Journal of Petrology Advance Access published online on August 24, 2007
Journal of Petrology, doi:10.1093/petrology/egm039
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Phase Relations and Melting of Anhydrous K-bearing Eclogite from 1200 to 1600°C and 3 to 5 GPa
Research School of Earth Sciences, Australian National University, Canberra, ACT 0200, Australia
Received November 23, 2006; Revised typescript accepted July 2, 2007
| ABSTRACT |
|---|
|
|
|---|
To investigate eclogite melting under mantle conditions, we have performed a series of piston-cylinder experiments using a homogeneous synthetic starting material (GA2) that is representative of altered mid-ocean ridge basalt. Experiments were conducted at pressures of 3·0, 4·0 and 5·0 GPa and over a temperature range of 1200–1600°C. The subsolidus mineralogy of GA2 consists of garnet and clinopyroxene with minor quartz–coesite, rutile and feldspar. Solidus temperatures are located at 1230°C at 3·0 GPa and 1300°C at 5·0 GPa, giving a steep solidus slope of 30–40°C/GPa. Melting intervals are in excess of 200°C and increase with pressure up to 5·0 GPa. At 3·0 GPa feldspar, rutile and quartz are residual phases up to 40°C above the solidus, whereas at higher pressures feldspar and rutile are rapidly melted out above the solidus. Garnet and clinopyroxene are the only residual phases once melt fractions exceed 20% and garnet is the sole liquidus phase over the investigated pressure range. With increasing melt fraction garnet and clinopyroxene become progressively more Mg-rich, whereas coexisting melts vary from K-rich dacites at low degrees of melting to basaltic andesites at high melt fractions. Increasing pressure tends to increase the jadeite and Ca-eskolaite components in clinopyroxene and enhance the modal proportion of garnet at low melt fractions, which effects a marked reduction in the Al2O3 and Na2O content of the melt with pressure. In contrast, the TiO2 and K2O contents of the low-degree melts increase with increasing pressure; thus Na2O and K2O behave in a contrasted manner as a function of pressure. Altered oceanic basalt is an important component of crust returned to the mantle via plate subduction, so GA2 may be representative of one of many different mafic lithologies present in the upper mantle. During upwelling of heterogeneous mantle domains, these mafic rock-types may undergo extensive melting at great depths, because of their low solidus temperatures compared with mantle peridotite. Melt batches may be highly variable in composition depending on the composition and degree of melting of the source, the depth of melting, and the degree of magma mixing. Some of the eclogite-derived melts may also react with and refertilize surrounding peridotite, which itself may partially melt with further upwelling. Such complex magma-genesis conditions may partly explain the wide spectrum of primitive magma compositions found within oceanic basalt suites.
KEY WORDS: eclogite; experimental petrology; mafic magmatism; mantle melting; oceanic basalts
| INTRODUCTION |
|---|
|
|
|---|
Mafic magmatism on Earth is primarily derived from melting of mantle peridotite, yet there is a growing consensus that multiple mantle components contribute to the production of many intraplate and some mid-ocean ridge magmas (Hofmann, 2003
Xenolith suites or exhumed mantle sections can provide important information on the compositions, melting behaviour and partial melt compositions of mantle eclogite (Hirschmann & Stolper, 1996
; Bodinier & Godard, 2003
; Pearson et al., 2003
), but undoubtedly the greatest insights into these processes are obtained from high-pressure, high-temperature experiments. Nonetheless, despite numerous experimental studies (e.g. Pertermann & Hirschmann, 2003b
; Kogiso et al., 2004
, and references therein) there remain fundamental gaps in our knowledge of the phase relations and melting conditions of eclogitic rocks as a function of pressure, temperature and bulk composition. Although a range of mafic compositions has been investigated, the effect that minor elements such as K, P and Ti have on phase relations and melt compositions is poorly understood. Technical limitations of available experimental apparatus have also hampered efforts to experimentally characterize mafic materials as a function of pressure and temperature. Experiments using the piston-cylinder apparatus allow precise control of pressure and temperature and have allowed detailed investigation of phase relations and melting of anhydrous eclogite or pyroxenite from 2·0 to 3·5 GPa (Yaxley & Green, 1998
; Klemme et al., 2002
; Pertermann & Hirschmann, 2003b
; Keshav et al., 2004
). However, most piston-cylinder systems are restricted to conditions of 3·5 GPa or lower. Multi-anvil devices have been employed for most experiments at higher pressure (> 4·0 GPa; Yasuda et al., 1994
; Tsuruta & Takahashi, 1998
; Tuff et al., 2005
), although temperature uncertainties for these experiments are commonly around 50°C. Therefore, details of phase changes over relatively small temperature intervals cannot be accurately resolved by multi-anvil experiments. For this reason the effects of temperature and pressure on phase relations and melt compositions of eclogite have been little studied.
