Journal of Petrology 2004 45(7):1339-1368; doi:10.1093/petrology/egh021
Journal of Petrology 45(7) © Oxford University Press 2004; all rights reserved
Partial Melting of Spinel Lherzolite in the System CaO–MgO–Al2O3–SiO2 ± K2O at 1·1 GPa
XI LIU* and
HUGH ST. C. O'NEILL
RESEARCH SCHOOL OF EARTH SCIENCES, AUSTRALIAN NATIONAL UNIVERSITY, CANBERRA, A.C.T. 0200, AUSTRALIA
RECEIVED
MAY 7, 2003;
ACCEPTED
JANUARY 22, 2004
 |
ABSTRACT
|
|---|
The compositions of multiply saturated partial melts are valuable
for the thermodynamic information that they contain, but are
difficult to determine experimentally because they exist only
over a narrow temperature range at a given pressure. Here we
try a new approach for determining the composition of the partial
melt in equilibrium with olivine, orthopyroxene, clinopyroxene
and spinel (Ol + Opx + Cpx + Sp + Melt) in the system CaO–MgO–Al
2O
3–SiO
2 (CMAS) at 1·1 GPa: various amounts of K
2O are added to
the system, and the resulting melt compositions and temperature
are extrapolated to zero K
2O. The ‘sandwich’ experimental
method was used to minimize problems caused by quench modification,
and Opx and Cpx were previously synthesized at conditions near
those of the melting experiments to ensure they had appropriate
compositions. Results were then checked by reversal crystallization
experiments. The results are in good agreement with previous
work, and establish the anhydrous solidus in CMAS to be at 1320
± 10°C at 1·1 GPa. The effect of K
2O is to
depress the solidus by 5·8°C/wt %, while the melt
composition becomes increasingly enriched in SiO
2, being quartz-normative
above 4 wt % K
2O. Compared with Na
2O, K
2O has a stronger effect
in depressing the solidus and modifying melt compositions. The
isobaric invariant point in the system CMAS–K
2O at which
Ol + Opx + Cpx + Sp + Melt is joined by sanidine (San) is at
1240 ± 10°C. During the course of the study several
other isobaric invariant points were identified and their crystal
and melt compositions determined in unreversed experiments:
Opx + Cpx + Sp + An + Melt in the system CMAS at 1315 ±
10°C; in CMAS–K
2O, Opx + Cpx + Sp + An + San + Melt
at 1230 ± 10°C and Opx + Sp + An + San + Sapph +
Melt at 1230 ± 10°C, where An is anorthite and Sapph
is sapphirine. Coexisting San plus An in three experiments help
define the An–San solvus at 1230–1250°C.
KEY WORDS: feldspar solvus; igneous sapphirine; mantle solidus; partial melting; systems CMAS and CMAS–K2O
|
INTRODUCTION
|
|---|
The simplest chemical system that contains all the major phases
of peridotitic upper mantle (olivine, orthopyroxene, clinopyroxene,
and an aluminous phase, one of plagioclase, spinel or garnet,
depending on pressure) is CaO–MgO–Al
2O
3–SiO
2 (CMAS). This simple system has been used to investigate the
basic phase relations of partial melting in the upper mantle
(O'Hara, 1968
; Kushiro, 1972
; Presnall
et al., 1979
; Longhi,
1987
; Gudfinnsson & Presnall, 1996
; Herzberg & Zhang,
1998
; Milholland & Presnall, 1998
; Presnall, 1999
; Liu &
Presnall, 2000
). However, with a typical four-phase lherzolite
assemblage, the initial melting in CMAS is isobarically invariant;
that is, the chemical potentials of all four components (CaO,
MgO, Al
2O
3 and SiO
2) in the melt are completely defined, as
in
 | (1) |
 | (2) |
 | (3) |
and, in the Sp-lherzolite stability field,
 | (4) |
Because all chemical potentials are known, the composition of the melt in equilibrium with all four solid phases is particularly valuable in constructing thermodynamic models of silicate melts. Unfortunately, the isobarically invariant melt composition is difficult to determine experimentally, because the five-phase assemblage exists only over an infinitely narrow temperature interval at a given pressure.
The traditional way around this problem is to bracket the isobaric invariant point by varying the experimental starting composition, to produce four-phase isobarically univariant assemblages (i.e. three solid phases plus melt as in Fo + Opx + Sp + Melt). Not only does this require a large number of experiments, but it is tedious to demonstrate equilibrium (Presnall et al., 1978; Liu & Presnall, 1990), as the solid phases (especially the pyroxenes) are multicomponent solid solutions with the potential to be compositionally unequilibrated. This method also requires some extrapolation to the isobaric invariant point, as this point is only bracketed.
Alternatively, it is possible to rely on experimental imperfections, such as temperature gradients or chemical impurities (e.g. some Na2O in the starting material or H2 diffusing through the Pt capsule to produce H2O in the experimental charge) to increase the effective variance of the system and obtain melt in equilibrium or quasi-equilibrium with four solid phases, which can then be analysed directly. This approach has been adopted by Presnall and colleagues (Presnall et al., 1979; Gudfinnsson & Presnall, 1996; Milholland & Presnall, 1998; Liu & Presnall, 2000). Although this seems empirically to work well for the system CMAS, this may not always be the case, and we have encountered insuperable difficulties in applying this approach to determining the effect of Cr2O3 on melting in the system CMAS–Cr2O3. Relying on semi-controlled imperfections to obtain a result is also intellectually unsatisfying, and the question arises of how much the imperfections affect results.
In this paper we report a logical improvement to this approach, which is to introduce another component into the system deliberately. The system is then studied as a function of the controlled concentration of this component, such that the composition of the CMAS isobaric invariant melt can be obtained by extrapolating to zero concentration of the added component. We chose K2O as the additional component (the K2O-method hereafter), as it is almost completely incompatible in all the solid phases (Ol, Opx, Cpx, Sp) at our chosen experimental pressure of 1·1 GPa. Having the extra component entering only the melt phase makes the extrapolation to the pure CMAS system simple. Also, K2O is an important constituent in several mantle-derived magma types (kimberlites, shoshonites, and so on), so that our data for the system CMAS–K2O generated as a by-product of our main purpose should be of some direct petrological relevance.
