partitioning of ree between minerals and coexisting melts...

American Mineralogisl, Volume 66, pages 242-259, 1981

Partitioning of REE between minerals and coexisting meltsduring partial melting of a garnet lherzolite

WnNov J. HlnnrsoN'

Geophysical Lab orat ory, Carnegie I nstitution of Washingtonlllashington, D. C. 20008

Abstract

Partition coeftcients (D:ru : Cf /C!, where C, is the concentration of element i in phases aand D) for Ce, Sm, and Tm between garnet, clinopyroxene, orthopyroxene, olivine, and melthave been determined at 35 kbar for 2.3,8, 20, and 37.17o meltrng of a garnet lherzolite nod-ule with chondritic nsn abundances. Partition coeftcients increase as the degree of partialmelting increases. From 2.3 to 87o melting, this increase is largely a consequence of non-Henry's law behavior of nBB in minerals. At melt percentages > 9Vo, changrng temperatureand melt composition, as well as non-Henry's law behavior, also influence the values of nnrpartition coefficients. The total increase in D;ggdu"t' from 2 to 38Vo melt may be up tolUVo for some minerals and REE, and the assumption made in petrogenetic modeling of con-stant partition coefficients is therefore questionec.

Experimentally determined nEE abundances in the 2.3 and 8Vo melts can be adequatelymodeled with an equilibrium partial melting model and a melting reaction, determined byMysen (1977a), in the system CaO-MgO-AlrOr-SiOr:

garnet + 0.67 diopside * 0.14 enstatite = 0.22 forsterite + l.6l liquid

nnn abundances cannot be calculated at higher percentages of melting because the meltingreaction is not known.

The LREE enrichment rn the 2.3Vo melt is 22 times chondritic abundance, and this melt haschondrite-normalized CelTm:4.18. The generation of nes abundances typical of most al-kali basalts (Ce enrichment 40-200 and CelTm 2-15) is not possible even by low percentagesof partial melting of a garnet lherzolite unless the material has been enriched in lnrB relativeto chondrites.

Introduction

The enrichment of trace elements such as light rareearth elements (rner) in alkali basalts implies thatthese magmas may be generated by small percent-ages of partial melting of a source material with achondritic REE pattern. Likely source materials aregarnet lherzolite and spinel lherzolite. Gast (1968),Kay and Gast (1973), and Shimizu and Arculus(1975) suggested that the necessary LREE enrichmentand unr,B depletion in alkali basalts could be mod-eled by small percentages of melting of a garnet pe-ridotite source. It has been shown by Mysen andHolloway (1977), however, that spinel lherzolite willnot yield melts containing the appropriate REE con-

I Present address: NASA Johnson Space Center, Mail Code SN6,Houston. Texas 77058.

tents for alkali basalts unless a prior metasomaticevent is invoked.

The nnp concentrations in melts that may be gen-erated from a model garnet peridotite source arelargely determined by the relative abundances of gar-net and clinopyroxene in the source (Kay and Gast,1973), by the magnitude of partition coefficients forREE between these minerals and the melts, and bythe stoichiometry of the melting reaction (Mysen,1977a). In the absence of a definitive set of partitioncoeffi.cients for natural garnet peridotite minerals andcoexisting melts, data from other systems (e.g., experimentally determined values in simple and complexsynthetic systems: Shimizu and Kushiro, 1975; b-ving, 1978; Mysen, 1978a) must be extrapolated tothe problem at hand. It has been found that Earnet/liquid partition coefficients and, to a lesser degree,clinopyroxene/liquid partition coeftcients vary as a

0003-o04x/8 l /030/.4242$02.ffi

HARRISON: PARTITIONING OF REE

function of pressure, temperature, and phase compo-sition (e.9., Woodo 1976; Harrison, 1978; l979a,b; h-ving, 1978; Irving and Frey, 1977, 1978; Mysen,l978a,b), as well as the REE concentration in thecrystals (Wood, 1976; Harrison, 1977, 1978; Mysen,1978a). Because physical conditions and phase com-positions change during a melting event (Mysen andKushiro, 1977), it is unlikely that partition coeffi-cients will remain constant. Thus, model calculationsof ReB abundances in melts that employ constantvalues of partition coefficients, determined fromother systems, will be imprecise.

Partition coefficients for three REE between olivine.orthopyroxene, clinopyroxene, garnet, and melt havebeen determined for 2.3, 8,20, and 37.77o melting ofa natural garnet lherzolite to assess the magnitude ofvariation of partition coefrcients during a meltingevent. Additionally, ReB enrichments in the meltshave been calculated to evaluate the possibility thatthe RBB characteristics of alkali basalts may be gen-erated by small degrees of melting of a garnet lherzo-lite with chondritic nnr abundances.

Experimental methods

The starting material was a sheared garnet lherzo-lite nodule, puN 16l l, described by Nixon and Boyd(1973). This nodule was chosen because its major ele-ment composition is considered representative of un-depleted upper mantle material and its REE pattern isessentially chondritic (Shimizu, 1975).

The starting material was crushed to < 5 plm andfired at 1150'C at /(Ot : l0-e atm for 18 hr to de-compose secondary hydrous minerals. Three samplesof the dried, powdered material were spiked with ra-diotracers for trace element analysis by autoradiogra-phy (Mysen and Seitz, 1975). Cerium-I4l, samar-ium-151, and thul ium-l7l are sui table l ight,intermediate, and heavy REE, respectively. These Bisotopes were added in dilute acidic solutions (1.716ppm 'a'Ce, 1.277 ppm '''Sm, and 0.0536 ppm '''Tm).

The spiked 5amples were ground thoroughly underalcohol and stored at ll0oC prior to use.

Partitioning studies require selected percentages ofmelt coexisting with known mineral assemblages.Previous determinations of the melting relations ofPHN 16ll have been made with platinum or plati-num-gold capsules as sample containers, and ironloss from both minslals and melts is incurred (Mysenand Kushiro, 1977; Wendlandt and Mysen, 1978).Mysen and Kushiro (1977) reported 207o iron lossfrom samples at 20 kbar and 1550'C and sub-stantially greater loss at 35 kbar. Because the sub-

stitution of Fe for Mg results il an expansion of thecrystal structure in many mantle minerals, Fe-bear-ing minslals will probably tolerate more REE thanFe-free minerals. It was decided that an attempt toreduce Fe loss should be made to obtain more realis-tic partition coefficients. A redetermination of themelting relations was carried out by use of a doublecapsule configuration comprising a1 innsr graphitecapsule and an outer platinum capsule.

An additional sample of powdered material wasspiked with 15 ppb "'W in NftOH solution ftindlysupplied by Dr. G. Cowan, Los Alamos ScientificLaboratory). The technique of determining amountsof melting in which a trace element partitions en-tirely into the melt phase was described originally byMysen and Kushiro (1977) for the element 'oC. Thiselement will be taken up by the innq glapfui1e cap-sules, however, and thus tttW was selected as an al-ternative (Wendlandt and Mysen, 1978).

The nBB contents of minerals and melts were de-termined by autoradiography (Mysen and Seitz,1975). Partition coefficients (Df/b: [concentration ofelement i in phase al/lconcentration of element i inphase bl) were then calculated, in addition to a chon-drite-normalized nrn pattern for the melts. As dis-cussed by Mysen and Holloway (1977), a patternequivalent to the chondrite-normalized pattern maybe obtained by dividing thp radioactive nnn contentof the melt by the initial radioactive REE spike. Theaccuracy of trace element concentrations determinedby autoradiography has been discussed by Mysen(1978a), and all aspects ofthe procedure used in thiswork are identical with those described by him.

The standards necessary for autoradiography werealbite glasses, prepared at 20 kbar and 1600oC, con-taining known amounts of cerium, samarium, or thu-lium. A sample of pHN 16l l, completely melted at 16kbar and 1650'C (containing 15 ppb "'W and 5.31wl.Vo H2O), was prepared as a standard for determi-nation of degrees of melting.

All experiments were conducted with a solid-me-dia, high-pressure apparatus (Boyd and England,1960). A piston-out technique was used througbout.No corrections for pressure loss due to friction weremade, and nominal pressures quoted are precise to+0.5 kbar (Eggler, 1977). Temperatures were mea-sured with Pt,oo-PtnoRh,o thermocouples and wereautomatically controlled to +loC (Hadidiacos,1972). No corrections were made for the effect ofpressure on thermocouple electromotive force.

Samples were contained in 2-mm-diameter graph-ite capsules that were stored at I lOoC prior to use.


These were contained in 3-mm platinum capsules,which were sealed by welding. The double capsulecon-figuration was employed to minimize the accessof HrO to the dried samples during capsule assemblyand experimental runs. Glass-sleeved furnace assem-blies were used in all experiments to reduce furtherthe access of hydrogen to the samples from HrO re-leased from the talc sleeve.