In this paper, we present results of piston-cylinder experiments conducted from 3·0 to 5·0 GPa and from 1200 to 1600°C on a composition representative of altered mid-ocean ridge basalt (MORB). This material is expected to be a significant component of subducting oceanic crust and hence may be present in appreciable quantities in the mantle. Melting of eclogite under these pressure–temperature conditions may contribute to the genesis of many oceanic magma suites (Kogiso et al., 2003
; Pertermann & Hirschmann, 2003a
; Ito & Mahoney, 2005
; Sobolev et al., 2005
; Yaxley & Sobolev, 2007
). We define phase relations, solidus and liquidus temperatures, and melt compositions over the entire pressure range. These data provide a basis for evaluating the effects of pressure and temperature on mineral and melt compositions of anhydrous eclogite. These results, when considered with other experimental data, can be used to advance our understanding of melting of mantle heterogeneities.
| EXPERIMENTAL SETUP |
|---|
|
|
|---|
Starting material synthesis
For all experiments we have used a synthetic starting material that is representative of sea-floor-altered MORB (Table 1; see also Yaxley & Green, 1994
|
Experimental techniques
All experiments were conducted in end-loaded Boyd–England type piston-cylinder apparatus at the Research School of Earth Sciences, Australian National University (ANU). Experiments at 3·0 and 4·0 GPa were conducted in standard 200 T presses, whereas the 5·0 GPa experiments employed a 500 T, ultrahigh-pressure press. For all experiments a small amount (10–20 mg) of GA2 sample was placed with or without vitreous carbon spheres (see below) within graphite capsules (inner diameter c. 1 mm), which were subsequently placed in 3·5 mm Pt capsules. Prior to final sealing of the Pt, the open capsules were heated to 200°C in an oven for a minimum of 2 h (mostly >10 h) to ensure completely anhydrous conditions in the capsules. After removal from the oven the capsules were immediately welded closed. All experiments used standard 1·27 mm salt–Pyrex assemblies with straight graphite heaters and MgO spacers above and below the capsules. Care was taken to position the capsules within the hotspot of the assembly and within 1 mm of the thermocouple tip. For this assembly, temperature gradients across the sample are expected to be less than 10°C based on calibration experiments using several thermocouples. Temperature was controlled using type-B thermocouples (Pt94Rh6/Pt70Rh30) and a Eurotherm temperature controller and is precise to within 3°C. Pressure is converted from load and is accurate to 0·1 GPa.
Experiments were initially heated to 600°C with minimal confining pressure (
0·2 GPa) to soften the Pyrex before temperature and pressure were increased simultaneously until desired run conditions were reached. Because of the high run temperature and low-friction assemblies, all friction in the assemblies is expected to have been lost within the first 3–4 h of the experiments. As all experiment durations were
10 h (Table 2) no friction corrections were applied to the pressure measurements. Experiments were terminated by turning off power to the apparatus, resulting in quenching to below 200°C in <10 s. Recovered capsules were mounted in epoxy and polished to expose the sample. Capsules containing carbon spheres were reimpregnated with epoxy once exposed to prevent plucking from the capsule during polishing.
|
Use of vitreous carbon spheres
Processes of quench modification and insufficient melt pooling in experimental charges have long plagued experimentalists attempting to accurately determine the compositions of low-degree partial melts (e.g. Hirose & Kushiro, 1993
In this study, experiments conducted close to solidus temperatures were loaded with a layer of vitreous carbon spheres (50–120 µm in diameter) at the base of the graphite capsule, prior to loading of the GA2 powder. The vitreous carbon sphere layer did not exceed 30% of the inner volume of the graphite capsule in any experiments, but in most cases the layer was at least four spheres in thickness (Fig. 1a). The carbon sphere layer serves as an effective melt trap, as in all experiments above solidus conditions melt isolated from the solid residue could be identified as thin films around the carbon spheres or filling pores within the spheres. Spheres close to the crystallized starting material often contained tiny mineral phases in addition to melt, so only melt rims deep within the carbon sphere layer were targeted for analysis (Fig. 1b). As all experiments were run in graphite capsules, the addition of the carbon spheres does not add any additional chemical components to the experimental samples. Furthermore, potential problems relating to disequilibrium melting of mineral mixes (e.g. Falloon et al., 1996
) are not relevant here as we used a homogeneous synthetic starting material.
|
Analysis of mineral and melt compositions
Sections through the experimental samples were examined using a JEOL 6400 scanning electron microscope (SEM), housed at the Electron Microscopy Unit of the ANU. Mineral grains and quenched melt pools from the experiments were analysed for major elements using an EDS detector mounted to the SEM. Some additional mineral analyses were performed by wavelength-dispersive spectrometry (WDS) using a Cameca SX 100 electron microprobe at the Research School of Earth Sciences, ANU, and the results obtained using the two methods were indistinguishable within analytical precision. For the EDS analysis, accelerating voltage, beam current and counting time were set to 15 kV, 1 nA and 100 s, respectively. Element concentrations were standardized against known mineral standards produced by Astimex Scientific Limited.
