Introduction

In a recent study of the basalt-peralkaline rhyolite volcanic complex Boseti (Main Ethiopian Rift; Brotzu et al. 1974, 1980), Ronga et al. (2010) found that the rhyolites had formed by prolonged crystal fractionation (90–95%) of alkali basaltic magmas. In detail, the rhyolites represent several, closely related, liquid lines of descent, differing, for example, in the level of Al-undersaturation and in Zr abundances. Ronga et al. (2010) presented chemical analyses of matrix glasses in the rhyolites which are among the most Fe-rich (FeO* up to 16 wt.%) and Al-poor (Al2O3 as low as 2.4 wt.%) yet recorded in peralkaline rhyolites. The glasses potentially hold important information on the composition of the natural minimum point to which peralkaline silica-oversaturated magmas evolve and the mineral-melt relationships operating at that point. In this paper, we provide further data on the matrix glasses, with the specific aims of assessing their significance in the evolution of peralkaline melts, and outlining, for the first time, mineral/melt relationships in such highly evolved compositions. We also comment briefly on the role of magma mixing in the evolution of the Boseti rhyolites.

Geological setting

The Boseti complex is a young volcano in the northern sector of the Main Ethiopian Rift (Fig. 1). It is composed of two coalescing main edifices, Gudda (2447 m a.s.l.) and Bariccia (2132 m a.s.l.). The volcanological history of Boseti has been described by Di Paola (1972), Brotzu et al. (1974, 1980) and Ronga et al. (2010). Activity at Boseti started with the emplacement of Pleistocene pre-caldera volcanic rocks and eruptions from centres peripheral to the main activity. Eruption of the pre-caldera rocks resulted in the formation of the main volcanic edifice and are represented by basaltic lava flows, spatter and cinder cones and peralkaline rhyolitic lava flows. This phase of activity ended with the emplacement of pantelleritic ash and pumice fall deposits. Contemporaneously with this stage, peripheral volcanism formed domes and composite cones. Caldera formation followed, now recognisable only in the western sector of the complex, resulting in the collapse of the main edifice; the younger, Gudda, edifice started to form. This post-caldera phase emplaced the largest volumes of magma in the Boseti complex. It started with the emplacement of pantelleritic lava flows which built up the Gudda edifice (Pleistocene-Holocene) and continued with eruption of peralkaline rhyolitic pumice deposits and lavas.

Fig. 1
figure 1

a Geological sketch map of the Boseti volcanic complex, redrawn from Brotzu et al. (1978). b Location of Boseti in the rift valley

Full size image

Almost simultaneously with the formation of Gudda, the Bariccia volcano (Pleistocene-Holocene) formed in two stages. The first stage saw the emplacement of trachyte flows followed by peralkaline trachytic and rhyolitic pumice deposits. The second stage was marked by the emplacement of pantellerite lava flows and pyroclastic deposits.

The eruptive products of Boseti display a marked bimodality, a feature found in other young central volcanoes in the Main Ethiopian Rift, such as Fantale (Gibson 1974) and Gedemsa (Peccerillo et al. 2003) (Fig. 1). Whilst arguing that the rhyolites were formed by prolonged fractional crystallisation of basalts, Ronga et al. (2010) explained the absence of intermediate rocks as a result of physical discrimination during eruptive processes.

Samples and analytical methods

Four samples of peralkaline rhyolites were used in this study, B350, B354, B355 and B375, which were erupted as part of the earliest, pre-caldera, stage of activity in the complex. Whole-rock analyses of all four (Table 1), and matrix glass analyses of three, were presented in Ronga et al. (2010). The samples were selected to give a range of relationships between whole-rock, glass and phenocryst phases. Phenocryst and matrix glass compositions were determined by electron microprobe at the Inter-Institute Analytical Complex at IGMiP Faculty of Geology, University of Warsaw, using a Cameca SX-100 microprobe equipped with four wavelength detectors. The accelerating voltage was 15 kV and the probe current was 20 nA for pyroxene, amphibole, olivine and spinel, and 15 kV and 10 nA and beam spot diameter of ~5 μm for feldspar, to reduce Na loss. For glass analyses, 15 kV and 6–10 nA and a dispersed spot of ~10–20 μm were used. Counting times for most elements were 20 s at peak and 10 s at background. Na was determined first, followed by Si. Fluorine in glass was analysed separately, using 15 kV and 40 nA with a dispersed spot. The standards used for glass analyses were: wollastonite for Ca, rutile for Ti, orthoclase for K and Al, synthetic Cr2O3 for Cr, albite for Na, diopside for Si and Mg, hematite for Fe, rhodocrosite for Mn, tugtupite for Cl and synthetic fluor-phlogopite for F. For clinopyroxene and olivine, the standards were: wollastonite for Si and Ca; other elements as for the glasses. Diopside was used as the Ca standard in feldspar analyses, with barytes for Ba; other elements as for the glasses. The ‘PAP’ φ(ρZ) program (Pouchou and Pichoir 1991) was used for corrections. Representative and/or average analyses are presented in Tables 2, 3 and 4; the full data sets are given in Supplementary Tables 1 and 2, available as electronic supplementary material to this paper.

