Journal of Petrology Advance Access originally published online on March 21, 2006
Journal of Petrology 2006 47(7):1317-1343; doi:10.1093/petrology/egl011
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Experimental Crystallization of a High-K Arc Basalt: the Golden Pumice, Stromboli Volcano (Italy)
1 UNIVERSITÁ DI PALERMO, DIPARTIMENTO DI CHIMICA E FISICA DELLA TERRA (CFTA) VIA ARCHIRAFI, 36, 90123 PALERMO, ITALY
2 INSTITUT DES SCIENCES DE LA TERRE D'ORLEANS, UMR 6113 1A RUE DE LA FEROLLERIE, ORLEANS CEDEX 2, FRANCE
RECEIVED JUNE 17, 2005; ACCEPTED FEBRUARY 22, 2006
 |
ABSTRACT |
|---|
The near-liquidus crystallization of a high-K basalt (PST-9 golden pumice, 49·4 wt % SiO2, 1·85 wt % K2O, 7·96 wt % MgO) from the present-day activity of Stromboli (Aeolian Islands, Italy) has been experimentally investigated between 1050 and 1175°C, at pressures from 50 to 400 MPa, for melt H2O concentrations between 1·2 and 5·5 wt % and
NNO ranging from –0·07 to +2·32. A drop-quench device was systematically used. AuPd alloys were used as containers in most cases, resulting in an average Fe loss of 13% for the 34 charges studied. Major crystallizing phases include clinopyroxene, olivine and plagioclase. Fe–Ti oxide was encountered in a few charges. Clinopyroxene is the liquidus phase at 400 MPa down to at least 200 MPa, followed by olivine and plagioclase. The compositions of all major phases and glass vary systematically with the proportion of crystals. Ca in clinopyroxene sensitively depends on the H2O concentration of the coexisting melt, and clinopyroxene Mg-number shows a weak negative correlation with NNO. The experimental data allow the liquidus surface of PST-9 to be defined. When used in combination with melt inclusion data, a consistent set of pre-eruptive pressures (100–270 MPa), temperatures (1140–1160°C) and melt H2O concentrations is obtained. Near-liquidus phase equilibria and clinopyroxene Ca contents require melt H2O concentrations <2·7–3·6 and 3 ± 1 wt %, respectively, overlapping with the maximum frequency of glass inclusion data (2·5–2·7 wt % H2O). For olivine to crystallize close to the liquidus, pressures close to 200 MPa are needed. Redox conditions around NNO = +0·5 are inferred from clinopyroxene compositions. The determined pre-eruptive parameters refer to the storage region of golden pumice melts, which is located at a depth of around 7·5 km, within the metamorphic arc crust. Golden pumice melts ascending from their storage zone along an adiabat will not experience crystallization on their way to the surface. KEY WORDS: basalt; pumice; experiment; phase equilibria; Stromboli
|
INTRODUCTION |
|---|
Knowledge of the physico-chemical conditions prevailing in magmas is a prerequisite for modelling igneous systems and volcanic eruptions. Possible approaches to reach this goal include: (1) geothermometry and geobarometry on phenocryst and cumulate assemblages from quenched eruption products (e.g. Luhr & Carmichael, 1980
; Ghiorso & Sack, 1991; Lindsley & Frost, 1992; Salvioli-Mariani et al., 2002); (2) melt inclusion studies (e.g. Roggensack et al., 1997; Metrich et al., 2001; Wallace, 2005); (3) experimental simulations of crystal–liquid equilibria (e.g. Johnson et al., 1994). During the past 20 years, experimental studies have increasingly been used to constrain the pre-eruptive parameters (P, T, fO2, fS2, fH2O) of subvolcanic magma bodies (e.g. Rutherford et al., 1985; Martel et al., 1998; Scaillet & Evans, 1999; Costa et al., 2004). However, this approach has mainly concerned silicic to intermediate (i.e. rhyolitic–andesitic) systems, virtually excluding more mafic compositions. This situation reflects the technical difficulties associated with the experimental crystallization of mafic melts in the presence of volatiles (e.g. Sisson & Grove, 1993a). Added to this are problems specific to mafic magmas, including the definition of liquid compositions and the possibility of complex open-system polybaric crystallization histories (e.g. O'Hara, 1977). Despite these difficulties, there is a need to apply the experimental approach to mafic magmas, in particular to complement data from melt inclusion studies. This study details the experimental simulation of near-liquidus crystal–liquid equilibria for a high-K basaltic melt from an active volcano, Stromboli (Aeolian Islands, Italy). |
VOLCANOLOGICAL AND PETROLOGICAL BACKGROUND |
|---|
 TOPABSTRACTINTRODUCTION VOLCANOLOGICAL AND PETROLOGICAL... EXPERIMENTAL STRATEGY AND...EXPERIMENTAL METHODSANALYTICAL METHODSEXPERIMENTAL RESULTSDISCUSSIONVOLCANOLOGICAL IMPLICATIONSCONCLUSIONSREFERENCES |
|---|
Stromboli, the northernmost volcano of the Aeolian archipelago, southern Italy (Fig. 1), lies on 20 km thick continental crust (Morelli et al., 1975
). It was built during the last 200 kyr in six cycles of activity, separated from each other by caldera or flank collapses (Pasquarè et al., 1993). Stromboli magmas typically show a trend of K enrichment with time, from early erupted calc-alkaline, intermediate high-K calc-alkaline and shoshonites to late K-rich products, the maximum K enrichment being attained in the late K-series (13-6 ka BP). Products younger than 6 ka, including those emitted during present-day activity, are shoshonitic to high-K basalts (Francalanci et al., 2004). Coupled isotopic and elemental variations suggest that the parental magmas originate from a metasomatized mantle source. These magmas feed one or more crustal reservoirs in which they evolve by fractional crystallization, assimilation and mixing. These reservoirs, discontinuously refilled and tapped, provide the magmas associated with the present-day activity (Francalanci et al., 1989, 1993, 2004).
