Skip Navigation


Journal of Petrology Advance Access originally published online on August 31, 2005
Journal of Petrology 2005 46(12):2495-2526; doi:10.1093/petrology/egi063
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
46/12/2495    most recent
egi063v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (3)
Right arrowRequest Permissions
Google Scholar
Right arrow Articles by GURENKO, A. A.
Right arrow Articles by LINDSAY, J. M.
Right arrow Search for Related Content
GeoRef
Right arrow GeoRef Citation
Social Bookmarking
 Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

© The Author 2005. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oxfordjournals.org

A Melt Inclusion Record of Volatiles, Trace Elements and Li–B Isotope Variations in a Single Magma System from the Plat Pays Volcanic Complex, Dominica, Lesser Antilles

ANDREY A. GURENKO1,2,*, ROBERT B. TRUMBULL1, RAINER THOMAS1 and JAN M. LINDSAY3

1 GEOFORSCHUNGSZENTRUM POTSDAM, SECTION 4.2, TELEGRAFENBERG, 14473 POTSDAM, GERMANY
2 MAX-PLANCK-INSTITUT FÜR CHEMIE, ABTEILUNG GEOCHEMIE, POSTFACH 3060, 55020 MAINZ, GERMANY
3 SEISMIC RESEARCH UNIT, UNIVERSITY OF THE WEST INDIES, ST. AUGUSTINE, TRINIDAD

RECEIVED JANUARY 30, 2004; ACCEPTED JUNE 16, 2005


    ABSTRACT
 TOP
 ABSTRACT
 INTRODUCTION
 GEOLOGICAL SETTING
 ANALYTICAL TECHNIQUES
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Glass inclusions in plagioclase and orthopyroxene from dacitic pumice of the Cabrits Dome, Plat Pays Volcanic Complex in southern Dominica reveal a complexity of element behavior and Li–B isotope variations in a single volcanic center that would go unnoticed in a whole-rock study. Inclusions and matrix glasses are high-silica rhyolite with compositions consistent with about 50% fractional crystallization of the observed phenocrysts. Estimated crystallization conditions are 760–880°C, 200 MPa and oxygen fugacity of FMQ + 1 to +2 log units (where FMQ is the fayalite–magnetite–quartz buffer). Many inclusion glasses are volatile-rich (up to 6 wt % H2O and 2900 ppm Cl), but contents range down to 1 wt % H2O and 2000 ppm Cl as a result of shallow-level degassing. Sulfur contents are low throughout, with <350 ppm S. The trace element composition of inclusion glasses shows enrichment in light rare earth elements (LREE; (La/Sm)n = 2·5–6·6) and elevated Ba, Th and K contents compared with whole rocks and similar or lower Nb and heavy REE (HREE; (Gd/Yb)n = 0·5–1·0). Lithium and boron concentrations and isotope ratios in melt inclusions are highly variable (20–60 ppm Li with {delta}7Li = +4 to +15 ± 2{per thousand}; 60–100 ppm B with {delta}11B = +6 to +13 ± 2{per thousand}) and imply trapping of isotopically heterogeneous, hybrid melts. Multiple sources and processes are required to explain these features. The mid-ocean ridge basalt (MORB)-like HREE, Nb and Y signature reflects the parental magma(s) derived from the mantle wedge. Positive Ba/Nb, B/Nb and Th/Nb correlations in inclusion glasses indicate coupled enrichment in strongly fluid-mobile (Ba, B) and less-mobile (Th, Nb) trace elements, which can be explained by fractional crystallization of plagioclase, orthopyroxene and Fe–Ti oxides. The {delta}7Li and {delta}11B values are at the high end of known ranges for other island arc magmas. We attribute the high values to a 11B and 7Li-enriched slab component derived from sea-floor-altered oceanic crust and possibly further enriched in heavy isotopes by dehydration fractionation. The heterogeneity of isotope ratios in the evolved, trapped melts is attributed to shallow-level assimilation of older volcanic rocks of the Plat Pays Volcanic Complex.

KEY WORDS: subduction; volcanic arcs; igneous processes; melt inclusions; SIMS; trace elements; lithium and boron isotopes; diffusion


    INTRODUCTION
 TOP
 ABSTRACT
 INTRODUCTION
 GEOLOGICAL SETTING
 ANALYTICAL TECHNIQUES
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
One of the greatest geochemical challenges in understanding arc magmatism is to identify the components involved in, and to evaluate their separate contributions to, the origin and evolution of the erupted magmas. This information is essential for global cycling models and for assessing what materials and how much of them are fed into the deeper mantle at subduction zones. The widely accepted conceptual model for generation of volcanic arc magmas envisions production of a primary magma by melting of peridotite in the mantle wedge, the melting process being enhanced by addition of water and other fluxing components from the subducting slab (Arculus, 1994Go; Pearce & Peate, 1995Go; Tatsumi & Eggins, 1995Go). Identifying the source components is difficult enough for near-primary magmas, but these are rarely erupted and one usually has to deal with intermediate compositions with the added complexity of magma evolution processes including fractional crystallization, crustal contamination and magma mixing. As a consequence, the erupted rocks will commonly be some mixture of crystals and melts that can have partially separate histories of evolution or even source, before being finally aggregated in the eruption column. A viable way forward is to combine conventional whole-rock and mineral studies with analyses of melt, fluid and solid phases trapped and isolated in host phenocrysts. Furthermore, melt inclusion analyses offer the only direct way to determine pre-eruptive concentrations and isotope ratios of volatile components that are lost to the atmosphere and undergo isotopic fractionation during shallow degassing and eruption.

This paper discusses an application of the melt inclusion approach toward understanding the origin and evolution of intermediate and silicic arc magmas from southern Dominica, which is the volcanically most productive island in the Lesser Antilles arc (Wadge, 1984Go) and one where explosive silicic eruptions have been especially prominent. Our focus is the late Pleistocene Plat Pays Volcanic Complex (PPVC) in southern Dominica, an area that has been well studied geologically (Lindsay et al., 2003Go) and on which a comprehensive geochemical and isotopic study of whole-rock samples has recently been completed (Lindsay et al., 2005bGo). For our in-depth study of mineral and melt inclusions we chose fresh pumice from an airfall eruption of the Cabrits dome, one of 12 dacitic domes that formed within the PPVC by magma resurgence after eruption of the 34 ka Grand Bay Ignimbrite. The whole-rock geochemical and radiogenic isotope compositions of the domes and the Grand Bay Ignimbrite are virtually indistinguishable (Lindsay et al., 2005bGo), so the results of our analysis of the Cabrits dacite are significant for magma genesis in the PPVC as a whole. We determined the full spectrum of major, trace and volatile (S, Cl, H2O) element concentrations by in situ microanalysis using electron microprobe analysis (EMPA), secondary ion mass spectrometry (SIMS) and confocal laser Raman spectroscopy. Special effort was made to determine the Li and B isotopic ratios in the melt inclusions because of their utility as tracers of the volatile elements and the slab contribution to subduction zone magmatism (e.g. Leeman et al., 2004Go). To date, there has been only one other study of B isotopes in the Lesser Antilles Arc, from Martinique (Smith et al., 1997Go), and ours is the first study to determine Li isotope ratios. Combined with the mineral and whole-rock compositions, these data are used to identify the composition and T, P and conditions in the magma prior to eruption and to evaluate the source components and physical processes that contributed to magma genesis. The results demonstrate that a diversity of magma compositions were present within a single dacitic system. This diversity is obscured through amalgamation and mixing in the bulk pumice but is preserved in mineral-hosted inclusions.


