37
The Canadian M ine ralo g i st
Vol. 34, pp.3748 (1996)
STANDARDS
MICRO.PIXE
ANALYSISOFSILICATE
REFERENCE
FORTRACENi, Gu, Zn, Ga,Ge, As, Rb, Sr, Y Zr, Nb, Mo AND Pb,
WITH EMPHASISON Ni FORAPPLICATION
OFTHENi.IN.GARNET
GEOTHERMOMETER*
JOHN L. CAMPBELL anp WILLIAM J. TEESDALE
Guclph-Waterlno
ProgramforGraduate
Workin Plrysics,
Universityof GuelplaGuelphOnarto NIG 2Wl
BRUCEA. KIARSGAARD
Geolagical
Surveyof Canada601BoothStreet,Ottawa,OntarioKIA OE8
LOUIS J. CABRI
CANMET, 555Booth Street,Ottawa.Ontario KlA OGI
AssrlAsr
Trace-elementanalysisis a powerfrrltool for studyingnumerousprocessesof researchinterestin theEarth Sciences;aswell,
it hasimportantapplicationsof interestto the mineral explorationcommunity.Here,we presentthe resultsof a comprehensive
studyof five internationallyrecognizedsilicatereferencestandards(BIIVO-I, G)R-5, GSR-5, MAG-I andGSD-50) utilizing
micro-PDG analysis.Resultsof trace analysesfor Ni, Cu, 7a, Gu Cte,As, Rb, Sr, Y, Zr, Nb, Mo and Pb are presented.
Measureddata for the five standaxdsexhibit agreementwithin *107o, and commonly better, with recomrnendedlevels of
concenmtion in the 5G400 ppm range. Detection limits (for a typical 3- to zl-minuteanalysisof an accumulatedcharge
of 2.5miooCoulombs)areintherangeof
2-l0ppn,withprecisionsof l-l\Vo. AdetailedstudyoftheX-rayspectrainthe
visinity of the Fe, Ni and Co lines illustratesthat the GUPD( softwareis able to correctly fit the complex spectraand thus
provide accurateNi concentrations.Further tests on Ni in gamet are repored, in the context of curent application of the
Ni-in-garnetgeothermometer
in diamondexploration;it appearsimportantto resolvethe differencebetweenthe two published
calibrationsof this importantgeothermometer.
Keywords:micro-PD(Eanalysis,geostandards,
traceelements,acccuracy,glasses,garnet,Ni-in-gamet geothermomeler.
Sovnrens
Lanalyse chimiquede matdriauxpour leur teneuren 6l6mentstracesconstitueun outil puissantdansl'6tudedesnombreux
processusgdologiques,et sesapplicationsdansles programmesd'explorationmini&e sontimportatrtes.Nous pr6sentonsici les
G)R-5, GSR-5, MAG-I et
r6sultatsd'une6tudeint6gr6ede cinq dtalonsde silicate i circulation intemationale@IM-l,
GSD-50) utilisant la m6thodemicro-PD(E.Nous prdsentonsles r6sultatsd'analysespour le Ni, Cu, Zn, Ga, Ge, As, Rb, Sr, Y,
Z,r,lr[b, Mo et Pb. I,es concentrationsnesur6esconcordenti LIVoprds,et dansplusieus cas,i moins que l0Vo pr0s,avecles
concentrationsrecomrnand6es,
dansI'intervalle 5H00 ppm. I,e seuil de ddtection,pour le cas d'une a:ralysetypique, d'une
dur6ede 3 a 4 minute,set ayantune chaxgeaccumul& de 2.5 microCoulombs,est entre2 et 10ppm, avecune pr6cisiond'entre
I et I07o.Une 6tudeddtaill6edesspectresde rayonsX dam la rdgion desraiesFe, Ni et Co montreque le logiciel GUPD( peut
reproduirecorrectementla complexitddesspectres,et ainsi meneri une d6terminationpr6cisede la concentrationdu Ni. Nous
prdsentonsles r6sultatsd'une6valuationde la teneuren Ni d'6chantillonsde grenat,dansle contextedu gdothermombtre
fond6
sur cet 6l6men! et de sesapplicationsI I'explorationpour le diamanl Il semblemaintenantessentielder6soudreles diff6rences
entrelqs deuxcalibragesde ce g€othermomdffeimportant d6ji dansla litt6rature.
(Iraduit par la R6daction)
Mots-cl6s:analysemicro-PD(E,6talons,6l6mentstraces,justesse,verres,grenat,gdothermom6trie.
* GeologicalSwvey of Canadacontributionnunber 24795.
38
TIIE CANADIAN MINERALOGIST
ItrxopucnoN
Mcrobeam face-elementtechniquesarewell suited
for analysis of minerals and glasses(or their microcrystalline equivalents).In this respect micro-PD(E,
a non-desfiuctivetechnique,allows specimensto be
studied using both electron and proton microprobe
analysis for major, minor and trace elements.
Apptcations of the datainclude studiesof solid-liquid
andliquid-liquid partition coefficientsfor petrogenetic
trace-elementmodeling of magmaticprocesses(for an
up-to-datereview, see Foley & van der I-aan L994).
Mcro-PD(E analysisof experimentalrun-productsto
determinetrace-elementpartitiening includesthe work
of Sweeneyet al. (7992), Adam e/ aI. (1993) and
Vicenzi et al. (1994).Mcro-PD(E mineral-melt partitioning studies have also been undertakenin glassy
volcanicrocks (e.9.,Stimac& Hickmott l994,Ewm
& Green1994).Another main useof micro-PD(Eis to
characterizetrace-elementsignaturesof mineralphases
from specific types of magmaticor ore-fornringfluid
systems. Relevant micro-PD(E studies on apatite,
zircon,titanite,carbonatesandsulfidesarereviewedby
Halden et al. (1995). TmFortantmicro-PD(E studies
have also been undertakenon a variety of mantle
phases(e.9.,olivine, gamet,spinel, clinopyroxeneand
orthopyroxene)for characterizationof tace-element
concentation (Moore et al. 1992,Gnffin et al, L989).
Applications of this work include the garnet/olivine
Ni-exchangetlennometer (Grfnn et al. 1989,Griffrn
& Ryan 1995), utilized for sfudies of diamond
explorationand associatedmantle petrology (e.g., the
conditionsof diamondformation).
BAcKGRoUND
It wonuarroN
The partitioning ofnickel betweenchromianpyrope
Garne| andolivine in kimberlite- or lamproite-derived
mantle xenoliths exhibits a well-defined dependence
on temperatureofequilibration. The relative variation
of nickel content arnong grains of garnet in such
xenoliths is very much greater than that in olivine.