Table 1 Major (wt.%) element analyses of study rocks
Full size table
Table 2 EMP analyses of alkali feldspars
Full size table
Table 3 Representative EMP analyses of olivine and clinopyroxene phenocrysts
Full size table
Table 4 EMP analyses of matrix glass and melt inclusions
Full size table

Petrography and nomenclature

The samples studied have CIPW normative quartz (q) between 17% and 31%. Taking the trachyte-rhyolite boundary at the conventional 10% q, all samples are classified as rhyolites. Their peralkalinity indices (PI = mol. (Na2O + K2O)/Al2O3) range from 1.22 to 1.84 (Ronga et al. 2010) and according to the classification scheme of Macdonald (1974a), they are pantellerites.

Comendites, less peralkaline rhyolites with generally lower FeO* and higher Al2O3 contents, are also present in the suite (Ronga et al. 2010). Below, we give a short petrographic description of each sample.

B350

Dark brown and grey glassy components of different composition are patchily and streakily intermingled (Fig. 2a). Subhedral to euhedral alkali feldspar phenocrysts up to 2 mm across occur in both components but are more common in the darker type. Some feldspars contain melt inclusions (Fig. 3b). There are rare, green pyroxene phenocrysts. A sulphide phase occurs mainly as inclusions in feldspar and pyroxene; one microphenocryst in the matrix has an unusual torpedo shape (Fig. 3a). Both darker and lighter components contain small (<10 μm) patches of glass, especially associated with feldspar phenocrysts.

Fig. 2
figure 2

Photomicrographs showing textural evidence of magma mixing in Boseti rhyolites. a Streaky mingling of lighter and darker components, B350. b At least three components in B354. c Dark border forming boundary between lighter and darker components in B354. d Mingling of darker and lighter components partially obscured by devitrification of both components. Plane-polarised light used in all

Full size image
Fig. 3
figure 3

BSE images of selected mineralogical features. a Unusual, torpedo-shaped sulphide microphenocryst in B350. b Partially devitrified melt inclusion in feldspar phenocryst in B350. c Carbonate with glass rims, interstitial to alkali feldspar in matrix, B375. d Matrix wedge between two feldspar phenocrysts, B375. Glass forms pools in the matrix and along contact with phenocrysts. Other phases are aegirine-augite (px), sanidine (Kfs), quartz (Q), aenigmatite (Aen) and ilmenite (Ilm)

Full size image

B354

There are three glassy matrix components of different composition (Fig. 2b,c). A dark brown component (1) is streakily intermingled with a grey material (component 2). The third component is very dark, almost black, and mainly occurs mingled with component 1. In places, component 1 forms a rim between component 3 and a fourth material, which is light-coloured, flow-banded and charged with microlites of pyroxene and aenigmatite. The various components are mixed at scales between tens of mms and cms. Euhedral to subhedral phenocrysts, sometimes partially resorbed, of alkali feldspar up to 1 mm occur in components 1 and 4.