    GEOLOGICAL SETTING
 TOP
 ABSTRACT
 INTRODUCTION
 GEOLOGICAL SETTING
 ANALYTICAL TECHNIQUES
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Subduction of the North American plate under the Caribbean plate resulted in the 850 km long chain of islands that forms the Lesser Antilles volcanic arc. An excellent thorough review of Cenozoic volcanism of the arc has been given by Macdonald et al. (2000)Go. Dominica is located in the central group of volcanic islands in the arc (Fig. 1a) and it is the volcanically most productive as well as the potentially most hazardous one, with nine centers classified as potentially active (Lindsay et al., 2005aGo) and with a record of explosive ignimbrite activity (Lindsay et al., 2003Go). With the exception of uplifted Pleistocene conglomerates and coral banks on the west coast (Sigurdsson & Carey, 1991Go), the island of Dominica is made up entirely of Cenozoic volcanic rocks and their erosional products. The oldest record of volcanism is from deeply eroded basaltic lavas and breccias of Miocene age. Pliocene volcanism comprises mainly basaltic to basaltic-andesite stratovolcanoes (Cochrane-Mahaut and Foundland: Briden et al., 1979Go; Bellon, 1988Go). Quaternary volcanic centers are concentrated in southern Dominica. The rocks are characteristically silicic, with compositions dominated by silicic andesite and dacite, although an exception is Morne Anglais (26–28 ka), which erupted basalts and basaltic andesite very similar in composition to those of Foundland (Lindsay et al., 2005bGo). Among the Quaternary volcanics are three large ignimbrites: the Roseau Tuff (c. 30 ka: Sigurdsson, 1972Go; Carey & Sigurdsson, 1980Go), the Grand Savanne Ignimbrite (>22 ka to >40 ka: Sparks et al., 1980aGo, 1980bGo), and the Grand Bay Ignimbrite (39 ka: Lindsay et al., 2003Go). The Grand Bay eruption is thought to have formed the Soufrière depression in southern Dominica (Lindsay et al., 2003Go), within and around which are the Morne Plat Pays stratovolcano and some 12 dacitic dome complexes whose ages range from about 36 ka to 450 years BP (Fig. 1b). Lindsay et al. (2003)Go introduced the term Plat Pays Volcanic Complex to include the Grand Bay ignimbrite and all centers associated with the Soufrière depression, and a close magmatic affinity among these units was affirmed by the geochemical and isotopic study by Lindsay et al. (2005b)Go. Continuing hot spring activity and a large swarm of volcanic earthquakes in 1998–2000 demonstrate that the magma system is still active, and indeed the Soufrière depression is regarded as the most imminent volcanic hazard on the island (Lindsay et al., 2003Go, 2005a).



View larger version (31K):
[in this window]
[in a new window]
 
Fig. 1. (a) Location map of Dominica showing the main Quaternary volcanic centers [from Lindsay et al. (2003)Go] and the Cabrits dome studied in this paper. (b) Simplified geological map of the Plat Pays Volcanic Complex in southern Dominica, modified from Lindsay et al. (2003)Go.

 
Petrography and bulk composition of the Cabrits Dome volcanic rocks
The Cabrits dome is a typical representative of the group of 12 dacitic domes that formed in and around the Soufrière depression after eruption of the Grand Bay Ignimbrite (Fig. 1b). It is one of the oldest of the post-collapse domes, with a 13C date of 35·8 ka, compared with 38·8–38·6 ka for the Grand Bay Ignimbrite itself (Lindsay et al., 2003Go). The main deposits from Cabrits and the other PPVC domes comprise block and ash flows, lava screens and basal units of pumiceous airfall and surge deposits. Pumice clasts in the airfall unit are up to 43 cm in diameter, typically light grey in color with a reddish oxidized rind. Their crystal content is about 15–20%, the dominant phenocrysts being plagioclase and orthopyoxene in a proportion of approximately 2:1; there are also minor to variable amounts of clinopyroxene or hornblende and Fe–Ti oxides. The sample D-JL18, which was selected for the melt inclusion study, contains fresh matrix glass and abundant, naturally quenched melt inclusions large enough for multiple SIMS analyses (60–150 µm) (Fig. 2). The host plagioclase (Pl) and orthopyroxene (Opx) grains are 0·5–3 mm in size; rare microphenocrysts (<0·3 mm) of the same phases are also present. Equant crystals of Fe–Ti oxides up to 200 µm across are present both in the matrix and as inclusions in other minerals. The glassy matrix commonly shows a discontinuous lamination and phenocrysts are aligned, probably a result of compaction and partial welding of the pumice (Fig. 2a). Orthopyroxene, plagioclase and titanomagnetite contain abundant primary melt inclusions, almost all of which are naturally quenched to clear glass, with or without a shrinkage bubble. A very few inclusions contain small daughter crystals, and such inclusions were not used for analysis (Fig. 3b–d). Inclusions of crystals within the phenocrysts include apatite and Fe–Ti oxides in Opx and Pl, Opx inclusions in Pl, and Pl inclusions in Opx (Fig. 2b and d).



View larger version (133K):
[in this window]
[in a new window]
 
Fig. 2. Photomicrographs showing groundmass texture and various types of melt and crystal inclusions in orthopyroxene and plagioclase phenocrysts. (a) Fluidal glassy matrix composing the groundmass of the dacite pumice D-JL18 from the Morne Cabrits dome (Plat Pays Volcanic Complex, Dominica). (b) Two glass inclusions in one orthopyroxene phenocryst containing gas bubble and combined with magnetite crystals; crystalline inclusions of plagioclase, magnetite and numerous needle-like apatite are also present. (c) Series of primary glass inclusions in plagioclase core. (d) Plagioclase-hosted glass inclusion combined with needle-like apatite crystal. Ap, apatite; Mt, magnetite; Pl, plagioclase; Gl incl, glass inclusion.