A focused beam was used for all mineral analyses and care was taken to avoid overlap with mineral inclusions or quench overgrowths on grain rims. For the melt analyses a defocused beam of >25 µm2 was used where sufficiently large quenched glass pools were formed; typically in experiments with >25% melting. In these cases, quench modification of the glass pools adjacent to mineral grains could be avoided during analysis. Analyses conducted using this technique resulted in melt compositions with major-element totals of between 97 and 101%. For experiments containing low melt fractions, quenched glass preserved as thin (1–5 µm) rims around carbon spheres or in the pores of the carbon spheres were analysed using a focused beam. The low beam current used (1 nA) and anhydrous nature of the experiments reduce the likelihood for Na loss during analyses. Nonetheless, Na concentrations were continually monitored during acquisition to evaluate any Na loss. Sodium contents remained relatively constant in all cases, indicating that Na loss was negligible. In many cases the melt films around the carbon spheres were too thin to avoid overlap of the beam with the carbon spheres, which resulted in low major-element totals. Similar results were obtained from analysis of similar experimental run products by Pertermann & Hirschmann (2003b
) and Wasylenki et al. (2003
). Although carbon could not be quantified, increased height of the carbon peak in the EDS spectra was observed for many of the analyses. Nonetheless, for individual experiments, analyses of melt with major-element totals over 90% produced consistent element ratios and hence similar compositions when normalized to 100%. As an example, data for run MH25 are presented in Table 3. Therefore, only analyses that had totals of over 90% prior to normalization were used for calculating average melt compositions. For comparative purposes, melt compositions from all experiments were normalized to 100% major-element oxides. Similar procedures were used by Pertermann & Hirschmann (2003b
) and Wasylenki et al. (2003
).
|
In all but two experimental runs the graphite capsules appeared to have remained intact during the experiments, which effectively prevented contact of the sample with the outer Pt capsule. However, in experiments MH15 and MH22 the graphite capsules had undergone some shearing and veins of melt could be observed extending through the graphite towards the Pt capsules. Melt compositions from these experiments contained noticeably lower Fe contents compared with other experiments conducted at similar pressure–temperature conditions and reasonable mass balancing of Fe could not be made. Melt from these experiments are deemed to have lost Fe as a result of contact and Fe alloying with the Pt capsule. There is a progressive variation in the Fe content of the quenched melt veins cutting the graphite capsules, with the most Fe-depleted melt located close to the contact with the Pt capsule. As Fe diffusion in high-temperature melts is very rapid, the preservation of this Fe zonation indicates that the shearing of the graphite capsules and Fe loss from the melt occurred at a late stage of the experiments, possibly during run quenching. In this case, Fe loss from the melt is unlikely to have significantly affected the residual mineral compositions. Melt compositions have been determined across the entire range of investigated conditions and these melt compositions vary systematically with pressure and temperature (see below). Therefore, by interpolating between melt compositions from experiments conducted under similar conditions (interpolation between runs MH17 and MH14 for run MH15, and interpolation between run MH23 and the GA2 bulk for run MH22) we calculate that the melt from experiments MH15 and MH22 lost 2·7 and 1·5 wt% FeO, respectively. Recalculation of the melt compositions from these two experiments to higher Fe contents not only is consistent with concentrations expected from the degree of partial melting, but also allows for reasonable calculations of mass balance.
Phase proportions of all experiments were calculated based on major-element mass balance of all phases against the GA2 starting composition using least-squares regressions. Microsoft Excel® and Analyse-it® were employed to calculate phase proportions in experiments containing four or fewer phases. Both programs gave very similar results. As K2O is highly incompatible and is below detection limits in garnet and clinopyroxene (see below), weighting of K2O was increased to allow for more accurate calculation of melt proportions. In contrast, for the low-temperature experiments that contain more than four phases these programs were unable to calculate accurate phase proportions. For these experiments, phase proportions were first estimated based on petrographic observations and extrapolations from phase proportions of the higher temperature experiments. These estimated proportions were then adjusted iteratively until the best fit (minimum residuals to GA2) was obtained. As a cross check, incompatible trace-element (e.g. Cs, Rb, U) concentrations of the melts were also scaled against the composition of GA2 to provide an estimate of melt proportion (Table 2). The trace-element data will be presented elsewhere, but in all cases there is excellent agreement between the calculated melt proportions and constraints from the trace-element data. Details of the experimental run conditions, phase relations and phase proportions are presented in Table 2.
Attainment of equilibrium
Many aspects of the experimental protocol were specifically designed to optimize equilibrium at run conditions, including using a homogeneous glass starting material, careful assembly of experiments, conducting long duration experiments under stable pressure and temperature conditions, and ensuring rapid quenching at run termination. Nonetheless, all experimental products were carefully scrutinized to evaluate attainment of equilibrium. Within error, no systematic variations in mineral compositions were found across the width or length of the experimental charges, although a few experiments conducted close to the solidus contained some garnets grains with cores slightly higher in Ti than the rims. These garnet cores are interpreted to represent garnet growth before complete equilibrium was reached, and so we consider only the garnet rims as part of the equilibrium mineral assemblages. Mineral zoning (exclusive of quenching effects) was not found in any other minerals. Temperatures calculated from coexisting garnet–clinopyroxene pairs using the geothermometer of Ellis & Green (1979
) are within 50°C of the nominal run temperature for all experiments, and within 25°C of the nominal run temperature for all but four experiments (Table 2). These values are within the minimum uncertainty of ±50°C for the geothermometer (Ellis & Green, 1979
). Melt compositions were also homogeneous throughout the capsules of individual experiments. Melt compositions around the carbon spheres were found to be similar to melt pools present in the mineral matrix, although the latter melt pools were more variable in composition as a result of quench modification. In these cases only analyses of melt from the carbon spheres were used to calculate representative melt compositions.