Griffin er al. (1989) and Griffin & Ryan (1995)
determinedthe Ni contentof garnet grains iz situ by
the micro-PD(Etechnique(Campbellet al. I989,Ryan
et al. 1990).This approachinvolves the recording in
energy-dispersion
modeof the X-ray spectrumemitted
under bombardmentby a 3-MeV beam of protons.
Misro-PD(E is well suitedto this applicationbecause
its low dsfsgfiealimit for nickel. A 5-minute
measurement
using some10 nA of currentfocusedin a
20 x 20 pm spot leads to a detectionlimit of about
5 ppm. For comparison,the concentation range is
0-150 ppm in the grains of interest. The work of
ffiffin et aI. (1989) and Griffin & Ryan (1995) has
resulted in the development of an empirical geothermometer.By assuminga constantnickel contentof
2900 ppm in mantleolivine, the "nickel therrnometet''
provides an estimate of equilibration temperature
solely on the basisofmeasurednickel concentrationin
garnet$ains derivedfrom the kimberlite or lamproite
host. Although the precise calibration of the thermometer is being challenged(cf, Canil 1994, Griffin &
Ryan 1995), the nickel thermometeris a powerful
addition to the array of tools availableto evaluatethe
diamond potential of garnet-bearingigneous formations. Furthermore,additionalftace-elementdata(e.g.,
Y, Zro Sr, Ga) obtained by PDG analysis of gamet
provide chemical signaturesinferred to be related to
specific processesin the mantle (e.9., metasomatism)
which can affect the likelihood of diamond preservation.
There are only a small number of micro-PD(E
facilities suitablefor the analysisof mineral specimens
around the world, and only a subset of these have
developedexperiencein the quantitative analysis of
mineral grains.Thereis, asyet, only a modestquantify
of documentationof the resultsof micro-PD(Eanalysis
of intemationally recoenized geochemicalreference
standards.Ryan er al. (1990) and Cz,amanskeet al.
(1993) have analyzed weil-chaxacterizedsilicate
glasses,including a number of fused geological
standardreferencematerials.
The determinationof ftace concentrationsof Ni in
silicates,however,presentsa specialchallenge.As the
PD(E specfrumof BHVO-I basalticglassin Figure I
demonsfiates,the very weak NiKa X-ray line lies on
the flank of the very intenseFeKB line. The intensity
ratio of about 100:1reflects the typical concentations
of severalpercentfor Fe andseveraltensofppm for Ni
associatedwith chromianpyrope.Even a small error in
the fitting of the Fe lines in the spectum or of the
underlying continuum background could cause
significant error in the iderred concentrationof Ni,
which hasimFortanlimFlicationsfor Ni thermomeffy.
The first purposeof this studyis to testthe accuracy
of micro-PD(E analysisof five referencestandardsfor
ten trace elements commonly used in petrogenetic
modeling. Our secondpu4)oseis to test our microPD(E methodologyfor the specific caseof Ni across
therangeof concentrationsgermaneto Ni thermometry
of gamet.
MsrHoDs
Sampledescription and.preparation
Five geochemical reference materials were
employed. These comprised the USGS materials
MAG-I (marinemud), GXR-5 (shale)and BHVO-I
(basalt), the Chinesematerial GSR-5 (soil), and the
artificial glass GSD-50. Two of the present authors
participatedin themicro-PD(Eanalysisof BIIVG-I by
Czamanskeet al. (1993), but a re-analysis was
consideredjustified by various minor improvements
effectedsincein our methodology.Samplesof the first
39
MICRO-PDG ANALY$S OF SILICATE REFERENCESTANDARDS
J
IIJ
z
z
1d
E
o
tr
ITJ
oo re
F
z
3
o
o
100
12
10
t4
t6
18
ENERGY(kev)
Hc. 1. Micro-PD(E specfiumof a silicate glasscontainingapproximately120ppm nickel, fitted asdescribedin the t€xt.
tbreematerialswere preparedby fusion on an iridium
strip heaier; in each case, three 0.25-9 samples of
powder were individually fused. The three resulting
glasseswere then groundtogetherin an agatemortar,
andfrom this material,0.25 g was removedand fused.
The 0.25 g product of the secondfusion was reground
and refused. Each fusion was done in air at a
temperature4.00oCbelow the melting point of iridium,
and lasted some 3-4 minutes. The pieces of glass
chosenfor micro-PD(E analysiswere taken from the
region firthest from the face that was in contactwith
the iridium strip, in order to minimize possible
contaminationby iridium or by tracemetalsin the strip.
Details of tle tusion of the USGS BHVO-I basalt
material have been published by Cz,amatskeet al.
(1993). The artificial glassGSD-50 is one of a set of
standardspreparedby the CorningGlassWorks for the
U.S. GeologicalSurvey (Myen e/ al. L%6). Portions
of eachglassweremountedin epoxyblocls, polished
and carbon-coatedto provide a conductingpath for the
beamof chargedparticles.
The practice adopted by Gladney & Roelandts
(1988, 1990)to generaterecommendedconcenfiations
for geologicalreferencematerialsis an iterative one.
The first stepis to calculate,for a given ftace element
in a given material,the meanand standarddeviationof
all availablepublishedconcentations derived from a
selectedset of analytical techniques.Any values that
depart from the mean by more than two standard
deviationsarethenrejected,following which the mean
and standarddeviation are recomputed.The values
recommendedby Myers et al. for the GSD-50 glass
were arrived at in a different m4nner. We have
thereforesubjectedthe raw dataof Myers et al. to tJie
procedureof Gladney& Roelandtsin order to provide
recommendedconcenhations;in doing so, we have
resfticted ourselvesto data listed by Myers et aI. w
"quantitative" and have excluded"semi-quantitative"
data.In the caseof GSR-S,Xie et al. (1989)provided
only a recommendedconcentrationwith no attached
uncertainfy;we havethereforeagainusedthe approach
of Gladney& Roelandtsto provide valuesof concen-
40
TIIE CANADIAN MINERAI,OGIST
tation and uncerainty consistentwith those of the
other standardsusedhere.
Micro-PIXE analysis
Micro-PD(E analysiswas conductedat the Guelph
Scanninghoton Microprobe, using a L2 nA beam of
3 MeV protons;the spotsizewas typically 15 pm2.An
Oxford InstrumeulsSi(Li) detector of approximately
80 mm2 area located 25 mm from the specimen
provided the most efficient possible geometry. The
delector'senergyresolution,expressedin the conventional manner as the full-width at half height of the
MnKcr line, was 145 eV. The I25 trtm.Mylarabsorber
normally usedto suppresslow-energybremsstahlung
background was supplemented by an aluminum
absorber(nominal thickness250 pm) whosefunction
was to reduce the iron contribution to the X-ray
specftum and thereby permit better statistics to be
amassedfor the trace element peaks than would
otherwisebe the case.
