Melt inclusions from Marianas arc lavas: implications for the composition and formation of island arc magmas

Transcription

1 Chemical Geology 183 (2002) Melt inclusions from Marianas arc lavas: implications for the composition and formation of island arc magmas Adam J.R. Kent a, *, Tim R. Elliott b a Analytical and Nuclear Chemistry Division, L-202 Lawrence Livermore National Laboratory, Livermore, CA 94551, USA b University of Bristol, Wills Memorial Building, Queen s Road, Bristol, BS8 1RJ, UK Abstract We have measured the major and trace element compositions of a suite of olivine-hosted melt inclusions in basaltic lavas from the islands of Agrigan and Guguan (one sample each) from the Mariana arc, part of the larger Izu Bonin Mariana system. The two lava samples examined show distinctly different chemical signatures that are considered to represent the dominance of sediment melt (Agrigan) or fluid (Guguan) components derived from the subducting slab and transferred to the mantle wedge, and thus provide an opportunity to examine melt inclusions from both fluid- and sediment melt-dominated arc melting systems. Un-homogenized melt inclusions from both samples examined, as well as other lavas from the same islands, contain unusual subsilicic amphiboles as mineral inclusions within melt inclusions. Textural evidence, as well as the compositions of laboratory-melted inclusions, suggests that these amphiboles may have been present as an early-formed phenocryst phase and trapped alongside melt within inclusions. Major element compositions of homogenized melt inclusions vary substantially, and in part are attributable to accumulative amphibole contributions. Inclusions that appear to contain little or no amphibole component, however, typically have higher MgO and FeO T and lower CaO and Al 2 O 3 contents than their host and associated lavas. Melt inclusions also display a large range of incompatible element abundances. Although the relative abundances of incompatible elements are generally consistent with the composition of lavas from Agrigan and Guguan, melt inclusion compositions are also substantially more variable than lava compositions from these islands. Melt inclusions from the two samples analyzed have Ba/La and [La/Sm] N ratios (chemical indicators of addition of slab-fluid and slab-derived melt components, respectively) similar in magnitude to the entire field of Mariana lava compositions. In general inclusions from Agrigan have higher La/Nb, Th/Nb, U/Nb, [La/Sm] N, [La/Yb] N,K 2 O/TiO 2 and lower Cl/K 2 O, Ba/La, Ba/Nb, Ba/Th and Ce/ Ce* than Guguan melt inclusions. These signatures are consistent with addition of slab-sediment-derived melt and slab-derived fluid to the melting systems beneath Agrigan and Guguan, respectively. We suggest that the large variations apparent in melt inclusion compositions largely represent differences in the flux of slab-derived components to the mantle wedge beneath individual arc melting systems, although small-scale variations in the depletion of the mantle wedge could also play a role. Our data also indicates that the composition of the slab-derived sediment melt supplied to the mantle wedge beneath Agrigan may be compositionally heterogeneous, although a larger data set is required to examine this in further detail. The presence of melt inclusions of highly variable composition in relatively evolved olivine phenocrysts (Fo 64 Fo 83 ) show that, regardless of the ultimate cause of the incompatible element variations in melt inclusions, discrete melts of distinctly different compositions persist without substantial intermixing within the magmatic plumbing systems beneath Agrigan and Guguan (and presumably other arc volcanoes) during long intervals of magmatic differentiation. This suggests a model of small batches of magma * Corresponding author. Current address: Danish Lithosphere Center, Øster Voldgade 10 L, 1350 Copenhagen K, Denmark. Fax: address: ajrk@dlc.ku.dk. (A.J.R. Kent) /02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved. PII: S (01)