B355

This sample contains 20% of alkali feldspar, clinopyroxene, olivine, FeTi-oxide and apatite phenocrysts. Alkali feldspar is dominant, forming euhedral to subhedral plates up to 4 mm long. Some contain melt inclusions. Green pyroxene prisms are up to 3.3 mm long. Prismatic olivine grains (≤1.5 mm) are commonly partly resorbed. Equant oxides (≤0.2 mm) tend to occur in clusters with the other mafic phenocrysts. Apatite crystals (≤0.1 mm) are invariably included in olivine and pyroxene. Sulphide crystals also occur in pyroxene. The matrix, composed of the same phases as the phenocrysts plus quartz, is largely hypocrystalline, with grain sizes ranging from microphenocrysts to microlites, and contains rare, small (<20 μm) pools of colourless glass.

B375

This sample contains two, darker and lighter, patchily mingled glassy matrix components of different composition (Fig. 2d). Composed of the same minerals, the darker component is richer in pyroxene microcrysts. The same phenocryst assemblage (15% modally) occurs in both components. It is dominated by subhedral alkali feldspar up to 3 mm in size, commonly forming clusters. Some show shadowy extinction. Prismatic clinopyroxene phenocrysts, up to 1 mm long, are zoned to bright green rims. There are rare olivine, aenigmatite, FeTi-oxide and sulphide (micro)phenocrysts. Although the matrix is almost completely crystalline, pristine glass occurs in two forms: as small (≤20 μm) irregular patches attached to the rims of feldspar phenocrysts and as larger (≤100 μm) pools in the matrix (Fig. 2d). The glass is associated, in the matrix, with alkali feldspar, aegirine-augite, aenigmatite, quartz and tiny grains of ilmenite (Fig. 3d). Small pools of carbonate are scattered in the matrix (Fig. 3c).

Since alkali feldspar is the sole phenocryst phase in B354, it can be assumed that it was the liquidus mineral in all samples. However, the fact that the feldspar may include phenocrysts of the mafic phases suggests that it was joined by them in the crystallizing assemblage. Although many phenocrysts are euhedral and unzoned, some subhedral partly resorbed crystals are found in three of the rocks, consistent with the evidence of magma mingling in those samples.

Mineral chemistry

Alkali feldspar

The total compositional range of the new data (Table 2; Supplementary Table 1) is Or28.1–40.6 Ab58.2–70.6An 0.0–1.9, i.e. compositionally the feldspars straddle the anorthoclase-sanidine boundary (Fig. 4). Calcium levels are very low (An0-2). Additionally, Ronga et al. (2010) recorded phenocryst core compositions of Or19.1Ab80.7An0.3 in B355, and Or15.5Ab81.1An3.5 in B175 (comendite; not analysed here). There is very considerable overlap between specimens but there are small differences in average composition in different rocks and/or different glass in the same rock, e.g. B355 Or30.4 Ab68.1An1.5; B375 light glass Or33.9Ab65.6An0.5, dark glass Or28.9Ab70.3An0.7; B350 light glass Or33.3Ab66.6An0.1, dark glass Or34.6Ab65.4An0.1; B354 dark glass Or36.4Ab63.6An0.0, light glass Or35.5Ab64.5An0.0 (Table 2). The value for B375 (dark glass) is probably anomalous; it refers to data from one crystal in immediate contact with a melt inclusion. Individual crystals show internal variation of up to 2–8 mol% Or; in most cases, the variation is not systematic but slight rimward enrichment in either K or Na is occasionally seen. Areas of high Fe2O3 contents up to 5.08 wt.% occur as either cores (B350) or rims (B375) to some phenocrysts (Table 2); they normally appear brighter on BSE images. So far as we know, these are the highest Fe values yet recorded in feldspars in peralkaline rhyolites, although Troll and Schmincke (2002) have recorded feldspars with up to 4 wt.% Fe2O3 in comendites from Gran Canaria. The most extreme Fe-enrichment occurs in a feldspar phenocryst in B350, where the feldspar contacting a melt inclusion has 8.45 wt.% Fe2O3* (0.29 apfu), decreasing to 0.91 wt.% (0.03 apfu) at ~10 μm from the inclusion. The high-Fe feldspar gives an acceptable formula (Table 2; Supplementary Table 1) and we do not believe that its composition is an artefact of its closeness to the melt inclusion.

Fig. 4
figure 4

Feldspar phenocrysts from Boseti plotted in the An-Ab-Or diagram. The shaded field is for feldspars from the wider sample coverage by Ronga et al. (2010)

Full size image

In the feldspars, the Fe almost exactly balances the Al-deficiency, i.e. (Al + Fe3+) ≈ 1 apfu. The entry of Fe into the feldspar is apparently controlled by composition, particularly by host-rock peralkalinity, and not by such parameters as fO2.