 


View larger version (35K):
[in this window]
[in a new window]
 
Fig. 3. Selected features of geochemical variations and Sr–Nd–Pb isotopic ratios of whole-rock samples from the Plat Pays Volcanic Complex, southern Dominica, from Lindsay et al. (2005b)Go. Data from the Cabrits dome are shown by stars, other units by open triangles. (a) All PPVC units have very similar radiogenic isotope composition, shown here compared with the wide range found in other Lesser Antilles islands (data from GEOROC databank: http://georoc.mpch-mainz.gwdg.de/georoc). (b) The multielement diagram normalized to the composition of sample D-JL18 emphasizes the extreme homogeneity of PPVC rocks and justifies the use of this sample as representative for the PPVC magmas. Other features of the PPVC whole-rock compositions are illustrated in combination with melt inclusion data in the results section.

 
The silicic andesites and dacites erupted from the PPVC are extremely homogeneous in composition, including the Plat Pays stratovolcano, pumice from the Grand Bay Ignimbrite, and the post-collapse domes, of which Cabrits is a typical example. To illustrate this point and to put the composition of the Cabrits dome into perspective, selected features of the geochemical and isotopic composition of these rocks are illustrated in Fig. 3. For a full description of the PPVC and other southern Dominica centers the reader is referred to Lindsay et al. (2005b)Go.


    ANALYTICAL TECHNIQUES
 TOP
 ABSTRACT
 INTRODUCTION
 GEOLOGICAL SETTING
 ANALYTICAL TECHNIQUES
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Electron microprobe analysis
Major elements in minerals and glasses were analyzed using the CAMECA SX-100 electron microprobe at the GeoForschungsZentrum Potsdam (GFZ). Accelerating voltage was 15 kV, and the beam current and size were varied: 20 nA and 1–5 µm for minerals, and 10 nA and 10–15 µm for glasses. Counting times were 20 s for Na and 40 s for other major elements. The CAMECA set of standards (mostly synthetic and natural minerals and oxides) was used for routine calibration, and the Smithsonian set of natural glasses and minerals (Jarosewich et al., 1980Go) was used to monitor the stability of the instrument during analytical sessions. Typical analytical uncertainties obtained from replicate measurements (n = 38) of the USNM 111240/52 VG-2 basaltic glass were 0·2–0·5% for SiO2; 0·5–1·4% for Al2O3, MgO and CaO; 0·7–3·2% for TiO2, FeO and Na2O; 5–10% for K2O and P2O5; and c. 15% for MnO. Sulfur and chlorine in inclusion glasses were analyzed in the same block as the major elements, using 60 s counting times on the peaks and 30 s for background. Under these conditions, the detection limit was around 250–350 ppm for both S and Cl. As monitor samples to control precision and accuracy of S and Cl measurements, we used VG-2 basaltic glass (0·134–0·137 wt % S and 0·029–0·032 wt % Cl; Dixon et al., 1991Go; Thordarson et al., 1996Go) and KE12 pantellerite glass (0·323 wt % Cl; Mosbah et al., 1991Go). Values obtained during this study were 0·149 ± 0·012 wt % S and 0·029 ± 0·008 wt % Cl (n = 38) for the VG-2 glass, and 0·324 ± 0·010 wt % Cl (n = 10) for the KE12 glass. These agree with the reference values within the ±2{sigma} uncertainty. Sulfur values similar to our results (0·147 ± 0·003 wt %) were obtained for VG-2 by N. Metrich (personal communication, 2003). The total analytical error for S and Cl, including both precision and accuracy of measurements, was better than 10%, or up to 20% for low concentrations (<500 ppm).

Hydrous Si-rich glasses are susceptible to Na loss under the analytical conditions applied (e.g. Morgan & London, 1996Go). We tested for this effect using the reference albite (Micro-Analysis Consultants Ltd, Cambridgeshire, UK), basaltic (USNM 111240/52 VG-2) and rhyolitic glasses (USNM 72854 VG-568), as well as two selected Dominica glass inclusions. Glasses were exposed for varying duration under a 15 kV, 12 nA and 10 µm beam of the JEOL JXA-8200 electron microprobe and the results are shown in Fig. 4. Under these conditions, nearly no Na loss was found for VG-2 glass, whereas up to 50% loss of Na from albite occurred after beam exposure times of 40–120 s, and about 25% Na loss occurred in VG-568 glass for 30–60 s beam exposure (Fig. 4a). The Dominica inclusions appear stable under the electron beam during the first 20 s of exposure, and then show similar behavior to the VG-568 rhyolite (Fig. 4b). We conclude that under the conditions applied for glass inclusion analyses, Na loss is less than the estimated analytical error of 3·2% for Na2O.



View larger version (36K):
[in this window]
[in a new window]
 
Fig. 4. Results of testing for Na loss during electron microprobe analysis of glass inclusions. (a) Na Ka intensity obtained over 120 s counting time on the albite reference material (Micro-Analysis Consultants Ltd, Cambridgeshire, UK) and Smithsonian USNM 111240/52 VG-2 basaltic and USNM 72854 VG-568 rhyolitic glasses (Jarosewich et al., 1980Go). (b) Na Ka intensity on two Dominica glass inclusions; shaded field shows counting time of Na during routine analyses of glass inclusions. Analyses were carried out using a JEOL JXA-8200 electron microprobe (MPI, Mainz) with 15 kV, 12 nA and 10 µm electron beam (see text). Significant Na loss (up to 50%) is observed in albite after 40–120 s and ~25% Na loss in VG-568 glass between 30 and 60 s. Nearly no Na loss was found in VG-2 glass. The Dominica inclusions appear stable for at least the first 30 s of beam exposure and then behave similarly to VG-568 glass. We conclude that Na loss during analyses of glass inclusions with 20 s exposure does not exceed the relative analytical uncertainty of 3·2% for Na2O.