All runs were conducted at 35 kbar for a durationof I hr. The high temperatures necessary to achieveanhydrous melting preclude longer run durations.An estimation of minimum distances of nnn diffu-sion calculated with difusion coefficients of aboutl0-rr cm2 sec-' at 1600'C for Rer in garnets (Harri-son and Wood, 1980) indicates that a run duration ofI hr is more than adequate to attain equitbrium inthe small crystals grown in these experiments (15-25trlm diameter). Reversal experiments for RBB datawould require that the individual minerals be synthe-sized containing calculated amounts of Ren and re-equilibrated with a REE-free melt. The melts undergoextensive crystallization on quenching from 35 kbar,and because of the problems inherent in electron mi-croprobe analysis of stable phases coexisting withquenched phases (e.g., Mysen and Kushiro, 1977\, noattempt has been made to analyze the major elementcompositions of minerals and melts. Independentsynthesis of phases for reversal experiments is there-fore precluded.

Results

Phase relations of nodule pnn 16II

The 35 kbar anhydrous melting relations of nodulePHN 16ll as a function of the weight percentage ofpaftial melting are shown in Figure l. Experimentaldetails are presented in Table l. Also shown, forcomparison, are melting relations of the same noduledetermined by Mysen and Kushiro (1977), withoutcontrol of iron loss.

The general features of the melting relations arecomparable in both cases. The present experimentsyield a lower estimated solidus temperature, l57S"C,compared with 1625'C estimated from the experi-ments of Mysen and Kushiro (1977). Three stablephase assemblages are observed (phase identificationis based on optical microscopy): olivine * ortho-pyroxene * clinopyroxene + garnet * liquid, olivine+ orthopyroxene + clinopyroxene * liquid, and oli-vine * orthopyroxene * liquid. A fourth phase as-semblage, olivine * liquid, is inferred from the ob-servations of Mysen and Kushiro (1977\. These

phase fields extend over different percentages ofmelting and are stable at systematically lower tem-peratures than the same phase fields in the systemwhere iron loss occurred.

The first phase to melt out with increasing temper-ature is garnet, which disappears within the initialll%o nelting (1580"-l590oC), whereas in the experi-ments of Mysen and Kushiro (1977\, garnet persistsover a melting interval of about 40"C (1620'-1660'C) before it disappears at 25Vo melting. Themuch reduced stability field of garnet in these experi-ments may be interpreted in terms of the lower tem-perature stability of an iron-bearing garnet relative toa pyrope-rich garnet.

The slope of the melting interval in which garnet isa liquidus phase is very nearly independent of tem-perature. Mysen and Kushiro (1977) and Wendlandtand Mysen (1978) have shown that melting in natu-ral systems, for example PHN 16ll, is analogous tomelting in synthetic systems such as CaO-MgO-AlrOr-SiOr. In the system cMAs, with four com-ponents and five phases (olivine, orthopyroxene,clinopyroxene, garnet, and liquid) the initial meltinginterval is isobarically invariant. The nodule puN16ll approaches this behavior much more closely(i.e.,Ihe melting curve is flatter) when Fe is retainqdin the system (Fig. l).'

The second phase to melt out is clinopyroxene.The melting interval in which the stable phase as-semblage is olivine * orthopyroxene + clinopyrox-ene + liquid extends from 9 Io 40Vo melting over atemperature range of 1590o to 1635"C. As in the gar-net in the lowest melting interval, the presence of Fereduces the stability of clinopyroxene (Kushiro,1968) so that it is no longer present at 4OVo melt com-pared with 45Vo melt in the experiments of Mysenand Kushiro (1977). The reduced stability of garnet

2 It is likely that the reduced melting temperatures observed in thepresent expcriments result from the r€tention of iron in the system.It is possible, however, that the presence ofvolatile species such asCO2 and CO (formed by reaction between the graphite capsuleand oxygen in the sealed platinum capsule) and perhaps H2O orCHa (formed by hydrogen diffusion into the capsules) may reducemelting temperatures. Generous estimates of 0.1 w.% of volatilespecies could result in the formation of a few tenths of a percent ofmelt, several hundred degrees below the observed solidus. Such asmall amount of melt would not be detectable and would have anegligible effect on melt compositions at higher degrees of melt-ing. Because the difference in melting temperatures in this studyand that of Mysen and Kushiro (1977) increases with degree ofrnelting, it is believed that better iron retention is the most satisfac-tory explanation for the lowered melting temperatures.

HARRISON: PART:ITIONING OF REE

| 800

| 750

r700

| 600

| 5 5 0

results in an extension of the stability of the phase as-semblage olivine + orthopyroxene + clinopyroxene+ liquid to lower temperatures (1590'C) and lowerpercentages of melt (9Vo) than in the experiments ofMysen and Kushiro (1977).

The steeper slope of the interval otvine * ortho-pyroxene * clinopyroxene + liquid may indicate iso-baric univariant melting behavior (four significantcomponents and four phases, as discussed by Mysenand Kushiro, 1977). At temperatures greater thanl640oC, orthopyroxene and olivine are present in themelting interval and are stable to at least 70Vo meltand 1660'C. The temperature above which olivinealone remains on the liquidus is unknown. The pro-gressively steeper slope of the melting interval re-flects a further degree of freedom relative to the melt-ing interval for ol iv ine * orthopyroxene +clinopyroxene * liquid, as a consequence of the lossof another phase (four significant components andthree phases).

On the basis of the redetermined melting relationsof nodule pHN 1611, four melt percentages were se-lected for Rnn studies: 2.3Vo melt (1580'C), 8.0Vo melt(1585'C), 20Vo melt (1620'C), and 37.7Vo melt(1635'C). The first two melts coexist with the min-

. ' \ 35 kbo r , d r y-Mysen ond Kush i ro (1977

p lo t i num copsu les

o l + o p x + c p x + l i q

o l+ opx + cpx + g t + l i q

6 5 0

L)o

o)=ob lo.Eo)

t'-

6050403020ro500 Lo 70 80 90 tooP e r c e n t o g e m e l f

Fig. l. Melting curves of nodule PHN 16ll at 35 kbar pressure (anhydrous). Upper curve determined by Mysen and Kushiro (197?)with no control on iron loss to noble-metal capsules. Lower curve determined in this investigation with graphite capsules. Size ofsymbols includes uncertainties in temperature (tl0'C) and determination of percentage of liquid (+lo). Arrows indicate conditionsselected for partitioning experiments.

eral assemblage olivine + orthopyroxene * clinopy-roxene + garnet (Fig. l). As discussed above, themelting reaction of this assemblage is essentially iso-barically invariant so that, to a first approximation,partitioning behavior may be examined independ-ently of pressure (35 kbar), temperature (1580'-1585"C), and melt composition; i.e., as a function ofthe amount of melting alone.

The second two melts (20 and 37.1Vo) coexist withthe assemblage olivine + orthopyroxene + clinopy-roxene. Because this assemblage melts in an essen-tially isobaric, univariant fashion, partitioning of theREE may not be evaluated independently of temper-

Table l. Melting relations of nodule PHN 16ll at 35 kbar

P . T - R u n .

l u t r ,R u n N o , . . - : d u r a t r o n .

k b a r _ c n e I EMinera logy

1 6 1 1 - W 1 2 3 5 1 6 6 0 6 01611-W 6 35 1645 6016r1-W 5 35 1635 601611-W 4 35 1625 601611-W 8 35 1615 60

1611-W 9 35 1605 601611-W 10 35 1590 601611-W 25 35 1585 601611-W 11 35 1580 60

7 I . 0 4 ! 1 . 5 6 0 1 , o p x , 1 i q4 8 . 1 3 r 0 . 0 3 o 1 , o p x , l i q3 7 . 7 3 1 0 . 0 3 0 1 , o p x , c p x , I i q2 I . 3 2 r 0 . 0 3 o 1 , o p x , c p x , l i q1 8 . 6 9 l o . 4 2 0 1 , o p x , c p x , 1 i q

1 4 . 6 8 1 0 . 5 9 0 1 , o p x , c p x , l i q8 . 5 0 l O . 2 2 o I , o p x , c p x , l i q8 . 0 0 1 0 , 1 9 0 1 , o p x , c p x , g r , l i q2 . 2 9 1 0 , 0 4 0 1 , o p x , c p x , g t , f i q

246 HARRISON: PARTITIONING OF REE

Table 2. Rare earth element partitioning data

Run No, D

kbarT .

