Journal of Petrology

Journal of Petrology Advance Access originally published online on December 7, 2005
Journal of Petrology 2006 47(4):647-671; doi:10.1093/petrology/egi088

This Article Abstract FREE Full Text (PDF) All Versions of this Article: 47/4/647    most recent egi088v1 Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Search for citing articles in: ISI Web of Science (13) Request Permissions Google Scholar Articles by DASGUPTA, R. Articles by STALKER, K. Search for Related Content GeoRef GeoRef Citation Social Bookmarking          What's this?

© The Author 2005. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oxfordjournals.org

Immiscible Transition from Carbonate-rich to Silicate-rich Melts in the 3 GPa Melting Interval of Eclogite + CO₂ and Genesis of Silica-undersaturated Ocean Island Lavas

RAJDEEP DASGUPTA^*, MARC M. HIRSCHMANN and KATHRYN STALKER

DEPARTMENT OF GEOLOGY AND GEOPHYSICS, UNIVERSITY OF MINNESOTA, 310 PILLSBURY DRIVE SE, MINNEAPOLIS, MN 55455, USA

RECEIVED APRIL 25, 2005; ACCEPTED OCTOBER 27, 2005

	ABSTRACT

TOP ABSTRACT INTRODUCTION EXPERIMENTAL TECHNIQUES EXPERIMENTAL RESULTS DISCUSSION REFERENCES

 
We explore the partial melting behavior of a carbonated silica-deficient eclogite (SLEC1; 5 wt % CO₂) from experiments at 3 GPa and compare the compositions of partial melts with those of alkalic and highly alkalic oceanic island basalts (OIBs). The solidus is located at 1050–1075 °C and the liquidus at 1415 °C. The sub-solidus assemblage consists of clinopyroxene, garnet, ilmenite, and calcio-dolomitic solid solution and the near solidus melt is carbonatitic (<2 wt % SiO₂, <1 wt % Al₂O₃, and <0·1 wt % TiO₂). Beginning at 1225 °C, a strongly silica-undersaturated silicate melt (34–43 wt % SiO₂) with high TiO₂ (up to 19 wt %) coexists with carbonate-rich melt (<5 wt % SiO₂). The first appearance of carbonated silicate melt is 100 °C cooler than the expected solidus of CO₂-free eclogite. In contrast to the continuous transition from carbonate to silicate melts observed experimentally in peridotite + CO₂ systems, carbonate and silicate melt coexist over a wide temperature interval for partial melting of SLEC1 carbonated eclogite at 3 GPa. Silicate melts generated from SLEC1, especially at high melt fraction (>20 wt %), may be plausible sources or contributing components to melilitites and melilititic nephelinites from oceanic provinces, as they have strong compositional similarities including their SiO₂, FeO^*, MgO, CaO, TiO₂ and Na₂O contents, and CaO/Al₂O₃ ratios. Carbonated silicate partial melts from eclogite may also contribute to less extreme alkalic OIB, as these lavas have a number of compositional attributes, such as high TiO₂ and FeO^* and low Al₂O₃, that have not been observed from partial melting of peridotite ± CO₂. In upwelling mantle, formation of carbonatite and silicate melts from eclogite and peridotite source lithologies occurs over a wide range of depths, producing significant opportunities for metasomatic transfer and implantation of melts.

KEY WORDS: carbonated eclogite; experimental phase equilibria; partial melting; liquid immiscibility; ocean island basalts

	INTRODUCTION

TOP ABSTRACT INTRODUCTION EXPERIMENTAL TECHNIQUES EXPERIMENTAL RESULTS DISCUSSION REFERENCES

 
Silica-undersaturated lavas are characteristic of many intraplate magmatic provinces, including those on continents (e.g. Wilson et al., 1995; Janney et al., 2002) and on many oceanic islands (e.g. Clague & Frey, 1982; Hoernle & Schmincke, 1993; Kogiso et al., 1997). Such lavas have conventionally been believed to originate from small degrees of partial melting of fertile peridotite, possibly in the presence of small amounts of CO₂ ± H₂O (e.g. Eggler, 1978; Hémond et al., 1994). However, detailed experimental studies documenting a link between such partial melts and alkalic lavas are incomplete, and many questions remain regarding their petrogenesis.

Alkalic lavas from oceanic islands have a wide range of compositions (Fig. 1) that presumably reflect a spectrum of sources and processes. As noted recently by Hirschmann et al. (2003) and Kogiso et al. (2003), key compositional characteristics of some alkalic oceanic island basalts (OIBs) differ significantly from liquids likely to be descended from those generated in existing experimental partial melting studies of peridotite (e.g. Hirose & Kushiro, 1993; Walter, 1998). This may be simply because such magmas are generated at very small degrees of melting of garnet peridotite, and experiments at the appropriately low melt fraction have yet to be achieved. On the other hand, associations between these lavas and isotopic signatures of crustal recycling (e.g. Hoernle et al., 1991; Kogiso et al., 1997; Janney et al., 2002; Jørgensen & Holm, 2002; Doucelance et al., 2003; Workman et al., 2004) have induced exploration of the hypothesis that the major element compositions of these lavas reflect a component derived from partial melts of pyroxenite or eclogite (Hirschmann et al., 2003; Kogiso et al., 2003). Similarly, a role for CO₂ should be considered, as lavas from these provinces show evidence for high CO₂ (e.g. Dixon et al., 1997), and their mantle source regions may be affected by CO₂-rich fluids and/or carbonatitic melts (Hauri et al., 1993; Saal et al., 1998; Coltorti et al., 1999; Kogarko et al., 2001; Neumann et al., 2002). In some localities, alkalic OIBs are associated with crustally emplaced carbonatites (e.g. Allègre et al., 1971; Hoernle et al., 2002).