2 264 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) migrating and differentiating beneath arc volcanoes, with mixing occurring only at a very late and probably shallow stage. D 2002 Elsevier Science B.V. All rights reserved. Keywords: Melt inclusion; Mariana; Composition; Formation; Island arc magma 1. Introduction The processes that govern the production of magmas at convergent margins exert a fundamental influence over the composition and geochemical evolution of the earth s crust and mantle. The primary source of geochemical information about these processes is the chemical and isotopic composition of lavas and related magmatic rocks produced within convergent margin settings, and thus subduction-zone-related lavas have been subject to considerable geochemical scrutiny (see overviews in Gill, 1981; Thorpe, 1982; Tatsumi and Eggins, 1995). To date studies of the compositional variations in island arc lavas and those from other convergent margin settings have largely relied upon analyses of whole rock and/or glass samples that represent the erupted end-products of magmatism, although samples have almost certainly experienced considerable modification since magma generation (e.g. Davidson, 1997). Few arc lavas have MgO>6% and many contain abundant phenocrysts, suggesting that processes such as crystal fractionation, magma mixing and crustal assimilation are likely to have significantly changed the composition relative to primitive melts produced within the mantle wedge. Melt inclusions hosted within phenocryst phases from arc lavas provide a potential means to circumvent some of these difficulties. Melts trapped within early-formed phenocrysts (typically olivine) may be physically shielded from the fractionation, assimilation and mixing processes that occur prior to eruption and can considerably alter magmatic compositions. In particular, as shown in studies of other mantle-derived rocks (e.g. Sobolev and Shimizu, 1993; Kamenetsky, 1996; Kamenetsky et al., 1997, 1998; Saal et al., 1998; Shimizu, 1998; Sours-Page et al., 1998; Kent et al., 1999a; Sobolev et al., 2000) melt inclusions may provide access to melt compositions that are not expressed in the composition of erupted lavas. These variations reflect processes such as melting and melt transport processes (e.g. Sobolev and Shimizu, 1993; Shimizu, 1998; Sours-Page et al., 1998) as well as geochemical heterogeneity within mantle source regions (e.g. Saal et al., 1998; Kamenetsky et al., 1997, 1998; Sobolev et al., 2000). The presence of compositional variations in melt inclusions that are not evident in lavas suggests that much of the variations within individual melting systems may be homogenized and removed by magma mixing processes prior to eruption (e.g. Kent et al., 1999a). Previous studies of melt inclusions in subduction zone-related lavas have largely concentrated on volatile contents (e.g. Sisson and Layne, 1993; Roggensack et al., 1996; Sobolev and Chaussidon, 1996; Sisson and Bronto, 1998; Newman et al., 2000) and/ or major element compositions (Lee and Stern, 1998; Danyushevsky et al., 2000; Schiano et al., 2000). Trace element studies of melt inclusions hosted in lavas from subduction zone environments, particularly lavas from island arcs, are less common, although melt inclusions from subduction-related lavas have been shown to contain enriched components related to mass transfer from the subducting slab to the overlying mantle wedge. For example, a study of olivinehosted melt inclusions from boninite lavas dredged from the northern Tonga trench by Sobolev and Danyushevsky (1994) identified a subduction-related H 2 O and LILE-enriched component in many melt inclusions. Kamenetsky et al. (1997) analyzed melt inclusions from lavas erupted from near-axis seamounts along the Valu Fa Ridge in the southern Lau Basin thought to have formed during the southward propagation of the Eastern Lau spreading centre. Melt inclusions from these lavas record the variable addition of a subduction-related LILE, Pb- and Cl-rich fluid component to the mantle source. Here, we have focussed on the trace element compositions of melt inclusions from a more typical island arc setting and have examined olivine-hosted melt inclusions from two islands in the Mariana arc. The Mariana arc is well-studied and exhibits well documented and systematic geochemical variations,

3 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) making the context of chemical variations within melt inclusions more readily interpretable (e.g. Meijer, 1982; Stern and Ito, 1983; Woodhead, 1988; Elliott et al., 1997) Geological and geochemical background of the Mariana Arc The Mariana Arc forms part of the 2500 km Izu Bonin Mariana (IBM) arc system produced over the last 45 Ma by westward subduction of the Pacific Plate beneath the Philippine Sea Plate (Meijer et al., 1983; Fig. 1). From south to north, the Mariana Arc can be divided into the Southern Seamount, Central Island, and Northern Seamount provinces (Bloomer et al., 1989). Studies of the Mariana Arc and the larger IBM system have been instrumental in the development of our present understanding of the processes of melt generation and chemical cycling in island arc environments (e.g. Dixon and Batiza, 1979; Stern, 1979; Stern et al., 1988; Woodhead, 1988, 1989; Stolper and Newman, 1994; Elliott et al., 1997; Peate and Pearce, 1998; Bryant et al., 1999). Previous studies of Mariana arc lavas have suggested that specific geochemical variations in arc lavas can be related to addition of (at least) two discrete subducted components to the mantle wedge (e.g. Elliott et al., 1997; Peate and Pearce, 1998; Bryant et al., 1999), although this interpretation is debated (e.g. Lin et al., 1990). These two slab additions are most readily interpreted as an aqueous fluid from the altered basaltic oceanic crust and a silicate melt of the subducted sediment pile, and we will subsequently use fluid and sediment melt as a convenient short hand for these geochemical signatures. Additionally, there is evidence in some element systematics of a melt depletion history within the Mariana mantle wedge (Woodhead et al., 1993; Elliott et al., 1997). The dominant control on lava compositions appears to be the amount of sediment melt added to Fig. 1. Map of the Mariana arc showing the locations of islands relevant to this study. Triangles represent other major volcanic edifices along the arc. The island of Guam, which represents an earlier phase of arc volcanism, is also shown. The grey line behind the arc marks the axis of current back-arc rifting. Also shown are the locations of three Ocean Drilling program drill holes to the east of the arc used to constrain the bulk subducted sediment composition (map modified from Elliott et al., 1997).

4 266 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) Table 1 Summary of the key geochemical characteristics of lava samples from Agrigan and Guguan used for this study SiO 2 MgO Ba/La Ba/Nb Ce/Ce * [La/Sm] N 143 Nd/ 144 Nd 238 U/ 230 Th AGR-8b Agrigan lavas GUG Guguan lavas Bulk Sediment Ranges of compositions for representative sample suites from the same volcanoes and bulk subducted sediment from Elliott et al. (1997) have also been given for comparison. the mantle wedge, which appears to be variable along the arc (e.g. Peate and Pearce, 1998). For example, Mariana Arc lavas show variable LREE enrichment, with the most enriched lavas showing marked negative Ce anomalies and lower 143 Nd/ 144 Nd that reflect the influence of the subducting sediment (Table 1; Elliott et al., 1997). The fluid component is most evident in lavas derived from sources with the smallest sediment input (least LREE enriched), in which the fluid signature is consequently less masked. Thus, lavas with modest LREE enrichment and 143 Nd/ 144 Nd close to MORB prominently display elevated ratios of fluid soluble elements (e.g. most notably Ba, Pb and Sr, and more modestly Rb, K and U) to fluid-insoluble elements (e.g. Th, REE, Nb, Ta). This gives rise to elevated ratios of large ion lithophile elements (LILE) relative to both high field strength elements (HSFE) e.g. Ba/Nb, and LREE, e.g. Ba/La. Fig. 2 illustrates Fig. 2. Ba/La vs. [La/Sm] N for lava samples from the Mariana arc including data for lavas from the islands of Guguan and Agrigan used for this study. Symbols are explained in the accompanying legend. Data for Guguan and Agrigan samples, the estimate of bulk sediment composition, compilations of other Mariana samples and the MORB and OIB fields are taken from Elliott et al. (1997).

