Age and magnetic polarity of samples from the Mull igneous centre

The first radiometric age data for Mull lavas were obtained by Evans, (1969), R. D. Beckinsale [unpublished data, reported by Mussett et al., (1972)] and Mussett et al., (1972). Samples from Ardtun and Bunessan, southwest Mull, gave K-Ar ages of 61·2 ± 2·0 Ma, 60·9 ± 2·0 Ma and 59·7 ± 2·0 Ma (1[sgr] errors), respectively, whereas a suite of lavas from Gribun in western Mull gave K-Ar ages ranging from 63 Ma to 46 Ma. The Gribun lavas appear to have suffered variable degrees of alteration and argon loss, which probably accounts for the wide range of ages. Basic dykes from the northwest coast of Mull also were dated by K-Ar methods (Evans, 1969; Mussett et al., 1972). These rocks gave ages ranging from 62 Ma to 45 Ma, again reflecting argon redistribution associated with partial alteration of the samples.

More recently, Mussett, (1986) dated 20 whole-rock samples from Mull using ⁴⁰Ar/³⁹Ar incremental heating methods. Data obtained for 10 of these samples did not meet the acceptance criteria of Lanphere & Dalrymple, (1978) but data for 10 other samples were interpreted as reliable estimates of the true crystallization ages. The Mull lavas gave ⁴⁰Ar/³⁹Ar ages ranging from 61·1 ± 1·4 Ma to 59·4 ± 0·9 Ma (1[sgr] errors), consistent with the K-Ar ages obtained for Ardtun and Bunessan lavas. In addition, several acidic rocks were analysed. The Loch Uisg granophyre (Centre 1) gave an age of 58·1 ± 1·6 Ma, and the Beinn a' Ghraig granophyre and Loch Bà ring dyke gave ages of 57·0 ± 0·8 and 56·5 ± 1·0 Ma (1[sgr] errors), respectively.

A recent palynological study of interlava sediments from Tertiary Hebridean lavas (Bell & Jolley, 1997) placed the initiation of volcanic activity on Mull at 55 Ma, i.e. ~4 my younger than the ⁴⁰Ar/³⁹Ar ages of the lavas. However, as pointed out by Kerr & Kent, (1998) the reliable ⁴⁰Ar/³⁹Ar ages (Mussett, 1986) for the Mull lava succession were not discussed by Bell & Jolley, (1997). Moreover, new ⁴⁰Ar/³⁹Ar dating studies (J. G. Fitton & L. Chambers, personal communication, 1997) of the Mull lava succession have confirmed (within error) the ages previously determined by Mussett, (1986). It would thus appear that the palynological age for the inception of the Mull lavas is in error, and more work is required before this technique can be used as a reliable dating technique for the Tertiary Hebridean lava successions.

The ⁴⁰Ar/³⁹Ar ages outlined above suggest eruption of the lava sequence and emplacement of the intrusive centre over a fairly short period of time, perhaps 4-5 my, beginning at ~61 Ma. This is broadly consistent with the results of palaeomagnetic studies of Mull lavas and dykes (e.g. Ade-Hall et al., 1972; Hall et al., 1977; Mussett et al., 1980; Dagley et al., 1987). The palaeomagnetic data suggest a polarity sequence of R-N-R or R-N-R-N-R, the latter possibly being the result of thermal overprinting (Dagley et al., 1987). The R-N-R-N-R sequence could correspond to chrons 26R-26-25R-25-24R on the timescale of Berggren et al., (1995), i.e. the period from ~61 Ma to ~55 Ma.

Lava succession

Bailey et al., (1924) divided the Mull lavas into two main types, the `Plateau magma type' and the `Non-porphyritic Central type', on the basis of a few chemical analyses and a multitude of petrographic observations. These two magma types were subsequently recognized world-wide and became the alkaline and tholeiitic types, respectively (Tilley, 1950; Wager, 1956). On Mull, this nomenclature has stood the test of time, with the Plateau magma type becoming the Mull Plateau Group (MPG, Morrison et al., 1980) and the Non-porphyritic Central type becoming the Central Mull Tholeiites (CMT, Kerr, 1995b). In recent years, a flow-by-flow chemical study (Kerr, 1993a, 1995b) of 600 lavas from the Mull succession has identified a third main type of lava on Mull, the Coire Gorm (CG) magma type (Figs 3 and 4).

Figure 3. Plot of total alkalis against silica for the Mull Plateau Group (MPG), Coire Gorm magma type (CG) and the Central Mull Tholeiite type (CMT). Also shown are fields for the Skye Main Lava Series (SMLS) and the Skye Preshal More (PM) type [from Thompson et al., (1972)] along with fields for the Antrim lava succession (ALS) and the Causeway tholeiites (CT) [from Lyle, (1980, , 1988) and Lyle & Patton, (1989)]. Classification boundaries are from Le Bas & Streckeisen, (1991).

Figure 4. (a-d) Plots of major and trace elements against F/(F + M) [Fe₂O₃*/(Fe₂O₃* + MgO), where * denotes total iron] for the three magma types of the Mull Tertiary lava succession. Also shown are 1 atm fractionation trends modelled using the TRACE 3 program of Nielsen, (1988) [see Kerr, (1993a) for details]. (e, f) Plots of (Sm/Yb)_cn vs (La/Nd)_cn and Ce/Y vs Ti/Zr for the Mull Plateau Group (MPG), Coire Gorm (CG) and Central Mull Tholeiite (CMT) magma types. The full geochemical data set used in this paper is available on the Journal of Petrology world wide web pages (http://www.oup.co.uk/petroj/hdb/Volume_40/Issue_06/dataset/).

It should be emphasized that the three groups of Mull lavas identified by Beckinsale et al., (1978) on the basis of 18 lava samples do not correspond to the three main magma types noted above. These workers proposed that each of their groups was derived by different degrees of partial melting of a lithospheric mantle source, with no crustal input and only minimal fractional crystallization. This interpretation now appears outmoded; extensive geochemical studies of the Mull lavas have suggested that the source of the lavas was asthenospheric mantle, and that the bulk of Mull magmas were variably contaminated by Lewisian (and/or Moine) crustal rocks before eruption (e.g. Morrison et al., 1980; Thompson et al., 1982, , 1986; Kerr, 1994, 1995a, 1995b; Kerr et al., 1995). Thus, Group 1 of Beckinsale et al., (1978) has been identified by subsequent workers as uncontaminated Plateau Group lavas, the Beckinsale et al. Group 2 as Plateau Group lavas contaminated with Moine schist and their Group 3 as Plateau Group lavas contaminated with Lewisian gneiss.

The sequence of Tertiary lavas exposed on Mull can be summarized as follows. The Mull Plateau Group forms the lower ~730 m of the lava pile, and crops out throughout northern and western Mull [see Fig. 1 and Kerr, (1995a) for further details]. The Coire Gorm lavas (at least 250 m thick) stratigraphically overlie the Plateau Group lavas and are exposed only at the top of Ben More. Several of the earliest Coire Gorm lavas are intercalated with the final more evolved lava flows (trachytes) of the Plateau Group. The Central Mull Tholeiite lavas are at least 900 m thick (Skelhorn et al., 1969) and occur in southeastern Mull, and within the Beinn Chaisgidle caldera (Centre 2) (Fig. 1). As lavas of the Central Mull Tholeiite type are not found within the main lava succession on Ben More, it is reasonable to assume that they were erupted after the Coire Gorm lavas.