Olivine

Our new data for olivine phenocrysts in B355 (Table 3; Supplementary Table 2) show a very narrow compositional range, Fo1.5–2.4, comparable to the value of Fo1.1 reported by Ronga et al. (2010). The total range in the Boseti rhyolites is Fo10-1, the minerals becoming more Fe-rich with increasing whole-rock peralkalinity (Ronga et al. 2010). MnO contents of olivine are up to 5.71 wt.%, in comendite B175 (Ronga et al. 2010). Although it has been known since the work of Carmichael (1962) that olivine in peralkaline rhyolites is Mn-rich, the Boseti values are, so far as we know, the highest yet recorded. CaO abundances range from ~0.6 wt.% to 0.8 wt.% (0.022–0.027 apfu) and decrease with increasing Fa content of the olivine (c.f. Ronga et al. 2010).

Using average olivine and melt compositions, the melt-liquid exchange distribution coefficients (KD) for Fe-Mg is 0.50. This fits in very well to the overall trend of increasing KD with increasing glass peralkalinity in peralkaline trachytes and rhyolites (Fig. 5). They are also in accord with the observation that olivine KD values vary with bulk rock composition, from ~0.3 in basalts (Roeder and Emslie 1970; Ulmer 1989) to as high as 0.68 in silicic rocks (Sisson and Grove 1993; Kilinc and Gerke 2003).

Fig. 5
figure 5

olivine-meltKDFe-Mg plotted against peralkalinity index of the coexisting glass for peralkaline trachytes and rhyolites. Data sources: Boseti, this paper; Pantelleria, Mahood and Stimac (1990); Olkaria natural rocks, Marshall et al. (2009); Olkaria experimental glasses, Scaillet and Macdonald (2003); Menengai, Macdonald et al. (2011)

Full size image

Clinopyroxene

The compositional range in the clinopyroxene phenocrysts is small, Ca45.8–42.2Mg5.1–0.2Fe49.1–57.5 crystals becoming more Fe-rich with increasing whole-rock peralkalinity (Table 3; Supplementary Table 2). Crystals are usually very homogeneous; occasionally, reversed zoning is seen, e.g. Ca43.3Mg1.2Fe55.5 to Ca45.6Mg2.9Fe51.6 from core to rim in a crystal in B355, whereas Ronga et al. (2010) recorded strong core to rim Fe-enrichment (Ca40.9Mg10.8Fe48.3 to Ca42.3Mg0.5Fe57.3) in a phenocryst from pantellerite B345 (not studied here). Levels of Na2O are low, <2 wt.%, even in the most Fe-rich crystals. Ronga et al. (2010) reported aegirine-augite rims (with Na2O up to 5.2 wt.%) to hedenbergite cores in Boseti pantellerites but, in this study, aegirine-augite and aegirine were found only as matrix phases (Supplementary Table 2). This is in accord with the general situation in peralkaline rhyolites; hedenbergite is the normal pyroxene phenocryst even in the most strongly peralkaline rocks, aegirine-augite being recorded only rarely (Macdonald et al. 2011). This is probably related to the relatively low fO2 at which the magmas evolve, at or close to FMQ (Scaillet and Macdonald 2001, 2003, 2006; White et al. 2005, 2009; Ren et al. 2006; Di Carlo et al. 2010). A microphenocryst in B355 has a relatively magnesian core (Ca44.9Mg21.1Fe34.0), which may represent a higher-temperature phase of magmatic evolution or mixing of the pantellerite with a more trachytic magma.

The extremely low MgO contents in both the pyroxene and glass in strongly peralkaline pantelleritic rocks, commonly around the analytical detection limits, makes it difficult to determine melt-liquid exchange distribution coefficients (KD) for Fe-Mg accurately. Using the average rim compositions of pyroxene phenocrysts and of matrix glass (below) in B355, the Fe-Mg KD is 0.14. This is in line with values determined for other peralkaline trachytes and rhyolites from Pantelleria, Italy (0.12–0.16; Carmichael 1962; Mahood and Stimac 1990; Di Carlo et al. 2010). These KD values are lower than those in mafic and intermediate rocks, which are normally in the range 0.20–0.30 (Sisson and Grove 1993; Putirka et al. 2003).