5 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) the effect of both slab components on the chemistry of the lavas. Sediment-derived melt has both elevated [La/Sm] N and Ba/La relative to MORB, but the addition of fluid further has an important impact on the Ba budget of the lavas, having a larger relative effect on the Ba/La ratio for sources with a smaller amount of melt added La. The distinctive arcuate field in Fig. 2, with ends far from MORB compositions is consistent with the importance of both slab-derived components, and with variable sediment melt addition but near constant fluid contribution. The samples selected for this study lie in different positions on the array shown in Fig. 2. Lavas from Guguan are derived from a sediment poor source and clearly show the effect of fluid addition, whilst the Agrigan lavas come from a sediment rich portion of the mantle wedge. From examination of melt inclusions from these samples, we hope to investigate the input of slab-derived components in both fluid- and sedimentdominated melting systems. 2. Analytical methods Olivine and clinopyroxene mineral separates were prepared from lava samples AGR-8b and GUG-13 using standard heavy liquid and magnetic mineral separation techniques. Final fractions were purified by hand picking. Observations of melt inclusions in olivine and clinopyroxene indicate that the majority of these contain crystalline phases, indicating that post eruption cooling rates of the sampled lavas were slow enough to allow crystal formation within melt inclusions. In order to obtain a homogenous glass for major and trace element analysis, crystals were heated for 10 min at redox conditions corresponding to the quartz fayalite magnetite buffer in a gas-mixing furnace at 1100 C (the temperature was chosen as an estimate of the liquidous temperature of these melts) and were quenched in water. Due to the relatively small amounts of olivine separated from these lavas, only one heating experiment was conducted. The short heating duration was selected in an attempt to prevent loss of volatile species (particularly H) via diffusion during reheating (e.g. Sobolev and Chaussidon, 1996), although subsequent measurements showed hydrogen contents of these lavas to be low (<< wt.% H 2 O; see below) compared to melt inclusions in other arc-related rocks (e.g. Sisson and Layne, 1993; Sisson and Bronto, 1998; Sobolev and Chaussidon, 1996). For examination and analysis, olivine grains were mounted in epoxy disks and polished to reveal grain interiors. Melt inclusions exposed by polishing were identified and documented using reflected and transmitted light and backscattered electron imagery. In addition to inclusions that had been heated and homogenized using the procedure described above, unheated olivine crystals were also mounted, polished and examined to determine the mineral constituents of unmelted inclusions. Clinopyroxene crystals from both samples were also prepared in the same fashion however subsequent examination and analysis revealed that clinopyroxene-hosted inclusions had assimilated a significant amount of clinopyroxene during laboratory heating and thus these inclusions are not included as part of this study. Major element compositions of melt inclusions were analyzed using a JEOL 833 electron microprobe at Lawrence Livermore National Laboratory following the analytical techniques described in Kent et al. (1999a,b). Analyses were conducted with an accelerating voltage of 15 kv and a 10-nA electron beam (defocused to 10 mm diameter for glass analyses). A standard glass (BHVO-1) was analyzed throughout the analysis sessions and provides a long-term monitor of accuracy and reproducibility. The averages of 12 BHVO-1 analyses (Table 2) are all within 2r of the accepted values for this standard (with the exception of FeO T which is just outside 2r). The long term reproducibility (2r) of SiO 2, MgO 3%, Al 2 O 3,Na 2 O, CaO, FeO T,TiO 2, are 5%; K 2 Ois 10%, P 2 O 5 20%; and Cr 2 O 5 and MnO are 30%. Trace element concentrations of melt inclusions were measured using a modified Cameca 3f ion microprobe at Lawrence Livermore National Laboratory, CA. For analysis a focussed beam of O ions accelerated at 12.5 kv was used to sputter material from selected inclusions, and sputtered ions were accelerated at a nominal 4.5 kv through a double focussing mass spectrometer. Ion beam intensities were measured with an electron multiplier. Analyses were conducted at a mass resolving power of 500 and, in order to suppress isobaric interferences from molecular ions and matrix effects, with an energy

