| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Journal of Petrology Volume 41 Number 2 Pages 257-283 2000
© Oxford University Press 2000
Melting of Refractory Mantle at 1·5, 2 and 2·5 GPa under Anhydrous and H2O-undersaturated Conditions: Implications for the Petrogenesis of High-Ca Boninites and the Influence of Subduction Components on Mantle Melting
SCHOOL OF EARTH SCIENCES, UNIVERSITY OF TASMANIA, GPO BOX 252-79, HOBART, TAS. 7001, AUSTRALIA
Received July 6, 1998; Revised typescript accepted July 29, 1999
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONHIGH-Ca BONINITE PRIMARY MAGMA...RATIONALE OF EXPERIMENTAL STUDYEXPERIMENTAL AND ANALYTICAL...EXPERIMENTAL RESULTSDISCUSSIONREFERENCES |
|---|
Boninites are an important ‘end-member’ supra-subduction zone magmatic suite as they have the highest H2O contents and require the most refractory of mantle wedge sources. The pressure–temperature conditions of boninite origins in the mantle wedge are important to understanding subduction zone initiation and subsequent evolution. Reaction experiments at 1·5 GPa (1350–1530°C), 2 GPa (1400–1600°C) and 2·5 GPa (1450–1530°C) between a model primary high-Ca boninite magma composition and a refractory harzburgite under anhydrous and H2O-undersaturated conditions (2–3 wt % H2O in the melt) have been completed. The boninite composition was modelled on melt inclusions occurring in the most magnesian olivine phenocrysts in high-Ca boninites from the Northern Tongan forearc and the Upper Pillow Lavas of the Troodos ophiolite. Direct melting experiments on a model refractory lherzolite and a harzburgite composition at 1·5 GPa under anhydrous conditions (1400–1600°C) have also been completed. Experiments establish a P, T ‘melting grid’ for refractory harzburgite at 1·5, 2 and 2·5 GPa and in the presence of 2–3 wt % H2O. The effect of 2–3 wt % dissolved H2O produces a liquidus depression in primary boninite of
112 ± 19°C at a given temperature. The H2O-bearing melts, recalculated to 100 wt % anhydrous, are 2–6 wt % higher in MgO, 1–2 wt % higher in SiO2 and 1–1·5 wt % lower in FeO, compared with nominally anhydrous melts at the same P and T. These differences are consistent with a change in the melting reaction, resulting in a higher contribution of orthopyroxene to the melt phase, compared with anhydrous conditions. We conclude that high-Ca boninite petrogenesis requires temperatures as high as 1480°C at depths of 45 km in the mantle wedge; these are constraints for any proposed model of intra-oceanic subduction zones. A comparison of the results from the boninite–harzburgite reaction experiments with the direct melting experiments on refractory lherzolite and harzburgite indicates that the influence of subduction components (included in the composition of the added model boninite) is to cause high-pressure melting cotectics to move towards the olivine apex (i.e. to relatively higher pressures) of the molecular normative projection from diopside onto the base of the ‘basalt tetrahedron’ [Jd + CaTs + Lc–Qz–Ol] compared with anhydrous melting of normal mantle in the absence of a subduction component. The subduction component involved in high-Ca boninite petrogenesis in addition to H2O has relatively high Al2O3 and Na2O contents. The experimental data from this and other studies empirically quantify the absolute effect of dissolved H2O (0·2–21 wt %) on the liquidus depression of olivine-saturated basaltic melts with <3 wt % total alkalis as follows:
 |
9 relative percent. KEY WORDS: high-Ca boninites; mantle melting; H2O-undersaturated melting; olivine liquidus depression; anhydrous melting
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONHIGH-Ca BONINITE PRIMARY MAGMA...RATIONALE OF EXPERIMENTAL STUDYEXPERIMENTAL AND ANALYTICAL...EXPERIMENTAL RESULTSDISCUSSIONREFERENCES |
|---|
Magma genesis at intra-oceanic subduction zones is fundamentally different from that occurring at mid-ocean ridges, as a result of the presence of H2O and other volatiles derived from the dehydrating subducted slab, the possible involvement of melts derived from the slab, and the presence of refractory mantle sources in the mantle wedge above the subducting slab (e.g. Gill, 1981
; Woodhead et al., 1993; Tatsumi & Eggins, 1995; Turner et al., 1997; Green & Falloon, 1998).
H2O not only acts as an incompatible element, and is therefore a ‘flux’ for melting, but can also form a supercritical fluid capable of fractionating and variably enriching subduction zone mantle sources in other incompatible elements. H2O affects the phase relationships of fertile mantle peridotite in terms of both lowering the solidus in pressure–temperature (P–T) space and changing the nature of melting reactions (Green, 1973b, 1976; Gaetani & Grove, 1998; Green & Falloon, 1998). Subduction zone magmas in general have geochemical features consistent with a significant role of H2O in magma genesis, i.e. higher degrees of mantle melting and enrichment in mobile LILE (large-ion lithophile elements) compared with MORB (mid-ocean ridge basalts) (e.g. Gill, 1981).
Boninites are a rare subduction-related magma type with higher SiO2 and H2O, and lower TiO2, compared with island arc tholeiite suites. They lack plagioclase in rocks more mafic than andesite, and contain very magnesian olivine (Ol) phenocrysts (up to Fo94, Crawford et al., 1989). It is generally accepted that the petrogenesis of boninite requires T significantly higher (>100–300°C) than can be reasonably expected (<800°C at <50 km) in a mantle wedge on the basis of geophysical models (Hsui & Toksov, 1979). Also, boninite petrogenesis is believed to involve melting of refractory mantle sources (harzburgitic consequent to prior extraction of MORB) through addition of an incompatible element enriched phase dominated by H2O. Boninites are in fact the most H2O-rich magma type known from intra-oceanic subduction zones (Danyushevsky et al., 1993; Sobolev & Chaussidon, 1996). The so-called ‘boninite paradox’, is that the highest T magmatic suite in intra-oceanic arcs also has the highest H2O content (Sobolev & Chaussidon, 1996). Boninite suites are therefore an ‘end-member’ subduction-related suite as they (1) are the most H2O rich, and (2) require the most refractory sources. A detailed understanding of boninite petrogenesis is important as a constraint on the geodynamics of subduction-related magmatic processes.