FeTi-oxides

In peralkaline trachytes and rhyolites, the spinel and rhombohedral phases occur in broadly the same range of rocks, covering the spectrum of whole-rock compositions. It is seldom clear in peralkaline rhyolites what stabilises one oxide over the other (Macdonald et al. 2011). In the Boseti rocks, an oxide phase occurs as phenocrysts only in B355, where it takes two forms: in one, titanomagnetite (X usp 70.4) rims ilmenite (X ilm96.5) and in the second, it is a matrix phase (X usp 66.8) of much later crystallization than the ilmenite (X ilm 94.9–95.6) enclosed in an adjacent olivine. It might appear, therefore, that at Boseti the titanomagnetite is stabilised, relative to ilmenite, in lower temperature, more peralkaline melts. However, tiny rounded grains of ilmenite occur in the matrix of B375 (Fig. 3d) and must have crystallized (although not necessarily in equilibrium) when the residual melt was extremely peralkaline.

Apatite

Analyses of euhedral apatite crystals included in a fayalite phenocryst in B355 are given in Supplementary Table 1. They are fluorapatites with up to ~5% of the britholite-(Ce) component via the substitution scheme Ca2+ + P5+ = REE3+ + Si4+. Published analyses of apatite phenocrysts in peralkaline rocks are scarce. Macdonald et al. (2008a) found up to 35% britholite component in apatites in comendites of the Olkaria complex, Kenya. In contrast, Mahood and Stimac (1990) found lower enrichment in the britholite component in apatites from Pantellerian trachytes and pantellerites, with REE + Si ≤0.5 apfu, and Macdonald et al. (2011) recorded solid solution towards britholite-(Ce) up to 7 mol% in apatites from pantelleritic trachytes of the Menengai volcano, Kenya. The reasons why apatites from the more strongly peralkaline (pantelleritic) rocks are less REE-enriched than those from the more mildly peralkaline (comenditic) rocks, at least as exemplified by the Olkaria suite, are not yet known.

Sulphide

Due to its small size, we were not able to produce quantitative analyses of the sulphides. Semi-quantitative analyses indicate, however, that it is pyrrhotite, which has also been recorded as microphenocrysts in peralkaline rhyolites from Tejeda volcano, Gran Canaria (Crisp and Spera 1987), Pantelleria (Lowenstern et al. 1993; White et al. 2005) and Eburru, Kenya (Ren et al. 2006).

Geochemistry

Analyses of glass matrices and of glass (melt) inclusions in feldspar phenocrysts are given in Table 4 and Supplementary Table 1. Before describing the variations, we assess the possibility that the Na2O contents do not represent the pristine values. Figure 6 shows the relationship between Na2O and FeO* in non-hydrated peralkaline obsidians (Macdonald 1974a). Extrapolation of the relationships suggests that at 16 wt.% FeO*, glasses should contain ~8–9 wt.% Na2O. This is exemplified by the glasses synthesised from an Eburru pantellerite (Scaillet and Macdonald 2006). Many of the Boseti matrix glass analyses fall below the band defined by the obsidians. Four explanations seem possible. First, the Boseti rhyolitic magmas were relatively Na-poor; this possibility is unlikely because the Boseti basalts are actually sodic (Ronga et al. 2010). Second, the glasses have lost variable amounts of Na during microprobe analysis; Morgan and London (2005) have shown that peralkaline glasses with high contents of F and Cl are particularly prone to Na migration, a point made also by Mungall and Martin (1996). Although we were very careful about maximising the probe conditions for Na analysis, we cannot totally preclude some Na loss under the beam. Third, the microprobe beam included microlites. Since feldspar and aegirine-augite are common microlites and both contain Na, this effect is unlikely. Fourth, the glasses have experienced variable degrees of secondary hydration. Three of the whole-rocks of our rocks have LOI between 1.1 wt.% and 2.1 wt.% (Ronga et al. 2010); since most of that is incorporated in the matrix glass, it is possible that Na has been lost by this mechanism. An additional complication is that although melt inclusions in phenocrysts may have avoided secondary hydration, their compositions may have been modified by post-entrapment crystallization, undetected in our BSE images.

