 , O1 . 0

0 . L4 , O

, , 1

4 7 . 8J . I

4 4 . 52 . 6

Modal Conposltions

20102050

IO102060

REE Contents*

0 . 4 50 . 9 10 . 9 9

0 . 1 7o . 7 70 . 9 5

0 , 0 4 10 . 4 5 10 . 7 2 1

*Rejat ive to averaqe chondt iXe (schnit t et aL., 1963, j964).

of the minerals in such a peridotite fractionate sig-nificantly between the individual REE.

Petrological implications

Stable continental shields

Direct evidence for heterogeneous distribution oftrace elements in the upper mantle is found in garnetperidotite nodules from kimberlite and related rocksin southern and central Africa (Shimizu, 1975a, b;Ridley and Dawson, 1975; Mitchell and Carswell,1976). Evidence for isotopic heterogeneities is givenby Ridley and Dawson (1975) and Erlank and Shi-mizu (1977). Petrographic descriptions of second-arily-grownr phlogopite and potassic richterite aregiven by, for example, Dawson and Powell (1969),Erlank (1973), Aoki (1974), and Erlank and Rickard(1977).

Data supporting an interpretation of metasomaticalteration of peridotite trace element contents andisotope geochemistry in the mantle prior to transportof the samples of such mantle to the earth's surfaceare provided by Ridley and Dawson (1975), Rhodesand Dawson (1975), and Erlank and Shimizu (1977).The REE data of Shimizu (1975a) are also best inter-preted in terms of metasomatism in the mantle priorto sampling by ascending kimberlite magma.

As mentioned previously, two types of garnet pe-ridotite occur in the portions of the upper mantle

I Secondary growth is taken to mean formation of a mineralsubsequent to equilibration of four-phase peridotite in the uppermantle. The growth could have occurred in the mantle prior toascent in a magma or as a result of interaction between the noduleand the transporting magma.

- 1 n no r v v

o

I

R u rn-L!; o ao 1 0

280 MYSEN: TRACE-ELEMENT PARTITIONING

P rC e N d

T b H o T m L uGd Dy Er Yb and

Fig. 5. Published REE patterns of garnet peridotite nodules inkimberl i te and related rocks. Ref. I , Mitchel l and Carswell (1976);ref. 2, Shimizu (1975a); ref. 3, Ridley and Dawson (1975).

released fluid will equilibrate with the fertile garnetperidotite and migrate into the neighboring depletedperidotite with which the fluid also equilibrates. Themajor- and trace-element patterns of depleted garnetperidotite presumably reflect these two processes.

The most important aspects of this model may beassessed by combining the present REE partitioningdata with information on melting behavior of garnetperidotite (Kushiro and Yoder, 1974; Mysen andKushiro, 1977) and mobility of HrO in the uppermantle (Mysen et al., 1977).

The garnet peridotite nodule 16ll (Nixon andBoyd, 1973) is a likely candidate for the fertile mantlematerial (e.9., Shimizu, 1975a). According to Mysenand Kushiro (1977), partial melting of garnet perido-tite of a composition such as that of l6ll is closelymimicked by the melting behavior observed in simplesystems like CaSiOs-MgSiOr-AlrOs-SiO, (curs). Byemploying the phase-equilibrium data from cnrns(Kushiro and Yoder, 1974), Mysen (1977b) showedthat the stoichiometry of the equation expressing themelting of garnet peridotite is approximated by theformulation

0.63 Ga + 0.41 Cpx f 0.09 Opx : 0.13 Ol + Liq( l )

The modal composition of the idealized model gar-net peridotite, GPI (Table 4), is similar to that ofnodule l61l (Nixon'and Boyd, 1973). In the fol low-ing calculations, composition GPI represents the fer-tile garnet peridotite, and compositions representingvarious degrees of depletion are derived by meltingGPl with a stoichiometry according to expression( I ). With the stoichiometry of equation ( I ), l6.l per-cent melt can be generated from GPl under invariantconditions. This amount of melting represents themaximum degree of depletion of garnet peridotiteG P I .

The equations that describe element concentra-tions in a metasomatized rock are derived in theAppendix. For a source rock with element concen-trations R? and a metasomatized rock with elementconcentration R3 before metasomatism, the concen-trations of i in the bulk rock (R\) and the crystal-line residue [(Gt)M] are:

RY : R?(1 - Y" X.') + R? X, T"Di -& (D i - l )

tULllE

!(l,. NoE 4

cI

o'c!co

O

(2)

(3)

P m E uSm

(C ' i ' ) ' - RY

(l - y" *.r* wIn these equations, X. and X* are the mass propor-tions of fluid in source rock and metasomatized rock,respectively. The bulk partition coefficients for sourceand metasomatized rocks are Di and DM1. Finally, theparameter IZ" is the ratio of fluid remaining in themetasomatized rock to equilibrate with it divided bythe mass proportion of fluid entering the rock toinitiate the metasomatic alteration.