However, at present the T of boninite formation is disputed. The controversy has arisen because of differing opinions on the nature of boninite parental or primary magma compositions, especially the MgO content of the boninite magmas, which is reflected in the composition of Ol phenocrysts. As the MgO content of mafic magma is a fundamental control on its liquidus T (e.g. Ford et al., 1983), it is vital that the MgO content of parental or primary boninite magmas is correctly established. Equally important is the effect H2O has on lowering the liquidus T of MgO-rich boninite magmas. Establishment of formation T of primary boninite is a vital first step in constructing reasonable models of origin, i.e. do boninites require deep mantle upwelling, or alternatively can they be produced by contact melting of metasomatized cold lithospheric mantle at shallow depths in the mantle wedge?
At present there is a view that boninite magmas require T ranging from 1150 to 1350°C, based on several experimental studies on boninite compositions ranging from 7 to 16 wt % MgO with experimental H2O conditions ranging from anhydrous to water saturated (Tatsumi, 1981, 1982; Tatsumi & Maruyama, 1989; Umino & Kushiro, 1989; Van der Laan et al., 1989).
On the other hand, on the basis of detailed studies of melt inclusions hosted by magnesian Ol phenocrysts in high-Ca boninite (HCB) suites, Sobolev et al. (1993) and Sobolev & Danyushevsky (1994) concluded that HCB primary magmas have MgO contents of 19–24 wt %, H2O contents of 1–2 wt %, and T of formation of between 1450 and 1550°C. Such high Ts are not accounted for by current geophysical models of mantle wedge thermal structure and hence, if these high Ts are correct, it has very important implications for the geodynamics of intra-oceanic subduction zones. For example, Danyushevsky et al. (1995) proposed that such high Ts require the involvement of mantle plumes.
The conclusions of Sobolev et al. (1993) and Sobolev & Danyushevsky (1994) were based on: (1) comparing the compositions of the established primary or parental HCB magma compositions with those of experimentally determined, anhydrous partial melts of fertile mantle compositions with an estimate of the effect of H2O on the partial melt compositions; (2) an estimate, based on available experimental data, on the effect of H2O in lowering the liquidus T of the very magnesian HCB melt compositions.
Here we attempt to refine the T estimates of HCB petrogenesis by establishing a ‘melting grid’ for refractory peridotite (harzburgite) affected by a subduction-related component (SC) under both anhydrous and H2O-undersaturated conditions, and by directly estimating the liquidus depression caused by the presence of H2O on such highly magnesian primary magma compositions.
Our experimental results confirm that remarkably high Ts (1480°C) at relatively shallow depths (45 km) in the mantle wedge are required for HCB petrogenesis.
|
HIGH-Ca BONINITE PRIMARY MAGMA COMPOSITION |
|---|
|
TOPABSTRACTINTRODUCTIONHIGH-Ca BONINITE PRIMARY MAGMA...RATIONALE OF EXPERIMENTAL STUDYEXPERIMENTAL AND ANALYTICAL...EXPERIMENTAL RESULTSDISCUSSIONREFERENCES |
|---|
Sobolev & Danyushevsky (1994)
presented a detailed mineralogical, geochemical and experimental study of an HCB suite from the northern termination of the Tonga Trench (Sharaskin et al., 1983; Falloon et al., 1987, 1989; Falloon & Crawford, 1991; Danyushevsky et al., 1995). Primary melt compositions for this suite were established on the basis of an experimental study of melt inclusions in Ol phenocrysts, and numerical modelling of the reverse of fractional crystallization. The approach of Sobolev & Danyushevsky (1994) was to search for the most magnesian Ol phenocrysts present in the boninites, by crushing the samples and mounting 200–300 Ol grains for microprobe analysis. The most magnesian Ol were then examined for suitable melt inclusions for optically controlled homogenization experiments. Successfully homogenized melt inclusions were analysed by electron microprobe. The primary melts established by this technique have high MgO contents ranging from 19 to 24 wt %. |
RATIONALE OF EXPERIMENTAL STUDY |
|---|
|
TOPABSTRACTINTRODUCTIONHIGH-Ca BONINITE PRIMARY MAGMA...RATIONALE OF EXPERIMENTAL STUDYEXPERIMENTAL AND ANALYTICAL...EXPERIMENTAL RESULTSDISCUSSIONREFERENCES |
|---|
Sobolev et al. (1993)
and Sobolev & Danyushevsky (1994) estimated the P and T of primary HCB magma genesis (2–2·5 GPa, 1450–1550°C; 2–3 GPa, 1380–1550°C, respectively) on the basis of:
- comparison with anhydrous partial melting experiments on fertile peridotite compositions (Takahashi & Kushiro, 1983; Falloon & Green, 1987, 1988) using the normative projection of Walker et al. (1979);
- the assumption that the mantle melting cotectics, defined by the anhydrous experimental data, were linear in the Walker et al. (1979) projection and would continue to be so towards more refractory mantle compositions;
- an estimate of the effect of 2 wt % H2O additions based on the experimental studies of Green (1973b, 1976) and Kushiro (1990), which were also performed on fertile peridotite compositions.
The rationale of our experimental study is to determine more accurately the P and T of primary HCB magma genesis by determining a ‘melting grid’ appropriate for refractory mantle in the presence of an SC, and to directly determine the effect of 2 wt % H2O on the liquidus T of primary HCB magma compositions.