Spectrawererecordedat ten spotson eachspecimen
except for BIIVO-I, in which case fiive spots were
studied. The accumulated proton charge for each
specfiumwas 2.5 microCoulombs,and the time taken
was typically between3 and4 minutes.
Full details of the use of the NIST standard
referencematerial SRM 1155 (a molybdenum steel
alloy) to calibrate the systemme given by Campbell
et al. (1993) and by Czamanskeet al. (1993). T\e
essenceof the approach is that the 2.38VoMo n
the SRM determinesthe insftumentalcorstant for our
systenL and the &.SVo hon gtves a very accurate
determination of the thickness of the aluminum
absorber.
FIc. 2. The "top haf' digital
filter used to suppress
background,togetherwith
its effect upon a simple
spectrum.
Dan red.uction
Data were processedusing tle GIIPX software
package developed at Guelph for PD(E analysis
(Maxwell et al. L989,1995).The processinvolves two
main steps.First, a model specEumis fitted to the
measuredspectrumto ascertaintlte peak areas.Then
these areas are converted to element concentrations
using the measuredinsfrumentalconstant,the known
characteristicsof the detector and the absorber,and
maftix correctionsthat accountfor the role of fhe major
elementsin slowing down the incident proton$ and
at0enuatingthe emitted X-rays. In order to compule
thesematix corrections,GUPD( requfues,in addition
to its database,the major-elementconcentrationsin the
specimens;theseweretakenfrom the literature(Myers
et al. L976,Govindarajul989,Xie et aI 1989,Gladney
& Roelandts1988,1990).
X-ray
It is in the freament of the energy-dispersed
spectrumthat approacheswithin the PD(E community
tend to differ. Most codesbuild the model spectrum
using Gaussiatrpeaks superposedon a continuum
background;in some cases,an analytical expression
(e.9., a high-orderexponentialpolynomial) is used as
backgroundmodel, whereasin others,a peak-removal
algorithmis usedto generatethe background,which is
subsequentlyemployedin numerical form as part of
the model.The model spectrumis then matchedto the
fitting, using
measuredone by nonlinearleast-squares
chi-squaredasthe goodness-of-fitcriterion.
GUPX differs from the majority of codes in its
treatmentof continuum background.Its approachis
based on the properties of the simple digital filter,
shown in Figure 2, along with its effect upon a very
simplespectrum.This "top hat" filter methodis widely
4l
MCRO.PDG ANALY$S OF SILICATE REFERENCESTANDARDS
used in the freafinent of electron microprobe X-ray
spcctra(McCarthy & Schamber1981);its effect is to
suppre$slinear background entirely and to modify
Gaussianpeak shapes.I1 wiil be an effective remover
of backgroundover an entire specftumprovided that
the continuumis locally closeto being linear over any
region of width equivalentto the filter width; the latter
is chosento equal one full-width at half-height at the
specffumcenter.
The intensepeaks due to major elementssuch as
iron exhibit sipificant low-energy tailing. The peak
modelin GUPD(includesquasi-exponential
featuresto
describethese,but in the large-areadetector that is
necessaryfor tace-element analysis, tleir intensity
is observedto changeover a single day of measurement. Such changesare reportedalso by other PD(E
investigators(,arsson et al. L989,Ryan er al. 1990).
One way !o deal in general with the rather poorly
known tailing featuresoand in particular with their
time-dependence,is to de-emphasizetheir influence
upon the fit" We desqribein detail elsewhere(Maxwell
et al.1994) a procedureto do this. It augmentsthe
sonventionalstatisticalenor in channelcontentby an
amountthat diminishesexponentiallyfrom the centroid
on the low-energy side of intensepeaks, and propagatesthe augmentederror into the channelweights.An
altemative is to define, following Berrjar:r'ins1 al.
(1988)"an error in the thicknesso-fthe absorbingfilter,
and to propagatethis into tle weights; becausethis
enor is largest for low-energy peaks of the major
elements,it de-emphasizes
the influence of the entire
iron peaks upon the goodness-of-fi1.Although we
considerthe first approachpreferable,we observethat
the two methodsgive concentrationsfor nickel and
other fraceelementsthat differ by lessthan 1 ppm.
M
7a+L
C1r
335tlE
Z^
32*3
@
47!4
Ge
r
A8
4t2
Rb
Sr
Y
7r
Nb
Overall resultsfor thirteen trace ekments
Table I displays the entire set of results in terms
of the mean concentrationsover ten spots and the
correspondingstandarddeviations. The ratio (h) of
the standarddeviationto the error of a singlemeasurement provides a measureof the homogeneityof the
distribution of eachelementin eachmaterialanalyzed.
Table I includes the mean-h values obtained by
averagingover the traceelementsin eachspecimen.It
seemsthat the fusions achievedgood homogeneity.
There are two exceptions.In GSR-5, the value of h
found for Zr is 5.L, indicating that the fusion had not
successtullyredistributedthe Zr. ln GSR-5, G)R-5
andMAG-I, tracesof Ir areobserved,andits distribution is highly heterogeueous;the h is presumably
derived from the filament. The results were not
includedin the calculationofh.
Our results have been corrected for the loss of
volatiles (e.g.,bound water) during the fusion, and so
they may be compared directly with the nominal
values,which have been taken from the literature, as
indicatedin Table 1. The agteemenlis generallygood
with some specific exceptions.In the three materials
fused on the iridium strip, the measuredconcenfration
ofZt is high by 20 - I40Vo,suggestiveofZ contamination in the glass.This may b€ contastedto the cases
of BIIVG-I and GSD--50.where the measuredand
nominal resultsarein good agreementIn MAG-I, the
measuredconcentrationof. Za is low by a factor of
three,and the measuredconcentrationof Sr is high by
25Vo;we haveno explanationfor thesetwo anomalies
in MAG-I. Ge. Mo and Pb were detectedonlv in
MEATII'RED (M) AND NOMTNAL (N) CONCENTRATIONSOF fiilRTEEN TRACE EI-EfIEIVIS*
TABLE I.
M
RFsuLTs
NM
75tE
3y.t3/.
