E L S E V I E R Chemical Geology 121 (1995) 105-130
CHEMICAL GEOLOGY
INCLUDING
ISOTOPE GEOSCIENCE
Trace-element partitioning between garnet and clinopyroxene in mantle-derived pyroxenites and eclogites: P - T - X controls
Suzanne Y. O'Reilly W.L. Griffin b aSchool of Earth Sciences, Macquarie University, Sydney, N.S. W. 2109, Australia
bCSIRO Division of Exploration and Mining, Box 136, North Ryde, Sydney, N.S. W. 2113, Australia
Received 26 May 1994; revision accepted 25 October 1994
Abstract
Trace-element abundances in coexisting clinopyroxenes and garnets from five suites of mantle-derived garnet pyroxenite and eclogite xenoliths and two suites of eclogite inclusion pairs in diamonds have been determined by electron microprobe (Ti) and proton microprobe (Ni, Zn, Ga, Sr, Y, Zr). The sample sets provide garnet--clinopyroxene pairs from a range of depths on contrasting geothermal gradients (40-90 mW m 2) from South Africa, and western and eastern Australia. These data are used to assess the effects of phase composition, pressure (P) and temperature (T) on the partitioning of each element between garnet and clinopyroxene.
Ni partitioning is moderately dependent on XMg and strongly dependent on T. Dzn (Zncpx/Znc,,t) is ~ 0.9 + 0.1 and may show a slight increase with T, but is independent of P. Ga partitioning is strongly dependent on X cpx as well as P and T. Sr and Y distributions are mainly controlled" " c, nt oy aca and are weakly dependent on Tbut not on P. Zr partitioning is mainly dependent on T; Dzr decreases with increasing X c°x and ' Gnt posslbly Xca , but shows no P effect. DT~ decreases markedly with increasing P (and possibly T) and has a weak tendency to increase with increasing X cox.
These results on trace-element partitioning in natural, equilibrated systems provide data for the modelling of melting and metasomatic processes in the upper mantle. They also provide a test dataset to evaluate experimental determinations of partition coefficients, and some relationships (e.g., Dzr, DNI) may be used to cross-check thermobarometry calculations where spurious calculated Fe 3 +/Fe 2 ÷ ratios may give invalid temperature estimates.
1. I n t r o d u c t i o n
Element partit ioning between coexisting mineral pairs provides data fundamental to crystal chemistry, geothermobarometry and the mathematical modell ing of geological processes such as partial melting, frac- tional crystall ization and metasomatism. Considerable effort therefore has been expended on the experimental measurement of major- and minor-element partitioning as a function of pressure (P ) and temperature (T)
under mantle conditions. However, these studies com- monly have problems with the attainment of equilib- rium, and the necessity of doping the charge to reach analytical detection limits leads to the use of unnatural compositions. Furthermore, few experiments on minor- element partitioning have assessed the effect of com- positional variability in the major-element composit ion of the phases involved.
Studies of element distribution in mantle-derived xenoliths can provide critical data to support experi- mental studies, although they also must face the prob- lem of assessing the degree of equilibrium between
106 s. E O'Red/y, W.L. Griffin / Chemical Geology 121 (1995) 105-130
phases. The natural samples have other limitations; many xenolith suites represent samples taken along a single geotherm, where P and T vary sympathetically in some regular fashion. The natural processes of sam- pling in any one xenolith suite therefore provide a rel- atively narrow range of well-correlated P - T whereas in the laboratory, P and Tcan be varied independently, and P - T combinations unlikely to occur in nature are routinely used to study P-Teffects. However, selection of xenolith samples from areas with different geother- mal regimes can provide both a useful range of P - T combinations, and many xenoliths appear to be derived from depths beyond the reach of routine experimental studies. Finally, a wider range of complex composi- tions is available in xenoliths, while only a small range can be studied in a time-consuming laboratory research program. Analysis of element distribution in such xeno- lith assemblages therefore can provide valuable data which can guide experimental work and which in turn can be used to evaluate the validity of experimental results.
In this study, we have determined trace-element abundances by electron microprobe (Ti) and proton microprobe (Ni, Zn, Ga, Sr, Y, Zr) in coexisting cli- nopyroxenes and garnets from mantle-derived garnet pyroxenite (i.e. low-Jd Cpx) and eclogite (i.e. Jd-rich Cpx) xenoliths and from inclusion pairs in diamonds. All of these minerals are low in Cr; partitioning between Cr-rich garnet, orthopyroxene and chrome diopside in garnet peridotite xenoliths will be discussed elsewhere (W.L. Griffin et al., in prep.). The live suites of xenoliths and two diamond-inclusion suites used here represent a variety of mantle pressure (P) and temperature (T) regimes from a variety of geothermal profiles and therefore can be used to assess the effects of P, T and mineral composition on trace-element par- titioning. The samples have been selected to be as free of alteration and disequilibrium microstructures (e.g., zoning) as possible.
The aims of this work are: (1) to establish the con- centration ranges of selected trace elements in mantle- derived clinopyroxene+low-Cr garnet pairs; (2) to assess as far as possible from these natural samples the effects of pressure, temperature and composition on clinopyroxene/garnet partition coefficients; (3) to pro- vide realistic constraints on partition coefficients for use in modelling of melting and metasomatic processes in the upper mantle; and (4) to establish a framework
for evaluation of experimental determinations of par- tition coefficients.
2. Samples
Xenoliths in kimberlites and basalts are (in princi- ple) useful for partition-coefficient studies because they commonly record the ambient T at the depth of entrainment, quenched in by rapid ascent (Boyd, 1976; O'Reilly, 1989) and eruption. Garnet-clinopyroxene pairs encapsulated in diamonds (and not in contact) record T at the time of their entrapment in the growing diamond; this may be significantly different from the ambient mantle temperature at the time of entrainment in the kimberlitic magma (Griffin et al., 1992, 1993). The sample suites used here were chosen to provide a wide range of P and Tconditions (from single pipes as well as collectively), and include samples that come from different geotherms, i.e. different combinations of P - T conditions (Fig. 1 ).
2.1. Xenoliths from South African kimberlites
These samples are xenoliths from the kimberlites at Roberts Victor [ including some described by Viljoen et al. (1994)], Newlands and Bobbejahn mines
Q,,
10
2O
3O
40
50
60
70
80
90
T:=C
200 600 1000 1400 1800
O~,~,~. ~ . J grit pyroxenites 5O
' ~ ~ diamond _ Argyleeclogite- sure _ ~k,~lncluslons ? _ Inclusions In d i a m o ~ 250
O0
i50 ~. 8
Fig. 1. P and T fields represented by the the xenolith suites studied. References are given in the text.
