GEOLOGICAL SETTING

The Comores archipelago is situated in the northern part of the Mozambique Channel (11-13°S, 43-46°E), between East Africa and Madagascar. It trends NW-SE and comprises four main volcanic islands; Grande Comore, Moheli, Anjouan and Mayotte (Fig. 1). Grande Comore, the westernmost and largest island, is composed of two shield-volcanoes, Karthala and La Grille (Fig. 1) in undissected and youthful stages of growth. The large shield-volcanoes of Moheli and Anjouan islands are more mature and deeply eroded. The fourth island, Mayotte, represents a still more deeply eroded volcano, with an embayed coastline implying relative subsidence (Esson et al., 1970). Geochronological data confirm the morphology and volcanological evolution of the Comore Islands, giving progressively older ages from Grande Comore (still active) to Mayotte. Volcanic activity occurred at 3·9-0·36 Ma, 5-0·48 Ma and 7·7-1·49 Ma for Anjouan, Moheli and Mayotte, respectively (Emerick & Duncan, 1982, , 1983; Nougier et al., 1986).

Figure 1. (a) Location of Grande Comore island in the Comores archipelago; (b) La Grille and Karthala volcanoes.

Two hypotheses for the origin of Comores archipelago have been proposed: (1) lithospheric migration above a relatively stationary `hot spot', active during the last 10 my (Flower, 1972; Hajash & Armstrong, 1972; Emerick & Duncan, 1982; Class et al., 1996; Claude-Ivanaj & Allègre, 1996); (2) reactivation in different periods of a very slow spreading axis transected by numerous fractures (Upton, 1982; Nougier et al., 1986).

Three main stages of volcanic activity have been recognized for the active volcano, Karthala, associated with major fissuring, whereas two have been recognized on La Grille (Upton, 1982; Bachèlery & Coudray, 1993). Karthala is composed of alkali basalts (normative ne 3-8%), which are commonly ankaramites or oceanites (Strong, 1972). La Grille volcano is morphologically more evolved than Karthala and the two volcanic stages are identified by variably eroded and weathered lava flows, together with various cinder cones mainly located near the summit. The La Grille lavas are distinctly more undersaturated than those of Karthala, mainly ranging from basanites to olivine nephelinites (Strong, 1972). Mantle-derived spinel lherzolite and wehrlite xenoliths occur in the La Grille lavas: the samples discussed in this paper were taken from a lava flow at Gula Ivoini on the northern coast of Grande Comore.

ANALYTICAL METHODS

Whole-rock major and trace element compositions of GC1, GC2, GC3 and GC4 samples (Table 1), were determined at the Istituto di Mineralogia, Università di Ferrara (pressed powder pellets), using a Philips PW 1400 X-ray fluorescence (XRF) spectrometer. Data on precision and accuracy have been reported by Leoni & Saitta, (1976). Major and trace element compositions of GC11/2, GC11/4, GC11/7 and GC11/9 (Table 1) were determined at the Department of Geology and Geophysics, University of Edinburgh, on glass fusion discs and pressed powder pellets, respectively, using a Philips PW 1400 XRF spectrometer.

Table 1. Major oxide (wt %), trace element (ppm) abundances, normative and modal compositions of La Grille xenoliths

Mineral major element analyses were carried out at the Department of Geology and Geophysics, University of Edinburgh, using a Cameca Camebax electron microprobe in wavelength-dispersive mode. The acceleration voltage was 15 kV with beam current of 10 nA and peak counting time of 20 s for each element. The standards were natural and synthetic minerals and glasses; matrix corrections were made by PAP procedures. Major element analyses of glasses were obtained in a similar fashion, using a raster beam (20-100 µm²) to reduce alkali migration. Many of the glass analyses could only be performed by use of a fully focused beam (~2 µm). In this case, both the counting time (10 s) and the beam current (8 nA) were reduced to minimize alkali loss.

Analyses for Rb, Sr, Y, Zr, Nb, Ba and rare earth elements (REE) of clinopyroxenes and glasses were carried out in situ on polished, gold-coated thin sections using a Cameca imf-4f ion microprobe (SIMS) at the Department of Geology and Geophysics, University of Edinburgh. Measurements were made with an 8 nA ¹⁶O^- primary beam with a net impact of 15 keV, and focused to a spot of ~15 µm in diameter. Molecular ion interferences were discriminated using an energy filtering technique (Zinner & Crozaz, 1986). Intensities of all masses were measured over 10 cycles for each analysis point, with an acquisition time of 10 s per mass per cycle. To remove instrumental effects, the ion intensities were normalized to Si. Corrections were made for overlaps of rare earth oxide ions (MO⁺); the BaO overlap on Eu was calculated assuming that excess counts at mass 154 were ¹³⁸Ba + ¹⁶O. Because of the small size of some of the clinopyroxene grains, overlap of the primary beam onto surrounding glass was sometimes unavoidable. As Ba and Nb strongly partition into the glass these elements were used as a monitor of glass contamination. Where minor overlap is suspected both Ba and Nb were not reported.

PETROGRAPHY

The La Grille xenoliths, which contain spinel as the only aluminous phase, fall into the Group I xenolith category defined by Frey & Green, (1974). Their modal compositions, reported in Table 1, were obtained with a mass balance program using whole-rock, mineral and glass compositions (Table 2). On the basis of their petrography and textures, the xenoliths were subdivided into three groups: one group of lherzolitic rocks (Lh Group) and two groups of wehrlites (Wh1 and Wh2 Groups). Sample GC1, which lies on the dunite-wehrlite boundary, with 8% of modal clinopyroxene, has been grouped in Wh2 (Fig. 2). Textures vary from protogranular to porphyroclastic, according to the definitions of Mercier & Nicholas, (1975), with some xenoliths showing textural transition between the two. In the protogranular xenoliths, olivine and orthopyroxene occur as large crystals (up to 2 mm), with curvilinear grain boundaries. Kink-banding is common. Smaller crystals of clinopyroxene (up to 0·6 mm) and spinel (up to 0·4 mm) are typically interstitial or included within the orthopyroxene. The xenoliths with porphyroclastic texture have large porphyroclasts of olivine and orthopyroxene (generally kinked) with small neoblasts of strain-free olivine and clinopyroxene.