We show that the K2O-method produces a result in good agreement with the previous work of Presnall and colleagues (Presnall, 1976; Presnall et al., 1979; Walter & Presnall, 1994), and our improved accuracy allows for some refinement of this work.
|
EXPERIMENTAL AND ANALYTICAL TECHNIQUES
|
|---|
Four types of experiments were performed:
- conventional direct partial melting experiments (DEs) in the system CMAS;
- sandwich experiments using the K2O-method (KEs), in which the solid assemblage Fo + Opx + Cpx + Sp was placed at the top and bottom of the capsule, with a K2O-containing melt composition as the ‘filling’;
- reversal experiments (REs-1) to study the crystallization of the CMAS melt composition deduced from the KEs;
- reversal experiments (REs-2) with Fo added; all but one of these (C-1789) used the sandwich configuration.
Starting materials
Table 1 summarizes the starting materials used in this study. Compositions were made from high-purity oxides (SiO2, Al2O3 and MgO) and carbonates (CaCO3 and K2CO3). An important strategy was to use pre-synthesized Opx and Cpx with compositions similar to those expected in the run products.
Mix9 was prepared by combining a crystalline mixture of pure
forsterite (Fo) and a crystalline mixture of Sp + Opx + Cpx
in a proportion of 1:5 by weight. The Fo was synthesized at
1400°C and atmospheric pressure for 69 h; the mixture with
Sp:Opx:Cpx in weight proportions of 1:2:2 was made in a 5/8
inch piston-cylinder press using a 3·5 mm diameter Pt
capsule and a talc–Pyrex assembly at 1280°C and 1·1
GPa for 48 h. The phase proportion in the final Mix9 was close
to 1:1:2:2 (Fo:Sp:Opx:Cpx). Crystalline mixture SEM02-1 containing
Fo:Sp:Opx: Cpx = 1:1:1:1 (by weight) was made by crystallizing
the decarbonated oxide mix at 1300°C and 1·1 GPa
for 48 h in a Pt capsule. Mixtures SEM02-3, SEM02-4, SEM02-8
and SEM02-6 were melted at 1400°C and 1 atm for 20 min and
then quenched to glass.
All mixtures were checked for compositions with electron microprobe (Ware, 1991). Mix9 was later stored in an oven at 110°C and other starting materials were stored in another oven at 150°C.
Piston-cylinder assemblies
All experiments were made in a conventional half-inch piston-cylinder apparatus (Boyd & England, 1960). The salt–Pyrex pressure assembly used in the preliminary DEs was described by Klemme & O'Neill (1997), except that, because of the generally lower pressure and temperature regime of this study, fired pyrophyllite replaced the MgO spacer underneath the capsule and mullite was used instead of high-purity alumina for the thermocouple tube. Further modification for the assembly used in the KEs and the REs were made because of possible H2 diffusion into the Pt capsule causing contamination of the runs by H2O. In these experiments the Pt capsule was held in an Fe2O3 sleeve, which was in turn surrounded by an alumina sleeve. At each end of the alumina sleeve, a ruby disc (0·5 mm thick) separated the Fe2O3 sleeve from other parts of the assembly. This structure prevents the Fe2O3 sleeve from being reduced by the graphite heater, or contaminating the thermocouple. Alumina spacers and then MgO spacers were positioned next to both ruby discs, to enhance the mechanical stability of the assembly. The thermocouple was protected by a combination of high-purity alumina tubing in the hot part of the assembly (
4 mm long), followed by mullite tubing above this. The Fe2O3 sleeve was made by cold pressing in a steel die and then sintered at atmospheric pressure and 850°C for 3 h, using acetone as a binder.
These salt–Pyrex assemblies have low friction and no pressure correction is required, considering the high temperatures and long run times used in this study (Green et al., 1966; Bose & Ganguly, 1995; Klemme & O'Neill, 1997).
Experimental procedures
For each experiment, 8–10 mg starting materials were loaded into a Pt capsule. The capsules used in the DEs were stored at 110°C for 6–8 h before welding. The Pt capsules for the KEs and REs were stored at 150°C for 6–8 h, then held in a steel block that had been pre-heated to 750°C while they were welded (Robinson et al., 1998).
All experiments were performed using the ‘piston-out’ method, i.e. the pressure was first raised to several MPa, then the temperature was raised to c. 450°C to soften the Pyrex sleeve. The pressure was then increased to c. 0·05 GPa higher than the desired pressure, the temperature was raised to the nominal temperature of the run, and finally the pressure was lowered to the required pressure (Johannes et al., 1971). Pressures were continuously monitored and adjusted, if necessary, which allowed each run to be controlled within ±0·02 GPa of the nominal pressure. Temperature was measured and controlled with Pt94Rh6–Pt70Rh30 thermocouple (type B), previously calibrated against the melting point of gold at atmospheric pressure; possible pressure effects on the e.m.f. of the thermocouple were neglected. The tip of the thermocouple, the upper ruby disc and the whole Pt capsule containing the experimental charge were all carefully placed in the 5-mm-long hot spot of the experimental assembly. Although temperature during experiments was controlled to ±1°C, the true temperature uncertainties are estimated to be c. ±5–10°C, mainly because of slight differences in sample position and power density through the graphite heater. Such an uncertainty is consistent with the variations in melt composition and other experimental parameters observed in these experiments. A discussion of the superior merits of type B thermocouples over other Pt/Rh thermocouples or W/Re thermocouples at temperatures near 1300°C has been given by Klemme & O'Neill (2000b).