View larger version (64K): [in this window] [in a new window]   Fig. 1. Compositions of melilitites and nephelinites from oceanic islands, and melilitites from continental provinces compared with high-pressure experimental partial melts of various potential mantle assemblages. Also included for comparison are ocean island basalts (OIB) with HIMU signatures and other primitive alkalic OIBs from the compilation of Kogiso et al. (2003). Oceanic melilitite and nephelinite [after classification of Le Bas (1989)] lavas are from Austral-Cook (Dupuy et al., 1988, 1989; Caroff et al., 1997; GEOROC database: http://georoc.mpch-mainz.gwdg.de/georoc/), Canaries (Hoernle & Schmincke, 1993; Demény et al., 2004), Cape Verde (Jørgensen & Holm, 2002; Doucelance et al., 2003; GEOROC database), Comoros (Nougier et al., 1986; Späth et al., 1996), Fernando de Noronha (Weaver, 1990; GEOROC database), Hawaii (Clague & Frey, 1982; Clague & Dalrymple, 1988; Maaløe et al., 1992; Dixon et al., 1997), Samoa (Hawkins & Natland, 1975; Palacz & Saunders, 1986; Workman et al., 2004) and Trindade (Oversby, 1971; Weaver, 1990; GEOROC database). Compositions of mafic (MgO >4 wt %) alkalic OIBs are from Azores, Tristan de Cunha, Society, Samoa, Pitcairn, Marquesas, Cook–Austral, and Canaries. Ocean island lavas with HIMU signatures are from Mangaia, Rurutu, and Tubuai from Cook–Austral and St. Helena. Selected data for continental melilitites are from Brey (1978), Frey et al. (1978), Dawson et al. (1985), Hegner et al. (1995), Wilson et al. (1995), Nielsen et al. (1997), Ivanikov et al. (1998), Janney et al. (2002), and Schultz et al. (2004). All natural lava compositions are normalized volatile-free. Experimental silicate-rich partial melts are from fertile peridotite between 2·5 and 7·0 GPa (Takahashi & Kushiro, 1983; Takahashi, 1986; Hirose & Kushiro, 1993; Kushiro, 1996; Walter, 1998), garnet pyroxenite between 2·0 and 5·0 GPa (Hirschmann et al., 2003; Kogiso et al., 2003), peridotite–basalt mixture between 1·5 and 2·0 GPa (Kogiso et al., 1998), and peridotite + CO2 at 3·0 GPa (Hirose, 1997). Arrows represent olivine fractionation trends of up to 35% from partial melts of Hirose (1997), using Fe2+–Mg KD = 0·33–0·39, determined from the same experiments (Hirose, 1997). This model of olivine extraction assumes fractionation of high-pressure liquids at shallow depths. Fractionation at intermediate pressures may involve cpx; however, such a process would not bring the trajectories closer to the target lava compositions.

 
Highly silica-undersaturated lavas such as melilitite and melilititic nephelinite (Le Bas, 1989) occur on some oceanic islands (e.g. Clague & Frey, 1982; Maaløe et al., 1992; Hoernle & Schmincke, 1993; Jørgensen & Holm, 2002; Doucelance et al., 2003; Demény et al., 2004). These lavas are characterized by much lower silica contents (Fig. 1) than those found in partial melting experiments of volatile-poor peridotite (Fig. 1). Many also carry distinctive trace element and radiogenic isotopic signatures similar to the HIMU mantle component (e.g. Hoernle et al., 1991; Jørgensen & Holm, 2002; Doucelance et al., 2003) and, thus, may incorporate recycled crustal material at their source. Melilitites and melilitic nephelinites from continents are similar, both in their major element compositions and their signatures of crustal recycling (e.g. Wilson et al., 1995; Janney et al., 2002). Also, there is a strong association in the field between melilitites, CO₂, and carbonatites (e.g. Schultz et al., 2004). Although volumetrically subordinate to more typical alkali basalt–basanite–nephelinite suites, these silica-poor alkalic lavas are of particular interest because experimental studies have so far failed to demonstrate equilibrium of melilititic melts with a four-phase (olivine + orthopyroxene + clinopyroxene + garnet) natural peridotite assemblage (Brey & Green, 1975, 1977; Brey, 1978; Brey & Ryabchikov, 1994).

Experiments investigating the possible role of carbonated eclogite in the petrogenesis of alkalic and highly alkalic OIB have not been performed previously. Here we present the results of partial melting experiments for a natural eclogite that has been modified by addition of carbonates (SLEC1; 5 wt % CO₂) at 3 GPa. In previous studies, we examined production of carbonatitic liquids from this carbonated eclogite (Dasgupta et al., 2004, 2005a). In the present study, we extend these experiments to higher temperature to document the transition from lower temperature carbonatite partial melts to carbonated silicate partial melts and to investigate possible relationships between these melts and the origin of alkalic (alkali basalt–basanite–nephelinite) and highly alkalic (melilitite–melilite nephelinite) intraplate lavas. First, we briefly review some relevant constraints on the petrogenesis of these lavas.

Origin of alkalic oceanic island basalt suites
It is well established that partial melting of peridotite ± CO₂ at high pressure and sufficiently low melt fraction gives rise to nepheline-normative liquids with compositional similarities to alkali basalt–basanite–nephelinite suites common on oceanic islands (e.g. Takahashi & Kushiro, 1983; Hirose & Kushiro, 1993; Hirose, 1997; Green & Falloon, 1998). However, in detail, liquids generated from existing experiments on peridotite ± CO₂ are distinct from many natural lavas in some important respects. For example, the lavas have lower Al₂O₃ and higher FeO^* and TiO₂ at a given MgO concentration than liquids likely to be derived by crystal fractionation from partial melts of volatile-poor peridotite or from partial melts of carbonated peridotite (Fig. 1; see also Kogiso et al., 2003). This may be because existing experiments on carbonated peridotite (Hirose, 1997) reflect relatively high degrees of melting. The compositions of small-degree partial melts of carbonated peridotite may possibly be more appropriate as parents for natural alkalic OIB suites.

As pointed out by Kogiso et al. (1998, 2003) and shown in Fig. 1, lavas with strong HIMU isotopic signatures of recycled oceanic crust are enriched in CaO and CaO/Al₂O₃ relative to experimental partial melts of volatile-poor peridotite. These must reflect contributions from CaO-rich melts, which may plausibly be related to the character of the HIMU source. Partial melts of carbonated peridotite are highly calcic, and, therefore, could play a role in HIMU genesis, although, as mentioned above, the lower FeO^* and TiO₂ and higher Al₂O₃ concentrations of carbonated peridotite partial melts from the experiments of Hirose (1997) are not appropriate for parental melts of the HIMU or other alkalic OIB lavas.