Fig. 6
figure 6

Plot of Na2O against FeO*. The shaded field encloses the compositions of 95% of non-hydrated peralkaline silicic obsidians compiled by Macdonald (1974a). The arrowed trend is for the glasses synthesised by Scaillet and Macdonald (2006) in a pantellerite obsidian from the Eburru volcanic complex. The majority of Boseti glasses fall below the obsidian field, suggesting that they have lost some Na relative to their pristine melt compositions. Data from Supplementary Table 1

Full size image

To deal with the problem of Na loss, we use the alternative measure of peralkalinity, FK/A (mol. (Fe+K)/Al, with all Fe calculated as Fe2+), where FeO*, K2O and Al2O3 are considered to be relatively immobile (White et al. 2003). An excellent correlation between PI and FK/A was found for the comenditic obsidians of the Olkaria complex (Marshall et al. 2009) and there is also an excellent correlation between the parameters in the experimental glasses from Eburru (Scaillet and Macdonald 2006). Revised Na2O values for the Boseti glasses, were calculated from the FK/A ratio and are in good agreement (to within 0.5 wt.%) with those estimated from the Na2O-FeO* plot (Fig. 6). The revised values were then used to calculate revised PI, which are given in Table 4.

We obtained one analysis of a melt inclusion and only one satisfactory matrix glass analysis in the largely devitrified B355. The analyses are broadly similar (Table 4), the melt inclusion perhaps being slightly less evolved, with lower SiO2 and higher Al2O3 and CaO contents, consistent with trapping during the earliest stages of crystallization. In B350, the darker and lighter glasses are compositionally different, most notably in the Al2O3 (7.88 and 5.36 wt.%) and FeO* (8.89 and 11.03 wt.%) abundances. Melt inclusions in feldspar phenocrysts from the dark and lighter glass areas are similar and have been averaged together in Table 3; they are transitional in composition between the darker and lighter matrix glasses. Except for CaO, where the averages are 0.42 wt.% (lighter) and 0.20 wt.% (darker), there are no significant differences between the darker and lighter matrix glasses in B354, as shown both by random analyses in each type and by two detailed profiles across the boundary between them (Supplementary Table 1). One profile revealed a point analysis of a glass component which is less evolved (q 19.6%) than the other glasses and the whole-rock composition (q ~30%). A melt inclusion is similar to the matrix glasses. There are also no significant differences between the darker and lighter components in B375. The most extreme glass compositions are those forming small (<100 μm) pools rimming feldspar phenocrysts and in the matrix, the main features being the low Al2O3 (average = ~2.3 wt.%) and high FeO* (average ~16.8 wt.%) contents. The average PI is 7.0, considerably higher than the values for the lowest-temperature experimental glasses synthesised from Eburru (3.7; Scaillet and Macdonald 2006) and Pantelleria (3.2, Di Carlo et al. 2010) pantellerites. The relatively low SiO2 abundance (~65 wt.%) leads to the glass being classified as trachyte in the TAS system (Le Bas et al. 1986), despite a q value of 32.1%, a useful reminder of the limitations of using SiO2 as a differentiation index in peralkaline silicic rocks.

The Boseti glasses are moderately Cl rich, up to 0.5 wt.%. Abundances in the darker glass are slightly higher than in the lighter glass in both B350 and B354 (Table 4). We have only limited F data, which vary from 0.10 to 0.52 wt.% (Table 4). Differences between the light and dark glasses in B350 and B375 are within analytical error. The F content of the residual glass in B375 (0.5 wt.%) is much lower than expected from closed-system fractionation of the host-rock. It is possible that F has been degassed in a volatile phase but the glasses are non-vesicular. Furthermore, F partitions strongly into the melt phase during crystallization of peralkaline rhyolites (Webster et al. 1995; Barclay et al. 1996). Neither has a F-rich phase crystallized, unless we have not recognised fluorite, which has been found as phenocrysts in the Olkaria comendites (Marshall et al. 1998).

