Earnscleuch

A sodic anorthosite xenolith (Lau24) comprises essentially unzoned oligoclase crystals, in a well-equilibrated fabric with triple grain boundaries meeting at angles of ~120°. Sporadic development of melt patches appears to indicate a second original phase, probably a pyroxene. The sample is thus a rather pure, sodic anorthosite lacking magnetite, zircon or apatite.

Feldspar

Reconnaissance electron and ion microprobe data for anorthoclase megacrysts from Elie Ness and a neighbouring vent at Brownhills were presented by Mason et al., (1982). In the feldspars investigated in the present study, veins and/or patches appear to have been secondarily altered. The unaltered (primary) feldspars analysed by ion microprobe for trace elements (Table 2) are all anorthoclases, with the exception of sanidine from Loch Roag. The unaltered areas generally have very constant major and trace element chemistry and only in one sample (Loch Roag, RSM1) was strong zoning observed (sanidine to anorthoclase). Feldspar in GRB117 exhibits some zoning (core to rim) from 15 to 7% An.

There are differences of up to three orders of magnitude between the LREE (specifically, La and Ce) contents of the analysed feldspars (Fig. 4). The feldspars can be subdivided on the basis of their REE, Ba and Sr contents into: (1) a group with high LREE (>20 ppm La) and low Ba and Sr contents (RP6, BBL24 and RSM1); (2) a group with very low La and Ce contents (<1 ppm) and moderate Ba and Sr contents (HB87, CYD10 and CYD13); (3) those with intermediate levels of LREE (1-12 ppm La) and moderate to high contents (>1500 ppm) of Sr and Ba (namely, EN2B, FD86, GRB117 and LR57). It is clear that the xenoliths (RSM1, RP6 and BBL24) containing corundum and Nb oxides are also those that contain the most REE-rich feldspars, although these are distinctly poor in Ba and Sr. The Coalyard Hill (CYD10 and CYD13) and Holmbyre (HB87) feldspars differ from those of the corundum-bearing assemblages in having very low REE contents as well as being very poor in CaO (Fig. 4). Although the analysed feldspars are monomineralic megacrysts, other feldspars from the same vents (with similar major element chemistry) are accompanied by biotite, apatite and calcite (the last possibly secondary after pyroxene).

Figure 4. Chondrite-normalized REE patterns for anorthoclases and sanidines from Scottish megacryst-xenolith assemblages. The three diagrams are arranged in order of decreasing REE content, the most enriched being from corundum-bearing samples at Ruddon's Point (RP6), Loch Roag (RSM1) and Burn-between-the-Laws (BBL24). Ion microprobe data. It was not technically possible on the ion microprobe to determine REE heavier than Eu. Yttrium values are included for reference. Chondrite values used in normalization are from Anders & Grevesse, (1989).

Some outer rims of individual crystals exhibit various degrees of secondary alteration, typically along veins; in general, such alteration can be readily distinguished by an increase in the potassium content of the feldspar from 1-4% K₂O to 7-9% K₂O. Two samples also show replacement of anorthoclase by very pure K-feldspar. The major and trace element compositions of these secondary feldspars are given in Table 2. It is clear, however, that the type of alteration varies between megacrysts and, also, that more than one type of alteration may be present within a single crystal. Cathodoluminescence (CL) studies of BBL24 revealed three different types of feldspar (Fig. 3c and d): the (primary) cores of the feldspar crystals show green luminescence whereas diffuse secondary alteration along cleavage planes and irregular fractures gave rise to feldspar that luminesces (dull) dark blue. The xenolith is traversed (as is commonplace for other xenoliths and megacrysts) by fine-scale veinlets of fairly pure calcite (bright orange CL) containing small quantities of Mn and Fe. Two generations of veinlets are identifiable: earlier, slightly broader veinlets are symmetrically rimmed by feldspar exhibiting bright (pale) blue CL. This `bright blue feldspar' has been generated adjacent to the calcite veinlets from both the dull blue and green luminescing feldspar. All are then traversed by fine-scale late-stage veinlets that are not associated with any alteration of their feldspar side-walls. Both varieties of blue-CL feldspar have notably higher K/Na values than the feldspar luminescing green. The change from green to blue cathodoluminescence is well defined in both major and trace element chemistry. There is a notable increase in K, with concomitant rise in Rb and Cs. An increase in Ba can also be explained, in part, by the increased ease of partitioning of Ba into feldspar as the Or content increases. It should be noted, however, that the (green) host feldspar is unusually poor in Ba. In contrast, CaO decreases from the green to the blue luminescing feldspar, as do Sr, Fe and LREE. The `bright blue' vein feldspar is not readily distinguishable from the dull blue luminescing feldspar in terms of major elements. However, whereas elements which strongly follow K (e.g. Rb and Cs) are unchanged, there is marked increase in Li, B, Mg, Ca, Ti, Sr, Nb and Fe. For some elements there is, therefore, a reversal of the trend observed between green and dull blue luminescing feldspars. Although it is not known whether the increase in trace elements is due to sub-microscopic inclusions, it is clear that the strong change in luminescence is accompanied by a large increase in Fe and Ti.