|
Stromboli is famous for its persistent activity, established 1400–1800 years ago (Rosi et al., 2000
), which consists of intermittent low-energy explosions (3–6 events/h) throwing crystal-rich scoria, gas and ash to heights <150 m above the summit vents. This steady-state activity, reflecting the equilibrium between emissions and magma supply from below, is periodically interrupted by major explosions (Barberi et al., 1993; Speranza et al., 2004), which occur unpredictably 0·5–3 times/year, and consist of much higher jets than normal, resulting in fallout of meter-sized ballistic blocks and lapilli showers within a distance of several hundreds of meters from the craters upon the settled areas of Ginostra and Stromboli (Fig. 1). Eruptive paroxysms represent the least frequent eruptive manifestations at Stromboli (the last occurred in April 2003), but are also the most violent. They consist of fallout of bombs and lithic blocks (up to 10 tons) able to reach Ginostra and Stromboli villages, showers of scoria, tephra accumulation, glowing avalanches and tsunamis (Barberi et al., 1993; Metrich et al., 2005). The distinctive feature of major explosions and eruptive paroxysms is the eruption of different products representing two different magmas: a highly vesicular pumice, yellowish in color (hereafter designated as golden pumice), and a black scoria. The scoria, which is volumetrically far more abundant than the pumices, contains 
50 vol. % phenocrysts (olivine, clinopyroxene, plagioclase), whereas the pumices are typically crystal-poor (10 vol. %, e.g. Metrich et al., 2001). These magma types are often intermingled at the hand-specimen scale and phenocrysts from the black scoria may be found within the golden pumice.
Despite their differences in crystal content, the black scoria (BS) and golden pumice (GP) bulk-rock compositions are similar to each other (high-K shoshonitic basalt, Francalanci et al., 2004). Their olivine and clinopyroxene phenocrysts divide into two compositional groups. The first group consists of crystals with relatively primitive characteristics (olivines with Fo >78 and up to 91, some containing Cr-spinel inclusions; diopsidic high-Cr, low-Ti clinopyroxenes with Mg-number up to 0·9, Metrich et al., 2001; Bertagnini et al., 2003; Francalanci et al., 2004). The second group includes olivines with Fo mostly between 68 and 74 and clinopyroxenes with Mg-number mostly between 0·7 and 0·8. Compositions from the first group tend to be associated with the GP and those from the second group with the BS (Metrich et al., 2001; Francalanci et al., 2004). Compositional gaps separate the two groups, suggesting that two main olivine + clinopyroxene crystallization stages are recorded in present-day Stromboli magmas. In contrast, plagioclase crystallization appears to be restricted to shallow levels (e.g. Francalanci et al., 2004). Plagioclase in BS shows a large range of compositions, and is complexly zoned from An60 to An88. Calcic compositions (An75–90) are observed in growth zones surrounding plagioclase xenocrysts inherited from the BS magma and partially reacted in the GP magma (Landi et al. (2004).
Important insights on magmatic volatiles are given by the study of olivine-hosted melt inclusions (MI). In golden pumices, MI have elevated volatile concentrations (1·8–3·4% H2O, 707–1887 ppm CO2, 610–2500 ppm S, 980–2640 ppm Cl, 641–1364 ppm F; Metrich et al., 2001; Bertagnini et al., 2003). These volatile-rich inclusions yield minimum entrapment pressures of 350 MPa (Bertagnini et al., 2003), when calculated using the solubility model for H2O–CO2 fluid mixtures of Papale (1999). In contrast, black scoria are characterized by volatile-poor MI (e.g. H2O 
0·2%), suggesting trapping of an extensively degassed melt at very low pressures.
Phenocryst and MI data strongly suggest that the crystal-rich scoria represents a shallow-level, degassed, crystal mush (the resident magma). This magma feeds lava flows such as those during the 2002–2003 eruption, and normal Strombolian activity. The resident magma is periodically injected with a volatile-rich, crystal-poor, magma emitted as golden pumice during paroxysms and major explosions (e.g. Francalanci et al., 2004).
|
EXPERIMENTAL STRATEGY AND STARTING MATERIAL |
|---|
Near-liquidus crystal–liquid equilibria were experimentally determined to constrain the P–T–H2O–fO2 conditions of the golden pumice magma. Four experimental variables are considered: pressure, temperature, melt H2O content and fO2. Because minimum entrapment pressures calculated from melt inclusion volatile contents cluster near 350 MPa (Bertagnini et al., 2003
), 400 MPa was selected as the upper bound of the experimental pressure range. The lower bound was set at 50 MPa. Temperature was varied between 1050 and 1175°C. The effect of varying temperature on phase equilibria was investigated systematically at 400 MPa and that of varying pressure systematically at 1100°C. In most experiments (i.e. fixed P–T–fH2), charges with different melt H2O contents were run together to determine the effect of H2O on phase equilibria. Oxygen fugacity was not systematically varied, but the experiments cover an fO2 range larger than 2 log units.