(grouped below as "Rov ic" ) , Kaalvallei (K.S. Vil- joen, 1994 and in prep.) and Monastery Mine (Moore et al., 1991). Several extreme compositions, including grospydites and kyanite- or corundum-bearing eclogi- tes, are included in these suites. Diamond-bearing eclo- gites come from the Roberts Victor, Orapa and Excelsior mines (Robinson et al., 1984; Hatton and Gurney, 1987; McCandless and Gurney, 1989). The area in which many of these kimberlites erupted had a well-constrained xenolith-based geotherm, corre- sponding approximately to Pollack and Chapman's (1964) model of conductive geotherm for 40-mW- m -2 heat flow (Finnerty and Boyd, 1987) to which clinopyroxene/garnet temperatures may be referred to obtain approximate pressure estimates (Table 1 ). Dia- mond eclogites are stable only at P > 40 kbar on this geotherm.
2.2. South African and Australian diamond inclusions
These samples represent clinopyroxene-garnet pairs, not in contact, from single diamonds. The dia- mond inclusions (hereafter referred to as DI) from Argyle diamonds were described by Griffin et al. (1988) and the data on the South African DI are given by Gurney et al. ( 1979, 1986a, b), Moore and Gurney (1989), Rickard et al. (1989) and J.J. Gurney (unpub- lished data). Both areas had a 40-mW-m -2 geotherm at the time of eruption (Finnerty and Boyd, 1987; Jaques et al., 1990), but the high temperatures recorded by some DI pairs may reflect growth of the diamonds during essentially isobaric thermal pulses, rather than derivation from extreme depths (Griffin et al., 1992; Fig. 1 ).
2.3. Garnet clinopyroxenites from Bullenmerri and Gnotuk Maars (B / G), Victoria, Australia
The petrology of these xenoliths (hereafter called the B/G suite) has been described by Griffin et al. (1984), and the elevated geotherm defined by these xenoliths is discussed by O'Reilly and Griffin (1985). The importance of this suite is that the xenoliths reflect significantly lower P at any given T than either of the other suites. Thus a T of 1000°C will be reached at ~ 40-kbar pressure in the South African xenolith suite, but at ~ 12 kbar in the B / G suite (Fig. l ). Comparison of these suites therefore provides a means of separating
the effects of pressure and temperature on element par- titioning. Trace-element abundances and residence sites for mantle-derived spinel lherzolite xenoliths from this locality were reported previously (O'Reilly et al., 1991). Two xenoliths from another locality in eastern Australia are also included within this group as they lie on the same geotherm and extend the P range (samples 69-27 and 38926 from the Delegate locality; Griffin and O'Reilly, 1986).
3. Problems
Problems encountered in interpreting the trace-ele- ment data include the usual decisions on appropriate thermobarometry calculations (Finnerty and Boyd, 1987) and the possible effects of disequilibrium.
We have used the Ellis and Green (1979) garnet- clinopyroxene thermometer to estimate temperatures for each mineral pair. This thermometer, and its various modifications (Krogh, 1988) are extremely sensitive to the ferric/ferrous iron ratio in the clinopyroxene, which is in turn sensitive to analytical error, especially in Si (Neumann, 1976). The Fe 3 ÷/Fe 2 ÷ ratio has been estimated for each analysis by normalization to four cations. In most of the xenoliths used here, the calcu- lated Fe 3 ÷ content of both clinopyroxene and garnet is within the analytical errors; Fe 3 + therefore is taken as zero and the T estimates are accepted as reliable. In some cases the calculated Fe 3 ÷ contents of the pyrox- enes are significant, and lead to improbably low tem- peratures, as noted by numerous earlier workers. Re-analysis does not remove this effect, and we inter- pret at least part of the high Fe 3 + contents as the result of late-stage (deuteric?) oxidation of the pyroxenes. In these cases (underlined in Table 1) we have arbi- trarily taken an average value between the temperatures calculated assuming zero and maximum Fe 3 + contents; this procedure typically changes the estimated T by 100-150°C and may produce more scatter in several of the plots shown below. However, removal of these samples does not produce much improvement in the regressions, and the Tuncertainty does not significantly affect the deliberately broad conclusions drawn from the data.
An approximate pressure has been estimated for each sample (Table 1) by referring T to the appropriate
108 S. K O 'Reilly, W.L. Griffin/Chemical Geology 121 (1995) 105-130
Table 1 Representative data for clinopyroxenes and garnets
Eastern Australia pyroxenites
Sample GNSI GNWI D R I O I 6 2 BMI71 G N 4 7 GN649(1) 69-27 38926 Locality Lakes Bullenmerri/Gnotuk, western Victoria Delegate, N.S.W.
112 S. E O'Reilly, W.L. Griffin/Chemical Geology 121 (1995) 105-130
geotherm (Fig. 1) ; these have at least as much uncer- tainty as the T estimates; they are included primarily to emphasise the significant differences between the B /G suite and the kimberlite-derived suites.
Potential causes of disequilibrium include metaso- matism by the host magma (e.g., basalt or kimberlite) or associated mantle events, which could preferentially change the trace-element concentrations in one phase, or change Fe 3 + /Fe 2 + (e.g., B allhaus et al., 1991 ) and hence lead to spurious temperature estimates. In addi- tion, short-term heating in the mantle or in the magma may result in incomplete trace-element diffusion, although the zoning expected from this process has not been observed here. DI pairs may not have been pre- cipitated at the same growth stage (Griffin et al., 1992, 1993) and thus may not represent equilibrium pairs.
All of these features also may contribute to scatter in the plots shown below.
4. Methods
The proton microprobe analyses for Ni, Zn, Ga, Sr, Y and Zr have been carried out using the CSIRO's HIAF facility, as described by Griffin et al. (1992, 1993) and in more detail by Ryan et al. (1990). These procedures provide rapid, standardless quantitative analysis of volumes ca. 30 × 30 × 30/xm. Typically 3- 10-point analyses were collected on each phase, any outliers (judged to result from analysis of inclusions or alteration material below the surface of the grain) were discarded, and the selected analyses were digitally summed to form a single spectrum with improved sta-
Table 2 Electron microprobe data for clinopyroxenes and garnets in Table 1
Sample Locality
Eastern Australia pyroxenites
GNS1 GNWI DRIOI62 BMI71 Lakes Bullenmerri/Gnotuk, western Victoria
between Grit and Cpx (Ellis and Green, 1979) will produce a general decrease in XC~ " with increasing T; this is in fact observed in the South African and Aus- tralian xenolith suites (Fig. 2), despite the scatter induced by the variation in bulk M g / ( M g + Fe) within each suite. However, the diamond-inclusion suites show no obvious correlation between v cpx and T, L'Mg which suggests they are derived from a wide range of bulk compositions.
X cp× ranges from zero to 0.7 (Fig. 3); the highest values are found in the pyroxenes of some grospydites and kyanite- or corundum-bearing eclogites. In the absence of a buffering assemblage (e.g., plagio- clase + quartz) the bulk composition of the rock deter- mines the maximum jadeite content of the pyroxene,
but the jadeite content of clinopyroxenes still is broadly dependent on pressure, because of the changes in Grit/ Cpx ratio with increasing P. The African xenoliths show a broad positive correlation between X cpX and T (Fig. 3) ; this reflects the rapid increase in P relative to T along the cratonic geotherm. The correlation is best displayed by the Kaalvallei eclogites and the "Rovic" suite, where two distinct trends are visible, correspond- ing to the Group-I (high-Jd) and Group-II (Iow-Jd) eclogites of MacGregor and Carter (1970).