The origins of feldspar CL have been the subject of much disagreement in the literature. Blue CL has been variously attributed to activations involving Ti (Mariano, 1988), Ga (De St Jorre & Smith, 1988), REEs (Laud et al., 1971) and also to structural defects (Walker, 1985). Unpublished data (D. Stirling & A. Finch, personal communication, 1997) indicate a relationship between the intensity of blue CL in some feldspars and Al content. This may result not from activation, but rather substitution of Al into feldspar, which may promote structural defects, which cause luminescence. Other substitutions may similarly give rise to such defects and hence to blue CL. In the present example, we note that the areas of brightest blue CL are those of feldspar with the greater concentration of dopant ions. Green CL in feldspar is generally considered to arise from a combination of the blue and a yellow excitation. Yellow CL is believed to result from Mn²⁺ activation (Walker, 1985) although we observed no relationship between the Mn content of these feldspars and the intensity of the inferred yellow excitation.

The variations in the BBL24 feldspars contrast to those observed in RP6 (Table 2), where alteration initially involved an increase in Fe and alkaline earths, without changing the bulk chemistry or REE contents. More pervasive alteration led to a marked drop in the alkaline earths, including Ba, with a very large increase in Fe. Whereas La and Ce decrease only slightly, Eu falls substantially. This is consistent with the supposition that Eu exists principally as Eu²⁺ within the feldspar lattice, showing strong behavioural similarity to Sr²⁺.

The trace element chemistry of the Coalyard Hill megacrysts is markedly different from that of the other feldspars studied. Recrystallization along veins has produced pure end-member sanidine with very low concentrations of alkaline earth elements and LREE as well as Fe and Ti. In the K-feldspar vein in CYD13, even Ba is remarkably low in comparison with the primary host feldspar, despite the preference of this element for the Or end-member.

It is clear that the original feldspars were subject to changes after their crystallization and it is inferred that the high K feldspars formed at relatively low temperatures along cracks and grain boundaries. As the alteration is always accompanied by a marked K increase, yet the bulk feldspars are not particularly K rich, there is no reason to suspect that all of the feldspar has been altered or that the extremes in REE content observed between megacrysts are secondary in nature. In brief, we conclude that these compositional modifications occurred in the protoliths before incorporation as xenocrysts in the host basalts.

