Experimental Simulation of Closed-System Degassing inthe ... · Experimental Simulation of...

Experimental Simulation of Closed-SystemDegassing in the System Basalt^H2O^CO2^S^Cl

PRISCILLE LESNE1*, SIMON C. KOHN1, JON BLUNDY1,FRED WITHAM1, ROMAN E. BOTCHARNIKOV2 ANDHARALD BEHRENS2

1SCHOOL OF EARTH SCIENCES, UNIVERSITY OF BRISTOL, WILLS MEMORIAL BUILDING, QUEENS ROAD,

BRISTOL BS8 1RJ, UK2INSTITUT FU« R MINERALOGIE, LEIBNIZ UNIVERSITA« T HANNOVER, CALLINSTRASSE 3,

D-30167 HANNOVER, GERMANY

RECEIVEDJULY 7, 2010; ACCEPTED MAY 10, 2011ADVANCE ACCESS PUBLICATION AUGUST 4, 2011

Magma degassing processes are commonly elucidated by studies of

melt inclusions in erupted phenocrysts and measurements of gas dis-

charge at volcanic vents, allied to experimentally constrained models

of volatile solubility. Here we develop an alternative experimental

approach aimed at directly simulating decompression-driven,

closed-system degassing of basaltic magma in equilibrium with an

H^C^O^S^Cl fluid under oxidized conditions (fO2 of 1·0^2·4 logunits above the Ni^NiO buffer). Synthetic experimental starting

materials were based on basaltic magmas erupted at the persistently

degassing volcanoes of Stromboli (Italy) and Masaya

(Nicaragua) with an initial volatile inventory matched to the most

undegassed melt inclusions from each volcano. Experiments were

run at 25^400MPa under super-liquidus conditions (11508C).Run product glasses and starting materials were analysed by electron

microprobe, secondary ion mass spectrometry, Fourier transform in-

frared spectroscopy, Karl-Fischer titration, Fe2þ/Fe3þ colorimetry

and CS analyser. The composition of the exsolved vapour in each

run was determined by mass balance. Our results show that H2O/

CO2 ratios increase systematically with decreasing pressure, whereas

CO2/S ratios attain a maximum at pressures of 100^300MPa. S is

preferentially released over Cl at low pressures, leading to a sharp in-

crease in vapour S/Cl ratios and a sharp drop in the S/Cl ratios of

glasses. This accords with published measurements of volatile con-

centrations in melt inclusion and groundmass glasses at Stromboli

(and Etna). Experiments with different S abundances show that

the H2O and CO2 contents of the melt at fluid saturation are not af-

fected.The CO2 solubility in experiments using both sets of starting

materials is well matched to calculated solubilities using published

models. Models consistently overestimate H2O solubilities for the

Stromboli-like composition, leading to calculated vapour compos-

itions that are more CO2-rich and calculated degassing trajectories

that are more strongly curved than observed in experiments.The dif-

ference is less acute for the Masaya-like composition, emphasizing

the important compositional dependence of solubility and melt^

vapour partitioning. Our novel experimental method can be readily

extended to other bulk compositions.

KEY WORDS: experiments; solubility; degassing; basalt; Stromboli;

Masaya

I NTRODUCTIONArc magmas are characterized by high concentrations ofthe volatile species H2O, CO2, SO2, H2S and HCl (e.g.Symonds et al., 1994). To understand the role played bythese volatiles in volcanic processes, such as magma degas-sing and eruption, their solubilities in silicate melts andtheir partitioning between coexisting melt and vapourphases need to be known. Of particular importance is abetter understanding of how the concentrations and ratiosof H2O, CO2, sulphur species and chlorine dissolved inmagmas and volcanic gases can be used to both recon-struct and forecast the evolution of magmatic systems.

*Corresponding author. E-mail: [email protected]

� The Author 2011. Published by Oxford University Press. Allrights reserved. For Permissions, please e-mail: [email protected]

JOURNALOFPETROLOGY VOLUME 52 NUMBER 9 PAGES1737^1762 2011 doi:10.1093/petrology/egr027

Downloaded from https://academic.oup.com/petrology/article-abstract/52/9/1737/1437269/Experimental-Simulation-of-Closed-System-Degassingby gueston 16 September 2017

Previous experimental studies on volatile contents invapour-saturated basaltic melts and vapour/melt partition-ing have focused on the solubilities of single volatile speciesor binary mixtures such as H2O^CO2 (e.g. Dixon et al.,1995; Dixon, 1997; Papale, 1999; Newman & Lowenstern,2002; Behrens et al., 2004; Botcharnikov et al., 2005a, 2006;Papale et al., 2006; Behrens et al., 2009; Shishkina et al.,2010; Lesne et al., 2011a, 2011b), H2O^S (Carroll &Rutherford, 1985, 1988; Luhr, 1990; Carroll & Webster,1994; Mavrogenes & O’Neill, 1999; Clemente et al., 2004;Moune et al., 2009) and H2O^Cl (e.g. Webster et al., 1999,2009; Signorelli & Carroll, 2000; Stelling et al., 2008).Recently, experimental studies have been extended to sili-cate melts containing C^O^H^Cl (Botcharnikov et al.,2007; Alletti et al., 2009) and H^O^S^Cl species(Botcharnikov et al., 2004;Webster et al., 2009). The behav-iour of H2O and CO2 in basaltic systems has been well stu-died (e.g. Dixon et al., 1995; Dixon, 1997; Papale, 1999;Newman & Lowenstern, 2002; Behrens et al., 2004;Botcharnikov et al., 2005a, 2006; Papale et al., 2006;Behrens et al., 2009; Shishkina et al., 2010; Lesne et al.,2011a, 2011b). All these studies have shown that CO2

degasses predominantly from the basaltic melt at highpressures, and H2O starts to degas to any significantdegree only at lower pressures. The addition of S to H2O-and Cl-bearing fluid-saturated melts tends to decrease theCl concentrations in the melt (Botcharnikov et al., 2004;Webster et al., 2009), with corresponding increases in theCl content of the fluids; that is, it tends to increase thefluid/melt partition coefficient of Cl (Dfl=melt

Cl ).The motivation of this study is to provide a dataset

for volatiles dissolved in the melt and coexisting fluidphase as a function of pressure for a fixed bulk compos-ition, providing for the first time experimental datafor the case of closed-system, equilibrium degassing inthe system basalt^H2O^CO2^S^Cl. We focused on twosynthetic basaltic compositions based on tephra eruptedfrom two active, passively degassing, subduction-related volcanoes: Stromboli (Italy) and Masaya(Nicaragua). