trace element partitioning between plagioclase and melt

PII S0016-7037(00)00389-6

Trace element partitioning between plagioclase and melt: Investigation of dopant influenceon partition behavior

ILYA N. BINDEMAN1,* and ANDREW M. DAVIS

2

1Department of Geology & Geophysics, University of Wisconsin, Madison, Wisconsin 53706, USA2Enrico Fermi Institute and Department of Geophysical Sciences, University of Chicago, Chicago, Illinois 60637, USA

(Received March26, 1999;accepted in revised form March13, 2000)

Abstract—We present results from an ion microprobe study of REE-doped and natural concentrationplagioclase-basalt run products of Drake (1972) that carries on from our earlier study (Bindeman et al., 1998).The goals of this work are (1) to determine plagioclase/melt partition coefficients for all REE for four analyzedplagioclase compositions (An40–80); and (2) to determine whether doping with REE influences partitioncoefficients of other REE and other trace elements. In combination with our analyses of Sr-doped runs(Bindeman et al., 1998), the new data allow us to compare partition coefficients (Di’s) of trace elements atnatural concentrations with those in REE-doped and Sr-doped runs. In these comparisons, runs have the samerun temperature and compositions, but different doping element(s).

We find that ln(DREE) decreases as a linear function of REE atomic number, in contrast to mostplagioclase-melt partition experiments. The slopes of the ln(Di) vs. %An and RT ln(Di) vs. %An depen-dencies of REE and monovalent and divalent cations increase from smaller to larger ions of each valencegroup.DLa/DY andDLa/DLu increase by more than 5 times as plagioclase composition changes from An80 toAn40. Within each valence group, slopes on ln(Di) vs. %An and RT ln(Di) vs. %An plots increase linearlywith ionic radius, with the trivalent REE showing the steepest slope vs. ionic radius dependence. Applicationof the elastic modulus model of Blundy and Wood (1994) to our results yielded good results, confirming thatelastic properties determine partition behavior. We also presentDi’s for U.

We find no detectable effects of trace element doping onDi’s of 30 trace elements of varying size andcharge. This implies that coupling between different trace elements is not a significant process duringpartitioning innaturalsystems, even for the case of heterovalent substitutions (such as REE). However,DREE

andDY in REE-doped runs are 30–100% higher thanDREE in undoped runs and Sr-doped runs. Doping withthousands of ppm of three selected REE (it does not matter which three) affectsDi’s of Y and all REE in theseruns, including those at natural concentration levels. We suggest that substituting trace cations largely competefor sites of substitution with the mineral-forming cations Na and Ca. We speculate that lowerDREE inREE-doped runs are the result of a change in the REE substitution mechanisms at doped concentrations of allREE which leads to a different value of the Henry’s law constant.Copyright © 2000 Elsevier Science Ltd

1. INTRODUCTION

Trace element partition coefficients (Di’s) are important incharacterizing and modeling primary and evolving terrestrialand extraterrestrial melts based on the trace element composi-tions of igneous minerals and the melts they were in equilib-rium with when they crystallized. A key concern in partitioningstudies is the influence of trace element concentration onDi’s(non-Henry’s Law behavior), as trace elements are often dopedto wt.% levels in experiments. Doping was necessary to makeion or electron microprobe measurements easier (or possible),especially in earlier studies. In addition, the dependence ofDi’son trace element concentration must be known when compar-ing natural systems with strongly differing trace element con-tents (e.g., tholeiites vs. pegmatites).

Another potentially important aspect of partitioning studiesis maintaining charge neutrality. Doping with trace elements ofdifferent charge (or oxidation state) may change the coupledsubstitution mechanism(s) that maintains charge balance. Sev-eral possible exchange reactions have been proposed for in-tracrystalline REE31 charge compensation in feldspars, cli-

nopyroxenes and other common silicates (Kneip and Liebau,1994; D’Arco and Piriou, 1989; see Discussion below).

The formation of clusters within crystal structures that re-sults from coupled substitution (e.g., Navrotsky, 1978) maycomplicate concentration dependence of partition coefficients.In addition, the issue of defect equilibria at low (natural)concentration levels in experiments and nature has receivedmuch attention and remains controversial (see Drake and Hol-loway, 1978; Watson, 1985; Beattie, 1993; and Urusov andDudnikova, 1998 for reviews). Higher values of the Henry’sconstant and partition coefficients at low concentrations havebeen attributed to partitioning into defect sites in minerals. Inmost of these experiments one or a few REE were added to achemically pure system and a departure from Henry’s lawbehavior was observed. Harrison and Wood (1980) have shownthat Henry’s law fails at;1 ppm for garnet-melt REE parti-tioning. Hoover (1978) showed it for Sm and Tm partitioningbetween plagioclase and melt. Urusov and Dudnikova (1998)have presented compelling evidence for “trapping effects” ofmicroimpurities in nonsilicate and, more recently, silicate sys-tems (see Urusov and Kravchuk, 1978; Urusov and Dudnikova,1998).

Recently, Beattie (1993; 1994) argued that there is insuffi-* Author to whom correspondence should be addressed.

Pergamon

Geochimica et Cosmochimica Acta, Vol. 64, No. 16, pp. 2863–2878, 2000Copyright © 2000 Elsevier Science LtdPrinted in the USA. All rights reserved

0016-7037/00 $20.001 .00

2863

cient evidence to claim that Henry’s law constants are differentat doped vs. natural concentration levels. He suggested thatpreviously reported results (e.g., Harrison and Wood, 1980;Mysen, 1978) on the concentration dependence of reportedDi’s are likely to be analytical artifacts of theb-track autora-diography technique used in those studies. However, a varietyof other techniques used to measure trace elements at lowconcentrations suggest that point and dislocational, equilibriumor nonequilibrium defects do exist at;0.1–10 ppm level oftrace element concentration (see Navrotsky, 1978; Urusov andDudnikova, 1998). These defects may participate in exchangereactions with trace elements, increasing the value of partitioncoefficients at ppm levels of concentration. Defect sites canalso be produced up to levels of several mole % as a result oftrace element heterovalent substitutions, for example, as aresult of REE-doping of feldspars (e.g., Kimata, 1988; D’Arcoand Piriou, 1989; Kneip and Liebau, 1994).

It is not clear how experimental results with synthetic ma-terials apply to complex natural systems in which minerals maycontain tens to hundreds of ppm of total REE and thousands ofppm of all trace elements combined. Geochemical evidence,such as the constancy of Zr/Hf and Rb/K ratios in rocks (e.g.,Watson, 1985), and the absence of odd–even anomalies inchondrite–normalized REE patterns, suggest that Henry’s lawis largely obeyed in fractionation processes in nature, even if itis not obeyed for some elements in certain experiments.

The ion microprobe is a reliable microbeam technique usedto measure trace elements at sub-ppm levels of concentration,in the domain of potential failure of Henry’s law. The experi-mental run products from the Drake (1972) plagioclase/naturalbasalt partitioning study provide an excellent set of samples toaddress these issues. In order to determine partition coefficientswith the electron microprobe, Drake (1972) conducted manyequivalent runs (with the same temperature and same plagio-clase and melt compositions) doped with a variety of elements(Sr, Ba, or 1 to 4 selected REE). A few blank runs were alsoconducted with no doping. Since each of these equivalent runswas doped with one or few elements, it is possible to directly

compare partition coefficients of all ion microprobe measurableelements dependingonly on the doping element. For example,partition coefficients for Sr can be studied in Sr-doped, Ba-doped, and REE- and Y-doped run series and in undoped runseries. Such an approach allows us to consider not only thepartitioning of different trace elements at doped vs. undopedlevels (e.g., Bindeman et al., 1998), but also to check thedopant influence on partition coefficients ofundopedelements.

We report here new ion microprobe analyses of the REE-doped and undoped plagioclase–basalt run products of Drake(1972). We made ion microprobe analyses for concentrations ofa variety of undoped elements in three undoped runs (see Table1). We also analyzed three more REE- and Y-doped samples, inaddition to those reported in Bindeman et al. (1998). In com-bination with our analyses of Sr-doped runs, these new dataallow us to compareDi’s of trace elements in doped, Sr dopedor undoped inequivalentruns.

2. ANALYTICAL TECHNIQUES

Analytical details are given in our first paper (Bindeman et al., 1998),and descriptions of experimental conditions are given in Drake (1972)and in Drake and Weill (1975). Table 1 summarizes conditions for theirruns that were analyzed in this study.