Rundura t ion , Wt %

min meltGarnet , *

pPn

c l ino- o r tho-pyroxene, pyroxene, O l iv ine ,

ppm ppn pPmLiqu id ,

ppm

1611-Ce 20

I6LL-Ce 2L

L6LL-Ce 27

L6LL-Ce 26

1611-Sn 22

1611 -Sn 23

l6I1-Sn 30

1611-Sn I

r611-Tn 13

1611-Tm 14

1611-Trn 28

161l-Tm 29

35

1580

1585

L620

1635

1580

15 85

1620

16 35

1580

t5 85

r620

16 35

2 . 2 9

8 . 0 0

20 . 00

3 7 . 7 3

2 . 2 9

8 . 0 0

2 0 . 0 0

3 7 . 7 3

2 . 2 9

8 . 0 0

20 . 00

3 7 . 7 3

0 . 4 1 7+0. 016

0 . 1 7 3+0. 010

4 . 7 3 2+0. 082

3 . 7 0 4+o.092

L . 7 9 4+0. 063

L . 6 2 5+o ,022

tL.337+0 . 210

4 . 7 0 5+0 . 113

3 .614+ o . o 7 2

3 . 1 3 6+0 .113

4 . 0 3 6+0 .L42

3 . 569+Q .O7 4

2 . 4 6 6+0 . 064

t . 6 3 4+0 .039

o . 2 9 6+0. 009

o .289+0. 005

0 . 2 6 0+0. 007

0 . 2 4 3+0. 005

2 . 7 9 7+0 .L92

L .O92+0.o29

0 . 8 2 6+0. 043

n , d . * *

0 . 6 7 2+0 .026

0 . 6 3 6+0 .032

o . 4 5 2+0. 021

0 . 3 r4+0. 007

0 . 0 6 8+0 .0002

0 . 0 6 1+0 .0006

0 . 0 5 2+0 .0006

o . 0 4 6+0 .0002

1 .550 55 .L69+0 .020 +0 .953

o . 6 2 r L 9 . 6 4 9+0 .012 +0 .521

0 . 5 1 8 1 1 . 5 1 6+0 .049 +0 . I 28

0 . 3 l r 6 . 1 5 1+ 0 . 0 1 1 + 0 . 1 3 9

o . 4 7 9 1 6 . 1 4 8+0 .031 +0 .194

0 . 384 11 . 538+0 .024 +0 .242

0 . 2 9 4 6 . 3 2 1+0 .014 +0 .290

0 . 2 2 2 3 . 1 8 3+0 .019 +0 .018

0 . 0 5 0 1 . 3 7 0+0 .001 +0 .043

0 .045 1 281+0 .001 +0 .035

0 . 0 3 7 0 . 8 8 1 5+0 .001 +0 .033

0 .036 0 .6L94+0 .0007 +0 .010

35

35

35

35

35

35

35

ature and compositional variables. The temperaturedifference between 20 and 37.7Vo melt is only l5oC,however, and can be realistically ignored.

A very low percentage of melt (e.g., A.SVo) was notused in partitioning experiments, even though such amelt would be more enriched in rRre than a meltproduced by 2Vo melting. The melting interval imme-diately above the solidus will not show isobaricallyinvariant behavior because pHN 16ll contains a fewtenths of a percent of alkalies and these preferentiallyenter the first formed melt, having high activities inthis melt. As a result, additional degrees of freedom(> four components and five phases) are probable inthe initial melts. The presence of alkalies results in asteepening of the melting slope imnediately abovethe solidus (Mysen and Kushiro, 1977), and the low-ered temperatures, in addition to the high activitiesof alkalies in the melts, will have a significant efecton partitioning of REE. Because the melting curve hasvery slight temperature dependence between 2 and8Vo melt (Fig. l), it is assumed that the concentration

of alkalies is dilute relative to that of the major ele-ments in these melts and that the natural system canthus be modeled by a four-component system.

Rare earth element partitioningPartition coemcients and REE concentrations in

phases are presented in Table 2. The concentrationsgiven in each phase have been calculated to includethe stable isotope present in the starting material(Shimizu, 1975) as well as the radioactive spikeadded.

Garnet-melt partition coeficients. Partition coeffi-cients for Ce, Sm, and Tm between garnet and meltare shown in Figure 2A: D..: 0.008+0.003 at both2.3 ands%omelt, D.^:0.293t0.009 at2.3Vo melt and0.321+0.014 Lt 8Eo melt, and D'- : 1.309+0.09 at2.3Vo melt and 1.26810.052 at 8Vo melt. Also shown,for comparison, in Figure 2B, are some other deter-minations of DE*'z*"t' for similar pressures, temper-atures, and compositions. In general, the agreement


Table 2 (continued)

Run No. Dgar / Ltq

,cpx/ l iq

,opx / t iq oor/Liq Calcu la tedEF

1611-Ce 20

1611-Ce 21

I6LL-Ce 27

1611-Ce 26

1611-Sm 22

1611 -Sm 23

1611-Sm 30

1611-Sn I

1611-tn 13

1611-Tm 14

1611-Tm 28

r611-Tn 29

0 . 0 0 8 7+0 .0003

0 .0088+0 .0005

o . 2 9 3+0. 009

o .32L+0. 014

1 . 309+0 .090

1 . 2 6 8+0 . 052

o .206+0. 005

0 . 2 3 9+0. 009

0 . 314+0. 010

0 . 5 1 0+o.o22

0 . 2 5 0+0. 009

0 . 309+0. 010

0 . 3 9 0+0. 021

0 . 5 1 3+0. 032

o .2 t6+0. 006

o . 2 2 6+0. 007

0 .295+0. 013

0 . 310+0. 008

0 . 051+0 .004

0 . 056+0. 005

0 . 0 7 2+0. 004

0 . 0 4 2+0. 002

0 . 055+0 .004

0 . 0 7 r+0. 004

0 . 0 9 9+0. 003

0 . 050+0 .002

0 . 0 4 8+0 .005

0 . 059+0. 005

0 . 0 5 9+0. 002

0 . 0 2 8+0. 001

0 . 031+0. 001

0 . 045+0 .001

0 . 049+0. 002

0 . 0 3 0+0. 002

0 . 0 3 3+0. 0003

0 .047+0 .004

0 . 0 7 0+0 .006

0 . 0 3 7+0. 001

0 . 0 3 5+0. 002

o . 0 4 2+0 .003

0 . 0 4 6+0. 002

2 2 . I 4

7 . 9 3

4 . 6 2

2 . 4 7

r0 . 68

7 . 6 3

4 . 1 8

2 , r 0

5 . 3 0

4 . 9 5

3 . 4 r

2 . 4 0

1 0 . 5

6 . 8 5

5 . 5

4 . 9

15

xconcentrations cafcuTaXed to include BlEnrictnent factor = (radioactive REE in

**Nox detexmined.

isotope + sXabTe isoXotrreliquid) / (radioactive REE

( f rom data o f Sh imizu , 1975) .L t t - t e L t l t t t i l r t ^ j .

between investigations is good for light and inter-mediate REE. Shimizu and Kushiro (1975) found D..: 0.021 and Ds* : 0.217 for a garnet of compositionpyropessgrossularrs at 30 kbar and, l275oc. Wood(1976) found D"- : 0.197t0.013 in a duplication ofthe experiments of Shimizu and Kushiro (1975). Har-rison (1977, 1978, 1979b) and Harrison and Wood(1980) have observed that crystal trace element con-tent may control Dfig;'tu"tt within certain limi15 a1dthat the Ca2* content of pyrope garnets is more im-portant in controlling the magnitude of DEgg*z*.t,than small changes in melt composition.

Partition coeftcients for gRrB are much more in-fluenced by garnet trace element content and garnetcalcium content than the light and intermediate nre.Partition coemcients for Ce, Sm, and Tm between agarnet selected from a nodule from kimberlite, pHN1925 (Nixon and Boyd, 1973), and a coexisting meltat l300oC and 30 kbar are shown in Figure 28 (Har-

rison, 1978, 1979b) to illustrate this point. When 50ppm REE are contained in the garnet (pyropero), Dr_: 0.88. The partition coemcient increases to 1.5 whenthe garnet contains 20 ppm REE and to 2.5 when thegarnet cont4ins l0 ppm REE. The increasing valuesare a consequence of non-Henry's law behavior ofREE in garnets at low trace element concentrations(Harrison, 1977, 1978). Partition coefficients deter-mined in this study and those determined by Shimizgand Kushiro (1975) and Harrison (1978, 1979b) forgarnet crystals ssil4ining 20 ppm REE are in agree-ment.

Values of DgE["z-d' have been shown to be sub-stantially greater when the melt composition is moresilicic (see data and discussions of Irving and Frey,197 7, 197 8; Irving, 1 97 8; Mysen, 197 8a); however, forthe basic melts produced in these experiments(Mysen and Kushiro, 1977) it is believed thar thepartition coemcients determined in this study are re-

248 HARNSON: PARTITIONING OF REE

o.o4

o.oo

GARNET/L IOUID

2o/o mell(35kbor , l58OoC)

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

Fig. 2A. Partition coemcients for Ce, Sm, and Tm betweengarnet and melt produced by 2.3 and 8% partial melt of PHN 16l 1.Error bars alo.

alistic values when garnet compositions and traceelement concentrations are taken into consideration.

Clinopyroxene-melt partition coefTcients. Partitioncoemcients for Ce, Sm, and Tm between clinopyrox-ene and a melt produced by 2.3, 8, 20, and 37.7Vomelting are shown in Figure 3. Values of Dc"increasefrom 0.206+0.005 at 2.3Vo melt to 0.510*0.022 at37.l%o melt, D"- increases from 0.250+0.009 at 2.3%melt to 0.513+0.032 at37.1Vo melt, and D.- increasesfrom 0.216+0.006 at 2.3Vo nelt to 0.310+0.008 at37;lVo melt.