Golden pumice sample PST-9 (Table 1) was selected as the starting material. It is the same sample as used previously in 1 atm continuous cooling experiments by Conte et al. (2004). PST-9 comes from a 10 cm thick layer erupted in the period AD 800–1600 (M. Pompilio, personal communication, 2001). This GP layer was found in a trench dug on the summit area of the volcano, at a height of 795 m above sea level (a.s.l.); the sample was kindly supplied by Dr M. Pompilio (Fig. 1). PST-9 is strongly vesicular (60 vol. % vesicles) and slightly porphyritic (11 vol. % crystals, within the range of other pumices), and its glassy groundmass is free of microlites. Modal proportions (vol. %, vesicle-free basis) are 89% glass, 8% clinopyroxene, 1–2% olivine, 1–2% plagioclase. Mass-balance calculations (Albarède, 1995) using the major element compositions of the bulk-rock and the phenocrysts (assumed homogeneous) are in good agreement with the point counting data for glass (91%), clinopyroxene (8%) and olivine (1%). For plagioclase, a slightly negative proportion is obtained, consistent with the very low abundance of plagioclase in the pumice. Metrich et al. (2001) found vol. % plagioclase 
clinopyroxene in other golden pumices, but noted that most crystals are in fact inherited from the crystal-rich resident BS magma and that golden pumice melts are nearly aphyric.
|
PST-9 is a K-basalt with 49·4 wt % SiO2 and 1·85 wt % K2O (Table 1). TiO2 (0·79 wt %) is low, as expected for a mafic arc melt, whereas CaO and CaO/Al2O3 are both high (respectively 12·7 wt % and 0·81, Table 1). Although golden pumices emitted over the last few centuries are all compositionally very similar (e.g. Francalanci et al., 2004
), PST-9 has the highest MgO (7·96 wt %), Cr, Ni and CaO/Al2O3. These geochemical characteristics indicate that PST-9 is relatively primitive, although its Mg-number (0·64, calculated with FeOt) is low for a typical primary mantle melt. Rare earth element (REE) abundances are high for such a mafic rock (La = 197 x C1). REE patterns are light REE (LREE) enriched (LaN/YbN = 15·3), with a slight negative Eu anomaly (Eu/Eu* = 0·84).
Clinopyroxenes in PST-9 have Mg-number (calculated with FeOt) mostly between 0·7 and 0·8. One analyzed microphenocryst has an Mg-number of 0·89 and 0·83 wt % Cr2O3 (Table 1), within the range of the diopsidic group (Francalanci et al., 2004). More evolved compositions, typical of crystals inherited from the BS magma, were also found (Table 1, column 4) and correspond to xenocrysts incorporated in the GP melt during eruption. Olivines show complex zoning patterns, with rims in the range Fo85–88 and cores in the range Fo70–84 (Table 1). Compositions with Fo <78–80 (Table 1, column 6) probably correspond to xenocrysts (Metrich et al., 2001). Scanning electron microscope (SEM) observations reveal olivine crystals crowded with melt inclusions (e.g. Metrich et al., 2001; Bertagnini et al., 2003), some of which have crystallized to Al-rich clinopyroxene, a K-rich phase (probably a mica) and sulphides. Plagioclase crystals often show sieve textures and thin (<15 µm) skeletal rims, and are all considered as xenocrysts derived from the crystal-rich BS magma (Landi et al., 2004). They have compositions between An70 and An80, with values up to An86 (Table 1). Apatite and Ti-magnetite (13–21·5 wt % TiO2, 63·5–77 wt % FeO, 4·5–6·5 wt % Al2O3, 3·5–5 wt % MgO, 0·5–2·3 wt % Cr2O3) were found as inclusions in clinopyroxene. The glassy groundmass is homogeneous with 49·6 wt % SiO2, 2·07 wt % K2O, 6·41 wt % MgO and CaO/Al2O3 = 0·66 (Table 1).
|
EXPERIMENTAL METHODS |
|---|
Charges and equipment
PST-9 pumice was first ground in an agate mortar to
50 µm, then fused in air at 1400°C, 1 atm in a Pt crucible. Two cycles of melting of 2–4 h each (with grinding between them) were performed, yielding a homogeneous glass whose composition was checked by electron microprobe (Tables 1, 4 and 5). The glass was then crushed to 10–50 µm and stored in an oven. About 30 mg of glass powder was loaded in either Au90Pd10 or Au70Pd30 capsules (15 mm length, 2·5 mm internal diameter, 0·2 mm wall thickness), depending on the experimental temperature. Ag70Pd30 tubing was also used in a few cases. Variable quantities (0·6–2 µl) of distilled water were introduced with a microsyringe at the bottom of the capsule before adding the glass powder. Capsules were sealed by arc welding, keeping them in a liquid nitrogen bath to prevent water loss. They were then put in an oven for several hours and reweighed to check for leaks.
|
|
All experiments were carried out in the same internally heated pressure vessel, working vertically and pressurized with Ar–H2 mixtures obtained by sequential loading of H2 and Ar at room temperature (Scaillet et al., 1992
). Most experiments were performed with an initial H2 pressure of 1 bar. Experimental fH2 was measured using Ni–Pd–O sensors (see below). Total pressure was recorded by a transducer calibrated against a Heise Bourdon tube gauge (uncertainty ±20 bars). A double-winding molybdenum furnace was used, allowing near-isothermal conditions in the 2–3 cm long hotspot (gradient <2–3°C/cm). Temperature was measured using three thermocouples (either type S or K) and recorded continuously (uncertainty ±5°C). Overall, the run duration range was 5–25·5 h, but except for run 4 (5 h), the others lasted for an average of 18 h.
A drop-quench technique, modified after Roux & Lefevre (1992), was systematically used. Experimental capsules (up to a maximum of six) plus the fH2 sensor capsule were placed together in a thin alumina tube that served as a sample holder. The alumina tube was hung in the furnace hotspot by a thin (0·2 mm) Pt wire. At the end of the experiment, the Pt wire was fused electrically, allowing the sample holder to fall into the cold part of the vessel. With this device, nearly isobaric quench rates of 100°C/s were achieved. Dropping of the sample holder causes the cold bottom gas to heat, and a successful quench is indicated by a peak of several tens of bars on the pressure readout. When the quench was unsuccessful, the vessel was opened, and the experiment was restarted and run for a few additional hours before quenching was attempted again. Only runs quenched with this technique are reported in Tables 2 and 3.
|
|
At the end of the experiment, capsules were weighed to check for leaks and then opened. For each PST-9 capsule, fragments of the run product were mounted in epoxy and polished for SEM observations and electron microprobe analyses. Some capsules were also prepared for determination of their Fe concentration by electron microprobe. Glass chips from supra-liquidus charges were selected for H2O determination by Karl-Fisher titration (KFT). The metallic pellets in the sensor capsule were recovered, mounted in epoxy and then analyzed by electron microprobe.