Table 2. Representative microprobe analyses of minerals and glasses of La Grille xenoliths

Figure 2. Modal composition of the La Grille ultramafic xenoliths.

Three pyrometamorphic textures (Pike & Schwartzman, 1976), superimposed on both the protogranular and porphyroclastic textures, have been distinguished on the basis of size and relative proportions of the secondary minerals (olivine, clinopyroxene and spinel) and glass (Table 3).

Table 3. Main petrograhic features of La Grille xenoliths

Type A has irregular, fine-grained (<0·10-0·15 mm) rims (or veins) developed on primary orthopyroxene, comprising clinopyroxene (~70 vol. %; cpx2) and olivine (~25 vol. %; ol2), ± glass (~0-2 vol. %), ± irregularly shaped spinel (~3-5 vol. %; sp1). In this texture, orthopyroxene shows clear evidence of destabilization and reaction processes (Fig. 3a).Type B comprises elongate, partly glassy (~5-8 vol. %) patches, containing a secondary assemblage of clinopyroxene (~75 vol. %; cpx2), olivine (~10 vol. %; ol2), and spinel (~7-10 vol. %; sp2: Fig. 3b). In this texture, the clinopyroxene and olivine grains are distinctly larger (0·1-0·6 mm) than in Types A and C. Secondary spinels are typically subidiomorphic or globular.Type C possesses secondary, very small clinopyroxenes (0·1-0·3 mm; ~50 vol. %; cpx2), sometimes aligned and in optical continuity, olivines (~6 vol. %; ol2), and irregularly shaped spinels (~4 vol. %; sp2), within glass (~40 vol. %) (Fig. 3c). The restricted occurrence of the clinopyroxene and the optical continuity suggest that it may have grown at the expense of a primary phase (possibly an earlier clinopyroxene).

Figure 3. (a) Type A texture: tiny crystals of olivine and clinopyroxene are forming at the expense of a large orthopyroxene. Interstitial glass pools are also present but not visible at this scale. (b) Type B texture: abundant glass together with secondary olivines, clinopyroxenes and subidiomorphic spinels. (c) Type C texture: small, commonly oriented, clinopyroxene aggregates together with small secondary olivines and brown glass (plane-polarized light, field of view 4 mm).

All the glasses observed in these samples are clear, with only rare evidence for incipient crystallization. Skeletal salitic clinopyroxenes, a few micrometres long, were observed only in one sample (GC11/7), where they are clearly the result of a quenching process. The colour, morphology and textural relationships between the glasses and associated minerals are highly variable, probably reflecting different degrees of equilibration with respect to the migration capability.

In the Type A texture, glass-producing reactions have occurred at the boundaries of, or within, the orthopyroxene whereas in Type C the original host is not preserved but is presumed to have been clinopyroxene. Glassy patches, light to dark brown in colour, are small (10-20 µm), lenticular with [Phi] angle of ~40-50°, and confined to the recrystallized areas. No veins or micro-channel connections have been observed (Fig. 3a and c). On the other hand, in the Type B texture, the glassy patches, pale yellow in colour, are somewhat larger (100-200 µm). They exhibit a lower [Phi] angle (~20-40°; Fig. 3b) and are located only at the edges and corners of olivine (both primary and secondary) crystals. Veins and network channels are also commonly observed (Fig. 3b).

MINERAL CHEMISTRY

T-P-f(O₂) CONDITIONS

Temperature estimates for La Grille xenoliths were obtained using the two-pyroxene geothermometer of Brey & Köhler, (1990) (BK) for lherzolites and by the O'Neill & Wall, (1987) spinel-olivine geothermometer (ONW) for lherzolites and wehrlites. Temperatures obtained from the spinel-olivine geothermometer are, on average, 100-150°C lower than those derived from the two-pyroxene thermometer (BK: 988-1071°C; ONW: 908-993°C; Table 4). This is probably due to the much faster Mg-Fe diffusion in spinel and olivine, which results in lower closure temperatures (Witt-Eickschen & Kramm, 1997). To evaluate the P-T condition of glass formation in equilibrium with the La Grille peridotites, a simple geothermobarometer involving a heterogeneous equilibrium between solid (ol-opx-sp) and melt phases was also tested. This method (BC) was originally calibrated by Bacon & Carmichael, (1973) to determine the P-T conditions under which melts are generated from lherzolitic sources, and here the algorithm is rewritten, modifying the activity formulation. The two equations which describe the exchange equilibria Mg-Si and Mg-Al are

	(1)
	(2)

In the calculations, activity formulations of solid phases were derived from the Wood & Kleppa, (1981), Kawasaki & Matsui, (1983) and Nell & Wood, (1991) models (see also below), whereas the aSiO2^liquid and aAl2O3^liquid were inferred from the values provided by Robie et al., (1978). This geothermobarometer gives a range of T values (BC: 930-975°C), comparable with those given by the O'Neill & Wall, (1987) method.

Table 4. T-P-f(O₂) estimates for La Grille xenoliths

There is a generalized decrease in temperature from the lherzolites to the wehrlites. Lh temperatures average between 908° and 993°C, Wh1 temperatures between 839° and 993°C, and Wh2 temperatures between 816° and 840°C (Fig. 4; Table 4).