Equations (2) and (3) show that the element con-centration of a metasomatized portion of the uppermantle is a complex function of a variety of poorly-known parameters. The results of varying these pa-rameters are shown in Figure 6. In this figure, GPI isconsidered the source of metasomatizing fluid. Thepressure in all calculations is 40 kbar. The influenceof pressure on the element concentrations of themetasomatized rock would be the result of pressure-dependence of

' the crystal-crystal partition coeffi-

cients, and the result of changes in modal proportionsof minerals in the rocks because of the pressure-dependence of solid solutions in the minerals (e.g.,MacGregor, 1974).It. was argued above that the pres-sure-dependence of crystal-crystal partition coeffi-cients may not be large. No such dependence wastherefore considered in the calculations. According toMacGregor, the proportion of garnet would increase

MYSEN: TRACE.ELEMENT PARTITIONING 281

oorcoEo

co()(l)(L

24o0

2 roo

| 800

| 500

| 200

900

600

300

o

A

C e l T mI

II

p l ' /" t /

/ cets^

, " 2Y 'ce/ rn

--t-ut'

/ / ./ Ce/Sm. '--ttcl.tgyrZ U'- t

t u i

lcf trM

B

\'\ Rll\\\

t..Rf

tcfr tM t . .

rcf rlM

CelTm

roxene content of the rock. Garnet is the only mineralin garnet peridotite that shows significant fractiona-tion of individual REE. Its presence in the metasoma-tized rock will tend to oppose light-REE enrichmentin the rock. The less garnet there is in depleted GPl,the less important this effect will be. Because thegarnet content of the residual garnet peridotite de-creases as the degree of melting increases, the effect ofmetasomatism on the REE pattern of the metasoma-tized rock increases. The effect of the ratio of mass offluid in source rock and metasomatized rock at F :0.16 and 40 kbar is shown in Figure 6B. The morefluid relative to rock that enters the depleted perido-tite, the greater the degree of enrichment. Put differ-ently, the smaller the mass of rock to be metasoma-tized for a fixed amount of fluid, the morepronounced is the effect of metasomatism on thisrock.

Finally, it is geologically unreasonable that therock system in question is closed to HrO. Con-sequently, only a portion of the fluid entering rock Mremains in the rock and equilibrates with it. As theresult, y, < l. Lowering of the value of I" lowers theeffective value of ,Yon thus resulting in a reduction of

2xs,/xM

Fig. 6. Degree of change of light REE of metasomatized garnet peridotite [bulk REE (crystals * vapor) - RY REE content ofcrystals - (Ci')t] relative to CelSm and CelTm of unaltered, depleted garnet peridotite. (64) As a function of degree of depletion[expressed as proportion of fractional melt extracted (F) prior to metasomatism] with X./XM: l, Ri = l. (68) As a function of X./Xvw i t h F = 0 .16 and R i : l .

o.o5 0.ro o.t5 0 |Degree o f dep le l ion

at the expense of aluminous orthopyroxene with in-creasing pressure. In GPI the relative increase ofgarnet in the range 20-40 kbar would be about l0percent. This variation will not significantly affect thevalue of the bulk partition coefficients. It is con-cluded, therefore, that the results of the calculationsapproximate the results of metasomatic alteration ofgarnet peridotite in the upper mantle in the pressurerange of garnet peridotite nodules sampled by kim-berlite.

All the curves in Figure 6 represent percent changeof enrichment factors, Ce/Sm and CelTm. relative totheir values after extraction of 5, l0 and 16 percentfractional melt as calculated according to the stoi-chiometry of equation (l) using the equations ofShaw (1970). The original values of these enrichmentfactors are given in Table 4.

The effect of modal composition of rock M (de-pleted GPI) at X"/X*: I and Yu: I is shown inFigure 6,4. The results show that, as the peridotiterock to be metasomatized becomes more depleted,the degree of light-REE enrichment increases rapidly.The main change of GPI during depletion is a lower-ing of garnet content and, to a lesser degree, clinopy-

282 MYSEN: TRACE.ELEMENT PARTITIONING

X"/X* as evaluated in Figure 68, so that X"/X* isproportional to Y,.