Corundum

The Loch Roag corundums show asterism as well as fine-scale concentric oscillatory zonation (see Irving, 1986). In thin section, they are colourless except for brown (euhedral) core regions and brownish coloration caused by oriented micro-inclusions of rutile (Frondel, 1954) along radiating zones whose axes are 60° apart (Fig. 5). We infer that these inclusions lie in the basal plane, perpendicular to the second-order prism faces (1120), bisecting the colour zoning at right angles (see Tait, 1959). As Ti₂O₃ and Fe₂O₃ (haematite) are both isostructural with corundum, significant solid solution may be expected. However, ilmenite-corundum systems often demonstrate broad solvi (see Sheikh & Irvine, 1993) as a result of entropy changes associated with ordering of the ilmenite. According to Frondel, (1954), Ti³⁺, substituting for Al³⁺, exsolves from corundum as rutile during slow annealing in an oxidizing environment. In thin section the brownish coloration in the crystals may be due to the presence of both rutile needles and haematite plates exsolved from the corundum. Electron microprobe analysis from core to rim of one crystal (LR1), ~30 mm diameter, reveals a minor variation in Fe/Al ratio which correlates with the large number of fine-scale oscillatory zones. The total Fe content (expressed as `FeO'), ranges from a minimum of 0·91% at the edge to a maximum of 2·08% at the core. Ti (as TiO₂) and MnO contents are generally at, or below, the electron microprobe detection limits but have their highest values (0·093% and 0·019%, respectively) in the Fe-rich core. These variations are supported by the ion probe measurements (Table 3). Corundum can dissolve up to 20% Fe₂O₃ in air at 1 atm PO₂ at 1400°C , falling to ~10% at 1100°C (Muan & Gee, 1955) and it is concluded that the Fe-rich core crystallized at higher temperatures than the remainder of the crystal. No minor element differences could be detected between the colourless and the brownish radiating zones in the LR1 corundum.

Figure 5. Photomicrograph of corundum (LR1) from Loch Roag. The crystal has ~25 mm diameter and shows concentric oscillatory zonation, radiating (brownish) coloured zones and a dark Fe-rich core.

The blue colour in corundum (as viewed megascopically) has been variously linked to trace amounts of Fe²⁺, Fe³⁺, Ti³⁺ and Ti⁴⁺. Ferguson & Fielding, (1971) attributed the colour to Fe²⁺-O-Ti⁴⁺ charge transfer but these conclusions were not supported by Matson & Rossman, (1988), who favoured Fe²⁺-O-Fe³⁺ super-exchange. Our observations that the intensity of the blue colour appears independent of Ti content support a charge transfer mechanism involving Fe alone. The fine-scale oscillatory zoning in the Loch Roag corundums points to crystallization from a melt in which P, T and/or composition were fluctuating during megacryst growth.

Ion microprobe analyses were made of corundums from three localities (samples BBL24, Burn-between-the-Laws; RP6, Ruddon's Point; RSM2, Loch Roag). Representative trace element analyses are presented in Table 3: analyses are selected to indicate the full range in chemical zoning and are given in order of increasing Ti content for each sample. In all four analysed corundums, the principal substituting elements are Fe and Ti. Although up to 300 ppm Si was observed, it is not clear how much may be ascribed to surface contamination. Na and K values may also be too high as a result of surface contamination and probably do not exceed 2 and 4 ppm, respectively.

Cathodoluminescence studies of the BBL24 corundum demonstrate the presence of zonation (not otherwise discernible), with luminescence changing from deep to lighter brown from the cores to the rims (Fig. 3b). Oscillatory zoning is manifest in terms of alternating light and dark CL bands. An ion microprobe traverse was made across one such zone (Fig. 6). Analyses were made by moving the sample beneath the primary beam, in 10 µm steps. At each step a single measurement was made of the secondary ion counts at masses 9 (Be), 13·5 (Al)²⁺, 48 (Ti), 56 (Fe) and 93 (Nb). Count rates, normalized to Al²⁺, were converted to concentrations by reference to point analyses made in the core of the crystal. The broad change from darker to lighter brown CL colouring correlates with the decrease in Fe from centre outwards. Whereas the Fe content is highest in the core, the rim also shows notable Fe and Ti enrichment. Although Be (<3 ppm) follows Ti closely, Nb variations are even more extreme (although still tending to correlate with Ti) and there is notable Nb deficiency in the outer rim. We conclude that the incorporation of individual elements changed during growth, either because of concentration changes in the melt from which the corundum was growing, or because of changes of the physical conditions attending growth. Whereas Fe³⁺ can fit within the corundum structure, the Ti⁴⁺ does not appear to be balanced by any coupled substitution. The poor fit of Ti within the corundum lattice at lower temperatures and pressures can also be inferred from the exsolution of fine needles of rutile in some crystals.