The B/G pyroxenes show lower X cp× at any T than those from the South African xenoliths; this reflects the higher geotherm (lower P at any T) of the B /G suite. Although the bulk Na content of the B/G xenoliths is comparable to those of many African eclogites, the
higher T at any P results in a more aluminous Cpx, a higher Cpx/Gnt ratio, and a lower X cpx on average.
The African DI pyroxenes show a clear decrease in X cox with increasing T; this suggests that T increased more rapidly than P for this suite, and is consistent with other indications that the diamonds grew during ther- mal pulses which affected rocks distributed over a rel- atively limited range of P (Griffin et al., 1992, 1993). The Argyle DI pyroxenes show no correlation between X cpx and T, but have generally higher X cpx at any T than the African DI pyroxenes; this presumably is a bulk-composition effect.
xGnt appears to show no correlation with T (Fig. 4) ; Ca the fact that the B /G samples are indistinguishable from the African eclogites in this plot suggests that xG,t also is independent of P, and is controlled pri- Ca marily by bulk composition.
5.2. Nickel
Ni contents range from < 10 to 200 ppm in garnet, and from < 10 to 1500 ppm in clinopyroxene; the broad fan on the correlation plot (Fig. 5a) indicates a signif- icant range in DNi [D;= (ppm of element i in Cpx) / (ppm of element i in Gnt)]. In both phases, there is a broad positive correlation between Ni content and XMg (Fig. 5b and c). The plot o f DNi VS. 1000/T (Fig. 5d) shows a broad negative correlation between DNi and T, showing that Ni preferentially enters the Cpx, and that this tendency increases as T decreases. A plot of DNi against vcpx (Fig. 5e) shows an overall positive cor- ,l. Mg relation; however, closer inspection of this plot reveals that some individual suites actually show a negative correlation between DNi andX cpx (most clearly shown by the diamond--eclogite suite), so that the plot consists
Fig. 2. X~," plotted against 1000/T for all the suites of clinopyrox- ene-garnet pairs. Descriptions of these suites and a discussion of the plot are given in the text. B/G refers to garnet pyroxenites from the Bullenmerri/Gnotuk locality in eastern Australia and includes one sample from a New South Wales locality (Delegate). Rotfic refers to xenolith suites from the kimberlites at Roberts Victor, Newlands and Bobbejahn mines. Diamond-bearing eclogites come from the Roberts Victor, Orapa and Excelsior mines. Subsequent diagrams use the same symbols.
80 1500 1300 1100 900 (°C)
+
+ 60 +
• x ++
• i • • ~ X ~0 X A X
I ~ X ~ 4- 4-
~o ~ ~ ~ oo~ o• #. + ~ . O +
o CII I I
o• ° 0 i i i 0 ,
0.5 0.6 0.7 0.8 0.9 1,0
1 0 0 0 / T , K
Fig. 3. Jadeite component in clinopyroxene plotted against 1000/T for all samples. Symbols used are the same as defined for Fig. 2.
x
o 40
of several broad en 6chelon negative trends. In detail, samples with a similar T range fall along trends that can be described by the equation DNi(corr)= DN~+y(XC~"-0.6), where y = 0 for T=1400- 1500°C, y = 10 for T= 1200-1300°C and y = 2 0 for T= < 1200°C. The corrected DNi-Values (omitting three values that became < 0) are plotted in Fig. 5f;
the corrected regression shows a significantly reduced scatter. We conclude that the partitioning of Ni between Gnt and Cpx in these samples shows a moderate dependence on XMg, and a strong dependence on T. The B/G samples remain on the high side of the scatter even after correction Cp× for XMg , which might indicate a small inverse correlation between DNi and P at constant T. However, the possible effect is too small to be reli- ably demonstrated by this dataset.
5.3. Zinc
Zn contents range from < 3 to 150 ppm in both Gnt and Cpx; the correlation plot (Fig. 6a) shows a broad scatter around a 1:1 line. The highest Zn contents are found in the African diamond inclusions. The Argyle DI data are not used here, because the surfaces of many grains had been contaminated by Zn from the surgical tape used for scanning electron microscopy (SEM) mounting (Griffin et al., 1988). Zn contents show a broad correlation with X M g in both phases (Fig. 6b and c). There is no obvious correlation of Dz,, with either
~.Gnt (Fig. 6e and f). However, the high-Ca X M g o r z,~Ca
X G"t >0.40) garnets of grospydites tend to take up Ca
slightly more Zn than other garnets, so that these rocks typically have anomalously low Dzn values.
The plot of Dz, vs. 1000/T (Fig. 6d) shows a broad band between Dz° = 0.7 and Dz, = 1.5, regardless of T, with the grospydites (and two low-Ca garnets) extend- ing to lower Dz,. In detail, the eclogites from Monas-
1500 1300 1100 900 (°C) 6 0 , ,
50
4O
o 30
¢0 O 20 X
10
• +
A +
Ib , ~ ~lI+ + I +
[] I i ,~ + x x xWr~ X . x ~ +
O X 'lI~.lO
I I~ + ~0~__ I
O X Z~ u ~
0 i i ~ I 0.5 0.6 0.7 0.8 0.9
1000/T,K
1.0
Fig. 4. Xc, in garnet plotted against 1000/T for all samples. Svmbt:.,ls used are the same as defined for Fig. 2.
Fig. 5. Varia t ion o f Ni p p m a n d / o r DNi with key chemical var iables and 1000/T. DNi(COIT ) =DNi Cpx +y(XMg - - 0 . 6 ) , where y = 0 for T = 1 4 0 0 - 1500°C v = 10 for T = 1200-1300°C and y = 20 for T = < 1200°C, and is discussed in the text. Symbols used are the same as defined for Fig. 2.
118 S. E 0 'Reilly. W.L. Grig'fin / Chemical Geology 12I (1995) 105 130
Fig. 6. Plots of Zn ppm in clinopyroxene and garnet and/or D z" against key chemical parameters and 1000/T. Symbols used are the same as defined for Fig. 2.
S. K O'Reilly, W.L. Griffin/Chemical Geology 121 (1995) 105-130 1 1 9
tery Mine, the Group-I eclogites from Kaalvallei and the African DI pairs show weak positive correlations between Dzn and T. The B/G samples fall in with the other suites, and show no dependence of Dzn on T; this suggests that the effect of P on Dzn is small.
We conclude that Dzn is typically ~ 0.9 + 0.1 ; it may increase slightly with T, but is independent of P. The only identifiable bulk-composition effect is the greater uptake of Zn by some high-Ca garnets. These regular- ities make the partitioning of Zn a sensitive test for contamination and for disequilibrium induced by meta- somatism.