Analytical methods
At the end of a run, the sample was sectioned longitudinally, mounted in epoxy and polished using a series of diamond pastes. Run products were carbon-coated and analysed on a Cameca electron microprobe at the Research School of Earth Sciences (RSES), ANU and/or on a JEOL 6400 scanning electron microprobe in energy dispersive mode (EDS) at the Electron Microprobe Unit (EMU) at ANU. Coexisting phases in all run products were identified by back-scattered electron imaging. Beam current was 1 nA, accelerating voltage was 15 keV and ZAF correction was applied in all analyses (Ware, 1991). A beam spot size of 1 µm was used for all crystalline phases, and both 1 µm and 10 µm beam spot sizes were used for glass analyses. Calibration was based on optimization to a large number of standards; to check this calibration and also as a monitor of any drift between analysing sessions, we repeatedly analysed three glass standards, GOR132G, T1G, and KL2G, which have comparable compositions to the phases present in this study (Jochum et al., 2000). The 185 analyses over 24 analytical sessions are summarized in Table 2. There is excellent agreement between our electron probe analyses and the recommended values.
Besides the major oxides CaO, MgO, Al
2O
3, SiO
2, and K
2O expected
in the run products, we also routinely analysed for FeO, Cr
2O
3,
and Na
2O in all phases. Either diffusion of Fe through the Pt
capsule from the Fe
2O
3 sleeve or a leak in the capsule could
result in Fe contamination. Na
2O appears to be a good indicator
of the purity of the starting materials. It concentrates in
the melt, and 0·1–0·3 wt % was found in
the glasses in this study, depending on the proportion of melt
to solid in the experiment. Such a concentration is close to
the limit of detection by EDS, and has been ignored as it has
a minimal effect on the phase relations (Walter & Presnall,
1994
; Liu
et al., in preparation).
Lithium and boron have potentially been cryptic contaminants of previous piston-cylinder studies, as only recently have these elements become amenable to microbeam analysis. Li salts are sometimes used in noble-metal fabrication processes, and B could come from the Pyrex glass (it apparently diffuses readily through Pt). To check for any such contamination, Li and B were measured in the glasses in runs C-1422, C-1448 and C-1565 (see Table 3 for experimental conditions) by laser ablation–inductively coupled plasma–mass spectrometry at RSES, ANU. The Li contents were 4·1 ± 0·2 ppm, 4·1 ± 0·1 ppm and 2·8 ± 0·4 ppm, respectively. The B contents were slightly higher at 53 ± 2, 54 ± 2 and 52·3 ± 7 ppm. We assume these levels have a negligible effect on phase relations.
Fe2O3 sleeve and H2O contamination
Following Robinson
et al. (1998)
, Fe
2O
3 sleeves were used as
H
2-getters surrounding the Pt capsule in all experiments apart
from the first few reported in this study (the DEs). The idea
is that an oxidized substance such as Fe
2O
3 should react with
H
2, which may be produced in the piston-cylinder assembly by
contaminant moisture reacting with the graphite heater, for
example. The getter should minimize the ingress of H
2 into the
Pt capsule by diffusion, where it can react with silicates (e.g.
alloying some Si into the Pt in FeO-free systems), thus forming
H
2O. According to Robinson
et al. (1998)
, the use of Fe
2O
3 sleeves
coupled with careful drying of capsules before and during welding
can reduce the H
2O content in experimental melts from
1 wt %
to
0·1 wt %.
The Fe2O3 sleeves after two experiments (C-1417 and C-1566; see Table 3 for experimental conditions) were checked by powder X-ray diffraction at the Geology Department, ANU and were found to be >99 wt % Fe2O3. This does not mean that the sleeves were unnecessary, however, as analysis of the Pt capsules (using the Cameca electron microprobe in wavelength-dispersive mode at RSES, ANU) showed an Fe diffusion gradient. An example is given in Fig. 1 (1340°C, 1·1 GPa, 72 h). Whereas Fe metal reached the middle point of the capsule wall, the experimental charge remained untouched. Longer run times or higher temperatures would result in Fe reaching the inner wall of the capsule, hence the need to analyse Fe routinely in run products.
View larger version (11K):
[in this window]
[in a new window]
|
Fig. 1. Profile of Fe diffusion into a Pt capsule wall. Experimental conditions: 1·1 GPa, 1340°C, 72 h. Curve is fitted by eye.
|
|
C-1621 contains only melt (see
Table 3 for experimental conditions)
and is ideal for using transmission infrared spectroscopy to
determine the water content. Details of this procedure as implemented
in this study will be reported by Liu
et al. (in preparation).
The average water content from seven analyses is 120 ppm, which
is substantially lower than the 0·1% reported by Robinson
et al. (1998)
and the 1100 ppm by Falloon
et al. (1999)
; Baker
et al. (1996)
and Falloon
et al. (2001)
estimated an even larger
amount, 0·5–1%, in their nominally anhydrous experiments.
If the same total of H
2O as in run C-1621 were concentrated
into the melt phase in those runs with smaller proportions of
melt, the H
2O concentrations might be higher than in C-1621,
but it seems unlikely that levels of H
2O in any run with the
Fe
2O
3 sleeves would exceed 0·1%.
Attainment of equilibrium
There is a fundamental dilemma to be addressed in designing experiments aimed at obtaining the compositions of partial melts at modest degrees of melt fraction. To optimize equilibration between melt and crystals, it is obviously advantageous to have the melt distributed between crystals, to maximize mutual contacts. This is the usual texture formed in simple experiments (the DEs of this study) with less than about 20–30% melt, in the absence of a temperature gradient. However, maximizing the contact between melt and crystals also maximizes the probability of quench modification by precipitation on the rims of crystals. In addition, melt so distributed is difficult to analyse accurately, as melt pockets are small and irregularly shaped. This latter problem becomes worse as the melt fraction decreases. The experimental solution to these problems has been to devise ways of separating the melt from the crystals, as in sandwich experiments (as in this study and Stolper, 1980; Takahashi & Kushiro, 1983; Fujii & Scarfe, 1985; Falloon & Green, 1987, 1988; Robinson et al., 1998) or by extraction into diamond aggregates or similar (Baker et al., 1992; Johnson & Kushiro, 1992; Kushiro & Hirose, 1992; Hirose & Kawamoto, 1995; Hirose, 1997). Another possible tactic is to exploit the temperature gradient present in many high-pressure experiments, particularly in multi-anvil experiments, to separate melt towards the hot end of the capsule (Takahashi, 1986). However, the inevitable consequence of any separation is to impair the chances of attaining equilibrium between melt and crystals.