Figure 6. Ion microprobe traverse of corundum crystal from Burn-between-the-Laws sample BBL24 showing variation in Fe, Ti, Nb and Be.

The spatial resolution of the ion probe beam was insufficient to resolve the fine-scale oscillatory zoning observed in CL. A second, higher spatial resolution, traverse was made along the same line by moving the sample in 2 µm (rather than 10 µm) steps beneath the primary beam and using a field aperture to select only ions from the central few microns of the sputtered area. Although the losses in secondary ion transmission with increase in spatial resolution were such that only Ti and Fe intensities could be measured, a clear correlation was observed between the CL oscillatory zones and Ti (and hence also Nb and Be) concentrations. The brightest CL correlated with the lowest Ti zones. It is accordingly inferred that the CL is quenched by the presence of Fe and Ti and that the intrinsic luminescence is due to other factors.

In contrast to the feldspar crystals, however, which are generally uniform in major and trace element composition (when not subject to later alteration), the corundums provide evidence for rapidly changing physico-chemical conditions during growth. The presence of disrupted growth zoning and the rounded morphology of many of the crystals suggest that the corundums formed early and were partially resorbed before the crystallization of the feldspar.

CL of corundum may be due to several causes and the brown luminescence, rather than a single colour from the visible spectrum, hints at two or more distinct emissions. Activation of corundum CL by transition metal ions such as Ti³⁺ is known (Reisfeld et al., 1987), but a variety of defect centres in corundum are recognized (e.g. Kotomin et al., 1994) and these may also relate to luminescence. Ti³⁺-doped alumina has been identified for laser applications and is characterized by emissions at 410 and 740 nm. Ion probe analysis demonstrates that the alternating light and dark brown CL zones correlate with changes in Ti content, suggesting some form of Ti-related activation mechanism. Emission in the violet or blue part of the visible spectrum is clearly not present, and a complex activation mechanism, possibly involving multivalent Ti and/or charge transfer to Fe, may be responsible for the observed CL.

Apatite

The feldspar megacrysts occasionally contain inclusions of magnetite and apatite (e.g. FD86). Furthermore, some xenoliths show anorthoclase in an equilibrated textural relationship with apatite, magnetite and clinopyroxene (e.g. samples LR57 and GRB117). These observations prompted the analysis of apatite and magnetite in samples FD305 (apatite-magnetite xenolith, Fidra) and MLL6 (apatite-magnetite-salite xenolith, Gribun). Data are presented in Table 4.

The compositions of two of the apatites (FD86 and GRB117) are virtually identical and the associated feldspars are also remarkably similar (Table 2). It may consequently be inferred that, although FD86 is a virtually monomineralic anorthoclase megacryst, it was derived from a feldspar-pyroxene-apatite-magnetite rock similar to the FD86 xenolith. The apatites of samples LR57, MLL6 and FD305 have similar REE contents (Table 4; Fig. 7) and are, accordingly, likely to have come from similar sources. This conclusion is supported by the compositional similarity of the pyroxenes in MLL6 and FD305. The two Gribun apatites show wide variation in absolute abundances, with GRB117 having a ca. * 14 enrichment in LREE relative to MLL6. A less extreme, but still wide variation in values is shown by the Fidra apatites (FD86 and FD305). Values for the Loch Roag apatite are commensurate with those given by Menzies & Halliday, (1988). The majority of the samples analysed have REE contents notably higher than the apatite megacrysts reported by Irving & Frey, (1984).

Figure 7. Chondrite-normalized REE patterns for apatites from Fidra (FD), Gribun (MLL and GRB) and Loch Roag (LR). Chondrite values for normalization from Anders & Grevesse, (1989).