6 268 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) Table 2 Measured major and trace element composition of BHVO-1 standard glass (accepted values from Govindaraju, 1984) Measured concentrations ± 2r Accepted value SiO TiO Al 2 O Cr 2 O FeO CaO MnO MgO Na 2 O K 2 O P 2 O Total Sr Y Zr Nb Ba La Ce Pr Nd Sm Eu Dy Er Yb Th U Ce/Ce* Measured concentrations are average of five separate analyses made during the course of analytical work for this study. All errors reflect 2r of the repeated analyses. 1 and 20 s per scan. Raw ionic intensities were normalized to 42 Ca, and data were reduced with reference to electron probe measured CaO contents using sensitivity factors determined from analysis of NBS 612 glass. Isobaric interference of BaO + and LREEO + on HREE + peaks were corrected using the procedure outlined in Zinner and Crozaz (1986) and Fahey et al. (1987) and REEO + /REE + correction factors were determined from measurement of REEdoped clinopyroxene glass standards. Analyses of basalt glass standard BHVO-1 were used throughout each analysis session to monitor accuracy of the analysis procedure. Measured concentrations of this standard are shown in Table 2 and for all elements analyzed concentrations lie within ± 2r of the accepted values (Table 2). Errors in measured concentrations are largely due to counting statistics and are estimated to be (2r): ± 5% for Sr, Y; ± 10% for Zr, Ba, La, Ce, Pr, Nd, Sm, Eu; ± 15% for Nb, Th, Dy, Er, Yb and ± 20% for U. In addition repeat measurements of BHVO suggests that the reproducibility of Ce/Ce* ratios are ± 0.05 at 2r. Water contents of melt inclusions were also investigated using a separate SIMS measurement, following the techniques outlined in Kent et al. (1999a). Water contents of all analyzed inclusions were found to be extremely low (<< wt.%). At this level, the large correction for machine background (equivalent to 0.1 wt.% H 2 O) prohibited accurate measurement of water contents and thus we have not reported the water contents of individual melt inclusions. offset of 80 ± 30 V relative to the voltage at which the intensity of 16 O + dropped to 10% of its maximum value on the low energy side of the energy distribution. Samples were gold coated prior to analysis to mitigate the effect of charge buildup. Analyses were obtained using a 5 20 na primary O beam (focussed to mm in diameter) and during analysis a 30-mm diameter field aperture was centered over the point to be analyzed. Trace element analysis involved 5 20 sequential scans of 16 O, 30 Si, 42 Ca, 86 Sr, 89 Y, 90 Zr, 93 Nb, 138 Ba, 139 La, 140 Ce, 141 Pr, 142 Ce, 143 Nd, 145 Nd, 146 Nd, 147 Sm, 148 Sm, 151 Eu, 152 Sm, 153 Eu, 154 Sm, 161 Dy, 162 Dy, 163 Dy, 166 Er, 167 Er, 168 Er, 171 Yb, 172 Yb, 173 Yb, 174 Yb, 232 Th, 238 U for count times ranging between 3. Sample description For this study, melt inclusions have been examined from two different lava samples: GUG-13 from the island of Guguan and AGR-8b from Agrigan. Both samples are olivine-clinopyroxene phyric basalts (SiO 2 51 wt.%, MgO 3.5 wt.%; Elliott et al., 1997) and are part of a suite of relatively young ( < 10 ka) mafic lavas collected by TRE. These samples were selected because they contain olivine phenocrysts (although modal olivine contents of these lavas are less than a few percent, commensurate with the relatively low MgO contents of these lavas), and because they have chemical and isotopic compositions consistent with considerable input from sediment-

7 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) derived melt (AGR-8b) and slab-derived fluid (GUG- 13) components (see discussion above). Major and trace element and Nd, Sr, Pb and U Th isotopic data for these samples are reported in Elliott et al. (1997), and the geochemical characteristics of both lavas are summarized in Table Melt inclusions Typical examples of melt inclusions are shown in Fig. 3. Melt inclusions occur in both olivine and clinopyroxene in the two samples studied and melt inclusions from both samples are similar. Olivinehosted inclusions are generally equant and ovoid with smooth walls, range between 20 and 60 mm in diameter and often contain small shrinkage bubbles (e.g. Fig. 3A,B). Clinopyroxene-hosted inclusions have irregular shapes and can be considerably larger in size than those in olivine, generally in the range mm in the largest dimension (Fig. 3D). Observations of unhomogenized olivine-hosted inclusions show that several different types of inclusions are apparent on the basis of the identified mineral constituents and textures. Although small (<30 mm) glassy inclusions are occasionally observed, the majority of unhomogenized olivine- Fig. 3. Backscattered electron micrographs of melt inclusions from AGR-8b and GUG-13. Scale bars are given for individual photos. (A) olivine-hosted melt inclusion containing spinel, amphibole, clinopyroxene and glass; (B) olivine-hosted melt inclusion containing a prominent shrinkage bubble as well as euhedral and subhedral clinopyroxene crystals in a matrix of intergrown plagioclase and glass; (C) small olivinehosted melt inclusion containing spinel, subhedral amphibole and a small amount of residual glass; (D) large irregular-shaped clinopyroxenehosted melt inclusion containing spinel (which itself contains two small melt inclusions), glass and a shrinkage bubble. Abbreviations: cpx clinopyroxene, plag plagioclase.