The problems that this can cause are well illustrated in this study by two experiments, C-1580 and C-1576 (Table 3; see Fig. 2a), in which the initial composition chosen for the ‘filling’ part of the sandwich turned out not to be in equilibrium with the Fo + Opx + Cpx + Sp assemblage at the ends of the sandwich (the ‘bread’), despite the long run times used in these experiments. Instead, the run products consisted of three zones: (1) the initial Fo + Opx + Cpx + Sp assemblage, which in these two runs is unequilibrated with respect to the pyroxene compositions (particularly Opx; see Table 4); (2) the ‘filling’ part, comprising melt plus An + Sapph + San (C-1580) or, in the lower-temperature experiment (C-1576), melt plus An + Sapph + San + Sp + Opx, the latter of which has a very different composition from the Opx in the Fo + Opx + Cpx + Sp assemblage; (3) a narrow reaction zone of Opx, no more than 20 µm wide, between the two other zones (see below). These textures are shown in Fig. 2a; for comparison, the texture developed in a successfully equilibrated sandwich experiment is given in Fig. 2b. The conspicuous inability of the melts in the ‘filling’ to re-equilibrate with the solid Fo + Opx + Cpx + Sp assemblage in these two runs may be due to their high SiO2, and hence high viscosity.
View larger version (101K):
[in this window]
[in a new window]
|
Fig. 2. Electron back-scatter images showing the texture of experiments: (a) C-1576, a disequilibrium sandwich experiment in which the original glass zone, now crystallized to Sp + Opx + An + San + Sapph + Melt, is separated from the Fo + Opx + Cpx + Sp zone by a layer of Opx (labelled ‘Reaction zone’); (b) C-1448, a typical experiment in which equilibrium between the original zones is inferred. (See Table 3 for experimental conditions and Table 4 for phase compositions.)
|
|
The large difference in the composition of Opx in the different
zones of C-1576 is instructive. Because the composition of a
solid solution phase depends on its chemical environment, it
is unlikely to be the same in the two different parts of a sandwich
experiment unless these parts are in equilibrium. We therefore
paid particular attention in this study to analysing pyroxene
compositions in the ‘filling’ part of the sandwich
experiment, to check that they were indeed the same within analytical
error as in the ‘bread’ layers. Some crystallization
in the ‘filling’ part is inevitable unless the initial
glass composition is exactly that expected at equilibrium; in
fact, as regards attainment of equilibrium, an absence of crystals
in the ‘filling’ would be an ambiguous result, in
the same way that ‘no reaction’ in the determination
of a subsolidus univariant reaction indicates either full equilibrium
at the
P–
T condition or merely sluggish reaction kinetics.
Clearly, a key aspect of any partial melting experiment in which the melt is separated from the crystals is demonstration of equilibrium between melt and crystals. We shall point to some other runs where we believe such equilibrium was not obtained, based on the following criteria (as discussed above): (1) careful examination of the interface between the melt-rich and the crystal-rich parts; (2) comparison of compositions of solid-solution phases in the two parts. On the basis of these criteria, equilibrium was obtained in the majority of the sandwich experiments reported here. Application of these criteria may not be possible in diamond-aggregate or similar types of melt extraction experiments, in which case the results from these experiments must be viewed with a degree of caution.
Several lines of evidence indicate that local equilibrium was approached closely in the experiments of this study. In addition to being the same in different parts of the sandwich, the compositions of Opx and Cpx are generally fairly homogeneous: the most heterogeneous oxide component, Al2O3, in the most heterogeneous phase, Cpx, generally has a variation of <0·5 wt % one standard deviation, which is comparable with that observed in other simple-system experiments. The amounts of Al2O3 in both Opx and Cpx show small but smooth variations with temperature, and are in good agreement with the results of previous studies (see below). Similarly, the average difference between temperatures calculated from the Ca-geothermometer of Nickel et al. (1985) and the nominal experimental temperatures is only 23°C, consistent with equilibrium Ca exchange between Opx and Cpx.
The achievement of local equilibrium in this work is expected because the duration of our experiments is generally longer than that used in previous comparable studies. The results of previous near-liquidus experiments in the system CMAS (Presnall et al., 1978; Sen & Presnall, 1984; Liu & Presnall, 1990) suggested that a period of several hours is all that is required to establish reversals of phase boundaries. Walter & Presnall's (1994) experiments in the system CMAS–Na2O found that 48 h was long enough for the attainment of equilibrium at solidus temperatures. Most of our run durations were longer than this.
|
EXPERIMENTAL RESULTS
|
|---|
Table 3 summarizes the starting materials, the run conditions,
the techniques used, and the results in terms of phases present,
for each experiment. The compositions of these phases are given
in
Table 4. Calculated temperatures from the two-pyroxene geothermometer
of Nickel
et al. (1985)
for those experiments containing both
Opx and Cpx are shown in
Table 3. There is a good correlation
between calculated and experimental temperatures, although with
a systematic offset towards the calculated temperatures being
higher at low experimental temperatures, with a mean difference
of 23°C (see
Fig. 5).
The initial DEs bracket the solidus in the system CMAS between 1300°C (C-1555) and 1310°C at 1·1 GPa. The run at 1300°C contains a small amount of melt (too small to analyse). The melt may have been produced as a result of H2 diffusion into the capsule during the run, producing a trace of water by reducing SiO2 to Si, which alloys with Pt (Chen & Presnall, 1975). At slightly higher temperature (1310°C, C-1556), one phase (Cpx) disappears in agreement with the phase rule. We did not observe the four solid phases coexisting with melt over a range of temperature (20°), as did Presnall (1976) and Presnall et al. (1979). This is consistent with only small amounts of chemical impurities (such as H2O) in our run products, and minimal temperature gradients.
Fourteen experiments were carried out using the K2O-addition method (KEs). The full Sp-lherzolite assemblage coexisting with melt is observed in seven experiments between 1260 and 1310°C. The temperatures of these KEs are plotted against the K2O contents of the melts in Fig. 3. Within experimental uncertainty the relationship is linear, at least to 8 wt % K2O and 1270°C. Extrapolating K2O to zero gives a solidus temperature of 1319°C for the system CMAS. The change of melt composition with K2O is shown in Fig. 4. On a simple oxide weight percent (wt %) basis, SiO2 increases linearly with K2O, MgO and CaO both decrease linearly, and Al2O3 remains approximately constant, although in detail it appears to go up then down, with a maximum at 6 wt % K2O. The melt composition at the solidus in the CMAS system is found by extrapolation to zero K2O, as for the solidus temperature: we obtain (in wt %) SiO2 49·09, Al2O3 20·14, MgO 15·45 and CaO 15·33. The compositions of both Opx and Cpx at the CMAS solidus can be deduced similarly (see Table 5).
View larger version (22K):
[in this window]
[in a new window]
|
Fig. 3. Relationship between nominal experimental temperature and K2O content of melt for the KEs. The uncertainty in experimental temperatures is assumed to be ±10°. A linear correlation between nominal experimental temperature and K2O content of the melt is observed at low K2O contents (<8·5 wt %) for melts in equilibrium with the Sp-lherzolite phase assemblage. The solidus temperature of 1319°C for the Sp-lherzolite phase assemblage in the system CMAS at 1·1 GPa was obtained by linear regression of K2O vs T for experiments with K2O <7 wt %, and extrapolating the resulting regression parameters to zero K2O.
|
|
View larger version (20K):
[in this window]
[in a new window]
|
Fig. 4. Variation diagram showing oxide trends for melts in the KEs. Symbols as in Fig. 3. One standard deviation of the oxide content is only occasionally larger than the symbols. There is a strong linear correlation between each of SiO2, Al2O3, MgO, and CaO with K2O. Linear extrapolation of the data at <7 wt % K2O to zero K2O yields the solidus melt composition of 49·09 wt % SiO2, 20·12 wt % Al2O3, 15·45 wt % MgO and 15·33 wt % CaO.
|
|
View this table:
[in this window]
[in a new window]
|
Table 5: Compositions of Melt, Opx and Cpx in the system CMAS at 1·0 GPa or 1·1 GPa in the isobaric invariant assemblages Fo + Sp + Opx + Cpx + Melt and Sp + Opx + Cpx + An + Melt: comparison with literature data
|
|
We then tested these results by making up a glass with this
composition and locating its liquidus and solidus (REs-1). The
experiment at 1340°C was above the liquidus, whereas two
runs at 1320°C (C-1639 and C-1565) crystallized Sp + Opx
+ Cpx ± An, with the residual melt being close to that
in the starting material (
Table 4). The compositions of the
pyroxenes in C-1639 are in good agreement with the forward experiments,
whereas those in C-1565, with a shorter run time and An present,
have slightly higher Al
2O
3 (
Table 4). These experiments confirm
directly the peritectic nature of the melting reaction, as Fo
was not crystallized. They also imply that the solidus of the
phase assemblage Sp + Opx + Cpx + An is only slightly lower
than that of the phase assemblage Fo + Sp + Opx + Cpx, in agreement
with Kushiro (1972)
. Thus the run at 1310°C (C-1566) is
almost completely crystallized. Although in one experiment (C-1565)
melt coexists with four crystalline phases, the temperature
interval over which this occurs is clearly very narrow (an infinitely
narrow temperature interval is expected from the phase rule).
The REs-1 demonstrate empirically that simple crystallization experiments cannot in practice reverse the partial melting equilibrium when this is peritectic, as here (i.e. the melt is in reaction relationship with Fo). Accordingly, we tried reversal crystallization experiments with added Fo. These experiments (called REs-2) directly constrain the solidus of the Sp-lherzolite phase assemblage in the system CMAS at 1·1 GPa to be 1315 ± 10°C. The experiment at this temperature, C-1781, produced four crystalline phases plus melt; however, the melt composition is slightly less magnesian than the composition deduced from the KEs (Table 5), and is intermediate between this composition and that in equilibrium with Opx + Cpx + Sp + An. The Al2O3 content of the pyroxenes is also slightly higher than in other comparable runs. This raises the question of whether equilibrium between Fo and the ‘filling’ part of the sandwich was truly established. Clearly, equilibrium was not established in a similar experiment at the slightly lower temperature of 1310°C, C-1768; here the ‘filling’ crystallized completely to Sp + Opx + Cpx + An, with An separated from Fo by a narrow reaction zone of Opx + Cpx + Sp. The further implications of this experiment for the interpretation of results from melt-extraction experiments (such as the diamond-aggregate technique) will be discussed below.
Table 5 compares the result of this study with previous work in CMAS at similar pressures in the literature. There is a difference of 30°C between the solidus as determined here and the solidus defined by run 116-3 of Presnall (1976), which contains Fo + Opx + Cpx + Sp + Melt at 1350°C and 1·1 GPa. However, another run with the same starting composition reported by Presnall et al. (1979), their 122-8 at 1327°C and 1·1 GPa, contained Fo + Opx + Melt and is therefore above the solidus, which would be in agreement with our result. There is also good agreement as regards the solidus melt composition and the solidus pyroxene compositions. It may be noted that Presnall (1976) reported 8·4 ± 0·2 wt % Al2O3 for his Opx based on five analyses, but noted two outliers at 9·6 and 5·5 wt % Al2O3, indicating that his Opx was somewhat heterogeneous in composition. The higher amount (9·49 wt %) later reported by Walter & Presnall (1994) following re-analysis of this experiment may be due to the inclusion of high-Al2O3 outliers. The solidus temperature and melt composition reported by Kushiro (1972) at 1·0 GPa agree slightly less well, the main difference in melt composition being in CaO (Table 5). Kushiro (1972) also reported data for the phase assemblage Sp + Opx + Cpx + An + Melt; in agreement with this study, this isobaric invariant point occurs at essentially the same temperature at a melt composition only 1 wt % lower in MgO than Fo + Sp + Opx + Cpx + Melt.
We attempted to locate the isobaric invariant point in the system CMAS–K2O by exploring lower temperatures. An isobaric invariant point should exist at which sanidine joins Fo + Opx + Cpx + Sp in equilibrium with melt. Our first attempts consisted of two runs, C-1580 (1250°C) and C-1576 (1230°C), with the same initial melt composition SEM02-08 as used at 1260 and 1270°C. However, this melt composition crystallized An + Sapph + San at 1260°C, joined at 1250°C by Sp and Opx, the latter with very high Al2O3 (14 wt %). This assemblage is clearly out of equilibrium with the Fo + Opx + Cpx + Sp layers, with a narrow reaction zone of about 10–20 µm separating the two assemblages (see Fig. 2a).
With this experience, we tried again with a new initial melt composition (SEM02-10). This time equilibrium was approached throughout the charge, but Fo was reacted out, producing Sp + Opx + Cpx + An + Melt (with a trace of residual Fo) in C-1701 at 1240°C, and the isobarically invariant assemblage Sp + Opx + Cpx + An + San + Melt in C-1708 at 1230°C. A third attempt with another initial melt composition (SEM02-13) produced Sp + Opx + Cpx + San + Melt with a trace of residual Fo in C-1779 at 1250°C.
These and the higher-temperature results in the system CMAS–K2O are summarized in Fig. 6, a projection from Fo + Di onto the An–Or–Qz plane. Although we failed to locate the isobaric invariant point Fo + Sp + Opx + Cpx + San + Melt at 1·1 GPa directly, this point, which is eutectic-like on this projection, must occur at a melt composition between runs C-1585 at 1260°C and C-1779 at 1250°C, at a slightly lower temperature (i.e. 1240 ± 10°C).
|
DISCUSSION
|
|---|
Compositions of Opx and Cpx
The compositions of both Opx and Cpx in the experimental run
products vary in CaO and Al
2O
3 contents.
CaO in pyroxenes
The partitioning of CaO between Opx and Cpx is the basis of two-pyroxene geothermometry, the main means of estimating temperatures of equilibration in lherzolitic assemblages. To compare our data with previous work in the system CMAS, we have used the formulation of Nickel et al. (1985) to calculate temperatures from our experimental run products (Table 3 and Fig. 5a). The calculated temperatures are in good agreement with nominal experimental temperatures, although in detail there is a weak systematic tendency for the former to be higher at low experimental temperatures. The average difference between the calculated temperature and the nominal experimental temperature for the experiments reported here is 23° with a standard deviation of 17°. We also tested two other pyroxene geothermometers, equations (9) and (10) of Brey & Köhler (1990). Equation (9), which uses the exchange of Ca between Opx and Cpx, yields higher calculated temperatures, with an average difference of 65°C between the calculated temperatures and the experimental temperatures (Fig. 5b). Conversely, equation (10), which uses only the amount of Ca in Opx coexisting with Cpx, gives lower calculated temperatures, with an average difference of –39° (Fig. 5c). These equations were intended for use with chemically complex natural systems, but were formulated as simply as possible for ease of use. Figure 5b and c indicates that this simplicity may result in a loss of accuracy; consequently, more rigorous albeit tediously complicated two-pyroxene geothermometers could potentially give better results.
View larger version (23K):
[in this window]
[in a new window]
|
Fig. 6. Defining brackets for the isobaric invariant point Fo + Sp + Opx + Cpx + San + Melt in the system CMAS–K2O at 1·1 GPa. The melt composition of this invariant point must be located between C-1779 (on the univariant line Sp + Opx + Cpx + San + Melt) and C-1585 (on the univariant line Fo + Sp + Opx + Cpx + An + Melt). Arrows show the direction of fall in temperature. The solidus of this invariant point should be lower than 1250°C but slightly higher than 1230°C, the invariant point Sp + Opx + Cpx + San + An + Melt.
|
|
Shown in
Fig. 7 are differences between observed and calculated
temperatures using the geothermometer of Nickel
et al. (1985)
for experiments in the system CMAS at pressures from 1 atm to
3·4 GPa, from Longhi (1987)
, Walter & Presnall (1994)
,
Gudfinnsson & Presnall (1996)
, Klemme & O'Neill (2000
a)
,
and this study (CMAS ± K
2O). Similar diagrams using the
two geothermometers of Brey & Köhler (1990)
, their
equations (9) and (10), are shown in
Fig. 8, with the addition
of the CMAS–Na
2O data from Walter & Presnall (1994)
.
These plots raise a number of issues regarding potential problems
with the experimental database in the CMAS system, in addition
to the problem with the geothermometers. One consideration is
the choice of thermocouple. The Nickel
et al. (1985)
and Brey
& Köhler (1990)
geothermometers are based on experiments
by those workers using type B Pt/Rh thermocouples, as in this
study, whereas the data from Presnall and co-workers were obtained
using W/Re thermocouples. Thus the discrepancy with increasing
pressure seen in
Figs 7 and
8 may be due to an error in the
pressure dependence of the Brey–Köhler geothermometers,
or to differences in the effect of pressure on the e.m.f. of
Pt/Rh thermocouples compared with W/Re thermocouples.