Recently it has been shown that high-pressure partial melting of silica-deficient garnet pyroxenite (MIX1G) also generates silica-undersaturated lavas similar to alkalic OIBs (Hirschmann et al., 2003; Kogiso et al., 2003). However, MIX1G does not produce melts with silica and alumina as low as primitive alkalic OIB and, thus, Kogiso et al. (2003) suggested that a source composition more silica deficient than MIX1G or a carbonated garnet pyroxenite could be a better source lithology for OIB parental melts. Keshav et al. (2004) argued that garnet pyroxenite is not a plausible source lithology for alkalic OIB suites because partial melting trends are transverse to variations of alkalic OIBs on MgO–Al₂O₃ and CaO–SiO₂ diagrams. This would be a valid consideration if varying degrees of isobaric partial melting were the principal cause of compositional variation in such suites. But if the alkali OIB major element trends chiefly reflect some other process, such as fractionation and/or accumulation of mafic phenocrysts, then the obliquity of the partial melting and lava trends is not relevant.

Origin of melilitites and melilitite nephelinites
In addition to low SiO₂, melilitites and melilitic nephelinites are characterized by high TiO₂, FeO^*, CaO, Na₂O and CaO/Al₂O₃ (Fig. 1). These cannot be derived by partial melting of peridotite in the absence of volatiles (e.g. Takahashi & Kushiro, 1983), but liquids with some of these compositional characteristics can be generated by partial melting of carbonated peridotite (Eggler, 1978; Hirose, 1997). For example, MgO-rich partial melts of carbonated peridotite produced at 3 GPa (Hirose, 1997) may evolve by olivine fractionation to liquids that match the SiO₂, CaO, Na₂O contents, and CaO/Al₂O₃,of melilitites and nephelinites (Fig. 1). However, these experimentally produced liquids are too poor in TiO₂ and FeO^* and too rich in Al₂O₃ at a given MgO content to account for ocean island melilitite, melilite nephelinite, and continental melilitites (Fig. 1). Thus, compared with partial melts of peridotite or carbonated peridotite from known experiments, both common alkalic OIB and melilititic lavas seem to require sources capable of generating higher TiO₂ and FeO^* and lower Al₂O₃. This may be why inverse experiments on the liquidus phase relations of olivine melilitite + H₂O + CO₂ for a range of upper mantle P, T, XH₂O, XCO₂, and fO₂ conditions failed to locate saturation with a lherzolite or harzburgite assemblage (Brey & Green, 1977; Brey, 1978). Instead, liquidus assemblages of primitive olivine melilitites at moderate to high pressures (0·5–3 GPa) include clinopyroxenite, garnet clinopyroxenite or wehrlite (Brey & Ryabchikov, 1994). Consequently, highly silica-undersaturated lavas may not be derived from sources consisting solely of typical carbonated natural lherzolite.

A key point is that much of the discussion regarding the origin of alkalic lavas has been based on comparisons between experimental partial melts and natural lavas projected into pseudo-ternary or pseudo-tetrahedral normative compositions (O'Hara, 1968). Although this is a powerful form of analysis, it has the potential to obscure key differences between experimental and natural liquids. For example, such projections do not address whether experimental partial melts have suitable FeO^* for a given MgO concentration (Herzberg & O'Hara, 2002) or whether they are sufficiently rich in TiO₂ to be parental to natural melilititic–nephelinitic lavas. For this reason, comparisons in this paper will be made chiefly through major element variation diagrams.

	EXPERIMENTAL TECHNIQUES

TOP ABSTRACT INTRODUCTION EXPERIMENTAL TECHNIQUES EXPERIMENTAL RESULTS DISCUSSION REFERENCES

 
Starting material
The starting material SLEC1 (Table 1) was prepared by mixing a natural eclogite xenolith powder (66039B; Dasgupta et al., 2004) with reagent grade and natural carbonates (Dasgupta et al., 2004). The silicate fraction (66039B) of our starting material falls along the thermal divide in the Fo–CaTs–Qtz system (at the crossover of CaTs–En and Fo–An; Kogiso et al., 2004a) and thus is silica-deficient compared with typical ocean crust. It was selected because its major element composition accounts for plausible subduction zone modification of typical ocean crust (Kogiso et al., 2003; Dasgupta et al., 2004, 2005a). CO₂ (5 wt %) was introduced by adding a mixture of natural siderite, magnesite and reagent grade CaCO₃, Na₂CO₃, and K₂CO₃ to keep the Ca:Mg:Fe:Na:K unmodified with respect to the base silicate fraction and thus to mimic the process of ocean-floor carbonation (e.g. Alt & Teagle, 1999; Nakamura & Kato, 2004) and possible modification during subduction (Dasgupta et al., 2005a, and references therein)

View this table: [in this window] [in a new window]   Table 1: Composition of the starting material

 
Experimental procedure
All experiments were carried out at 3 GPa using an end-loaded piston-cylinder apparatus, ¹/₂ inch assembly with BaCO₃ pressure cell, and Pt–graphite double capsules at the University of Minnesota, following the calibration of Xirouchakis et al. (2001) and procedures as described by Dasgupta et al. (2004, 2005a). After the experiments, capsules were embedded in epoxy, ground longitudinally, and polished to a ¹/₄ µm finish, using diamond pastes, for textural and compositional analyses. Because quenched carbonate melts are delicate and water-soluble, the samples were polished without water.

Analytical techniques
Wavelength-dispersive spectrometry (WDS) point analyses of the resulting phases were performed using a JEOL JXA8900R electron microprobe at the University of Minnesota. Accelerating voltage was set at 15–20 kV for all the phases. Analyses of the silicate and oxide minerals were performed using a fully focused 20 nA beam. Carbonate melts were analysed with a defocused beam of 1–6 µm and a current of 1–3 nA, depending on the dimension of the interstitial quenched melt pool. For runs where immiscible silicate melts did not segregate to form a large melt pool (A413, A418, and A423), a beam diameter of 1–6 µm and a current of 2–5 nA were employed to measure the silicate melts. We attempted to sample clean glass, but when this was not available we used a broad (up to 10 µm) beam to reintegrate quenched melt regions with exsolved Fe–Ti oxides. To obtain a reasonable estimate of the melt compositions from these experiments, we averaged as many spot analyses as possible by repolishing a new surface after each microprobe session for up to two to three times and reanalysing the quenched melt. Relatively large quenched silicate-melt pools for runs A429 and A426 were analysed with a 30 µm, 10 nA beam. Counting times for all elements were 20 s on peak and 10 s on each background for minerals and silicate melts, and 10 s on peak and 5 s on backgrounds for quenched carbonate melt. Analytical standards were natural cpx, garnet, olivine, ilmenite, feldspar, chrome spinel, and natural basalt glass for minerals and quenched silicate glass. For quenched carbonate mats, Ca and Mg were standardized on natural dolomite and Fe on siderite.