These values are comparable with data from othersystems (for review see Cullers et al., 1973; Irving,1978; Mysen, 1978a). Values of Dg'z'a' fall in therange 0.05-0.4; D"^, 0.09-0.7; Dr-, 0.04-0.4 for ba-

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

Fig. 28. Partition coefficients determined in this study (Fig. 2A)compared with other experimentally determined values of

Df;;;o'-ot. Values for garnet conposition pyropeee with l0' 20'

and 50 ppm nnr in the garnet at 1300"C and 30 kbar are fromHarrison (1979b). Thc data indicated by dashed lines are from

Shimizu and Kushiro (1975).

saltic-andesitic melts (the higher values are associ-ated with more acidic melts). The melts that coexistwith the clinopyroxenes in this study are probably al-kali picritic (up to 9Vo melt) and olivine tholeiitic (10-40Vo melt\ (Mysen and Kushiro, 1977), so it appearsthat the values of Dio;f-"' obtained here are some-what higber than would be expected. Seitz (1974) de-termined ftn-l"tt : about 0.09 at 20 kbar in equilib-rium with a haplobasaltic melt in the system An-Ab-Di. Mysen (1978a) found 4oil/"" :0.29 at 15 kbarand 1450'C for clinopyroxene on the liquidus ofKilauea l92l basalt. Grutzeck et al. (1974) deter-

o.8

o.

9-q

0 u J(,^E o.08

o.o6

GARNET/MELT


mined D[ionsidc'zmett : 0.10, Dr- : 0.26, and Dr. : 0.30aI l265oc and I atm for diopside crystals coexistingwith a glass in the system Abs-An3o-DLs.

There are, however, no determinations of Digl-.uat high pressures with which to compare these data.Drake and Holloway (1978) noted an apparent pres-sure dependence of ftP*/"tt in a hydrous, Fe-freesynthetic andesitic composition; D"- is observed toincrease from 0.75 at 2 kbar and 1050'C to 0.85 at 5kbar and l050oC, and D"- is found to be 2.68 at 12kbar and 900'C in a natural andesite. Some of theincrease may be related to compositional differences,however. If a similar pressure dependence does occurin these experiments, where crystals coexist with an-hydrous basaltic melts, then it is not surprising that

the observed partition coefficients are much higherthan those determined in lower pressure experimentson basaltic compositions.

Clinopyroxene-garnet ratios. Ratios of nEs in clino-pyroxene to REE in garnet are as follows: For 2.3Vomelt, Dgre"'"" : 23.'75, fr'/eun't: 0.85, fix/eanet -0.16. For 8qo melt, D\^re" 't:27.28, rffi*/e"n'r: 0.96,PP*te" " ' : 0.18.

The ratios are shown in Figure 4 with a range ofvalues calculated from nBn analyses of peridotitenodules for comparison. The latter data have beenaccumulated from the analyses of Haskin el a/.(1966), Nagasawa et al. (1969), Reid and Frey (1971),Philpotts et al. (1972} Ridley and Dawson (1975),Shimizu (1975), and Mitchell and Carswell (1976).

r .o

o.8

0.9

o.7

0.6

0 .3

o.?

C L I N O P Y R O X E N E / L I Q U I D

20 kbor , 95O"C, Mysen (1978o)hoploondesi t ic mel l + d iopside

A/i . {

/ t . .

\\

\

o, " \ l o tm, l25O'C, Gru tzeck e t o t , (1974)

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

E'

fct

) l+ o.5i iao

o . l

o LLo

o.4

Fig' 3' Partition coefficients for Ce, Sm, and Tm between clinopyroxene and melts produced by 2, 8,20, and 37.7vo partial melting.Error bars are +lo' Also shown are data of Grutzeck et at. (1974) for diopside coexisting with a melt in the system Ab-An-Di at I atmand l2-50'C and data ofMysen (1978a) for diopside coexisting with a hailoandesitic melt at 20 kbar and 950.C. The triangle denotes avalue for Ds- at 15 kbar, l450oC, for diopside coexisting with a basaltic melt determined by Mysen.

250 HARRISON: PARTITIONING OF REE

0 .01

Fig. 4. Clinopyroxene/garnet nnE ratios for 2 .3 and 8Vo melt. A range of values calculated for nre analyses of peridotite nodules is also

shown (see text).

o

o!

a u td c r

The experimentally determined values in this studyare in agreement with ratios from natural systems.

Orthopyroxene-melt partition coefrcients. Figure 5shows the values obtained for DinJ/"" as a functionof the amount of melting. There is no analysis forD.*;t^"t' at 37.7Vo melt. The partition coefficients in-crease with degree of melting. Values of Dc. increasefrom 0.051+0.004 at 2.3Vo melt to 0.072+0.004 at20Vomelt, D"- increases from 0.042+0.002 at 2.3Vo melt to0.099+0.003 at 37.7Vo melt, and D'* incteases from0.050+0.002 at 2.3vo melt to 0.059+0.002 at 37;|Vomelt. Except at 37.7Vo melt, there is relatively littlefractionation of light or heavy REE by the crystal.The (Ml) site onto which Ree substitute is sub-stantially smaller than the REE ions (ionic radii, 1.09,1.04. and 0.96A in octahedral cooidination for Ce,

Sm, and Tm, respectively, and 0.80A for Mg; Whitta-ker and Muntus, 1970), and although thulium issmaller than cerium it is unlikely that it approachesthe Mg site size closely enough to have a substantialinfluence on D"*"u relative to Dr*"t. Fractionation ofthe REE at 37.7Vo melting is probably due to liquidcomposition (e.9., Mysen et al.,1979).

There is only one other investigation of partition-ing of REE at high pressure between orthopyroxeneand liquid. Mysen (1978a) investigated the behaviorofnsn at 10 and 20 kbar and 1025" and 1075'C. Themelt with which his orthopyroxene crystals coexistedwas haplobasaltic and contained ll-20 ,ttt.Vo HzO.Mysen's (1978a) data are shown in Figure 5 for com-parison and are in agreement with partition coeffi-cients determined in this study at low percentages of

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

o.09

o.08

o.07

o.06

melting. As there are differences in pressure, temper-ature, and composition (particularly HrO content ofthe melt) between the two studies, the fact that thesedata are comparable must be fortuitous. Partitioningdata at lower pressures between orthopyroxenes andmelts and vapors have been reviewed by Cullers et a/.(1973),Irving (1978), and Mysen (1978a). This studyprovides the first high-pressure data (35 kbar) onpartition coefficients for nns between orthopyroxenesand melts; because the effect of pressure has not beenevaluated, comparison with lower pressure data maynot be meaningful.

Olivine-melt partition coeftcients. Partition coeffi-cients for Ce, Sm, and Tm between olivine and meltas a function of the amount of melting are shown inFigure 6. The value of D.. increases from0.028+0.001 at2.3Vo melting to 0.049+0.002 at 37.7Vo


O R T H O P Y R O X E N E / L I Q U I O

37.7olo

20"h

lOzsoC, 20 kbor ,hop loboso l f i c met i (Mysen, t978o)

251

o. l

!t5cr

o,co,xo

CIoG

o

; o.o5)I): Ho.04i Eo

o.03

o.o2

o.or

oLo Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm yb Lu

Fig. 5' Orthopyroxene/melt partition coemcients for Ce, Sm, and Tm at2.3,8,20, atd37.7Vo partial melting. Error bars are tlo. Alsoshown are data ofMysen (1978a) in the system plagioclase62-forsteritel5-silica2 at 20 kbar and 1025.C.

melting, D.^ increases from 0.030+0.002 zt 2.3Vomelting to 0.070+0.006 at37.7Vo melting, and Dr- in-creases from 0.037t0.001 at 2.3Vo melt ing to0.046i0.002 at 37 .7Vo melting. The values of D^r, areclose to those determined by Mysen (1978a). For 20and 37.7Vo melt, DREE are substantiarlly highsl thanthose determined by Mysen (1978a). As discussedabove, Mysen's data are obtained at 20 kbar com-pared with 35 kbar in this study, and the crystalscoexist with hydrous melts (20 wt.Vo HrO). The dataobtained at low pressure are reviewed by lrving(1978) and Mysen (1978a), and phenocryst-matrixdata are summarized by Cullers et al. (1973). Thevalues determined in this study are among ft6 high-est of any olivine-basaltic melt REE partition coeffi-cients and are also determined under the highsstpressure and temperature conditions.


O L I V I N E / L I Q U I D

\ ,oru.a, 20 kbor,hop loboso l l i c me l l (MYsen, 1978o)

6 l t t I t t t t t t t t ' ' '-Lo

Ce Pr N<l Pm Sm Eu Gd Tb Dy Ho Er Tm Yb Lu

Fig. 6. Olivine,/melt partition coefrcients for Ce, Sm, and Tm at 2.3, 8,20, and 37 .7Vo partial m€lting. Also shown, for compariso4 are

the data of Mysen (1978a) for fonterite coexisting with a haplobasaltic, hydrous melt at 20 kbar and 1075'C.

o. ro

o.o9

o.o8

o.o7

o.o6

!,

tc,

o,c';14

= UoEo o.o4

o.03

o.o2

o.o I

Liquids. In Figure 7 the enrichment of radioactiveCe, Sm, and Tm in the melts relative to the initialspike is depicted; this experinentally determined en-richment factor, EF (Table 2), is equivalent to thechondrite-nonnalized REE pattern of partial melts ofthe nodule (Mysen and Holloway,1977).