8 270 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) hosted inclusions contain crystalline phases. Many inclusions contain small euhedral clinopyroxene crystals in a microcrystalline matrix of intergrown feldspar and glass (Fig. 3B). These inclusions also often show evidence for considerable post-trapping growth of olivine around the margin of the inclusions. Another common group of inclusions contains a uniform residual glass phase as well as euhedral subhedral crystals of spinel ± clinopyroxene ± orthopyroxene ± amphibole (Fig. 3A,C). In many of these inclusions, all these crystalline components are evident (e.g. Fig. 3A), however a subset of these inclusions contain amphibole ± spinel as the only crystalline phases (Fig. 3C). Amphibole crystals are only observed in olivine-hosted melt inclusions and not in clinopyroxene-hosted inclusions, or as mineral inclusions in olivine or clinopyroxene. An important observation is that, especially in this latter group of inclusions, the volume proportion of amphibole and spinel often appears to be greater than the amount of these crystals that could conceivably grow from the inclusion after trapping (e.g. Fig. 3C). This suggests that amphibole and spinel may have been trapped as crystalline phases alongside melt in melt inclusions. For spinel, this observation is relatively common (e.g. Roedder, 1983), however for amphibole this observation is unique (to the best of our knowledge) and indicates that amphibole could have been present as an early-crystallizing phase. Although we note that petrographic observations are primarily made from observation of polished surfaces, and may not completely consider the full volume of the melt inclusions, viewing in transmitted light also reveals that relatively large equant amphibole crystals occur in a subset of melt inclusions. Chemical evidence outlined below is also consistent with the suggestion that amphibole was trapped as a crystalline phase in these inclusions. After heating to 1100 C melt inclusions consist of homogenous glass with or without a residual spinel phase. Because inclusions were re-heated in bulk at a single temperature, and over a relatively short time period, we do not assume that melt inclusions fully equilibrated with their host olivine. In addition, individual melt inclusions may represent different melt batches that fractionated olivine to a different degree, and thus for comparison we have corrected measured inclusion compositions to a constant MgO content of 8 wt.%. This is normally done by incrementally adding or subtracting olivine that is in Fe Mg exchange equilibrium with the trapped melt (e.g. Sobolev, 1996). However, for this study the added complication of possible assimilation of co-trapped amphiboles (see discussion below) prevents this correction being made in this way. Instead, we simply added or removed olivine of the same composition of the host olivine adjacent to each melt inclusion until the calculated MgO content of each inclusion equaled 8 wt.%. Although this procedure may introduce some uncertainties into the composition of melt inclusions, for most inclusions (all except two) this involved addition of a modest amount of olivine, and commensurately minor (3 10%) dilution of the concentrations of incompatible elements. Two inclusions have MgO contents greater than 8 wt.% and removal of olivine from these increased the concentration of incompatible elements. For one inclusion (AGR-8b ), this increase in concentration was also relatively minor (13%), however inclusion GUG had a measured MgO content of 16 wt.% and required removal of a substantial amount of olivine and an increase in incompatible trace element concentration of 49%. In addition, differences in the amount of olivine added to or subtracted from measured melt inclusions will not alter the relative abundances of incompatible elements, and thus we have largely based our discussion and interpretation of melt inclusion compositions upon incompatible element ratios. 4. Results Results of major and trace element analyses of olivine-hosted melt inclusions and their mineral hosts are shown in Table 3. Olivine phenocrysts from each lava sample analyzed for this study show a range of compositions. Olivines from GUG-13 are typically more Mg-rich than those from AGR-8b, with fosterite contents ranging between and wt.%, respectively. CaO contents of olivine phenocrysts in both samples are similar and largely uniform (AGR-8b: wt.% and GUG-13: wt.%). The major element compositions of melt inclusions in both AGR-8b and GUG-13 are variable, and most

9 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) Table 3 Major and trace element composition of olivine-hosted melt inclusions from samples AGR-8b and GUG-13 Agrigan AGR-8b Guguan GUG SiO TiO Al 2 O Cr 2 O FeO CaO MnO MgO Na 2 O K 2 O P 2 O Cl Total Fo a FeO T (8) b Dil.(8) c Sr Y Zr Nb Ba La Ce Pr Nd Sm Eu Dy Er Yb Th U [La/Sm] N Ba/Nb Ba/La La/Nb [La/Yb] N Th/U Ce/Ce* a Fosterite content of the olivine, which hosts the inclusion, measured mm from the inclusion. b FeO T calculated at a constant MgO = 8 wt.% using the method outlined in the text. c Dilution factor for trace element concentrations calculated to a constant MgO = 8 wt.%. Number represents (concentrations at MgO = 8wt.%)/(measured concentration). major element oxides (e.g. TiO 2,Al 2 O 3, FeO T, MgO, CaO, Na 2 O and K 2 O as well as Cl) vary in concentration by a factor of two or greater (Table 3). SiO 2 contents also vary substantially ( wt.%). The compositions of melt inclusions from both samples analyzed overlap considerably, although inclusions from AGR-8b typically have higher K 2 O than inclusions from GUG-13. Although the differences in major element oxide concentrations are reduced after calculation to a constant MgO = 8 wt.%, considerable variations are still apparent. In general, melt inclusions range from those that have lower SiO 2, CaO and

10 272 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) Al 2 O 3 and higher FeO T than the host lavas to those with substantially lower SiO 2 and higher FeO T and TiO 2. These latter inclusions (e.g. AGR-8b-6-2, 6-3; GUG , 1-2, 1-6) have SiO 2 contents as low as 41 wt.% and FeO T and TiO 2 contents that range up to greater than 24 and 2 wt.%, respectively. Variations are also evident in the relative abundances of incompatible elements from analyzed melt Fig. 4. Primitive mantle-normalized plots of lava and melt inclusion compositions from Agrigan and Guguan. Normalizing data from McDonough and Sun (1995). (A) Agrigan; (B) Guguan.