To confirm textural interpretations regarding the presence of carbonate and silicate melts, high-resolution imaging of the run products was performed in a JEOL-JSM-6500F field emission gun scanning electron microscope in the IT Characterization Facility of University of Minnesota.

	EXPERIMENTAL RESULTS

TOP ABSTRACT INTRODUCTION EXPERIMENTAL TECHNIQUES EXPERIMENTAL RESULTS DISCUSSION REFERENCES

 
Experimental conditions and corresponding phase assemblages are listed in Table 2 and micrographs of the run products are presented in Fig. 2. Phase proportions inferred from mass-balance calculations are also given in Table 2 and are plotted in Fig. 3 as a function of temperature. Compositions of minerals and melts are listed in Tables 3–7 and plotted in Figs 4–7.

View larger version (85K): [in this window] [in a new window]   Fig. 2. Secondary electron images from high-resolution field emission gun scanning electron microscope (a and c) and back-scattered electron images from electron microprobe (b, d, e, and f) of experimental charges. (a) Run A375 (3 GPa, 1150 °C, 6 h): carbonate-rich melt (Lc) occurs along grain edges and at triple grain junctions of garnet, cpx, and ilmenite [see also Dasgupta et al. (2004, fig. 2b)]. (b) Run A423 (3 GPa, 1350 °C, 22 h): immiscible carbonate-rich (Lc) melts are present at the interstices of garnet and cpx. (c) Run A418 (3 GPa, 1315 °C, 24 h): both carbonate-rich (Lc) and silicate-rich (Ls) melts are present along with residual garnet and cpx. Blebs and stringers of Fe–Ti oxides within domains of silicate-rich melts are interpreted to be quench products. Gr, graphite capsule. (d) Run A423 (3 GPa, 1350 °C, 22 h): coexisting carbonate-rich (Lc) and silicate-rich (Ls) melts are observed as well as residual cpx and garnet. Immiscible silicate-rich melt is concentrated towards the top of the capsule, whereas conjugate carbonate-rich (Lc) melts are confined to interstitial melt pools distributed across the charge. (e) Run A429 (3 GPa, 1375 °C, 22 h): quenched mat of carbonated silicate melt is composed of quenched cpx (qcpx) and quenched carbonate (qcarb). Also shown are residual garnet and cpx with interstitial quenched carbonate melt. (f) Run A426 (3 GPa, 1400 °C, 21 h): a mat of cpx (qcpx), carbonate (qcarb), and rutile (qrut) is interpreted as a quench product of a single melt phase present during the experiment. Residual garnet and minor cpx (not shown) are also present.

 

View larger version (25K): [in this window] [in a new window]   Fig. 3. Modes (expressed as weight fractions) of garnet, cpx, ilmenite, calcio-dolomitic solid solution, carbonatitic melt, and silicate-rich melt from mass-balance calculation of SLEC1 partial melting experiments at 3 GPa. The solidus is between 1050 and 1075 °C and the liquidus is at 1415 °C. Temperature interval (dashed lines) for coexistence of immiscible carbonate (Lc) and silicate (Ls) melt is also shown. The melting interval of carbonated eclogite SLEC1 is close to 350 °C.

 

View larger version (19K): [in this window] [in a new window]   Fig. 4. Compositions of carbonate-rich partial melts from SLEC1 partial melting experiments. Error bars are ±12 °C and ±1 (wt %). Two melts are stable in the temperature interval between the two vertical dashed lines. A definite change in carbonate melt composition can be noted from the change of slope in the oxide (wt %) vs temperature trends with the appearance of immiscible silicate melt.

 

View larger version (15K): [in this window] [in a new window]   Fig. 5. Compositions of silicate-rich partial melts from SLEC1 partial melting experiments. Concentrations of SiO2, TiO2, Al2O3, FeO*, MgO, CaO, and Na2O are plotted on a volatile-free basis, and that of CO2 is by difference of the average electron microprobe total from replicate analyses and 100%. Error bars are ±12 °C and ±1 (wt %).

 

View larger version (20K): [in this window] [in a new window]   Fig. 6. Compositions of garnet from SLEC1 partial melting experiments. Error bars are ±12 °C and ±1 (wt %). The absence of an oxide error bar for the 1010 °C data points indicates that fewer than three data points were available to average. Three dashed vertical lines mark the temperature of the solidus, initiation of silicate melt formation, and the liquidus. (Note the relatively small variation of CaO and TiO2, whereas Mg-number increases with rising temperature as evidenced by increasing MgO and decreasing FeO* after the silicate-rich melts appear.)

 

View larger version (20K): [in this window] [in a new window]   Fig. 7. Compositions of clinopyroxene from SLEC1 partial melting experiments. Error bars are ±12 °C and ±1 (wt %). The absence of oxide error bars for the 1010 °C data points indicates that fewer than three data points were available to average. Three dashed vertical lines mark the temperature of the solidus, initiation of silicate melt formation, and the liquidus.

 

View this table: [in this window] [in a new window]   Table 2: Summary of 3 GPa experiments: run conditions, phase assemblages and phase proportions

 

View this table: [in this window] [in a new window]   Table 3: Composition of carbonate-rich melt

 

View this table: [in this window] [in a new window]   Table 4: Composition of silicate-rich melt

 

View this table: [in this window] [in a new window]   Table 5: Composition of garnet

 

View this table: [in this window] [in a new window]   Table 6: Composition of clinopyroxene

 

View this table: [in this window] [in a new window]   Table 7: Composition of ilmenite

 
Phase assemblages and textures
Near solidus phase relations for SLEC1 have been reported by Dasgupta et al. (2004). At 3·0 GPa, the subsolidus assemblage (1050 °C) consists of garnet, cpx, ilmenite, and calcio-dolomitic solid solution. At higher temperatures, quenched carbonate melts (1075–1375 °C) and silicate melts (1225–1400 °C) are present. Above the solidus, at 1075 °C the appearance of calcio-dolomitic melt coincides with the disappearance of crystalline carbonate. Carbonatitic melt persists with residual garnet, cpx, and ilmenite to at least 1175 °C (Fig. 2a). At 1225 °C, a coexisting silicate glass appears. Ilmenite disappears between 1225 and 1275 °C. Carbonate and silicate melts coexist at least up to 1375 °C (Fig. 2c and d) and at 1400 °C a single CO₂-rich silicate melt is observed (Fig. 2e and f). Garnet and cpx are observed in all experimental charges, but based on phase proportion trends (Fig. 3), cpx is inferred to disappear within 10–20 °C of the liquidus and garnet is inferred to be the remaining phase at the liquidus, which is estimated to be at 1415 °C.