Also shown in Figure 7 are calculated enrichmentsof nne in 2.3 and 8Vo partial melts. The equation ofShaw (1970) for equilibrium partial melting (which isthe case defined by these experiments) was used. Par-tition coefficients used in the calculations were thosedetermined in this study (Table 2). The real phaseproportions contributing to the melt are unknown,but as a first approximation the reaction stoichio-metry for invariant melting in the system CaO-MgO-AlrOr-SiOu may be used. Mysen (1977a) hascalculated the following reaction stoichiometry based

on mineral analyses and calculated liquid composi-tions in the system cMAs at high pressure:

garnet + 0.67 diopside + 0.14 enstatite

=0.22 forsterite + l.6l liquid.

The initial modal composition of the nodulewas l07ogar:rret, ll%o chnopyroxene, 20Vo otthopy r oxenLa, 6AVoolivine (Nixon and Boyd, 1973; Shimi2u, 1975). Thecalculated enrichments of Ce, Sm, and Tm in the2.3Vo melt are, respectively, 15, 10.5, and 5.5 (CelTm:2.7) compared with the experimentally determ.inedenrichments of 22,10.7, and 5.3 (Ce/Tm:4.15).

The lower value for cerium enrichment calculatedfrom the reaction stoichiometry of the iron-free, syn-thetic system implies that more garnet must be enter-ing the melt relative to clinopyroxene in the natural


1.3"/oa _ _ _ _ _ _ _ _ _ _ a _ _

_ _ _ _ _ _ -

253

a - - - - a

/F-<

x-. - . -x

M y s e n o n d H o l l o w o y ( 1 9 7 7 )

Th is work , exper imen lo l vo luesTh is work , mode l vo lues

2.3%m'el l

oIo

E I OoFE 8o

clrl 6

8 o/o m€-lf

l

Fig. 7. Enrichment factors of Ce, Sm, and Tm in melts produced by 2.3, 8,20, and 37.7vo paftial melting of pHN 16l I at 35 kbar,denoted by triangles. Also shown are the emichment factors for 2.3 and 8vo paftiil melts calculated for equilibrium partial meltingdenoted by crosses (see text). Data from Mysen and Holloway (1977) are indicated by dashed lines; enrichment factors for 1.3, 15, 35,and 50Vo partial m€lts at 20 kbar for pnN 16ll are shown.

system; i.e., the presence of Fe has a greater effect inreducing the stabitty of garnet than it does in reduc-ing the stability of clinopyroxene. Alternatively, theinitial modal mineralogy may not be l}Vo clinopyrox-ene and lOVo garnet as estimated by Nixon and Boyd(1973) and Shinizu (1975) but would have garner inexcess of lOVo and clinopyroxene less than 1070.3 Thefirst alternative is probably more reasonable. Of note,however, is the close agreement that exists betweenthe, experimentally determined patterns of nBn en-richment and the calculated values. This agreementlends support both to the application of an invariant-

3 The use of other estimates of the modal mineralogy of ruN16ll (e.g., Boyd and McCallister, 1976) does not improve the fit ofthe data significantly.

type melting interval when garnet is present and tothe application of the melting reaction stoichiometrydetermined in crvres by Mysen (1977a) to more com-plex, natural systems.

Thle 8Vo partial melt has a less fractionated pattern.Experimentally determined enrichment factors forCe, Sm, and Tm are, respectively, 7.9, 7.6, and, 4.95(Ce/Tm: 1.6).Cerium enrichment is controlled bythe percentage of melt present, i.e., the cerium con-tent at 8Vo melt is diluted relative to 2.3Vo melt.whereas thulium enrichment is controlled by the gar-netlmelt partition coeftcients and remains relativelyconstant in the garnet melting interval. The lowervalue of Ce/Tm at 8Vo melt relative to 2.3Vo melttherefore reflects the increased percentage of meltand the lowering ratio of garnet/cltnopyroxene. The

_ _ _ . a

2O7"melf

rz.zzJrr

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


calculated REE enrichment in an 87o panial melt is inclose agreement with the experimentally determinedpattern. It is to be expected that most deviation fromthe ideal behavior predicted from the CaO-MgO-AlrO3-SiO, reaction stoichiometry would be ob-tained at the lowest amounts of melting as initialmelts will be relatively alkali and iron rich.

Between 20 and 37.7Vo meltrng, clinopyroxene isprogressively removed. It is not surprising that the20Vo melt is slightly tner enriched, Ce/Tm : 1.34,whereas the 37.1Vo melt, close to the limit of clinopy-roxene stability, has CelTm = 1.0. The difference innEn enrichment in the two percentages of melting re-flects the dilution produced by increasing the amountof melt because neither orthopyroxene nor olivinefractionates REE signifi sanlly (Dioi/-"' and DiYB"' areboth small).

No theoretical calculations of nnE enrichment arepossible at20 and37.7Vo meltrng because the meltingreaction is unknown. Quench problems preclude mi-croprobe analysis of phase compositions from whichthe melting reaction could be estimated.

Also shown in Figure 7 are the nen enrichmentpatterns of 1.3, 15, 35, and 5OVo partial melts of nod-ule pHN 16ll at 20 kbar determined by Mysen andHolloway (1977). At 20 kbar, garnet is replaced byspinel; consequently the heavy rare earth depletion inlow percentage melts is not observed. The l.3%o melthas CelTm : 1.3 compared with Ce/Tm: 4.2 inthis study at2.3Vo melting although the total REE en-richment is higher at 20 kbar. The two sets of dataare in close agreement at 35 and 37.77o melt. At 20kbar and 35Vo melt no clinopyroxene is melting; at 35kbar and 37.77o melt, clinopyroxene is at the end ofits stability field. In view of the small pressure depen-dence of olivine/melt and orthopyroxene/melt parti-tion coefficients for REE (Mysen, 1978a), the agree-ment between the data in this paper and those ofMysen and Holloway (1977) implies that the meltcompositions at 20 and 35 kbar cannot be very differ-ent from each othero a conclusion also reached byMysen and Kushiro (1977).

Discussion

Partition coeficientsfor REE between Sarnet lherzolite

minerals and melts

The partition coefficients determined in this work

for three nEE between olivine, orthopyroxene, clino-

pyroxene, garnet, and various peroentages of melting

at high pressures and temperatures provide a coher-ent set of data that is complementary to the studies insynthetic systems by Mysen (1978a). Values of D*.'are, in general, close to those determined at similarpr€ssures and temperatures in synthetic systems andfall within the much broader range of values definedfrom analyses of natural systems.

Partition coefficients vary as a function of degreeof partial melting of garnet lherzolite. This relation-ship is predictable in view of the changing melt com-positions (Mysen and Kushiro,1977; Wendlandt andMysen, 1978), incre6ing temperatures (1580"C at2.3Vo melt to l635oC and37.7Vo melt), and changingtrace element concentrations in all phases. Evidentlythe assumption made in most petrogenetic modelsthat Di[td/-'r' are constant is unrealistic. In the ab-sence of more comprehensive partitioning data, mod-eling is constrained to this assumption. It is impor-tant to note that the variation in ffij"t/-'r' withthe amount of melting is frequently negligible, e.9.,Dltnct/nett (Fig. 2A), fswm& between 2.3 and 8Vo melt(Fig. 3), ft!:{*'" between 2.3 and 8Eo melt, D1o'r^"trbetween 20 and 37.7Vo melt'(Fig. 5), ry#''- between2 and 8Vo nelt, and D;tz-a' between 2O and 37.7Vomelt (Fig. 6). For some REE and some minerals theassumption that D*.r are constant during magmageneration and evolution may be realistic.

Partition coefrcients at 2.3 and 8Vo melting. Between2.3 and 8Vo melting, both temperature (1580o andl585oc, respectively) and melt composition (meltingis invariant-like at constant pressure) are essentiallyconstant. It has been observed that .ffi"ifv"" may in-crease between the two melting fractions, however.Two variables may contribute to this increase. First,the FelMg in crystals and melts changes, the ratiofalling as the amount of melting is increased (seedata of Mysen and Kushiro, 1977). Simple site-sizeconsiderations indicate that REE should substitutemore readily onto an Fe-bearing site than an Mg-bearing site. Considering the influence of crystalstructure alone, it is to be expected that as meltingproceeds and the minerals become more Mg-rich

ft;";"tz'"u will decrease. The reverse trend is ob-served; however, the melt Fe/Mg decreases at a fas-ter rate than the crystal ratio, and the effect of meltcomposition on D*"r cannot be evaluated. Second,the nnn concentrations in crystals and melts decreaseas the amount of melt increase (Table 2). If Henry'slaw is not obeyed, DEE";"tz-"tt will increase as the nBeconcentrations decrease. Mysen (1978a) has observedthat Henry's law is obeyed only at very low concen-trations of nrB in olivines and orthopyroxenes (< l-2


ppm), and Harrison (1978) has found that Henry'slaw is not obeyed for REE concentrations < 10-20ppm in clinopyroxenes and garnets. It is suggestedthat the observed increases h D"r. between 2.3 and8Vo melting are due to non-Henry's law behavior ofREE. Partition coefficients for thulium are essentiallyconstant, and it is suggested that as a result ofthe lowconcentrations in the crystals (Table 2), Henry's lawbehavior is obtained for this element. The increase inD3;^av-d' is not greatly in excess of experimental er-ror, and the assumption of constant partition coeffi-cients in model calculations should not result in sig-nificant error. Partition coefficients for samariumshow a signifisanl increase as the melting event pro-ceeds; evidently the Sm concentrations in the phasesare outside the limits of Henry's law.