| RESULTS |
|---|
|
|
|---|
Textural observations and phase proportions
In experiments conducted below or close to the solidus, the GA2 starting material recrystallized to form fine-grained (1–10 µm) eclogite dominated by clinopyroxene and garnet, with minor quartz–coesite, rutile and feldspar. In these experiments quartz–coesite is commonly observed in the matrix assemblage and as inclusions in garnet (Fig. 2a). With increasing temperature and, to a lesser extent, pressure the grain size of the experimental charges increases (Fig. 2), with individual clinopyroxene and garnet grains reaching hundreds of micrometres in size in some experiments containing high melt proportions (> 40%). In experiments with melt proportions in excess of 10–15%, melt appears to wet all grain boundaries and is completely interconnected throughout the charge (Fig. 2c). Phase proportions for all experiments are given in Table 2. From the quartz–coesite transition of Bose & Ganguly (1995
|
The solidus for GA2 is located at close to 1230°C at 3·0 GPa, around 1260°C at 4·0 GPa and around 1300°C at 5·0 GPa, giving a solidus slope of 30–40°C/GPa (Fig. 3). At 3·0 GPa garnet, clinopyroxene, quartz, rutile and feldspar are all present up to 1260°C, but quartz, rutile and feldspar are completely melted out by 1280°C. At higher pressure, feldspar is rapidly melted out above the solidus, and the width of the temperature interval in which rutile is stable above the solidus appears to diminish with pressure (Fig. 3). Coesite is stable up to 70°C above the solidus. The liquidus surface is less well constrained, although projections of melting percentages with temperature (Fig. 4) indicate that the liquidus at 3·0 and 4·0 GPa are located at approximately 1460 and 1520°C, respectively. GA2 is completely liquid at 5·0 GPa and 1600°C. These results constrain the slope of the liquidus to around 70°C/GPa. Melting intervals range from 230°C at 3·0 GPa to around 300°C at 5·0 GPa.
|
|
Phase proportions calculated from mass balance (Table 2, Fig. 4), reveal distinct inflections in melt productivity (that is, the rate of increase of melt fraction) as a function of temperature. At 3·0 GPa melting increases rapidly above the solidus until quartz, feldspar and rutile are exhausted at 1280°C. Melt productivity then decreases markedly until the melt proportion reaches around 30%, at which point rapid melting again resumes until the liquidus is reached (Fig. 4a). At 4·0 GPa initial melting close to the solidus eliminates both feldspar and rutile. Again, melt productivity with garnet and clinopyroxene (± coesite) remains low until c. 25% melting is reached, after which melt productivity increases up to the liquidus (Fig. 4b). At 5·0 GPa, a similar trend in melting systematics can be discerned, although it is noteworthy that the increase in melt productivity at around 25–30% melt fraction is less pronounced. Proportions of both garnet and clinopyroxene decrease progressively with increasing melt fraction and show little variation with pressure (Fig. 5). The only notable exception is in experiments that contain less than 20% melt, where a distinct increase in garnet mode as a function of pressure is apparent (Fig. 5b). Near-liquidus experiments (MH20 and MH22) and projection of phase proportions (Fig. 4) indicate that garnet is the liquidus phase at all investigated pressures.
|
Mineral compositions
Garnet
Garnet is present in all subliquidus experiments and in most cases forms porphyroblastic grains of homogeneous composition. Representative analyses are presented in Table 4. Garnets from all experiments are almandine (20–40 mol%) and pyrope (40–60 mol%) rich, with around 20% grossular component. Compositional changes with pressure are minimal, with only a slight decrease in almandine content observed with increasing pressure. In contrast, large compositional changes in garnet accompany increasing temperature or melt production. Trends of increasing garnet Mg-number with temperature (Fig. 6a) compare closely with the trends of melt productivity (Fig. 4) at the respective pressure conditions. Most notably, the kinks in the garnet Mg-number trends at Mg-number 56–58 (Fig. 6a) directly correspond to the prominent kinks in the melt proportion trends at 25–30% melting (Fig. 4). Therefore, there is a direct linear relationship between garnet Mg-number and percentage melting (Fig. 6b). Reduction of grossular contents with increasing melt proportion is relatively minor. All analysed garnets also have relatively high TiO2 contents (0·5–1·5 wt%) and detectable Na2O. In contrast, K2O was below detection limits (0·2 wt%) in all experiments.
|
|
Clinopyroxene
Clinopyroxene is the dominant mineral phase to crystallize from GA2 in all experiments containing <40% melt. At higher degrees of melting garnet is the most abundant residual phase (Fig. 4). In all cases, clinopyroxenes are Al2O3 rich (13·5–15·7 wt%) and omphacitic (Table 5), and are comparable in composition with clinopyroxenes reported from other experimental studies of dry eclogite melting (e.g. Yaxley & Green, 1998
|
|
Calculated molecular formulae for clinopyroxene always give low cation totals (Table 5), which are not an analytical artefact but are attributed to the presence of a Ca-eskolaite (Ca-Es) component (Wood & Henderson, 1978
|
|
Feldspar
Fine-grained feldspar was identified in experiments up to 1260°C at 3·0 GPa and in subsolidus experiment MH32 at 5·0 GPa, but could only be quantitatively analysed in the 3·0 GPa experiments (Table 6). At 3·0 GPa, subsolidus feldspar (MH1) is sodian sanidine with minor Ca contents (San41Ab50An9). Subsolidus feldspars of similar composition were reported for experiments on anhydrous granite at high pressure and temperature (Green & Lambert, 1965
). In our experiments feldspar compositions change dramatically above the solidus, becoming progressively depleted in K and enriched in Na and (to a lesser extent) Ca with increasing melt fraction (Fig. 8). Comparison of the phase relations (Fig. 3) and the projected feldspar compositional trends with temperature indicates that the feldspar would be essentially K-free at the point of complete feldspar exhaustion, as shown in Fig. 8.