Discussion

Significance of glass analyses

Our new glass analyses, and analyses of a range of highly evolved natural and experimental glasses and whole-rocks, are shown on an FeO*-Al2O3 plot in Fig. 7. Tie-lines connect whole-rocks and matrix glasses. Predictably, the matrix glasses are, with one exception, more evolved (higher Fe, lower Al) than their host rock. The exception is B354, where the higher Al2O3 in the glass cannot be explained by crystallization of alkali feldspar, the only phenocryst phase. It may be that the whole-rock composition reported by Ronga et al. (2010) has been wrongly labelled.

Fig. 7
figure 7

Glasses from various suites plotted on an FeO*-Al2O3 diagram. The fields of comenditic trachyte (CT), comendite (C), pantelleritic trachyte (PT) and pantellerite (P) are from Macdonald (1974a). Tie-lines connect matrix or experimental glasses and host whole-rocks. For Boseti rocks with >1 glass component, the values used here are averages weighted by the estimated abundance of each component. Data sources: Boseti, whole-rocks, Ronga et al. (2010), glasses, this paper; Pantelleria, whole-rock and most FeO*-rich experimental glass, Di Carlo et al. (2010, no. 6–7); matrix glasses and melt inclusions from pantellerites of Pico Alto, Terceira, Mungall and Martin (1996); Fantale obsidian, Ethiopia, Lacroix (1930). The shaded field encloses the field of natural pantellerite obsidians, updated from Macdonald (1974a)

Full size image

Our new data (Table 4) confirm the range of glass compositions established by Ronga et al. (2010), particularly the exceptionally low Al2O3 contents and high FeO* contents. Broadly similar glass compositions have been recorded in natural eruptive rocks and in experiments on pantellerites (Fig. 7). Lacroix (1930) presented an analysis of a pantellerite obsidian from the Fantale volcano, Ethiopia, with 6.26 wt.% Al2O3 and 10.75 wt.% FeO*. Melt inclusions in fayalite phenocrysts and intratelluric glass in partially crystallised pantellerites of the Pico Alto volcano, Terceira, Azores, have Al2O3 as low as 5.00 wt.% and FeO up to 13.29 wt.% (Mungall and Martin 1996). The most evolved glass produced experimentally from an Eburru pantellerite had 5.22 wt.% Al2O3 and 12.27 wt.% FeO*, the glass representing 53.8% residual melt (Scaillet and Macdonald 2006). Similarly, Di Carlo et al. (2010) recorded that the most evolved experimental glass from a Pantellerian pantellerite had 4.72 wt.% Al2O3, FeO* 12.8 wt.%, with 30% modal glass.

The fact that natural and experimental glasses broadly converge on a composition around 5 wt.% Al2O3 and 13 wt.% FeO* seems to suggest that they have approached the true near-minimum melt composition of peralkaline oversaturated systems (c.f. Scaillet and Macdonald 2006). In detail, there is a range of compositions; at 5 wt.% Al2O3, FeO* contents range from ~12 wt.% to 14 wt.% (Fig. 7) and SiO2 from 64 wt.% to 68 wt.%. There is probably not, therefore, a unique minimum point, the range reflecting such factors as the point at which quartz begins to crystallize (perhaps controlled by P, pH2O and the F/Cl ratio), and whether magnetite or ilmenite was the stable oxide. The glass in B375 (Al2O3 2.39 wt.%) indicates that even more extreme melt compositions can be reached, albeit in tiny amounts. The ~5 wt.% Al2O3 compositions can then be seen as effective minima, i.e. compositions reached in experiments and occasionally as matrix glasses but which have been erupted as a rock only once, at Fantale volcano, Ethiopia (Lacroix 1930). It is important to stress that these highly evolved compositions are reached by the fractionation of alkali feldspar + fayalite + hedenbergite + oxide ± quartz assemblages, the dominant assemblage in peralkaline trachytes and rhyolites.

The scarcity of such strongly peralkaline melts as eruptive rocks is probably a result of relatively high density caused by their high FeO* and low SiO2 contents. Allied to their low volumes, this inhibits crustal ascent and the melts get trapped at depth (Scaillet and Macdonald 2006). Eruption may require explosive activity; the most differentiated compositions may therefore occur as pyroclastic, especially ash fall, deposits. This has been shown at the Menengai trachyte volcano (Leat et al. 1984) and the Olkaria rhyolitic complex (Marshall et al. 2009) but we are unaware of any published, systematic geochemical study of the fall deposits in a pantellerite centre.