Run products were first studied under reflected light and then withsecondary and backscattered electron imaging on a JEOL JSM-5800LVscanning electron microscope. Electron microprobe analyses for majorelements were made on a Cameca SX-50 electron microprobe using anaccelerating voltage of 15 kV and a defocused beam of 10–25 nA (;2mm) to minimize sodium loss. Two-mm step profiling was performedalong and across several crystals within selected runs. We found thatcrystals are homogeneous within error with respect to mole % An andthe major elements K, Fe, Mg, and Ti. Cathodoluminescence imagingof plagioclase in REE-doped, Sr-doped, and undoped runs also showedhomogeneity within and among crystals in each run product analyzed.Given the;10 mm ion microprobe beam diameter, we could onlyanalyze (and report here) experimental products of in which the crystalsize exceeded 15–20mm, where it was possible to avoid overlap ontoadjacent glass during the ion microprobe sputtering.

Ion microprobe analyses were made using the modified AEI IM-20instrument at the University of Chicago (e.g., Hinton et al., 1988;Simon et al., 1994; MacPherson and Davis, 1994; Bindeman et al.,

Table 1. Experimental conditions and products of REE-doped and undoped runs.a

Run T (K) System Doped element %An, avg%

Total REE (ppm)

Plag Glass

Undoped99-1 1460 MP11 20%Ab 51.3 16 88

100-1 1530 MP11 10%An 69.2 22 86101-1 1570 MP11 40%An 71.3 7 50

REE-doped94-7 1426 MP11 50%Ab La 44.7 6633 25,26399-7 1460 MP11 20%Ab Nd, Dy, Er 53.0 1208 31,15599-2 1460 MP11 20%Ab Sm, Eu, Gd 53.0 2368 39,81099-5 1460 MP11 20%Ab La, Ce, Y, Lu 52.5 3053 25,867

100-4 1530 MP11 10%An La, Ce, Y, Lu 67.7 2899 37,267100-6 1530 MP11 10%An Nd, Dy, Er 68.5 729 19,041101-4 1572 MP11 40%An La, Ce, Y, Lu 75.5 2443 38,709101-3 1572 MP11 40%An Sm, Eu, Gd 75.0 3439 56,373101-7 1572 MP11 40%An Nd, Dy, Er 75.0 882 16,601

a MP1 is a natural basaltic andesite from Mountain Pass, Oregon; Ab and An are synthetic albite and anorthite. %An was measured on theUniversity of Chicago electron microprobe. Added REE are the sum of indicated doped REE. Total ion microprobe-measured REE in undoped runsare the sum of La, Ce, Pr, Nd, Sm, Eu, and Y (HREE were below detection limits in plagioclase).

2864 I. N. Bindeman and A. M. Davis

1998). Calcium-normalized ion yields were obtained from a variety ofnatural and synthetic silicate standards. Previous experience (Hinton etal., 1988; MacPherson and Davis, 1994) has shown that there is littlevariation of ion yields in different silicate matrices under the energy-filtering conditions used in this work.

Since Di’s of HREE heavier than Eu are normally below 0.01,HREE concentrations in typical igneous plagioclase and undoped ex-perimental products are at the 10–100 ppb level. In addition, heavyREE have isobaric mass interferences from oxides of lighter REE andBa (such as BaO on Eu and NdO and SmO on Dy). These effects areespecially severe in high LREE/HREE phases such as plagioclase, soHREE were determined only at doped (80–200 ppm in plagioclase)concentration levels. Y, however, has a higher cosmic abundance thanthe HREE and no significant interferences and can be determined withreasonable precision even at the sub-ppm concentrations typical ofnatural plagioclases. Since Y has the same radius and valence as Dyand Ho (Shannon, 1976), it was used as a proxy for HREE.

3. RESULTS

3.1. Partition Coefficients of HREE and U

We have reanalyzed experimental run charges doped withHREE in order to accurately determineDi’s for these elements,since the HREE were near electron microprobe detection limitsin the earlier Drake and Weill (1975) study (Tables 2–4). In ourelectron microprobe reanalyzes (Bindeman et al., 1998), detec-tion limits are 50–100 ppm; Drake and Weill did not givedetection limits. Together with reanalyzed LREE-doped runs,the new analyses provide a consistent set of partitioning datafor most of the REE at doped concentration levels for a rangeof plagioclase compositions. The partition coefficients for Gd,Dy, Er, and Lu and the linear regression of ln(DHREE) vs. %Anare presented on Figure 1 and are compared with the earlierDrake and Weill (1975) measurements and with existing vol-canic phenocrysts/matrix determinations. We find thatDGd issimilar to the Drake and Weill (1975) value, whileDDy andespeciallyDEr and DLu are increasingly discordant and thecharacter of their %An dependence is different.

Concentrations of U in plagioclase are near detection limitsin most runs but we find that our determinations ofDU arewithin the range of natural plagioclases from volcanic rocks(see Fig. 1) and observe that ln(DU) has a strong dependence on%An, as is the case with other cations with high charge andlarge ionic radii (e.g., Bindeman et al., 1998).

We calculated the slopes and intercepts of the RT ln(Di) vs.%An relationship using Williamson’s (1968) regression routinewhich weights each analysis according to its uncertainty, andyields uncertainties in the slope and intercept:

RT ln~DU) 5 24846 195%An2 74526 12,552,

RT ln~DDy) 5 21476 27%An2 35,0756 1925,

RT ln~DY) 5 2326 8%An2 59,7006 592,

RT ln~DGd) 5 2946 29%An2 30,2346 1834,

RT ln~DEr) 5 21646 28%An2 40,0876 1992,

and

RT ln~DLu) 5 2806 7%An2 33186 515.

The linear fit is thermodynamically justified (e.g., Blundy andWood, 1991) and the empirical dependence of lnDi on %An is

of good practical value for choosing aDi for measured %An ina natural plagioclase. These approximation parameters for Dy,Y, Gd, Er, Lu at doped concentration levels supersede thosereported in our earlier study (Bindeman et al., 1998) and arerecommended for use.

3.2. ln(Di) vs. %An and RT ln(Di) vs. %AnDependencies as a Function of Trace Element Sizeand Charge

A comparison of the slopes of the ln(Di) dependencies on%An for the REE shows a regular decrease of slope fromstrongly negative (La) to weakly positive (Lu) (Fig. 2a). Bin-deman et al. (1998) observed a similar effect for plagioclase/melt partition coefficients for monovalent and divalent cations.Those cations that are similar, or smaller, in size than that of theM-site (e.g., Li and Mg) exhibit a positive slope in their ln(Di)vs. %An dependence, while almost all other elements shownegative slopes.

The REE are a group of trace elements that exhibit gradualdecrease in atomic radii with increasing atomic number, and theobserved smooth changes in partitioning behavior are to beexpected. In order to explain the fan-like trend shown in Figure2a, we used the thermodynamic elastic modulus approach ofBrice (1975) elaborated by Blundy and Wood (1994) for par-titioning studies. In this model, at each temperature and pres-sure, partition coefficients of trace elements of each valencegroup can be described as a function of three parameters: theoptimum radius of the lattice site in plagioclase structure,r0;the “strain-compensated” partition coefficients,D0, for strainfree substitution; and the Young’s modulus,E, of plagioclase.The elastic modulus model describes a near parabola whencalculated partition coefficients are plotted on a ln(Di) vs. ionicradius graph. Experimentally and naturally determined partitioncoefficients of trace elements lie along inverted parabolas onln(Di) vs. ionic radius plots when grouped by valence (e.g.,Onuma et al., 1968).

Application of the model to our data yielded a similarlyfan-like trend between LREE and HREE with decreasing An(Fig. 2b). The difference in the slopes of the ln(Di) vs. %Andependence for LREE vs. HREE leads to partition coefficientratios, such asDLa/DLu andDLa/DY, that are very sensitive toplagioclase composition (Fig. 3): anorthite exhibits the smallestDLa/DY and DLa/Y ratios and albite the largest ones. Thissuggests that plagioclase of variable composition is capable ofchanging the geochemically indicative LREE/HREE ratios byup to one order of magnitude for a reasonable An range (e.g.,An80 to An40) within a single zoned grain of plagioclase. Thereconstruction of parent melt REE patterns assuming constantpartition coefficients ratios, therefore, can be quite misleading.

The RT ln(Di) vs. %An dependence resembles the ln(Di) vs.%An dependence, since the temperature difference betweenhigh and low temperature experiments is only 144°C and is inthe high temperature interval where temperature differenceshave smaller effects (see Table 1). This translates to a less than10% variation in RT ln(Di) parameters as compared to thosefor ln(Di). The slopes of the RT ln(Di) vs. %An dependenciesincrease with increasing ionic radius for REE and monovalentand divalent cations entering theM site (Bindeman et al., 1998)(Fig. 4). For divalent cations (Mg, Ca, Sr, and Ba) there is an

2865Investigation of dopant influence

Tab

le2.

Pla

gioc

lase

com

posi

tions

from

RE

E-d

oped

and

undo

ped

runs

.O

xide

conc

entr

atio

nsar

egi

ven

inw

t.%an

del

emen

tco

mpo

sitio

nsar

egi

ven

inpp

m.

The

dopa

ntco

ncen

trat

ions

are

unde

rline

d.U

ncer

tain

ties

are

base

don

the

grea

ter

ofco

untin

gst

atis

tics

orre

plic

ate

anal

yses

and

are

only

give

nw

hen

they

exce

ed5%

ofth

eam

ount

pres

ent.

Upp

erlim

itsar

e,

2s.

99-1

100-

110

1-1

94-7

99-7

99-2

99-5

100-

410

0-6

101-

410

1-3

101-

7

#an

al.

43

313

23

23

32

22

An%

51.5

60.

866

.26

0.9

71.3

60.

639

.86

0.5

50.7

60.

250

.66

0.7

49.1

61.

553

.96

0.5

66.3

60.

569

.96

0.3

71.8

62.

170

.16

0.7

Na 2

O5.

503.

372.

906.

515.

515.

386.

133.

973.

503.

003.

133.

19A

l 2O

329

.732

.333

.225

.229

.329

.129

.131

.932

.132

.333

.232

.8S

iO2

50.5

47.7

46.2

57.4

51.4

52.1

50.1

46.7

47.8

48.2

44.8

46.3

CaO

12.3

14.8

16.2

8.56

11.8

11.4

12.3

15.3

14.9

15.0

17.1

16.2

FeO

1.43

1.29

1.09

1.00

1.25

1.18

1.41

1.22

1.05

0.81

0.91

0.93

¥Ln

2O

30.

000.

000.

000.

770.

140.

280.

020.

320.

090.

030.

400.

10Li

11.5

22.5

61.

831

.44.

3811

.960.

614

.124

.230

.950

.111

.621

.310

.5B

e1.

606

0.33

1.38

60.

181.

216

0.30

0.12

460.

035

1.22

60.

201.

376

0.09

0.68

360.

044

0.79

60.

240.

4636

0.08

80.

4526

0.06

71.

521.

47B

3.43

60.

671.

466

0.56

3.33

60.

391.

356

0.25

4.94

9.26

2.8

7.12

60.

606.

696

0.62

4.26

60.

9025

617

2.25

60.

296.

596

0.73

Mg

1017

1047

1036

658

7736

5110

136

215

834

9796

124

1032

616

093

876

082

883

8P

956

1576

635

456

1612

6616

826

1690

614

716

1756

618

436

2952

615

35.7

66.

367

611

K12

0610

156

7310

1110

0712

9413

5916

11612

714

826

216

1163

1060

9296

104

1228

678

Ti

513

626

3686

2043

3631

424

4916

3441

5631

463

3846

5837

932

3635

2606

2133

5R

b0.

826

0.43

,0.

670.

806

0.29

0.95

60.

230.

956

0.48

0.69

60.

341.

006

0.42

0.81

60.

67,

0.64

,0.

481.

346

0.73

0.99

60.

34S

r10

7680

462

698

610

9610

9811

1191

26

7180

455

061

4635

647

Y0.

431

60.

043

0.65

560.

063

0.52

60.

120.

916

0.11

1.00

60.

280.

4896

0.07

914

069

186

642

0.66

60.

1412

667

0.75

160.

079

0.58

960.

077

Zr

0.45

60.

220.

356

0.10

0.49

60.

190.

776

0.18

0.22

760.

083

0.40

60.

140.

586

0.28

1.50

60.

780.

2626

0.09

40.

3346

0.08

60.

3746

0.06

60.

416

0.12

Ba

1286

882

.184

614

207

147

185

147

95.56

7.8

81.8

59.1

756

1671

612

La2.

116

0.15

1.64

60.

151.

706

0.19

6572

3.58

2.20

60.

2821

926

119

1135

663

1.89

60.

1110

791.

616

0.16

1.93

60.

24C

e3.

176

0.19

2.75

60.

302.

726

0.49

11.9

60.

96.

402.

836

0.21

6876

4114

436

953.

326

0.23

1110

2.36

60.

252.

786

0.21

Pr

0.36

060.

046

0.37

360.

046

0.36

560.

039

1.16

60.

120.

686

0.20

0.47

60.

110.

436

0.11

0.42

60.

150.

336

0.09

0.23

60.

10.

4756

0.06

20.

4706

0.09

2N

d1.

146

0.11

1.39

60.

131.

406

0.14

4.20

60.

3459

312

.16

6.1

1.74

60.

681.

216

0.32

303

0.79

60.

375.

263.

366

1651

Sm

0.45

860.

083

0.50

960.

091

0.42

60.

120.

5866

0.07

3,

0.79

1268

0.62

60.

160.

396

0.29

,0.

790.

596

0.11

1686

0.39

60.

21E

u0.

1046

0.06

40.

1736

0.05

50.

1236

0.07

80.

726

0.15

0.31

60.

1463

70.

546

0.10

1.45

60.

410.

116

0.10

0.87

460.

067

984

,0.

33G

d46

376

9D

y33

626

713

7E

r27

915

984

.26

5.9

Lu34

.413

56

3012

86

9U

0.10

360.

038

0.15

160.

084

0.06

160.

030

0.07

760.

030

0.07

260.

030

,0.

058

,0.

044

0.04

260.

021


Tab

le3.

Gla

ssco

mpo

sitio

nsfr

omR

EE

-dop

edan

dun

dope

dru

ns.

Oxi

deco

ncen

trat

ions

are

give

nin

wt.%

and

elem

ent

com

posi

tions

are

give

nin

ppm

.T

hedo

pan

tco

ncen

trat

ions

are

unde

rline

d.U

ncer

tain

ties

are

base

don

the

grea

ter

ofco

untin

gst

atis

tics

orre

plic

ate

anal

yses

and

are

only

give

nw

hen

they

exce

ed5%

ofth

eam

ount

pres

ent.

Upp

erlim

itsar

e,

2s.

99-1

100-

110

1-1

94-7

99-7

99-2

99-5

100-

410

0-6

101-

410

1-3

101-

7

#an

al.

33

32

22

24

22

22

Na 2

O5.

613.

633.

237.

325.

615.

446.

874.

463.

843.

804.

153.

89A

l 2O

315

.717

.018

.815

.217

.217

.117

.618

.017

.520

.620

.820

.5S

iO2

62.5

60.8

58.0

62.5

57.5

56.3

55.1

54.1

58.4

50.6

47.5

53.2

CaO

4.65

6.76

8.34

3.63

4.92

5.11

5.41

7.74

7.22

9.21

9.86

9.21

FeO

5.13

5.59

5.35

3.65

4.93

5.26

5.46

5.31

4.72

5.17

5.06

5.00

¥Ln

2O

30.

010.

010.

012.

863.

594.

661.

263.

612.

202.

226.

711.

92Li

36.2

74.0

106

17.96

1.2

44.0

52.66

3.4

95.2

107

1856

1046

.894

.441

.6B

e1.

816

0.11

1.33

60.

131.

456

0.14

0.43

160.

040

2.08

60.

242.

081.

836

0.13

1.02

1.01

60.

080.

9556

0.05

01.

476

0.14

1.87

60.

20B

28.2

61.

622

.86

1.5

22.2

6.78

60.

4021

.86

1.5

43.1

62.

851

.025

.062.

224

.56

1.4

37.1

65.

215

.928

.7M

g24

212

2351

624

225

1740

2611

8522

109

2330

922

378

2174

421

940

2214

322

204

2357

9P

924

6236

4554

5633

1184

674

940

872

880

4746

4161

0687

556

3426

1946

3K

7845

7547

7161

5786

7680

7043

1033

889

9881

5888

3979

2684

69T

i57

6156

6255

5144

1155

0855

8757

7754

3754

0456

5053

9454

88R

b26

.320

.019

.361.

428

.76

1.7

56.9

29.0

36.86

2.1

23.1

38.36

4.2

20.8

61.

122

.731

.2S

r34

038

432

929

5633

391

386

423

4526

2540

036

638

038

5Y

14.9

15.5

16.6

14.26

1.2

26.9

20.06

2.1

5910

6229

24.0

5570

24.2

23.7

Zr

102

112

119

81.0

106

102

107

103

107

123

105

114

69

Ba

330

296

306

312

341

367

364

368

625

301

323

372

335

La8.

248.

036

0.47

9.20

2433

7629

9013

.26

0.7

9.58

9944

6481

9.74

60.

7476

7510

.411

.86

0.6

Ce

20.0

20.5

22.4

25.36

1.4

38.4

21.1

5015

1268

028

.311

680

24.5

29.3

Pr

2.47

2.48

60.

192.

626

0.18

2.82

4.88

60.

473.

276

0.26

3.45

60.

224.

606

0.33

2.80

60.

273.

576

0.41

3.60

4.07

60.

27N

d9.

2610

.511

.510

.960.

648

46,

8210

.36

0.6

11.7

61.

128

9510

.76

0.93

1526

1672

81S

m4.

693.

556

0.18

3.00

60.

162.

696.

186

0.71

2027

95.

839.

862.

04.

361.

67.

676

0.53

2679

83.

306

0.74

Eu

0.71

60.

101.

046

0.12

1.41

60.

11,

0.19

2.55

60.

2895

087.

3716

.26

1.1

2.13

60.

2614

.014

260

2.62

60.

25G

d10

0236

508

1531

4D

y11

197

8415

4662

Er

1511

177

3146

57Lu

4997

1187

713

785

U2.

626

0.19

2.06

60.

151.

546

0.11

1.01

60.

081.

376

0.25

0.91

60.

160.

8686

0.06

43.

926

0.28


Tab

le4.

Pla

gioc

lase

/mel

tpa

rtiti

onco

effic

ient

sfr

omR

EE

-dop

edan

dun

dope

dru

ns.

The

dopa

ntpa

rtiti

onco

effic

ient

sar

eun

derli

ned.

Unc

erta

intie

sar

epr

opag

ated

from

unce

rtai

ntie

sof

conc

entr

atio

nsin

plag

iocl

ase

and

glas

san

dar

eon

lygi

ven

whe

nth

eyex

ceed

5%of

the

amou

nt.

Upp

erlim

itsar

e,

2s.

99-1

100-

110

1-1

94-7

99-7

99-2

99-5

100-

410

0-6

101-

410

1-3

101-

7

Li0.

317

0.30

560.

024

0.29

60.

2456

0.01

80.

2726

0.01

90.

2686

0.02

00.

254

0.28

80.

2716

0.01

40.

247

0.22

50.

252

Be

0.88

60.

191.

046

0.17

0.84

60.

220.

2886

0.08

50.

586

0.12

0.65

660.

048

0.37

360.

035

0.77

60.

240.

4576

0.09

40.

4736

0.07

41.

036

0.10

0.78

760.

089

B0.

1226

0.02

50.

0646

0.02

50.

1506

0.01

90.

1996

0.03

80.

2266

0.01

90.

2136

0.06

60.

1396

0.01

20.

2676

0.03

40.

1746

0.03

80.

696

0.48

0.14

260.

019

0.23

060.

026

Na

0.98

10.

930

0.89

70.

889

0.98

30.

989

0.89

30.

889

60.

048

0.91

40.

789

0.75

460.

067

0.82

0M

g0.

0420

0.04

450.

04276

0.00

250.

0444

60.

0042

0.04

660.

010

0.03

580.

04376

0.00

570.

0474

60.

0074

0.04

280.

0343

0.03

730.

0355

Al

1.90

1.90

1.76

1.66

1.70

1.71

1.65

1.78

1.83

1.57

1.59

1.60

Si

0.80

80.

785

0.79

60.

919

0.89

50.

925

0.91

00.

863

0.81

90.

954

0.94

40.

871

P0.

1036

0.01

70.

1236

0.05

70.

0826

0.03

10.

1066

0.01

50.

0876

0.01

70.

1036

0.01

70.

0816

0.01

90.

1186

0.03

90.

0716

0.04

80.

0946

0.02

80.

1046

0.01

90.

1456

0.02

4K

0.15

40.

1356

0.01

00.

141

0.17

460.

009

0.16

80.

193

0.15

660.

013

0.16

560.

025

0.14

30.

120

0.11

760.

013

0.14

560.

009

Ca

2.68

2.19

1.94

2.366

0.19

2.39

2.23

2.26

60.

111.

982.

071.

621.

741.

76T

i0.

0891

60.

0046

0.06

506

0.00

370.

0780

60.

0056

0.09

626

0.00

550.

0892

60.

0064

0.07

436

0.00

570.

0801

0.07

160.

011

0.07

010.

05726

0.00

620.

0482

60.

0039

0.06

11F

e0.

279

0.23

00.

204

0.27

50.

253

0.22

46

0.02

20.

258

0.22

960.

021

0.22

260.

012

0.15

70.

181

0.18

560.

010

Rb

0.03

160.

016

,0.

033

0.04

260.

015

0.03

316

0.00

820.

0166

60.

0084

0.02

460.

012

0.02

760.

011

0.03

560.

029

,0.

017

,0.

023

0.05

960.

032

0.03

260.

011

Sr

3.17

2.10

1.90

3.346

0.39

2.80

2.84

2.63

2.026

0.19

2.01

1.50

1.61

60.

091.

68Y

0.02

896

0.00

290.

0423

60.

0044

0.03

156

0.00

720.

0641

60.

0092

0.03

760.

010

0.02

446

0.00

470.

0238

60.

0017

0.02

986

0.00

670.

0274

60.

0059

0.02

276

0.00

130.

0311

60.

0033

0.02

486

0.00

33Z

r0.

0044

60.

0021

0.00

3086

0.00

091

0.00

426

0.00

160.

0096

60.

0022

0.00

2156

0.00

078

0.00

396

0.00

140.

0054

60.

0027

0.01

466

0.00

760.

0024

560.

0008

80.

0027

260.

0007

10.

0035

560.

0006

30.

0037

60.

0011

Ba

0.38

960.

024

0.27

70.

2776

0.04

80.

6616

0.03

90.

4326

0.02

30.

504

0.40

30.

2596

0.02

80.

272

0.18

30.

2006

0.04

40.

2116

0.03

7La

0.25

660.

020

0.20

560.

022

0.18

460.

021

0.27

060.

035

0.27

160.

019

0.22

960.

030

0.22

060.

013

0.17

560.

012

0.19

460.

018

0.14

10.

1556

0.01

50.

1636

0.02

2C

e0.

1596

0.01

00.

1346

0.01

50.

1216

0.02

20.

4706

0.04

30.

1676

0.00

90.

1346

0.01

10.

1376

0.00

90.

1146

0.00

90.

1176

0.00

90.

0950

0.09

660.

011

0.95

060.

0075

Pr

0.14

660.

020

0.15

060.

022

0.13

960.

018

0.41

260.

043

0.13

960.

043

0.14

260.

034

0.12

460.

032

0.09

160.

033

0.11

860.

034

0.06

360.

019

0.13

260.

018

0.11

660.

024

Nd

0.12

360.

012

0.13

360.

013

0.12

160.

013

0.38

660.

038

0.12

260.

004

,0.

730.

1686

0.06

60.

1046

0.02

90.

105

0.07

460.

036

0.03

460.

022

0.09

086

0.00

71S

m0.

0986

0.01

80.

1446

0.02

70.

1396

0.04

10.

2186

0.02

9,

0.13

0.06

250.

1066

0.02

80.

0406

0.03

1,

0.19

0.07

760.

016

0.06

290.

1176

0.06

8E

u0.

1486

0.09

30.

1666

0.05

60.

0886

0.05

6,

0.64

0.12

360.

056

0.06

700.

0736

0.01

40.

0896

0.02

60.

0536

0.04

50.

0623

60.

0049

0.06

90,

0.13

Gd

0.04

616

0.00

320.

0502

Dy

0.03

000.

0318

0.02

93E

r0.

0185

0.02

060.

0181

60.

0013

Lu0.

0068

70.

0114

60.

0025

0.00

9276

0.00

069

U0.

0396

0.01

50.

0736

0.04

10.

0396

0.02

00.

0766

0.03

00.

0536

0.02

4,

0.06

5,

0.05

10.

0107

60.