Figure 4. [Delta]log f(O₂)-T diagram for primary and secondary parageneses of La Grille ultramafic xenoliths. Lh Group: [closed diamond], GC11/2; [open diamond], GC11/9; x , GC2. Wh1 Group: [closed triangle], GC11/7; [open triangle], GC3; [closed square], GC4. Wh2 Group: [closed circle];, GC1; [open circle], GC11/4.

P estimates were made using the Köhler & Brey, (1990) geobarometer (KB) and the Bacon & Carmichael, (1973) ol-opx-sp-gl geothermobarometer (BC). As far as the KB barometer is concerned, some criticism has arisen on account of the faster Ca diffusion in olivine than in pyroxene (Witt-Eickschen & Kramm, 1997). Nevertheless, it remains, as yet, the most suitable barometer for spinel-facies peridotitic rocks (Köhler & Brey, 1990; Luhr & Aranda-Gómez, 1997; Witt-Eickschen & Kramm, 1997). In this respect, it may be useful to note that results obtained from the two independent methods are in good agreement, giving most of the P values between 7 and 12 kbar (Table 4). The calculated T-P values cannot be used for reconstructing the geothermal gradient beneath Grande Comore because of possible temperature anomalies caused by metasomatism.

Oxygen fugacity [f(O₂)] values, reported in Table 4, were obtained from two heterogeneous equilibria:

(1)

(Wood et al., 1990; fW),

(2)

(O'Neill & Wall, 1987; fONW).

A crucial point for f(O₂) estimates is the determination of the Fe³⁺/Fe²⁺ ratios of the various phases. The Fe³⁺ content of the orthopyroxenes and spinels has been calculated using the Carswell & Gibbs, (1987) and O'Neill & Navrotsky, (1984) methods, respectively, assuming all the iron in olivine to be Fe²⁺. The amount of Fe³⁺ in the spinel solid-solution is difficult to evaluate, considering the high dilution of Fe₃O₄ component in the upper-mantle spinels. The Fe₃O₄ activity has been calculated using experimental data from Nell & Wood, (1991).

The two methods give comparable results, with fONW indicating slightly more oxidizing conditions than those obtained from fW, although in most of the cases values are within the precision limits (±0·3log units) calculated for both methods. Based on fONW, the f(O₂) calculated values for the xenoliths range between -0·56 and +2·68 [Delta]log unit with respect to the quartz-fayalite-magnetite (QFM) buffer, with the lowest values being recorded within the Lh Group (-0·56 to 1·29[Delta]log units), and the highest within the Wh2 Group (1·28-2·68[Delta]log units: Table 4).

GLASS AND CLINOPYROXENE COMPOSITIONS

The compositions of La Grille xenolith glasses are reported in Table 2 and are represented in an alkali-silica diagram (Fig. 5a). When compared with other xenolith glass occurrences (Fig. 5b), the La Grille glasses show compositional similarities, e.g. high SiO₂, Al₂O₃ and alkali contents, and low MgO, FeO and CaO contents. However, the Na₂O contents in some samples are unusually high (up to 13·5 wt % for GC11/4) and the glasses are chemically homogeneous within any single xenolith.

Figure 5. Alkali-silica diagram for glasses entrained in the La Grille peridotites (a) compared with glass occurrences from other silicate- and carbonatite-metasomatized continental and oceanic lithospheric mantle (b). L, Mt Lessini (Italy); Sr, Sardinia (Italy); A, Antarctica (Siena & Coltorti, 1993); CV, Cape Verde Islands; C, Canary Islands (Siena et al., 1991; Bonadiman, 1994); Y, Yemen (Chazot et al., 1996a); M, Central Mexico (Liang & Elthon, 1990); H, Hawaii (Sen et al., 1996); K, Kerguelen Island (Schiano et al., 1994); S, Western Samoa (Hauri et al., 1993); Sh, Southeastern Siberia (Ionov et al., 1995); Sp, Spitsbergen (Ionov et al., 1993); WE, West Eifel, Germany (Edgar et al., 1989). Symbols as in Fig. 4.

In the Lh Group, glasses are rare (invariably texturally associated with destabilized orthopyroxene) and are qz- or hy-normative (Fig. 6). They have the highest SiO₂ (58-70 wt %) and MgO (1·3-3·9 wt %) contents, withNa₂O and K₂O varying from 4·2 to 7·8 wt % and 0·36 to 2·18 wt %, respectively. TiO₂ contents are very low in the GC11/2 glasses (0·16-0·21 wt %) but higher in GC11/9 (1·5-1·8 wt %: Table 2). In this group, only sample GC11/2 was analysed for trace elements (Table 5, Fig. 7). The chondrite-normalized incompatible element patterns have Ba and Rb values of ~30 * chondrite. Nb, La and Ce have similar normalized values at ~150 times chondrite and a decreasing trend for all the other elements [up to 5-7 times chondrite for heavy REE (HREE)]. Sr, Zr and Ti (and, to a lesser extent, Y) show notable negative anomalies (Sr* 0·21-0·31, Zr* 0·02-0·06, and Ti* 0·2-0·4), with an extremely low Zr content (~100 times less than the other samples). The total REE content is between 60 and 200 ppm, with (La/Yb)_N between 27 and 50.

Table 5. SIMS trace element analyses of clinopyroxenes glasses from La Grille xenoliths; trace element contents of the calculated carbonatites shown in Fig. 11 are also reported

Figure 6. Normative compositions (Lc + Ne or Qz + Opx) vs Na₂O/K₂O ratio for the glasses of La Grille peridotites. `Opx-free' separates lherzolite from wehrlite fields. (See text for further explanations.) Field of La Grille basanites is reported from Bachèlery & Coudray, (1993). Symbols as in Figs 4 and 5, but most of the labels have been replaced by crosses (for sake of clarity). Normative minerals: Lc, leucite; Ne, nepheline; Qz, quartz; Opx, orthopyroxene.