As previous petrological studies on HCB petrogenesis suggest that the residual mantle at the P and T of magma generation is clinopyroxene (Cpx) free (Crawford et al., 1989; Falloon et al., 1989; Sobolev & Danyushevsky, 1994), we have performed reaction experiments between a model primary HCB and a model refractory harzburgite. The rationale of the experiments is to allow the boninite to equilibrate with harzburgite at various P and T values, thus defining the positions of Ol + orthopyroxene (Opx) + liquid (L) cotectics for refractory peridotite affected by an SC. The use of reaction experiments ensures that large areas of glass, in the case of nominally anhydrous experiments, or glass and quench pyroxenes in the case of hydrous experiments, are available after the experiment, which can be analysed free from the effects of metastable quench crystallization on primary crystal phases (Fig. 1a), and can also be analysed by Fourier transform IR (FTIR) spectroscopy for H2O contents.
|
|
We should be able to demonstrate that at some P and T the primary HCB identified by melt inclusion studies is saturated in mantle minerals (specifically Ol and Opx), and the composition of the saturated melt is close to that determined by the melt inclusion technique. Two significant technical problems that could affect the outcome of the melt inclusion studies are: (1) overheating of the inclusions during homogenization causing estimated primary magma compositions to have incorrectly high MgO contents; (2) poor quenching, especially in H2O-bearing MgO-rich melt compositions. An additional problem is the limited availability of suitable sized melt inclusions in Ol phenocrysts of the most magnesian compositions.
Our boninite–harzburgite reaction experiments allow us to determine the effect of an SC on melting mantle. Petrological studies of subduction zone magmatic suites clearly document the influence of SCs on the geochemistry of erupted magmas (e.g. Turner et al., 1997). Although geochemical tracers can be used to identify the various potential sources (e.g. sediment) of the SC (Turner et al., 1997), it is difficult to constrain the composition of the SC independently. The simple addition of H2O as an analogue for the SC in peridotite melting experiments (Green, 1976), although providing data on the effects of H2O as a pure component on mantle melting, may not be directly applicable to natural subduction-related magmatic suites. An advantage of boninite–harzburgite reaction experiments is that we are adding a melt component closely matching a natural magma type, which is itself the result of subduction zone melting. Therefore in our experiments the actual SC involved in HCB genesis is present in the composition of the added boninite itself, even though we cannot independently isolate it. To determine the effect of this SC on mantle melting we also performed melting experiments on a refractory lherzolite and harzburgite under anhydrous conditions with no added SC to compare with the boninite–harzburgite reaction experiments.
|
EXPERIMENTAL AND ANALYTICAL TECHNIQUES |
|---|
|
TOPABSTRACTINTRODUCTIONHIGH-Ca BONINITE PRIMARY MAGMA...RATIONALE OF EXPERIMENTAL STUDYEXPERIMENTAL AND ANALYTICAL...EXPERIMENTAL RESULTSDISCUSSIONREFERENCES |
|---|
Experimental techniques
Starting compositions
All starting compositions used in the experimental study are presented in Table 1. The harzburgite composition (HZ, Table 1) was modelled on the composition of a natural harzburgite from the Troodos ophiolite (sample 2PKI-38, Table 1, Sobolev et al., 1993
). The composition of 2PKI-38 was recalculated to an mg-number of 94 and 40 wt % Ol of Fo93·5 was subtracted. The starting composition was prepared from a mixture of analytical grade oxides and carbonates (Ca, Na, K), ground under acetone in an agate mortar. This mixture was pelletized and sintered overnight (16–20 h) at 950°C. An appropriate amount of synthetic fayalite was then added to the sintered mix and the mixture was again ground under acetone, before storage in glass vials in an oven at 110°C.
|
The HCB composition (BON, Table 1), modelled on the HCB compositions established by the studies of Sobolev et al. (1993) and Sobolev & Danyushevsky (1994), was prepared by addition of appropriate amounts of analytical grade oxides (SiO2, Al2O3 and MgO) and synthetic fayalite to a hydrous natural glass, also from the Troodos ophiolite (sample A-46, Table 1, Danyushevsky et al., 1993). This mixture was ground under acetone in an agate mortar before storage in glass vials in an oven at 110°C. The resulting starting composition (BON, Table 1) has H2O content of 2·4 wt %, identical to that determined for Tongan HCB primary magmas by melt inclusion studies (Sobolev & Danyushevsky, 1994). The advantage of using a natural glass as a source for H2O is that it eliminates any uncertainty in the exact amount of starting H2O and therefore allows us to monitor changes in H2O contents during our experiments.
Starting compositions T-4347 and TQ-40 were prepared in a similar manner to HZ above. Composition T-4347 is the glass composition from run T-4347 (Table 2) and was used for reversal experiments at 1·5 GPa (see text below). Composition TQ-40 is a refractory lherzolite (minus 40 wt % Fo91·9) composition modelled on the natural Tinaquillo Lherzolite (Jaques & Green, 1980). The rationale of using Ol-depleted compositions in experimental studies has been presented by Falloon & Green (1987).
|
For anhydrous experiments, the BON starting composition was dehydrated by firing at 1000°C in an Ar atmosphere for 24 h.