11 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) Fig. 5. Plots of trace element concentrations in melt inclusions and lavas from AGR-8b and GUG-13. Symbols are given in the accompanying legend and in Fig. 2. Plotted elemental concentrations are calculated to a constant MgO = 8 wt.% by the method described in the text. (A) Ba (8) vs. La (8) ; (B) Sm (8) vs. La (8). inclusions. In general, melt inclusions from each lava sample have relative abundances of incompatible elements in melt inclusions from each sample examined are broadly consistent with the composition of the host lava, although absolute concentrations vary substantially. This is shown in Fig. 4 where we have compared the primitive mantle normalized abundance of incompatible trace elements in melt inclusions and host lavas. Melt inclusions from AGR-8b have normalized concentrations that range to both higher and lower than those of Agrigan lavas. Melt inclusions from GUG-13 have normalized concentrations that range from similar to Guguan lavas to those that are distinctly lower. Both lavas from Agrigan and inclusions from AGR-8b tend to have elevated concentrations of light over heavy REE, negative Ce anomalies and are strongly depleted in Nb relative to Ba, U, K and REE. Inclusions from GUG-13 are broadly similar to Guguan lava compositions in that they have flat to slightly depleted LREE abundances, show little or no sign of negative Ce anomalies, and in general are relatively less depleted in Nb compared to other fluid immobile elements (e.g. lower Th/Nb, smaller negative Nb anomaly). Inclusions from AGR-8b show a large range of relative Sr abundances (Fig. 4), and the degree of Sr enrichment appearing to be the largest in inclusions with the lowest overall REE concentrations. For both samples, incompatible trace elements concentrations tend to correlate positively (Fig. 5) although absolute abundances may vary by more than a factor of 10 (e.g. Ba and La concentrations in melt inclusions from AGR-8b range between and ppm). Abundances of incompatible trace elements also tend to be low in inclusions that contain high FeO T and low SiO 2 (Table 3; Fig. 6). The ranges of several commonly-used incompatible element ratios ([La/Sm] N, [La/Yb] N, Th/U, Ba/ Nb, Ba/La, Ba/Th, La/Nb, Ce/Ce*, Cl/K 2 O, K 2 O/ TiO 2 ) in melt inclusions are summarized and compared to lavas from the islands of Guguan and Agrigan in Table 5. Two important observations can be made from this comparison: (i) the range in the relative abundance of incompatible element ratios from melt inclusions is generally greater, sometimes considerably, than the variation evident within the whole rock compositions of lavas from the island from which each sample originated; (ii) despite these large variations the composition of melt inclusions and lavas from the same island show many compositional consistencies (also see Figs. 6, 9 11). As shown in Table 5 and Fig. 6, both melt inclusions and lavas from Agrigan, in comparison to those from Guguan have generally higher La/Nb, Th/Nb, U/Nb, [La/Sm] N, [La/Yb] N, K 2 O/TiO 2 and generally lower Cl/K 2 O, Ba/La, Ba/Nb, Ba/Th and Ce/Ce*.

12 274 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) Fig. 6. FeO T(8) vs. trace element concentrations (A F) and incompatible element ratios (G M) for melt inclusions from AGR-8b and GUG-13. FeO T(8) and all trace element concentrations are calculated to a constant MgO = 8 wt.% using the method described in the text. Symbols are the same as described for Fig. 5.

13 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) Discussion 5.1. Possible presence of amphibole as an early-melt phase Our observations of unhomogenized melt inclusions in olivine from both samples studied indicate that a Fe Ti-rich and Si-poor amphibole is present as a euhedral or subhedral crystalline phase in many olivine-hosted melt inclusions from both AGR-8b and GUG-13 (Fig. 3) and similar amphiboles have also been observed in olivine-hosted melt inclusions from other lava samples from Agrigan and Guguan. Electron microprobe analyses indicate that amphibole inclusions have highly unusual compositions compared to those commonly reported from arc basalts and associated rocks (typically hornblende and related Fig. 7. Total Al vs. Si cation contents (calculated on the basis of 23 oxygen) for amphiboles from melt inclusions in AGR-8b and GUG- 13. Also plotted are compositions of subsilicic amphiboles from metamorphic and hydrothermal rocks from Shimizaki et al. (1984) and Sawaki (1989). Symbols are explained in the accompanying legend. Table 4 Composition of amphibole mineral inclusions in samples AGR-8b and GUG-13 (1) AGR-8b (2) GUG-13 (3) GUG-6 a (4) #5(Y) b SiO TiO Al 2 O Cr 2 O FeO T CaO MnO MgO Na 2 O K 2 O bd 3.1 Cl Total Cation Proportions Si Al iv Al vi Ti Cr Fe Fe Mn Mg Ca Na K Amphibole from olivine-hosted inclusion from Guguan sample GUG-6 (sample details in Elliott et al., 1997). Sadanagaite from the Yuge skarn deposit, Japan (from Shimazaki et al., 1984). calcic amphiboles). Amphibole inclusions are Si- and K-poor (with SiO 2 contents as low as 27 wt.%) and Al Fe Ti-rich (Table 4) and appear to be a variety of subsilicic amphibole (amphibole where the additional Al substitution has occurred for Si in the T site beyond the usual [Al 2 Si 6 ] substitution limit; Leake et al., 1997). To our knowledge subsilicic amphiboles have previously only been reported from metamorphic and hydrothermal settings (e.g. Shimizaki et al., 1984; Sawaki, 1989), although we note that the amphiboles we have identified within Marianas melt inclusions have consistently higher TiO 2 and MgO and lower K 2 O than reported from metamorphic and hydrothermal subsilicic amphiboles (Table 4; Fig. 7). Textural evidence suggests that amphibole within melt inclusions may have been present prior to melt inclusion formation and thus trapped with melt during melt inclusion formation. If this is true, then it is significant because it records the presence of an unusual composition amphibole as an early-melt phase and this may have ramifications for the origin and evolution of arc magmas. We have observed several inclusions that contained large crystals of amphibole and relatively small amounts of residual glass (e.g. Fig. 3C), although as discussed above these observations are largely limited to two dimensions. In several cases, the volume of amphibole appears too large (Fig. 3C) to have crystallized solely from the trapped melt. In addition, the euhedral and subhedral