Figure 7. Chondrite-normalized trace element patterns of clinopyroxenes and glasses of La Grille ultramafic xenoliths. Normalizing values for Ba, Rb, Nb, Sr, Zr, Ti, Y are from Thompson et al., (1983) and REE are from Boynton, (1984).

In the Wh1 Group the glasses are clear and occupy interstices between the olivine crystals. The compositionsare fairly constant within each sample but show some differences between samples GC11/7 and GC3 (SiO₂ 48-51 wt %; MgO 2·1-3·1 wt %; Na₂O 5·8-6·7 wt %; TiO₂ 2·3-2·9 wt %) and GC4 (SiO₂ 53-55 wt %; MgO 3-3·2 wt %; Na₂O 10-12 wt %; TiO₂ 1·2-1·8 wt %). The wehrlite glasses are silica undersaturated with normative ne between 10 and 35%. Na₂O/K₂O ratios increase with rising undersaturation (Fig. 6). The glasses have higher Na₂O and TiO₂ values than those of the lherzolite group, whereas the K₂O contents are comparable with those of the latter. Only for sample GC4 was a full analysis of the glass made. Further (incomplete) trace element data for samples GC11/7 and GC3 show comparable values. The chondrite-normalized patterns resemble that of GC11/2, but indicate higher contents for all trace elements, less pronounced Sr (Sr* 0·33-0·34) and Ti (Ti* 0·28-0·31) anomalies, and negligible Zr anomalies. The total REE content varies from 378 to 422 ppm with (La/Yb)_N of 17-18 (Fig. 7).

In the second wehrlite group (Wh2), the glasses have the most undersaturated compositions of the whole xenolith population, with 35-42% ne (Fig. 6) reflecting the very high Na₂O contents (11-14 wt %) and low SiO₂ contents (53-56 wt %). The TiO₂ and K₂O contents are also among the lowest for the whole glass population. The trace element contents and normalized patterns are very different from those of the Lh and Wh1 glasses. Together with marked Sr, Zr, Ti (and Y) (Sr* 0·3-0·7; Zr* 0·07-0·34; Ti* 0·03-0·12) negative anomalies, the glasses have very low Rb and very high Nb contents (up to 1320 ppm; Table 5, Fig. 7). They also have the highest REE contents ([Sigma]REE = 526-985 ppm), coupled with the most fractionated patterns [(La/Yb)_N = 37-191].

In Fig. 6, where the normative compositions of La Grille glasses are plotted against Na₂O/K₂O ratios, a decrease in the degree of saturation from the Lh Group to the Wh2 Group can be observed. Na₂O/K₂O ratios appear to be higher (up to 36; Table 2) than in any other reported glass analyses. Glasses with comparably high silica-undersaturated characteristics (up to 24% ne) have been described for xenoliths from Spitsbergen (Ionov et al., 1993), Cape Verde (Bonadiman, 1994) and Hawaii (Sen et al., 1996). However, in these instances the Na₂O/K₂O ratios remain low (below three). All the other glasses analysed to date have had much lower Na₂O contents and also lower Na₂O/K₂O ratios. For the La Grille glasses, the remarkable increase of Na₂O (and undersaturation degree) accompanies the disappearance of orthopyroxene in the peridotitic xenoliths.

Several similarities between the trace element contents of the glasses and those of the clinopyroxenes can be observed (Table 5). There is an increase in the total REE content of the clinopyroxenes from the Lh to the Wh2 Groups (27-90 ppm in Lh; 77-170 ppm in Wh1 and 70-276 ppm in Wh2) as well as, although less pronounced, in the (La/Yb)_N ratio (Lh cpx 4-12; Wh1 cpx 3-6; Wh2 cpx 3-17). Moreover, the almost ubiquitous, negative anomalies for Sr, Zr, Ti (and Y), in the chondrite-normalized clinopyroxene patterns are similar to those of the glasses from the same xenolith groups (Fig. 7).

Several chemical parameters suggest that equilibrium was approached in many of the secondary phases. These include the observed distribution of Ti between clinopyroxene, glass and spinel (Fig. 8a and b) and Ca between clinopyroxenes and secondary olivines (Fig. 8c), as well as the calculated clinopyroxene-glass partitioning coefficients which give results similar to those of Green, (1994) and Chazot et al., (1996b). Moreover, as shown above, geothermobarometric estimates, based on minerals or minerals and glass are broadly similar, suggesting a tendency toward equilibration between crystals and glass.

Figure 8. Ti contents in spinels (a) and glasses (b) vs Ti contents in clinopyroxenes, and CaO contents in olivines (c) vs CaO contents in clinopyroxenes from La Grille ultramafic xenoliths. In (c), dashed line and continuous line refer to secondary and primary olivine, respectively. Ti in spinels and clinopyroxenes is expressed as atoms per formula unit (a.f.u.) * 100, whereas TiO₂ in glass is expressed as mole fraction * 100.

MECHANISM FOR LA GRILLE GLASS FORMATION

No traces of hydrous minerals, such as amphibole and phlogopite, were found in the La Grille xenoliths. Thus the hydrous minerals decompression melting hypothesis (Francis, 1976, 1990; Stosch & Seck, 1980) is inapplicable for these xenoliths. In the same way, none of the studied xenoliths shows petrographic evidence for host magma infiltration (Ellis, 1976; Mertes & Schmincke, 1985; Garcia & Presti, 1987). This possibility could also be discounted by the remarkable difference in chemical characteristics between host basalts and glass compositions, such as Na₂O contents and Na₂O/K₂O ratios (see also Fig. 6). The lack of any hydrous phases would also appear to rule out the in situ melting model of Chazot et al., (1996a), which also fails to account for some chemical features. The absence of crystallites or clouding in the glass also rules out crystal fractionation processes.