49*9
4Ot4
3615
5013
53t4
2A*2
nd
Pb
nd
h
1.0
M
40*4
a*3
26X2
55t6
d
nd
tJ(1.1)6
3813
/5.6*3.4
U!2
1.410.4
206*t2
90111
26*l
xn13
14.3*2.5
8t2
13,*4
206t5
25X2
155t6
t45t.2
nd
nd
nd
trd
r1t2
4!2
4ll3
212!4
39*2
11otE r?0x3
tt9x2
r6jt1,5 16i1.5
!3
2lEX4 r40t,l0 24+20
7.6!1.7 6.7!2.6 l3!2
Mo
M
a1 +7
0.9
MN
46X5
42x3
42t2
3tl
35t2
ad
0.411 94X2
!d
9.2*.r.2
3!t2
lLx2
149*6
9jl
403t? 64x3
l46tt5
39E*3
28t3
23.5t0.5 n.6xl'1 36l:1
126tL3
170t3 rD*21 4tx2
l1L1
l8*1
t9xz
41*3
33t2
nat
51t3
nd
0.8
l.o
53t8
30:t3
13{tt6
m.4!t.5
119t6
ltrt3
11313
2,*2
l2rX2
1361?
10Jt5
2tt2
54t7
46f'7
4L*6
n!5
4L*4
@*7
4614
46t3
42x2
46'*.t
r*5
t lvtra&redvalm m @ d dddad &t lsdou &r l0 spot& Noniml vatr8 e reaomded mcetrcim d endrrt
dqbim
I clad""f &Rooladrg(f990Il rfFdst (1989), Gtsdfft&RffledrE (198t),.MIlgBetd, (lto;'r&
trd&nn 0e46Zr
od€(L
42
THE CANADIAN
MINERALOGIST
{En
400
FF
Cu
300
120
90
200
60
o
100
30
ol
0
200
1@
300
60
E
g30
160
+,
Ga
45
l.}l
Rb
't20
80
ili*
E15
o
40
I
a
U
C
o
P
o
O
15
500
30
Sr
37s
40
60
45
120
80
an
E
=
160
60
a
Y
45
+
E
g
120 't50
90
400
250
't2s
30
F---fu
litl
U
15
T
a
0
0
125
250
375
500
250
Zr
200
150
#
60
45
+
Nb
40
,+,
2n
15
50
30
100
20
50
10
a
_
qq
l-cl
FIH
0
0
0
s0
0
100 150 200 2w
10
20
30
50
40
(ppm)
NominalConcentration
Ftc. 3. Graphicalcomparisonofmeasured and recommendedconcentrationsof Cu, Zn,
Ga, Rb, Sr, Y, Zr, Nb in five geochemicalreferencematerials.
GSD-50, wheregood agreementwithnominalconcentations was observed.The resultsfor Cu, Zn, Ga"Rb,
Sr, Y, ?, Nb are summarizedgraphicallyin Figure 3,
andthosefor Ni are discussedseparatelybelow.
Resultsfor nicl<el
Resultsfor Ni are summarizedin Table 2, againin
terms of the mean concenftationsand the standard
deviations from ten spots. The result for Ni is
influencedby whetheror not the neighboringelement
cobalt is included in the list of elementswhoseX-ray
lines are to comprise tle model spectrum. This
influencearisesbecausethe CoKBline is very closein
TABII2.
DEPENDB{G
ON TITBMANNR. OFlITItr{G COBALP
F{%)
tNilFo
AM
Co
BNi
C}{
Co
N6iel
M
N@irlt
Co
3.15
n
5.6r
6.5
u.513.7 39.214.0
r7!t3
13110
76.415.1 37.3i4.8
76.2X4.1 35.914.8
lltll
6t6
?5jt
nx3
2rt2
29.9trz
'56 ert f6 dsqtplion of fil
lroced@
.t B, c.
tRdio i! in ets of
t4m M relaltve to % Fq
OFMCELCTpn)
5.66
8.9
9,4
t4
4L2t4.L
183t56
4.9
11
47,7 !4,7
4r3t4.3
!18.7*6.3
0.0
q.6t4.6
40.4j4.1
118.7*5.6
45.7X4.6
3r8
16+15
1tt2
53it
a4tr.6
3ltl9
t21!2
6x2
0,0
55*6
S!
MCRO-PXE ANALYS$ OF SILICATE REFERENCESTANDARDS
energyto the NiKct,line. The nominal concentrationof
Co in the five glassesranges from 21 to 45 ppm.
Becausethe CoKcl line is superimposedon the lower
flank of the intenseFeKp line, the detectionlimit of Co
is very high (180 - 3110ppm). Nonetheless,
the fit
(referredto as fit A) in some casesreports non-zero
concentrationsof Co (10-40 ppm). If Co is then
omitted from the elementlist and the fit repeated@),
the observedconcentations of Ni decreaseby a few
ppm, as shownin Table 2.
Theseobservations,coupledwith the low levels of
Co in the five referencestandards,tend to supportthe
omissionof Co routinely in our fits to the PIXE specta
of the fusedglasses.However,in the caseof chromian
pyropegarnet thereis somedangerin adoptingthis as
a generalassumption.Analysis of mantlegametfor Co
hasbeenundertakenin only a few studies.Shimizu &
Allbgre (1978) determinedCo concentrationby SIMS
in garnet from 14 mantle xenoliths, and reported
concenftaiionsin the range 33 - 92 ppm Co. In
particular,they reported50 ppm Co in mantlexenolith
PHN1611.Merlet & Boudinier (1990) also studied
xenolith PHN1611, and indicated similar concenfrations of Co, in the rangeof 57-:75ppm (asdetermined
by high-precision wavelength-dispersionelectron
microprobe).Hemig et al. (1986)analyzed31 samples
of gamet from a variety of "cold" and 'tot" gamet
themolitexenolithsfor Co by SMS. Co concenfrationg
were found to rangefrom 66 b Ln ppm. They noted
that these levels are L0 to 50Vo higher than those
reportedby Shimizu & Allbgre (1978). Hervig et al.
(1986) suggestedthat the differencebetweenthe two
studiesis related to their use of a nonoptimal intermineral matrix-correction scheme, It can thus be
suggestedon the basisofprevious work that typical Co
concentations in mantle gamet are in the order of
3G-100ppm.