Partition coeficients at 20 and 37.77o melting. Val-ues of Dffg';tz-"tt vary much more between 20 and37.7Vo melting than they do at the low percentages ofmelt. Unlike the lower melting interval, the variationin melt composition and, to a lesser extent, temper-ature (16200-1635'C) is signifimnl, and so the twosets of partition coefficients cannot be rigorouslycompared either with each other or with the lowerpercentage melt experiments.

Temperature may be considered an insignifcantvariable because of the small melting interval in-volved (15'C). Values of D*"" in these experimentsappear to increase with increasing temperature al-though there is some consensus that D*.. decreasewith increasing temperature (e.g., Irving, 1978;Mysen, 1978a). This increase must be interpreted interms of chatrging melt and mineral compositions.Because melt compositions in these experiments arenot characterized, no further discussion can usefullybe made. Mysen and Kushiro (1977) suggested thatthe melt that coexists with the phases olivine + or-thopyroxene + clinopyroxene at 35 kbar is olivinetholeiitic (10-39Vo melt). Ir is evident from Figures 2-6 that the difference in some partition coefficients be-tween 20 and3SVo melt is not so great as that between8 and 20Vo melt; D;o;/-"', fttlrTl', &rrd D;mz** do nothave significantly different values between 20 and,37.1Vo nelt. In view of the well-documented depen-dence of partition coefficients on melt composition(Watson, 1976; Indng and Frey, 1978; Ryerson andHess, 1978), it is concluded that the melt compositionbetween 20 and 37.7Vo melttlLg probably is relativelyconstant. The bigger diference in partition coeffi-cients between 8 and 20Vo melt is related to a majorchange in melt composition that results from the lossof garnet as a liquidus phase at 9Vo melttng.

Application of these partition coeficients to othersystems

The determination of DREE for natural, garnet-lherzolite minerals and coexisting melts at high pres-sure and temperature in a normal REE concentrationrange provides a set ofdata that should be applicableto many problems of magma origin and evolution.The principal lirnitation of these data is that theypertain to a volatile-free system.

Anhydrous melting of peridotite undoubtedly pro-duces many continental and oceanic basalts (Kush-iro, 1972; Yoder, 1976; Mysen and Kushiro, 1977;Presnall et al.,1978), and for such cases the partitioncoefficients determined here may be applicable whensuitable corrections for pressure and temperature dif-ferences are made.

Melting in the mantle in the presence of volatilessuch as CO, and HrO may generate more unusualmagmas at low percentages of melt (e.g., kimberlites,carbonatites, ultra-alkalic basalts; Eggler and Hol-loway, 1977; Mysen and Boettcher, l975a,b; Wend-landt and Mysen, 1978; Wendlandt and Eggler,l980a,b). The presence of CO, and HrO in themagma sourc€ region will place serious lirnitationson the use of these partitioning data; the volatileschange melt structures (e.9., Mysen, 1977b) andphase relationships (Eggler, 1978; Mysen and Boett-cher, l975a,b) and reduce melting temperature(Mysen and Boettcher, 7975a,b; Wendlandt andMysen, 1978). Watson (1976, 1977) arrd Mysen andKushiro (1978) have shown that changes in meltstructure have a strong influence on partition coeffi-cients. For small amounts of melting where liquidsmay be volatile-saturated (Eggler and Holloway,1977) and alkali-rich (Mysen and Kushiro, 1977;Eggler, 1974,1977), the strong preference ofnnn forboth HrO and CO, vapor phases relative to mineralsand melts (Mysen, 1979; Wendlandt and Harrison,1979) and the influence of alkalies on REE solubilitiesin melts (Robinson, l974a,b) place limitations on theuse of the data obtained here.

At high percentages of melting, volatiles are di-luted in the melts and their effects on partitioningmay be much reduced. The data obtained here, whencorrected for the reduced melting temperatures of avolatile-bearing peridotite (Mysen and Boettcher,l975a,b; Wendlandt and Mysen, 1978), may thus beapplicable to high percentages of melting of such aperidotite.

The effect of temperature on partition coefrcientsis not well defined; experimental studies of temper-ature dependency invariably involve changes in an-


other parameter (Weill and McKay,1975; Harrison,1977; Leeman and Lindstrom, 1978). Watson (1977)determined that Dilf"' decreases with increasingtemperature and has evaluated this effect independ-ently of melt compositional changes. Becausechanges in temperature of melting in this nodule willprobably produce changes in melt composition also,it is impossible to predict how the partition coeffi-cients determined in these anhydrous experimentswould vary under different temperature regimes.

Implicationsfor the genesis of alkali basalts

It has been suggested that alkali basalts may begenerated by only a few percent of partial msfting ofa garnet-bearing source, owing to their enrichment inLREE (Gast, 1968; Kay and Gast, 1973; Shimizu andArculus, 1975). Alkali basalts have chondrite-nor-malized Ce/Tm of 2-15 and tnnp chondrite-normal-ized abundances of 40-2N (Schilling and Winches-te11969; Kay and Gast, 1973; Shimizu and Arculuso1975; Frey et al.,1978). Values of CelTm determinedin these experiments are 4.18 (2.3Vo nelt), 1.60 (8Vomelt), 1.35 (2ovo melt), and 1.03 (37.7Vo melt), andrRBn chondrite-normalized abundances are < 22.Evidently, the LREE enrichment typical of alkali ba-salts cannot be generated even by very small degreesof anhydrous partial melting of a garnet lherzolitesource with chondritic np,B abundances, on the basisof the experiments presented in this paper. Two pos-sible explanations arise. First, the crystaVmelt parti-tion coefficients, in particular, garnet/melt partitioncoefficients for HREE, used i1 66dsling alkalibasalt genesis need modifying (e.g., Kay andGast, 1973; Shimizu and Arculus, 1975). Investiga-tions of .Fffi"/*"r'have shown that order-of-magnitudevariation may result from pressure, temperature, and

Table 3. REE enrichme*. i;lfft

from data of Morgan el a/.

phase composition changes (Shimizu and Kushiro,1975; Harrison, 1978; Irving and Frey, 1978). Be-cause the melt REE pattern is principally controlledby the value of DggS'tz-dt for small amounts of melt-ing, it is crucial that D*r", and in particular D"""r,should be closely constrained in petrogenetic models.

Second, the mantle source region for alkali basaltsmay not have chondritic nen abundances but may beI-nBB enriched. The nBB enrichments calculated inthis paper are equivalent to those that would be ob-tained if the nBB pattern for nodule pHN 16ll ischondritic; Shimizu (1975) derived such a pattern byanalysis of garnet and clinopyroxene. An initial nnnpattern of 3-5 times LREE enrichment and 1.5-2times HREE enrichnent would yield nnr abundancestypical for alkalic magmas if the partition coefficientsdetermined in this work and reaction stoichiometryof Mysen (1971a) are used. A mantle nen enrichmentof this magnitude, also predicted from the theoreticalanalysis of Minster and Alldgre (1978) andFrcy et al.(1978) and isotopic studies by Menzies and Murthy(1980), may be produced by a metasomatic event in-volving, for example, a COr-HrO fluid phase(Mysen, 1979; Wendlandt and Harrison, 1979).

If psN 16l I has undergone metasomatism (as sug-gested by its KrO content of 0.14 wt.Vo), and inter-granular deposition ofnnB has occurred, a REE anal-ysis by phase separation will yield erroneous results.A recent whole-rock analysis by Morgan et al. (1980),reproduced in Table 3, shows that nodule PHN 16llis enriched in rneB 3.5 times relative to chondritesand enriched in Hnne 2 times relative to chondrites.The experimentally determined enrichment factorsmay be used to calculate the REE abundances thatcould be obtained from the enriched source (or, in-deed, any source). The new values for nodule pHN16l l (Table 3) are similar to those for alkali basalts.

Conclusions

l. Partial melting of a garnet lherzotte nodule,PHN 16l l, for which iron loss to noble-metal capsulesis avoided, occurs at temperatures at least 30oC lowerthan melting in the same nodule when iron loss is notavoided. Additionally, the phase field for garnet ex-tends to only ll%o melting, and the phase field fofclinopyroxene extends to only 39Vo melttng.