In applying the above considerations to the pos-sible metasomatic alteration of REE patterns of pe-ridotite in the portions of the upper mantle sampledby kimberlite, some information on the values of theparameters discussed above is needed. Mysen andKushiro (1977) argued that because garnet peridotitemelts under near invariant conditions, partial meltingof a source rock such as GPI will continue until oneof the crystalline phases is exhausted. Consequently,depletion of GPI to F : 0.16 (Table 4) is likely. TheREE content ofl fertile garnet peridotite is likely to bein the range l-3 relative to average chondrite (e.g.Shimizu, 1975a; Loubet et al., 1975). From the resultsshown in Figure 6, it appears that XM ) X" to resultin significant alteration of REE patterns of depletedgarnet peridotite.

In Figure 7, calculated REE patterns of metasoma-tized depleted garnet peridotite are compared withthe range of REE abundances found in depleted gar-net peridotite. It appears from the results (Fig.7) thatX* ) X. to generate light-REE enrichment in meta-somatized, depleted garnet peridotite. The range ofabsolute REE abundance is covered by varying the

Fig. 7. REE patterns of metasomatized, depleted garnetperidotite (f : 0 16) at 40 kbar compared with range of valuesfrom nodules ( f rom Fig. 5) . GPI-DEP-R3 wi th F : 0.16and initial Rj; GPI-DEP-M = GPI-DEP after metasomatism.

Fig. 8. Published chondrite-normalized (Schmitt et al., 1963,1964) REE patterns for island arc andesites.

R! between the values commonly suggested for fertilegarnet peridotite.

Island arcs

Another region of the upper mantle that may con-

tain significant amounts of HzO is the source region

of melts beneath island arcs. High water contents of

lavas in island arcs (up to 5 percent or more) have

been indicated by studies of Anderson (1975) andMuenow et al. (1977), suggesting the presence of

significant amounts of water in the source region of

these melts. The trace-element geochemistry of ande-

sitic lavas indicates enrichment in large tithophile

elements (LIL) and perhaps l ight REE (Fig. 8), which

suggests unusually large amounts of LIL elements in

the source of island arc andesites or a different miner-

alogy compared with the source of oceanic basalts

that do not show light-REE enrichment.Most models for the formation and evolution of

andesitic melts involve the presence of HrO (see

Boettcher, 1973, for a summary of these models). It

has also been suggested that H2O may play a role ingoverning the trace element contents of andesitic

melts (Jakes and Gill, 1970; Fyfe and McBirney,

1975). Arguments have also been put forth for melt-

ing of a descending oceanic plate under essentiallyanhydrous conditions (e.g., Marsh, l976a,b). Marsh

argued that dehydration of the plate occurred at a

shallower depth than that of andesite formation. Fur-

l..lJLrJ

!oN

oE

I

o

E

1L o P r P m E u T b H o T m L u

C e N d S m G d D y E r Y b

thermore, he believed that HrO was essentially immo-bile in a crystalline upper mantle. Consequently, theeffect of water on the melting relations of peridotitecalled upon by other authors (as summarized byBoettcher, 1973) for the formation of andesite was anunlikely model. He also argued that the effect offrictional heating of the slab-mantle interface was solarge that melting would occur there under any cir-cumstance. The thermal models used in those argu-ments (such as calculated by Toksoz et al.,1971, andTurcotte and Schubert, 1973, for example) use partialmelting on the interface as a boundary condition.Petrological arguments for partial melting of a de-scending slab of basaltic composition based on suchthermal models are therefore of questionable valuebecause partial melting of the slab was one of theassumptions used for the thermal models.

Delaney and Helgeson (1978) cast further doubt onmodels for melting of the slab. They calculated thethermodynamic parameters of probable dehydrationreactions that may take place in the descending platebeneath island arcs. These variables were combinedwith various thermal models to calculate the depthand temperature where such dehydration may occur.The pressure and temperature ranges given byDelaney and Helgeson were 25-40 kbar and 400.-600oC, respectively. It should be noted that basalt+ H2O solidi are always at temperatures greater than600oC (e.g., Robertson and Wyllie, l97l; Lambertand Wyllie, 1972; Allen et al., 1975\. If the results ofDelaney and Helgeson are correct, dehydration of adescending plate of basaltic bulk composition mayoccur at depths ranging from 70 to 120 km withoutresulting in melting of the slab material. providedthis fluid can migrate into the overlying peridotiticwedge from the slab itself-as also argued by Jakesand Gill (1970) and Fyfe and McBirney (1975)-thetrace-element content of the overlying wedge may beaffected. The expected effect on REE patterns ofpartial melts from a metasomatized peridotitic wedgeoverlying dehydrating basaltic rocks in the descend-ing plate will be considered here, and the validity ofsome of the preceding arguments may be assessed.