5.4. Gallium
Ga contents range from 3 to 46 ppm in Cpx and from 4 to 25 ppm in garnet. Cpx typically contains more Ga than coexisting Gnt (Fig. 7a), but in detail some indi- vidual suites scatter across the general trend, reflecting a large range in DGa. A sapphirine granulite xenolith from Delegate (Griffin and O'Reilly, 1986) is anom- alous in that the garnet contains 45 ppm Ga, whereas the coexisting Cpx contains only 8 ppm. It is interesting to note that Ga does not "fol low" AI, in the sense of being more abundant in the more aluminous phase, even though the chemical "coherence" of Ga and AI is widely accepted in the literature (e.g., McKay and Mitchell, 1988). Ga contents of Cpx are well correlated with X cpx both overall and within individual suites (Fig. 7b), suggesting that Ga is involved in the charge- balance mechanism for Na substitution, as noted by Griffin et al. (1988). The B/G samples fall above the general trend (i.e. high Ga relative to Jd), which sug- gests that Ga also may be involved in the Tschermak's substitution together with A1.
The plot of DG~ vs. 1000/T (Fig. 7c) shows a broad scatter with an overall negative slope (i.e. DGa increases with T). However, this trend is largely defined by some Argyle DI data at high T, and some Kaalvallei data at low T; most of the data lie in a band that shows a weak positive slope. In particular, the data from Monastery and Roberts Victor and the diamond eclogites show positive slopes (although there are many outliers). As might be expected, there is a strong overall correlation ofDG= withX cpx (Fig. 7d); the data from B/G lie well off this trend, and the data from the Argyle DI lie at a steep angle to it, since they show a wide variation in Dca at uniformly high X cpx. Separa-
tion of the data on this plot by T shows that DGa broadly increases with decreasing T at constant X cpX .
We have used the data for the South African eclo- gites, in the range T= 1100-1200°C, to derive a cor- rection for the effect of X cpx on D~a at constant T: DGa(corr) = D~a-0 .02(%Jd) . This simple procedure corrects all D~a to X cpX = 0; only 4 values give Dca < 0. A plot of D~(cor r ) vs. 1000/T (Fig. 7e and f) actually shows more overall scatter than a plot of the uncor- rected data (Fig. 7d). However, the scatter within each suite is reduced, and several interesting points stand out. The B/G data show no clear T dependence, but 6 of the 7 points lie well above the African eclogite trend. The diamond-inclusion data from Argyle lie along a trend normal to that defined by the African eclogites, while the African DI data show no clear correlation of DGa with T (Fig. 7f).
These contrasting trends can be reconciled by the following interpretation (Fig. 7f): (1) The T depend- ence over a narrow P range is given by the Argyle DI data, i.e. D~a increases with T; (2) the P dependence at constant T is shown by the difference between the African eclogites and the B/G suite, i.e. DGa increases as P decreases; (3) the broad decrease in Doa with increasing T observed in the African eclogite suites reflects the covariation of T with P along a cratonic geotherm, i.e. P increases relatively more rapidly than T, as noted above. The large degree of scatter within the African xenolith and DI data may reflect differences between the geotherms in each locality at the time of eruption, a larger degree of disequilibrium than is sug- gested by the other data, or compositional controls that have not been recognized.
We conclude that D ~ is strongly dependent on xCpx because Ga is involved in the charge-balance Jd ,
mechanism for Na (and for tetrahedral AI-Si substi- tution in high-Tpyroxenes). Doa also shows strong P - T dependence; Ga preferentially enters Cpx with increasing T and Gnt with increasing P.
5.5. Strontium
Sr is strongly concentrated in Cpx, where contents range from 3 to 1000 ppm (Fig. 8a); few garnets con- tain > 10 ppm Sr except for the Argyle DI garnets, in which Sr concentrations can reach 60 ppm. There obvi- ously is a large range in Dsr, but the uncertainty in the ratio is large in many cases due to the analytical uncer-
Fig. 7. Plots o f G a ( p p m ) a n d / o r Dc~, in c l i n o p y r o x e n e aga ins t Ga ( p p m ) in garnet , %Jd in c l i nopy roxene and 1000/T. Dc,,(corr) = D m . - 0 . 0 2 ( % J d ) is plotted against 1000/T in (e), and (f) clarifies that plot by showing the trends of selected suites. Symbols used are the same as defined for Fig. 2.
S.Y. O'Reilly, W.L. Griffin I Chemical Geology 121 (1995)105-130 121
Fig. 8. Plots of Sr ppm in clinopyroxene and garnet and Dsr against relevant parameters ~LS discussed in the text. :~vmbols used are the same as defined for Fig. 2. (d) shows the "Rovic" dataset subdivided on the basis ofT.
tainty in the analysis of Sr in Gnt (Table 1 ). Excluding some outliers (especially the low-temperature mem- bers of the " R o v i c " suite), Ds,. shows a broad overall negative correlation with T (Fig. 8b). There also is a strong negative correlation between Ds~• and ~vcmc~,, at least up to ~ 25% Gross (Fig. 8c), reflecting a greater uptake of Sr by Ca-rich garnets over this range. In general, the high-T samples within the "Rov i c " eclo- gite suite have higher Ds,. at a given X~.',; ~ than the lower- T garnets (Fig. 8d). Using the medium-T group from this suite, we derive a correction for the effect oi'" .,~m -'~ c,,
" 3 ~ y G n t on Ds~: Ds r (co r r )=Ds , .+ 1 . . . . c~, • After removal of this effect, there is still a weak negative correlation between Ds,. and X cv~ at low Jd contents (Fig. 8e) but the data are too scattered to provide a reliable correction for this effect.
The plot ofDs~(corr ) vs. 1000/T (Fig. 8f) shows a reduced scatter (compared to Fig. 8b), and hence a somewhat more pronounced negative correlation of Dsr with T; however, this trend is not definitive within any one suite, with the possible exception of the diamond eclogites (n = 4) and the African diamond inclusions.
We conclude that Ds~ is strongly controlled by X~'~tc+,, especially at low grossular contents. There also appears to be a weak T effect on Ds~, with more Sr entering Gnt as T increases. No P effect is discernible in the limited data available from the B / G suite.
5.6. Y t t r i u m
The Y contents of Gnt range from < 2 - 1 2 0 ppm; most Cpx contain < l0 ppm and many contain < 1 ppm (Fig. 9a). There clearly is a large range of Dr, but as for St, the low Y contents of the Cpx produce relatively large uncertainties on some Dv-values. There is a good correlation between the Y content of Gnt and XC,,t in some suites, notably the African DI and the Ca
Monastery eclogites, but no correlation is obvious in the other suites (Fig. 9b). There is no apparent corre- lation between Y and Na shown in the plot of Y vs. %Jd (Fig. 9d). The behaviour of Dy with T varies widely among the sample suites (Fig. 9c) ; the African DI show a broad positive correlation between Dy and 7", the Monastery suite shows a negative correlation, and the other suites show none. Overall, the data define a large scatter with a slight negative correlation between Dy and 7+. However, there is an overall nega- tive correlation between Dy and xC'mc~, (Fig. 9d), which
is essentially independent of T. We have used the Afri- can D1 data to derive a correction for the effect of x Gn t Ca on Dy: D v ( c o r r ) = D y + 140X(C~p ~. Removal of this effect greatly reduces the scatter in the data, and leaves a very small positive correlation between Dy and T (Fig. 9e). The B /G data/'all within the range of the other suites.