14 276 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) form of amphibole inclusions also suggest that they were an early-crystallizing phase, crystallizing after spinel (e.g. Fig. 3C) but prior to clinopyroxene which appears to partially mantle equant amphibole in some inclusions (e.g. Fig. 3A). This is in contrast to the generally late position of amphibole crystallization in mafic igneous systems (e.g. Bowen, 1928). Amphibole crystals trapped with melt in melt inclusions also provide an explanation for the unusually high FeO T,TiO 2 and low SiO 2 contents of some laboratory-homogenized inclusions (Table 4; Fig. 8). Laboratory heating of melt inclusions containing cotrapped amphibole at higher temperatures and/or lower pressures (homogenization experiments were conducted at 1100 C and atmospheric pressure, although higher local pressures may exist within melt inclusions) than those at which amphibole will melt, resulting in a glass that is a mixture of melted amphibole and the original trapped glass. Although we have not conducted heating experiments to determine the melting temperature of trapped amphiboles, Fig. 8. SiO 2(8) vs. FeO T(8) for melt inclusions from samples AGR-8b and GUG-13. Symbols as for Figs. 5 and 7. Elemental concentrations have been calculated to a constant MgO = 8 wt.% by the method described in the text. The arrow represents the effect of assimilation of amphibole calculated using inclusion AGR-8b-7-1 as an initial composition and amphibole analysis 3 from Table 4 for the composition of the assimilant. Tickmarks show 25% increments in assimilation. we can calculate the chemical effect of assimilation of such an amphibole on a basaltic melt inclusion. In Fig. 8, where we have plotted the FeO T contents of melt inclusions against SiO 2 (with all oxide concentrations corrected for olivine fractionation using the method described above). On these plots, we also show a trajectory approximating the effect of assimilation of amphibole (using the amphibole compositions measured in unhomogenized inclusions; Table 4). Although variations in the compositions of amphibole and trapped melt mean that this trajectory is not necessarily unique, and we note that clear correlations are not apparent between FeO T and SiO 2 in all inclusions, assimilation of subsilicic amphibole could potentially explain the high FeO T and low SiO 2 contents of many melt inclusions. If this is the case, then the majority of melt inclusions have compositions consistent with assimilation of relatively small amounts of amphibole ( << 25%), however, the most FeO T -rich melt inclusion compositions are consistent with assimilation of up to 50% (by weight) of a Fe Ti-rich amphibole. Amphibole has been proposed to be a mineral phase in the melting systems of subduction-related basalts and associated rocks (e.g. Sisson and Layne, 1993), although the ubiquitous 226 Ra excesses observed in arc lavas suggest amphibole is unlikely to be a major residual phase during melting (e.g. Gill and Williams, 1990; Turner et al., 2000). Amphibole is also uncommon as a phase in Mariana lavas (even in rather evolved compositions) or entrained cumulate nodules (Dixon and Batiza, 1979; Stern, 1979; Meijer, 1982; Meijer and Reagan, 1981; Woodhead, 1988). The only reported amphiboles from the Mariana islands come from Sarigan (Meijer, 1982; Meijer and Reagan, 1981) and these are hornblendes in composition. Our measured amphibole compositions are also substantially different to those formed during experimental melting of subducted sediment (Johnson and Plank, 1999). Our recognition of amphibole as a potential early melt phase, although it only appears to be volumetrically minor, may be important for understanding major and trace element systematics in arc basalts during melting and the early stages of differentiation. At high pressures amphibole is known to be stable in water-saturated basaltic melt at temperatures less than C (e.g. Hilyard et al., 2000), and it may be that deep fractionation occurs with amphib-