A case where amphibole, clinopyroxene, and gar-net coexist with hydrous vapor in the slab is con-sidered. Because K$g'gt'-cn* is near I (Mysen,1977a), the vapor-crystal partition coefficients ofclinopyroxene and amphibole are likely to be simi-lar. Therefore, clinopyroxene * amphibole is con-sidered as one mineral and referred to as Cpx inthe following discussion. The fluid is considered tobe in equilibrium with both the dehydrating rock of

REE Content6**

1 1 0 1 01 1 0 1 01 1 0 1 0

*Modaf spinel evenTg divided bexreen 07 and opx fot caJculationputtrbses.

*xReiative to avetage clondrite.

MYSEN: TRACE-ELEM ENT PARTITIONING 283

eclogitic mineral assemblage and the peridotite. Anuncertainty in these calculations is the temperature.According to Cullers et al. (1973), the vapor-crystalpartition coefficients for REE increase with decreas-ing temperature at least at I kbar. If this trend alsoapplies to REE partitioning at higher pressures,and the temperature of dehydration and metaso-matism is lower than I100'C (the experimentaltemperature), the bulk partition coefficients usedhere are minimum values. Consequently, the magni-tude of the change of REE contents and patterns inperidotite beneath island arcs after passage of themetasomatic front represents minimum values.

A case where spinel peridotite (SPl; see Table 5)overlies a dehydrating slab containing amphibole,clinopyroxene, and garnet is considered. Dehy-dration of an amphibole-rich rock of bulk composi-tion similar to oceanic tholeiite is likely to occur inthe pressure range 20-30 kbar, resulting in a rock ofeclogitic mineralogy. The fluid is considered to be inequilibrium with both the dehydrating rock of ec-logitic mineral assemblage and the peridotite. Theamount of released fluid (X"), the amount of fluid inSPI after metasomatism (Xr), and the ratio of garnetto clinopyroxene in the eclogite will affect the extentto which the REE pattern of the spinel peridotiteoverlying the descending plate will be affected. In thefollowing calculations the amount of released water isvaried between 0.25 and 1.0 weight percent. Theamount of fluid in SPI is also varied between 0.25and I weight percent and the modal Ga/(Ga + Cpx)between 0.2 and 0.5 (Table 5). From the viewpoint ofmodeling bulk compositions of oceanic basalt in

Table 5. Compositional data on eclogite and garnet-freeperidotite in model island arc environment

sPl

SiO2A1z0:ueoCaO

GacPxoPxO I

CeSnTh

4 1 . 32 . 6

r0 ,030 .1

5 . r20 .820 ,7

2 1 . 81 8 , 3

5L.21 0 . 123,2r 5 . 5

5 0 . 11 t 1

2 4 . 21 3 , 0

llodal compoeltions

20 30 40 5060 50

10 1010 1010 10

MYSEN: TRACE-ELEMENT PARTITIONING

cMAS, of the four compositions used, E4 is the mostsimilar to basalt. Prior to dehydration, the slab mate-rial is considered to have an unfractionated REEpattern with REE abundances equal to ten times thatof average chondrite (Schmitt et al., 1963, 1964).Such REE contents are typical for oceanic tholeiite(e.9., Frey et al., 1974).In an open system where HrOcan move freely, the value of )/" [equation (2)] is lessthan l. Lowering of I/" to values less than I wouldresult in a shift of absolute REE abundances of themetasomatized peridotite to lower values. It is as-sumed, however, that as HrO migrates into the spinelperidotite and the peridotite begins to melt, all avail-able HrO will dissolve in the melt. Consequently, thesystem is effectively closed to HrO and f, : 1.

The REE patterns of metasomatized spinel perido-tite SPI are shown in Figure 9. The results show thatan increasing proportion of garnet in the source rockresults in a slight increase in the relative l ight-REEenrichment of SPl. A lowering of the proportion offluid in the source rock (X.) relative to the proportionof f luid in the metasomatized rock results in a slightincrease in l ight-REE enrichment of SPI (Fig. l0).Lowering of Xy results in a more rapid reduction ofdegree of l ight-REE enrichment, however (Fig. 2).The latter case would be encountered if the total massof SPI exceeds that of the eclogite source, whereas inthe former case (X" ( Xrra), the mass of eclogitesource is less than that of metasomatized SPl.