Clinopyroxene

Compositions of the most highly evolved B-Group pyroxenes from Scotland are presented in Table 1. Whereas there are no precise geobarometric indicators, the relatively jadeitic compositions of some (Al₂O₃ up to 10%, Na₂O >2%) point to a high-pressure paragenesis in the 5-15 kbar range (Thompson, 1974). As noted above, the pyroxenes in LR57 and MLL6 have similar major element compositions, with relatively high Al₂O₃ contents (6-9 wt %). Although the absolute REE contents differ by a factor of ~2, the overall normalized patterns are comparable (Fig. 8). In contrast, whereas the absolute LREE concentrations in the GRB117 pyroxene resemble those of LR57, the former shows a more sinusoidal pattern. Rather less complete REE data for a pyroxene megacryst from Loch Roag were given by Menzies & Halliday, (1988). The REE abundances for the analysed Loch Roag and Gribun clinopyroxenes are notably higher than those of other pyroxene megacrysts quoted by Irving & Frey, (1984).

Figure 8. Chondrite-normalized REE patterns for Group B megacryst-xenolith pyroxenes from Gribun (MLL and GRB) and Loch Roag (LR). Chondrite values for normalization from Anders & Grevesse, (1989).

Nb- and Nb-REE-rich oxides

The Nb-rich oxides (columbite, ilmenorutile, yttro-niobate and samarskite) occur exclusively in the peraluminous assemblages. Electron microprobe analyses of ilmenorutile, yttro-niobate and samarskite were given by Aspen et al., (1990). Samarskite has ~34% Nb₂O₅, 20% ThO₂ and 22% REE₂O₃. Further data for samarskite were presented by Hinton & Upton, (1991) and new analyses are given in Table 5, together with analyses of columbite. Nb-REE-rich oxides, analogous to yttro-niobate and samarskite, are known from inclusions in basalt-hosted corundums around the world (Guo et al., 1996b). Whereas the three analysed samples (Table 5) have varying proportions of Nb, REE and Th, the chondrite-normalized REE patterns (Fig. 9) are distinctive in having La Ce and marked negative Eu anomalies.

Figure 9. Chondrite-normalized REE patterns for yttro-niobate (RP6), samarskite (RSM1) and columbite (BBL24). Chondrite values for normalization from Anders & Grevesse, (1989).

Guo et al., (1996b) noted that of all the inclusions recorded from basalt-hosted corundums, Nb-Ta oxides are the most abundant and persistent. They pointed out that the columbites have a distinctive composition, with high TiO₂ (up to 17 wt %) and low Ta₂O₅ (typically <4%) comparable with columbites from carbonatites. However, the columbites in this study (BBL24) have TiO₂ ~3·4% with Ta₂O₅ from 3·5 to 4·9%.

Corundum-bearing parageneses

Coenraads et al., (1990) listed numerous localities world-wide where sapphire occurs in volcanic terrains. Syngenetic inclusions (e.g. U-pyrochlore, thorite and columbite), as well as CO₂-filled inclusions, indicate that the sapphires grew in environments rich in incompatible elements and volatiles. Dating the zircons in the sapphires showed them to be neo-contemporaneous with the basalt magmatism. The `growth-zoning', CO₂-rich fluid inclusions, and high inclusion homogenization temperatures (>650°C) in sapphires from Queensland led Stephenson, (1976) and Irving, (1986) to infer high temperatures and high pressures for their origin. Coenraads et al., (1990) concluded that the sapphires grew from highly evolved magmas in the deep crust or upper mantle. Guo et al., (1996b) concurred that the sapphires grew from highly fractionated silicate-rich melts but proposed a mid-crustal provenance. Sutherland et al., (1998), however, argued for a lithospheric mantle origin.