Draper & Green, (1997) have demonstrated that felsic glasses, with compositions comparable with those of the La Grille xenoliths, could be in equilibrium with harzburgite, and close to equilibrium with lherzolite, at temperatures as low as 850°C and pressures between 10 and 20 kbar (either under anhydrous conditions or in the presence of a CO₂-H₂O fluid). Although some glass compositions may match this model, in general it is unreliable to believe that the wide compositional range (even in the same sample), which characterizes most of the xenolith glasses world wide, could only be obtained by this kind of process. As far as La Grille glasses are concerned, there are several reasons leading us to discount this model, such as (1) the difficulty in balancing the chemical budget, (2) the glass textural positions and their relationships with mantle minerals and (3) the observed reaction phenomena, which are not expected for equilibrium melting.

Several experimental studies on the migration of fluids through mantle matrices indicate that most of the La Grille glasses are unstable in the textural situations in which they occur (Type A and C textures), with only the Type B texture suggesting a tendency to a stable configuration. Melt is morphologically stable (and substantial permeability or interconnectivity possible) when located at ol-ol-ol edge regions (prismatic shape), or at ol-ol-ol-ol corner regions (tetrahedron shape). Conversely, melt is morphologically unstable in all edge and most corner situations involving at least one clinopyroxene, or, to a lesser extent, orthopyroxene (Toramaru & Fuji, 1986). As a result, melts are expected to migrate towards areas having the most stable configurations, and such redistribution will occur over a time-scale no longer than a few weeks (Jin et al., 1994). Moreover, in our opinion, it is not feasible to maintain clear glass for a long time in the mantle withoutcrystallization or chemical diffusion occurring. For these reasons, we believe that these glassy patches represent melt acquired at mantle depths, shortly before entrainment, and quenched during rapid transport to the surface, as also stated by Yaxley et al., (1997).

The petrographic and textural features of the La Grille xenolith glasses can be explained with a model involving interaction between metasomatic fluids and a primary mantle assemblage. Migration of melts through peridotite is strongly constrained by the value of the dihedral angle [Phi], which is a direct expression of the balance between the surface tensions of the solid grains and melts (Bulau et al., 1979; Waff & Faul, 1992). Watson & Brenan, (1987) and Watson et al., (1990) found large dihedral angles (>60°) for fluid mixtures of H₂O and CO₂, suggesting that peridotites should be almost impermeable to these components. Conversely, highly undersaturated silicate and carbonatite magmas present the lowest dihedral angles, conferring the greatest capability to permeate and metasomatize mantle materials (Hunter & McKenzie, 1989; Watson, 1991). Major and trace element compositions of the La Grille glasses vary widely among the three groups of peridotite xenoliths, but common features include (1) the very high alkali contents for all samples, (2) the very high Nb and REE contents, particularly for Wh2 Group glasses, and (3) the (almost) ubiquitous Sr, Zr, Ti and Y negative anomalies. To the best of our knowledge, there are no alkaline silicate magmas which can account for these striking chemical features. The most promising metasomatic candidate is a carbonate-rich melt. The remarkable chemical homogeneity of glasses within a single sample, which suggests a high diffusion rate, lends support to this hypothesis. Besides other parameters such as T, P, concentration, and the charge, size and structural position of the ions, diffusion rates are strongly dependent on melt viscosity, which is several orders of magnitude lower in carbonatite than in silicate melts (Treiman, 1989).

Modelling possible metasomatic reactions

To test the model, mass balance calculations were performed, assuming that primary mineral compositions in undepleted lherzolites react with a carbonatite melt, to yield the observed secondary minerals and glasses. These calculations make use of a carbonatite magma composition which is that of a melt shown experimentally by Wallace & Green, (1988) to be in equilibrium with a mantle paragenesis. It fails, however, to explain the high TiO₂ content of some of the glasses, as the TiO₂ content of the carbonatite is very low (Nelson et al., 1988; Wallace & Green, 1988). Thus, we were obliged to postulate the presence of one or more TiO₂-bearing minerals (e.g. phlogopite and amphibole) in the primary assemblage. Of these, phlogopite was preferred, as the several attempts to model the reactions using amphibole produced poor residual values, mainly because of the high K₂O (and TiO₂) contents in some of the glasses.

The formula used in the mass balance model is

± x1 ol1 ± x2 opx + x3 cpx1 ± x4 sp1 ± x5 ph + x6 carb = x7 ol2 + x8 cpx2 ± x9 sp2 + 1GL

where xi is the mass balance coefficient of the phase i; ol1, opx, cpx1, sp1, ph and ol2, cpx2, sp2 are the compositions of the primary and secondary minerals, respectively, carb is the Wallace & Green, (1988) carbonatite composition and GL is the composition of the glass. The least-squares residual is always <1 and generally ~0·2-0·3. The mass balance coefficients obtained for the various samples are summarized in the following reactions:

for the Lh Group:

(GC11/2) 6·5 opx + 0·08 sp1 + 0·03 ph + 1·0 carb = 4·6 ol2 + 1·9 cpx2 + 0·01 sp2 + 1 GL

(GC11/9) 4·8 opx + 0·2 sp1 + 0·2 ph + 0·7 carb = 3·7 ol2 + 1·3 cpx2 + 0·04 sp2 + 1 GL

for the Wh1 Group:

(GC11/7) 22·6 ol1 + 4·6 opx + 1·1 cpx1 + 0·04 sp1 + 0·2 ph + 1·1 carb = 25·7 ol2 + 2·9 cpx2 + 1GL

(GC3) 6·6 ol1 + 4·9 opx + 3·1 cpx1 + 0·03 sp1 + 0·2 ph + 1·0 carb = 10·4 ol2 + 4·5 cpx2 + 1 GL