Quenched carbonate melts consist primarily of feathery mats of dolomite–ankerite_ss crystals (see Dasgupta et al., 2004, fig. 2b) and are present at triple grain junctions and along grain edges of residual garnet, cpx, and ilmenite (Fig. 2a). For relatively high-temperature runs (>1225 °C), acicular crystals of quench cpx and needles of rutile are also identified within the interstitial pools of carbonate melts. Immiscible silicate melts quench primarily to glass. However, variable amounts of exsolved Fe–Ti oxide are observed within quenched silicate glass at 1225, 1275, 1315, and 1350 °C (Fig. 2c and d). At 1375 and 1400 °C, the quenched silicate melt is an aggregate of quench-cpx, carbonate and rutile (Fig. 2e and f), and devoid of any glass.

Textural relations do not indicate any definite preferential wetting of mineral grains by carbonate or silicate melts (Fig. 2c and d); however, quench-overgrowths on residual garnet and cpx (Fig. 2b–e) suggest selective wetting of grain edges by silicate melt. In runs A410 (1225 °C), A413 (1275 °C), A418 (1315 °C), and A423 (1350 °C), interstitial silicate melt is observed to be present towards the top of the capsule only, where carbonate melts are confined to interstitial melt pools distributed across the charge (Fig. 2d, Table 2). This also indicates that the silicate melts form an interconnected, permeable, network under conditions where carbonate melt pockets are isolated. In run A429 (1375 °C), complete separation of quenched pools of silicate melt from a zone of garnet + cpx + interstitial carbonate melt is observed (Fig. 2e; Table 2), which further supports our inference of diminished mobility of carbonate melts when silicate melts are present.

Approach to chemical equilibration
The experiments reported here are unreversed and the back-scattered electron images show zoning in garnet for runs up to 1350 °C, reflecting incomplete chemical equilibration. However, an approach to equilibrium can be assessed as follows. All the experiments have been mass-balanced and for 10 new experiments whose phase compositions are reported here for the first time, the average sum of squared residuals (r²) is 0·74. Considering uncertainties in analyses of alkalis and estimation of CO₂ in quenched carbonate melts, these residuals are considered to be acceptable. The average r² is reduced to 0·50 when Na is left out of the mass-balance calculation. Garnet compositions (rim compositions for runs up to 1350 °C) also show systematic shifts with increasing temperature. Finally, generalized thermometers based on Fe²⁺–Mg partitioning between garnet and clinopyroxene (Ellis & Green, 1979; Krogh-Ravna, 2000) were compared with known experimental temperatures. Based on average analyses of garnet and clinopyroxenes, calculated temperatures fall within 65 ° of the nominal run temperatures. We consider this result to be supportive, given that average mineral compositions of the charge and not the texturally coexisting garnet–cpx pairs were used.

Role of quench modification of partial melt compositions
Formation of overgrowths on residual minerals and consequent modifications of partial melt compositions during quench is a long-standing problem in high-pressure, high-temperature experiments in mafic–ultramafic systems. Observations of variable proportions of quench-rims on cpx and garnets in the present study indicate some modifications of interstitial melt compositions that were stable during the experiments. However, we believe that any such effects are minor and that the reported average compositions (Tables 3 and 4) are representative of equilibrium melt compositions because: (1) both silicate and carbonate melts show systematic compositional evolution as a function of temperature (discussed in the ‘Phase compositions’ section); (2) mass-balance calculations, especially for the silicate melt-present runs, produce satisfactory sum of residuals squares (0·1–0·8); (3) values of garnet–silicate melt and cpx–silicate melt Fe^*–Mg K_D values show excellent agreement with those obtained from existing partial melting experiments of eclogite or garnet pyroxenite (e.g. Yaxley & Green, 1998; Hirschmann et al., 2003; Kogiso et al., 2003; Pertermann & Hirschmann, 2003a). Also, Fe^*–Mg K_D values for garnet–carbonate melt and cpx–carbonate melt are similar to those observed by Yaxley & Brey (2004) between 3·0 and 3·5 GPa.

Phase compositions
In the following section we summarize the compositional evolution of five phases observed—carbonate melt, silicate melt, garnet, cpx, and ilmenite—as a function of temperature across the 3 GPa melting interval of SLEC1. The composition of subsolidus crystalline carbonate has been given by Dasgupta et al. (2004, 2005a) and is not repeated here.

Carbonate melt
Quenched carbonate melts are broadly calcio-dolomitic, with Ca-numbers [molar Ca/(Ca + Mg + Fe)] ranging from 0·55 to 0·60, and show systematic compositional variations throughout the melting interval of SLEC1 at 3 GPa (Table 3, Fig. 4). With increasing temperature from 1075 to 1375 °C, SiO₂ increases gradually from 2 to 5 wt %. From 1075 to 1175 °C, concentrations of Al₂O₃, TiO₂, FeO^*, CaO and MgO, and Mg-number shows little variation. With the appearance of silicate-rich melt at 1200 °C, there are noticeable changes in most of the major element oxide vs temperature trends. With increasing temperature from 1225 to 1375 °C, carbonate melts become richer in Al₂O₃, TiO₂ and FeO^*, and poorer in MgO and CaO. Mg-numbers of immiscible carbonate melts diminish with temperature.

The low measured concentrations of Na₂O in the carbonate melts (0·3–1·8 wt %) are probably artefacts of polishing damage, as mass-balance calculations indicate deficits of the same. Reconstruction of the carbonate melt compositions, using mass balance to achieve zero residuals for Na₂O (Yaxley & Green, 1996) indicates that the carbonate melts are Na₂O-rich, with concentrations ranging between 4·5 and 8 wt % (Table 3).