Partial melts of similar major-element compositionderived from SPI before and after metasomatismwould have different REE patterns (Fig. l l). In Fig-ure l l, the compositions of partial melts from SPI

Xs = o 'otX u = o ' o t

Ce Sm Tm

Fig. 9. REE patterns of metasomatized spinel peridotite as afunct ion ofgarnet content ofsource rock. In th is f igure, X" : 0.01,xM : 0 .01 , yv = l .

S P I - E 4

Ce Sm Tm

Fig. 10. REE patterns of SPI after metasomatic alteration byfluid from source rock E4 as a function of proportion of fluid in

source rock (X") and metasomat ized rock (X") .

and GPI (where GPI is garnet peridotite of similarbulk composition to SPI) with an unfractionatedchondrite-normalized REE pattern equal to unity arecompared with the patterns for partial melts derivedfrom metasomatized SPl. The stoichiometry of theequation describing the melting of garnet-free perido-tite is based on data of Kushiro (1969) as derived byMysen (1977c),

0.54 Cpx + 0.54 Opx : 0.08 Ol + L iq (4)

whereas equation (l) describes melting of garnet pe-ridotite. The crystal-liquid partition coefficients forREE are from Mysen (1977a). The liquid composi-tions are those of an accumulated fractional meltcalculated with the formulations of Shaw (1970). Theuse of a batch-melting model does not significantlyaffect the results. In this figure, 5 and 10 percentmelting (f : 0.05 and 0.10, respectively) are consid-ered.

Metasomatic alteration of the REE contents ofspinel peridotite source rock for andesitic melts re-sults in a l ight-REE enrichment of this rock that is sogreat that partial melts from such a rock would bemore enriched in l ight REE and also have higherabsolute REE abundances than partial melts fromgarnet peridotite that was not metasomatically al-tered prior to melting (Fig. I l). The similarity of thecalculated patterns and those for natural andesite inisland arcs is striking. It is concluded, therefore, thatmetasomatized spinel peridotite deriving its HzOfrom dehydration of a descending oceanic plate withan eclogitic mineral assemblage is a l ikely model forandesitic formation in island arcs. The model leads to

l o

Rl',l

RT

MYSEN. TRACE-ELEMENT PARTITIONINC

where X" is the proportion of s that is converted tofluid, (Cit)" and (CY)' are concentrations of i incrystal and vapor in source rock, respectively. Ri isthe total content of element, i, in rock, s.

For the metasomatized rock. M. the mass bal-ance is:

R'" = RP (l - X') * (CY)"X' (4)

where R1D and RiM are the concentrations of ele-ment, i, before and after the fluid entered rock M.Xy is the proportion of fluid in rock M. If only aportion of the fluid (f") remains in the metasoma-tized rock, equation (4) becomes:

(Ci')' (l - Y"XM) + (L\)MY"Xu : Rl' (5)

where (Cit)M and (Cj)M are the concentrations of iin crystals and coexisting vapor, respectively.

From equat ions (1) and (3):

(qf :Dl- X. (D? - l )

From equations (4) and (6), the total content ofelement. i. in metasomatized rock. M:

RY:R3( l -Y ,X* ) r Ri XM Y.Di- x" (Di - l )

From equations (2), (5) and (7), the element con-tents of coexisting rock M and vapor can be calcu-lated:

(ci")' :

Fig. l l . REE pat terns of part ia l mel ts f rom Spl and Gplcompared with range of REE abundances in island arc andesite(data from Gill,1974;Kay,1977;Yajima et al., 1972;Thorpe et at.,1976). Curve A: 5 percent accumulated f ract ional mel t f rom Spl-E4 with X, = 0.0025, X, : 0.01; Curve B: Same as A, bur withl0 percent melt; Curve C: Same as B, but with X" : 0.01, Xu =0.0025; Curve D: 5 percent partial melt from unaltered Gpl; CurveE: 5 percent partial melt from unaltered SPl.

agreement between both REE data and informationon phase equil ibria of wet peridotite under the appro-priate pressure conditions.

Appendix

Bulk partition coefficients are defined:

Dl, = >(R/ I{,'*) : (g'),"i / ^ \ ^ " ) : ( C , J ( l )

DMr: Z(py//c/xt) - (q ') '""i/ ̂\,^.) : n.,f e)

where Pi and P\ etc. are modal proportions ofphases in source rock (s) and metasomatized rock(M), respectively. The partit ion coefficients, Kyt't -

CYapor/Cetvst"r, are weight ratios of element, i, invapor and crystal.

From mass balance, for the source rock. s:

E

o

E so

R?

R^1

(6)

(7)

(8)( l-Y,xM)-w

Equations (7) and (8) are used in the calculations inthe text.