(GC4) 8·9 opx + 3·6 cpx1 + 9·5 sp1 + 0·1 ph + 2·2 carb = 5·8 ol2 + 7·7 cpx2 + 9·8 sp2 + 1 GL

for the Wh2 Group:

(GC1) 7·1 opx + 3·5 cpx1 + 0·9 sp1 + 1·6 carb = 4·9 ol2 + 6·0 cpx2 + 1·1 sp2 + 1 GL

(GC11/4) 4·2 opx + 19·8 cpx1 + 0·6 sp1 + 1·1 carb = 3·7 ol2 + 20·3 cpx2 + 0·7 sp2 + 1 GL

The compositions of the phases used as reactants, together with the secondary phases (products) in each sample, are reported in Table 6, and the relative percentages of melted and recrystallized phases are shown in Fig. 9. These results agree with the observed petrographic characteristics, as: (1) evidence for incipient melting of orthopyroxene is often observed, and orthopyroxene never occurs as a reaction product [as has been noted by, for example, Green & Wallace, (1988), Yaxley et al., (1991) and Dalton & Wood, (1993)]; (2) clinopyroxene grows mainly at the expense of orthopyroxene, as observed in the three textural types described above, converting harzburgites or clinopyroxene-poor lherzolites to clinopyroxene-rich lherzolites or wehrlites; (3) spinel is preferentially melted, reducing its total volumetric percentages; this accords with the observations that the modal spinel content is usually low (<0·5%) (Table 1).

Table 6. Major element compositions of the reactant and product phases in metasomatic reactions modelling (see text for explanation)

Figure 9. Results of mass balance calculations between a migrating carbonatitic fluid (Wallace & Green, 1988) which reacts with a primary assemblage and gives rise to secondary parageneses (see text for further explanation). Ol, olivine; Cpx, clinopyroxene; Opx, orthopyroxene; Sp, spinel; Ph, phlogopite; Carb, carbonatite.

In the modelling we have assumed, at least for major elements, that a constant (carbonatitic) metasomatizing agent reacts with a variety of ultramafic rocks ranging from lherzolite or clinopyroxene-poor lherzolite to harzburgite, in each case containing variable contents of phlogopite. However, phlogopite may have been absent from the primary assemblage, or, alternatively, the reactions may have been confined to a restricted micro-zone, as suggested by the position and morphology of the glasses in Wh2 samples (Fig. 3c).

The model can be developed further. Using the trace element contents of the glasses and clinopyroxenes, and the modal proportions of primary and secondary minerals resulting from the mass balance calculations, the trace element contents of the metasomatizing agent have been calculated (Table 5). Two patterns, one for Lh and Wh1 Groups and the other for Wh2 Group, have been obtained, assuming different roles for the phlogopite. As shown above in the equations and in Fig. 9, phlogopite must have played an important part only in the genesis of the Lh and Wh1 glasses. Further support for phlogopite participation comes from Fig. 10, which shows that the trace element contents of the Lh and Wh1 glasses are (more or less) intermediate between those of carbonatites and phlogopite, whereas the trace element contents of the Wh2 glasses are closely comparable with those of carbonatites. The two calculated patterns for the inferred metasomatizing melts (Fig. 11) are very similar, and, most significantly, they closely resemble the trace element patterns of carbonatites from Tchivira-Bonga (Coltorti et al., 1993), Jacupiranga (Nelson et al., 1988) and Oldoinyo-Lengai (Dawson, 1989), as well as that of the experimental carbonatite obtained by Sweeney et al., (1995).

Figure 10. Chondrite-normalized trace element patterns of glasses of La Grille ultramafic xenolith compared with trace element contents of carbonatite from Tchivira-Bonga (Angola, Coltorti et al., 1993), Jacupiranga (Nelson et al., 1988), Oldoinyo-Lengai (Dawson, 1989) and from an experimental run (Sweeney et al., 1995). Analyses of phlogopites (*) from mantle peridotites (Western Victoria; O'Reilly et al., 1991) and from experimental work (*) are also reported (LaTourrette et al., 1995). Symbols: (a) Lh and Wh1 Groups: [closed triangle], GC11/2; [closed square], GC4. (b) Wh2 Group: [closed circle], GC1; [open circle], GC11/4.

Figure 11. Normalized trace element patterns of carbonatites obtained from mass balance calculations. [closed triangle], Lh and Wh1 Groups; [closed circle];, Wh2 Group. As in Fig. 9, for the former a primary assemblage containing phlogopite was inferred. (See text for further explanation.)

From the mass balance calculations, it was possible to calculate K_d values for the trace element distribution between clinopyroxenes and the inferred carbonatitic melt. These calculated values are reported in Table 7 and depicted in Fig. 12, together with the experimental data from Green et al., (1992), J. Adam et al. (unpublished data, 1993), Klemme et al., (1995) and Sweeney et al., (1995), and data from a natural occurrence (Hauri et al., 1993). The similarity between the Grande Comore patterns and that from Samoa is remarkable. A good comparison with the experimental data is also observed, although experimental K_d values for Ti and, to a lesser extent, Zr are not consistent with anomalies varying from positive to negative (Fig. 12). This variability may be due to analytical uncertainty in the experiments, where the data are systematically affected by high standard deviations (Klemme et al., 1995; Sweeney et al., 1995).

Table 7. Trace element clinopyroxene-carbonatite (calculated) partition coefficients from La Grille xenoliths compared with values from both experimental and natural occurrences

Figure 12. Distribution coefficients (K_d) between clinopyroxenes and calculated carbonatites for La Grille peridotites, compared with available K_{d cpx/carb} from both experimental and natural occurrences.