Control of redox conditions
Each run included a Ni–Pd–O sensor capsule, which served to determine fH2. These consisted of two pellets of NiPd metal mixtures (each with different Ni/Pd ratio) plus NiO, placed in either a Pt or a AuPd capsule in the presence of excess H2O (Taylor et al., 1992). Analysis of the composition of the metallic phase after the experiment allows the fO2 of the sensor system to be determined (Pownceby & O'Neill, 1994). The fH2 of the sensor (and by inference that of the experiment, as fH2 is identical for all capsules) is then obtained from the water dissociation equilibrium, using the fO2 determined above, the dissociation constant of water (Robie et al., 1979) and the fugacity of pure water at the experimental P and T (Ferry & Baumgartner, 1987; Holloway, 1987; Papale, 1997). Tables 2 and 3 list sensor compositions (mole fraction of Ni in the metallic phase, XNi) and the corresponding fH2 for each experiment.
For a given experiment (i.e. constant P–T–fH2), the fO2 of each charge is not constant but varies along with aH2O (or fH2O). The latter was determined for each charge from the H2O content of the quenched glass, using the thermodynamic model for H2O solution in multicomponent melts of Burnham (1979). The oxygen fugacity of each charge is then calculated from the water dissociation equilibrium, using the fH2 and fH2O determined above, and the dissociation constant of water (Robie et al., 1979). Typical uncertainty on log fO2 is <0·25 log units (e.g. Scaillet et al., 1995; Martel et al., 1999; Scaillet & Evans, 1999; Costa et al., 2004). In this study, fO2 values are expressed as deviations from the NNO (nickel–nickel oxide) buffer (NNO values), calculated at the P and T of interest.
|
ANALYTICAL METHODS |
|---|
All charges were systematically examined by SEM in back-scattered electron mode to assist in the identification of the phases and to evaluate the importance of quench crystallization. Instruments from the universities of Palermo (Cambridge LEO 440) and Orléans (JEOL WINSET JSM 6400) were both used. Electron microprobe analyses of mineral and glass phases were performed with either the Cameca Camebax or the Cameca SX-50 of the joint BRGM–CNRS facility at Orléans. Analyses were carried out under an acceleration voltage of 15 kV, counting times of 10 s and a sample current of 6 nA, except for metallic sensor phases and capsules, which were analyzed under 20 kV and 20 nA. For glasses, a defocused beam of 10 µm was used, and for minerals a focused beam of 1–2 µm. Silicate minerals were used as standards. For the oxides, the relative analytical errors are 1% (SiO2, Al2O3, CaO), 3% (FeO, MgO, TiO2) and 5% (MnO, Na2O, K2O, P2O5). Phase proportions, FeO and K2O losses were calculated for each charge using a least-squares mass-balance routine computed after Albarède (1995)
, using electron microprobe compositions of the starting material and phases coexisting in the charge. The regression was based on eight major oxides, excluding MnO, P2O5 and H2O.
A total of six supra-liquidus, bubble-free, experimental glasses was selected for H2O analysis by Karl-Fischer titration, using equipment and procedures identical to those described by Behrens et al. (1996). These glasses served to calibrate the ‘by difference’ method for the measurement of the H2O content of crystal-bearing glasses (Devine et al., 1995). For each electron microprobe session, the difference from 100% of electron microprobe analyses was calibrated against the dissolved glass H2O content, using the supra-liquidus glasses of known H2O content as standards. The uncertainty in the determination of the H2O concentration with this method ranges from ±0·5 wt % to around ±1·0 wt % depending on glass analytical totals.
|
EXPERIMENTAL RESULTS |
|---|
For presentation purposes, the experimental results have been divided into two groups, corresponding respectively to the 400 MPa isobaric and the 1100°C isothermal sections. For each group, experimental conditions and results are detailed in Tables 2 and 3, and experimental compositions in Tables 4 and 5, respectively. In total, 16 experiments corresponding to 34 charges are reported.
Evaluation of quench crystallization and Fe loss
Stable phases encountered include clinopyroxene, olivine, plagioclase and an oxide phase, found in one charge (10-2) and possibly in another (10-1, Table 2). When present, clinopyroxene is relatively small in size (10–20 µm, Fig. 2a and b). It may form clusters that mimic the shape of larger crystals. Olivines are large (30 µm), with tabular to equant habits (Fig. 2b). Plagioclase is tabular, usually very thin (5 µm) and difficult to analyse by microprobe (it could not be analyzed in charge 19-3). Large plagioclase crystals (60–120 µm in length) were encountered only at 50 MPa (charge 12-3, Fig. 2c).
|
Despite the use of the rapid-quench device, quench phases were detected by SEM in eight charges out of the 34 reported (Tables 2 and 3). Quench crystallization is marked in SEM images by the appearance of very thin needles that heterogeneously nucleate on clinopyroxene (Fig. 2d). Quench crystallization is correlated with, and can be evaluated from, K2O loss calculated by mass balance. This indicates that the dominant quench phase is most probably phlogopite, consistent with its abundance in preliminary experiments performed with relatively slow quench rates. In charges 16-7, 16-6, 15-1, 15-3 and 19-4, all with quench phases detected by SEM (Fig. 2d; Tables 2 and 3), K2O losses are small (i.e. <10% relative), indicating no significant influence of quench crystallization on phase compositions. The highest K2O losses were found in run 7. K2O loss is up to 36% relative in charge 7-1, decreasing regularly with the melt H2O content from charge 7-1 to 7-3, no quench phase being found in charge 7-4 (Table 2). Charges 7-1 and 7-2 are the only ones for which quench crystallization has a detectable influence on glass composition. Fe–Mg crystal–liquid exchange coefficients (Kd) in charges from run 7 and in those quenched without apparent problems (Tables 2 and 3) are in the same range. Overall, this indicates that quench crystallization has a minor influence in this study.