In micro-PD(E analysisof garnetsfor Co at these
levels of concenfration,it is important to assessthe
influenceof the CoKp X-ray line on the NiKclline. We
thereforemodified our fitting code to freat the CoKct,
and,l$ lines as representingtwo separale'oelements",
i.e.. the lines were not constrained in' the model
spectrumto maintain the intensity ratio quotedin the
literature (Maxwell et al. 1989). Any concentation
reportedvia CoKa canthenbe ignored,given the high
limit of detection.But for CoKp, the limit of det€ction
is 40 ppm, and so there is the possibility to observe
lower concentrationsof Co than in the conventional
approach.This approachis refened to as C; considering that the typical range of concentrationin
mantle gamet is 30-100 ppm (close to or greater
than the 40 ppm limit of detection), approachC is
recommended
for the analysisof mantlegarnet.In each
ofthe C-typefits to the glassspectra,the concentration
of Co is reportedat a level below the limit of detection,
andthe Ni value is slightly lower than in fits A andB.
The Ni datareportedin Table 1 arethoseof methodC.
43
The variation in Ni contentarnongthe three setsof
results gives some crude measureof the systematic
uncertaintycausedby the proximity of the cobalt and
nickel lines.
DscusstoN
Compartson of micro-PIXE
to other microbeamtechniques
ln common viith other anatytical techniques[e.g.,
X-ray fluorescence (XRF), instrumental neutronactivation analysis (INAA), inductively coupled
plasma - atomic emission spectroscopy(ICP-AES),
inductively coupled plasma - mass spectrometry
(ICP-MS)I and microbeam trace-elementanalytical
lschniques [e.9., secondary-ionmass spectrometry
(SMS), laser-ablationmassspectrometrywith inductively coupled plasma (LAM-ICP-MS), synchrofton
X-ray fluorescence(SXRFll, micro-PD(E has its own
inherent set of strenglhs and weaknessesin terms
of methodologyand application.Recentdiscussionsof
microbeamanalyticaltechniquesareprovidedby Reed
(1990)andGreen(L994).Onemajor strengthof microPD(E (as comparedto SIMS and LAM-ICP-MS) is
that it is a nondestructivetechnique. Furthermore,
micro-PD(E is essentially a fundamental-parameter
techniquethat requiresa singlenormalizationfactor or
instrumentalconstant,which may be determinedfrom
a simple few-element stlndard (e.9., NIST 1155, as
describedabove). In contast SMS instrumentsare
calibratedby utilizing working curvesdevelopedfrom
relevantstandards,Oncestandardsandworking curves
are obtained, however, SIMS has the advantageof
being able to analyze for a wide range of elements
(to ppm detectionlimits) with good spatial resolution
(Green 1994). The LAM-ICP-M$ lsshnique is
currently at an early stage of development(Green
1994).Insnumentcalibrationutilizes spikedreference
standards.Recent work (e.9., Jackson et al. t992,
Perkins et al. 1993, Fryer et aI. L995,Ludden et al.
1995, Jeffries et al. 1995) illusffate the potential for
analysis of glassesand a variety of mineral phases
for a numberof elementsat low levels of delection.
Trace elemcnts(exceptNi)
Both the precisionand the accuracyof micro-PD(E
may be estimatedfrom the data of Table 1. Precision
is expressedas the ratio (percentage)of standard
deviation to meanmeasuredconcentrationfor the ten
measurementson a given element in a given glass.
Inspectionshowsthat it is typically LVofot concentations around400 ppm, 2-37o for concenfrationsin the
vicinity of 100 ppm, and 5-L0Vofor concentrations
around50 ppm. Becausethe specimenshave a homogeneousdistribution of traceelements,the precisionis
determinedmainly by countingstatistics,andtherefore
44
if the measurement
time is increasedby a given factor
(at constant beam-cunent), then tlre precision will
improve as the squareroot of that factor. We have
est'matedthe accuracyof the determinationof concentration for each elementby the simple expedientof
averagingthe percentagedifferencebetweenmeasured
and nominal concenffationsover the five glasses,Two
types of data are omitted in this calculation.
Concentrationsbelow 10 ppm are omitted because
their statisticalerror intoduces too much uncertainty.
The few casesof obviouslarge discrepanciqsalso are
omitted:theseareZn in MAG-I" andboth Sr andZ in
the threeglassesfusedon the Ir suip. In the caseof Sr,
the measuredconcenfationsare on averagesome2J%o
high in thesethree caseso
but they differ by less than
LVofrom nominal concentrationsin the remainingtwo
glasses.For the elementsCa,7n, Ga"Ge,Rb, Sr, Y, Nb
and 7.r, the average determinedby this method is
x,LIVo;onTyone measurementis availablefor eachof
Ge,Mo andPb,but the 107oestimateholdsfor tle first
andthird of these.
Analysisfor nickel
Figure4 present$a summaryof the resultsfor nickel
concentations. For the standardsGSR-5. G)(R-5
and BHVO-I, the agreementbetweenmeasuredand
nominal concentrationsis excellent. The results for
MAG-I and GSD-50 are lower than the expected
values,but for the latter, there is overlapbetweenthe
uncertaintybars of the measuredand nominal concentration. It should be noied that the recommended
149
;
120
o.
c
100
wN6r(
,/
lu
F80
o
c
o
o60
!
fr
,/
I
f
t!2
o
E
_ 9xR-5
/GSD-50
#
r /*-.
GSR-5 t?/
MAG-1
/
4
0
20
zm
60
80
100
120
1&
RecommendedConcenMon (ppm)
Flc. 4. Graphicalcomparisonof measuredand recommended
Ni concentrationsin geochemicalreference mat€dals.
The regressionis y = -3.2 + 1.053X with corelation
coefficient= 0.988.
Ni @lr6hali@
MsgEed
BCR.I
t3
13 t 3
AGV.I
166
15 X2
Rf@€taL0ggo)
GSP.1
l0
121
6 X2
Rr@€ral(1990)
BIIVGI
Ryatr et aL(190)
131
Canm.&e 4 af (1993)
123
&ar@d6
et af (f93)
concenfrationsfor GSD-50 are basedupon measurements performedprior to 1976: these concentations
are therefore somewhatless well-founded than those
of the other referencematerials employed here. For
completeness,other publishedPD(E results for Ni in
geochemicalreferencestandardsaregiven in Table 3.
The micro-PD(E analysis for Ni merits more
extensivediscussion,partly becauseof the tecbnical
diffrculty that is evident in fiaing the spectr4 and
partly becauseof the imporknce of the Ni thermometer
as a tool in diamondexploration.The Ni resultsmust
first be consideredin the establishedcontext of the
overall accuracyfor other traceelements;the accuracy
figures given in the precedingparagaph suggestthat
for nickel between 50 and 150 ppm, the accuracy
would be in theraage2-l0%o,the uppervalueapplying
to the lower concenEations.However, this argument
fails to considerthe specialproblem of overlap with
the iron lines that is encounteredin the Ni case.The
potential for error might be expectedto increaseas
the Ni/Fe concentrationratio (R) decreases.