2. Partition coefficients for Ce, Sm, and Tm be-tween garnet, clinopyroxene, orthopyroxene, olivine,and a coexisting melt have been deterrrined for 2.3,8, 20, and 37.7Vo meltrng of the nodule. Crystal-meltpartition coefficients vary as a function of theamount of melting. Variations between 2 and 8Vo

Enr lchnent FacCorsREE, CI -norna l i zed+

Ppn* 2.32 neTt 8Z melt 202 nelt 37.7i( nel t

L a 0 . 8 9 0c e 2 . 2 5 0N d 1 , 6 5 0S m O . 4 7 0E u 0 , 1 7 3G d 0 , 0 6 5Tb 0. 100H o 0 . 1 3 0fr no daraY b 0 . 3 3 0l u 0 . 0 5 2

3 . 6 3 93 . 5 2 73 4833 . O 5 22 . 9 8 20 3r8o . 2 6 72 . 2 9 32 . 1 0 0 * *7-9992 . O 4 8

7 8 . 1

3 2 . 6

1 1 . 1

2 6 9

2 3 . 2

1 0 . 4

1 2 - A

7 . 2

4 . 7

6 - 4

5 . 0

*Mta o t Morgan e t a1 . ( f980) .

+Notnalizeil to cl average of Evensen et al- (1978)-**Estinateil value.

melting may be ascribed to non-Henry's law behav-ior of REE in the minerals as the melt compositionand temperature remain essentially constant. In-creasing partition coefficients from 8 to 37.7Eo meltreflect changes in melt composition, temperature(1585"-l635oC), and non-Henry's law behavior ofnrn. The total increas€ in Dfi6";tu^d, may be up to afasior of 2 ftom 2 to 37 .7Eo melt; however, some valuesof ffi;tz-d' do not change signifisan11y.

3. The assumption that .fiEtd/"r'remains constantduring partial melting (and, presumably, fractionalcrystallization) is an oversimplification. Within cer-tain melting intervals and for some REE, however,this assumption does not necessarily result in verypoorly constrained models of basalt petrogenesis. It isparticularly important that Df,gi"/-'r'be well defined.

4. The REE patterns of the melts investigated donot show the fractionation and LREE enrichment typ-ical of alkali basalts, and it is considered unlikelythat such magmas may be generated by small per-centages of melting of a garnet-bearing source withchondritic nEn abundances. A garnet-bearing sourcewith 3-5 times enrichnent in IREB and 1.5-2 timesenrichment in HREE relative to chondritic abun-dances would produce melts containing nnn abun-dances appropriate for alkali basalts. A metasomaticevent may produce the necessary nen enrichment ofthe mantle source region.

5. The REE patterns of 2.3 and 8Vo melts may bemodeled adequately with an equilibrium partialmelting model and a melting reaction that has thestoichiometry

sarnet + 0'67 dionside +{'lt;liJ#'f + r.6r riquid

in the system CaO-MgO-AlrO3-SiO, (Mysen, l77a),providing further con-firmation of the principle thatmelting behavior in natural systems may be modeledby simple systems.

Acknowledgments

The experiments described in this paper were com-pleted while the author was a predoctoral fellow atthe Geophysical Laboratory, Carnegie Institution ofWashington. The manuscript was prepared while theauthor was a visiting postdoctoral fellow at the Lunarand Planetary Institute, Houston, Texas, which is op-erated by the Universities Space Research Associa-tion under contract NASW-3389 with Nnse, and aNASA-NRC Research Associate at the Johnson SpaceCenter, Houston, Texas. The paper has benefitedfrom the critical reviews of Drs. A. A. Finnerty, A. J.

HARRISON: PARTITIONING OF REE 257

Irving, B. O. Mysen, D. Phelps, S. A. Rawson, R. F.Wendlandt, and H. S. Yoder, Jr. This paper is Lunarand Planetary Institute Contribution No. 425.

References

Boyd, F. R. and England, J. L. (1960) Apparatus for phase equi-librium measurements up to 50 kilobars and temperatures up to1750'C. Joumal of Geophysical Research, 65, 7 4l-7 48.

Boyd, F. R. and McCallister, R. H. (1976) Densities of fertile andsterile garnet peridotites. Geophysical Research Letters, 3, 509-512.

Cullers, R. L., Medaris, L. G., and Haskin, L. A. (1973) Experi-mental studies of the distribution of rare earths as trace ele-ments among silicate minerals and liquids and water. Geochi-mica et Cosmochimica Acta,37, 1499-1512.

Drake, M. J. and Holloway, J. R. (1978) Henry's law behavior ofSm in a natural plagioclase/melt system and importance of ex-perimental procedure. Geochimica et Cosmochimica Acta, 42,679-683.

Eggler, D. H. (1974) Efect of CO2 on the melting of peridotite.Camegie Institution of Washington Year Book, 73,215-224.

Eggler, D. H. (19'17) Calibration of a Pyrex solid-media assembly.Carnegie Institution of Washington Year Book, 76, 656-658.

Eggler, D. H. (1978) The effect of CO2 upon partial melting of pe-ridotite in the system Na2O-CaO-AI2O3-MgO-SiO2-CO2 to 35kb with an analysis of melting in a peridotite-CO2-H2O system.American Joumal of Science, 278,305-343.

Eggler, D. H. and Holloway, J. R. (1977) Partial melting of perid-otite in the pr€senc€ of H2O and CO2: Principles and review. InH. J. B. Dick, Ed., Magma Genesis, Oregon Department of Ge-ology and Mineral Industries Bulletin, 96, 15-36.

Evensen, N. M., Hamilton, P. J., and O'Nions, R. K. (1978) Rare-earth abundances in chondritic meteorites. Geochimica et Cos-mochimica Acta. 42. 1199-1212.

Frey, F. A., Green, D. H., and Roy, S. D. (1978) Integrated mod-els of basalt petrogenesis: A study of quartz tholeiites to olivinemelilitites from southeastern Australia utilizing geochemicaland experimental petrological data. Journal of Petrology, 19,463-513.

Gast, P. W. (1968) Trace element fractionation and the origin oftholeiitic and alkaline magma types. Geochimica et Cosmochi-mica Acta, 32, 1057-1086.

Grutzeck, M., Kridelbaugh, S., and Weill, D. (1974) The distribu-tion of Sr and REE b€tween diopside and silicate liquid. Geo-physical Research Letters, l, 273-275.

Hadidiacos, C. (1972) Temperature controller for high-pressureapparatus. Carnegie Institution of Washington Year Book, 71,62U622.

Harrison, W. J. (1977) An experimental study of the partitioningof samarium between garnet and liquid at high pressures(abstr.). In papers presented to the International Conference onExperimental Trace Element Geochemistry, Sedona, Arizona,1917, p.4142.

Harrison, W. J. (1978) Rare earth element partitioning betweengarnets, pyroxenes, and melts at low trace element concentra-tions. Carnegie Institution of Washington Year Book, 77, 682-689.

Harrison, W. J. (1979a) An experimental investigation of the par-titioning of a REE, Sm3+, between garnets and msl$ a1 highpressures and temperatures. In Progress in Experimental Petrol-

258 HARRISON: PART:ITIONING OF REE

ogy, Fourth Progress Report ofRcsearch Supported by lrnc, p.59-67.

Harrison, W. J. (1979b) An Experimental Investigation of the Par-titioning of Rare Earth Elements between Gamets and Liquidsat High Pressure and Temperature with Particular Reference toNon-Henry's Law Behavior. Ph.D. Thesis, University of Man-chester, England.

Harrison, W. J. and Wood, B. J. (1980) Partitioning of REE be-tw€en garnets and liquids: The role of defect-equilibria. Contri-butions to Mineralogy and Petrology, 72, 145-155.

Haskin, L. A., Frey, F. A., Schmirt, R. A., and Smith, R. H. (1966)Meteoritic, solar and tcrrestrial rare earth element distributions.Physics and Chemistry of the Eanh, '1,167-321.

Irving, A. J. (1978) A review of experimental studies of crystal/liq-uid trace element partitioning. Geochimica et CosmochimicaActa. 42.743-771.

Irving, A. J. and Frey, F. A. (1977) Experimental partitioning oftrace elements between garnet and hydrous acidic melt (abstr.).In Papers presented to the International Conference on Experi-mental Trace Element Geochemistry, Sedona, Arizona, p. 59-6 1 .

Irving, A. J. and Frey, F. A. (1978) Distribution of trace elementsbetw€en garnet megacrysts and host volcanic liquids of kimber-litic to rhyolitic compositions. Geochimica et CosmochimicaActa, 42,771-789.

Kay, R. W. and Gast, P. W. (1973) The rare earth content and ori-gin of alkali-rich basalts. Journal of Geology, 81, 653-682.

Kushiro, I. (1968) Synthesis and stability ofiron-free pigeonite inthe system MgSiO3-CaMgSirOu at high pressures. Carnegie In-stitution of Washington Year Book, 67, 80-83.

Kushiro, I. (1972) Partial melting of synthetic and natural perido-tites at high pressures. Camegie Institution of Wtshington YearBook, 71, 357-362.

Leeman, W. P. and Lindstrom, D. J. (1978) Partitioning of Ni2*between basaltic and synthetic melts and olivines-an experi-mental study. Geochimica et Cosmochimica Acta, 42,817-833.

Menzies, M. and Murthy, V. R. (1980) Nd and Sr isotope geo-chemistry of hydrous mantl€ nodul€s and their host alkali ba-salts: Implications for local heterogeneities in metasomaticallyveined mantle. Earth and Planetary Science Letters,46,323-334.

Minster, J. F. and Alldgre, C. J. (1978) Systematic use of trace ele-ments in igneous processes. III: Inverse problem ofbatch partialmelting in volcanic suites. Confiibutions to Mineralogy and Pe-trology, 68,37-52.