|
|
Melt compositions
Average major-element compositions of melt from all melting experiments are presented in Table 7 and Fig. 9. Melt compositions vary significantly as a function of melt fraction and pressure. Compositions range from dacitic to andesitic with FeO, MgO, CaO and Al2O3 behaving compatibly and K2O, TiO2 and P2O5 behaving incompatibly. At melt fractions below 40% there is considerable variation in melt composition as a function of pressure. Trends in SiO2 and MgO do not vary significantly with pressure, whereas FeO and CaO more readily enter the melt at higher pressure. There is a pronounced decrease in melt Al2O3 and Na2O contents with increasing pressure. Most notably, Na2O is incompatible at 3·0 GPa, with Na2O contents increasing until feldspar is melted out, whereas Na2O is compatible in the solid residue at 5·0 GPa (Fig. 9). The TiO2 content of the dacitic melts also varies significantly with pressure, ranging from around 2·5 wt% TiO2 at 3·0 GPa to 5·5 wt% TiO2 at 5·0 GPa. Peak TiO2 contents tend to be close to melting out of rutile. K2O is highly incompatible at all pressure conditions, but is slightly less incompatible at 3·0 GPa where feldspar is still present. P2O5 is incompatible at all pressures.
|
|
| DISCUSSION |
|---|
|
|
|---|
Phase relations and element partitioning
In this study we have conducted experiments over a large pressure and temperature range using the piston-cylinder apparatus. Therefore, our experimental data allow detailed evaluation of phase relations and compositions of anhydrous eclogite as a function of pressure and temperature. As with all high-pressure (>2·0 GPa) studies of mantle rocks of mafic composition, the dominant mineral phases of GA2 are garnet and clinopyroxene. Both of these phases become progressively Mg-rich and Ti- and Fe-depleted with increasing melt fraction. At very high melt fractions clinopyroxene is lost from the residue, leaving garnet as the sole liquidus phase of GA2 at all investigated pressure conditions.
The most striking feature of the clinopyroxene compositions is the increase in jadeite and Ca-eskolaite components and decrease in Ca-Tschermaks and enstatite–hedenbergite components with increasing pressure (Fig. 7). A direct outcome of these compositional changes is a shift to more aluminous clinopyroxene compositions at higher pressure and high melt fractions (Fig. 5c). The increase in jadeite stability effects a strong shift towards partitioning of Na into clinopyroxene rather than melt with increasing pressure (Fig. 10). In contrast, clinopyroxene–melt partitioning of Ti appears to decrease only slightly with increasing pressure (Fig. 10), despite significantly higher TiO2 content in the melt at higher pressure (Fig. 9). At subsolidus to near-solidus conditions increased stability of jadeite with pressure is also likely to affect coexisting feldspar compositions. At the investigated conditions pure albite will break down to jadeite plus quartz–coesite at around 3·5 GPa (Holland, 1980
). In our experiments at 3·0 GPa feldspar contains significant Na2O, and becomes progressively K2O depleted with increasing melting (Fig. 8), indicating that the albite component may be crucial for feldspar stability above the solidus at P < 3·5 GPa. Above 3·5 GPa the albite component is expected to react to form jadeite (Holland, 1980
), so the subsolidus feldspars are likely to be K-rich. As K-feldspar appears to be rapidly melted out, this compositional change with pressure results in closing of the feldspar field against the solidus with increasing pressure (Fig. 3). The correlation of melt TiO2 content with pressure at low melt fractions (Fig. 9) also indicates that rutile may be more readily melted out with increasing pressure, which is consistent with the determined phase relations (Fig. 3) and the assertions of Klemme et al. (2002
). In contrast, pressure does not seem to have an effect on quartz–coesite stability with respect to the solidus, at least under the conditions investigated.
|
Comparison with previous studies
Previous experiments on dry melting of eclogite or garnet pyroxenite included a large range of starting compositions and hence have collectively resulted in a large variation of potential melting conditions (e.g. Kogiso et al., 2004
Yasuda et al. (1994
), Pertermann & Hirschmann (2003a
, 2003b
) and Yaxley & Sobolev (2007
) conducted experiments on MORB or oceanic gabbro starting materials (NAM-7, G2 and Gb108, respectively) that have lower K2O contents than GA2 (Table 1). The solidus temperatures for 3·0–5·0 GPa quartz–coesite-bearing eclogite obtained from these studies are at least 50–150°C higher than for GA2. The difference in melting conditions is almost certainly due to the higher K2O content of GA2, which stabilizes subsolidus sanidine feldspar and allows melting to begin at lower temperatures. This is also consistent with the very low solidus temperatures and phase assemblage of alkali basalt JB1 reported by Tsuruta & Takahashi (1998
) and Wang & Takahashi (1999
). Those researchers identified sanidine, rutile and quartz–coesite as subsolidus phases below 5·0 GPa, and determined solidus temperatures between 3·0 and 5·0 GPa as below 1200°C and 1300°C, respectively.