The geochemical differences between the three main Mull lava types are illustrated in Figs 3 and 4 (see also Table 2). The element ratios, (Sm/Yb)_cn, Ce/Y, La/Nd and Ti/Zr, were selected because they are not affected significantly by crustal contamination (although higher La/Nd ratios may be caused by contamination). In general, variations on these diagrams mainly reflect differences in the degree of partial melting or in the nature of the mantle source region [as well as the effects of low-pressure fractional crystallization in the case of falling F/(F + M) and very low Ti/Zr values]. The Plateau Group lavas represent a transitional tholeiitic-alkalic series, ranging in composition from high-magnesium (14 wt % MgO) basalt to trachyte [Figs 3 and 4a-d and Kerr, (1993a)]. The basalt and hawaiites of the Plateau Group mostly have Ti/Zr ratios of <105 and (Sm/Yb)_cn ranging from 1·6 to 3·5 (Fig. 4e and f). The Coire Gorm basalts have major element compositions that are similar to those of the Plateau Group basalts, but possess flatter rare earth element (REE) patterns and have Ti/Zr > 100 (Fig. 4). The Central Mull Tholeiite basalts have generally higher CaO, and lower abundances of incompatible trace elements, when compared with the Plateau Group and Coire Gorm lavas (Fig. 4a-d), (La/Sm)_cn ranges from 0·6 to 1·5, (Sm/Yb)_cn < 1·9 and Ti/Zr ratios of >100 in the Central Mull Tholeiites (Fig. 4e and f).

Table 2. Representative chemical analyses of the Mull lava succession

The different trace element ratios noted above suggest a change in the depth of partial melting with time, from deep, garnet-present melting (Plateau Group lavas) to shallow melting with little residual garnet in the source (Coire Gorm and Central Mull Tholeiite lavas). A similar observation has been noted for the Skye Tertiary dyke swarm (Mattey et al., 1977) and lava succession (Wood, 1979; Thompson et al., 1980; Ellam, 1992), the early Tertiary Rockall Trough and Hatton Bank lavas (Morton et al., 1988) and in basalts from East Greenland (Fram & Lesher, 1993). On Mull, this change was interpreted by Kerr, (1994, 1995b) to indicate partial melting below progressively thinning lithosphere as the magmatic system developed.

Subdivision of the Mull Plateau Group lava sequence

Kerr, (1995a) subdivided the Plateau Group lavas into nine chemically distinctive sub-types (Tables 2 and 3). These sub-types are believed to reflect modification of the primary mantle-derived magmas by processes such as fractional crystallization, crustal assimilation and contamination by lithospheric mantle rocks. The earliest Plateau Group sub-type to be erupted, known as the Staffa magma sub-type (I; Table 3), appears to be extensively contaminated by Moine and Lewisian metasediments (Thompson et al., 1986; Kerr, 1999). Some of the later, more-alkalic high-magnesium basalts (IV) and basaltic hawaiites (VII) have assimilated small amounts of silicic crust (<5% contaminant; Kerr et al., 1995) from the Lewisian basement. However, not all of the high-magnesium basalts are crustally contaminated; a sub-type (V) with ~13-14 wt % MgO and low Ba/Nb ratios (<15) has also been found (Table 3). These lavas are inferred to provide the closest approximation to the composition of the primitive asthenospheric melts (Kerr, 1995b). The more evolved basalts and hawaiites (VI and VII) are generally less contaminated with Lewisian crust (Ba/Nb <25; [epsilon]_Nd > + 5) when compared with the high-magnesium basalts (Kerr et al., 1995). These, and the mugearites (VIII), benmoreites and trachytes (IX) appear to be related to the least contaminated Plateau Group basalts (V) by simple fractional crystallization (e.g. the modelled trends in Fig. 4a-d). Similarly, the basalts, intermediate rocks and trachytes belonging to the Skye Main Lava Series are related by simple fractional crystallization (Thompson et al., 1980).

Table 3. Distinctive chemical features of magma sub-types found within the lavas of the Mull Plateau Group

Cone sheets

Each of the three igneous centres on Mull is associated with a separate suite of cone sheets (Fig. 2), which range in composition from basic to acidic (Table 4; Bailey et al., 1924; Thomson, 1986). Bailey et al. subdivided the cone sheets into three broad groups: early basic cone sheets (EBCS), early acidic-to-intermediate cone sheets (EACS) and late basic cone sheets (LBCS), which are essentially ferrodioritic in composition. Subsequent work by Thomson, (1986) revealed that the picture is more complicated than this: Centre 1 cone sheets comprise early basic and early intermediate-to-acidic varieties; Centre 2 cone sheets comprise both early and late, basic and intermediate-to-acidic varieties; Centre 3 cone sheets comprise late basic and late intermediate-to-acidic varieties. Cone sheets from all three intrusive centres preserve evidence of magma mixing [see below and Thomson, (1986)]. For the purposes of this study, we have used 4 wt % MgO to divide the more basic cone sheets from the more acidic varieties.

Table 4. Representative geochemical analyses of cone sheets associated with the Mull central complex

Basic cone sheets

There are significant field and petrographic differences between the EBCS and LBCS. The EBCS generally are coarse grained and contain olivine (often pseudomorphed). Individual cone sheets are generally thicker than those of the LBCS, which contain no olivine and are generally fine grained. The two suites also differ with respect to their chemistry (Table 4; Thomson, 1986). Figure 5a and c reveals that at equivalent MgO contents, the LBCS possess levels of incompatible trace elements that are slightly higher than those measured in the EBCS. This is particularly evident in the concentrations of the REE (Table 4 and Fig. 5c). Plots of incompatible elements and isotopic ratios (Fig. 6) and major elements against MgO (Fig. 7) also serve to illustrate these differences. Thomson, (1986) concluded that the differences in trace element contents between the EBCS and LBCS are not related to varying degrees of crustal contamination. She suggested that the two cone sheet suites represent different degrees of partial melting, or are related by fractional crystallization.

Figure 5. Primitive mantle- and chondrite-normalized (Sun & McDonough, 1989) multi-element and REE plots showing data for: (a) early and late basic cone sheets and basic Mull Plateau Group lavas, all with F/(F + M) = 0·68-0·75; (b) early and late acidic cone sheets and acidic Mull Plateau Group lavas, all with MgO = 3·0-2·5 wt %; (c) early and late basic cone sheets, the Beinn Buie Gabbro and the Central Mull Tholeiite and Coire Gorm lavas-also shown are Plateau Group lavas with F/(F + M) values of 0·55-0·80; (d) early and late acidic cone sheets and Mull Plateau Group lavas with <3·5 wt % MgO. Data sources: Mull lavas-see caption to Fig. 4; cone sheets -Thomson, (1986); Beinn Buie Gabbro -Skelhorn et al., (1979).

Figure 6. (a-d) Incompatible trace element ratio and Sr-Nd isotope plots showing data for the Mull cone sheets (Thomson, 1986), and fields for the lavas (see caption to Fig. 4 for data sources) and Mull granites (Walsh et al., 1979). Also shown is a field for the Loch Scridain sills (Preston et al., 1998), which belong to the Mull early cone sheet series. The field for acidic Lewisian gneiss is taken from Hamilton et al., (1979), Moorbath & Thompson, (1980), Dickin, (1981) and Thompson et al., (1982). Other Lewisian data are from Kerr et al., (1995). The field for Moine schists is taken from Preston et al., (1998). DM in (b) represents depleted asthenospheric mantle. For clarity, fields have been drawn round the symbols for the early and late granites in (c) and (d). All the given symbol legends apply to all four diagrams in this figure.

Figure 7. (a-d) Plots of selected major elements and ⁸⁷Sr/⁸⁶Sr_(t) vs MgO for the Mull cone sheets and granites (symbols as in Fig. 6), with fields for the Mull lava types, the Loch Bà ring dyke and the granophyres or intermediate rocks and gabbros of the Glen More (GM) ring dyke. Also shown in (b) is `QFM - 4', a 1 atm fractionation trend modelled by TRACE 3 (Nielsen, 1988) with the fO₂ set at 4 log units below the QFM (quartz-fayalite-magnetite) buffer. Data for the Loch Bà ring dyke are from Sparks, (1988). Data for the Glen More ring dyke are from Seedhouse, (1994); other data sources are given in captions to Figs 4- 6.

One of the most notable features of the LBCS, in comparison with the EBCS, is a marked enrichment in Fe, Ti and P (Fig. 7), as a result of the late onset of Fe-Ti oxide and apatite fractionation. Thomson, (1986) attributed this to crystallization of the LBCS under conditions of low fO₂, when compared with the EBCS. In this regard, it is noteworthy that some of the Tertiary plugs in north Mull and Morvern (Kerr, 1997), together with the youngest of the basic dykes (see below), display compositions that are similar to those of the LBCS.

On a primitive mantle-normalized multi-element plot (Fig. 5a), basic cone sheets from Mull show strong negative Nb anomalies. Elsewhere in the Hebrides, such anomalies in basalts are inferred to reflect crustal contamination, particularly when associated with low ²⁰⁶Pb/²⁰⁴Pb and high ⁸⁷Sr/⁸⁶Sr ratios (e.g. Thompson et al., 1982, , 1986; Kerr et al., 1995). In the Mull cone sheets, Nb contents have been reduced by assimilation of crustal rocks; however, the large-ion lithophile elements (Rb, Ba, K, Sr) show substantial increases over their original concentrations in these rocks, resulting in greatly exaggerated Nb anomalies. Crustal influences are seen also in the Nd-Sr isotopic compositions of the Mull cone sheets (Thomson, 1986). For example, Fig. 6b shows that these rocks possess high ⁸⁷Sr/⁸⁶Sr and low ¹⁴³Nd/¹⁴⁴Nd ratios relative to depleted mantle compositions. Given these features, Moine metasediments appear to represent the most likely contaminants of these cone sheets (Fig. 6b). This idea is supported by studies of cone sheets at Loch Scridain (Fig. 2; Kille et al., 1986; Thompson et al., 1986; Preston & Bell, 1997; Preston et al., 1998), which belong to the EBCS and are probably related to Centre 1 magmatism. The Loch Scridain cone sheets generally have ¹⁴³Nd/¹⁴⁴Nd ~0·5119 (Preston et al., 1998), comparable with Nd isotope ratios measured for Moine rocks [see Fig. 6b, and Thompson et al., (1986) and Thomson, (1986)]. These cone sheets also contain numerous partially melted xenoliths of Moine metasediments (Kille et al., 1986; Preston et al., 1998).

Major gabbroic and granitic intrusions

The major intrusions of the Mull central igneous complex mostly take the form of ring dykes and stocks (Bailey et al., 1924). With four exceptions [the Ben Buie and Corra-bheinn layered gabbros, and the An Cruachan and Gaodhail augite diorites; Fig. 2 and Skelhorn et al., (1969, , 1979)], the major intrusions are granitic or are mixed-magma bodies containing a significant granitic component. Only a very small amount of whole-rock geochemical data are currently available for the major basic intrusions, and these come from the chilled margin of the Ben Buie gabbro, the last major intrusion of Centre 1 [Skelhorn et al., (1979) and Table 5]. Samples of this gabbro have high-Ca tholeiitic compositions and flat to LREE-depleted chondrite-normalized REE patterns (Fig. 5c). The Ben Buie gabbro is thus compositionally similar to some of the EBCS and the youngest preserved lavas. On the basis of these characteristics, these rocks are considered to belong to the Central Mull Tholeiite magma type (see above).

Table 5. Representative geochemical analyses of major intrusions of the Mull central complex

Detailed petrological studies by Pankhurst et al., (1978) and Walsh et al., (1979) showed that the Mull granitoids can be divided into two compositionally distinct groups: those of Centre 1 (`early granites') and those belonging to Centres 2 and 3 (`late granites'). The compositions of the early and late Mull granites are shown in Figs 6c-d, 7d, 8 and 9, and Table 5. The late granites have high K₂O, Na₂O, Al₂O₃, MnO, Zr, Y and Nb, and low CaO, P₂O₅, TiO₂, Sr, V, (Ce/Yb)_cn and initial ⁸⁷Sr/⁸⁶Sr relative to values for the early granites (Walsh et al., 1979). A significant overlap and obvious trends between the compositional fields of the granites and the acidic cone sheets of Centres 1 and 2 (e.g. Figs 7- 9) suggest that the early and late granites are related to their respective acidic (and basic) cone sheet magmas by fractional crystallization with variable amounts of contamination by Moinian and/or Lewisian crust. Centre 1 acid magmas appear to be the most contaminated by crust (highest initial ⁸⁷Sr/⁸⁶Sr; Fig. 7d), whereas Centre 2 magmas may contain a smaller proportion of crust (lowest ⁸⁷Sr/⁸⁶Sr; Fig. 7d) (Walsh et al., 1979) than both Centres 1 and 3.

Figure 8. (a-d) Plots of selected major and trace elements vs SiO₂ for the acidic Mull cone sheets and granites (data sources and symbols as in Fig. 6).

Figure 9. (a) Total alkali-silica diagram showing the same data as in Fig. 8a-d, in addition to the early and late basic cone sheets, a field for the Mull Plateau Group lavas and the alkalic-tholeiitic divide [after Miyashiro, (1978)]. (b) Primitive mantle-normalized (Sun & McDonough, 1989) multi-element plot comparing the compositions of the early granites and early acidic cone sheets with the late granites and late acidic cone sheets. [Data sources and symbols for both (a) and (b) are as in Fig. 6.]

Petrographic observations and geochemical modelling suggest that the crystallization sequence for the early magma suite (basic and acidic cone sheets and granites, predominantly from Centre 1) initially involved plagioclase and pyroxene, with small amounts of olivine. As crystallization continued, Fe-Ti oxide(s) joined the fractionating assemblage at ~4 wt % MgO, followed by apatite at ~3 wt % MgO. The point at which Fe-Ti oxide(s) and apatite started to fractionate is somewhat difficult to estimate from Fig. 7a and b because of extensive mixing between basic and acidic members of the early magma suite (Thomson, 1986). In the late magma suite, comprising most of the Centre 2 and Centre 3 rocks, the onset of Fe-Ti oxide and apatite fractionation was suppressed until ~3 wt % MgO and ~2 wt % MgO, respectively. Therefore, because plagioclase and clinopyroxene were the main fractionating phases over a wide compositional interval in the late acidic suite, these cone sheets and granites have abundances of Sr, CaO (Fig. 8) and V (not shown) that are lower than those of equivalent rocks from Centre 1. At the point of apatite and Fe-Ti oxide fractionation in the late acidic magmas, the residual melt was highly enriched in P, Ti, V and Fe, such that these phases were able to form a significant proportion of the fractionating assemblage. Consequently, relative to the early magma suite, removal of these elements from the melt was more efficient. The net result is that the late granites of Centres 2 and 3 have generally low concentrations of P₂O₅, TiO₂, Fe₂O₃* and V when compared with those of Centre 1 (Walsh et al., 1979).