 
Secondary ion mass spectrometry
Li, Be and B concentrations, Li and B isotopic composition
Glass inclusions were analyzed for Li, Be and B concentrations and Li and B isotope ratios using the CAMECA IMS6f instrument at the GFZ Potsdam, following techniques described by Chaussidon & Libourel (1993)Go, Chaussidon & Jambon (1994)Go, Chaussidon et al. (1997)Go and Chaussidon & Robert (1998)Go. Element concentrations and isotope ratios were measured in subsequent analytical sessions. Standards and unknown samples were sputtered with a nominal 12·5 kV 16O primary beam of ~2–10 nA current and ~15–20 µm in diameter. The energy slit was centered and opened at 25–50 V. A 150 µm contrast aperture and a 750 µm field aperture were used, giving an ~60 µm field of view. Secondary ions were accelerated by 10 kV and analyzed at a mass resolving power M/{Delta}M {approx} 1650–1750. The isotopes 6Li+, 7Li+, 9Be+, 10B+, 11B+ and 30Si+ were counted for 8 s, 2 s, 4 s, 8 s, 4 s and 2 s, respectively. The ion intensities were corrected for the electron multiplier dead time (14 ns). We analyzed only high-energy (>80 eV) secondary ions during measurement of element concentrations, whereas low-energy ions (>0 eV) were analyzed during 7Li/6Li and 11B/10B measurements. Typical ion intensities observed under these conditions were 200–300 cps for 7Li+ and 11B+, 10 cps for 9Be+ and 3 x 104 cps for 30Si+ during analyses of element concentrations at high-voltage offset (HVO) of –80 V, and 1 x 104 cps for 6Li+, 2 x 105 cps for 7Li+, (1–2) x 103 cps for 10B+, and (4–8) x 103 cps for 11B+ at HVO = 0 V. The measurements were started after a 300 s unrastered preburn to remove the gold coating and possible surface contamination. Each measurement included 70–100 cycles. Element concentrations relative to SiO2 and isotope ratios were calculated using calibration curves obtained by linear regressions from a set of six reference glasses (Table 1). The observed internal precision and external reproducibility of Li/Si, Be/Si and B/Si ratio measurements was better than 5%. Variations of primary beam intensity, between 2 and 10 nA depending on element concentrations in a given reference material, were found to have no systematic influence on the relative sensitivity factors [RSF, defined as the ratio of total ionic intensity for a given element to the ionic intensity of Si divided by the respective element ratios in atomic concentration for the material; i.e. (Li+/Si+)/(Li/Si)]. The Li and B isotopic ratios are expressed as {delta}7Li and {delta}11B values relative to NBS-LSVEC (7Li/6Li = 12·0192 ± 0·0002) and NBS 951 (11B/10B = 4·04558 ± 0·00033) reference materials, respectively (Flesh et al., 1973Go; Spivack & Edmond, 1986Go). The instrumental mass fractionation (IMF) values (Table 2) used to calculate 7Li/6Li and 11B/10B ratios were determined from analyses of the NIST SRM 610 and 612 reference glasses ({delta}7Li = 31 ± 1{per thousand} and {delta}11B = –1·1 ± 0·8{per thousand} for both glasses; Dalpe et al., 2001Go; Kasemann et al., 2001Go; T. Zack, personal communication, 2004) and the CRPG-CNRS GB4 internal reference glass ({delta}11B = –12·9 ± 1·0{per thousand}; Chaussidon & Jambon, 1994Go; Chaussidon, 1995Go). The applied instrumental settings and analytical conditions resulted in a typical internal precision (1{sigma}) ranging from 0·6 to 1·7{per thousand} for both 7Li/6Li and 11B/10B ratios (Table 2), and the average external reproducibility (1{sigma}) was ±2{per thousand}.


View this table:
[in this window]
[in a new window]
 
Table 1: SIMS reference materials used for calibration of Li, Be and B

 

View this table:
[in this window]
[in a new window]
 
Table 2: Instrumental mass fractionation during SIMS measurements of Li and B isotopes in reference glasses

 
H2O and trace elements
The concentrations of H2O and selected trace elements [rare earth elements (REE), Nb, Th, Sr, Y, Zr, V and Cr] in glass inclusions were determined using the CAMECA IMS3f instrument at the Max Planck Institute for Chemistry (MPI), Mainz. Analytical conditions for H2O were similar to those described by Sobolev (1996)Go and Sobolev & Chaussidon (1996)Go; that is, 12·5 kV accelerating voltage for O primary ion beam, 4·5 kV secondary accelerating voltage, –80 V offset and M/{Delta}M = 300, as no species interfering with 1H mass are expected. The energy slit was centered and opened to 25 V. A 150 µm contrast aperture and a 750 µm field aperture were used. The analyses were performed in three blocks containing six cycles each of scans over 1H, 30Si and 47Ti masses, each counted for 2 s. Titanium was monitored to detect and if necessary correct for overlap with the host mineral in the case of small (<40 µm) inclusions. Orthopyroxene and plagioclase hosts were repeatedly analyzed throughout the analytical session to monitor the H2O background level, and analysis was started when H2O concentration on orthopyroxene was equivalent to, or lower than, 0·03 wt %. Typical count rates were (3–5) x 104 cps on the 1H peak, (2–4) x 102 cps on the 47Ti peak and (2–3) x 104 cps on the 30Si peak. The external reproducibility based on reference glasses with H2O contents ranging from 0·11 to 8·5 wt % was better than 5% (Table 3). Estimates of accuracy obtained by multiple measurements of selected reference glasses (analyzed as unknowns and not used to assess the calibration line) are within 5% of the accepted value.


View this table:
[in this window]
[in a new window]
 
Table 3: H2O calibration by SIMS

 
The analysis of trace elements employed similar instrument settings as for H2O, except that a larger field aperture was used (1800 µm). Each analysis consisted of five scans, starting from 16O mass (using for magnet adjustment), then over the sequence 30Si, 39K, 44Ca, 47Ti, 51V, 52Cr, 88Sr, 89Y, 90Zr, 93Nb, each mass for REE from 133 to 180 and finally 232Th. Oxide interferences, for example, light rare earth element (LREE) oxides interfering with heavy rare earth elements (HREE), were corrected using peak deconvolution (e.g. Zinner & Crozaz, 1986Go; Fahey et al., 1987Go). Instrument drift was controlled and necessary correction applied based on daily replicate analyses of KL2-G reference glass (Jochum et al., 2000Go; Table 4). Estimated analytical error was better than 6% for all elements, except Gd, Tb, Lu, Hf and Th (between 10 and 15%).


View this table:
[in this window]
[in a new window]
 
Table 4: Precision and accuracy of trace element analyses by SIMS

 
Raman spectroscopy
In many samples, H2O analyses were also made by confocal laser Raman spectroscopy at GFZ Potsdam, following techniques described by Thomas (2000)Go. The instrument used is a Dilor XY Laser Raman Triple 800 mm spectrometer fitted with an Olympus optical microscope (80x long working distance objective). Raman spectra were collected with a Peltier-cooled CCD detector. The 488 nm line of a coherent Ar+ laser (Innova 70-3), with a power of 320 mW, was used for sample excitation. The beam diameter was about 2 µm, which produces an excitation volume of c. 10 µm3. To obtain H2O concentrations dissolved in the glass, the ratio of the 3550 cm–1 band intensity to the intensity at 490 cm–1 was measured for each glass inclusion. Each measurement consisted of two spectra taken at different spectrometer positions during five accumulations, each of 50 s acquisition time. Integral intensities instead of peak heights were used, and two integral limits of 3100–3750 cm–1 and 300–700 cm–1 were applied. The method gives an average accuracy of ±0·25 wt % for H2O concentrations of a few wt % (Thomas, 2000Go). The results of Raman and SIMS analyses on the same inclusions agree fairly well, the range of discordance being 0·1–1·6 wt %.