Run assemblies and temperature control
All experiments were performed using standard piston-cylinder techniques in the High Pressure Laboratory housed in the School of Earth Sciences, University of Tasmania (UT). All experiments used talc–Pyrex assemblies (except runs 4–11, Table 2, which used NaCl–Pyrex assemblies) with graphite heaters and a W97Re3/W75Re25 (W/Re) thermocouple (calibrated against the melting point of Au and the e.m.f. of a Type S, Pt/Pt90Rh10 (Pt) thermocouple, at atmospheric P in an Ar atmosphere). Temperatures were controlled to within ±1°C of the set point using a Eurotherm type 818 controller. No P correction was applied to the thermocouple calibration. For the majority of experiments starting compositions were loaded into inner graphite capsules and then sealed by welding into large capacity (3·5 mm o.d.) platinum capsules. Boron nitride and sintered alumina components were used as internal spacers and surrounds for experiments, with experiments using graphite sealed in platinum capsules. Anhydrous experiments using graphite capsules only used fired pyrophyllite and alumina spacers, and mullite and alumina surrounds (runs 4–11, Table 2). The thermocouple entered the assembly via a pure alumina sheath and was protected from the graphite or platinum capsule by a 1 mm alumina disc. All experimental components and starting materials were stored in a oven at 110°C. Experiments were performed using the hot piston-out technique (Johannes et al., 1971). Pressures are accurate to within ±0·1 GPa.
Analytical techniques
FTIR spectroscopy
Infrared spectroscopy was used to monitor any possible significant changes in H2O contents during the experiments. Measurements were performed using a Bruker IFS 66 spectrometer with attached optical microscope (all reflecting optics) and Bruker Opus/IR reduction software, housed in the Central Science Laboratory (CSL), UT. Run products were doubly polished (30–70 µm thick, using Superglue for bonding to a standard thin-section glass slide, during polishing). Diameters of analysed areas were usually 60–90 µm. During each analysis, 100 scans were collected with the resolution of four wavenumbers between 4000 cm-1 and 2400 cm-1. H2O contents were estimated using the main OH-stretching peak at 3500 cm-1, following the calibration of Danyushevsky et al. (1993).
Unfortunately, the above sample preparation technique results in poor analytical accuracy (mainly because of difficulties in estimating sample thickness and the necessity to subtract the glue spectrum). Additional problems during analyses of H2O-bearing runs result from: (1) the unknown density of the quenched material; (2) the unknown, and variable proportions of glass and quenched crystals (Fig. 1); (3) changes in the 3500 cm-1 peak shape caused by the presence of quenched crystals in the analysed area.
The analysed H2O contents in the hydrous experiments presented in this paper were usually within ±0·5 wt % from the estimates made on the basis of mass balance and initial H2O contents. As a result of the above technical problems, we consider the mass balance estimates to be more accurate, and we used the IR data only to ensure that no significant changes in H2O content occurred during the experiments.
In the case of the nominally anhydrous experiments, H2O analyses using FTIR spectroscopy were more precise because of the absence of quench crystallization (Fig. 1a), and we therefore consider the analysed H2O contents for nominally anhydrous experiments to be reliable.
As our experiments were performed in graphite capsules, we also need to consider the possibility of dissolved CO2 in our experimental glasses. Holloway et al. (1992) demonstrated that the dissolved CO2 concentration in experimental basaltic melts, run in graphite capsules, depends on three important factors: (1) the concentrations of ferric and ferrous iron in the starting material; (2) the amount of any additional CO2 source added to the starting material; (3) the amount of hydrogen infused into the sample during the experiment.
The first factor above is of particular relevance to our study, as part of the BON starting material is a natural glass (A-46, Table 1) and consequently there will be an initial, but undetermined, amount of Fe2O3 in our BON starting composition. CO2 is produced via reaction between the graphite capsule and Fe2O3 in the melt according to the reaction
 |
If we assume that reaction (1) proceeds to completion in our experiments, then it is possible for us to calculate the maximum amount of CO2 we can expect to analyse by FTIR spectroscopy. We have performed this calculation for runs T-3464 and T-3462 (Table 2). If the natural glass A-46 equilibrated at oxygen fugacities equivalent to the NNO (nickel–nickel oxide) oxygen buffer then CO2 contents should vary between 0·15 (T-3464) and 0·26 (T-3462) wt %. If the natural glass A-46 equilibrated at oxygen fugacities equivalent to the QFM (quartz–fayalite–magnetite) oxygen buffer then CO2 contents should vary from 0·10 (T-3462) to 0·082 (T-3464) wt %. For the purposes of analysing the quenched melt phase in runs T-3464 and T-3462 for dissolved CO2 by FTIR spectroscopy, the run products were unmounted from their glass slides by dissolving the Superglue in acetone. The very small thickness of the samples resulted in relatively high detection limits for CO2 components in the melt (0·07 wt %). However, no CO2 was detected in either run above the detection limit. This suggests that no petrologically significant amounts of CO2 was present in our run products.
Another potential problem of significance for our experiments is hydrogen diffusion out of our experimental capsules as a result of reaction of H2O and the graphite capsule. This has the potential to produce significant amounts of CO2 contents in our experimental glasses via the reaction
 |
For example, if we had 0·5 wt % H2O involved in reaction (2), then we would expect 0·61 wt % CO2 in our experimental glasses. As we have checked for CO2 using FTIR spectroscopy with very good detection limits (see above), we consider that reaction (2) has not produced any petrologically significant amounts of CO2 in our experimental glasses. The effectiveness of reaction (2) in producing CO2 in experimental glasses will depend on the magnitude of the chemical potential gradient for H2 between the capsule and the furnace assembly. For example, Green (1973a) and Brey & Green (1975) performed a limited number of H2O-saturated experiments in the presence of graphite using an Ol melilitite and basanite composition, respectively. These experiments did not come to equilibrium and significant amounts of CO2 were found to be dissolved in the melt (D. H. Green, personal communication, 1999). This result can be simply explained by the presence of a large amount of added water (>20 wt %) creating a relatively high chemical potential for H2 inside the capsule relative to the furnace assembly. In our experiments, we do not have fluid-saturated conditions, having only 1–2 wt % H2O present, and therefore the chemical potential gradient for H2 between the capsule and the furnace assembly appears to be insignificant, as demonstrated by measured CO2 contents below or at detection limits in runs T-3464 and T-3462 (see above).