Comparison with other glass occurrences

Comparison with xenolith glass ccompositions in the literature must take into account the difficulty of analysing alkali-rich glasses. It is known that microprobe analysis of glasses may be affected by alkali loss or migration during analyses (Edgar et al., 1989; Heinrich & Besch, 1992; Schiano et al., 1994; Zinngrebe & Foley, 1995). There is also the possibility that analyses may give low totals because of the presence of volatiles that were not determined. The possibility of small inclusions being incorporated in the analyses is also high.

In Harker variation diagrams the La Grille glasses, as well as those from various xenolith localities around the world, show generalized trends (Fig. 13a and b). There is distinct negative correlation between FeO (and MgO) and SiO₂ contents, as has been observed by Edgar et al., (1989) for glasses from West Eifel (Germany). The steeply decreasing trend for FeO with respect to MgO, with increasing SiO₂, leads to a crude positive correlation between mg-number and SiO₂. This observation reduces the likelihood that the liquids resulted from any crystal fractionation processes, for which the reverse trend would be anticipated. Correlations with other elements are not so well defined, but there is a general decrease of CaO, and TiO₂ with rising SiO₂. Alkalis, particularly K₂O, show a random distribution with respect to SiO₂ (Fig. 13b), whereas an enrichment would be expected in a simple fractionation model.

Figure 13. Harker diagrams for glasses from La Grille peridotites and from the same localities as reported in Fig. 5. Symbols and abbreviations as in Figs 4 and 5. Arrow simulate an orthopyroxene melting process producing a secondary assemblage of olivine and clinopyroxene ± spinel.

Although formation of glass in the peridotites from both oceanic and continental environments has been ascribed to widely different metasomatic agents ranging from alkali-silicate (e.g. Liang & Elthon, 1990; Siena et al., 1991; Sen et al., 1996) to carbonatite (e.g. Hauri et al., 1993; Siena & Coltorti, 1993; Bonadiman, 1994; Ionov et al., 1994, , 1996), all may result from similar petrogenetic processes. Although there is vigorous debate as to whether the secondary minerals become more MgO rich than the primary minerals as a result of metasomatic reactions there is, in several xenolith populations (including that of La Grille), a tendency for the secondary olivines and clinopyroxenes to have the higher mg-numbers. Furthermore, some secondary spinels are more Cr rich (and occasionally more Fe³⁺ rich) than their primary counterparts (e.g. Canary Islands, Siena et al., 1991; Antarctica, Beccaluva et al., 1991; Spitsbergen, Ionov et al., 1993; Sahara, Dautria et al., 1992). In the La Grille glasses the inverse correlation between FeO and SiO₂ can be modelled if the glass is considered as a reaction product between `primitive' mantle materials and a metasomatic melt or fluid that caused orthopyroxene melting and recrystallization of olivine and/or clinopyroxene with mg-numbers similar to, or higher than, those of the primary minerals. The greater the difference between the mg-numbers of the primary and secondary minerals, the steeper will be the inverse FeO-SiO₂ trend. The presence of orthopyroxene in the primary paragenesis will also affect the silica-saturation degree of glasses, as observed in Fig. 6, but would not by itself modify the Na₂O/K₂O ratio.

Evidence from trace element contents of clinopyroxenes

Comparison of the trace element contents of the Grande Comore clinopyroxenes with those of other metasomatized xenoliths [ranging from supposed alkali-silicic metasomatism (e.g. West Eifel, Germany; Dish Hill, California; Massif Central, France: Johnson et al., 1996; NE Brazil and Paraguay: Rivalenti et al., 1996) to carbonatitic (e.g. Samoa: Hauri et al., 1993; Mongolia: Ionov et al., 1994)] demonstrates that the Grande Comore clinopyroxenes are remarkable in having (1) high REE contents (comparable with those of Samoan secondary clinopyroxenes, at 90-480 ppm) and (2) extreme Zr and Ti negative anomalies (Ti* 0·14-0·29; Zr* 0·03-0·34 in Lh GC11/2). It may be noted that xenoliths thought to be reaction products of a carbonatite melt show greater enrichments in REE and greater depletion in Ti and Zr (and, to a lesser extent, Sr) than those thought to have been metasomatized by alkaline silicate melts, irrespective of tectonic setting. In consequence, the Ti/Eu together with (La/Yb)_N ratios may be taken as indicators of carbonatite metasomatism (Rudnick et al., 1993; Klemme et al., 1995), the Ti/Eu usually being <1500, and the (La/Yb)_N usually >3-4. The La Grille clinopyroxene compositions plot well within the field for carbonatite metasomatism (Fig. 14).

Figure 14. (La/Yb)_N vs Ti/Eu ratios of clinopyroxene from La Grille ultramafic xenoliths, compared with clinopyroxene from continental and oceanic mantle domains which have been affected by silicate and carbonatite metasomatism. References: Samoa, Hauri et al., (1993); Mongolia, Ionov et al., (1994); Hawai, Sen et al., (1996); NE Brazil includes data from Pico de Cabuçi e Fernando de Noronha, Rivalenti et al., (1996); ETB (Eastern Transylvanian Basin), Vaselli et al., (1995); Dish Hill (California) and West Eifel (Germany), Johnson et al., (1996); Leura and Noorat (Western Victoria, Australia), Rivalenti et al., (1996); Yemen, Chazot et al., (1996a, 1996b).

SUMMARY AND CONCLUSIONS

Patches of glass with compositions varying from SiO₂ oversaturated (up to 24% qz), to strongly undersaturated (up to 40% ne) are common within the La Grille mantle xenoliths. The textural relationships between the primary minerals and their reaction products (secondary minerals and glass) permit distinctions to be made between three pyrometamorphic types (Types A, B and C). In these, olivine, clinopyroxene, spinel and glass are always present, although in variable proportions. The glasses are inferred to be products of metasomatic reaction between an alkali-carbonatitic melt and a mantle paragenesis, which mainly involved orthopyroxene, spinel and ± clinopyroxene, and possibly minor phlogopite. Orthopyroxene and, to a lesser extent, spinel were the first phases to react, changing the bulk compositions towards more clinopyroxene-rich assemblages and, ultimately, to wehrlite (Yaxley & Green, 1996).