The importance of Fe loss was evaluated from the mass-balance calculations (Tables 2 and 3). FeO loss is 13% on average. Half of the charges have Fe losses 10% relative, six have Fe losses >20% and one has >25% loss. In general, the smallest Fe losses are associated with Au90Pd10 and the highest with Au70Pd30 capsules. However, fO2 also has a detectable influence, Fe losses becoming higher for lower NNO. Fe losses in the same range as in this study were found previously with AuPd alloys (Kawamoto & Hirose, 1994; Gaetani & Grove, 1998; Pichavant et al., 2002b), and this confirms that Fe loss can be minimized, but not suppressed, with AuPd tubing. Fe losses in Ag70Pd30 capsules (charges 19-3 and 21-4, Tables 2 and 3) are similar to those in AuPd capsules for the same experimental conditions (charges 19-2 and 3-1, Tables 2 and 3). Pre-saturating the capsules with Fe was not attempted in this study because fO2 varies between charges for a given experiment (fH2 is the buffered parameter, not fO2). In addition, pre-calculating the fO2 of a given charge is difficult, as this depends on the melt H2O content, a variable that does not relate directly to the amount of water loaded in the capsule when crystals are present. Therefore, capsule Fe pre-saturation was considered impractical. As a substitute, some experiments were performed with FeO-enriched starting mixtures (i.e. PST-9 glass + about 10% FeO) to test for the influence of variable bulk FeO (run 14, Table 2). Results are consistent with other charges at 1150°C, 400 MPa (charges 3–1 and 21–4, the latter run with Ag70Pd30 tubing) in terms of phase assemblages and location of the saturation curves.
Evaluation of equilibrium
All the experiments from this study are of the crystallization type and attainment of equilibrium has not been tested from reversals. Nevertheless, several lines of evidence can be used to evaluate the approach to equilibrium.
(1) Experimental durations (18 h on average, except for one run of 5 h) are in the range of previous phase equilibrium studies on hydrous basalts (Sisson & Grove, 1993b; Barclay & Carmichael, 2004). Baker & Eggler (1987) found that 24 h was sufficient for phase assemblage and melt composition to reach a steady state at 1060°C, 1 atm for a mafic andesite composition. Bartels et al. (1991) obtained reproducible phase assemblages and proportions after 6 h in near-liquidus anhydrous experiments on a partially crystalline high-alumina basalt at 1280°C, 1·2 GPa. Because crystallization kinetics are dramatically faster in the presence of H2O (quench crystallization is most important in charges that are the most H2O-rich, e.g. charges 7-1 and 7-2), experimental durations, as used in this study, were such that equilibrium proportions of crystals/phase assemblages should have been obtained.
(2) Crystal morphologies (euhedral, equant or tabular) suggest growth at small degrees of undercooling (e.g. Muncill & Lasaga, 1987). The distribution of crystals in the charges is homogeneous and no crystal settling was recognized (Fig. 2). The difficulties encountered in quenching the charges (see above) can be taken as indications of easy crystal nucleation and growth in our experiments.
(3) Crystals and glass phases are broadly chemically homogeneous for a given charge. Electron microprobe analyses for hydrous glasses, olivines and plagioclases have standard deviations in the same range as or lower than analytical dispersions (Tables 4 and 5), suggesting homogeneous compositions. For clinopyroxene, some chemical heterogeneity, mainly related to core–rim zonation, is apparent in some SEM images. Standard deviations of clinopyroxene analyses for a given charge may exceed 1% (i.e. higher than the analytical dispersion) for Al2O3 and more rarely for SiO2, FeO, MgO and CaO (e.g. charges 7–1, 12–3 and 16–7, Tables 4 and 5). Chemical dispersions of the same type and magnitude as in this study for clinopyroxene have been found in other phase equilibrium studies of hydrous basalts (e.g. Sisson & Grove, 1993a, 1993b).
(4) Crystal–liquid exchange coefficients (Kd) calculated from the experimental data for olivine, clinopyroxene and plagioclase are in agreement with values from the literature under comparable conditions (see below for details).
(5) Sums of residuals from the mass-balance calculations range from values <1 to >5 and strongly correlate with Fe loss (Tables 2 and 3). Charges with Fe losses <10% relative have 
R2 <1 (0·83 on average) and a constant bulk silicate composition has been essentially maintained in those experiments. In contrast, charges with Fe losses exceeding 20% have R2 >4.
(6) Electron microprobe traverses performed on one Au70Pd30 capsule from run 5 (Table 2) yielded relatively low Fe concentrations in the metal (0·1–0·3 wt % Fe), with the highest values being found near the inner rim of the capsule. This, together with the fact that charges run with FeO-enriched starting mixtures have Fe losses in the same range as the others, suggests that the experimental capsules are not fully saturated with respect to Fe.
To summarize, a state of partial equilibrium (and not bulk equilibrium, as the capsules are not saturated with Fe) is approached in the charges as shown by the textural–morphological features and the compositional data for crystals and glasses, despite the slight chemical heterogeneity noted for clinopyroxenes. Differential Fe loss caused some dispersion in the bulk silicate composition. Nevertheless, this dispersion does not significantly affect our results: the different charges run at 1150°C, 400 MPa (runs 3, 14 and 21) and 1100°C, 200 MPa (runs 15 and 19) yield phase assemblages and saturation curves that are mutually consistent despite variable Fe losses (Tables 2 and 3). We, therefore, conclude that our experiments have reached a state of equilibrium sufficient to discuss the crystallization of Stromboli magmas.