The late granites of Centre 3 are, by and large, the most evolved of the three centres (e.g. Figs 8 and 9). Low Ba and K₂O in some of the more evolved rocks suggest that the parent magmas may have begun to fractionate alkali feldspar. The most fractionated granite of all (74·6-76·7 wt % SiO₂; Walsh et al., 1979) is a large dyke (Toll Doire; Fig. 2) associated with the main portion of the Beinn a' Ghraig granophyre. One of these Toll Doire granite samples has a Zr content significantly lower than that of other Toll Doire samples (Fig. 8c), and at such high degrees of differentiation this may be indicative of zircon fractionation.

Figure 9a shows the slightly more alkaline nature of the late cone sheets and late granites, with many of the samples plotting above the alkali-tholeiite divide; this is in contrast to the early cone sheets and associated granites, which almost all plot below the divide. This is consonant with higher incompatible trace element contents (at equivalent MgO contents) in the late cone sheet and granite series as opposed to the early series (Figs 5 and 8). These features suggest that the parental magmas of the late series were derived either from a more enriched mantle source region or by smaller degrees of mantle melting.

Major mixed-magma intrusions

Glen More ring dyke

The Glen More ring dyke (Fig. 2) is the youngest intrusion in Mull Centre 2. At its western termination the ring dyke consists of coarse- to medium-grained olivine gabbro passing upwards into quartz dolerite, diorite, granodiorite, and granophyre (Bailey et al., 1924; Bor, 1951). Although poorly exposed, boundaries between these units appear to be gradational. Evidence of magma mixing occurs in the form of acidic patches, veins and enclaves in the gabbro and diorite, and sub-rounded clots and xenoliths of intermediate (ferrodioritic) composition in the granophyre [for details of field and petrographic relationships, see Bailey et al., (1924)].

H. H. Thomas and E. B. Bailey [in Bailey et al., (1924)] proposed that the Glen More ring dyke formed by in situ differentiation, resulting in the segregation and upward filtration of residual acidic material through the solidified gabbro-dolerite [see also Koomans & Kuenen, (1938)]. This suggestion was refuted on petrographic grounds by Holmes, (1931) and Fenner, (1937), who believed that the ring dyke originated by mixing between intermediate and acidic magmas. Bor, (1951) offered a compromise solution, in which fractionation of a gabbroic mineral assemblage produced a residual liquid of intermediate composition; subsequently, this magma became hybridized by mixing with acidic liquid. The acidic melt was considered to have formed by differentiation of a batch of basaltic magma intruded after the gabbro-dolerite body had crystallized. Field observations by Marshall, (1984) and least-squares modelling of whole-rock and clinopyroxene compositions (Seedhouse, 1994) broadly support Bor's hypothesis.

The first major study on the Glen More ring dyke since Bor, (1951) was carried out by Seedhouse, (1994). Thirty samples for major and trace element analysis (see Table 6) were collected from throughout the height of the ring dyke. Chemically, the gabbros and diorites from the ring dyke possess compositions which are similar to those of the LBCS, in that they display a tendency towards high Fe₂O₃*, TiO₂ and P₂O₅ and are slightly more enriched in incompatible trace elements when compared with the EBCS (Fig. 7a-c; Table 6). On first consideration, it appears that the parent magmas of the Glen More ring dyke were more magnesian than the most mafic representatives of the LBCS. Seedhouse, (1994) has shown that olivines from the most mafic gabbro (sample GM13; Table 6) possess maximum forsterite contents of 72, which would have been in equilibrium with a magma containing 5-6 wt % MgO. Seedhouse proposed that the more mafic gabbros of the Glen More ring dyke contain cumulus olivine and augite, and that the parent magma of the gabbros and diorites was akin to the composition of sample MT3 (Table 6), which contains 5·4 wt % MgO. Fractionation of olivine, augite and plagioclase from this parental magma produced the diorites and the cumulate gabbros (Fig. 7). On primitive mantle-normalized multi-element plots (Fig. 10), the ring dyke samples have high Rb/Nb ratios (comparable with those of the LBCS) and are most probably contaminated with Moine metasediments. As for the LBCS, fractionation of Fe-Ti oxides and apatite was suppressed until <4 wt % MgO and ~2 wt % MgO, respectively. Figure 10b-c shows the increasing influence of plagioclase, apatite and Fe-Ti oxide fractionation, in the form of negative Sr, P and Ti anomalies, with increasing degree of magmatic evolution. The granophyres of the Glen More ring dyke are, not surprisingly, very similar in composition to the other granites of Centres 2 and 3 (see Fig. 10b). It is probable that these granophyric magmas were derived from a parental magma similar to the Glen More diorite, in a deeper, probably older magma chamber (Seedhouse, 1994).

Table 6. Representative major and trace element analysis of gabbros, granophyres and intermediate rocks from the Glen More Ring dyke

Figure 10. Primitive mantle-normalized (Sun & McDonough, 1989) multi-element plots showing: (a) data for the Glen More ring dyke gabbros, with a field for the late basic cone sheets (LBCS); (b) the range of compositions for Glen More ring dyke granophyres, with fields for the late granites and late acidic cone sheets; (c) granophyres, intermediate rocks and gabbros of the Glen More ring dyke. Data sources as in Figs 6 and 7.

It can be seen from Fig. 7 that the granophyric and more intermediate rocks plot along rather tightly constrained trends. Three lines of evidence suggest that magma mixing is the most likely explanation for these trends (Seedhouse, 1994): (1) the upper part of the granophyre has a chilled margin of the same composition, indicating that the granophyric magma was intruded into the ring dyke and did not form by in situ differentiation; (2) the intermediate rocks have a blotchy heterogeneous nature; (3) the margins of clinopyroxene and plagioclase crystals in diorite enclaves display resorption textures where they are in contact with the granophyres. Mixing has occurred between a residual dioritic magma and a separate granophyric magma intruded from below. The blotchy appearance of the intermediate rocks means that mixing has not been complete or vigorous. Possible reasons for this, and the mechanics of magma mixing in relation to the Glen More and Loch Bà Ring dykes will be discussed below.

Basaltic dykes

The Mull regional dyke swarm is one of the largest in the Hebrides, extending southeastward across southern Scotland and northern England (e.g. Tyrrell, 1917; Holmes & Harwood, 1929), and northwestward to the Outer Isles (Jehu & Craig, 1925). On Mull, the structure, distribution and emplacement style of the dykes was studied by Sloan, (1971) and Jolly & Sanderson, (1995). These workers showed that the mean thickness and spacing of dykes increase with distance from the central intrusive complex, consistent with a decrease in the amount of crustal extension (~21% to 2%). Mafic dykes occur in all parts of the swarm, but felsic dykes (felsites, quartz porphyries, trachytes) are largely confined to the area close to the intrusive complex. On the basis of field relationships, it is thought that intrusion of the dykes occurred throughout the period of Tertiary igneous activity.