Alkali element enrichment (mainly Na) and trace element anomalies (mainly Nb, Zr, and Ti) in the glasses suggest that the metasomatic agent was an ephemeral alkali-rich carbonatite. Evidence for its passage was provided solely by the reaction products: the carbonate-rich melt was entirely fugitive and lost by the system. The absence of any hydrated phase in the Grande Comore xenolith assemblages, together with the very high alkalis (particularly Na), Nb and LREE contents, leads us to discount the model of in situ amphibole melting (i.e. closed system) proposed by Chazot et al., (1996a) for the Yemen lithospheric mantle. Thus we are compelled to invoke an external agent which added chemical components to the system, as is also supported by our modelling of the metasomatic reactions.

The best fit in the mass balance calculations was obtained using the carbonatite composition obtained experimentally by Wallace & Green, (1988). Moreover, the equations obtained for the Lh, Wh1 and Wh2 Groups permit the estimation of the trace element contents of the hypothetical carbonatite responsible for the metasomatism. These are very similar to the trace element compositions of natural carbonatites, e.g. Oldoinyo-Lengai (Dawson, 1989) and Tchivira-Bonga (Coltorti et al., 1993). Consequently, we were able to calculate the trace element distributions between clinopyroxene and carbonatite under upper-mantle P-T-f(O₂) conditions and the results show good agreement with the (few) reported natural and experimental partitioning coefficients (Table 7).

In the Grande Comore xenoliths, textural evidence and mass balance calculations indicate a progressive increase in the amount of glass from the lherzolites (where the reacting protolith is still preserved) to wehrlites (which represent the final reaction products). There was a concomitant increase in the oxygen fugacity from Lh to Wh1 and Wh2 Groups with f(O₂) varying between -0·9 and +1·3[Delta]log units in the former and between +0·7 and +2·7[Delta]log units in the latter. These values are comparable, on the one hand, with values estimated for abyssal peridotites (-3 to +0·5[Delta]log units; Wood et al., 1990), which may be taken as unmetasomatized protolith, and, on the other, with those from oceanic silicate-metasomatized mantle xenoliths (-1 to +3[Delta]log units; Siena et al., 1991; Amundsen & Neumann, 1992), suggesting that a possible correlation between metasomatism and oxidation state also pertains for carbonatite-rich melt metasomatism. Redox data for carbonatite-metasomatized peridotite xenoliths from oceanic islands are, as yet, limited to this study and thus do not allow us to outline any general relationship between the oxidizing conditions deduced and the carbonatite melt assumed in our modelling. More data are required relating to carbonatite melts and their reaction with depleted oceanic mantle before a realistic hypothesis can be formulated. The metasomatizing melts or fluids are believed to have arisen from the lithosphere-asthenosphere boundary. However, the ultimate origin of the CO₂ and H₂O, responsible for oxidation of the lithospheric mantle remains enigmatic. The oxidized nature of the Comorien lithospheric mantle might be due to recycled oxidizing material derived from a subducting slab at a former convergent plate margin (Mattioli et al., 1989; Wood et al., 1990).

Geobarometric indicators suggest that the La Grille peridotites equilibrated at 7-12 kbar, above the carbonate stability field. Carbonatite melt with Na as dominant alkali was shown experimentally to be in equilibrium with phlogopite lherzolite at 20-25 kbar and 950-1170°C (Sweeney, 1994) or, for higher Na/K ratios, with amphibole lherzolite at 21-30 kbar and 930-1080°C (Wallace & Green, 1988). Residual amphibole in the Comorien mantle has been invoked for the petrogenesis of the La Grille magmas (Class et al., 1996; Spath et al., 1996). In our opinion, the La Grille xenolith glasses represent quenched reaction products, i.e. we are dealing with a metasomatic event still occurring in the sub-oceanic mantle. Lithospheric metasomatism is recorded at the shallowest level (7-12 kbar), whereas the host magmas are generated at depths >22 kbar (Spath et al., 1996) and carbonatite generation may be anticipated between 21 and 30 kbar (Wallace & Green, 1988). It is thus possible that carbonatite genesis (with subsequent metasomatic reactions at shallower level) and magma generations are two effects associated with the same phenomenon, i.e. the rise of a deep mantle plume.

The apparently carbonatite-metasomatized La Grille xenoliths represent, as yet, only the second such reported occurrence from an oceanic environment, the first being that from Samoa (Hauri et al., 1993). The Samoan data suggest that carbonatite-rich metasomatizing agents originated from recycled crustal components in the convecting mantle, and may be correlated with the HIMU and EMII components of OIB sources (see Zindler & Hart, 1986; Weaver, 1991, and references within). A prevalent HIMU contribution was deduced by Spath et al., (1996) for the La Grille magmas, whereas recycled crustal material was suggested by Class et al., (1996) for the same lava suite on the basis of helium isotopes. The results of our study also indicate a possible relationship between carbonatite metasomatism in the sub-oceanic mantle and the HIMU signature in the Comorien magmatism.

ACKNOWLEDGEMENTS

We are grateful to Drs P. Hill and S. Cairns for help with electron-microprobe analyses, and to Dr J. Craven for assistance with the ion-probe analyses. Our thanks go also to Drs D. James and J. G. Fitton for XRF analyses, to Professor B. Harte and Dr R. Sweeney for helpful discussions, and to Professor L. Beccaluva for constructive criticism of the manuscript.

REFERENCES