Observations on the Loch Roag and Burn-between-the-Laws corundums demonstrate clearly that they grew in dynamic environments that gave rise to complex oscillatory zoning and episodic resorption. Rhythmic growth zoning in zircons from Elie Ness suggests fluctuations in REE, Th and U content of the parent melt (Hinton & Upton, 1991). Together with the oxygen isotope data and the association with geochemically extreme mineral phases, the evidence denies a meta-sedimentary origin for the peraluminous xenoliths, as has been proposed for some kimberlite xenolith suites (e.g. Mazzone & Haggerty, 1989). The peraluminous trachytic melts from which the Scottish corundum syenite xenoliths are inferred to have crystallized are likely to have had SiO₂ contents in the region of 60%, total alkalis approximating 10% and Al₂O₃ contents probably in excess of 20%.

Whereas xenolith-megacryst studies are normally handicapped by the inability to observe `field relationships', occasional tectonic slices of uplifted mantle may provide pertinent information. Possibly an analogy to the relationships under discussion is provided by an occurrence of `hornblende peridotite' in Plumas County, California, which is cut by a 5 m dyke of feldspathic rock (`plumasite') comprising 83·6% oligoclase and 16·4% pale blue-violet corundum (Barlow, 1915). The modal proportions are very similar to those of the anorthoclase-corundum xenolith (BBL24). The half-metre thick zircon syenite pegmatitic dykes in the (mantle tectonite) phlogopite peridotites of the Finero mafic-ultramafic complex in the Ivrea Zone of the Alps (Stahle et al., 1990) may also be analogous to the megacryst source rocks. These dykes have high Nb, Th, U and LREE, and are inferred to have crystallized from mantle-derived volatile-rich partial melts (Stahle et al., 1990). In the same region an REE-rich alvikite carbonatite occurs containing inclusions of alkali feldspars and syenitic metasomites.

Subdivision of the assemblages and inferences concerning parent melts

It is clear that large variations exist in the trace element chemistry of individual feldspars and this must reflect differences in the chemistry of the melt from which they formed, the trace element partitioning related to their major element chemistry, or temperature-pressure considerations. Thus although, in the extreme cases, there is a difference of a factor of 1000 in the La content between individual feldspars, observed relationships within each group suggest that a factor of 5-10 may be attributed to partitioning changes related to CaO content. It is unlikely that the remaining variation of ~200 in the La content for feldspars with similar CaO content can be attributed to either changes in melt major element chemistry or P, T conditions and must reflect variations in the REE content of the melt. The trace element subdivision of the feldspars referred to above can be used to subdivide the megacryst assemblages as a whole into three groups.

Group 1

The majority of the Group 1 samples are essentially monomineralic anorthoclases with CaO contents of <0·8%. They have moderate Sr and Ba contents and calculated melt compositions of ~100 ppm Sr and 250-700 ppm Ba (see Appendix for details of partition coefficients used). Although the REE abundances are 10-20 times lower than in Group 2 (see below and Fig. 10) the similarity of LREE patterns in the feldspars (Fig. 4) suggests comparable LREE-HREE values in the melt.

Figure 10. Estimated chondrite-normalized REE patterns for melts equilibrated with a selection of apatites and clinopyroxenes from the megacryst suites. Chondrite values for normalization from Anders & Grevesse, (1989).

Group 2

These feldspars are associated with clinopyroxene and/or apatite and generally have higher CaO contents (1·4-3 wt %). The coexisting melts must have been enriched in Sr (150-300 ppm) and highly enriched in Ba (900-1800 ppm). Overall, the melts calculated using clinopyroxene and apatite (Fig. 10) give similar patterns of LREE enrichment (La/Yb 40-80) and flat, to gently sloping, HREE patterns (see Appendix for details of partition coefficients used). Whereas no Eu anomaly is observed in the calculated melt pattern, there is a significant positive Eu anomaly in the feldspar; the melt must have been sufficiently oxidizing to limit the abundance of Eu²⁺. Although the overall degree of REE enrichment calculated is dependent on the choice of partition coefficient for pyroxene, it is within a factor of two of the calculated REE budget for GR117 (based on observed mineral proportions). The Zr and Nb contents of the pyroxenes are consistent with relatively high whole-rock Zr and Nb contents of 100-500 ppm.