400 MPa isobaric phase equilibria
A total of 20 charges were run at 400 MPa, covering the temperature range 1050–1175°C at 25°C intervals. Their H2O contents range between 2·2 and 5·5 wt % and they have NNO values between –0·07 and +2·32. The 400 MPa phase equilibria are represented in the T–melt H2O content diagram of Fig. 3. The H2O saturation curve is constructed after Burnham (1979). Mineral saturation curves are drawn from the experimental data in Table 2. The two charges run with the FeO-added starting mixture (14-1 and 14-2) are reported in Table 2, but they are not used in the construction of the phase diagram because their bulk composition differs strongly from PST-9. Clinopyroxene is the first phase to appear on the liquidus, followed by olivine and then plagioclase upon lowering either T or melt H2O content. At 1150°C, the liquidus is encountered for a melt H2O content of about 3·3 wt %, and at 1100°C for about 4·3 wt % H2O. Saturation curves have negative slopes in T–melt H2O space, as expected for anhydrous phases. They are progressively steeper in the order plagioclase > clinopyroxene > olivine, but the clinopyroxene saturation curve is poorly defined for H2O in melt >5 wt %. The diagram is characterized by a large primary clinopyroxene stability field in the high-T, low-melt H2O content region. Plagioclase is restricted to the low-T, low-melt H2O portion. At 1100°C, cotectic olivine + clinopyroxene crystallization occurs for melt H2O contents 3·5 wt %. For olivine to crystallize at temperatures above 1100°C, melt H2O contents <3 wt % would be required at 400 MPa. It should be noted that an oxide phase, too small to be analyzed by electron microprobe, is present in charge 10-2 at 1100°C. Because this oxide could not be analyzed, it was not included in the mass-balance calculations for that charge. As, a consequence, the calculations reveal a TiO2 loss of 10% relative, suggesting that the missing phase is an Fe–Ti oxide. In charge 10-1, no oxide was detected by SEM but mass-balance calculations also reveal a TiO2 loss of about 10% relative, which indicates that Fe–Ti oxide crystallization also probably occurred.
|
Crystallinities range between zero (supra-liquidus charges) and 53 wt % (Table 2). Clinopyroxene is by far the most abundant mineral phase, followed by plagioclase and olivine in decreasing abundance. The proportion of clinopyroxene reaches a maximum of 40 wt %, that of plagioclase 13 wt % and that of olivine 4 wt %. Both plagioclase and olivine crystallize in relatively constant amounts. Clinopyroxene:olivine weight ratios range between 6 and 13, increasing with progressive crystallization, and clinopyroxene:plagioclase ratios are around 3 (data in Table 2). Crystallization is marked mainly by an increase in the proportion of clinopyroxene at the expense of melt.
1100°C isothermal phase equilibria
At 1100°C, four pressures in addition to 400 MPa were investigated: 300, 200, 100 and 50 MPa, corresponding to a total of 14 charges (Table 3). They cover a range of melt H2O contents between 1·2 and 4·9 wt % and NNO values between +0·22 and 1·85. The 1100°C phase equilibria are portrayed in the P–melt H2O content diagram of Fig. 4, where the H2O saturation curve is from Burnham (1979). Mineral saturation curves are constructed from the data in Table 3, and from Table 2 for 400 MPa. Clinopyroxene is the liquidus phase for the three highest pressures investigated (400, 300, 200 MPa). With progressive crystallization (i.e. upon lowering melt H2O content under isobaric conditions), clinopyroxene is joined by olivine and then plagioclase. One critical aspect of the diagram is the positive slope of the clinopyroxene saturation curve (Fig. 4). In comparison, the slopes of the saturation curves for olivine and plagioclase are nearly vertical or slightly negative. This results in the shrinkage of the clinopyroxene primary field and in the reduction of the crystallization interval with decreasing pressure. At 200 MPa, the cotectic olivine + clinopyroxene assemblage is close to the liquidus. The available experimental data suggest that the clinopyroxene and olivine saturation curves may cross below 200 MPa, as drawn in Fig. 4. Thus, olivine could replace clinopyroxene on the liquidus from 150 MPa down to the saturation pressure for 3·5 wt % H2O in the melt (about 100 MPa, Fig. 4). Plagioclase crystallization is restricted to melt H2O contents <2·5 wt %. Only four-phase assemblages (clinopyroxene + olivine + plagioclase + L) were found at 50 MPa.
|
As expected from the change in phase relations, there are substantial variations in crystallinities with pressure. For charges having similar melt H2O contents, crystallinities slightly decrease with decreasing pressure (e.g. from 53 wt % at 400 MPa to 31 wt % at 50 MPa, charges 10-2 and 12-1, Tables 2 and 3). Clinopyroxene becomes less abundant at lower pressures whereas olivine and plagioclase become more abundant. At 100 and 50 MPa, clinopyroxene:olivine and clinopyroxene:plagioclase weight ratios are 3–5 and 1·5–3 (respectively 6–13 and 3 at 400 MPa).
Experimental compositions
Experimental compositions are reported in full (Tables 4 and 5) for all charges in Tables 2 and 3, but compositions from charges 14-1 and 14-2 (FeO-added experiments) are excluded from the following discussion as their bulk composition differs strongly from PST-9. Clinopyroxenes are diopsides to augites with Wo contents ranging between 42 and 49%, Fs between 4 and 15% and En between 41 and 47% (Tables 4 and 5). They have low TiO2 (0·27–0·86 wt %), and variable and relatively elevated Al2O3 (2·54–8·52 wt %) and Cr2O3 (0·04–0·70 wt %). Mg-number (calculated with FeOt) is in the range 0·75–0·88. The average clinopyroxene–liquid exchange coefficient (Kd cpx–liqFe–Mg) is 0·31 ± 0·06 when calculated with FeO = FeOt, and 0·35 ± 0·07 when calculated with FeO. In this latter case, the clinopyroxene FeO is determined from the structural formulae and the glass FeO from the expression of Kress & Carmichael (1991), using the data in Tables 2–5. These average Kd values are similar to those found in recent experimental studies of hydrous basaltic compositions at P <500 MPa (0·23–0·27, Sisson & Grove, 1993a; 0·28, Pichavant et al., 2002a). It should be noted that all clinopyroxene-bearing charges from Tables 2 and 3 have been averaged, with the exception of charges 14–1 and 14–2 (FeO-added experiments). Charges the most affected by either Fe loss (5–3) or quench crystallization (7–1, 7–2) have Kd values that do not significantly depart from the others (Tables 2 and 3).