Silicate melt
Low microprobe (93–96·5 wt %) totals in replicate analyses of quenched silicate melts indicate that SLEC1-derived partial silicate melts contain 3·5–7·0 wt % dissolved CO₂ (Table 4). Inferred CO₂ concentrations in the silicate melt increase from 3·5 ± 2·5 wt % at 1225 °C to 7·0 ± 1·75 wt % at 1350 °C, and then decrease to 5·5 ± 0·4 wt % at 1400 °C as the melt composition approaches the bulk composition of SLEC1 (5 wt % CO₂; Fig. 5). Inferred CO₂ concentrations are at the low end of solubilities measured in high-pressure experiments in natural and synthetic silicate systems (Brey & Green, 1976; Eggler & Mysen, 1976; Blank & Brooker, 1994; Thibault & Holloway, 1994; Brooker et al., 2001). They are lower than predicted from the 2 GPa solubility parameterization of Brooker et al. (2001). This is as expected from melts in equilibrium with carbonate melt, rather than CO₂ vapor. To facilitate comparison between analyses of silicate melts and those from other experiments or natural lavas, all the major element oxide concentrations of silicate melts discussed below, and presented in Table 4 and Fig. 5, are recalculated to microprobe totals of 100%.

Silicate partial melts of SLEC1 vary from melilitites at lower temperature to melilite nephelinites (Le Bas, 1989) at higher melt fractions. SiO₂ increases from 33·8 wt % at 1225 °C to 42·9 wt % at 1375 °C. The melt at 1400 °C has 43·3 wt % SiO₂ and approaches the composition of the starting material. From 1225 to 1400 °C Al₂O₃, MgO, CaO, and Mg-number show gradual increases, whereas TiO₂, FeO^*, and Na₂O show steady decreases with rising temperature and increasing melt fraction (Table 4, Fig. 5). The most interesting compositional features of the immiscible silicate partial melts are their extreme enrichments of TiO₂ and FeO^*, particularly at lower temperatures. At 1225 °C the silicate melt has 19 wt % TiO₂ and 25 wt % FeO^*. With increasing temperature and melt fractions, both TiO₂ and FeO^* diminish dramatically and approach the starting material concentrations for the respective oxides (Fig. 5).

Garnet
Garnet compositions are given in Table 5 and plotted in Fig. 6. Several oxides, including TiO₂, FeO^* and MgO, and Mg-number, show kinks in their trends around 1225 °C, the temperature of first appearance of immiscible silicate melt and of ilmenite breakdown. The Mg-number remains 60 from the solidus to 1225 °C and then increases smoothly to 75 at 1400 °C. With increasing temperature, TiO₂ contents increase from 0·6 to 1·0 wt %, while ilmenite is present, but are then reduced with increased melting and temperature to 0·4 wt % at 1400 °C.

Clinopyroxene
Compositions of aluminous clinopyroxene vary systematically across the melting interval of SLEC1 at 3 GPa (Table 6, Fig. 7). From the solidus (1050–1075 °C) to 1175 °C, their Mg-numbers vary little, from 77 to 75, but following the onset of silicate melting at 1225 °C, they increase steadily to 83 at 1400 °C. Al₂O₃ concentrations increase steadily from 4 wt % at the solidus to 7 wt % at 1400 °C; TiO₂ increases from 0·5 at the solidus to 0·9 at 1175 °C, but then decreases steadily, after the elimination of ilmenite from the residue, with increased degree of silicate melting, to 0·4 just below the liquidus. Na₂O concentrations drop from 2·1 wt % to 1·7 wt % across the solidus and then stay near-constant up to 1175 °C. From 1225 °C to 1400 °C, Na₂O drops from 1·7 to 1·0 wt % as the silicate melt fraction increases.

Ilmenite
At 3 GPa, the accessory Fe–Ti oxide phase in the melting interval of SLEC1 is stoichiometric ilmenite–geikielite solid solution (Table 7). From below the solidus (1050 °C) to 1175 °C, ilmenite composition varies little, with MgO concentrations of 6–7 wt % (Mg-number of 21–22). At 1225 °C, with the appearance of Fe–Ti-rich silicate melt, the MgO content of ilmenite increases sharply to 11 wt % (Mg-number of 38). Ilmenite is absent at or above 1275 °C.

	DISCUSSION

TOP ABSTRACT INTRODUCTION EXPERIMENTAL TECHNIQUES EXPERIMENTAL RESULTS DISCUSSION REFERENCES

 
Carbonate–silicate melt immiscibility during partial melting of carbonated eclogite
Our experiments provide clear evidence for coexisting carbonate and silicate liquids in the melting interval of SLEC1 carbonated eclogite at 3 GPa. This interpretation is supported both by textural evidence and by the compositions of coexisting liquids. The compositional evidence includes noticeable changes in the temperature vs major element oxide trend for carbonate melt at the appearance of immiscible silicate melt in SLEC1 (Fig. 4), and systematic shifts in Fe^*–Mg exchange equilibria between the two melts (discussed below). More generally, the relationship between the compositions of these coexisting melts can be illustrated in the temperature vs molar Ca/(Ca + Si) and molar Mg/(Mg + Ti) diagrams shown in Fig. 8. These clearly demonstrate the presence of a miscibility gap between the two conjugate melts in T–X (CaO–SiO₂ and MgO–TiO₂) space. The shape of the miscibility gap as observed in Fig. 8 probably indicates changes in structure of the silicate melts with increasing temperature. High initial TiO₂ contents indicate a polymerized melt structure with TiO₂ acting as a network-forming species (e.g. Dickinson & Hess, 1985). With rising temperature, the TiO₂ concentration in the silicate melts diminishes rapidly, accompanied by increasing SiO₂. An increase in CaO and Ca/Si, and decrease in Si/Al with increasing temperature probably enhances solubility in the melt (e.g. Kubicki & Stolper, 1995; Dixon, 1997; Brooker et al., 1999, 2001), as the silicate melts become less polymerized and approach the composition of the carbonate melt. The eclogite-derived carbonate-rich melt, in contrast, shows relatively limited compositional evolution as a function of temperature (Fig. 8), indicating very low solubility of SiO₂, TiO₂, and Al₂O₃ in its largely ionic structure.

View larger version (12K): [in this window] [in a new window]   Fig. 8. Temperature–composition diagram showing the miscibility gap between carbonate-rich and silicate-rich conjugate melts with respect to molar CaO/(CaO + SiO2) (a) and molar MgO/(MgO + TiO2) (b). The field boundary between one- and two-liquid fields is hand drawn and is not rigorously fitted. Closure of the miscibility gap from these T–X diagrams is consistent with textural evidence for the existence of two liquids at 1375 °C and one liquid (dark gray diamond) at 1400 °C. Experimental partial melt compositions from model (CMS–CO2, Moore & Wood, 1998) and natural (Hirose, 1997) lherzolite + CO2 systems are also included for comparison. The temperature for ‘silicate melting’ of carbonated peridotite is apparently too high to intersect the carbonate–silicate two-liquid field at 3 GPa. The molar ratios of melts from this study are calculated using Monte Carlo simulations where the microprobe error (1) of individual oxide concentrations from replicate analyses was randomly distributed about the average composition to generate 1000 solutions to the matrix equation; the plotted values and the corresponding error bars are the mean and 1 SD from the generated solution population.