This ratio
(in units of ppm NilFe) is thereforeincludedin Table2.
At the highestvaluesof R" encounteredin GXR-5 and
BIM-I,
measuredand nominal Ni concentrations
(obtainedby methodC) agreeat the l-2%olevel. Atthe
lowest value of R (GSR-S), there is agreementat
the 3Volevel.It is thereforedfficult to understandwhy
in the two caseswith intermediateR-values(MAG-I
and GSD-50), tle measuredvalues are, respectively,
Vl and LTVolower than the nominal values.It is worth
noting, however, that in GSD-50, the elementsRb,
Y and Mo show very similar discrepanciesto that for
Ni.
Fedorowich et al. (L995) have recently published
resultsof a LAM-ICP-MS/micro-PD(E study directly
relevant to the present work. Data from this study
providean interes:ng comparisonof the two analytical
techniques.A sfrengthof micro-PD(Eis the ability to
examinecertain transition elements,e.g., N| Cv, Zn,
Ga Ge, As, Co (as well as Rb andNb) at low concentrations.In contast, analysisof thesespecificelements
by I-AM-ICP-MS can be problematic for a number
of reasons.From a technical standpoint, problems
arise during analysis for transition elements by
LAM-ICP-MS owing to: (a) high backgroundscaused
MCRO-PDG ANALYSIS OF SITICATE REFERENCESTANDARDS
by polyatomic compounds of the major plasma
constituentsof the ICP [Ar, N, O, C, (tf)]; (b) interferencesby polyatomic ions from the major matrix
minerals 1e.g., Ncar6o on 56Fe,uca1,6o on 6t{i1,
(c) differential ablation-behaviorthanfor the lithophile
major elementscommonly used as internal standards
(e.9., Ca), i.e., the transitionelementsexhibit variable
continuousfractionationduring ablationin comparison
relative to the intemal standard(S.L. Jackson,pers.
comm.,1995).Figure3 of.Fryeret al. (1995)illustrates
the relative fractionation of the transition elements
relative to lithophile elementscommonlyusedasinternal standards(e.9.,Ca or Si).
On the basis of the above discussion,we are not
confident about the accuracyof the Ni value recommended for the STAG garnet by Fedorowich et a/.
(1995), as this gamet has only been analyzed by
ICP-MS, and theseresultshavenot beenconoborated
by anothertechnique (e.9., INAA). Furthermore,we
note that the Guelph micro-PD(E value of 69 ppm
publishedby Fedorowichet al. (L995)was determined
utilizing "Method A" (as outlined previously).
Application of the now preferred'Method C' for the
STAG gamet gives 65 ppm Ni with the standard
deviation for L5 different spots being t5 ppm. The
Ni/Fe ratio in this caseis 7, which correspondsclosely
to that of the GSR-5 glass @ = 6.5), where we
observedthe agreementbetweenmeasuredand nominal Ni concentationsto be within 3Vo.T\e measured
value,65 ppm, is approximately107ohigherthanthose
reported for STAG garnet 59 t 1 ppm Ni (by
LAM-ICP-MS), 56 t 2 ppm Ni (by solutionICP-MS)
and 57 * 2 ppm Ni (micro-PD(E analysisconducted
at the Commonwealth Scientific and Industrial
Organization,CSIRO, Sydney,Australia), as reported
by Fedorowichet al. (L995); seeFigure 5. Details of
werenot
the error estimatesin the othermeasurements
given. The differencebetweenthe Guelphand CSIRO
PD(E resultsmay be due to the two different software
data-reductionschemes@yan er al. L990, Maxwell
et al. L989,1995)employedby therespectivefacilities.
However, as shown in the presentstudy, the Guelph
micro-PD(E facility producesreliable Ni determinations over the rangeof concentrations30--130ppm that
is of interestfor mantle garnet.This tends to support
our result of 65 ppm for the STAG garnet. As an
independenttest of accuracy,we were able to analyze
a garnet megacryst (GHR-I) lent to us by Anglo
American ResearchLaboratories@ty), Johannesburg,
South Africa; this had been characterizedpreviously
by instrumental neutron-activation analysis. One
sample consisting of acid-leachedgrains gave a
Ni result of 12L ppm, whereas a second sample
consistingof unleachedgrainsgave 117 ppm; the error
estimate for these INAA results (1o) is t4%o. T\e
micro-PD(E value of 119.5 t 4.5 ppm (mean over
15 spots) agrees closely with the INAA value of
120.5t 3.5 ppm, as shownin Figure 5. The merit of
developing one or more well-characterized garnet
standardsis obvious.
E
c
-q 110
270
soluTtoN
tcP-Ms
tAM
lcP
PIXE
cslRo
45
PXE
GUELPH
PIXE
GUEII'H
Ftc. 5. Comparisonof measuredconcentra.tions
of Ni in STAG garnet and in GHR-I
megacryst.
46
TTIE CANADIAN MINERALOGIST
I rnplicationsfo r thermametry
Ni concentations are usually convertedto formation
temperatures using tle calibration established by
Single-spotanalysisof a set of 100 mantle garnet Griffin et al. (1989) and refined by Griffin & Ryan
grainsis the commonpracticefor the applicationof Ni (1995); this is shown in Figure 6a. For a single-spot
thermomety in diamond-explorationprograms. The analysis (with 3-4 minute data-collection period)
a
* Peridotltexonoliths
o Ollvineinctugions
In garnot graing
(Garnet Fldge, USAI
I
Perldotltematslts
A Olivine-garnolpairs In
Argyle diamonds
10
t*g
r*"S
o
z
B#
5 r.o
Hc. 6. (a) Ni concentration
ratio (distribution coefficien$ in coexistinggamet
and olivine as a function
of temperature (Griffin
et aI. 1989, Griffin and
Ryan 1995). Reproduced
by permission of C.G.
Ryan
and
Elsevier
SciencePublishers.(b) Ni
thermometer calibrations
according to Griffin &
Ryan (1995) and to Canil
(1994).Estimatesof measurementerror shown are
;ot u single 4-minute
micro-PD(Emeasurement
using the Guelphmethodology. Temperatureerrors
associated with typical
analysesof Ni in mantle
gametsat 15,40, 70, 100
and 130 ppm concentration are illustrated for
the calibration of Griffin
& Ryan (1995) with open
symbols and for tlat of
Cadl (1994) with fiIled
symbols. Heavy dashed
lines labeled 36 mWm2.