Possible role of carbonatite melts

Long et al., (1994) presented data on the Loch Roag megacrysts and pointed out that, as some of the pyroxenes and micas have mg-numbers too high to have been in equilibrium with a trachytic melt (pace Aspen et al., 1990) and that they equilibrated with a more basic melt, the Loch Roag megacrysts must be polygenetic. Considering coexisting carbonatite-silicate melts, Long et al., (1994) noted [from consideration of Hamilton et al., (1989)], that enrichment of the carbonate fraction in REE is favoured by high pressures and increased polymerization of the silicate melt. In the shallow upper mantle, the REE would preferentially concentrate (by a factor of 10 or more) into the carbonatite fraction of any conjugate melt pair. Long et al., (1994) commented on the similarity between the parental melts inferred for the Loch Roag megacrysts and carbonatite melts, and noted that the enrichment in Ba, K, Nb, LREE and Sr characterizing the pyroxenites points to possible influx of carbonatitic melts that may have been genetically related to the basaltic magmatism that hosted the xenoliths and megacrysts. In contrast to the on-craton Loch Roag situation, however, Long et al., (1994) inferred that the melts which coexisted with the Scottish off-craton megacrysts were not so markedly enriched in LREE. Whereas this may generally be true, our data show that the extreme LREE enrichment exhibited by the feldspars associated with the Loch Roag corundums is actually surpassed by those in the corundum-bearing xenoliths at Ruddon's Point and Burn-between-the-Laws in SE Scotland, remote from the craton.

According to Guo et al. (1996b), the compositions of columbite and U-pyrochlore in basalt-hosted corundum megacrysts suggest a source of carbonatitic character. Those workers concluded that the corundum genesis involved interaction of carbonatitic magma with alkali syenite or granite pegmatite compositions (magmas or rocks) at mid-crustal levels and temperatures of 400°C. However, although conceding that the evidence from these Scottish assemblages (particularly, but not exclusively, the peraluminous cases), supports the participation of carbonatitic melts, our data indicate shallow mantle sources at temperatures in excess of 700°C (Aspen et al., 1990), in agreement with the corundum crystallization temperatures of 685-900°C proposed by Sutherland et al., (1998).

A carbonatite association is consistent with the low content of divalent cations, including Mg, Sr, Ba and probably, Eu. Ca may be increased because the overwhelming concentration of Ca is such that even high partitioning of Ca into the melt will lead to a relatively high content of Ca in the silicate. The REE may be similarly enriched but with a noticeable change in the La/Ce ratio in the most REE-rich cases consistent with high La vs Ce fractionation into the carbonatite. A carbonatite would exclude Al, Ti, Zr and Th, increasing the elements in the silicate-oxide phases. Hf is taken into the silicate in preference to Zr (Fielding, 1992) providing a possible explanation of the high Hf content of the RSM1 zircons. Si, Nb and Ta will partition strongly into the silicate but this partitioning is less extreme than that for Al. U, P, together with Ba would preferentially stay in the carbonatite, thus offering possible explanation for the preponderance of phases with relatively high Th/U values. However, if carbonatite melts were involved in the genesis of the megacrysts then carbonate was subsequently wholly lost to the system.

Several workers (e.g. Meen, 1987; Green & Wallace, 1988; Wallace & Green, 1988; Lee & Wyllie, 1998) have concluded that reaction between carbonated nephelinite or carbonatite and mantle harzburgite or lherzolite produces clinopyroxene-rich wall rocks (wehrlite or olivine pyroxenite). Lee & Wyllie, (1998) noted that, at depths <70 km, wehrlite can coexist with carbonatite melts relatively enriched in Ca/Mg. Long et al., (1994) commented on the similarity between the parental melts inferred for the Loch Roag megacrysts and carbonatite melts, and noted that the enrichment in Ba, K, Nb, LREE and Sr characterizing the pyroxenites points to possible influx of carbonatitic melts. CO₂ degassing of the mantle was held by Menzies & Wass, (1983) to be responsible for migration of REE-rich fluids and localized transformation of lherzolite to pyroxenites veined by apatite ± kaersutite ± biotite ± diopside. More recently, Baker et al., (1998) concluded that amphibole ± apatite-bearing lherzolite xenoliths from Yemen have characteristics consistent with enrichment of shallow mantle by carbonatite melts and hydrous fluids.