We conclude that the partitioning of Y is controlled vG,,t and that there is only a weak almost entirely by ~c~, ,
tendency for Y to enter Cpx with increasing T; no P effect on partitioning is discernible.
5. 7. Z i r c o n i u m
Zr contents generally range from < 1 to 50 ppm in Cpx and from < 1 to 80 ppm in Gnt, with generally higher contents found in Gnt (Fig. 10a). Several Mon- astery eclogites have unusually high-Zr Cpx, and the Argyle DI garnets tend to contain anomalously high Zr. Dzr for most samples ranges from 0.1 to 3, and shows an overall negative correlation with T; a number of samples, especially from Monastery, scatter widely around this main trend (Fig. 10b). There is a broad overall negative correlation between Dz,. and L,vCpXjd (Fig. 10c), and this correlation is pronounced within several individual suites, notably the Argyle DI, B/G, and the eclogites from Kaalvallei and Roberts Victor. Examination of the African eclogite suite shows that in general Dz,. increases with Tat any X cox . We have used the intermediate-T group from this suite to derive a correction for the effect of vq,x z . Jd o n Dz,-: Dzr(corr) = Dzr + 14X cpX . After removal of this effect, there is no clear overall correlation between Dz,.(corr) and l,vc;"c:~+ (Fig. 10d). No correction has been applied for this effect; similarly, no correlation could be seen between Dz, (corr) and vcp× (not shown). ~ M g
The plot ofDzr(corr ) vs. 1000/T (Fig. 10e) shows a greatly reduced scatter compared with the raw data (Fig. 10b). Some of the high-T, high-Jd pairs from Argyle may have been overcorrected by the procedure applied. Two Monastery eclogites still fall well above the trend; these may reflect disequilibrium caused by late-stage metasomatic uptake of Zr by Cpx. The B /G samples are not distinguishable from the other suites.
We conclude that Dz~ is mainly controlled by T, with more Zr entering Gnt as T increases. Dz,. is lowered by
V Gt+t increased ,.vcP×ju , and possibly also by increased ~c,, •
S. E O'Reilly, W.L. Griffin / Chemical Geology 121 (1995) 105-130 123
10
8
U r-
D
>" 4
2
0
10
1
>-
.1
.01 0.5
0.5
0.4-
0.3 >.
0.2
0.1
0.0
0
O
El ~ x
n
ra o o n
m"'m ~ ~
O+
1500 i
Q
X X)NK
XO
I
4O
1300 t
0
I
80
Y In gnt
1100
a
I
120
900 (°C) I
c
x o oC~O x ¥&xx ~ +.~
0 O Ox~
X X El
I i I
0.6 0.7 0.8
1000/T,K
0.9
120
4
10
X m
o + •~
X I ~ & +
m o ~ ~ ' + • A " ÷ + •
I ! I I I
.10 .20 .30 .40 .50
XCa, gnt
8 -
~ . 6 - U e -
4 - ~ o
O O
2 - O
0 0
I
10
'° 1 [] o [] 1 x
+ x x M oox XOo + .1
0 0 X EIA O a ~ ~ 1 3 0
D & )0( ~ 0
, , , , , .01 .10 .20 .30 .40 .50 .60 0.5
XCa, gnt
n
><K X +
O
0 o 0 El
X • X [ : 3 ~ IP, - X
r'l O >0< [3 &
ZX ^ o ~ ~
z$ +
X
X X
20 30 40 50 %Jd, cpx
1500 1300 1100 I I ]
.60
^ f t . . ~_ a ¢~_a~ ~ o " - " " ~ " ~ d ' ~ o ~ % n =
LnDy = - 0.59 - 0.43(1000/1") R^2 = 0.005
I i i
0.6 0.7 0.8 1 0 0 0 / T , K
60
900 (°C) I
f
0.9
Fig. 9. Plots of Y (ppm) in clinopyroxene and garnet against key chemical parameters, and Dy against 1000/T. Dv(corr) = Dv + 140X~'," (as discussed in the text). Symbols used are the same as defined for Fig. 2.
Fig. 10. Variation o f Z r (ppm) and Dzr with key chemical parameters and 1000/T. Dzr(Corr) = Dz ,+ 14X~I TM as discussed in the text. Symbols used are the same as defined for Fig. 2.
S. E 0 'Reilly, W.L. Griffin / Chemical Geology 121 (1995) 105-130 125
No pressure effect on Dz,- can be recognized from these data.
6. Discussion and conclusions
6.1. Compositional dependence
5.8. Titanium
Ti contents range up to 5500 ppm in Cpx and 8000 ppm in Gnt; in most samples, the two phases have similar Ti contents (Fig. 1 la). However, the Ti con- tents of the B/G garnets are uniformly low, despite the relatively high Ti in Cpx. The Kaalvallei eclogites show two distinct groups on the basis of Ti content. There appears in general to be little dependence of D-ri on T (Fig. 1 lb), but the diamond-eclogite suite displays a pronounced negative correlation between DTi and T.
independent of Xc~ (Fig. Dvi appears in general to be - cnt
1 ld), but a group of relatively high-Ca samples in the Argyle and diamond-eclogite suites have anomalously low Dye. There is no overall correlation between Dw and X cpx, but a strong positive correlation is seen
within the diamond--eclogite, Monastery and Kaalval- lei suites, suggesting that X cpx may be an important
factor. Inspection of the data shows that for a given Jd content, D-r~ decreases with increasing T. Using the medium-T subset of the African eclogite suite, we have derived a correction for the effect of X cpx on D-ri:
DTi (corr) = D w - 9 2Y co x + 0.55. This algorithm cor- ~" ~ J d
ycp~ = 0.20. However, appli- r e c t s O y i t o a constant ,~ j~
cation of this correction produces no improvement in the overall correlation between DTi and 1000/T; several low-Ca samples from Argyle still fall offthe main trend to lower Dw (Fig. 1 le). Both raw and corrected Dw show a broad negative correlation with T, suggesting an increased preference of Ti for Grit at higher T. The extremely low Ti contents of the B/G garnets (leading to high /)TO suggest that D-r~ decreases significantly with increasing P at constant T, although this effect cannot be quantified with the present dataset. It there- fore is probable that the apparent dependence of Dv~ on T observed in Fig. 11 e simply reflects the relationship of T to P along a typical cratonic geotherm.
We conclude that Dv~ decreases markedly with increasing P, and also may decrease with increasing T. These effects are partially offset by a weak tendency for Dr~ to increase with increasing X cp~ .