Electron microprobe microanalysis
Compositions of glasses were analysed using a Cameca SX50 electron microprobe, housed in the CSL, UT, at 15 kV and 20 
A, using international standard USNM 111240/2 (basaltic glass) from Jarosewich et al. (1980). Counting times for all elements were 10 s for the peak and 5 s for the background on both sides of the peak. Glasses were analysed in scanning mode with area scans varying from 10 to 50 µm2. Ol and Opx analyses were obtained using a beam size of 1–2 µm and international standards USNM 122142 (augite) and USNM 111312/444 (Ol) (Jarosewich et al., 1980). The mineral phases (except Ol) in run T-4302 (Table 2) were obtained by energy dispersive microanalysis using a Cameca MICROBEAM microprobe housed in the Research School of Earth Sciences, The Australian National University (operating conditions 15 kV, 5 A).
As glasses in our H2O-bearing experiments quenched to a mixture of glass and quench pyroxenes (which grew directly out of the glass, Fig. 1b) it was necessary to systematically investigate the homogeneity of broad beam area scans across the entire melt layer. We systematically performed area scans in 3–4 different areas, and within each area we analysed progressively smaller areas (from 50 µm x 50 µm to 10 µm x 10 µm). In the majority of cases, we found that major elements remained constant, except for Na2O loss on smaller area scans. These results combined with good mass balance (Table 3) indicated that the glass composition before the quenching of our H2O-bearing runs could be determined with confidence.
However, in a few of our high-P H2O-bearing runs (T-3513, T-3519, Table 2) analyses of quench melt did not result in good mass balance, and therefore only Ol and Opx analyses are presented for these experiments in Table 3.
|
EXPERIMENTAL RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONHIGH-Ca BONINITE PRIMARY MAGMA...RATIONALE OF EXPERIMENTAL STUDYEXPERIMENTAL AND ANALYTICAL...EXPERIMENTAL RESULTSDISCUSSIONREFERENCES |
|---|
Attainment of equilibrium
Several lines of evidence can be used to evaluate the approach to equilibrium of our experiments. These are as follows:
- the maintenance of constant sample bulk composition is essential for equilibration, and has been demonstrated by materials balance (Table 3). The average square of the sum of the residuals for the 31 experiments reported here is 0·2 ± 0·2 (range 0·0009–0·6636), which compares favourably with other peridotite melting studies (e.g. Kinzler & Grove, 1992, 0·01–3·58; Falloon et al., 1997, 0·004–0·654; Kinzler, 1997, 0·01–1·15; Gaetani & Grove, 1998, 0·02–0·13).
- The achievement of consistent mineral–melt exchange equilibrium indicates a close approach to equilibrium. The average KD Fe–Mg values for Ol (0·34 ± 0·01) and Opx (0·31 ± 0·01) for our short runs (0·2–4 h) are all within 1
of our longer (10–69 h) experiments (Ol, 0·32 ± 0·02; Opx 0·31 ± 0·02) and compare favourably with average values reported by other workers (e.g. Gaetani & Grove, 1998, 0·34 ± 0·01 and 0·32 ± 0·02; Kinzler, 1997, 0·33 ± 0·02 and 0·33 ± 0·04; Robinson et al., 1998, 0·32 ± 0·02 and 0·32 ± 0·02; for Ol and Opx, respectively).
- The phase compositions are homogeneous. Unlike natural mineral mix starting materials, synthetic starting materials made out of sintered oxides react rapidly to produce homogeneous run products (Falloon et al., 1999b). This is demonstrated by the very low standard deviations on averages of both core and rims of Ol and Opx analyses presented in Table 3 (in general, each average Ol and Opx composition present in Table 3 represents equal numbers of core and rim analyses). All standard deviations for Si, Fe, Mg (Ol) and Si, Fe, Mg, Al, Ca (Opx) are within one decimal place (Table 3).
- The internal consistency of our experimental data also indicates a close approach to equilibrium. Ol, Opx and coexisting melt compositions in our experiments all systematically change with P and T. For example, the mg-number of both Ol and Opx increases with T, and the Al2O3 and CaO contents of Opx decrease with increasing T. The ratio of the initial mass of added liquid (MLi) to the final mass of liquid (MLf) shows an internally consistent and systematic change with T at a given P (Table 2). At T below the liquidus of the BON composition the mass of liquid decreases (MLi/MLf > 1), indicating that crystallization and re-equilibration of the bulk composition is occurring. At T above the liquidus of the BON composition, the mass of liquid increases (MLi/MLf < 1), indicating that melting and re-equilibration of the bulk composition is occurring.
- We have also performed time series experiments at 1·5 GPa at 1450°C demonstrating that equilibrium is achieved within 2 h (Table 2). Runs T-3385, T-3387 and T-3488 show systematic changes in melt and mineral compositions, and MLi/MLf with time (Tables 2 and 3). Most importantly, the experiments show that within 2 h the Ol KD Fe–Mg decreases from 0·38 to an equilibrium value of 0·34.
In summary, we believe the experimental data presented here closely approximate equilibrium phase assemblages.
In Fig. 1a–c we present photographs of our run products to demonstrate the difference in quenching between ‘anhydrous’ (Fig. 1a) and hydrous experiments (Fig. 1b and c), and to illustrate that our hydrous runs contain a significant amount of disseminated graphite throughout the quenched melt layer (Fig. 1b and c). The layer of graphite was apparent only after the experimental run products had been prepared as double-sided polished thin-sections for FTIR analysis (compare Fig. 1b with Fig. 1c). The broad bands of disseminated graphite appear to correlate with electron microprobe area scans that showed anomalously low Na2O contents. Although we do not have an explanation for the observed phenomenon, the graphite is unlikely to have quenched from the melt, as we do not observe any CO32- peaks in our FTIR spectrum. We therefore believe that the graphite is mechanically derived from the enclosing graphite capsule, the process being linked to, or enhanced by, the presence of H2O.