0054


Fig. 1. Partition coefficients of HREE and U as a function of %An. The solid lines are linear regressions through our ionmicroprobe-measured partition coefficients; dashed lines are linear regressions through Drake’s (1972) electron microprobe-measured partition coefficients. U is at its natural concentration level, HREE are at doped concentrations. Phenocryst/matrixpartition coefficients are taken from: Nagasawa and Schnetzler (1971); Schnetzler and Philpotts (1970); Vernieres et al.,1977; Nash and Crecraft, 1985; Higuchi and Nagasawa, 1969; Dudais et al., 1971; Dunn and Sen, 1994; Fujimaki et al.,1984; Francalanci (1989); Phinney, 1992; Worner et al., (1983).


excellent overall agreement between observedDi’s and thosecalculated with the Blundy and Wood (1994) model and theslopes of ln(Di) and RT ln(Di) vs. %An dependence (exceptfor Mg, which shows a slope of opposite sign). For monovalentcations (Li, Na, K, Rb) there is a good agreement with theBlundy and Wood (1994) model, with larger discrepancies forsmallest cation, lithium, for which the model predicts lowerDi

slightly beyond our analytical uncertainty for both values andslopes.

Both our measurements and the Blundy and Wood (1994)model give an increase of RT ln(Di) vs. %An slope withincreasing ionic radius and increasing cation charge, and themagnitude of slope variation within each valence group in-creases from monovalent to trivalent cations. These effects canbe understood by consideration of an Onuma diagram [ln(Di)vs. ionic radius) (Onuma et al., 1968). The thermodynamicsignificance of Onuma diagrams can be explained using therelation discovered by Brice (1975) on the dependence of thefree energy of the exchange of a trace element between crystaland melt with the magnitude of its strain of the crystal lattice.In Figure 5, parabolic curves, based on Eqn. 2 of Blundy andWood (1994), are fitted through partition coefficients of mostanorthite-rich (thick lines) and anorthite-poor (thin line) ofanalyzed runs (Bindeman et al., 1998). These curves show thatwith an increase in albite content the parabola maximum (in-dicating the size ofM-site) shifts to the right, because the sizeof the M-site in albite is larger than that in anorthite in accor-

Fig. 2. (a) The slope of the ln(DREE) vs. %An dependence decreases with decreasing REE atomic number (increasingionic radius) from Lu to La. MeasuredDREE from experiments doped with those REE are from this study (see Fig. 1) andBindeman et al. (1998). (b) Calculated slopes using the Blundy and Wood (1994) elastic modulus model based on themeasured Di of one middle REE, Nd. Formula (3) of Blundy and Wood (1994) was used to predict the values ofDREE. Thisapproach allowed us to avoid offsets and is best for comparing ofDi’s of neighboring REE (cf. Wood and Blundy, 1997).The size of theM site for strain-free substitution of REE31 (r0) for the four studied plagioclase compositions was takenfrom the maxima of the REE parabolas on Onuma diagrams (cf. Fig. 5) and are 1.224 Å for An45, 1.215 Å for An55, 1.200Å for An65 and 1.172 Å for An75. A linear fit to these values yieldedr0 5 1.29 foralbite and 1.15 for anorthite, whichis within the range determined from bulk elastic properties of plagioclase (Angel et al., 1988) and the Blundy and Wood(1994) partitioning data. We usedE andr0 values (in kbars) (2084) for albite and (1903) anorthite from Blundy and Wood(1994) and assumed thatE is a linear function of %An in order to obtainE values for our plagioclases of intermediatecomposition. Discrepancies for larger and smaller REE may result from the assumed linear relationship ofE andr0 with%An and/or partial partitioning of smaller ions (e.g., Mg) into theT site (e.g., Peters et al., 1995).

Fig. 3. DLa/Lu and DLa/Y increase with decreasing An content ofplagioclase. Filled symbols are ion microprobe-measured values, opensymbols are calculated values based on the Blundy and Wood (1994)model (see Fig. 2 for description). Note that both measured andcalculatedDLa/Lu and DLa/Y yield a similar increase with decreasingAn%. DLa/Yb and DLa/Y values based on anorthite-CAI melt experi-ments of Simon et al. (1994) and McKay et al. (1994) are shown forcomparison.


dance with larger Na–O than Ca–O interatomic distance (Angelet al., 1988). Partition coefficients of ions whose radius is closerto the parabola maximum (e.g., Na, middle REE) do not showa large increase;Di of larger ions (e.g., Rb and La) show asignificant increase, whileDi of smaller ions (e.g., Li and Lu)show a decrease.

The dependence of the slope on ionic charge is also ex-plained by consideration of Figure 5. Since the parabola fittedthrough the trivalent REE group is narrower than the parabolafor monovalent group, an equivalent shift of parabola max-ima to the right (larger ionic radius) produces a largerrelative effect of slope change per unit of ionic radiusincrease for trivalent than monovalent ions, explaining thebehavior shown in Figure 4.

To summarize, we observe a general tendency of increasingslope of the ln(Di) vs. %An and RT ln(Di) vs. %An depen-dencies with increasing ionicsizeandcharge. REE group andhigher-charged ions are the most sensitive to plagioclase com-position change, while alkalis are least sensitive. This has animportant general implication: partition coefficients of thelarger and higher valence ions of each valence group and theirratios (e.g., Fig. 3) may vary significantly with %An.

3.3. REE Spectrum Partitioning at Doped Concentrations

We find that for four studied plagioclase compositions, ln-(DREE) decreases linearly with increasing atomic number (Fig.6). We also observe that ln(DREE) exhibit steeper slope formore albitic plagioclase, in accordance with increasingly diver-gentDLREE vs. DHREE in Figure 2. Since Drake’s experimentswere conducted in air, Eu is almost entirely trivalent (e.g.,Drake, 1975; Wilke and Behrens, 1999), and behaves like therest of the REE. However, a very small Eu positive peak is stilldiscernible and is likely to be due to a small portion (;1%,

Wilke and Behrens, 1999) of total Eu that is present as Eu21

and partitions 20 to 100 times more efficiently into plagioclase,causingDEu

(total) be higher. On the contrary, the negative devi-ation of Ce from the straight line may be related to the fact thatCe is partly tetravalent, and is less favored by the plagioclasestructure. Schreiber et al. (1980) found that Ce41 can constituteup to 10–20% of total Ce in different basaltic systems inexperiments conducted in air.

Using Blundy and Wood’s (1994) model we are able toreproduce a linear drop inDREE with REE atomic number forfour plagioclase compositions (Fig. 6), which clearly demon-strates that REE at doped concentrations partition into theMsite. We conclude that a linear drop in ln(DREE) with REEatomic number is likely to be a general pattern of plagioclase-melt partitioning, and is not an artifact of our measurements ora specific result of the Drake experiments. Significantly, weobserve that the pattern of ln(DREE) becomes steeper withdecreasing anorthite content in plagioclase (Figs. 6 and 3) forboth measurements and calculations.

The existence of a linear ln(DREE) vs. ionic radius depen-dence is not an expected result. Most previous experiments andbulk phenocryst/matrix studies produced a progressively shal-lowing concave-up ln(DREE)-atomic number pattern, and thispartition behavior seems to be taken for granted by many users.However, such apparent partitioning behavior can be an artifactcaused by two offsetting errors. It is likely that there was anoverestimation of HREE in plagioclase (because of analytical

Fig. 5. An Onuma diagram, with near parabolic curves fitted throughpartition coefficients of the most anorthite-rich (thick lines) and anor-thite-poor (thin line) of the analyzed runs (Bindeman et al., 1998). Thetop of each parabola indicate the size of the site (M site), r0, and thepartition coefficients,D0, of the strain free substitution (e.g., Onuma etal., 1968). Equation 2 from Blundy and Wood (1994) was used fornonlinear fitting with three free parameters:r0, D0 and the Youngmoduli,E. Notice that the parabolas shift to the right and up, indicatingthe increase inr0 andD0 with decreasing %An. Vertical arrows showthe magnitude ofDi increase for trace elements at each ionic radius.Di

of ions whose radius is closer tor0 (e.g., Na) do not show a largeincrease inDi; the larger ions Rb and La show a significant increase inDi, while the smaller ions Li and Lu show a decrease inDi withdecreasing %An. This explains the lnDi vs %An seen in Figs. 2–4 andBindeman et al. (1998).