View larger version (17K):
[in this window]
[in a new window]
|
Fig. 7. Difference between experimental temperatures measured using either Pt/Rh or W/Re thermocouples, and calculated temperatures from the Opx–Cpx geothermometer of Nickel et al. (1985) for experiments in the systems CMAS and CMAS–K2O. The data are plotted as a function of the experimental pressure. There is a systematic difference between experimental temperatures measured using Pt/Rh thermocouples (this study; KO00a, Klemme & O'Neill, 2000a; L87, Longhi, 1987) and those using W/Re thermocouples (WP94, Walter & Presnall, 1994; GP96, Gudfinnsson & Presnall, 1996). The latter return lower calculated temperatures. It should be noted that the datum at 1 bar using a Pt/Rh thermocouple (from Longhi, 1987), where thermocouple contamination is not likely to be an issue, is consistent with the higher-pressure data using Pt/Rh thermocouples.
|
|
A further complicating issue is the possibility of oxidation
of W/Re thermocouples at lower pressures. To guard against this,
Walter & Presnall (1994)
used a nitrogen flow at low pressures
(
1·1 GPa) for their experiments in the system CMAS–Na
2O,
but this precaution does not appear to have been used in the
earlier experiments of Presnall and co-workers in the system
CMAS [re-analysed by Walter & Presnall (1994)
]. Oxidation
of W/Re thermocouples is thought not to be a problem at higher
pressures, as collapse of the alumina tubing around the thermocouple
wires should prevent entry of air. The calculated temperatures
from the lower-pressure experiments with W/Re thermocouples
with the N
2 flow agree fairly well with the experiments using
Pt/Rh thermocouples. However, earlier low-pressure experiments
using W/Re thermocouples but without a N
2 flow (system CMAS)
have lower calculated temperatures, consistent with thermocouple
drift caused by oxidation.
Al2O3 of pyroxenes in Sp-lherzolite
The alumina contents of Opx and Cpx in the Sp-lherzolite assemblage are controlled by the reactions
 | (5) |
and
 | (6) |
Previous work has shown that these equilibria are insensitive to pressure but strongly dependent on temperature, Al2O3 increasing with rise in temperature (Obata, 1976; Fujii, 1977; Danckwerth & Newton, 1978; Herzberg, 1978; Lane & Ganguly, 1980; Gasparik, 1984; Sen, 1985). Recently Klemme & O'Neill (2000a) re-fitted existing experimental data for reaction (5) together with their new data. The alumina contents of Opx coexisting with Fo + Sp in the systems CMAS and CMAS–K2O reported here agree well with this previous work (Fig. 9b), except for the data of Gudfinnsson & Presnall (1996), which plot at slightly lower values.
The Al
2O
3 contents of Cpx show a very similar trend (
Fig. 9a),
but with a little more scatter. This extra scatter may be due
to a slight pressure effect on this equilibrium, as seen in
the experiments of Gasparik (1984)
. The partitioning of Al between
Opx and Cpx does not vary with temperature (i.e. the temperature
dependence of the Al isopleths is similar in Opx and Cpx), as
demonstrated in
Fig. 10. In experiments in the system CMAS–Na
2O,
the Cpx might be expected to have additional Al from a jadeite
component (NaAlSi
2O
6). To test this, we also plotted (
Fig. 10b)
the data for the CMAS–Na
2O system (Walter & Presnall,
1994
) with the molar amount of Al associated with Na subtracted.
Remarkably, this results in a greatly increased scatter in the
plot, indicating that our expectation of increased Al associated
with Na must be incorrect.
View larger version (23K):
[in this window]
[in a new window]
|
Fig. 10. Partitioning of Al between Cpx and Opx for a Sp-lherzolite phase assemblage in the systems CMAS, CMAS–K2O and CMAS–Na2O. The CMAS–IV data are plotted in two ways: (a) with the Al associated with Na as the Jadeite component in Cpx (NaAlSi2O6: Jd) retained; (b) with this component subtracted. Clearly the latter procedure increases the scatter. (For data sources, see Fig 9.)
|
|
Al
2O
3 in pyroxenes, particularly in Cpx, is always the most
heterogeneous component in our experiments and anomalous Al
2O
3 contents are a sign of disequilibrium. An example is the subsolidus
reversal experiment C-1789, in which the Al
2O
3 of the Cpx remains
similar to that in the starting material, although the Opx in
this run has changed its Al
2O
3 content to a lower value.
Al2O3 in pyroxenes of other phase assemblages
Reactions (5) and (6) show that, in the absence of Fo, the MgAl2SiO6 (MgTs) component in Opx and the CaAl2SiO6 (CaTs) component in Cpx increase. Therefore, the pyroxenes in the phase assemblage Sp + Opx + Cpx should have higher Al2O3 contents than those in the phase assemblage Fo + Opx + Cpx + Sp, as observed in our experiments (Fig. 11).
View larger version (31K):
[in this window]
[in a new window]
|
Fig. 11. The effect of the coexisting phase assemblage on the solubility of Al2O3 in Cpx (a) and in Opx (b) from phase assemblage to phase assemblage. Lines are fitted by eye. Some data are slightly shifted horizontally to display them more clearly.
|
|
A single governing reaction for the Al
2O
3 content in both pyroxenes
coexisting with An + Sp can be written:
 | (7) |
The absence of either Sp or An would shift this reaction to the left-hand side so that MgTs in Opx and CaTs in Cpx decrease. Again, this is observed in our experiments (Fig. 11).
As the Al2O3 contents of pyroxenes depend on the identities of coexisting phases, the Al2O3 contents are useful indicators of equilibrium between the different parts of sandwich experiments. Although this kind of disequilibrium is clearly seen in a few experiments, in most runs the same Al2O3 contents in pyroxenes were indeed observed in the ‘filling’ and the ends of the sandwich.
The effect of K2O on melting relations of Sp-lherzolite in the system CMAS–K2O
Effect of K2O on melt compositions
We find that 1 wt % K2O in the melt depresses the solidus temperature of the Sp-lherzolite phase assemblage at 1·1 GPa by about 5·8°C (Fig. 2); concurrently, this amount of K2O decreases MgO by 1·30% and CaO by 1·03%, but increases SiO2 by 1·05% and Al2O3 by 0·28% (Fig. 3).
To compare the effect of K2O on melt composition with that of Na2O, we plot our experimental data in mole percent in Fig. 12, together with the data of Walter & Presnall (1994) in the system CMAS–Na2O, at 2·0 GPa. Walter & Presnall (1994) provided a global regression of their data to 3·5 GPa that shows that the difference in pressure between 1·1 and 2·0 GPa has a negligible effect in this context. For K, an increase of 1% molar increases Si by 0·95% and Al by 0·33%, and decreases Mg by 1·46% and Ca by 0·82%. For all cations, the effect of Na is similar in direction to that of K, but much weaker: a 1% molar increase in Na only increases Si by 0·1% and Al by 0·1%, and decreases Mg by 0·74% and Ca by 0·47%.