 
The well-defined temperature interval of coexisting carbonate and silicate partial melts for carbonated eclogite SLEC1 at 3 GPa is distinct from the behavior observed in carbonated peridotite (Hirose, 1997; Moore & Wood, 1998). At 3 GPa, partial melting experiments with fertile natural peridotite KLB-1 + 5 % magnesite (Hirose, 1997) and with model CMS–CO₂ lherzolite (Moore & Wood, 1998) indicate a continuous transition from carbonatite partial melts near the solidus to silica-undersaturated silicate melts at higher temperatures. A continuous transition is also observed at 3·2–8·0 GPa in model CMAS–CO₂ peridotite (Dalton & Presnall, 1998; Gudfinnsson & Presnall, 2005). The difference between the 3 GPa carbonated peridotite studies and our results for carbonated eclogite probably lies in the temperature at which silicate melt is first stabilized. This is illustrated in Fig. 8, which shows that the temperature maximum of the miscibility gap observed for SLEC1 occurs at 1400 °C. Carbonated peridotite studies (e.g. Hirose, 1997) encountered silicate melt only at temperatures above this miscibility gap, thereby allowing a continuous transition from carbonatite to silicate partial melts. We note the Ca/(Ca + Si)–T and Mg/(Mg + Ti)–T projections of the miscibility gap do not rule out a narrow interval of coexisting carbonatitic and melilititic melts for the study of Hirose (1997) below 1400 °C (Fig. 8).

The intersection of melt compositions with the silicate–carbonate miscibility gap in the melting interval of eclogite but not peridotite at 3 GPa presumably relates to the lower temperature of silicate melting for eclogite (e.g. Kogiso et al., 2004a). According to Kogiso et al. (2004a), the chief factors that allow eclogites and pyroxenites to have lower solidi than peridotite are lower Mg-numbers and higher alkali content. Our observations for SLEC1 also suggest that bulk compositions with residual rutile or ilmenite should also stabilize silicate melts at low temperature owing to high activities of TiO₂ and associated TiO₂ enrichments in near-solidus partial melts. Interestingly, Hammouda (2003) also reported immiscible coexisting carbonate and silicate melt for a Ti-free carbonated eclogite bulk composition at 6·0 GPa and 1300 °C, which suggests that other compositional parameters may also play a role in the stability of silicate-rich melts of eclogite at relatively lower temperatures. We believe, however, that at high pressure, the partial melting behavior of carbonated eclogite may be more similar to that of carbonated peridotite, with a gradual transition from carbonate-rich to silicate liquid compositions and without clear immiscibility. This is because we expect the temperature of first appearance of silicate melt to increase with increasing pressure whereas the region of carbonate–silicate melt immiscibility should diminish, owing to increasing solubility of CO₂ in silicate melt.

Isobaric melting phase relation of carbonated eclogite
Melting reactions
Applying the method of Walter et al. (1995) yields the following average melting reactions (in weight fractions) for the melting interval of SLEC1:

(1)

(2)

(3)

Equation (1) indicates that the principal phase contributing to the near-solidus carbonatitic melt is [cc–dol]_ss, with minor contributions from cpx and ilmenite. The fraction of carbonate melt changes little from the solidus until silicate melt appears, at which point the carbonate melt fraction is reduced as indicated by reaction (2). Of course, most of the mass of the silicate melt derives also from the silicate and oxide minerals, with a significant contribution of ilmenite (0·24 weight fraction), as required by the TiO₂-rich composition of the melts, and garnet. With increasing melt fraction, the garnet contribution diminishes and the cpx mode entering the silicate melt phase increases. Also, as more CO₂ is dissolved into the silicate melt, the fraction of immiscible carbonatitic melt gradually decreases and finally disappears at 1400 °C.

Solidus and liquidus temperatures—effect of carbonates
The solidus (1050–1075 °C) temperature observed for carbonate-bearing composition SLEC1 at 3 GPa is distinctly lower than that observed for comparable compositions of nominally volatile-free pyroxenite, which have solidus temperatures in the range of 1300–1500 °C (see Kogiso et al., 2004a, fig. 5). The liquidus (1415 °C) of carbonated SLEC1 is also slightly lower than for nominally volatile-free equivalent compositions (1450–1550 °C; Kogiso et al., 2004a, fig. 6). Because addition of carbonate has a larger effect on the solidus than on the liquidus, it also has the effect of increasing the eclogite-melting interval.

Expansion of the melting interval for carbonated eclogite is partly due to the very low temperature onset of carbonatite melt formation. However, the interval of silicate melt formation is also increased, as carbonated silicate melt appears at 1200 °C, more than 100 °C lower than silicate melt would be expected in nominally volatile-free eclogites and pyroxenites (Kogiso et al., 2004a, and references therein). A similar lowering of the temperature of initial silicate melt formation is observed for carbonated peridotite at 3 GPa; i.e. in experiments with KLB-1 + 5 % magnesite, Hirose (1997) observed silicate melts at 1400 °C, whereas the CO₂-free solidus for KLB-1 is near 1500 °C (Takahashi, 1986). With increasing pressure and consequent enhanced solubility of carbonate in silicate melts, carbonate may have a more pronounced effect on the temperature of first-silicate melt appearance. Thus, at 6 GPa, the first silicate melt production for the CMAS lherzolite occurs at 1875 °C (Presnall et al., 2002, and references therein), whereas ‘silicate melt’ (>20 wt % SiO₂ in the melt) in CMAS–CO₂ occurs at temperatures as low as 1405 °C (Dalton & Presnall, 1998). Thus, the effects of carbonate on partial melting of upwelling mantle are twofold: in addition to producing carbonatite melts at great depth, more extensive carbonated silicate melt formation commences at depths considerably greater than the volatile-free silicate solidus.