Fig. 4. Comparison of measured (filled symbols) and calculated[using the Blundy and Wood (1994) approach, open symbols] slopes ofthe RT ln(Di) 5 a XAn 1 b dependence for monovalent, divalent andtrivalent cations as a function of their ionic radii at eightfold coordi-nation (from Shannon, 1976). Note that the dependence becomessteeper with increasing valence number, meaning thatDi of highercharged ions are more sensitive to the anorthite content of plagioclase.


interferences and concentrations near the detection limit) inearlier electron microprobe-based experimental partitioningstudies, including Drake’s. On the other hand, in phenocryst/matrix bulk determinations of partition coefficients, microin-clusions of apatite, zircon and other accessory minerals as wellas clinopyroxene and amphibole are often present in plagio-clase and it is difficult to separate inclusions from plagioclaseusing conventional techniques. These minerals are stronglyenriched in HREE relative to LREE (e.g., Watson and Harri-son, 1984) and this may lead to overestimation ofDHREE inbulk measurements of natural plagioclases. However, someexperimental and volcanic studies do show a straight line inln(DREE)-atomic number coordinates. McKay et al. (1994) andSimon et al. (1994) found similar patterns between anorthiteand CAI-type melt in doped experiments analyzed by ionmicroprobe. Phenocryst/matrix partition coefficients by Phin-

ney and Morrison (1990) derived from neutron activation anal-yses also plot as a straight line.

3.4. Difference BetweenDREE in REE-Undoped andREE-Doped Samples

The analyzedDREE for undopedREE at natural concentra-tion levels determined on undoped and Sr-doped runs areplotted in Fig. 7 and compared withdoped DREEin REE-dopedruns as a function of REE atomic number. We find that un-doped REE in runs doped with Sr behave similar to REE inundoped runs (Fig. 7), signifying that there is no influence ofvariable concentrations of Sr on REE partitioning. For the threestudied plagioclase compositions,Di’s of undoped REE arehigher than that of doped REE. The difference inDi’s is notlarge for some LREE, but there is a tendency of ln(DREE) of

Fig. 6. MeasuredDREE for REE-doped runs compared withDREE calculated using the Blundy and Wood (1994) model;see Fig. 2 and text for discussion. Note the linear drop inDREE in both calculated and measured results and the steepeningof the slope with the decreasing An content of plagioclase.


undoped REE be systematically higher for several studiedplagioclase compositions for the whole spectrum of REE (Fig.7). Therefore, the difference between otherwise equivalent runs(same temperature and major element composition, same ionmicroprobe analytical conditions, but different doping element)is significant. Averaged undopedDREE are directly comparedto REE-dopedDREE in these equivalent runs as a function ofeach REE concentration (Fig. 8a). The;30 to 100% increasein DREE with decreasing concentration (and its slope) aresimilar for different REE in each run, and in between differentruns. The increase does not changeDREE relative to one an-other, and undopedDREE obey the same Onuma diagramrelationship as dopedDREEin equivalent runs (Bindeman et al.,1998). This regular and parallel fashion ofDREE increase withdecreasing concentration suggests that REE are partitionedstructurally at doped and natural concentration levels. (SeeTables 2–4.)

Since Drake normally doped his runs with 3 to 4 selectedREE, theDi’s of other,undopedREE in these REE-doped runsis interesting to consider (Fig. 9). La–Ce–Y–Lu, Sm–Eu–Gd,and Nd–Dy–Er run series were conducted (see Table 1) foreach plagioclase composition. For example, it is possible tocompareDSm in the La–Ce–Y–Lu and Nd–Dy–Er doping se-ries where Sm is at natural concentrations withDSm in theSm–Eu–Gd series in which Sm was a doping element. The goalis to check if doping with selected REE changesDREE of thewhole spectrum (including undoped REE in REE-doped runs)or the undoped REE follows the same partition behavior as inREE-undoped runs. In particular, if an undoped REE (X) in arun doped with other REE gives a lowerDX for doped con-centrations (as it would if the run were doped with rare earthelementX), the REE spectrum will be smooth. IfDX turns outto be high, as it would in a run not doped with any REE, apositive anomaly would appear in the REE spectrum at thatpoint. Figures 8b and 9 presentDREE for both doped andundoped REE in each run series for three plagioclase compo-sitions. Despite the significant analytical uncertainty, we ob-serve thatDREE values tend to follow the same pattern as fordoped REE. We conclude that there is an influence of dopingwith REE onDREE values for undoped REE. By comparingFigure 7 with Figure 9, or more clearly by comparing Figures

8a,b, it can be seen that doping with REE strongly subdues theeffect of undopedDREE increase with decreasing concentra-tion.

3.5. Partition Coefficients of Trace Elements in RunsDoped with REE, Y, Sr, and Ba and in UndopedRuns

Figure 10 presentsDi’s for other analyzed trace elements inorder to address the question of whether doping with REE or SrinfluencesDi’s of other trace elements at their natural concen-tration level. The position of a trace atom within a crystalstructure has been discussed in the literature. In order to min-imize near-order charge and size distortions caused by thesubstituting trace element, trace atoms have been proposed toform clusters with other trace atoms of a suitable size andcharge to compensate for the poor fit of the trace atom into thecrystal structure; even an embryonic phase consisting of severaltrace atoms has been proposed (Navrotsky, 1978; Urusov andDudnikova, 1998). In particular, the small ion Li was cited asa potential candidate to compensate for charge and size (Mooreand White, 1974).

We do not detect any influence on the presence or identity ofdoping elements on the partition coefficients of undoped ele-ments (apart from the REE discussed above). Remarkably, thisis true for elements with a variety of valences and ionic radii.This is especially clear for relatively abundant elements that aredetermined with high precision with the ion microprobe, suchas Ti, Mg, Fe, K, Sr, and Ba.

Considering the doped elements Sr and Ba, we also see nodifference inDSr between REE-doped, Sr-doped or undopedruns for several plagioclase compositions (e.g., Bindeman etal., 1998). This implies that from the;800 ppm (natural) to upto 11,000 ppm (doped) level the partitioning behavior of Srfollows the same Henry’s law constant. In one analyzed Ba-doped sample, we also found noDBa dependence on concen-tration or doping element (;100 ppm natural vs. 5500 ppmdoped), confirming electron microprobe results of Drake (1972).

Electron microprobe analyses of undoped samples and thosedoped with Sr, Ba or REE, show that feldspars containing up to0.5 wt.% (;0.3 mol%) of total REE remain stoichiometric

Fig. 7. REE and Y partition coefficients (61 s) at natural concentration levels of REE (;0.3–3 ppm) measured onSr-doped runs and undoped runs compared with those measured at doped concentration level in REE and Y doped runs.


Fig. 8. Partition coefficients of REE at dopedconcentrations (filled circles) vs. that of REE atnatural concentrations (empty circles) in equiva-lent runs. (a) Doped REE vs. REE at natural con-centrations in REE-undoped runs (determined onundoped and Sr-doped run products, open circles).(b) Doped REE (filled circles) vs. REE at naturalconcentrations in REE-doped runs (open circleswith crosses). Analyses are from Table 2 of Bin-deman et al. (1998) and from Drake and Weill(1975). Note thatDi at natural concentrations arealways higher for all REE; in runs doped with REE(b), undoped REE show less less-pronounced in-crease inDi with decreasing concentration than doDi values in runs not doped with any REE. Un-certainties are from Table 4.


within analytical accuracy (assuming that all these cations enterthe M site). No feldspars were found to be deficient inM-sitecations.

4. DISCUSSION

We observe thatDi’s of analyzed trace elements other thanREE are unaffected by the presence or identity of dopant

elements. All other analyzed trace elements do not show anychange in partitioning behavior as a result of doping with one,three, or even four different REE, Sr, or Ba. This argues againstsignificant REE coupling with other trace elements in partition-ing. It is improbable for a doped REE to encounter manymonovalent trace elements in the vicinity of each trivalent REEatom to achieve a charge balance; the most likely monovalent

Fig. 9. Partition coefficients of doped and undoped REE and Y. Partition coefficients at natural concentration levels ofREE measured on Sr-doped and undoped run series are shown as shadowed areas. Solid symbols indicate doped REE;corresponding open symbols indicate undoped REE. Note that undopedDREE’s tend to cluster around dopedDREE’s inREE-doped run series.

Fig. 10. Partition coefficients of trace elements between plagioclase and melt as a function of %An and doping element.Filled squares and solid regression lines are Di determined earlier on Sr-doped samples (see Bindeman et al., 1998). Circleswith dots are partition coefficients for undoped runs. Triangles are Di for REE and Y-doped runs. Note that there is nodifference in partition coefficients as a function of the doping element. We take this as the evidence against trace elementscoupling between themselved in order to compensate for charge and size misfits of the dopant.


charge-balancing ion is Na1, because of its much higher con-centration. Therefore, major elements Ca and Na are the mainparticipants in the exchange reactions with REE.