40 mWm2 and M
mWm2 are the temperatures at which the
graphite-diamond transition is inteaected by the
steady-stale geotherm
indicated. Garnet with
"Ni temperatures"to the
left of the known (or
assumed) geotherm are
interpretsd to be derived
from the stability field of
diamond.
2
o
(t
1400 1200
0.6
0.7
6oooc
800
1000
0.9
0.8
(
K
1000/T )
b
lnamw/m"
laomw/mr
I
fr
r
r
I
lgo.wlm,
I
I
I
I
I
z
Grifftt&eya0\
1m0
05
4.7
T(oC)
8m
0.8
0.9
1000/T(K)
1.1
MCRO.PDG ANALYSIS OF SILICATE REFERENCESTANDARDS
utilizing the Guelphmethodology,associatedstatistical
and fitting uncartainty in the determination of Ni
concentrationis in the 5-8 ppm range.The use of the
top-hat filter causesthis uncertaintyto be somewhat
larger than would be the case if a fitted continuum
backgroundwere used.However, the standarddeviation in concenfrationsmeasuredfor 10 or 15 spotsis
4-6 ppm. If Ni is distributedhomogeneously,which
we believe to be fhe case,then the standarddeviation
shouldbe equal to the fitting uncertainty.We deduce
that the fltting uncertainty may be slightly overestimated, and we therefore adopt the standard
deviation of tie multispot measurement as tle
statisticaluncertaintyof a singlemeasurement.
Thus single-spotanalysesof garnetsampleswith Ni
concentations in the range 15-130 ppm have associatedstatisticaluncertaintiesof approximately5 ppm.
Analysis of 100 grains of garnet to produce
temperature-frequency
stackhistograms,as utilized in
diamond-explorationprograms(e.9., Griffin & Ryan
1993) will, of course,tend !o smooth any statistical
errors in Ni determination. However, as noted by
Kjarsgaard(1992),applicationofNi concentrationdata
requires a well-constrained calibration of the Ni
thermometer.Figure 6b illustratesthe consequences
of
a 5-ppm variation in Ni determinationassociatedwith
a single spot analysis at the 15, 40, 70, 100 and
130ppm Ni concenffationlevel; the result is illustated
for both the publishedcalibrationsof the Ni thermometer,i.e.othoseof Griffin & Ryan (1995)andof Canil
(1994). ln the concentation mnge of approximately
40-70 ppm, temperatures derived from the two
calibrations differ by less than the uncertainfy of a
single-spot measurement.However, at lower and
higher Ni content,the divergenceof the two calibrations causestemlleraturedifferencesthat exceedthe
error of measurement.
Furtherrefinementof measurement uncertaintyotherefore, appearslssg imFortant
than a resolution of the difference betweenthe two
publishedcalibrations.
AcKNowLmcm/ENTS
47
Rmsnwcrs
Aneu, J., Gnmq, T.H. & Sm,S.H. (1993):hoton microprobe
determinedpartitioning of Rb, Sr, Ba" Y, Zt, Nb and Ta
betweenexperimentallyproducedamphibolesand silicate
melts with variableF content.Chen Geol. [W,2949.
T.M., Durn'v, C.J. & Rocms, P.S.Z. (1988):
BrNrervm:{,
Geochemicalutilization of nuclear microprobes.NacL
hstrum" MethodsPhys.Ras.830,454458.
CeMpsBr, J.L., Iltcuctr, D., MAxwEr, J.A. & TEspale,
W.J. (1993): Quantitative PD(E microanalysisof thick
specimens.NzcL lnstr, Methods Phys.Res.877, 95-L09.
Maxwnuc, J.A., TErsnem, W.J., WeNo, J.-X. &
CesRI, L.J. (1989): Micro-PDG as a complement to
electron probe microanalysis in mineralogy. )VzcJ.
Instrum.MethodsPhys. Res.M\ 347-356.
Caun, D. (1994):An experimentalcalibrationof the "Nickel
in Gamet'' geothermometerwith applicalions. Contrib.
Mineral. Petrol. ll7 " 410420.
CzAMANsIc, G.K., SIssoN, T.W., Cel"rrswr, J,L, &
TssDAI-e, WJ. (1993): Micro-PD(E analysisof silicate
referencestandards.Am Mineral. 78. 893-903.
Ewenr, A. & Grurrn, W.L. (1994): Application of proton
microprobedata to trace elementpartitioning in volcanic
rocks.Chzm-Geol. 117, 251-2U.
J.S., JAN, J.C., I(nnrucl R. & SoPUc& V.
FtsDoRowrcH,
(1995):Traceelementanalysisof gametby laser-ablation
microprobeICP-MS . Can Mineral.33, 469-480.
For,ny, S.F. & vnN om. l,aax, S.R. (194): Preface.InTrace
element partitioning with Apptcation to magmatic
Processes(S.F. Foley & S.R. van der Laan, S.R., eds.).
Chem Geol. (Spec.Issue)117, vii-xiv.
FhyE& B.J. JAcKsoN,S.E. & lpncmror, H.P. (1995): The
design,operationandrole of the laser-ablationmicroprobe
coupled with an inductively coupled plasma - mass
spectrometer(LAM-ICP-MS) in the earth sciences.Can
Mineral.33,303-312.
Gt-enrsv, E.S. & Ron-lNors, I. (1988): 1987compilationof
The work describedwas supportedin part by the
elemental concentration data for USGS BHVO-I,
Natural SciencesandEngineeringResearchCouncil of
MAG-I, QLO-I, RGM-I, SCo-I, SDC-I, SGR-I, and
Canada. We are grateful to A.R. Ramsden for
STM-I. GeostanlardsNewslexer12, 253-362.
supplying some of the now scarce GSD-50 and to
(1990): 1988compilationof elemental
&J.H. Gilles Laflamme for preparingpolished sections. concentrationdata for USGS geochemicalexploration
We thank T.W. Sissonsfor preparing the BIfVO-1,
reference naterials GXR-I to G)R--6. Geostandards
and G.K. Czamanskefor repolishingand recoatingit.
Newsletter14.21-1L8.
Our thanks are due to G. Read at Anglo American
ResearchLaboratories(Pty) LtA Johannesburg,
South
Govnqoenaru,K. (1989): 1989compilationofworking values
Africa, for permission to reproduce their Guelph
geostandards.
and sample description for n2
micro-PD(Edatafor the GIIR-I gamet,andto R. Hart
GeostandardsNewsletter13 Gpecial issuz),l'1I3.
at the SchonlandCente, Johannesburg,
for permission
to reproducehis INAA datafor this sampleof garnet. Gnmt, T.H. (1994): Experimentalstudies of trace-element
The commentsof the two reviewersand the editor are
partitioning applicableto igneouspetrogenesis- Sedona
much apprecialed.