‘Anhydrous’ peridotite reaction experiments
Melting of refractory lherzolite and harzburgite at 1·5 GPa
The results of our melting experiments at 1·5 GPa on the refractory compositions TQ-40 and HZ are presented in Fig. 2a and b, and compared with melting cotectics for more fertile lherzolite compositions MPY (Robinson et al., 1998) and MM-3 (Falloon et al., 1999b). Direct melting experiments on TQ-40 at 1400, 1450 and 1500°C (runs 9–11, Table 2) resulted in well-equilibrated run products in which there was a significant melt fraction (F = 25–50 wt %, Table 3). It was possible to obtain consistent glass analyses which appear to be free from the effects of quench modification, as a result of the presence of relatively large glass pools. We were able to obtain very good mass balances with very low residual sums (0·0009–0·1132, Table 3). Direct melting experiments on HZ at 1550 and 1600°C (runs 4 and 5, Table 2) resulted in well-equilibrated run products in which there was a low F (6–12 wt %, Table 3). Although it was only possible to analyse glass in small pools, selected area scans gave good mass balance and equilibrium KD Fe–Mg for Ol and Opx (Table 3). To confirm these compositions as equilibrium melts, reversal experiments were performed by reacting the composition of the melt phase in run T-4347 with HZ at 1550 and 1600°C (runs 7 and 8, Table 2). Both runs T-4349 and T-4351 resulted in well-equilibrated run products in which there was a significant F (23–31 wt %, Table 3). It was possible to obtain consistent glass analyses that appear to be free from the effects of quench modification, as a result of the presence of relatively large glass pools. We were able to obtain very good mass balances with very low residual sums (0·02–0·05, Table 3). The glasses in the reversal experiments are very close in composition to those obtained from the direct melting experiments on the HZ composition at 1550 and 1600°C (Fig. 2a and b) and we therefore believe that they represent equilibrium melts of the HZ composition at 1·5 GPa under ‘nominally’ anhydrous conditions.
|
, glass compositions in direct melting experiments on HZ; 
, glass compositions in reversal experiments of HZ melt compositions; 
, glass in direct melting experiments on TQ-40 in equilibrium with Ol + Opx; 
, glass composition in direct melting experiment on TQ-40 in equilibrium with Ol + Opx + Cpx; half-filled diamond in (b) is the composition of the eutectic melt in the system Fo–En–Qz from the study of Taylor (1973)Although we have not analysed our dry melting experiments on TQ-40 and HZ for H2O by FTIR spectroscopy, we believe our experiments have negligible H2O contents for the following reasons: (1) we have used exactly the same experimental techniques as in the study of Falloon et al. (1999b
), who determined that H2O contents were between 0·07 and 0·17 wt % for the experimental glasses chosen for FTIR analysis; (2) calculated anhydrous Ol liquidus T for our experiments using the Ol geothermometer of Ford et al. (1983) are all with 15°C of nominal experimental run T (see Fig. 7b, below). This suggests that our H2O contents are <0·1 wt % (see discussion below).
|
(c), high-P experiments using the Pt thermocouple; continuous line is the 1:1 line and dotted lines represent ±15°C from the 1:1 line. Data sources: Green et al. (1979)Assuming a linear relationship between F and T for the HZ composition at 1·5 GPa, the solidus (
0% F) lies between 1490 and 1495°C, and at 1510°C the HZ composition has 2% F. In run T-4350 (Table 2) we attempted to determine the composition of a near-solidus melt for the HZ composition at 1510°C by a reaction experiment using the T-4347 glass composition and HZ. Run T-4350 resulted in a well-equilibrated run product with very small melt pools. Glass area scan analyses on these pools varied systematically in major elements because of differing proportions of quench crystals and glass. Using the Ol geothermometer of Ford et al. (1983) we calculated an anhydrous Ol liquidus T for each individual glass scan. Calculated Ol liquidus T varied from 1445 to 1565°C. The composition of the melt fraction in run T-4350 was then calculated by linear regression, in similar manner to Falloon & Green (1988), using the calculated Ol liquidus T as the independent variable. The calculated composition of the melt in run T-4350 produced a very good mass balance with a very low residual sum (0·08, Table 3). The position of the calculated melt composition for run T-4350 is consistent with the higher-T melt compositions of HZ in Fig. 2a and b. Using the calculated glass composition of run T-4350 and the coexisting residual phases, mass balance indicates that the glass in run T-4350 represents 2·2% F (residual sum of squares is 0·2104). We therefore believe that the calculated glass composition in run T-4350 is a close approximation to an equilibrium near-solidus melt of the HZ composition at 1·5 GPa. In Fig. 2a and b the position of equilibrium melting cotectics for the HZ is compared with those for fertile to refractory lherzolite compositions within the molecular normative basalt tetrahedron. The model lherzolites MPY, MM-3 and TQ-40 vary systematically in their CaO/Al2O3 (0·776, 0·897 and 0·929, respectively) and CaO/Na2O (7·22, 11·52 and 16·56, respectively) values, and therefore the three compositions can be used to examine the changes in phase relationships for fertile mantle undergoing a process of isobaric dynamic melting. Higher CaO/Na2O values of a composition are reflected in higher Qz/[Jd + CaTs + Lc] values in Fig. 2b, whereas higher CaO/Al2O3 values are reflected in higher Di/[Jd + CaTs + Lc] values in Fig. 2a. As can be seen from Fig. 2b, the mantle melting cotectics for the three lherzolite compositions are almost coincident. However, the following differences between the melting cotectics of the three lherzolite compositions can be seen in an examination of Fig. 2a and b:
- as the position of near-solidus initial melts, in the basalt tetrahedron (Fig. 2a and b), is dependent on the CaO/Na2O value of the lherzolite composition (Falloon et al., 1997b), there are significant differences between the initial melts of the three lherzolite compositions. Initial melts from the most fertile composition (MPY) are strongly nepheline (Ne) normative (Fig. 2b), have relatively low normative diopside (Di) contents in Fig. 2a, and are relatively high in SiO2 (52·7 wt %) and Na2O (7·27 wt %) (Robinson et al., 1998). Initial melts from the slightly more refractory MM-3 composition are still Ne normative (Fig. 2b), have higher normative Di contents in Fig. 2a, and have intermediate SiO2 (49·21 wt %) and Na2O (3·81 wt %) contents (Falloon et al., 1999b). Initial melts from the refractory TQ-40 composition are inferred to have lower SiO2 and Na2O contents, to be Ol and hypersthene (Hy) normative in Fig. 2b, and to have the highest normative Di content in Fig. 2a.