Among the elements studied here, Zn shows the least compositional effect on partitioning. This probably reflects the fact that Zn substitutes simply for Fe in both phases, and the Zn contents of coexisting Cpx and Gnt tend to be very similar. There is some indication that Dz, decreases as Xca,~ t increases. The partitioning of Ni shows a weak dependence on XMg, reflecting the simi- larity of ionic radii between Ni and Mg. The partition- ing of Y and Sr appears to be controlled essentially by the Ca content of Gnt. The substitution of Sr into garnet probably is controlled primarily by ionic-radius effects, where Sr enters the larger site provided by Ca. Capo- ruscio and Smyth (1990) showed that the larger X site of grossular-rich garnets favours the uptake of inter- mediate rare-earth elements (MREE) relative to heavy rare-earth elements (HREE), and the ionic radius of Y~+ is most similar to that of Ho -~+. The substitution of Y into clinopyroxene must involve a charge balance, but our data do not identify the mechanism; defect substitution may be involved (Wood, 1974; B. Hensen, pers. commun., 1993). The data of Caporuscio and Smyth (1990) suggest that partitioning of Y also should be affected by X cpX, but this effect is not obvi- ous in our data, probably because of the overriding importance of XcC~, "t.
The partitioning of Ga between Gnt and Cpx is very strongly affected by yCpx --J0 , as well as being controlled by T and P. The P - T dependence ,~f ycp× within each ~ l i x jd
bulk composition makes these effects difficult to sep- arate. The strong correlation between the Na and Ga contents of Cpx suggests that Ga enters the Cpx lattice as part of the charge-balance mechanism. Both the Ca ~ Na and Na ~ [2]substitutions on the M2 site are compensated by trivalent cations substituting on the MI site, a contraction of the M1 polyhedron, and a more strongly negative M1 site potential (Caporuscio and Smyth, 1990; Oberti and Caporuscio, 1991 ) ; these features may favour the uptake of the Ga 3 ~ ion on M 1. The partitioning of Zr and Ti between Cpx and Grit
vcvx . this suggests that shows a weak dependence on l, jo , a similar mechanism affects the uptake of these ele- ments, even at low concentrations.
126 S. E O 'Reilly, W.L. Griffin / Chemical Geology 12 l (I 995) 105-130
6000
5000
x 4000 o
3000 P
2000
1000
I = a 4
10
1 I.- o o
e~
.1
.01
X X XX
x [3 i-i •
O + • o ~ ~x., ~ x
o , , i l ' x " ×
~+.~ A
++
0 2000 i i
4000 6000 in gnt
n
I
8 0 0 0
o
o o 0
0 +
0 +
" + , x t ++ + +
I I I
20 40 60 °/~1 d, ¢px
0 0 0 0
0 0 + &
o + O ~. ±x~O~+ ++ Ilaoc
+ A
all. A
i i i
0.5 0.6 0.7 0.8 0.9 1000/T,K
10 a
g l
.1 10000 0.5
~ 0 0 0
0 0 + A,
0 +
I I I
06 07 08 1000/T,K
10 I oo ° o d
O O + 4
& O +
A~ [ + + • , r
, 1 I I I I
80 0 10 20 30 40 50 60 XCa, gnt
0,9
Fig. I 1. Ti (ppm) in cl inopyroxene plotted against key chemical parameters and 1000/T. Dn(cor r ) = D i i - 2 . 2 X J ~ TM + 0.55 and accounts for
the effect of X.~'~ Px on D.r,. Symbols used are the same as defined for Fig. 2.
s. E 0 'Reilly, W.L. Griffin / Chemical Geology 121 (1995) 105-130 127
6.2. Temperature dependence
There appears to be little temperature dependence on the partitioning of Zn, Sr and Y between Gnt and Cpx; Dzn is essentially constant, and Dsr and Dy are controlled primarily by crystal-chemical factors. The partitioning of Ni and Zr shows a significant T effect, which overrides the modest effects of phase composi- tion; both elements partition preferentially into Grit at higher T. The partitioning of Ti may reflect a weak T effect, with D decreasing as T increases, but it is diffi- cult to separate this from the pressure effect with the available data.
The behaviour of the Argyle DI suite suggests that partitioning of Ga between Gnt and Cpx is strongly dependent on T, as well as on P and X cp~. At constant P and X cvX, D appears to increase with increasing T, as Ga preferentially enters Cpx.
6.3. Pressure dependence
The analysis in this study of the pressure dependence of partitioning is based on the difference in ambient geotherms for the African and B/G sample suites. For most elements, including Ni, Zn, Y, Sr and Zr, the B/G samples do not stand out from the African ones once compositional dependence has been removed. We therefore conclude that the partitioning of these ele- ments between Grit and Cpx in eclogites and pyroxe- nites shows no measurable P dependence.
The B/G garnets contain much less Ti than the coex- isting Cpx, in contrast to the African samples with similar T, and this indicates the existence of a depend- ence of Dw on P. This P effect may also account for the apparent decrease in Dw with increasing T within the African suite, since P increases rapidly relative to Talong the ~ 40-mW-m 2 paleogeotherm from which these samples were probably derived. Smith et al. ( 1991 ) showed that DT~ is higher in peridotite xenoliths sampled from elevated geotherms than in those from cratonic kimberlites; this suggests that a similar P effect is present in ultramafic rocks.
A similar situation applies in the case of Ga, after the dependence Of DG, on X cpX has been removed; the B/G samples show higher DGa than the African eclo- gites at similar T, implying that DGa decreases with increasing P. As for Ti, this P dependence may account for the general decrease of D with increasing T in the
African eclogite suites. The true T dependence of DG~ probably is shown by the Argyle DI pairs, where DG,~ increases sharply with increasing T. Scatter in these data may be due to lattice effects such as defect substi- tution, as strain may also be a significant factor but cannot be assessed by this dataset.
6.4. Applications o f partitioning data
None of these elements promise simply applicable alternative geothermometers or geobarometers for eclogitic systems. The corrections derived here for compositional effects are necessarily crude, the scatters about the D-Tregressions are still large, and in the case of Ga and Ti the strong P effects are not quantified.
However, these data do provide a series of tests for equilibrium between phases in xenolith samples, and hence can be used to improve the choice of appropriate samples for other studies, especially including isotopic analysis. Samples that fall well outside the trends shown here, after compositional effects are considered, probably are in disequilibrium, and the nature of the anomalies may provide guides to the processes that have produced the disequilibrium.
The broad patterns of trace-element partitioning described here also provide information on the envi- ronment of formation of Gnt-Cpx rocks. The partition- ing behaviour of Ga and Ti, in particular, is useful for recognizing eclogitic material from different P - T regimes. For example, the data on Ga partitioning in the Argyle DI pairs are consistent with the formation of these diamonds in thermal pulses occurring over relatively restricted ranges of P. Studies of other DI suites may reveal similar patterns. Eclogites or eclogitic DI pairs from asthenospheric P - T regimes, in which T increases adiabatically with P, should show anomal- ously low DG~ and DTi at high T, relative to the ' 'African eclogite" trends. Conversely, samples from strongly advective thermal regimes should stand out because of high DGa and Dw at relatively low T, as do the B/G samples.