    RESULTS
 TOP
 ABSTRACT
 INTRODUCTION
 GEOLOGICAL SETTING
 ANALYTICAL TECHNIQUES
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Mineral compositions and conditions of magma crystallization
Orthopyroxene phenocrysts are euhedral to subhedral and their compositions correspond to hypersthene (Wo1·5–2·5 En50–58 Fs40–48; Table 5). Titanium and aluminum contents range from 0·06 to 0·18 wt % TiO2 and from 0·4 to 0·8 wt % Al2O3, respectively. Both correlate positively with mg-number [100 x Mg/(Mg + Fetot), atomic ratio], which varies between 51 and 59 (Table 5). Plagioclase forms euhedral and elongated, commonly broken crystals with a total compositional range from An47 to An67. Grains typically show normal zoning with maximum within-grain variations of 13 mol % An (Table 5). The plagioclase inclusions in orthopyroxene are at the calcic end of the total range (An51–67), and similarly, orthopyroxene in plagioclase has relatively high Mg-number (En55) compared with the total range of En50–58. This mutual inclusion of relatively Ca-rich plagioclase and Mg-rich orthopyroxene suggests fairly early cotectic crystallization of the two phases. The Fe–Ti oxides comprise coexisting, commonly intergrown, magnetite and ilmenite. The compositions of Fe–Ti oxides were determined only on grains included in Opx to avoid the problem of subsolidus exchange during cooling. Magnetite [Mt; (Fe2+,Mg)Fe2O4] contains significant amounts of ulvöspinel [Usp; Fe2TiO4], the range being Mt64–71Usp24–31, whereas the spinel [Sp; (Fe2+,Mg)Al2O4] and chromite [Crt; (Fe2+,Mg)Cr2O4] components are negligible (Table 6). The coexisting ilmenite (Ilm; FeTiO3) contains varying amounts of hematite (Hmt; Fe2O3), with compositions ranging from Ilm85Hmt15 to Ilm88Hmt12 (Table 6). Apatite (Ap) is an accessory mineral that occurs in the groundmass of the pumice and as inclusions in Opx and Pl phenocrysts. We analyzed only the apatite inclusions, and found 1·2–1·7 wt % Cl and 1·4–1·7 wt % F, giving Cl/F ratios ranging from 0·83 to 1·04.


View this table:
[in this window]
[in a new window]
 
Table 5: Major elements, volatile abundances, Li, Be, B concentrations and Li and B isotopic composition of glass inclusions; compositions of Opx and Pl host crystals

 

View this table:
[in this window]
[in a new window]
 
Table 6: Representative compositions of magnetite and ilmenite inclusions, host minerals and derived temperature and redox conditions of crystallization*

 
Quartz (Qz) is absent from the phenocryst assemblage and from the groundmass, even though the composition of groundmass glass is rhyolitic (73–76 wt % SiO2; see below). One explanation for the lack of Qz could be eruption and quenching of the magma at a temperature exceeding the quartz-in curve. Mineral thermometry from the Cabrits sample yields 760–880°C (see below), and crystallization experiments with intermediate bulk compositions have shown Qz growth from residual melts with >70 wt % SiO2 at temperatures below 840°C (Barclay et al., 1998Go; Devine et al., 1998Go; Murphy et al., 1998Go). Another factor could be the relatively high H2O concentrations in the residual melt (up to 6 wt %; see below), as high H2O fugacity decreases the liquidus temperature of Qz in granitic systems (Pichavant, 1987Go, and references therein).

Estimates of the temperature and oxygen fugacity during magma crystallization were made from mineral equilibria among coexisting Pl, Opx and Fe–Ti oxides using the QUILF program of Andersen et al. [(1993); based on calibrations of Andersen & Lindsley (1988)Go and Frost et al. (1988)Go]. Because clinopyroxene is not part of the phenocryst assemblage and was not found as mineral-hosted inclusions, we used the QUILF program in a single pyroxene mode to calculate T and conditions using compositional data from orthopyroxene, titanomagnetite and ilmenite. We also tested the option of fixing the activity of SiO2 in the melt, which allows the composition of equilibrium Cpx to be calculated by QUILF. A similar approach was used by Murphy et al. (2000)Go to constrain conditions of magma crystallization at the Soufrière Hills (Montserrat). Our samples yielded the following temperature and oxygen fugacity estimates, calculated for 1 atm total pressure and the given mineral assemblages (Table 6): T = 796–878 ± 19°C and {Delta}FMQ = 0·8–1·1 ± 0·3 log units (where FMQ is the fayalite–magnetite–quartz buffer) for Mt + Ilm + Opx, when considering the activity of SiO2; T = 761–792 ± 36°C and {Delta}FMQ = 1·2–2·0 ± 0·4 log units for Mt + Ilm + Opx with no SiO2 activity constraint; and T = 768–821 ± 54°C and {Delta}FMQ = 1·2–1·6 ± 0·6 log units for the Mt + Ilm assemblage alone.

Our geothermometry results agree with temperature estimates for the PPVC rocks by Lindsay et al. (2005b)Go. That study reported magnetite–ilmenite geothemometry in the range of 810–840°C and log values from –13·2 to –12·5 ({Delta}FMQ = +0·8 to +1·6 log units) for several post-caldera domes including Cabrits. Hornblende–plagioclase geothermometry for the few samples having the appropriate assemblage (Cabrits dome, La Vue dome, Morne Plat Pays) yielded temperatures of 800–830°C, in the same range as the magnetite–ilmenite results. Slightly higher temperatures (840–890°C) were obtained from two-pyroxene QUILF geothermometry.

Composition of glass inclusions
Major and trace elements
The compositions of glass inclusions in the Opx and Pl hosts are given in Tables 2 and 3 together with representative analyses of groundmass glass and the whole-rock composition for sample D-JL18. For many of the glass inclusions, analytical totals are high (up to 103 wt %), but taking into account the individual uncertainties of the electron microprobe, SIMS and Raman methods, the totals are acceptably close to 100%. For discussion and plots presented in this section all major element compositions of inclusions and whole rocks were recalculated to 100% on a volatile-free basis. Comparisons with the PPVC whole-rock compositions are made from data reported by Lindsay et al. (2005b)Go.