Clinopyroxene crystallizes from a range of melt compositions and coexists with different mineral phases. Therefore, it is necessary to identify bulk chemical effects on clinopyroxene composition before the influence of experimental variables can be extracted. Clinopyroxene SiO2, Al2O3, Mg-number, TiO2 and Cr2O3 are strongly correlated with the melt fraction, determined from the mass-balance calculations. In high melt fraction charges, clinopyroxenes have high SiO2, Cr2O3 and Mg-number (and En), and low Al2O3 and TiO2. With advancing crystallization, they become progressively more Al2O3-, TiO2- and Fs-rich, and their Mg-number (and En) decreases (Fig. 5a and b), consistent with the chemical evolution of the coexisting melt (Fig. 6a). Clinopyroxenes with the most primitive characteristics (i.e. high SiO2, Cr2O3 and Mg-number) are found in charges close to the liquidus at 1100°C, 300 MPa (16–6, 16–7), 200 MPa (15–1, 15–3) and 100 MPa (9–1, Tables 3 and 5). The most evolved (i.e. with high Al2O3, TiO2 and FeOt) are from the most crystallized charges at 1050–1100°C, 400 MPa (10–2, 11–4, 11–7) and at 1100°C, 50 MPa (12–1, 12–2, Tables 2–5). FeO loss leads to an increase of clinopyroxene Mg-number but does not affect the Mg-number vs melt fraction trend, which is well defined from charges with minimal FeO losses. Chemical zonation in clinopyroxene, with MgO-rich cores and Al2O3- and FeO- rich rims, reproduces, at the scale of the charge, the chemical evolution between high and low melt fraction charges. Overall, the clinopyroxene composition is a sensitive indicator of the degree of crystallization of the melt.
|
|
The atomic concentration of Ca (and the Wo content) in clinopyroxene appears to be strongly correlated with aH2O or H2O content of the melt (Fig. 7). Clinopyroxenes have atomic Ca concentration decreasing with aH2O (see also Gaetani et al., 1993
), from 0·948 to 0·71. In detail, different trends can be distinguished, depending on the phase assemblage. The clinopyroxene + olivine trend is slightly offset toward higher Ca in comparison with the clinopyroxene trend, and the clinopyroxene + olivine + plagioclase trend is displaced toward lower Ca contents. This indicates that compositional factors, in addition to aH2O, also control clinopyroxene atomic Ca concentrations. Subdividing the data in Fig. 7 as a function of different temperatures or pressures does not reveal any significant effects of these parameters on the Ca content of clinopyroxene.
|
The effect of fO2 on clinopyroxene composition was investigated from plots of clinopyroxene Mg-number against
NNO. When examined on a run-by-run basis (i.e. selecting runs 6, 7, 10, 12 and 19, where clinopyroxene occurs over a substantial fO2 range, Tables 2 and 3), clinopyroxene Mg-number (calculated with FeOt) is correlated positively with fO2, and trends with similar slopes are observed between runs. However, although systematic, such trends probably cannot be attributed to the effect of fO2 alone. This is because, in a given run, fO2 variations between charges are obtained by changing the melt H2O content, which also affects the melt fraction and the melt composition. Thus, variations in fO2 are in fact coupled with variations in melt fraction. Therefore, to analyze the effect of fO2 on clinopyroxene composition, charges from different runs (i.e. different fH2) but with the same range of crystallinities need to be considered. The analysis was applied to high melt fraction (<10% crystals) near-liquidus charges. A total of eight charges fulfil this criterion, one at 1175°C (4-3, Table 2) and seven at 1100°C (6-1, 15-1, 15-3, 16-6, 16-7, 19-4 and 9-1, Tables 2 and 3). Fe losses are <10% except in charges 4-3 and 6-1. Clinopyroxene Mg-number ranges between 0·83 and 0·88 without any clear dependence on fO2 in the limited NNO range (from +0·41 to +1·7) covered. However, if Mg-number is calculated with FeO instead of FeOt, a correlation emerges, clinopyroxene Mg-number (from 0·88 to 0·92) being slightly negatively correlated with NNO (Fig. 8). In these near-liquidus charges, clinopyroxene Mg-number is not significantly affected by the presence of olivine (compare charges 16-7 and 16-6, Tables 3 and 5).