Most studies of eclogite melting have found that the melting interval is relatively small (< 200°C) and tends to decrease with increasing pressure (e.g. Yasuda et al., 1994
; Pertermann & Hirschmann, 2003a
). A notable feature of the phase relations determined for GA2 is that the melting interval increases with pressure from 230°C at 3·0 GPa to 300°C at 5·0 GPa, largely as a result of the low dT/dP of the solidus (Fig. 3). We attribute this difference in melting behaviour to depression of the solidus by the stability of accessory K-feldspar and rutile. With increasing pressure the stability fields of these phases shrink with respect to the solidus as Ti is increasingly partitioned into the melt (Fig. 9) and Na and K solubility in clinopyroxene and/or garnet increases (Fig. 10; Tsuruta & Takahashi, 1998
; Wang & Takahashi, 1999
; Harlow & Davies, 2004
). Tsuruta & Takahashi (1998
) also found a widening melting interval for alkali basalt JB1 with increasing pressure up to 5·0 GPa, which was also attributed to the presence of K-feldspar. However, between 6·0 and 7·0 GPa those workers observed a rapid increase in the solidus temperature, which correlates with a dramatic increase in K solubility in clinopyroxene and the loss of feldspar as a subsolidus phase (see also Wang & Takahashi, 1999
). At pressures above 7·0 GPa phase relations and liquidus and solidus temperatures for alkali basalt (Tsuruta & Takahashi, 1998
) and MORB (Yasuda et al., 1994
) (both forming coesite eclogite) are remarkably similar. Given the similarity in phase relations of alkali basalt JB1 and GA2 between 3·0 and 5·0 GPa, we expect that phase relations for JB1 and GA2 at P above 5·0 GPa are also comparable.
Several other experimental studies have focused on melting of quartz–coesite-absent garnet pyroxenites that are considered to be comparable with many mantle xenoliths (Ito & Kennedy, 1974
; Hirschmann et al., 2003
; Kogiso et al., 2003
; Keshav et al., 2004
). These compositions have appreciably lower contents of K2O, Na2O, TiO2 and SiO2, are substantially richer in MgO and undergo melting at significantly higher temperatures (>200°C) than eclogites of basaltic protolith. Kogiso & Hirschmann (2006
) investigated phase relations and melting of bimineralic eclogite with a bulk composition of SiO2-depleted MORB at 3·0 and 5·0 GPa. The solidus temperature for this composition was also found to be significantly higher (around 100°C) than that of other MORB-like compositions such as G2 (Pertermann & Hirschmann, 2003a
) or NAM-7 (Yasuda et al., 1994
). Of note, a feature of the Mg-rich and bimineralic garnet pyroxenites is a very narrow melting interval of <150°C (Ito & Kennedy, 1974
; Kogiso et al., 2003
; Keshav et al., 2004
; Kogiso & Hirschmann, 2006
), which indicates that melt is produced very rapidly at temperatures above the solidus. These results are consistent with phase petrology theory, which predicts higher solidus temperatures and smaller melting intervals as the bulk composition approaches simple two-phase mixtures of pyrope–almandine and diopside–hedenbergite (Kogiso & Hirschmann, 2006
).
Collective evaluation of the melting conditions of eclogite or garnet pyroxenite allows us to make some generalizations on high-pressure melting of mafic materials as a function of composition. Kogiso et al. (2004
) and Kogiso & Hirschmann (2006
) have already demonstrated that increasing alkalis (K2O and Na2O) will tend to reduce the solidus temperatures of eclogite or pyroxenite. However, such compositional changes seem to have little influence on liquidus temperatures. K-rich quartz–coesite eclogite will begin to rapidly melt at relatively low temperatures until minor phases (e.g. feldspar, rutile) are exhausted. After melting out of these minor phases, melt productivity significantly decreases until at a certain melt fraction the melting rate dramatically increases again (see Fig. 4). For GA2 this rapid increase in melting occurs at around 25–30% melting and corresponds to a clear change in garnet compositional trends (Fig. 6) and a reduction of variation in melt compositions as a function of pressure and temperature (Fig. 9). Similar abrupt changes from low melt productivity to high melt productivity have been reported from melting experiments on other alkali-rich mafic starting compositions (Tsuruta & Takahashi, 1998
; Tuff et al., 2005
). In particular, Tuff et al. (2005
) documented a rapid increase in melting of alkali ferropicrite (97SB68) at melt fractions of 20–30% from 3·0 GPa to 7·0 GPa. Such inflections seem to be absent or unresolvable from melting curves of alkali-poor eclogite or bimineralic pyroxenites (e.g. Ito & Kennedy, 1974
; Kogiso et al., 2003
; Pertermann & Hirschmann, 2003a
; Keshav et al., 2004
; Kogiso & Hirschmann, 2006
). In these cases, melt productivity is high and is exclusively or almost exclusively controlled by garnet and clinopyroxene. For our GA2 composition and the 97SB68 composition of Tuff et al. (2005
), rapid melting conditions at melt fractions of >30% also correspond to melting of a garnet–clinopyroxene residue. The change in melt productivity at these melt fractions corresponds to a change from melting conditions that are influenced by minor phases such as feldspar, rutile and quartz–coesite, to melting conditions that are controlled exclusively by garnet and clinopyroxene. Chemically, these minor phases reflect the importance of K2O, Na2O, TiO2 and SiO2 in controlling solidus temperatures. For GA2 the position of the melting inflection is parallel to the liquidus (Fig. 3) and is roughly consistent with the solidus temperature expected of a bimineralic garnet clinopyroxenite (Kogiso et al., 2004
; Kogiso & Hirschmann, 2006
). Based on these observations we label this inflection surface the garnet–clinopyroxene residue melting surface (Fig. 3).