Relationship to glasses in mantle xenoliths

The paragenesis of the megacryst associations may have some relationship to the formation of some of the salic and alkalic glass patches reported from mantle xenoliths. Compositions of such glasses vary widely [see Chazot et al., (1996) and references therein], with SiO₂ 52-65%, Al₂O₃ 18-23%, Na₂O 3-10% and K₂O 0·3-2·3%. Although the origins of these glasses may be diverse, Schiano et al., (1994) and Eiler et al., (1993) concluded that some glass inclusions represent small amounts of metasomatic melts with compositions dramatically different from those of the host magmas, originating at depth as part of an exotic migrating phase in the lithosphere. Draper & Green, (1997) proposed a two-stage model to account for the Si-Al-alkali-rich melts corresponding to mantle xenolith glasses. Stage 1 involves enrichment of the mantle (at ~45-90 km depth) in low-melting components by percolating alkali mafic liquids whereas Stage 2 involves subsequent low-degree partial melting to yield the liquids ultimately trapped as xenolith glasses. The calculated REE patterns from the apatite and pyroxene megacrysts (Fig. 10) are comparable with those described by Schiano et al., (1994) for mantle glasses. A chondrite-normalized La concentration of 200-400 falls steadily to ~10-20 at Yb and there is no Eu anomaly. Although the major element chemistry of the glasses is similar to the likely composition of the melts from which the pyroxene megacrysts crystallized, the CaO content is, however, too high for many of the samples studied.

Salic melts in the shallow mantle

Evidence that alkaline salic magmas can be produced within the sub-continental lithospheric mantle was reviewed by Aspen et al., (1990), who deduced that volatile fluxes interacting with the lithospheric mantle led to small-fraction melts from which the highly evolved, feldspathic rocks crystallized. That these melts were trachytic rather than benmoreitic (see Guo et al., 1996a) is evident from the high Fe/Mg nature of associated ferromagnesian silicate and oxide phases and the high (Na + K)/Ca of most of the feldspars. These melts appear to have had alkali:silica proportions buffered so that neither silica-oversaturated nor feldspathoidal compositions have yet been encountered. The most extremely fractionated(peraluminous) melts yielded alkali feldspar rocks with accompanying oxides but apparently lacking ferromagnesian silicates.

A petrogenetic problem arises in that the megacrysts and xenoliths studied indicate that the parental melts were, in some cases, peraluminous. The concept of melts yielding corundum being peralkaline (see Guo et al., 1996b) is clearly untenable. Normal fractional crystallization processes will not give rise to peraluminous residues [other than in calc-alkaline environments; see Cawthorn et al., (1976)]. Accordingly, if assimilation of aluminous (metasedimentary) material is excluded, the peraluminous condition may have been brought about by alkali loss. Whereas the melts parental to the B-group megacrysts must all have been relatively highly evolved [low mg-number, high (Na₂O + K₂O)/CaO, high Zr, P, REE], those from which the corundum-bearing assemblages formed were the most extreme (see Guo et al., 1996b) and their inferred very high LREE, Nb, Ta, U and Th concentrations suggest the involvement of carbonatite. The calcite veining (e.g. in xenolith BBL24), as well as the remarkable carbonate-filled tubules described from some anorthoclase megacrysts (Aspen et al., 1990) may point to the separation of a conjugate carbonatite melt from a carbonate-rich trachyte magma. If so, alkali loss leaving a peraluminous residuum may have been brought about by such a fugitive carbonatite melt. Sutherland et al., (1999) maintained that melts generated from amphibole-bearing mantle sources can produce corundum in Nb- and Ta-enriched residual liquids.