AcknowledgmentsCritical reviews by W. Harrison, J. R. Holloway, D. Rumble III,

J . -G. Schi l l ing, R. Vidale, R. J. Wendlandt , and H. S. Yoder, Jr . ,are appreciated.

References

Allen, J. C., A. L. Boettcher and G. Marland (1975) Amphiboles inandesites and basalt. I. Stability as a function of P-T-JOr. Am.Mineral.. 60. 1069-1085.

Anderson, A. T. (1975) Some basaltic and andesitic gases. Reu.Geophys. , 1 3 , 37 -57 .

Aoki, K. (1974) Chromian spinels in lherzolite inclusions fromHinome-gata, Japan. C ont rib. M ineral Petrol., 46, 249-256.

Boettcher, A. L. (1973) Volcanism and orogenic belts-the originof andesite. Tectonophysics, I 7, 231-244.

-, B. O. Mysen and P. J. Modreski (1975) Phase relat ionshipsin natural and synthetic peridotite-Hro and peridotite-Hro-CO, systems at high pressures. Phys. Chem. Earth,9, 855-867.

Boyd, F. R. and J. L. England (1960) Apparatus for phase equil ib-rium measurements at pressures up to 50 kilobars and temper-(ci ')" (l - x.) + (c)" x" : Ri (3)

286 MYSEN: TRACE-ELEMENT PARTITIONING

atures up to 1750'C. J. Geophys. Res.,65, '141-748.Burnham, C. W. (1967) Hydrothermal fluids at the magmatic

stage. In H. L. Barnes, Ed., Geochemistry of Hydrothermal OreDeposits, p. 34-67. Holt, Rinehart, and Winston, New York.

Cullers, R. L., L. G. Medaris and L. A. Haskin (1973) Experimen-tal studies of the distribution of rare earths as trace elementsamong silicate minerals and water. Geochim. Cosmochim. Acta,37 .1499-15t3 .

Dawson, J. B. and D. G. Powell (1969) Mica in the upper mantle.Contrib. Mineral. Petrol., 22, 233-23'l .

Delaney, J. M. and H. C. Helgeson (1978) Calculation ofthermodynamic consequences of dehydration in subductingoceanic crust to 100 kbar and 1000"C. Am. J.5ci. ,278,638-686.

Delaney, J. R., D. Muenow, J. Ganguly and D. Royce (1977)Anhydrous glass-vapor inclusions in oceanic pillow basalts(abstr.). EOS, J8, 530.

Eggler, D. H. (1973) Role of CO, in melting processes in themanlle. Carnegie lnst. Wash Year Book,72, 457467.

- (1974) Application of a portion of the sysrem CaAlzSLO"-NaAISLOT-SiO,-MgO-Fe-Or-H,O-CO, to genesis of the calc-alkal ine suite. Am. J. Sci. . 274.297-3t5.

- (1975) CO2 as a volat i le component of the mantle: thesystem MgrSiO.-SiOr-HrO-CO2. Phys. Chem. Earth, 9, 869-881 .

- (1976) Does COz cause partial melting in the low velocitylayer of the mantle? Geology, 2, 69-72.

- 2ni M. Rosenhauer (1978) Carbon dioxide in silicate melts.IL Solubilities of CO, and H,O in CaMgSLOe (diopside) liquidsand vapors at pressures to 40 kbar. Am. J. Sci. ,278,64-94.

Erlank, A. J. (1973) Kimberlite potassic richterite and the distribu-tion of potassium in the upper mantle. Extended Abstracts,International Kimberlite Conference, University of Cape Town,p. 103-106.

- and R. S. Rickard (1977) Potassic richterite bearing perido-tites from kimberlite and evidence they provide for upper mantlemetasomatism. Extended Abstracts. Second International Kim-berlite Conference, Santa Fe, New Mexico.

- and N. Shimizu (1977) Strontium and strontium isotopedistributions in some kimberlite nodules and minerals. ExtendedAbstracts, Second International Kimberlite Conference, Santa Fe,New Mexico.

Frey, F. A., W. B. Bryan and G. Thompson (1974) Atlantic Oceanfloor: geochemistry and petrology of basalts from Legs 2 and 3of the Deep-Sea Drilling Project. J. Geophys. Res.,79, 5507-5527.

Fyfe, W. S. and A. R. McBirney (1975) Subduction and the struc-ture of andesitic volcanic belts. Am. J. Sci.. 275-A. 285-297.

Gill, J. B. ( 1974) Role of underthrust oceanic crust in the genesis ofa Fijian calc-alkaline suite. Contib. Mineral. Peno\.,43,2945.