The difference in values betweenDREE in undoped vs. thosein REE-doped experiments (Figs. 7–9), suggests that the addi-tion of wt.% of REE to the plagioclase-basalt system affects theDi’s of REE and Y. Due to a geochemical similarity betweendifferent REE elements, adding selected REE changes (de-creases) partition coefficients ofall REE in these runs, eventhat of other undoped REE (e.g., Figs. 8b and 9). This mayindicate that doping with REE leads to their partition into sitesof similar size and charge that are suitable for REE and REE-like elements in plagioclase structure. The partition at dopedconcentration is structural and obeys predicted behavior forM-site partitioning (see Figs. 2, 5, and 6). However, partition atnatural concentration level also obeys Onuma diagram withMsite as a site of preferred substitution (see above and Bindemanet al., 1998). This implies that it is the plagioclase structure thatis primarily responsible for the difference inDi’s at differentconcentrations.

The partition coefficientDi of the plagioclase-melt traceelement exchange reaction with a constantK can be expressedthrough the Henry’s law constants of the trace element in themelt (kL) and plagioclase (kPl) (e.g., Wood and Fraser, 1978):

Di 5 K~kL!/~kPl).

Because we consider equivalent experimental runs,K is ex-pected to stay constant (for each exchange reaction). Oneexplanation of higherDi’s at natural concentrations is thatadding a few wt.% of REE in the melt may cause a significantdecrease inkL, thus affectingDi’s. However, this is less likelysince the heat of fusion of fictive REE-feldspar is much greaterthan the heat of its incorporation into the melt (e.g., Wood andBlundy, 1997). The other explanation, is that adding a fewthousand of ppm of total REE to plagioclase increaseskPl, and,correspondingly, decreasesDREE’s. However, in a regular so-lution model of REE substitution for Ca and Na at magmatictemperatures, the REE–Ca interaction parameters are not ex-pected to be more than 2–3 kcal/mole and a concentrationchange from ppm levels to 3 wt.% will produce less than a;10% change in partition coefficients (e.g., Beattie, 1993).Therefore, in order forDi’s at natural concentrations to besignificantly larger (30–100%), different substitution mecha-nisms must be considered.

Feldspar synthesis studies demonstrate possible substitutionmechanisms for trace elements in plagioclase. A variety ofsynthetic endmembers have been synthesized with B, Ga, Fe,Ni, Ge, Fe, Mg, and P entering the tetrahedralT site (Bychkovet al., 1989; Fleet et al., 1988; Galoisy and Calas, 1992), andRb, NH4, Tl, Sr, Ba, Pb, Eu, La (and other REE) in seven- tonine-coordinatedM site (Kneip and Liebau, 1994; D’Arco andPiriou, 1989). For REE substitution, synthesis experiments andcharge balance suggest that the following reactions operate:

2 Ca21 5 REE31 1 Na11, (1)

3 Ca21 5 2 REE31 1 vacancy, (2)

2 Ca21 5 REE31 1 Trace11. (3)

Substitution in theM site may be accompanied by theT-sitesubstitutions:

Ca21 1 Al31 5 Na11 1 Si41, (4)

Ca21 1 Si41 5 REE31 1 Al31. (5)

A significant REE–Ca solid solution has been produced anddemonstrates the feasibility of substitution mechanism (Eqn.1). Substitution mechanism (Eqn. 5) was shown to be lesslikely due to the Al-tetrahedral avoidance rule, however Isma-tov et al. (1985) synthesized REE-anorthite with Al/Si. 1,which shows a limited possibility of this mechanism. Remark-ably, the synthesized REE-bearing feldspar can tolerate a sig-nificant proportion of vacancies (Kimata, 1988; Kneip andLiebau, 1994), confirming the feasibility of substitution mech-anism (Eqn. 2). We argued above against the possibility thatexchange reactions with monovalent trace elements at naturalconcentration (Eqn. 3) serve as a mechanism of charge balanc-ing of REE. Since we cannot detect stoichiometric deficienciesor excesses in Drake’s experiments, we suggest that becausethe experiments used natural basalts, the availability of othertrace and minor elements, such as Fe, Mg, and Ti, may com-pensate for vacancies (e.g., Watson, 1985).

In order to account for higherDi’s at natural concentrations,we speculate that REE at natural concentrations may be in-volved in several substitution mechanisms (e.g., Eqns. 2 and 5),some of which can operate only at low concentrations. With theincrease in REE concentrations, the low-concentration equilib-ria become saturated and the relative importance of mecha-nisms other than Eqn. 1 vanishes. In particular, defect equilibria(Eqn. 2) may be saturated and Al-avoidance does not allowreaction 5 to operate, at higher concentrations.

Urusov and Dudnikova (Eqn. 6, 1998) and Harrison andWood (Eqn. 5, 1980) showed that when a defect equilibria aretaken into account with a certain constant of the reaction, alongwith a heterovalent trace element exchange reaction, theDi isan algebraic function of two constants of these reactions. Theresulting equations yield plateau of higherDi’s at naturalconcentrations, and plateau of lowerDi’s at doped concentra-tions, and a narrow transition zone. We suggest that the sameresult can be derived not only for defect equilibria, but also forthese exchange reactions operating at low concentrations,yielding two (or more) plateaus ofDREE as a function of REEconcentration. The same may also be true for other traceelements, especially those involved in heterovalent substitutionwith several possible reactions, but this seems not to be the casewith the homovalently substituting cations like Sr and Ba.Specifically designed experiments involving trace elementsother than REE and conducted equivalently at doped and un-doped concentrations may be necessary to clarify this questionin the future.

We conclude that partition coefficients determined at naturalconcentration levels of trace elements of the present study andreported in Bindeman et al. (1998) are to be preferred forgeochemical use. In this paper we discuss in details the role ofREE-doping on decreasing DREE by 30–100% and possiblereasons for this effect. We emphasize that the natural range ofconcentrations of REE in plagioclase (tholeiites to alkali-richpegmatites) is much smaller than that of doped experiments.Therefore, ourDREE measured at natural concentration levels


in experiments that are not doped with REE are most appro-priate to use.

Acknowledgments—We are grateful to Professor M. J. Drake for pro-viding samples of experiments for this study and for fruitful discussionsand to J. D. Blundy for recommendations on the approximation tech-nique. This work was supported by NASA Grant No. NAG5-4298 (toA.M.D.), and NSF Grant No. EAR 9417787 to A. T. Anderson.

REFERENCES

Angel R. J., Hazen R. M., McCormick T. C., Prewitt C. T., and SmythJ. R. (1988) Comparative compressibility of end-member feldspars.Phys. Chem. Minerals15, 313–318.

Beattie P. (1994) Systematics and energetics of trace-element partition-ing between olivine and silicate melts: Implications for the nature ofmineral/melt partitioning.Chem. Geol.117,57–71.

Beattie P. (1993) The occurrence of apparent non-Henry’s law behaviorin experimental partitioning studies.Geochim. Cosmochim. Acta57,47–55.

Bindeman I. N., Davis A. M., and Drake M. J. (1998) Ion microprobestudy of plagioclase-basalt partition experiments at natural concen-tration level of trace elements.Geochim. Cosmochim. Acta62,1175–1193.

Blundy J. D. and Wood B. J. (1994) Prediction of crystal-melt partitioncoefficients from elastic moduli.Nature372,452–454.

Blundy J. D. and Wood B. J. (1991) Crystal-chemical control on thepartitioning of Sr and Ba between plagioclase feldspar, silicate melts,and hydrothermal solutions.Geochim. Cosmochim. Acta55, 193–209.

Brice J. C. (1975) Some thermodynamic aspects of the growth ofstrained crystals.J. Cryst. Growth28, 249–253.

Bychkov A. M., Kotel’nikov A. R., Romanenko I. M., and SenderovE. E. (1989) Effect of isomorphic replacement of silicon by phos-phorus on structural peculiarities of feldspars.Geochem. Int.2,310–312.

D’Arco P. and Piriou B. (1989) Fluorescence spectra of Eu31 insynthetic polycrystalline anorthite: distribution of Eu31 in the struc-ture.Am. Min.74, 191–199.

Drake M. J. (1972) Ph.D. dissertation, University of Oregon.Drake M. J. and Holloway J. R. (1978) “Henry’s law” behavior of Sm

in a natural plagioclase/melt system: Importance of experimentalprocedure.Geochim. Cosmochim. Acta42, 679–683.

Drake M. J. and Weill D. F. (1975) Partition of Sr, Ba, Eu21, Eu31, andother REE between plagioclase feldspar and magmatic liquid: Anexperimental study.Geochim. Cosmochim. Acta39, 689–712.