15 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) ole before water is lost. Nevertheless, accounting for the unusual composition of the amphiboles remains problematic. Given the uncertainty in the major element compositions of trapped melt in inclusions, it is difficult to use major element analyses to investigate the origin of melts trapped within inclusions. Although some corrections can be made for olivine fractionation the major element compositional variations produced by assimilation of amphibole (directly on the SiO 2,TiO 2 and FeO T contents and on other elements by closure) are difficult to constrain given the variable composition of both the amphibole inclusions and the trapped melts. Variations relating to amphibole assimilation as well as those that result from partial melt and source compositional differences cannot be easily resolved. In addition, re-equilibration after trapping between trapped melt and the host olivine can also alter the major element (especially FeO T ) content of melt inclusion glasses (e.g. Danyushevsky et al., 2000). For this reason, we have largely restricted our examination of melt inclusion compositions to the examination of incompatible trace elements. Elemental partitioning data for subsilicic amphiboles are not specifically known. Measured partition coefficients for incompatible elements between amphibole and melt in basaltic systems are greater than for other phenocrysts phases but rarely greater than unity for the elements discussed below (e.g. Arth, 1976; Green and Pearson, 1985; Rollinson, 1993; Brenan et al., 1995a; LaTourette et al., 1995; Hilyard et al., 2000). Although no data are available for the subsilicic amphibole group, existing data suggests that, although the assimilation of amphibole may alter the concentrations of many trace elements by dilution, it is unlikely to have a great affect on the overall trace element budget of melt inclusions which will be dominated by the trapped melt. In addition, unless there are large differences in partitioning behaviour between different incompatible trace elements, most incompatible element ratios should be relatively insensitive to the amount of included amphibole. Although the amphibole inclusions that we have observed in unhomogenized inclusions are too small and too intimately associated with trapped glass (Fig. 3) to directly measure their trace element contents, we can test our data from homogenized melt inclusions to examine this suggestion. In Fig. 6, we have plotted the abundances of several trace elements against FeO T for melt inclusions, using this as an index of the assimilation of amphibole. For elements that have relatively low abundances in amphibole compared to melt, we expect that trace element concentrations will tend to be lower in inclusions that contain a higher component of assimilated amphibole. This appears to be the case for most of the trace elements examined. Inclusions with high FeO T have the lowest or amongst the lowest concentrations of K 2 O, Zr, Nb, Ba, La, Nd, Sm (and other REE), Th and U; Table 3; Fig. 6). Exceptions to this are Sr and Cl which both have high concentrations in some FeO T -rich inclusions. This suggests that both these elements have been contributed to some extent to the homogenized inclusion by melting of an amphibole phase. For Cl, this is consistent with the high Cl concentrations in amphibole inclusions measured by electron microprobe (up to 0.27 wt.%; Table 4). We have also plotted several commonly used incompatible element ratios against the measured FeO T contents of homogenized melt inclusions (corrected to MgO = 8 wt.%) from AGR-8b and GUG-13 (Fig. 6G M). Although substantial variations are apparent in many of these ratios, for the most part there is no correlation between the FeO T content and the [La/Sm] N, Ba/La, La/Nb, Ba/Nb, Ce/Ce* and K 2 O/TiO 2 ratios. This is consistent with our belief that assimilation of amphibole will have a relatively minor effect on most incompatible element ratios. One possible exception to this is the Cl/K 2 O ratio, which appears to correlate well with FeO T in inclusions from AGR-8b (although there is no relation apparent between FeO T and Cl/K 2 O in melt inclusions from GUG-13). We also note that, in addition to the lack of dependence between incompatible element ratios and the degree of amphibole assimilation, the range of variation in both incompatible element compositions and incompatible element ratios in melt inclusions from each sample is defined by inclusions with relatively low FeO T contents ( < 18 wt.%; Fig. 6). These FeO T contents are consistent with little or no ( < 25%) assimilation of amphibole (Fig. 8), and the large variations in incompatible element abundances in melt inclusions with low FeO T can thus be inferred to relate directly to variations in the composition of

16 278 A.J.R. Kent, T.R. Elliott / Chemical Geology 183 (2002) the trapped melt component, rather than being the result of addition of amphibole. This is even the case for the Cl/K 2 O ratio which does appear affected to some degree by amphibole assimilation. Cl/K 2 Oin inclusions with high FeO T range up to 0.3 but the Cl/ K 2 O ratios in inclusions with FeO T < 18 wt.% range between ; Fig. 6). Thus, although we have included inclusions that appear to have assimilated significant amounts of amphibole in the discussion below (e.g. GUG a, GUG , AGR-8b-6-2, AGR-8b-6-3), our overall conclusions, which are largely predicated on variations in incompatible element ratios, would be unchanged if we removed these inclusions from our data set Origin of melts trapped within inclusions An important observation from this work is that although melt inclusions from AGR-8b and GUG-13 have trace element compositions that are generally consistent with the range of compositions evident in lavas from Agrigan and Guguan, the compositions of melt inclusions from AGR-8b and GUG-13 are also substantially more variable than the range of lava compositions from each island. This comparison is summarised in Table 5 and the relation between melt inclusions and lava compositions is also shown for several different trace element ratios in Figs In general, lavas from Agrigan and melt inclusions from AGR-8b tend to generally have (or range to) higher La/Nb, Th/Nb, [La/Sm] N, [La/Yb] N, and K 2 O/TiO 2, and lower Ba/La, Ba/Nb, Ba/Th and Ce/Ce* than lavas from Guguan and melt inclusions from GUG- 13. The relation between melt inclusion and lava compositions are particularly well-illustrated by the plot of Ba/La against [La/Sm] N shown in Fig. 9. As Fig. 9. Variation of Ba/La with [La/Sm] N in melt inclusions and lavas from Agrigan and Guguan. Representative error bars (2r) are shown for each figure. Symbols are the same as for Figs. 2 and 5. Data sources are as for Fig. 2. we have already discussed, lavas from Guguan have relatively high Ba/La and low [La/Sm] N ratios, consistent with addition of a Ba-rich fluid to the mantle wedge, whereas Agrigan lavas have lower Ba/La and higher [La/Sm] N, consistent with a larger addition of LREE-enriched sediment melt to the mantle wedge. Melt inclusions from AGR-8b and GUG-13 are consistent with the variations evident in lavas collectively forming an arcuate field extending to high [La/Sm] N and high La/Ba. Inclusions from GUG-13 (Guguan) show a large range of Ba/La ratios (30 65) but relatively restricted (and MORB-like) [La/Sm] N ( ); inclusions from AGR-8b (from Agrigan) Table 5 Summary of incompatible element ratios from melt inclusions from AGR-8b and GUG-13 and lavas from Agrigan and Guguan [La/Sm] N [La/Yb] N Th/U Ba/Nb Ba/La Ba/Th La/Nb Ce/Ce * Cl/K 2 O K 2 O/TiO 2 AGR-8b melt inclusions Agrigan lavas a GUG-13 melt inclusions Guguan lavas Data for lavas from Elliott et al. (1997). a Cl not analysed for lava samples.