Huang, W. L. and P. J. Wyllie (1976) Melting relationships in thesystems CaO-COz and MgO-CO, to 33 kilobars. Geochim. Cos-mochim. Acta, 40, 129-132.

Irving, A. J. and P. J. Wyll ie (1975) Subsolidus and melt ing rela-tionships for calcite, magnesite on the join CaCOr-MgCO, to 36kb. Geochim. Cosmochim. Acta, 39,35-53.

Jakes, P. and J. Gill (1970) Rare earth elements and island arctholeiitic series. Earth Planet. Sci. Lett.,9, 17-28.

Kay, R. W. (1977) Geochemical constraints on the origin of Aleu-tian magmas. In M. Talwani and W. C. Pitmann lll,Eds., lslandArcs, Deep Sea Trenches and Back Arc Basins, p. 229-242.American Geophysical Union, Washington, D. C.

Kennedy, G. C., G. J. Wasserburg, H. C. Heard and R. C. Newton(1962) The upper three-phase region in the system SiOr-HrO.Am. J. Sci.. 260. 501-521.

Kushiro, I. (1969) The system forsterite-diopside-silica with andwithout water at high pressures. Am. J. Sci., 267-A,269-294.

- (1972) Effect of water on the composition of magmasformed at high pressures. J. Petrol., 13,331-334.

_, N. Shimizu, Y. Nakamura and s. Akimoto (1972) compo-sitions of coexisting liquids and solid phases formed upon melt-ing of natural garnet and spinel lherzolites at high pressures: apreliminary report. Earth Planet. Sci. Lett., 14, 19-25.

- and H. S. Yoder, lr. (1974) Formation of eclogite fromgarnet lherzolite: liquidus relations in a portion of the systemMgSiOr-CaSiOe-Al2Os at high pressure. Carnegie Inst. Wash.Year Book,73,266-269.

and M. Nishikawa (1968) Effect of water on themelting of enstatite. Geol. Soc. Am. Bull.,79, 1685-1692.

Lambert, I. B. and P. J. Wyllie (1972) Melting of gabbro (quartzeclogite) with excess water to 35 kilobars, with geological appli-cations. J. Ceol., 80, 693-708.

Loubet, M., N. Shimizu and C. J. Al legre (1975) Rare earthelements in alpine peridotites. Contrib. Mineral. Petrol., 53, l-12.

MacGregor, I. D. (1974) The system MgO-ALOg-SiOr: solubilityof AlzOs in enstatite for spinel and garnet peridotite composi-t ions. Am. Mineral. , J9, l l0-119.

Marsh, B. D. (1976a) Mechanics of Benioff zone magmatism. /m.Geophys. Union Monogr., 19, 337-350.

- (1976b) Some Aleutian andesites: their nature and source.J. Geol..84.2' l-45.

Mitchel l , R. H. and D. A. Carswell (1976) Lanthanum, samariumand ytterbium abundances in some southern African garnetlherzolites. Earth Planet. Sci. Len.,31, l'15-178.

Muenow, D., J. R. Delaney, A. Meijer and N. Liu (1977) Water-rich glass-vapor inclusions in phenocrysts from tholeiitic pillowbasalt r ims in the Marianas back arc basin (abstr.). EOS,58,530.

Mysen, B. O. (1977a) Experimental determination of cerium, sa-marium and thulium partitioning between hydrous liquid, gar-net peridotite minerals and pargasite. Carnegie Inst. Wash. YearBook. 76.558-564.

- (1977b) Magma genesis in peridotite upper mantle in lightof experimental data on partitioning of trace elements betweengarnet peridotite minerals and partial melts. Carnegie Inst.lAash. Year Book, 76,545-550.

- (1977c) The solubility of HrO and CO, in the predictedpressure and temperature range of magma genesis in the uppermantle and some petrological and geophysical implications.Reu. Geophys., 1J, 351-361.

- snd A. L. Boettcher ( 1975a) Melting of a hydrous mantle. I.Phase relations of natural peridotite at high pressures and tem-peratures with controlled activities of water, carbon dioxide andhydrogen. J. Petrol., 16,520-548.

- and - (1975b) Melt ing of a hydrous mantle. I I . Geo-chemistry of crystals and liquid formed by anatexis of mantleperidotite at high pressures and high temperatures as a functionofcontrolled activities ofwater, hydrogen and carbon dioxide../.Petrol. , 16.549=590.