|
Olivines have Fo ranging between 74·7 and 87·3 (Tables 4 and 5). The CaO content is relatively high (average 0·32 wt %, total range 0·22–0·63 wt %) and positively correlated with MnO (average 0·27 wt %, total range 0·13–0·39 wt %). The average olivine–liquid exchange coefficient (Kd ol–liqFe–Mg) is 0·26 ± 0·04 when calculated with FeO = FeOt, and 0·36 ± 0·04 when calculated with FeO, the glass FeO being determined as above. These average Kd values are close to those found in recent experimental studies of hydrous basaltic compositions at P <500 MPa (0·28–0·33, Sisson & Grove, 1993a
; 0·33, Pichavant et al., 2002a; 0·28, Barclay & Carmichael, 2004). As for clinopyroxene, all olivine-bearing charges from Tables 2 and 3 have been averaged; the charges most affected by either FeO loss or quench crystallization do not yield anomalous values (Tables 2 and 3). Olivine composition systematically varies with the degree of crystallization and the composition of the coexisting melt (Fig. 6b). Olivines with the highest Fo are from high melt fraction charges at 100 and 200 MPa (9-1 and 15-1, Tables 3 and 5). Those with the lowest Fo come from the most crystallized charges at 400 MPa, 1050°C (11-4 and 11-7, Tables 2 and 4). When plotted as a function of pressure, Fo contents define a bell-shaped curve with maxima at 100 and 200 MPa (Fig. 9). This is consistent with the change in phase relations with pressure noted earlier, and with the fact that, between 100 and 200 MPa, olivine crystallizes close to the liquidus (Fig. 4) with clinopyroxene. When olivine crystallizes much later than clinopyroxene (e.g. at 400 MPa, Fig. 3), Fo contents are limited to values <84. Olivines coexisting with plagioclase are the least Fo-rich, because plagioclase is restricted to charges with low melt fractions.
|
Plagioclase has An contents between 65·9 and 81·6, Or between 1·4 and 4·9, and an average Fe concentration of 1·26 wt % FeO (Tables 4 and 5). The plagioclase–liquid Ca–Na exchange coefficient (Kd plg–liqCa–Na) ranges between 1·6 and 2·5, in good agreement with values for hydrous basaltic compositions with comparable melt H2O contents (Sisson & Grove, 1993a)
. Plagioclase composition is primarily controlled by the degree of crystallization. When the proportion of crystals increases, An progressively decreases and Or increases. Plagioclase An content correlates negatively with pressure, the highest values being found in charges 12-3 (50 MPa) and 18-1 (100 MPa) at 1100°C (Tables 3 and 5). The lowest An contents are from the most crystallized charges at 400 MPa, 1050–1100°C (11-4, 11-7 and 10-2, Tables 2 and 4).
Compositions of experimental glasses can be discussed using CaO/Al2O3 as a differentiation index. Glass CaO/Al2O3 systematically decreases with decreasing melt fraction, reflecting the dominant influence of clinopyroxene on the liquid line of descent. Glasses have SiO2, Al2O3, Na2O, K2O and TiO2 progressively increasing, and CaO, FeOt, MgO and Mg-number progressively decreasing with differentiation (Fig. 10). The most differentiated glasses occur in the 1050°C, 400 MPa crystal-rich charges where SiO2 and K2O reach values up to 54·3 wt % and 3·4 wt %, respectively, within the shoshonite field (Peccerillo & Taylor, 1976). Mg-number drops from 0·64 to 0·44 (charge 11-4, 7·45% FeOt, 3·34% MgO; Table 4). The rate of decrease of Mg-number with CaO/Al2O3 depends on the phase assemblage, being progressively higher in the order clinopyroxene > clinopyroxene + olivine > clinopyroxene + olivine + plagioclase.
|
|
DISCUSSION |
|---|
Phase assemblages and crystallization sequence of Stromboli golden pumice
Only clinopyroxene, olivine and plagioclase have been found as major crystallizing phases in our experiments on hydrous golden pumice melt, corresponding to three types of phase assemblages: clinopyroxene, clinopyroxene + olivine and clinopyroxene + olivine + plagioclase. Clinopyroxene-absent assemblages (i.e. olivine, plagioclase, olivine + plagioclase) have not been encountered, although the data suggest the possibility that olivine crystallizes alone on the liquidus at 1100°C between 150 and 100 MPa (Fig. 4). In the 1 atm continuous cooling experiments of Conte et al. (2004)
, olivine was found as the sole crystallizing phase between 1175 and 1150°C, and pigeonite appeared together with clinopyroxene and plagioclase from 1150 down to 1100°C. In our experiments, an Fe–Ti oxide phase is present in one (10-2) and probably two (10-1) subliquidus charges that are among the most oxidized (NNO >1·5, Table 2). Both charges 10-2 and 10-1 have >10% crystals (10-2 with 53 wt % crystals is one of the two most crystallized charges), thus Fe–Ti oxide was not encountered at near-liquidus conditions. This suggests that, for NNO <1·5, Fe–Ti oxide is late in the crystallization sequence and follows clinopyroxene, olivine and plagioclase. It is also worth noting here that phlogopite has not been found as a stable phase in this study.
One critical result of the experiments concerns the fact that clinopyroxene precedes olivine in the crystallization sequence for all the P–T–H2O–fO2 conditions investigated in this study. Clinopyroxene persists as the liquidus phase down to at least 200 MPa; that is, in a pressure range that is unexpectedly low on the basis of previous experimental studies on basalts (e.g. Gust & Perfit, 1987; Falloon et al., 1999; Pichavant et al., 2002b). Clinopyroxene (and also orthopyroxene for primary melts from lherzolitic mantle) is expected to saturate at relatively high pressures on the liquidus of basalts. Olivine would saturate at lower pressures, and multiple saturation points are expected in the range 1–1·2 GPa or above if the system is hydrous (e.g. Pichavant et al., 2002b). In the case of PST-9, co-saturation of olivine and clinopyroxene on the liquidus is possible on the basis of the available data (Fig. 4), but only at very low pressures (P 150 MPa at 1100°C). Orthopyroxene is totally absent. These features are unusual for a relatively primitive basaltic composition (e.g. 8 wt % MgO in PST-9), but seem typical of ultra-calcic (i.e. ankaramitic) magmas from arc settings. In a recent study, Médard et al. (2004) established that a model nepheline-normative ultra-calcic melt, representative of those found in arc settings, is saturated at 200 MPa on its liquidus with clinopyroxene and olivine. Plagioclase was the third phase to crystallize and orthopyroxene was not encountered (Médard et al., 2004). Thus, the phase relations of PST-9 are similar in several critical aspects to the ultra-calcic <