The Cabrits dome and associated units of the PPVC erupted calc-alkaline, medium-K, silicic andesites to dacites based on their relationships of SiO2 to FeO/MgO, K2O, and K2O + Na2O (Miyashiro, 1974Go; Tatsumi & Eggins, 1995Go; Arculus, 2003Go). The total SiO2 range in whole rocks is 64–68 wt % (volatile-free) and the Cabrits dome samples including D-JL18 are in the silica-rich part of this range (Fig. 5). In contrast to the bulk rock, all glass inclusions have rhyolitic compositions, whether enclosed by plagioclase or orthopyroxene. The independence of inclusion composition from the host mineral is good evidence that there has not been significant effect on melt composition by post-entrapment crystallization of the host (see mineral vectors in Fig. 5). The excellent correspondence between inclusion compositions and the composition of the glassy groundmass (Fig. 5) also suggests that the inclusions are reasonable representatives of the trapped melt compositions. The gap between the melt inclusion or groundmass glasses and the bulk-rock composition is consistent with the idea that the glasses represent residual melts, from crystallization of the major phenocryst phases in the dacite, as described in more detail in the discussion section below.



View larger version (29K):
[in this window]
[in a new window]
 
Fig. 5. Major element compositions of Opx- and Pl-hosted glass inclusions and comparison with other volcanic rocks from the Plat Pays domes on Dominica (Lindsay et al., 2005bGo). (a) Total alkalis vs SiO2; (b) CaO vs SiO2; (c) FeO vs SiO2; (d) FeO vs TiO2. All concentrations are in wt % after recalculation on a 100% volatile-free basis. The inclusions show systematically more evolved compositions than those of erupted volcanic rocks and match the composition of the groundmass glass. Arrows show control vectors for Opx, Pl and Mt addition to melt, demonstrating that the whole-rock dacite can result from a mixture of earlier-formed crystals with rhyolitic residue melt and that the inclusion compositions are not significantly affected by host-mineral crystallization.

 
Trace element analyses of glass inclusions are given in Table 7 and selected features of element abundances and ratios are shown with respect to whole-rock compositions in Figs 6 and 7. With a few exceptions described separately (Li, H2O), there are no systematic differences in minor or trace element concentrations between inclusions trapped in plagioclase and in orthopyroxene. To organize the discussion it is useful to distinguish groups of trace elements according to their intercorrelations and relationship to bulk-rock compositions, as these features may relate to different components in the trapped melts or processes of melt evolution (see discussion). We consider the following points to be significant.



View larger version (24K):
[in this window]
[in a new window]
 
Fig. 6. Trace element variation diagrams illustrating the compositional range of Opx- and Pl-hosted glass inclusions (circles) and whole-rock samples from the Plat Pays Volcanic Complex (triangles) and Cabrits dome (stars; data from Lindsay et al., 2005bGo). Compositions are in ppm except where noted.

 


View larger version (16K):
[in this window]
[in a new window]
 
Fig. 7. Variation diagrams illustrating the compositional range of volatile and light elements in the Opx- and Pl-hosted glass inclusions from the Cabrits dacite. Compositions are in ppm except where noted.

 

View this table:
[in this window]
[in a new window]
 
Table 7: Trace element compositions of Opx- and Pl-hosted glass inclusions from Dominica analysed by SIMS

 
(1) The most internally coherent group of trace elements is the high field strength elements (HFSE), Ti, Nb, Zr and Y and the middle REE (MREE) to HREE (Sm to Lu). These elements show strong positive correlations among each other (R = 0·78–0·99), exemplified in Fig. 6 by the Zr–Nb, Zr–Y and Zr–Ti diagrams. It should be noted that the inclusion glasses plot off the whole-rock compositional trends in plots with TiO2 and, to a lesser extent, with Y. We believe this is due to some accumulation of Fe–Ti oxides and other minor phases in the whole rocks, as most other trace and major element variations in the inclusion glasses follow reasonable extensions of the bulk-rock compositions.

(2) A second and rather diverse group of trace elements shows weaker intercorrelations than the first, and a good correspondence between inclusions and bulk-rock compositional trends. These elements include Th, Ba, Sr, K and B. Examples of good positive correlations are: B and K (R = 0·84), B and Ba (R = 0·84), K and Sr (R = –0·88), and K and Th (R = 0·80). It should be noted that Sr behaves as a compatible element, with lower contents in the glasses than bulk rock, and Ba shows the opposite behavior (Fig. 6). This is consistent with plagioclase-dominated fractionation and the lack of sanidine or biotite in the crystal assemblage.

(3) Neither Li nor Be shows significant correlation with other elements in the inclusion glasses, with all correlations yielding values of R < 0·6. The concentrations of these elements are apparently controlled by other processes. The concentrations of Li in the inclusion glasses vary widely and are significantly lower in plagioclase-hosted inclusions than in those hosted by Opx (see discussion).

(4) The mid-ocean ridge basalt (MORB)-normalized multi-element diagram (Fig. 8) demonstrates the overall similarity in trace element pattern between inclusion glasses and bulk rocks, but at different concentration ranges. The glasses have higher contents of incompatible elements Ba, Th, K and LREE (La, Ce), lower Nb, Sr, P, Eu and Ti, and also lower but much more variable Y and HREE. The depletions in Ti, Eu and Sr of glasses relative to bulk rock may be explained by fractionation of plagioclase, apatite and oxide phases.



View larger version (34K):
[in this window]
[in a new window]
 
Fig. 8. Multi-element diagram normalized to N-MORB (Hofmann, 1988Go) and representing trace element compositions of Opx- and Pl-hosted glass inclusions in comparison with rock data from the Plat Pays Volcanic Complex (shaded field; data from Lindsay et al., 2005bGo); the inset represents chondrite-normalized (Anders & Grevesse, 1989Go) REE concentrations in glass inclusions.

 
(5) Chondrite-normalized REE patterns of glass inclusions and bulk rocks (Fig. 8 inset) are similar or even virtually identical but typically the inclusions have slightly higher LREE (La, Ce) contents and lower and much more variable MREE and HREE (Sm to Lu). The ‘V-shape’ patterns are exemplified by strong negative correlations between (La/Sm)n = 2·5–6·6 and (Gd/Yb)n = 0·5–1·0 (R = –0·89) or Sm/Yb (R = –0·80) ratios. Europium contents are highly variable in the inclusions, and many show significantly larger negative Eu anomalies than the PPVC whole rocks. It is interesting that the strongest Eu anomalies are found in inclusions hosted by orthopyroxene.