- 31fl I. Kushiro ( 1977) Compositional variations of coexist-ing phases with degree of melting of peridotite in the uppermantle. Am. Mineral., 62,843-865.

and T. Fujii (1977) Experimental studies of diffu-

MYSEN TRACE-ELEMENT PARTITIONING ?8',7

sion of water in crystalline peridotite and partitioning of sometrace elements between water-rich vapor and crystals in theupper mantle. Extended Abslracls, Second International Kimber-lite Conference, Santa Fe, New Mexico.

- and M. G. Seitz (1975) Trace element partitioning deter-mined by beta-track mapping-an experimental study using car-bon and samarium as examples../. Geophys Res., 80, 2627-2637.

Nakamura, Y. and I Kushiro ( 1974) Composi t ion of the gas phasein Mg,SiO.-SiOr-HrO at l5 kbar Carnegie Inst. Wash YearBook. 73. 255-259.

Nixon, P. H. and F. R. Boyd (1973) Petrogenesis of the granularand sheared ultrabasic nodule suite in kimberlites. In P. H.Nixon, Ed., Lesotho Kimberlites, p. 48-56. Cape and TransvaalPrinters, Ltd., Cape Town.

Oxburgh, E. R (1964) Petrological evidence for the presence ofamphibole in the upper mantle and its petrogenetic and geophys-ical impl icat ions. Geol . Mag., 101, l -19.

Rhodes, J. M. and J. B. Dawson (1975) Major and t race elementgeochemistry of peridotite inclusions from the Lashaine vol-cano, Tanzania. Phys Chem. Earth, 9,545-559.

Ridley, W. l . and J B. Dawson (1975) L i thophi le t race elementdata bearing on the origin of peridotite xenoliths, ankaramiteand carbonatite from Lashaine volcano, Tanzania. Phys. Chem.Earth, 9, 559-57 1 .

Robertson, J. K. and P. J. Wyl l ie (1971) Rock-watersystems, wi thspecia l reference to the water-def ic ient region. Am J.Sci . ,271,252-278.

Rubey, W. W. (1955) Development of the hydrosphere and atmo-sphere with special reference to the probable composition oftheearth's atmosphere in the crust ofthe earth. Geol. Soc Am. SpecPap , 62, 63 l -655.

Schmit t , R. A. , R. H. Smith, J. E. Lasch, A. W. Mosen, D. A.Olehy and J. Vasi levski (1963) Abundances of fourteen rare-earth elements, scandium and yttrium in meteoric and terrestrialmaller. Geochim. Cosmochim. Acta. 27. 577-622.

and D. A. Olehy (1964) Rare-earth, yt t r ium andscandium abundances in meteoric and terrestrial matter. II.Geothim. Cosmochim Acta. 28. 67-86.

Shaw, D. M. (1970) Trace element fractionation during anatexis.Geochim. Cosmochim. Acta. 34. 237-243.

Shimizu, N. (1975a) Rare earth e lements in garnets and c l inopy-roxenes from garnet lherzolite nodules in kimberlite. EarthPlanet . Sci . Let t . . 25.26-32.

- (1975b) Geochemistry of ultramafic inclusions from SaltLake Crater. Hawaii. and from southern African kimberlites.Phys Chem. Earth,9, 655-'l'71.

Thorpe, R. S. , P. J. Pot ts and P. W. Francis ( 1976) Rare earth dataand petrogenesis of andesite from North Chilean Andes. Coz-trib. Mineral. Petrol., 54, 65-78.

Toksoz, M. N., J . W. Minear and B. R. Jul ian (1971) Temperaturefields and geophysical effects of a downgoing slab. J. Geophys.Res , 76 , l l l 3 - l 138 .

Turcot te, D. L. and G. Schubert (1973) Fr ic t ional heat ing and thedescending lithosphere. J. Geophys. Res., 78,5876-5886.

Yajima, T., H. Higuchi and H. Nagasawa(1972) Variations of rareearth concentrations in pigeonitic and hypersthenic rock seriesfrom Izu-Hakone region, Iapan. Contrib. Mineral. Petrol., 35,235-245.

Yoder, H. S., Jr. (1950) High-low quartz inversion to 10,000 bars.Trans. Am. Geophys. Union, 31,827-835.

- (1969) Calc-alkaline andesites: experimental data bearingon the origin of their assumed characteristics. Oregon Dept.Geol. Miner. Ind. Bull.,65,77-89.

Ziel inski, R. A. and F. A. Frey (1974) An experimental study ofthe partitioning of a rare earth element (Gd) in the systemdiopside-vapor. Geochim. Cosmochim. Acta, 38, 545-565.

Manuscript receiued, April 24, 1978;accepted Iot publication, July 25, 1978.