16yearslater.Chem-Geol.I'Iil, l-36.
48
GRFF[{, W.L., CousEr.{s,D.R., RvaN, C.G., Ste, S.H. &
Strre& G.F. (1989): Ni in chromepyrope gamets:a new
geothermometer.
Contrib. Mineral. Petrol. 103, 19-202,
Mmrrr, C. & Bouonqm, I.-L. (1990): Elechon
determinationof minor and hace transition elementsin
silicate minerals: a new nethod and is application to
mineral zoning in the peridotitenodulePlX{161,1.Chen
GeoI.8.55-69.
& Ryar, C.G. (1993): Trace elementsin gamets
and chromites:evaluationof diamondexplorationtaxgets.
/n Diamonds:Exploration,SamplingandEvaluation@.A. Moonr, R.O., Gnml, W.L., GuRNEy,J.J., RvAN, C.G.,
Sheahan & A. Chater, eds.). Short Course Proc.,
Cousmls, D.R. SrE, S.H. & Sum, G.F. (1992): Trace
Prospectors and Developers Association of Canada,
element geochemisty of ihnenite megacrystsfrom the
Toronto,Ontario(187-211).
Monasterykimberlite, SouthAfrica.Lir&o,r 29, l-18.
(195): Trace elemetrtsin indicator
& minerals:areaselectionand target evaluationin diamond
exploration.J. Geochzm.
Explor. 53,3ll-337 .
IIATDET,N.M., Cel"rent?Ir,
J.L. & TEEsoarr"W.J. (1995):
PD(E analysis in mineralogy and geochemistry.Caz.
Mhzral.33,293-302.
Flrnvrc, R.L., Sri.rru,J.V. & DAwsoN,J.B. (1986):Lherzolite
xenoliths in kimberlite and basalts: petrogenetic and
crystallochemicalsignificanceof some minor and trace
elementsin olivines, pyroxene,garnetand spinel. Zrazs.
Roy.Soc.EdinburglqEarth Sci.77, 181-201.
Mvms, A.T., Havms, R.G., CoNNoR,J.J.,CoNra:D{,N.M. &
Rose, H.J., Jn. (1976): Glassreferencestandardsfor the
trace elementanalysisof geologicalmaf€rials- compilation ofinter-laboratory data. U.S. Geol, Surv.,Prof. Pap.
1013.
PmKrNs,W.T., Peance,NJ.G. & Jsrruss, T.E. (1993):Laser
ablationinductively coupledplasmamnssE)€ctrometry:a
new techniquefor the determinationof trace and ultraActa{I,
traceelementsin silicates.Geochim.Cosmochim475482.
RffD, SJ.B. (1990): Recent developmentsin geochemical
nicroanalysis. Chen, Geol. E3, l-9.
JAcKsoN,S.E., I-oncmrcu, H.P., Duro.m{c,G.R. & Fhvrn,
B.J. (1992): The applicationof laser-ablationmicroprobe RYAN,C.G., CousB,.ls,D.R., SrE,S.H., Gnnmr, W.L., Stnm,
- inductively coupled plasma - mass spectrometer
G.F. & CmvroN, E. (1990): QuantitativePD(E micro(LAM-ICP-MS) ta in situ trace-elementdeterminations
analysisof geological malerial using the CSIRO proton
in minerals.Can.Mincral.30. 1M9-1064.
microprobe. Nucl. Instru.m. Methads Phys. Res. M7,
55-71,.
JuFp;ms,T.E., Pmmw, W.T. & PEARcE,N.J.G. (1995):
Comparisonof infrared and ultraviolet laserprobemicro- Srmnzu, N. & Au-BcRE, C.S. (1978): Geochemisnryof
analysisinductively coupledplasrnaruss spectrometryin
transition elements in garnet lherzolite nodules in
mineral analysis.Analyst 12.0(4),L365-1.372.
kimberlites. Cowib. Mineral..Petrol. 47. 4l-50.
Kransceeno, B.A. (1992): Is Ni in cbromepyrope garnet a
valid diarnondexploration tool? Geol. Sun. Can, Pap.
!r?-18!3L5-322.
Snuac, J. & Ilrclo4orr, D. (1994): Trace elementpartition
coefncientsfor ilmenire, orthopyroxeneand pynhotite in
rhyolite determinedby micro-PD(Eanalysis.ChenLGeol.
tt7,313-330.
LARsso\ N.P.-O., TAppB, U.A.S. & MARrNssoN, B.G.
(1989): Characterizationof the responsefunction of a
Si(Li) detector using an absorber technique. .lfacl.
InstraftLMethodsPhys. Res.843, 574-580.
Swmwv, R.J., Gnmu, D.H. & SE, S.H. (1992): Trace and
minor elementpartitioning betweengametand amphibole
and carbonatiticmell Earth Plangt. Sci.Iztt, ll3, l-14.
LUDDE{,J.N., FsNc,Rur, Garmm., G,, SrX J., LANG,SItr,
FhANcrs,D., MAo{ADo, N. & Wu, Guornc (1995):
Application of I-AM-ICP-MS analysisto minerals.Caa
MtureraL.33.419434.
VrcEr.rz,E., GREN,T.H. & SE, S.H. (1994):Effect of oxygen
fugacity on trace-elementpafiitioning Ssleg€1immisgi$le
silicate melts at atmospheric pressure: a proton and
electronmicroprobestudy.Chem-Geol. 117, 355-360.
Maxws-1, J.A., CAM?BEI, J.L. & TsFsnALE,WJ. (1989):
The Guelph PD(E software package. Nucl, Instrwn
MethodsPlrys. Res.D/,3, 218-230.
Xm Xrmrnc, YAN MD{ccAr,WANGCsuNssu,Lr IleNzrorc
& Snw Htryw (1989): Geochemicalstandardreference
samples GSD 9-12, GSS 1-8, and GSR l-6.
GeostandardsNewsktte r 73. 83-179.
TEESDAB, W.J. & Caunnnr.l, l.L. (1994):
Compensationschemesfor peak-ta;t;nguncerlaintiesin
PDspectra" using the GIIPD( code. NucL InstnanMethadsPlrys. Res.B94, 172-179.
(1995): The Guelph PDG
& softwarepackageII. NucI. Iwtrum MethodsPhys. Res.
895,407421,.
McCenmy, J.J. & ScIrAI\Dm,F.H. (1981): Least-squares
fit
with digital fllter. Nat. Bur. Standard*,Spec.PubL M.
Received March I, 1995, revised mnnuscript
September 28, 1995.
accepted