Volatile concentrations
The concentrations of H2O in the inclusion glasses determined by Raman spectroscopy and SIMS mostly show good agreement; the relative difference between the two methods ranges from 2 to 16% and in three extreme cases reaches 21–31% (Table 5). We have taken the average value for the two methods as an estimate of true concentrations. The total range of H2O contents is 2–6 wt % and, generally, inclusions in Opx have higher water concentrations than plagioclase-hosted inclusions. With one exception, the former yielded 5–6 wt % H2O whereas all inclusions hosted by plagioclase have <5 wt % H2O. This difference is found in both SIMS and Raman results. Water contents do not correlate with incompatible elements or with other volatile components. Similar high water concentrations of 5·8 ± 0·7 wt % H2O were obtained for Mt. Pelée, Martinique by Martel et al. (1998)Go and the Soufrière Hills Volcano, Montserrat (4·3 ± 0·5 wt % H2O; Barclay et al., 1998Go). Chlorine concentrations in glass inclusions have a wide total range from 1700 to 2880 ppm (presented in Fig. 9). These values are similar to those in andesitic glass inclusions from the Izu volcanic arc (Straub & Layne, 2003Go) and examples of intermediate arc magmas from Guatemala (Sisson & Layne, 1993Go), St. Vincent (Heath et al., 1998Go) and Montserrat (Edmonds et al., 2001Go). Degassing of Cl is probably one reason for the variation in Cabrits melt inclusions, and is supported by the fact that the groundmass Cl content (1900 ppm; see Table 5) is at the low end of the inclusion range. The solubility of Cl in silicate melts is a complex function of pressure, temperature and melt composition (e.g. Metrich & Rutherford, 1992Go; Webster, 1992Go; Carroll & Webster, 1994Go; Webster et al., 1999Go; Signorelli & Carroll, 2001Go). Chlorine is less soluble in hydrous, low-alkali, siliceous melts than in basic and intermediate magmas (Webster et al., 1999Go; Signorelli & Carroll, 2001Go), and is thus likely to become oversaturated during differentiation.



View larger version (21K):
[in this window]
[in a new window]
 
Fig. 9. Chlorine and water contents of glass inclusions. Isobars of H2O and Cl solubility in felsic melts are from Webster (1997)Go and Webster et al. (1999)Go. Concentrations of H2O and Cl dissolved in glass inclusions suggest minimum pressure of phenocryst crystallization and melt inclusion trapping corresponding to ~200 MPa.

 
Because the solubilities of H2O and Cl in felsic melts are pressure dependent (Webster, 1997Go; Webster et al., 1999Go), their concentrations in the inclusions can be used to estimate the confining pressure at the time of trapping. Compared with the experimentally estimated solubilities from Webster et al. (1999)Go, the inclusion compositions correspond to a pressure of about 200 MPa, or a depth of c. 6 km, for Opx and Pl crystallization (Fig. 9). A similar estimate of ~5–6 km was obtained for the Soufrière Hills, Montserrat by Barclay et al. (1998)Go based on water solubility alone, using the model of Moore et al. (1998)Go and H2O contents of 4·3 ± 0·5 wt % in rhyolitic melts.

The sulfur contents in the Cabrits melt inclusions are low, less than the c. 350 ppm S detection limit for our electron microprobe configuration in all inclusions. Edmonds et al. (2001)Go reported very low S contents (average 70 ppm) in rhyolitic melt inclusions and matrix glass from resurgent dome samples at Soufrière Hills, Montserrat, and attributed them to loss of S by magma degassing before entrapment. This could be the case for Cabrits as well, but there is no direct evidence that sulfur concentrations in the magma were originally higher. Recent work on S solubility in rhyolitic melts at 800–1000°C (Clemente et al., 2004Go) showed that solubility is dominated by the sulfide species () and under the conditions of P, , , and FeO concentration they examined, the melts contained only 800–1000 ppm dissolved sulfur [for FeO <0·5 wt %, {Delta}NNO of –2 to –1 log units (where NNO is the nickel–nickel oxide buffer), log , log and –1 > log ]. We did not perform special studies to estimate fugacity of H2S, SO2 and S2 for the Dominica glass inclusions, but the calculated redox conditions of FMQ + 1 to FMQ + 2 log units and FeO contents of 1·3–2·3 (Tables 5 and 6) suggest that sulfur solubility in the Cabrits residual melt would have been under 1000 ppm, so degassing may not be necessary to explain the low concentrations.

Li and B concentrations and isotopic composition
The glass inclusions contain 20–60 ppm Li and 60–100 ppm B (Table 5, Figs 7 and 10). The concentrations of these elements are generally higher than the range for other island arcs shown in Fig. 10, including the study of Smith et al. (1997)Go from Martinique (boron only). At least part of this difference is due to the fact that our data represent high-silica residual melts whereas Smith et al. (1997)Go analyzed whole-rock samples of andesite to dacite composition. The high B and Li concentrations in Cabrits inclusions are similar to those found in rhyolitic groundmass and melt inclusions from caldera systems (Long Valley: Anderson et al., 2000Go; Schmitt & Simon, 2004Go, Jemez: Stix & Layne, 1996Go; La Pacana: Lindsay et al., 2001Go; Schmitt et al., 2003Go). The relationships of Li and B variations with other chemical components in the inclusions are not simple. Boron is one of the ‘group 2’ elements described above (K, Ba, Sr, Th), which show consistent positive correlations with one another. Boron also shows fair positive correlation with Cl and negative correlations with the HFSE Zr, Y, Nb and HREE (Fig. 7). Lithium behaves differently from B and shows generally poor correlations with other elements [also noted by Stix & Layne (1996)Go]. There also seems to be a systematic difference in the Li abundances depending on the nature of the host mineral, with 23–40 ppm for plagioclase-hosted inclusions and up to 58 ppm for inclusions in Opx (Fig. 7).



View larger version (29K):
[in this window]
[in a new window]
 
Fig. 10. Li and B concentrations and isotopic composition of Opx- and Pl-hosted glass inclusions shown with reference to potential magma source components. It should be noted that inclusions in Pl have systematically higher {delta}7Li values and lower Li concentrations than those in Opx. This difference is ascribed to post-entrapment diffusion and isotopic fractionation (see text). The MORB box and other reference fields are based on data of Chan et al. (1992Go, 1999Go), Chaussidon & Jambon (1994)Go, Chaussidon & Marty (1995)Go, Ishikawa & Tera (1997)Go, Smith et al. (1997)Go, Moriguti & Nakamura (1998)Go, Tomascak et al. (2000)Go and Straub & Layne (2002)Go. Assumed composition of AOC (altered oceanic crust) is 12 ppm Li with {delta}7Li = +11{per thousand} (Leeman et al., 2004Go) and 5·2 ± 1·7 ppm B with {delta}11B = +3·4 ± 1·1{per thousand} (Smith et al., 1995Go). Assumed composition of LAS (Lesser Antilles Sediment) is 55 ppm Li with {delta}7Li = –0·3 [median values of Bouman et al. (2004)Go] and 120 ppm B with {delta}11B = –10{per thousand} [for western Atlantic marine sediments near Martinique, from Smith et al. (1997)Go].