Hence, the effect of K
2O on modifying the composition of melts
multiply saturated with Fo + Opx + Cpx + Sp in the system CMAS–K
2O
is very different from that of Na
2O in the system CMAS–Na
2O.
This is illustrated in
Fig. 13, in which the compositions of
multiply saturated melts are plotted in the basalt tetrahedron
projected from Ol (
Fig. 13a) and Di (
Fig. 13b). The data for
the system CMAS–Na
2O were taken from Walter & Presnall
(1994)
at 2·0 GPa, the higher pressure being used because
Na
2O stabilizes plagioclase at the expense of spinel at only
2 wt % Na
2O at 1·1 GPa. The difference in the behaviours
of K
2O and Na
2O can be summarized under three points, as follows.
(1) The addition of K
2O drives the melt composition towards
the Qz-normative field (defined by the join AbAnOr–Hy
in
Fig. 13b), the melt becoming Qz-normative at
2 wt % K
2O.
The addition of Na
2O drives the melt composition towards the
Ne-normative field (defined by the join AbAnOr–Ol in
Fig. 13b),
the melt becoming Ne-normative at
2 wt % Na
2O. (2) K
2O
decreases normative Di, whereas Na
2O has no effect on this normative
component. (3) As a consequence, adding K
2O eventually produces
a corundum-normative melt at
8 wt % K
2O. Na
2O-rich melts never
become corundum-normative.
Effect of K2O on the partitioning of MgO between olivine and melt
The amount of MgO in a basaltic magma increases strongly with
rise in temperature. Several workers have attempted to quantify
this effect, using the partition coefficient for MgO between
Melt and Ol (Leeman, 1978
; Ford
et al., 1983
; Gudfinnsson &
Presnall, 2001
). However, such single-element partition coefficients
are generally expected to depend on other factors including
the details of the melt composition (e.g. O'Neill & Eggins,
2002
).
Figure 14 shows that adding K
2O to CMAS produces a very
different trend from that established by the systems CMAS, CMAS–Na
2O
and CMAS–FeO (compare Gudfinnsson & Presnall, 2001
,
fig. 2). Clearly, empirical geothermometers, with a less than
rigorous thermodynamic basis, have limited reliability, and
should never be applied to compositions outside those used in
their formulation.
The MgO-partitioning ‘magmathermometer’ of Ford
et al. (1983)
has been used recently by Danyushevsky
et al.
(1996)
, Falloon
et al. (1999
, 2001
) and Falloon & Danyushevsky
(2000)
. Falloon & Danyushevsky (2000)
found that this formulation
returns progressively larger errors in the calculated liquidus
temperatures of basaltic melts as alkalis (Na
2O + K
2O) exceed
3 wt %. Our data can be used to check this phenomenon further.
The result is shown in
Fig. 15. Apparently the Ford
et al. (1983)
geothermometer cannot reproduce the experimental temperatures
accurately either in the system CMAS or in the system CMAS–K
2O
(
Fig. 15a). We also checked the effect of pressure on the geothermometer,
with results shown in
Fig. 15b. Evidently pressure affects the
calculated temperature as well.
Melt properties and phase relations
The CIPW norms of the experimentally observed melt compositions
are summarized in
Table 6. Melts with K
2O contents below
4·1
wt % K
2O are
ol-normative. The melt becomes
qz-normative at
higher K
2O. Melts with >8 wt % K
2O are
co-normative.
View this table:
[in this window]
[in a new window]
|
Table 7: Regression coefficients for phases of variable compositions in experiments with Fo + Sp + Opx + Cpx + Melt
|
|
Melt compositions observed in this study are plotted in
Fig. 16.
In the system CMAS there are two isobaric invariant points
of interest, A (Fo + Sp + Opx + Cpx + Melt) and A
* (Sp + Opx
+ Cpx + An + Melt). Points B, C and D are isobaric invariant
points in the system CMAS–K
2O, for the phase assemblages
Fo + Sp + Opx + Cpx + San + Melt, Sp + Opx + Cpx + An + San
+ Melt and Sp + Opx + An + San + Sapph + Melt, respectively.
All these invariant points are peritectic. From point A to point
B the isobarically univariant phase assemblage is Fo + Sp +
Opx + Cpx + Melt, whereas from point B to point C it is Sp +
Opx + Cpx + San + Melt. At point C, the melt composition trend
is joined by another melt composition trend emanating from point
A
*, with the isobarically univariant phase assemblage of Sp
+ Opx + Cpx + An + Melt. The isobarically univariant phase assemblage
from point C to point D is Sp + Opx + San + An + Melt. Thermal
maxima must be located between B and C, and between C and D.
The temperature change is judged by our direct temperature observation
or by the method of Presnall (1986)
. One interesting property
of the system suggested by our experiments is that the univariant
solidus curves have a relatively steep temperature drop (1320–1240°C)
from A to B or from A
* to C, but the temperature interval between
A and A
* and the corresponding univariant solidus curves A–B
and A
*–C is very small.
View larger version (28K):
[in this window]
[in a new window]
|
Fig. 16. Projections from An + Or (a) and from Fo + Di (b) showing the melt compositions in the KE experiments (wt %). CIPW norms are from Table 6. Projection (a) is inappropriate for melts with high K2O content whereas (b) is inappropriate for melts with zero K2O. In the former case, the melt compositions contain zero normative ol but high proportions of normative an and or, which do not appear in the projection; conversely, in the latter case, the melts are f | |
TOP
ABSTRACT
INTRODUCTION
and 
) vs inverse temperature for the Sp-lherzolite phase assemblage in the systems CMAS, CMAS–K2O and CMAS–Na2O. Data sources: WP94, Walter & Presnall (1994)