Feldspar-melt relationships

Our aim of describing feldspar-melt relationships in the Boseti rocks is hampered by uncertainties over the Na2O contents of the glasses. Using, however, the Na2O values calculated from the FK/A ratios, some points may usefully be made. The range of Or contents in the phenocrysts (Or30.4–35.5, omitting the anomalous value in B375) approaches the value of Or36 recorded by Di Carlo et al. (2010) as coexisting with the most evolved glass in their experiments on a Pantellerian pantellerite. The range is also close to the minima between the anorthoclase and sanidine solid solution loops at 1 atm. (anhydrous) pressure (Or35; Schairer 1950) and at 1 kb water pressure (Or30; Bowen and Tuttle 1950). It has long been known (Bailey 1974; Roux and Varet 1975) that peralkaline rhyolites with PI up to ~2 are in equilibrium with feldspar of “minimum” composition. The Boseti and experimental Eburru and Pantelleria data indicate that this relationship seems to persist into the most strongly peralkaline compositions, with PI exceeding 3. Furthermore, the matrix feldspar coexisting with glass in B375 averages Or35 (Table 2; Supplementary Table 1); if we can assume that they are in equilibrium, the feldspar has remained at the minimum composition.

In Fig. 8, X is K/(K + Na) and Y is (Si/3Al)-1, representing the degree of silica-enrichment over an alkali feldspar composition (Roux and Varet 1975). Tie-lines connect feldspar phenocrysts and coexisting matrix glass in B350, B354 and B355. The feldspars are more potassic than the glass, demonstrating the orthoclase effect of Bailey and Schairer (1964) whereby crystallization of feldspar results in strong Na-enrichment in residual melts. The important point, demonstrated for the first time in a natural sequence, is that the orthoclase effect operates even when the melt has reached a PI over 3.

Fig. 8
figure 8

Plot of X (K/(K + Na)) against Y ((Si/3Al)-1) (Roux and Varet 1975) to show the relationship between the alkali ratio and silica-enrichment in the Boseti matrix glasses and Or content of the coexisting alkali feldspar phenocrysts. Tie-lines connect matrix glass and feldspar; D and L refer to darker and lighter glasses in mixed rocks

Full size image

The composition of the matrix glass in B375 is also of considerable interest. The feldspar-glass tie-line is at an angle to those in the other rocks, a result of the crystallization of aegirine and aenigmatite prior to glass formation (Fig. 8), muting the Na-enrichment. Whilst the glass points to the trend potentially followed during extreme fractionation of peralkaline rhyolite, we stress that we know of no natural volcanic sequence where compositional variations due to fractionation of aegirine and/or aenigmatite have been convincingly demonstrated (c.f. Macdonald 1974b).

Magma mixing

The occurrence of magma mixing in the Boseti peralkaline rhyolites is most clearly shown by the petrographic and geochemical evidence of mingling outlined earlier (Fig. 2), including: (i) the occurrence in B355 and B175 of feldspar phenocrysts of cores of Or16-19, and in B355 of a relatively magnesian core to a clinopyroxene phenocryst (Ca44.9Mg21.1Fe34.0), values consistent with crystallization from a more mafic trachyte or benmoreite; (ii) the presence of resorbed feldspar and olivine phenocrysts; and (iii) the local presence of glass in B354 less evolved than the bulk-rock. Magma mixing is very common in peralkaline silicic systems, having been recorded at, inter alia, Longonot, Kenya (Scott and Bailey 1986), Menengai, Kenya (Leat et al. 1984), SW Sardinia, Italy (Morra et al. 1994), Pantelleria, Italy (Ferla and Meli 2006), Gedemsa, Ethiopia (Peccerillo et al. 2003), and Olkaria, Kenya (Macdonald et al. 2008b). The process involves 2-, 3- or 4-component mixing of various combinations of basalt, mugearite, trachyte and rhyolite and occurs both in the magma reservoirs and in volcanic conduits. The mixing attests to the open nature of such systems. In noting the occurrence of rhyolite-rhyolite and rhyolite-trachyte/mugearite mixing at Boseti, we suggest that further study will find evidence of a basaltic component in many rocks. We further suggest that the intimate mixing of trachytic and pantelleritic magmas in the complex indicates the existence of compositionally zoned magma reservoirs.