Drake M. J. (1975) The oxidation state of europium as an indicator ofoxygen fugacity.Geochim. Cosmochim. Acta39, 55–64.

Dudais M. J., Schmitt R. A., and Harward M. E. (1971) Trace elementpartitioning between volcanic plagioclase and dacitic pyroclasticmatrix. Earth Planet. Sci. Lett.11, 440–446.

Dunn T. and Sen C. (1994) Mineral/matrix partition coefficients fororthopyroxene, plagioclase, and olivine in basaltic to andesitic sys-tems: A combined analytical and experimental study.Geochim.Cosmochim. Acta58, 717–733.

Fleet M. E. (1988) Tetrahedral-site occupancies in reedmergnerite andsynthetic boron albite (NaBSi3O8). Am. Mineral.77, 76–84.

Francalanci L. (1989) Trace element partition coefficients for mineralsin shoshonitic and calc-alkaline rocks from Stromboli Island (Aeo-lian Arc). Neues-Jahrbuch-fuer-Mineralogie, Abhandlungen.160,229–247.

Fujimaki H., Tatsumoto M., and Aoki K. (1984) Partition coefficientsof Hf, Zr, and REE between phenocrysts and groundmasses.J.Geophys. Res., (Suppl.)89, B662–B672.

Galoisy L. and Calas G. (1992) Network forming nickel in feldspar.Eos Trans.73, 361.

Harrison W. and Wood B. J. (1980) An experimental investigation ofthe partitioning of REE between garnet and liquid with reference tothe role of defect equilibria.Contrib. Mineral. Petrol.72, 145–155.

Higuchi H. and Nagasawa H. (1969) Partition of trace elements be-tween rock forming minerals and host volcanic rocks.Earth Planet.Sci. Lett.7, 281–287.

Hinton R. W., Davis A. M., Scatena-Wachel D. E., Grossman L., andDraus R. J. (1988) A chemical and isotopic study of hibonite-richrefractory inclusions in primitive meteorites.Geochim. Cosmochim.Acta 52, 2573–2598.

Hoover J. D. (1978) The distribution of samarium and tulium betweenplagioclase and liquid in the systems Di-An, Ab-An-Di at 1300°C.Carnegie Inst. Wash. Yrbook.77, 703–709.

Ismatov A. A., Yunusov M. Yu., and Nasyrova D. S. (1985) Solidsolution of the composition Ca7TrAl17Si15O64 having the anorthitestructure.Inorganic Materials21, 576–578.

Kimata M. (1988) The crystal structure of non-stoichiometric Eu-anorthite: an explanation of the Eu-positive anomaly.Min. Mag.52,257–265.

Kneip H.-J. and Liebau F. (1994) Feldspars with trivalent non-tetrahe-dral cations: Experimental studies in the system NaAlSi3O8–CaAl2Si2O8- LaAl3SiO8. Eur. J. Min.6, 87–98.

MacPherson G. J. and Davis A. M. (1994) Refractory inclusions in theprototypical CM chondrite, Mighei.Geochim. Cosmochim. Acta58,5599–5625.

McKay G. A., Le L., Wagstaff J., and Crozaz G. (1994) Experimentalpartitioning of rare earth elements and strontium: constraints onpetrogenesis and redox conditions during crystallization of Antarcticangrite Lewis Cliff 86010.Geochim. Cosmochim. Acta58, 2911–2919.

Moore R. J., and White J. (1974) Equilibrium relationships in thesystems Li–Co–O and Li–Ni–O.J. Mat. Sci.9, 1401–1408.

Mysen B. O. (1978) Limits of solution of trace elements in mineralsaccording to Henry’s Law: Review of experimental data.Geochim.Cosmochim. Acta42, 871–885.

Nagasawa H. and Schnetzler C. C. (1971) Partitioning of rare earth,alkali, and alkali earth elements between phenocrysts and acidicigneous rocks.Geochim. Cosmochim. Acta35, 953–968.

Nash W. P. and Crecraft M. R. (1985) Partition coefficients for traceelements in silicic magmas.Geochim. Cosmochim. Acta49, 2309–2332.

Navrotsky A. (1978) Thermodynamics of element partitioning: (1)systematics of transition metals in crystalline and molten silicatesand (2) defect chemistry and “the Henry’s Law problem.”Geochim.Cosmochim. Acta42, 887–902.

Onuma N., Higuchi H., Wakita H., and Nagasawa H. (1968) Traceelement partition between two pyroxenes and host lava.EarthPlanet. Sci. Lett.5, 47–51.

Peters M. T., Shaffer E. E., Burnett D. S., and Kim S. S. (1995)Magnesium and titanium partitioning between anorthite and Type BCAI liquid: Dependence on oxygen fugacity and liquid composition.Geochim. Cosmochim. Acta59, 2785–2796.

Petrov I., Agel A., and Hafner S. S. (1989a) Distinct defect centers atoxygen positions in albite.Am. Mineral.74, 1130–1341.

Petrov I., Bershov L. V., Hafner S. S., and Kroll H. (1989b) Order-disorder of Fe31 ions over the tetrahedral positions in albite.Am.Mineral. 74, 604–609.

Phinney W. C. (1992) Partition coefficients for iron between plagio-clase and basalt as a function of oxygen fugacity; implication forArchean and lunar anorthosites.Geochim. Cosmochim. Acta56,1885–1895.

Phinney W. C. and Morrison D. A. (1990) Partition coefficients forcalcic plagioclase; implications for Archean anorthosites.Geochim.Cosmochim. Acta54, 1639–1654.

Schnetzler C. C. and Philpotts J. A. (1970) Partition coefficients of rareearth elements between igneous matrix material phenocrysts II.Geochim. Cosmochim. Acta34, 307–340.

Phinney W. C. and Morrison D. A. (1990) Partition coefficients forcalcic plagioclase; implications for Archean anorthosites.Geochim.Cosmochim. Acta54, 1639–1654.

Schreiber H. D., Lauer H. V., Jr., and Thanyasiri T. (1980) The redoxstate of cerium in basaltic magmas: An experimental study of iron-cerium interactions in silicate melts.Geochim. Cosmochim. Acta44,1599–1612.

Shannon R. D. (1976) Revised effective ionic radii and systematicstudies of interatomic distances in halides and chalcogenides.ActaCrystallogr.A32, 751–767.

Simon S. B., Kuehner S. M., Davis A. M., Grossman L., Johnson M. L.,and Burnett D. S. (1994) Experimental studies of trace element


partitioning in Ca, Al-rich compositions: Anorthite and perovskite.Geochim. Cosmochim. Acta58, 1507–1523.

Urusov V. S. and Dudnikova V. B. (1998) The trace-componenttrapping effect: Experimental evidence, theoretical interpretation,and geochemical applications.Geochim. Cosmochim. Acta62,1233–1240.

Urusov V. S. and Kravchuk I. F. (1978) Trapping effect of microim-purity by the defect sites and its geochemical significance.Geochem.Int. N7, 963–978.

Vernieres J., Joron J.-L., Treuil M., Coulon C., and Dupuy C. (1977)Coefficient de partage de quelques elements en trace entre plagio-clase et verre dans les ignimbrites—Implications petrogenetiques.Chem. Geol.19, 309–325.

Watson E. B. (1985) Henry’s Law behavior in simple systems and inmagmas: Criteria for discerning concentration-dependent partitioncoefficients in nature.Geochim. Cosmochim. Acta49, 917–923.

Watson E. B. and Harrison T. M. (1984) Accessory minerals and the

geochemical evolution of crustal magma systems: A summary andprospectus of experimental approaches.Phys. Earth Planet. Int.35,19–30.

Wilke M. and Behrens H. (1999) The dependence of the partitioning ofiron and europium between plagioclase and hydrous tonalitic melt onoxygen fugacity.Contrib. Mineral. Petrol.137,102–114.

Williamson J. H. (1968) Least-squares fitting of a straight line.Can. J.Phys.46, 1845–1854.

Wood B. J. and Blundy J. D. (1997) A predictive model for rare earthelement partitioning between clinopyroxene and anhydrous silicatemelt. Contrib. Mineral. Petrol.129,166–181.

Wood B. J. and Fraser D. G. (1978)Elementary Thermodynamics forGeologists. Oxford University Press, 303 p.

Worner G., Beusen J. M., Duchateau N., Gijbels R., and SchminckeH.-U. (1983) Trace element abundances and mineral/melt distribu-tion coefficients in Laacher See Volcano (Germany).Contrib. Min-eral. Petrol.84, 152–173.


trace element partitioning between plagioclase and melt

Documents