The petrography and compositions of Mull-related dykes have been studied by Lamacraft, (1978), Thompson, (1982) and Kent, (1995; see also Table 7). The majority of the dykes analysed belong to the Plateau Group and Central Mull Tholeiite magma types, with a smaller number of Coire Gorm-type dykes (Table 7). The Plateau Group dykes generally have low MgO, Ni and Cr contents (typically 4-6 wt %, <100 ppm and <60 ppm, respectively), and high abundances of Fe₂O₃* and TiO₂ (up to 17 wt % and up to 3·4 wt %, respectively) (Fig. 11). With the exception of hawaiite samples, most of these dykes are depleted in Nb relative to primitive mantle. Coire Gorm-type dykes have MgO ~6-7 wt %, moderate Fe₂O₃* (12-14 wt %) (Fig. 11), and show relative depletion in Nb. The Central Mull Tholeiite-type dykes include high-magnesium samples (up to 11 wt % MgO; Kent, 1995), but usually are moderately evolved (7-8 wt % MgO, up to 250 ppm Ni and up to 550 ppm Cr) (Fig. 11). All analysed Central Mull Tholeiite-type dykes have <4 ppm Nb (Table 7).

Table 7. Representative geochemical analyses of basic dykes associated with the Mull Tertiary igneous centre

Figure 11. (a-b) Plots of Ni and Fe₂O₃* vs mg-number showing the compositions of dykes ([closed square]) exposed along the northern coast of Mull (Table 7 and R. W. Kent, unpublished data, 1994), compared with fields for the Mull lavas (Kerr, 1993a). A few Central Mull Tholeiite lavas possess Fe₂O₃* lower than that of the rest of the Mull lavas and dykes. It should be noted that Fig. 7a shows that several EBCS possess similarly low Fe₂O₃* contents at equivalent mg-number (see text for more details). (c) Nb vs Zr for the north Mull dykes and a subset of 100 basic Mull Plateau Group lavas. Both suites of rocks have been analysed for high-precision Nb at the University of Edinburgh (Plateau Group lavas courtesy of J. G. Fitton). (d) Plot of log (Ti/Y)_cn vs log (Zr/Y)_cn showing that the three magma types found within the Mull lava succession also exist within the dyke swarm of north Mull. `Haw' denotes hawaiite samples.

Chemical analyses of Mull-related Tertiary dykes in southern Scotland and northern England, including the Cleveland en echelon dyke suite, reveal that they too are of Central Mull Tholeiite type, i.e. tholeiitic basalts exhibiting enrichment in the LREE and with essentially flat chondrite-normalized heavy REE (HREE) patterns (Macdonald et al., 1988). These dykes possess high initial ⁸⁷Sr/⁸⁶Sr ratios (up to 0·7123; Moorbath & Thompson, 1980) and appear to be extensively contaminated by Moine metasediments. Thus, they are compositionally very similar to the Mull EBCS, and may even have been derived from the same magma reservoir. The lateral injection of ~85 km³ (Macdonald et al., 1988) of basaltic magma for up to 400 km from the Mull centre may be linked to the collapse of the Centre 1 caldera. This suggestion is supported by calculations showing that the volume of magma that would need to be removed from an underlying chamber to accommodate the collapse of the Centre 1 caldera is approximately the same as that inferred for the Cleveland dyke suite [see Macdonald et al., (1988)].

Plugs of north Mull and Morvern

In a recent study, Kerr, (1997) assessed the geochemistry and petrogenesis of the previously little-studied plugs of north Mull and Morvern. Field evidence suggests that the plugs may have acted as feeder conduits for the lava succession; moreover, the compositions of several of the plugs closely resemble the Central Mull Tholeiite and Plateau Group lavas. Despite substantial contact metamorphic aureoles, several Mull plugs have compositions that are not preserved in the lava succession. Certain of these plugs are low-Fe benmoreites and trachytes (Kerr, 1997) resembling evolved lavas from Skye (Thompson et al., 1972). The low-Fe plugs also are similar in composition to the Mull early acidic cone sheets (see above). Like these, the plugs are probably fractionates of a Central Mull Tholeiite-type magma.

Most of the more basic plugs (Table 8) resemble rocks of the Beinn Dearg More magma type (Thompson & Morrison, 1988), represented by a group of incompatible trace element enriched, late-stage dykes on Skye (Fig. 12). Relative to the Beinn Dearg More magma type, the more basic Mull plugs are less enriched in incompatible trace elements and less alkaline. In addition, these Mull plugs show a tendency towards Fe enrichment, not observed in the Beinn Dearg More dykes (Fig 12b). The trace element enriched character (relative to the Central Mull Tholeiite magma type; Fig. 12) of these more mafic Mull plugs is, however, reminiscent of that of the Late Mull magma type-as represented by the LBCS and the late Mull dykes (see previous section on cone sheets). Kerr, (1997) speculated that lavas with compositions similar to those of the Late Mull magma type plugs were erupted, but were near the top of the lava succession, and so have been eroded away.

Table 8. Representative chemical analyses of plugs associated with the Mull igneous complex

Figure 12. (a) Primitive mantle-normalized multi-element plot (Sun & McDonough, 1989) showing data for the late dykes (Late Mull magma type) cutting the Loch Bà ring dyke (Table 7), fields for basic plugs from Mull (Kerr, 1997), the Mull late basic cone sheets (Thomson, 1986) and the Beinn Dearg More (BDM) dykes from Skye (Thompson & Morrison, 1988). (b and c) Plots of Fe₂O₃* vs Zr and Ti/Zr vs Ba/Nb for the Mull late dykes and plugs and the Skye Beinn Dearg More dykes. Also shown are fields for the early and late basic cone sheets from Mull.

Some of the Late Mull magma type plugs have MgO contents well in excess of 10 wt % (Table 8). These rocksare enriched in Ba (e.g. Ba/Nb >25; Fig. 12a) and appear to have assimilated some Lewisian crust. The magmas supplying the high-magnesium plugs probably came through the crust well away from the large upper-crustal magma chambers of the Mull central intrusive complex, thus avoiding excessive fractionation and contamination by upper-crustal (Moine) rocks. This means that they are closer in composition to their parental magmas when compared with the LBCS, which as we have shown earlier are the more-evolved and contaminated representatives of the Late Mull magma type.

Dykes cutting the Loch Bà ring dyke

During mapping of the Loch Bà intrusion (Figs 1 and 2), E. B. Wright and J. E. Richey (unpublished field slips, dated 1910-1913) noted the occurrence of basaltic dykes cutting the ring dyke along its northwestern edge close to Loch Bà. These dykes represent the final pulse of basaltic magmatism on Mull. Discussed briefly by Bailey et al., (1924), the geochemistry of these late dykes has not been studied previously.

We have collected and analysed a representative set of samples, major and trace element data for which are given in Table 7 and Fig. 12. Two samples are low-alkali, high-Ca tholeiites belonging to the Central Mull Tholeiite magma type. The remaining samples have low to moderate MgO (3·6-6·7 wt %) and are enriched in incompatible trace elements relative to Central Mull Tholeiite-type basalts. These basaltic dykes can be classified as belonging to the Late Mull magma type (Fig. 12). In common with the LBCS, the Late Mull magma type dykes show an Fe-Ti enrichment trend, with two samples (LB7 and LB8) possessing >20 wt % Fe₂O₃* (Fig. 12) and >3·3 wt % TiO₂. These Late Mull magma type dykes have slightly high Ba/Nb (19-52) relative to sample LB6 (Central Mull Tholeiite type), suggesting that they have assimilated small amounts of Lewisian crust (Fig. 12c).

Magma mixing processes

The deeply dissected Tertiary volcanoes of the Hebrides form a classic area in which to study composite intrusions. Such bodies are believed to represent quenched magmatic systems in which intimate, but incomplete, mixing between physically and chemically diverse magma types has occurred (Harker, 1904; Skelhorn, 1959). Field and petrographic evidence suggests that the most primitive mixing end-member in many Hebridean composite igneous intrusions is a ferrodioritic liquid, derived from a parental tholeiitic magma by fractional crystallization processes (e.g. Harker, 1904; Bell, 1983; Marshall & Sparks, 1984; Sparks, 1988).