 
The isotopic composition of Li and B in the inclusion glasses is variable despite the fact that the inclusions come from a single sample of erupted magma. The values of {delta}7Li (+4 to +14{per thousand}) and {delta}11B (+6 to +13{per thousand}) in the Cabrits inclusions are considerably higher than MORB values and they overlap at the high end of the ranges reported from other volcanic arcs (Chaussidon & Jambon, 1994Go; Chaussidon & Marty, 1995Go; Ishikawa & Tera, 1997Go; Moriguti & Nakamura, 1998Go; Chan et al., 1999Go; Tomascak et al., 2000Go; Straub & Layne, 2002Go; Elliott et al., 2004Go). To our knowledge, no previous Li isotope study has been made in the Lesser Antilles, but the B-isotope study of Martinique by Smith et al. (1997)Go found much lower values of {delta}11B than our results (–5{per thousand} to +0·7{per thousand}), which were attributed to input of subducted terrestrial sediments to the magma source (see discussion). The variations in Li and B isotope ratios in the Cabrits inclusions do not correlate with each other and there is only a weak correlation of isotope ratios with the corresponding element concentrations (Fig. 10). The Li results suggest a complex behavior. Li isotope ratios do not correlate with other compositional variables in the inclusions but there is a systematic difference depending on the type of host mineral. The {delta}7Li values in Pl-hosted inclusions are higher than those in Opx with one exception (Fig. 10, Table 5), and duplicate analyses (e.g. inclusions 38-1, 75-1 and 76-1; Table 5) confirmed that this difference is not due to analytical error. Because the Pl-hosted inclusions also have lower Li and H2O concentrations, and taking into account that post-entrapment loss of hydrogen from inclusions is commonly observed (Sobolev & Danyushevsky, 1994Go), we suggest that the Pl-hosted inclusions may have suffered diffusive loss of Li and H, with the difference in isotope composition caused by higher diffusion rate for the lighter 6Li isotope as described by Richter et al. (2003)Go and discussed below. In contrast to Li, there is no dependence of B concentration or isotope ratio on the host mineral and the measured {delta}11B values show weak but significant correlations with Ba, Th, Nb and some of the REE or their ratios (Fig. 11).



View larger version (25K):
[in this window]
[in a new window]
 
Fig. 11. Boron isotopic composition of the Cabrits glass inclusions plotted against trace element concentrations and ratios.

 

    DISCUSSION
 TOP
 ABSTRACT
 INTRODUCTION
 GEOLOGICAL SETTING
 ANALYTICAL TECHNIQUES
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Melt inclusion–host rock relationships and implications for magma storage and assembly
An interpretation of melt inclusion compositions in terms of magmatic processes must first consider if the inclusions are representative of the trapped melts. The melt inclusions analyzed are large enough (many >50 µm) that concentration gradients in the melt around the growing crystals are unlikely to have affected their element contents and concentration ratios (Lu et al., 1995Go). Furthermore, the inclusion glass compositions are largely independent of the type of host mineral and the plots of inclusion data compared with orthopyroxene or plagioclase control lines in major element variation diagrams show that post-trapping crystallization or exchange with the host mineral walls was insignificant (Fig. 5). Therefore we conclude that the inclusion compositions can be treated as realistic estimates of the melts present during crystallization and trapping. A first-order observation that holds for all inclusions analyzed is their rhyolitic composition, whereas the host rock is a dacite and there are no known eruptions of rhyolite from the PPVC. We suggest that the rhyolitic glasses represent residual melts from the Cabrits magma system, and that the whole-rock composition results from a mixture of earlier-formed crystals and the melt phase (see also Naumov et al., 1996Go; Kamenetsky et al., 2000Go; Schmitt, 2001Go). Indications for this are the similarity of melt inclusions and groundmass glass compositions and the fact that inclusion data plot along extensions of the whole-rock variation trends (Figs 5 and 6). To test the hypothesis we calculated least-squares mixing models (e.g. Wright & Doherty, 1970Go) using the whole-rock composition and average analyses of orthopyroxene, plagioclase, titanomagnetite, ilmenite and apatite from the rock (Table 8). The results suggest that the gap between whole-rock composition and inclusion or matrix glasses represents about 50% of phenocryst accumulation. This estimate is about twice the observed crystal content of the erupted pumice, so there must have been considerable separation of crystals and melt before and/or during eruption.


View this table:
[in this window]
[in a new window]
 
Table 8: Major element mass balance calculations

 
A striking characteristic of the Cabrits dome and other volcanic units of the PPVC is their very homogeneous compositions at the scale of whole-rock samples (Fig. 3). This contrasts with the wide range of trace element and Li–B isotope composition of the trapped melts, and the variation in cation ratios of the host minerals (e.g. plagioclase phenocrysts with An47 to An67). Unfortunately, the systematics of host crystallization and inclusion trapping in the dacite were not regular enough to link melt inclusion compositions with particular growth stages. The implication, nevertheless, is that the erupted magma represents a mixture of melt fraction(s) and entrained phenocrysts that represent different local histories of magma differentiation and segregation processes within the Cabrits system.

The thermobarometry results from Opx, Pl and Fe–Ti oxide equilibria (Table 6) indicate that equilibration of the phenocryst assemblage took place over a considerable range of temperature, about 760–880°C, with oxygen fugacity 1–2 log units above the FMQ buffer. Pressure estimates are much less precise, but the measured H2O and Cl concentrations in inclusion glasses, compared with experimental solubilities (Fig. 9), indicate a pressure of about 200 MPa. This agrees with the 5–6 km depth of swarms of volcanic seismicity that indicate recent magma movement beneath the Soufrière depression (Lindsay et al., 2003Go).

Source components and controls on B–Li isotope variations
The large number of petrogenetic studies of the Lesser Antilles arc [for a thorough recent review see Macdonald et al. (2000)Go] reached a consensus view that primary magmas were derived from fluid-enhanced melting of a depleted mantle similar in composition to the MORB source above the subducting slab. Therefore, the major factor affecting the chemical variation of primary magmas in this setting is the composition of the slab-derived components and the relative proportion of slab vs mantle-wedge components in the final magma. The rhyolitic glass inclusions from the Cabrits dacite are far removed from any parental magma composition and can give only indirect information about the magma source. However, Li and B isotopes can be valuable for fingerprinting different slab components because the mantle Li and B concentrations are so low, and the isotope ratios are not much affected by magma differentiation (Tomascak et al., 1999Go). It is also worth pointing out that despite their high degree of differentiation, HREE and HFSE ratios in the Cabrits inclusions are similar to those from basaltic rocks of Morne Anglais, southern Dominica (e.g. Gd/Yb = 0·7–1·2 for inclusions vs 1·3–1·6 for Anglais basalt; Zr/Nb = 25–34 for inclusions and 29–39 for Anglais basalt; data from Lindsay et al., 2005bGo). The corresponding ratios in average N-MORB (Gd/Yb = 1·2 and Zr/Nb = 32 from Sun & McDonough, 1989Go) are not much different, so in terms of these relatively fluid-immobile incompatible elements, the residual melts retained