In sum, there are two conditions of rapid melting of eclogitic rocks with increasing temperature. The first is near-solidus melting that is dominated by minor phases (e.g. rutile, feldspar, apatite, quartz–coesite). The second condition corresponds to melting of the garnet–clinopyroxene residue, which approaches the solidus and melting behaviour of two-phase garnet pyroxenite. These two melting intervals of relatively high melt productivity are separated by an interval of low melt productivity in the pressure–temperature range where garnet and clinopyroxene are dissolving in the low-degree melt component (e.g. Fig. 4a).
Melt compositions as a function of temperature and pressure
At all pressure conditions investigated, low-degree partial melts of GA2 are dacitic with high K2O contents and K/Na ratios. Similar melt compositions were reported for low-degree melting of GA1 (Yaxley & Green, 1998
) and G2K (Pertermann & Hirschmann, 2003b
); anhydrous starting materials that are of similar compositions to GA2. For compositions with even higher alkali element contents and K/Na ratios, low-degree melts tend to approach rhyolitic compositions (Tsuruta & Takahashi, 1998
). At similar melt fractions, partial melts of low-alkali quartz eclogite tend to be andesitic with higher TiO2 contents (Pertermann & Hirschmann, 2003b
). Pertermann & Hirschmann (2003b
) attributed these differences to the presence of alkali elements in the melt, which serve to increase SiO2 and decrease TiO2 contents of the melt while saturated in quartz–coesite and rutile. For garnet pyroxenites that lie on, or to the forsterite side of, the garnet–pyroxene thermal divide in CMAS or high-pressure projections [see Yaxley & Green (1998
) and Kogiso et al. (2003
) for details] melt compositions will be silica-undersaturated, Mg-rich picrites near the solidus (or at the ol–opx + cpx + ga eutectic), becoming basaltic with increased melting of residual garnet and clinopyroxene (Yaxley & Green, 1998
; Kogiso et al., 2003
; Kogiso & Hirschmann, 2006
).
Based on high-pressure (3·0–7·5 GPa) multi-anvil experiments, Kogiso et al. (2003
) and Kogiso & Hirschmann (2006
) documented a general increase in the CaO/Al2O3 and decrease in the Na2O contents of partial melts of garnet pyroxenite with increasing pressure. These compositional changes were respectively attributed to the increase in garnet stability relative to clinopyroxene (Kogiso et al., 2003
) and increasing Na2O partitioning in clinopyroxene (Kogiso & Hirschmann, 2006
) with increasing pressure. Other than these two studies, the effect of pressure on eclogite partial melt compositions has not been investigated. From our experimental results we observe some striking compositional variations in low-degree (< 30%) partial melts of GA2 as a function of pressure (Fig. 9). With increasing pressure, FeO and CaO more readily enter the melt, whereas Na2O and Al2O3 become increasing compatible in the solid residue. These effects are a direct consequence of increasing stability of garnet and jadeitic clinopyroxene with increasing pressure, which is in accordance with the finding of Kogiso et al. (2003
) and Kogiso & Hirschmann (2006
). The mode of garnet increases with pressure at melt fractions below 20% (Fig. 5b), which tends to produce melt with significantly lower Al2O3 contents. Furthermore, the slight shift to higher Mg-number in garnet with pressure (Fig. 6) effects the increase of melt FeO content with pressure. The increase in jadeite component of clinopyroxene with pressure is accompanied by a reduction in the Ca-Tschermaks and diopside components (Fig. 7), and increased CaO partitioning into the melt. At high melt fractions, garnet modes do not vary with pressure (Fig. 5b). However, the enhanced stability of jadeite with pressure produces pyroxenes with higher Al2O3 contents (Fig. 5c), and hence coexisting melts remain Al2O3 poor in these conditions (Fig. 9). Progressive increase in jadeite in pyroxene at high pressure also significantly affects clinopyroxene–melt partitioning of Na2O (Fig. 10). At relatively low pressure (2·0–3·0 GPa) Na2O behaves incompatibly during partial melting (Fig. 10; Pertermann & Hirschmann, 2003b
), whereas at 5·0 GPa and low melt fractions, Na2O is more compatible in the solid residue. A similar pressure effect on Na2O concentration was also described for low-degree melts of lherzolite (Longhi, 2002
).