Two distinct types of magma mixing are observed within the igneous rocks of the Mull Tertiary volcano. First, there is what we term `observable magma mingling', i.e. clear field evidence of mixing, such as convolute or crenulated contacts between silicic and mafic rocks, or ragged inclusions and pillows of mafic and/or intermediate material within a silicic host. This type of mixing is exemplified by the Loch Bà and Glen More ring dykes. Similar bodies are found in several other British Tertiary igneous centres, notably St Kilda, Skye and Ardnamurchan. They can take the form of either net-veined intrusions, such as the Mullach Sgar complex on St Kilda (Marshall & Sparks, 1984), or mixed-magma bodies, such as the Glamaig granite and Marsco summit gabbro on Skye (Thompson, 1968, , 1980; Bell, 1983). Second, there is what can be described as near-complete or cryptic magma mixing (hybridization), where the only evidence of magma mixing comes from whole-rock composition or from petrographic evidence in the form of disequilibrium phenocrysts or partially resorbed and regrown phenocrysts. Petrographic evidence from the Plateau Group lavas, in the form of resorbed and regrown plagioclase phenocrysts, suggests that cryptic magma mixing was also an important process early in the evolution of the Mull igneous centre (Kerr, 1998). In the discussion below, we focus on the EBCS, which display chemical and petrographic evidence of magma mixing. In contrast to observable magma mingling, the `mafic' end-member in the EBCS is not of ferrodioritic composition, but rather is a moderately evolved basalt with 7-8 wt % MgO (Fig. 7). The silicic end-member is a diorite (Fig. 7- 9), i.e. a composition less evolved than that suggested to give rise to observable magma mingling (e.g. Bell, 1983).

In the following section, we explore possible explanations for the two styles of magma mixing identified within the Mull Tertiary igneous complex. We address two related questions:

(1) Why do the early cone sheets (Centre 1 and early Centre 2) display much greater evidence for cryptic magma mixing than the late cone sheets (late Centre 2 and Centre 3)?

(2) What were the principal factors that determined the various styles of magma mixing on Mull?

Aside from evidence of cryptic magma mixing, the Mull early cone sheets are more contaminated with upper-crustal rocks and extend to compositions that are more mafic or more evolved than those of the late cone sheets (Figs 7- 9). These two features are probably related, because the initially more mafic nature of the early cone sheet magmas may have permitted them to assimilate a larger volume of the country rock than was possible for the late cone sheet magmas [see Campbell, (1985) and Huppert & Sparks, (1985)]. Additionally, a high heat flux into the chamber from which the early cone sheet magmas were derived could have contributed to the efficiency of the magma mixing process. In the Mull late cone sheets, there appears to have been some mixing between magmas lying at either side of the apex on a plot of MgO vs TiO₂ (Fig. 7). In these rocks, mixing occurred between magmas with concentrations of TiO₂ ~1 wt % higher (at equivalent MgO contents) than those of the early cone sheets. The major element compositions of these mixing end-members were not appreciably different from one another.

In their modelling of the Mullach Sgar complex on St Kilda, Sparks & Marshall, (1986) identified a zone in TiO₂-MgO space (at high proportions of silicic magma relative to mafic) where mixing will not occur (Fig. 13). The Mull late cone sheets and the Loch Bà and Glen More ring dykes largely fit this model, but the Mull early cone sheets display a complete mixing spectrum right across the zone of `prohibited' mixing proposed by Sparks & Marshall, (1986) (Fig. 13). Figure 14, a plot of density vs MgO, illustrates why this might be the case. Some of the most mafic early cone sheets lie in a density minimum [when olivine has been joined on the liquid line of descent by plagioclase or clinopyroxene; see Stolper & Walker, (1980)]. With increasing fractionation of a plagioclase- and pyroxene-dominated assemblage, the density of a residual magma will increase to a maximum just before the onset of Fe-Ti oxide fractionation, after which the density drops sharply. Figure 13 shows two mixing zones: one for the magma mingling in the Glen More and Loch Bà ring dykes (Zone 1) and the other for the Mull early cone sheets (Zone 2). It is noticeable that the overall gradient for Zone 2 is much gentler than that for Zone 1. This suggests that the higher density of Hebridean Fe-Ti-rich magmas acted as a substantial thermodynamic barrier to mixing, in contrast to mixing between more silicic end-members and basalts lying at the density minimum. Although the viscosity contrast between the more evolved Fe-Ti-rich magma and the silicic magma will be less than that between the more mafic, density minimum, basalt and the silicic magma, relative density appears to have been the most important factor in determining the intensity and efficiency of mixing [see Oldenburg et al., (1989)]. In addition, we observe that the silicic end-member for the early cone sheet mixed magmas is less evolved (1-2 wt % MgO; Fig. 13b), and so less viscous and more dense, than the silicic end-member proposed by Sparks & Marshall, (1986) for magma mixing at Mullach Sgar, St Kilda.

Figure 13. (a) Plot of TiO₂ vs MgO showing data for the Mull cone sheets (Thomson, 1986), together with a tholeiitic liquid line of descent and a zone where mixing is `prohibited' [both from Sparks & Marshall, (1986)]. It should be noted that a significant number of the early cone sheets lie within this zone (see text for further discussion). (b) Plot of TiO₂ vs MgO showing the cone sheets plus data for the Loch Bà and Glen More Ring dykes (data sources as in Fig. 7). Two mixing zones are shown. Observable magma mixing occurs in Zone 1, whereas cryptic magma mixing occurs in Zone 2 (see main text).

Figure 14. Plot of density (g/cm³) vs MgO showing the Mull cone sheets (Thomson, 1986) together with a density trend calculated for the Mull Plateau Group lava samples of Kerr, (1993a). Densities were calculated using the method of Bottinga & Weill, (1970).

It is significant that magma mingling involving an Fe-Ti-rich magma is very commonly preserved in the ring dykes of the British Tertiary province (Sparks & Marshall, 1986). These ring dykes have formed as a result of the collapse of a central caldera into an underlying zoned magma chamber, followed by the intrusion of magma along the ring dyke fracture. Thus, mixing has been physically forced upon the system by caldera collapse and, in the case of Loch Bà, exsolution of gas, which was possibly caused either by pressure release or influx of a denser magma (see Thompson, 1980; Sparks, 1988).

In our discussion of the Mull-Morvern plugs, we noted that they preserve compositions more primitive (up to 11 wt % MgO) than those of the LBCS (up to 5·5 wt % MgO). We reasoned that this was because they were emplaced well away from the density filtration trap of the Mull central igneous complex. A similar explanation can be advanced as a reason why the EBCS extend to ~10 wt % MgO. In the early stages of activity in the central complex, the magmatic plumbing system may have been relatively simple, with few magma chambers. Thus, it would have been relatively easy for density-minimum basalts to ascend from Moho or lower-crustal magma chambers (where they had fractionated from primitive olivine tholeiites) into high-level magma reservoirs. As the plumbing system developed, more magma chambers (i.e. traps) would have developed in the crust, making it increasingly difficult for density-minimum basalts to pass into the upper crust without undergoing significant further fractionation (to form Fe-Ti-rich magmas).