Mitchell, R. H. and Carswell, D. A. (1976) Lanthanum, samariumand ytterbium abundances in some Southem African garnetlherzolites. Earth and Planetary Science Letters, 31, 175-178.

Morgan, J. W., Wandless, G. A., and Petrie, R. K. (1980) Earth'supper mantle: volatile element distribution and origin of sidero-phile element content. Lunar and Planetary Science XI, pp.740-742, Abstracts of Papers Submitted to the Eleventh Lunarand Planetary Science Conference.

Mysen, B. O. (1977a) Partitioning of nickel betwe€n pargasite,gamet peridotite minerals and liquid at high pressure and tem-perature. Carnegie Institution of Washington Year Book, 76,557-563.

Mysen, B. O. (1977b) The solubility of HrO and CO2 under pre-dicted magma genesis conditions and some petrological andgeophysical implications. Reviews of Geophysics and SpacePhysics, 15,351-361.

Mysen, B. O. (1978a) Experimental determination of rare earthelement partitioning between hydrous silicate melt, amphibole

and garnet peridotite minerals at upper mantle pressures andtemperatures. Geochimica et Cosmochimica Acta, 42, 1253-1263.

Mysen, B. O. (1978b) Limits of solution of trace elements in min-erals according to Henry's law: Review of experimental data.Geochimica et Cosmochimica Acta, 42, 871-885.

Mysen, B. O. (1979) Trace element partitioning between garnetperidotite minerals and water-rich vapor: Experimental datafrom 5 to 30 kbar. American Mineralogist, 64,27+288.

Mysen, B. O. and Boettcher, A. L. (1975a) Melting of a hydrousmantle. I. Phase relations of natural peridotite at high pressuresand temperatures with controlled activities of water, carbondioxide and hydrogen. Journal ofPetrology, 16,520-548.

Mysen, B. O. and Boettcher, A. L. (1975b) Melting of a hydrousmantle. II. Geochemistry of crystals and liquids formed by ana-texis of mantle peridotitc at high pressures and temperatures asa function of controlled activiti€s of water, hydrogen and car-bon dioxide. Journal of Petrology, 16,549-592.

Mysen, B. O. and Holloway, J. R. (1977) Experimental determina-tion ofrare earth fractionation patterns in partial melts from pe-ridotit€ in the upper mantle. Earth and Planetary Science Let-ters. 34. 231-237.

Mysen, B. O. and Kushiro,l. (1977) Compositional variations ofcoexisting phases with degree of melting of peridotite in the up-per mantle. American Mineralogist, 62, 843-865.

Mysen, B. O. and Kushiro, L (1978) The effect of pressure on thepartitioning of nickel between olivine and aluminous silicatemelt. Carnegie Institution of Washington Year Book, 77, 706-709.

Mysen, B. O. and Seitz, M. G. (1975) Trace element partitioningdetermined by beta-track mapping: An experimental study us-ing carbon and samarium as examples. Journal of GeophysicalResearch, 80, 2627 -2635.

Mysen, B. O., Virgo, D., and Seifert, F. (1979) Influence of meltstructure on element partitioning between olivine and melt andbetween orthopyroxene and melt at I atm. Carnegie Institutionof Washington Year Book, '18, 542-547.

Nagasawa, H., Wakita, H., Higuchi, H., and Onuma, N. (1969)Rare earths in peridotite nodules: An explanation ofthe geneticrelationship between basalt and peridotite nodules. Earth andPlanetary Science Letters, 5, 337 -381.

Nixon, P. H. and Boyd, F. R. (1973) Petrogenesis ofthe granularand sheared ultrabasic nodule suite in kimberlites. In P. H.Nixon, Ed., Lesotho Kimberlites, p. 48-56. Lesotho NationalDevelopment Corporation, Maseru, Lesotho.

Philpotts, J. A., Schnetzler, C. C., and Thomas, H. H. (1972) Pet-rogenetic implications of some new geochemical data on eclo-gitic and ultrabasic inclusions. Geochimica et CosmochimicaActa ,36 , l13 l -1166.

Presnall, D. C., Dixon, S. A., Dixon, J. R., O'Donnell' R. H''Brenner, N. L., Schrock, R. L., and Dycus, D. W. (1978) Liq-uidus phase relations on the join diopside-forsterite-anorthitefrom I atm to 20 kbar; their bearing on the generation and crys-tallization of basaltic magma. Contributions to Mineralogy andPetrology, 66, 203-220.

Reid, J. B. and Frey, F. A. (1971) Rare earth distributions inlherzolite and garnet pyroxenite xenoliths and the contributionof the upper mantle. Journal of Geophysical Research, 76,l l8,$-n96.

Ridley, W. I. and Dawson, J. B. (1975) Lithophile trace elementdata bearing on the origin of peridotite xenoliths, akermaniteand carbonatite from Lashaine volcano, N. Tanzania. Physicsand Chemistry of the Earth, 9, 559-569.

Robinson, C. C. (1974a) Multiple sites for Ef+ in alkali-silicateglasses. I. The principal six-fold co-ordinated site of Ef+ in sili-cate glass. Journal of Non-Crystalline Solids, 15, l-ll.

Robinson, C. C. (1974b) Multiple sites for EC* in alkali silicateglasses. II. Evidence of four sites for Gt'+. Journal of Non-Crystellin6 Solids, 15, 1l-29.

Ryerson, F. J. and Hess, P. C. (1978) Implication of liquid-liquiddistribution coefficients to mineral-liquid partilioning. Geochi-mica et Cosmochimica Acta, 42,921-932.

Schilling, J. G. and Winchester, J. W. (1969) Rare earth contribu-tion to the origin of Hawaiian lavas. Contributions to Mineral-ogy and Petrology, 23, 22-37.

Seitz, M. G. (1974) Fractionation of a rare earth element betweendiopside and a basalt melt at 20 kbar pressure. Carnegie Institu-tion of Washington Year Book, 73, 547-551.

Shaw, D. M. (1970) Trace element fractionation during anatexis.Geochimica et Cosmochimica Acta, 34,237-243.

Shimizu, N. (1975) Rare earth elements in garnets and clinopyrox-enes from garnet lherzolite nodules in kim!g1t1ss. Earth andPlanetary Science Letters, 25, 26-32.

Shimizu, N. and Arculus, R. J. (1975) Rare earth element conoen-trations in a suite ofbasanitoids and alkali olivine basalts fromGrenada, Lesser Antilles. Contributions to Mineralogy and Pe-trology, 50, 231-240.

Shimizu, N. and Kushiro, I. (1975) The partitioning of rare earrhelements between garnet and liquid at high pressures: Prelimi-nary experiments. Geophysical Research Letters, 2, 413416.

Watson, E. B. (1976) Two-liquid partition coefficients: Experimen-tal data and geochemical implications. Contributions to Miner-alogy and Petrology,56, l19-134.

Watson, E. B. (1977) Partitioning of manganese between forsteriteand silicate liquid. Geochimica et Cosmochirnica Acta, 41,1363-1374.

Weill, D. F. and McKay, c. A. (1975) The partitioning of Mg, Fe,

HARRISON: PARTITIONING OF REE 259

Sr, Ce; Sm, Eu, and Yb in lunar igneous systems and a possibleorigin of KREEP by equilibrium partial melting. Proccedings ofthe Sixth Lunar Science Conference, p. 1143-1158.

Wendlandt, R. F. and Eggler, D. H. (1980a) Investigations bearingon the origins of potassic magmas: I. Melting reLations in thesystem KAlSiOa-Mg2SiOa-SiO2-CO2 to 30 kilobars. AmericanJournal of Science, 280, 385421.

Wendlandt, R. F. and Eggler, D. H. (1980b) Investigations bear-ing on the origiru of potassic magmas: II. Stability of pblogopitein natural spinel lherzolite and in the system K2O-MgO-AI2O3=SiO2-H2O-CO2 as a function of volatile composition athigh pressures and temperatures. American Journal of Science,280,421459.

Wendlandt, R. F. and Harriso4 W. J. (1979) Rare earth elementpartitioning between coexisting immiscible silicate and carbotr-ate liquids and CO2: Implications for the formation of light rareearth enriched magnas. Contributions to Mineralogy and Pe-trology, 69, 4o9419.

Wendlandt, R. F. and Mysen, B. O. (1978) Melting phase relationsof natural peridotite + CO2 as a function of degree of partialmelting at 15 and 30 kbar. Carnegie Institution of WashingtonYear 8ook,78,756-761.

Whittaker, E. J. W. and Muntus, R. (1970) Ionic radii for use ingeochemistry. Geochimica et Cosmochimica Acta, 34, 945-956.

Wood, B. J. (1976) Partitioning of samarium between garnet andliquid. Carnegie Institution of Washington Year Book, 75,659-662.

Yoder, H. S., Jr. (1976) Generation of Basaltic Magma. NationalAcademy of Sciences, Washington, D.C.

Manuscript received, fune 16, I9E0;acceptedfor publication, December 4, 1980.

partitioning of ree between minerals and coexisting melts...

Documents