Fe^*–Mg partitioning between immiscible silicate melts, carbonate melts, cpx and garnet
The Fe–Mg compositions of coexisting carbonate and silicate liquids in our experiments show systematic variations (Fig. 9). With increasing temperature, Mg-numbers of carbonatite liquids (L_c) diminish whereas those of conjugate silicate liquids (L_s) increase (Fig. 9a). This is due to the marked temperature dependence of the Fe^*–Mg partitioning between garnet, cpx, and the two melts, with , decreasing with temperature (from 0·7 to 0·2 and from 1·3 to 0·3, respectively, from 1175 to 1375 °C) and , increasing with temperature (from 0·2 to 0·4 and from 0·4 to 0·7, respectively, from 1225 to 1375 °C). The resulting temperature dependence of K_D (Fe^*–Mg) between carbonate and silicate melt, shown in Fig. 9b, is similar to that documented from studies of carbonate melt–silicate melt immiscibility at 2–3 GPa (Baker & Wyllie, 1990; Lee & Wyllie, 1997; Brooker, 1998). When compared with data from a wider range of pressures, it is also apparent that there could be a pressure dependence for the K_D (Fig. 9b), with Fe²⁺ partitioning preferentially into carbonate melt at higher pressures. When immiscible silicate melt is not present, modest decreases of with increasing temperature (from 1·7 to 1·4 from 1075 to 1175 °C) are also observed. Yaxley & Brey (2004) also found diminishing , from 1·7 at 1180 °C to 1·2 at 1250 °C, in 3 GPa experiments on carbonated eclogite. On the other hand, in the absence of immiscible silicate melt, shows little variation (0·65–0·70) between 1075 and 1175 °C and is similar to values reported by Yaxley & Brey (2004) (0·8 at 1180 °C and 0·6 at 1250 °C).

View larger version (14K): [in this window] [in a new window]   Fig. 9. Variation of Mg-number for carbonate-rich and silicate-rich melts (a) and Fe*–Mg KD between carbonate and silicate melts (b) as a function of temperature from the partial melting experiments of SLEC1. Fe*–Mg KD values from other experimental studies on carbonate–silicate immiscibility are also included in (b) for comparison. Other data on carbonate melt–silicate melt immiscibility are grouped according to the nominal pressures of the experiments: 0·1–0·3 GPa, Kjarsgaard et al. (1995) and Kjarsgaard (1998); 0·4–0·5 GPa, Kjarsgaard et al. (1995) and Kjarsgaard (1998); 1·0–1·3 GPa, Lee & Wyllie (1997); 2 GPa, Baker & Wyllie (1990), Lee & Wyllie (1997), Brooker (1998) and Hammouda (2003). The error bars for Mg-number and Fe*–Mg KD are calculated using Monte Carlo simulation as described in the caption of Fig. 8. The errors for KD values from the present study are within the size of the symbols where error bars are not shown. Fe*–Mg KD between immiscible carbonate and silicate melt increases broadly with pressure and temperature.

 
Partitioning of Ti between coexisting phases during melting of carbonated eclogite
An interesting feature of the silicate partial melts of SLEC1 is the very high TiO₂ concentrations (up to 19·4 wt % TiO₂) derived from a bulk composition with modest total TiO₂ (2·16 wt %). To our knowledge, similarly TiO₂-rich melts have not been observed previously in partial melting experiments involving mafic or ultramafic lithologies. The cause of these enrichments must be related to low values of mineral–melt partition coefficients for Ti. In this section we discuss Ti partitioning between silicate partial melts of SLEC1 and cpx, garnet and carbonate-rich melt and evaluate them in the light of previous studies of silicate melt–garnet–cpx equilibria.

Both and from the present study depend strongly on temperature. diminishes from 0·15 ± 0·02 at 1400 °C to 0·05 ± 0·01 at 1225 °C; diminishes from 0·2 ± 0·04 to 0·05 ± 0·01 over the same interval. Thus, strong enrichments in TiO₂ may be related in part to stabilization of melt at low temperature. On the other hand, Pertermann & Hirschmann (2003a) noted trends of increasing and with decreasing temperature from a compilation of experimental partial melting studies of pyroxenites and peridotites. Consequently, the observed low values of and in our experiments cannot be caused by low temperature alone.

Another factor that may diminish and is TiO₂ enrichment in the liquid (Fig. 10). For example, in experiments in Ti-rich systems, low values of and are observed (Van Orman & Grove, 2000; Dwarzski & Draper, 2004; R. E. Dwarzski, personal communication, 2005) (Fig. 10). Similar decreases of D_Ti with TiO₂ concentration in the melt are observed for opx–silicate melt equilibria (Xirouchakis et al., 2001). This non-Henrian behavior of TiO₂ in silicate melts is probably a contributing factor to the large TiO₂ enrichments observed in the silicate partial melts from our experiments. However, it would be circular reasoning to conclude that TiO₂-rich melts are both the principal cause and an effect of low D_Ti. Other influences must also be considered.

View larger version (13K): [in this window] [in a new window]   Fig. 10. Plot of TiO2 partition coefficients (grey diamonds) between garnet and silicate-rich melt (a), cpx and silicate-rich melt (b) as a function of TiO2 concentration (wt %) in silicate melts. Also shown for comparison are partition coefficients from high-pressure partial melting experiments on peridotitic assemblage (, Walter, 1998; Longhi, 2002), MORB-like eclogite assemblage (, Yaxley & Green, 1998; Pertermann & Hirschmann, 2002, 2003a; Pertermann et al., 2004), silica-deficient garnet pyroxenite assemblage (, Hirschmann et al., 2003; Kogiso et al., 2003), and experiments in Ti-rich systems in relation with lunar basalt petrogenesis [, Van Orman & Grove (2000) for cpx–melt; Dwarzski & Draper (2004) and R. E. Dwarzski (personal communication, 2005) for garnet–melt]. The D values from this study are calculated using Monte Carlo simulations where the microprobe error (1) of TiO2 concentrations from replicate analyses of both crystals and melts was randomly distributed about the average composition of TiO2 to generate 1000 solutions to the matrix equation; the plotted values and the corresponding error bars are the mean and 1 SD from the generated solution population.

 
The concentration of SiO₂ in the silicate liquid may also affect D_Ti, owing to its effect on the activity coefficient of TiO₂. As is well established from rutile-saturation experiments (Ryerson & Watson, 1987; Pertermann & Hirschmann, 2003a), silica enrichment raises . Thus, the low silica concentration in the carbonated silicate liquids from our experiments leads to a low , and consequently to small and . The low silica contents of the silicate liquids are in turn attributable partly to the effects of CO₂ (e.g. Kushiro, 1975). Enrichments in FeO^* in the silicate liquids may further diminish