- The point at which Cpx is eliminated from the residue, as seen in Fig. 2b, moves progressively towards the Ol–Qz boundary of the tetrahedron, as lherzolite sources become more refractory. In Fig. 2a the Cpx-out point moves to higher normative Di as lherzolite sources become refractory and increase in CaO/Al2O3 value.
- The point at which Opx is eliminated from the residue also moves progressively towards the Ol–Qz boundary of the tetrahedron in Fig. 2b, as lherzolite sources become more refractory. In Fig. 2b, it can be seen that the Ol + Opx + L cotectics for the three lherzolite compositions are almost coincident; however, in Fig. 2a, these cotectics are clearly separate, because of the differing CaO/Al2O3 values of the lherzolite compositions. The more refractory compositions define Ol + Opx + L cotectics at progressively higher CaO/Al2O3 values.
Compared with the mantle melting cotectics for the lherzolite compositions, the Ol + Opx + L cotectic for the HZ composition plots in a significantly different position in the basalt tetrahedron, as can be seen from Fig. 2b. The progressive melting of sources more refractory than TQ-40 may have been expected to produce liquids in equilibrium with Ol + Opx lying on an extension of the Ol + Opx cotectics for lherzolite compositions towards the simple system eutectic forsterite (Fo) + enstatite (En) + L at 1·5 GPa (dotted curve, labelled 1 in Fig. 2b). However, as can be seen from Fig. 2b, melts from the HZ composition are displaced towards the Qz apex of the basalt tetrahedron and are Qz normative at T <1600°C. As well as this displacement, the orientation of the Ol + Opx + L cotectic within the basalt tetrahedron, as seen in Fig. 2b, is significantly different compared with the Ol + Opx + L cotectics for lherzolite sources. These differences in normative composition are also reflected in significantly different major element compositions as can be seen in Fig. 3. Melt compositions from HZ have significantly higher SiO2, lower Al2O3 and slightly lower CaO at a given MgO content compared with melts from lherzolite sources (Fig. 3). Therefore melt compositions from lherzolite sources cannot be used to infer the compositions of melts from refractory harzburgite compositions, because, as can be seen in Figs 2 and 3, there is a discontinuity in composition between melts from lherzolite and harzburgite sources. Liquids in equilibrium with Ol + Opx from more refractory sources than HZ are inferred to lie on an extension of the Ol + Opx + L cotectic for HZ towards the Fo + En + L eutectic at 1·5 GPa (dotted curve labelled 2 in Fig. 2b).
|
In Fig. 2a, melt compositions from HZ have high CaO/Al2O3 values and achieve lower normative Di contents than melts in equilibrium with Ol + Opx from lherzolite compositions.
Boninite–harzburgite reaction experiments
We performed two anhydrous reaction experiments at 1·5 GPa and five experiments at 2 GPa where the resulting run product was a melt in equilibrium with a refractory harzburgite residue (Tables 2 and 3). Four of these nominally ‘anhydrous’ experiments were analysed by FTIR spectroscopy for H2O contents, which vary from 0·2 to 0·7 wt % (Table 3). Probably some H2O remained in the starting composition BON as a result of incomplete devolatilization during firing at 1000°C (see experimental techniques section).
In Fig. 4a and b we compare the results of our reaction experiments with the mantle melting cotectics for the refractory lherzolite composition TQ-40 at 1·5 and 2 GPa and for HZ at 1·5 GPa. The glass compositions in equilibrium with Ol + Opx from our reaction experiments are consistent with the position of the anhydrous Ol + Opx ± Cpx cotectics determined for TQ-40 and HZ in the following two respects:
- as for the anhydrous cotectics for TQ-40 at 1·5 and 2 GPa, the 1·5 and 2 GPa Ol + Opx + L cotectics defined by the reaction experiments also show a consistent shift in normative Ol with P in the projection from Di (Fig. 4b);
- the Ol + Opx + L cotectics defined by the reaction experiments maintain the same orientation to those of HZ in Fig. 4a and b but are displaced towards the Jd + CaTs + Lc apex of the normative tetrahedron.
|
As can be seen from both Figs 2b and 4b, glass compositions from progressive melting of fertile mantle define smooth approximately curvilinear trends in the normative projection from Di, and glass compositions from the melting of refractory mantle (HZ and boninite reaction experiments) do not lie on the extrapolation of these trends towards the Ol–Hy join. Thus the assumption of Sobolev et al. (1993)
and Sobolev & Danyushevsky (1994) that the linear extrapolation of cotectics defined by fertile mantle compositions could be used to estimate the P–T conditions of boninite formation is invalid, and results in a significant overestimation of the P of primary HCB generation. Our results also demonstrate the important control of mantle bulk composition on the position of cotectics in the normative projection.