Melt SiO2 and MgO contents do not vary significantly with pressure. As expected, K2O is highly incompatible, although slightly less so at 3·0 GPa, whereas feldspar remains stable in the residue. TiO2 contents of low-degree melts increase dramatically with pressure; this is partly due to decreasing relative stability of rutile with increasing pressure (Fig. 3), but may also be due to decreasing TiO2 partitioning between clinopyroxene (and garnet) and melt with pressure. Results from our experiments and those of Yaxley & Green (1998
) indicate that Ti partitioning between clinopyroxene and melt decreases with increasing pressure and temperature, whereas data from Pertermann & Hirschmann (2003b
) at 2·0–3·0 GPa do not indicate any pressure dependence (Fig. 10). Other than slight differences in bulk starting compositions we are unable to explain the inconsistency between these results. However, our results are again consistent with data from lherzolite melting experiments, which show a slight increase in TiO2 contents of melts with increasing pressure (Longhi, 2002
). Longhi (2002
) also observed a decrease in Ti partitioning between clinopyroxene and melt with increasing melt fraction, which was largely attributed to changes in melt compositions.
At high degrees of melting of GA2 (> 40%) melt compositions become remarkably similar, regardless of the pressure (Fig. 9). The notable exception is the Al2O3, which remains relatively low in the 4–5 GPa melts as a result of stabilization of more aluminous clinopyroxene, as discussed above. At these high melt fraction conditions—above the so-called garnet–clinopyroxene residue melting surface—melting is completely controlled by garnet and clinopyroxene and the initial SiO2–TiO2–K2O-rich liquid fractions become increasing diluted. A similar behaviour at high melt fractions is expected for a large range of high-pressure mafic rock-types, as under these conditions garnet and clinopyroxene are the principal, if not only, residual phases.
Expressed in phase equilibria terms, solidus temperatures and near-solidus melts in eclogitic rocks are strongly dependent on the minor phases and degree of silica undersaturation relative to the garnet + clinopyroxene compositional plane. The degree of silica undersaturation in garnet pyroxenite or eclogite is reflected in the presence of quartz–coesite or olivine (oversaturated and undersaturated, respectively) and more subtly in the presence of rutile, rutile + ilmenite, or ilmenite, and in the eskolaite content of clinopyroxene (silica oversaturated). Coupled with the silica-saturation control on partial melting, the presence of minor phases (ilmenite–rutile, corundum–kyanite, K-feldspar–phlogopite, apatite, carbonate–graphite) have the major control on solidus temperature, melt fraction and melt composition at temperatures below the garnet + clinopyroxene solidus. Once the latter solidus is reached, liquids approach the garnet + clinopyroxene plane in their compositions (i.e. approach basaltic compositions) from either the silica-undersaturated or -oversaturated (andesitic) side. It should be noted that this emphasis on silica saturation at high pressure uses a different measure [i.e. a high-pressure normative mineralogy (OHara, 1968
; Yaxley & Green, 1998
)] from that appropriate at low pressures (CIPW norm and basalt tetrahedron; Yoder & Tilley, 1962
).
The contrasting behaviour of potassium and sodium at high P
A basic premise underpinning igneous geochemistry is that alkali elements Na and K are highly incompatible for a large range of igneous processes and hence in most cases these elements are considered to behave in a very similar manner [e.g. the total alkalis–silica (TAS) diagrams of Le Bas et al. (1986
)]. Large variations in K or Na are often observed in volatile-rich alkaline rock suites (e.g. Foley, 1992a
), but these rocks are relatively rare. In contrast, little attention has been paid to the behaviour of K and Na during volatile-deficient magma generation at high pressure, either directly from experimental data or indirectly through the geochemistry of oceanic basalts. In general, the K2O and Na2O contents of most basaltic alkaline rocks do directly correlate, but there are often significant variations in Na/K ratios, even within single rock suites (e.g. Stolper et al., 2004
). The explanation for these variations has remained elusive.
An interesting aspect of our melt chemistry results concerns the behaviour of Na2O and K2O at high pressure during partial melting. At 3·0 GPa both K2O and Na2O behave incompatibly during partial melting, but with increasing pressure Na2O becomes progressively retained in the solid residue, whereas K2O becomes increasingly incompatible as the stability of supersolidus feldspar diminishes (Fig. 9). Therefore, Na/K ratios in the melts vary substantially over a range of pressure conditions. As Na2O is also hosted by the jadeite component of clinopyroxene in lherzolite, we might expect increasing pressure to have a similar effect on Na/K ratios of melts of mantle lherzolite. However, the stability of super-solidus phlogopite in lherzolite to high temperatures and low water fugacities (Foley et al., 1986
; Conceição & Green, 2004
) means that garnet lherzolite melting involves relative partitioning of K and Na between melt, clinopyroxene and phlogopite. Our results with eclogite demonstrate a mechanism by which deep partial melting and melt extraction from K2O- and Na2O-bearing material can effectively fractionate K2O from Na2O. As melting of such rocks is likely to contribute to oceanic magmatism to some degree (e.g. Hirschmann & Stolper, 1996
; Hofmann, 2003
), this process may explain some of the Na and K variations observed in oceanic basalts. Some specific examples are discussed below. However, we do not suggest that the Na/K ratio of oceanic basalts could be used as an exclusive monitor of the depth of melting, as there are many other potential factors that may influence the geochemistry of these rocks, such as geochemical variation of source rocks, degree of partial melting and melt–peridotite reaction processes. These are also discussed in more detail below.
Eclogite in the mantle
Subduction of oceanic crust and delamination of dense mafic cumulates from the base of continental crust are probably the