These data also can be used to evaluate the validity of experimental determinations of partition coeffi- cients, and to guide the choice of experimental condi- tions and the compositions used in the experiments. For example, Green et al. (1989) report garnet/liquid and clinopyroxene/liquid partition coefficients for Zr and Y, derived from experiments on a basaltic composition
at 25 kbar and 1100°C. The garnets in the experimental charge are zoned, whereas the pyroxenes are homoge- neous; Dzr (Cpx/Gnt) -values calculated from their data range from 0.14 to 0.17; by comparison with the data shown in Fig. 10d and e, these values are low by perhaps a factor of 10. The values of Dv calculated for the same samples range from 0.08 to 0.12; after correc- tion for XcG2 t these values appear to be low by perhaps a factor of 5 (cf. Fig. 9). It seems likely that trace- element equilibrium was not attained in these experi- ments, even in the rims of grains, or that the analyses are flawed. A reanalysis of one of these samples by laser ablation microprobe-inductively coupled plasma mass spectrometry (LAM-ICP-MS) (Jenner et al., 1993) gave higher values of Dzr = 0.51 and Dy = 0.17, which still are lower than the data from natural samples at this T. Fujimaki et al. (1984) give Dzr=0.67 for coexisting garnet + clinopyroxene produced in an oli- vine tholeiite composition at 20 kbar and 1150°C. No X cpx is given, but it is expected to be low, and the raw value of Dzr falls within the range of our data (Fig. 10). Shimizu (1980) measured D s r in clinopyroxene- garnet pairs produced by experiments on high-A1 basalt and alkali basalt at 30 kbar and 1400-1500°C. His three Ds~-values range from 3.6 to 6.7 and are inversely cor- related with XcG2 t, which ranges from 0.16 to 0.27; these
values are consistent with the data shown in Fig. 8d and f. Similarly, Johnson (1993) reports data from a Kilauea basalt (Hawaii) composition at 1300-1470°C, 20-30 kbar (conditions well above even the south- eastern Australian geotherm of Fig. 1): DTi = 1.28; Dsr= 29: Dy = 0.13; Dzr=0.18. While these data can- not be corrected for phase composition with the data given in the abstract, the individual values fall within the ranges of those reported here. More data from such experiments will help to quantify the P dependence of
DTi.
As noted earlier, the estimation of temperature in Gnt-Cpx rocks, using Fe-Mg geothermometry, requires the calculation of Fe3+/Fe 2+ from stoichi- ometry. Since this calculation accumulates all of the analytical errors, the F e 3 + values typically are overes- timated; Fe 3 + also may be affected by late-stage alter- ation processes. The trace-element data may be used to cross-check the Fe 3÷ calculations. Specifically, the DNi-T and DZr-T relations described here can help to constrain the equilibration temperatures, and give an
indication of whether Fe-Mg temperatures derived by calculating Fe 3 + in Gnt and Cpx are realistic.
The general ranges of D~ established here for given values of P and T, and the qualitative dependence of D on bulk composition, provide important constraints on the values used in the mathematical modelling of partial melting, fractional crystallization and metasomatic processes in the mantle. Experimental data on clino- pyroxene/liquid partition coefficients are much more abundant than data on garnet/liquid coefficients; our data may be used to calculate appropriate garnet/liquid trace-element coefficients from the clinopyroxene/liq- uid data, and thus greatly extend the potential use of trace elements in modelling involving basaltic com- positions at high pressures.
Acknowledgements
John Gurney generously provided samples of dia- mondiferous eclogites and diamond-inclusion pairs which contributed significantly to the database, and Fanus Viljoen kindly provided samples from Roberts Victor and Kaalvallei. We have benefited from com- ments on the early drafts by Norm Pearson, Bas Hen- sen, Jing Guo, Ming Zhang, Heinz Stosch, Chris Ryan and Dmitri Ionov, and we thank them for their time and helpful discussions. Gordon Medaris and Doug Smith provided constructive reviews of the penultimate ver- sion. Jing Guo assisted with the data handling (a large task) and the diagrams. Chris Ryan provided invalua- ble help with the proton microprobe techniques and Tin Tin Win assisted with the proton microprobe analyses. This research was supported financially by the Austra- lian Research Council and the Macquarie University Research Grants Scheme.
References
Ballhaus, C., Berry, R.F. and Green, D.H., 1991. High pressure experimental calibration of the olivine-orthopyroxene-spinel oxygen barometer - - implications for redox conditions in the upper mantle. Contrib. Mineral. Petrol., 85: 85-94.
Boyd, F.R., 1976. Inflected and non-inflected geotherms. Carnegie Inst. Washington Yearbk., 75: 521-523.
Caporuscio, F.A. and Smyth, J.R., 1990. Trace element crystal chem- istry of mantle eclogites. Contrib. Mineral. Petrol., 105: 550- 561.
Ellis, D.J. and Green, D.H., 1979. An experimental study of the effect of Ca upon garnet-clinopyroxene Fe-Mg exchange equilibria. Contrib. Mineral. Petrol., 71: 13-22.
Finnerty, A.A. and Boyd, F.R., 1987. Thermobarometry for garnet peridotite xenoliths: a basis for mantle stratigraphy. In: P.H. Nixon (Editor), Mantle Xenoliths. Wiley, New York, N.Y., pp. 381-402.
Fujimaki, H., Tatsumoto, M. and Aoki, K., 1984. Partition coeffi- cients of Hf, Zr and REE between phenocrysts and ground- masses. Proc. 14th Lunar Conf., Part 2. J. Geophys. Res., 89(Suppl.): B662-B672.
Hatton, C.J. and Gurney, J.J., 1987. Roberts Victor eclogites and their relation to the mantle. In: P.H. Nixon (Editor), Mantle Xenoliths. Wiley, New York, N.Y., pp. 453-463.
Green, T.H., Sie, S.H., Ryan, C.G. and Cousens, D.R., 1989. Proton microprobe determined partitioning of Nb, Ta, Zr, Sr and Y between garnet, clinopyroxene and basaltic magma at high pres- sure and temperature. Chem. Geol., 74: 201-216.
Griffin, W.L. and O'Reilly, S.Y., 1986. Sapphirine in a mantle- derived xenolith from Delegate, Australia. Mineral. Mag., 50: 635-640.
Griffin, W.L., Wass, S.Y. and Hollis, J.D., 1984. Ultramafic xenoliths from Bullenmerri and Gnotuk maars, Victoria, Australia: petrol- ogy of a sub-continental crust-mantle transition. J. Petrol., 25: 53-87.
Griffin, W.L., Jaques, A.L., Sie, S.H., Ryan, C.G., Cousens, D.R. and Suter, G.F., 1988. Conditions of diamond growth: a proton microprobe study of inclusions in West Australian diamonds. Contrib. Mineral. Petrol., 99: 143-152.
Griffin, W.L., Gurney, J.J. and Ryan, C.G., 1992. Variations in trap- ping temperatures and trace elements in peridotite-suite inclu- sions from African diamonds: evidence for two inclusion suites, and implications for lithosphere stratigraphy. Contrib. Mineral. Petrol., 110: 1-15.
Griffin, W.L., Sobolev, N.V., Ryan, C.G., Pokhilenko, N.P., Win, T.T. and Yefimov, Y., 1993. Trace elements in garnets and chro- mites: diamond formation in the Siberian lithosphere. Lithos, 29: 235-256.