In summary, the emphasis placed by previous workers (e.g. Bell, 1983; Marshall & Sparks, 1984; Sparks & Marshall, 1986) on the importance of magma mixing in the petrogenesis of Hebridean Fe-Ti-rich intermediate magmas appears unjustified. Rather than being fundamental to the mixing process, the high density of Fe-Ti-rich end-member magmas may actually have inhibited their complete hybridization with more silicic liquids. Therefore, intrusive bodies displaying observable magma mingling (see above) are `frozen' examples of a very ineffective magma mixing process, which has been severely hindered by the large density contrast between an Fe-Ti-rich magma and a more silicic magma. In contrast, mixing between density-minimum basalts and diorites was more effective and complete because of the smaller density contrast between the two magmas (Figs 13 and 14).

Origin of Fe-Ti-P enriched intermediate magmas

The crystallization products of high-Fe-Ti-P intermediate magmas are found within the Plateau Group lava sequence and the late cone sheets of Mull Centres 2 and 3 (e.g. Bailey et al., 1924; Thomson, 1986). In this section, we focus on the processes that may have given rise to these rocks.

One of the main factors in controlling the onset of Fe-Ti oxide crystallization is oxygen fugacity (fO₂.) At relatively high fO₂ values (more oxidizing) Fe-Ti oxide will precipitate at higher MgO contents than when the magma has a lower fO₂ (more reducing). However, an explanation based solely on fO₂ arguments cannot explain the concomitant enrichment in P₂O₅ (as well as Fe and Ti), and so other factors must also play a role in determining the saturation point of Fe-Ti oxide components in the magma.

The existence of Fe (and Ti) enrichment trends in tholeiitic magma series has been recognized for some time (e.g. Fenner, 1929). More recently, it has been noted that P is enriched in many high-Fe-Ti magmas, suggesting a link between Fe-Ti enrichment and P enrichment (see, e.g. Juster et al., 1989; Toplis et al., 1994). Phosphorus is highly soluble in basaltic melts (Watson, 1979; Harrison & Watson, 1984), and due to the high electronegativity of the P⁵⁺ ion, forms very stable complexes with 3+ ions, particularly Fe³⁺ (Toplis & Carroll, 1994). Complexing between P and Fe³⁺ ties up these elements, resulting in a delay in the onset of apatite and Fe-Ti oxide fractionation. Critically, however, this complexing does not change the Fe³⁺/Fe²⁺ ratio of the melt.

We noted above that in the Mull LBCS, apatite fractionation commences at values of MgO lower than those at which Fe-Ti oxide is precipitated (i.e. 2·5 wt % as opposed to 3·0 wt % MgO; Fig. 7). If the onset of Fe-Ti oxide fractionation is a result of the breakdown of P-Fe³⁺ complexes, why does apatite fractionation not commence at the same time as that of Fe-Ti oxide? The reason may be that P forms bonds with other trivalent ions in the melt, for example, the REE and Al³⁺. Alternatively, the onset of Fe-Ti oxide fractionation may simply represent the build-up of free Fe³⁺ in the melt as a result of fractionation of olivine and pyroxene (which contain more Fe²⁺ than Fe³⁺). In this situation, the P-Fe³⁺ complexes may not break down until the commencement of apatite fractionation.

Following Juster et al., (1989) and others, Toplis & Carroll, (1994) have noted several effects of increasing P concentrations in a basaltic magma. In the context of Mull, the most important of these is the dissolution of olivine, which will reduce the Fe₂O₃/FeO ratio of the melt. Olivine dissolution can occur either by reaction with O₂ to form magnetite, or by reaction with SiO₂ (in the liquid phase) to form pigeonite (e.g. Juster et al., 1989; Toplis & Carroll, 1995, , 1996). The absence of pigeonite in the Plateau Group and Late Mull magma types is not only a reflection of their alkaline nature, but suggests also that the first reaction may be more significant in basaltic rocks of the Mull Tertiary igneous centre. It is important to note that olivine resorption increases with rising fO₂. Toplis & Carroll, (1995) suggested that olivine may react to form a magnetite component in the liquid, rather than the solid; however, if this reaction occurs in a highly oxidized magma chamber (such as that in which the Mull EBCS are inferred to have formed; Thomson, 1986), magnetite may have crystallized. Therefore, partial dissolution of olivine in the EBCS magmas may have been responsible for the early onset of Fe-Ti oxide fractionation. This in turn resulted in the EBCS magmas being less enriched in Fe-Ti-P when compared with the LBCS. Alternatively, the less enriched character of the EBCS may be due to the fact that they are inferred to have been more extensively contaminated with oxidized and hydrated upper-crustal rocks, relative to the LBCS (Thomson, 1986). Assimilation of such crust would have resulted in a more oxidized magma, thereby promoting the earlier (higher-MgO) onset of Fe-Ti oxide fractionation.

It is interesting to note that on Mull, it is the more alkalic igneous rocks (Plateau Group and Late Mull magma types) that display Fe-Ti-P enrichment, whereas the more tholeiitic Coire Gorm and Central Mull Tholeiite magma types do not show such pronounced enrichment (Figs 7, 12 and 13). A likely reason for this is that low degrees of partial melting of anhydrous garnet lherzolite will result in (alkalic) magmas that are enriched in Ti and P relative to concentrations of these elements in (tholeiitic) magmas formed by higher degrees of partial melting within the spinel stability field (e.g. Thompson, 1974). The Plateau Group and Late Mull magma types are believed to represent small-degree partial melts of garnet lherzolite, in contrast to the Coire Gorm and Central Mull Tholeiite magma types, which are inferred to result from larger degrees of melting (see below). Experiments by Hirose & Kushiro, (1993) show that partial melting in the spinel stability field will generate basaltic magmas which have, on average, FeO* concentrations ~2 wt % less than those of melts formed within the garnet stability field (see Klein & Langmuir, 1987). Thus, having started out with lower contents of FeO* (and presumably low Ti and P), the Coire Gorm and Central Mull Tholeiite parent magmas were less likely to have evolved to high-Fe-Ti-P liquids.

To summarize, a combination of factors, in addition to variable fO₂, may have been responsible for the Fe-Ti-P enrichment observed in the Plateau Group and Late Mull magma types. These include the depth of final melt segregation, the behaviour of Fe and P ions in the melt, partial dissolution of olivine in the melt and the depth of contamination of the parent magmas.

Temporal evolution of lithospheric contamination

Several contamination trends can be identified within the Mull lavas, dykes and cone sheets, as shown in Fig. 6. The Mull igneous centre does not appear to record a simple progression with time from lower- to upper-crustal contamination of basaltic magmas, as proposed for Skye by Dickin, (1981) and Thompson et al., (1982). The main temporal trends pertaining to crustal contamination are summarized in Fig. 15.

Figure 15. Schematic diagram illustrating the temporal evolution and source of crustal contamination of magmas of the Mull Tertiary volcano. Vertical bars indicate the range in ⁸⁷Sr/⁸⁶Sr_(t) for the Staffa magma sub-type, the Mull Plateau Group lavas and igneous rocks from Centres 1, 2 and 3. The continuous line connecting the bars shows the average ⁸⁷Sr/⁸⁶Sr_(t) for each group. The caption to Fig. 6 provides references for data used to calculate the average ⁸⁷Sr/⁸⁶Sr_(t) of Lewisian gneiss and Moine schist. The range in ⁸⁷Sr/⁸⁶Sr for Tertiary-Quaternary Icelandic lavas is from Hémond et al., (1993).