H2O-undersaturated peridotite reaction experiments
We performed three H2O-undersaturated reaction experiments at both 1·5 and 2 GPa and four experiments at 2·5 GPa where the resulting run product was a melt in equilibrium with a refractory harzburgite residue (Tables 2 and 3). Although runs T-3462 and T-3465 at 1·5 GPa and T-3461 and T-3464 at 2 GPa produced dunite residues, the compositions of these experimental melts are not significantly displaced from an Ol + Opx + L cotectic. We therefore infer that the T values of the respective experiments are close to the T at which Opx is eliminated from the residue for the bulk composition used. Accordingly, we use T-3462, T-3461 and T-3464 to help define the position of Ol + Opx + L cotectics at 1·5 and 2 GPa in normative projections for H2O-undersaturated conditions (Fig. 5a and b).
|
In Fig. 5a and b, we compare the glass compositions from the H2O-undersaturated reaction experiments with the Ol + Opx ± Cpx + L cotectics defined by the anhydrous experiments presented in Fig. 4a and b. As can be seen from Fig. 5b, H2O has a significant effect at both 1·5 and 2 GPa on the position of the Ol + Opx + L cotectics for refractory mantle in the projection from Di. The presence of H2O at a given T causes the Ol + Opx + L cotectics to move towards the Qz apex, but maintaining their orientation relative to the anhydrous cotectics. It can also be seen from Fig. 5b that the displacement in the Ol + Opx + L cotectics towards the Qz apex, at similar H2O contents, is greater at 1·5 GPa than at 2 GPa.
Calculated H2O contents obtained from our mass balance calculations (Table 3) suggest that H2O contents vary from 2 to 3 wt % in the range 1450–1350°C at 1·5 GPa, 1500–1400°C at 2 GPa and 1500–1480°C at 2·5 GPa. As discussed in the analytical technique section, the analysed H2O contents of hydrous experiments were usually within 0·5 wt % of the calculated values, but the latter are considered to be more accurate.
The shift in the Ol + Opx + L cotectics, defined by the boninite–harzburgite reaction experiments, at 1·5 and 2 GPa is consistent with a change in the nature of the melting reaction, resulting from an increased proportion of Opx contributing to the melt (Green, 1976; Gaetani & Grove, 1998). This change in the melting reaction probably reflects the depolymerizing effect of H2O on melt structure. As Opx is relatively rich in SiO2 and poorer in FeO compared with Ol, the increase in Opx to the melting reaction is reflected in higher SiO2, lower FeO and higher MgO contents compared with anhydrous melts (compared on an anhydrous basis) at the same T, as can be seen from Fig. 6a–c.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONHIGH-Ca BONINITE PRIMARY MAGMA...RATIONALE OF EXPERIMENTAL STUDYEXPERIMENTAL AND ANALYTICAL...EXPERIMENTAL RESULTSDISCUSSIONREFERENCES |
|---|
Olivine liquidus depression as a result of dissolved H2O
To calculate the effect of dissolved H2O on the Ol liquidus, we need to accurately calculate the anhydrous Ol liquidus of water-bearing glasses. In this study, we have checked and then employed the Ol geothermometer of Ford et al. (1983)
to calculate the anhydrous liquidus of H2O-bearing glasses in equilibrium with Ol. The Ford et al. (1983) geothermometer is empirically based on a data set of 648 experiments at atmospheric P and 99 high-P experiments on dry mafic and ultramafic magmas and related synthetic systems (Ford et al., 1983). The empirical model of Ford et al. (1983) reproduces the Ol liquidus T of their data set to ±1°C. Danyushevsky et al. (1996) have demonstrated that the Ford et al. (1983) geothermometer can also reproduce the experimental T of atmospheric-P experiments conducted on MORB compositions to an accuracy of ±10°C. In Fig. 7a–c, we compare the calculated Ol liquidus T for 229 glass compositions in equilibrium with Ol from published and unpublished experimental studies performed at UT from atmospheric P to 3 GPa and T between 1100 and 1650°C (references in the caption to Fig. 7).
The atmospheric-P experiments using the Pt thermocouple (Fig. 7a) closely conform to a 1:1 line with an accuracy of ±10°C. High-P experiments using the W/Re thermocouple (Fig. 4b) are also normally distributed around the 1:1 line, but display a slightly larger scatter than the atmospheric-P experiments. This scatter can be explained by a number of factors, including analytical errors, quench modification of experimental glasses, presence of small amounts of H2O or other volatiles, and non-equilibrium. High-P experiments using the Pt thermocouple are shown in Fig. 7c. The generally higher calculated Ol liquidus T values of these experiments are consistent with the well-known drift of the Pt thermocouple in the relatively reducing environment of the piston-cylinder apparatus (Holloway & Wood, 1988). This shift was not avoided even by using short run times (average 2·5 h for experiments using the Pt thermocouple, average 41 h for experiments using the W/Re thermocouple). An alternative explanation, which we consider less likely, is that the majority of high-P Pt experiments were not entirely anhydrous and had small amounts of dissolved H2O (0·2 wt % on average).
We are confident that the atmospheric-P experiments and the high-P W/Re experiments are nominally anhydrous and these data clearly demonstrate that the Ford et al. (1983) geothermometer can be used with confidence to calculate the anhydrous Ol liquidus T for basaltic magma compositions. Our analysis of the Ford et al. (1983) geothermometer demonstrates that positive and progressively larger errors start to occur in the calculated liquidus T of basaltic liquids using the Ford et al. (1983) geothermometer if the alkalis are >3 wt %. Our data set has an average total alkalis of 1·8 ± 1 wt % (range 0·37–5·88).
Using the Ford et al. (1983) geothermometer, we have calculated the Ol liquidus depression caused by dissolved H2O for experimental glasses from the literature (and this study; references in the caption to Fig. 8) in equilibrium with Ol where the H2O content has been analysed or can be reasonably estimated, and these data are presented in Fig. 8 (n = 106, H2O contents 0·2–21 wt %, alkalis <3 wt %). In addition to the factors contributing to errors in calculating the dry liquidus T (see above), additional errors in determining the exact H2O contents of the experimental glasses are also a cause of scatter in Fig. 8.
|