Gurney, J.J., Hams, J.W. and Rickard, R.S., 1979. Silicate and oxide inclusions in diamonds from the Finsch kimberlite pipe. In: F.R. Boyd and H.O.A. Meyer (Editors), Kimberlites, Diatremes and Diamonds. Am. Geophys. Union, Washington, D.C., pp. 1-15.
Gurney, J.J., Harris, J.W. and Rickard, R.S., 1986a. Premier Mine diamond inclusions. Univ. of Cape Town, Stellenbosch, Kim- berlite Res. Group, Intern. Rep. No. 2.
Gurney, J.J., Hams, J.W., Rickard, R.S. and Moore, R.O., 1986b. Silicate and oxide inclusions in diamonds from the Orapa mine, Botswana. Univ. of Cape Town, Stellenbosch, Kimberlite Res. Group, Intern. Rep. No. 4.
Jaques, A.L., Hall, A.E., Sheraton, J.D., Smith, C.B., Sun, S.-s., Drew, R.M., Foudalis, C. and Ellingsen, K., 1990. Composition of crystalline inclusions and C-isotopic composition of Argyle and Ellendale diamonds. In: Kimberlites and Related Rocks, Vol. 2: Their Mantle/Crust Setting. Geol. Soc. Aust., Spec. Publ. No. 14. Blackwells, Melbourne, Vic., pp. 966-989.
Jenner, G.A., Foley, S.F., Jackson, S.E., Green, T.H., Fryer, B.J. and Longerich, H.P., 1993. Determination of partition coefficients
for trace elements in high pressure-temperature experimental run products by laser ablation microprobe-inductively coupled plasma-mass spectrometry (LAM-ICP-MS). Geochim. Cos- mochim. Acta, 57:5099-5104.
Johnson, K.T.M., 1993. Experimental cpx/ and garnet/melt parti- tionong of REE and other trace elements. Eos (Trans. Am. Geo- phys. Union), 1993 Fall Meet., p.658.
Krogh, E.J., 1988. The garnet-clinopyroxene Fe-Mg geothermo- meter - - a reinterpretation of existing experimental data. Contrib. Mineral. Petrol., 99: 44-48.
MacGregor, I.H. and Carter, J.L., 1970. The chemistry of clinopy- roxenes and garnets of eclogite and peridotite xenoliths from the Roberts Victor mine, South Africa. Phys. Earth Planet. Inter., 3: 391-397.
McCandless, T.E. and Gurney, J.J., 1989. Sodium in garnet and potassium in clinopyroxene: criteria for classifying mantle eclo- gites. In: Kimberlites and Related Rocks, Vol. 2: Their Mantle/ Crust Setting. Geol. Soc. Aust., Spec. Publ., No. 14. Blackwells, Melbourne, Vic., pp. 827-832.
McKay, D.B. and Mitchell, R.H., 1988. Abundance and distribution of gallium in some spinel and garnet lherzolites. Geochim. Cos- mochim. Acta, 52: 2867-2870.
Moore, R.O. and Gurney, J.J., 1989. Mineral inclusions in diamond from the Monastery kimberlite, South Africa. In: Kimberlites and Related Rocks, Vol. 2: Their Mantle/Crust Setting. Geol. Soc. Aust., Spec. Publ. No. 14. Blackwells, Melbourne, Vic., pp. 1029-1041.
Moore, R.O., Gumey, J.J., Griffin, W.L. and Shimizu, N., 1991. Ultra-high pressure garnet inclusions in Monastery diamonds: trace element abundance patterns and conditions of origin. Eur- opean. J. Mineral., 3: 213-230.
Neumann, E.-R., 1976. Two refinements for the calculation of struc- tural formulae for pyroxenes and amphiboles. Nor. Geol. Tidsskr., 56: 1-6.
Oberti, R. and Caporuscio, F.A., 1991. Crystal chemistry of clino- pyroxenes from mantle eclogites: a study of the key role of the M2 site population by means of crystal-structure refinement. Am. Mineral., 76:1141-1152.
O'Reilly, S.Y., 1989. Australian xenolith types, distribution and transport. In: R.W. Johnson (Editor), lntraplate Volcanism in Eastern Australia and New Zealand. Cambridge University Press, Cambridge, pp. 249-253.
O'Reilly, S.Y. and Griffin, W.L., 1985. A xenolith-derived geotherm for southeastern Australia and its geophysical implications. Tec- tonophysics, 111: 41-63.
O'Reilly, S.Y., Griffin, W.L. and Ryan, C.G., 1991. Residence of trace elements in metasomatized spinel lherzolite xenoliths: a proton microprobe study. Contrib. Mineral. Petrol., 109: 98-~ 13.
Pollack, H.N. and Chapman, D.S., 1964. On the regional variation of heat flow, geotherms and lithospheric thickness. Tectono- physics, 38: 279-296.
Rickard, R.S., Hams, J.W., Gumey, J.J. and Cardoso, P., 1989. Mineral inclusions in diamond from the Koffiefontein Mine. In: Kimberlites and Related Rocks, Vol. 2: Their Mantle/Crust Set- ting. Geol. Soc. Aust., Spec. Publ. No. 14. Blackwells, Mel- bourne, Vic., pp. 1054-1062.
130 S. E O 'Reill_v, W.L. Gri~lin / Chemical Geology 121 (1995) 105-130
Robinson, D.N., Gurney, J.J. and Shee, S.R., 1984. Diamond eclogite and graphite eclogite xenoliths from Orapa, Botswana. In: J. Kornprobst (Editor), The Mantle and Crust-Mantle Relation- ships. Elsevier, Amsterdam, pp. 11-24.
Ryan, C.G., Cousens, D.R., Sie, S.H., Griffin, W.L. and Suler, G.F., 1990. Quantitative PIXE microanalysis of geological material using the CSIRO proton microprobe. Nucl. lnstrum. Methods, B47:55-7 I.
Shimizu, H., 1980. Experimental study on rare-earth element parti- tioning in minerals formed at 20 and 30 kb for basaltic systems. Geochem. J., 14: 185-202.
Smith, D., Griffin, W.L., Ryan, C.G. and Sie, S.H., 1991. Trace- element zonation in garnets from The Thumb: Heating and melt
infiltration below the Colorado Plateau. Contrib. Mineral. Petrol., 107:60 79.
Viljoen, K.S., 1994~ The petrology and geochemistry of a suite of mantle-derived eclogite xenoliths l~om the Kaalvallei kimberlite, South Africa. Ph.D. Thesis, University of the Witwatersrand, Johannesburg, 160 pp.
Viljoen, K.S., Robinson, D.N., Swash, P.M., Griffin, W.L., Otter, M.L., Ryan, C.G. and Win, T.T., 1994. Diamond- and graphite- bearing peridotite xenoliths from the Roberts Victor Kimberlite, South Africa. In: H.O.A. Meyer and O.H. Leonardos (Editors), Kimberlites, Related Rocks and Mantle Xenoliths. Comp. Pesq. Recurs. Miner., Spec. Publ., IA/93: 285-303.
Wood, B.J., 1974. The solubility of alumina in orthopyroxene coex- isting with garnet. Contrib. Mineral. Petrol., 46: 1-15.