Early Mull lavas belonging to the Staffa magma sub-type (Kerr, 1999) are Plateau Group lavas that have assimilated substantial amounts (20-30%) of Moine metasediments (Thompson et al., 1986). Why these early lavas should have progressed to high crustal levels and evolved by assimilation-fractional crystallization (AFC) processes is unclear, especially when one considers that many of the overlying Plateau Group lavas are contaminated with lower-crustal granulite-facies rocks (e.g. Morrison et al., 1980; Kerr et al., 1995). Thompson et al., (1986) suggested that one explanation for this was that the earliest magmas ascended along a zone of weakness associated with the Great Glen fault, which runs through southern Mull. This `express route' to the upper crust was not exploited by later magmas, possibly because the Great Glen fault may have become inactive at this time.

Following eruption of lavas belonging to the Staffa magma sub-type, the locus of contamination shifted to the lower crust (~0·9 GPa; Thompson, 1982; Kerr et al., 1995). Some Plateau Group magmas reached the surface having undergone virtually no contamination, and even the most contaminated Plateau Group lavas are believed to contain only 3-5% highly fusible silicic granulite-facies Lewisian gneiss (Kerr et al., 1995). This contamination is evidenced by the lavas having moderately radiogenic ⁸⁷Sr/⁸⁶Sr values (<0·7045; Figs 6 and 15) and very unradiogenic ²⁰⁶Pb/²⁰⁴Pb (Dickin, 1981; Kerr et al., 1995). As the most mafic (hottest) Plateau Group lavas are the most contaminated, Kerr et al. suggested the contamination mechanism was one of `assimilation during turbulent magma ascent' (see Huppert & Sparks, 1985). An alternative to this, which does not require turbulent flow in dykes supplying the Mull lavas, is an AFC-type process in which the ratio of rates of assimilation to crystallization (r) may be substantially greater than one (r = 2·0-2·7; Reiners et al., 1995). Although such a process allows relatively large amounts of crustal assimilation (up to 18% of the initial magma mass) with very little accompanying differentiation of the basaltic melt, it is doubtful if it is thermodynamically feasible (Huppert & Sparks, 1985).

Development of the Mull central intrusive complex at ~59 Ma marked a return to predominantly upper-crustal (Moine) contamination, evidenced by high ⁸⁷Sr/⁸⁶Sr and high Rb/Nb ratios in the Mull cone sheets (Figs 5a, 6 and 15; Thomson, 1986). Lead isotopic data are not available for most of the cone sheets and granites. However, Pb isotope analyses of the (Moine-xenolithic) EBCS around Loch Scridain (Thompson et al., 1986; Preston et al., 1998) suggest that although contamination is overwhelmingly by rocks of Moine composition, the EBCS have also assimilated small amounts of granulite-facies Lewisian gneiss. In detail, it appears that Mull Centre 1 cone sheets and granites have been crustally contaminated to a greater extent than those of Centres 2 and 3. One reason for this is that the Centre 1 magmas were among the first to invade the crust, and so could have melted a substantial proportion of the most fusible Moine rocks. This would have left less-fusible crust available as a contaminant for magmas associated with Centres 2 and 3 (see Thompson, 1982). In addition, Centre 1 magmas may have been more mafic (hotter) when they reached the upper crust, when compared with magmas of Centres 2 and 3. Figure 7d shows that a good negative correlation exists between ⁸⁷Sr/⁸⁶Sr and MgO in Centre 1 rocks, consistent with AFC-type processes (see Thomson, 1986). The lower initial ⁸⁷Sr/⁸⁶Sr ratios and Rb/Ba of Centre 2 and 3 rocks could indicate smaller amounts of contamination, or may be a reflection of deeper magma chambers (Fig. 15), partly in Moine rocks and partly in upper-crustal amphibolite-facies Lewisian rocks. Resolution of this issue awaits a Pb isotopic investigation of rocks from the Mull central igneous complex.

The Centre 2 cone sheets and granites have low ⁸⁷Sr/⁸⁶Sr (mean = 0·7070) relative to those of Centre 3 (mean = 0·7090; Fig. 6). Thomson, (1986) ascribed this to differing ratios of the mass of crust assimilated to the mass of material fractionated (see DePaolo, 1982). Although this is certainly plausible, several other explanations can be advanced. One possibility is that the basaltic magmas of Centre 2 were contaminated with a greater proportion of Lewisian crust than of Moine. The generally lower initial ⁸⁷Sr/⁸⁶Sr of Lewisian crust (Fig. 6b) could mean that Centre 3 magmas actually assimilated about the same volume of crust as the magmas of Centre 2, but that most of this crust consisted of less radiogenic Lewisian material. Another possibility is that Centre 2 magmas intruded a region of crust from which much of the readily fusible material had been stripped out by earlier magmas (those of Centre 1, the locus of which lies close to that of Centre 2; Fig. 2). In contrast, Centre 3 magmas intruded crustal rocks ~5 km to the northwest of the earlier intrusive centres. This crust may have seen little previous throughput of magma.

In summary, the development of the Mull magmatic complex appears to have been associated with a cycle of magma chamber (crustal contamination) locations ranging from initially shallow to lower crustal, followed by a progressive return to upper-crustal magma chambers. Basaltic melts belonging to each of the three main intrusive centres may have occupied magma reservoirs at different depths within the upper crust.

Mantle melting processes and sources

From the chemical evidence presented above, it is possible to piece together a magmatic sequence for Mull. Primitive magmas parental to the earliest basaltic lavas (the Plateau Group) had relatively steep chondrite-normalized HREE patterns [e.g. (Nd/Yb)_cn >2] and slightly down-turned chondrite-normalized LREE patterns [(La/Nd)_cn <1] (Figs 4e and 5c). These basalts were followed by Coire Gorm-type lavas, which possess essentially flat chondrite-normalized REE patterns. The Coire Gorm lavas were in turn succeeded by basalts belonging to the Central Mull Tholeiite magma type, a group which is also represented in the EBCS suite. The Central Mull Tholeiite magma type, when free from the effects of contamination, possesses relatively flat chondrite-normalized middle REE (MREE) to HREE patterns [e.g. (Eu/Yb)_cn <1·3] and is moderately depleted in the LREE [(La/Nd)_cn <= 1]. The final Mull magma type is the Late Mull magma type, found in the LBCS suite and the Mull-Morvern late dykes and plugs. Rocks of this magma type generally have chondrite-normalized MREE-HREE patterns [e.g. (Eu/Yb)_cn >1·4] that are steeper than those of the Central Mull Tholeiite magma type. The Late Mull magma type also are enriched in the LREE [(La/Nd)_cn >1] when compared with Central Mull Tholeiite-type basalts.

It is widely held (see, e.g. Wood, 1979; Ellam, 1992) that basaltic igneous rocks showing significant depletion in the MREE-HREE relative to chondritic abundances were derived from a garnet-bearing source region. During the evolution of the Mull Tertiary volcano, the mantle source of the basalts changed with time from garnet-bearing (Plateau Group lavas) to a garnet-poor, spinel-bearing source (Coire Gorm and Central Mull Tholeiite lavas, EBCS suite), and back to a garnet-bearing source (Late Mull magma type). Similar, but not identical, temporal geochemical trends have been established in basalts erupted elsewhere along the northwest European and East Greenland margins (see above). The origin of these trends is unclear, but could be related to variations in lithospheric thickness with time (Ellam, 1992; Fram & Lesher, 1993; Kerr, 1994), temporal differences in the efficiency of polybaric melt pooling (Brodie, 1995) or generation of basaltic melt oceanward of the rifted continental margins, followed by transport of melt in dykes to the site of final eruption (White, 1992b). We now examine each of these hypotheses in turn.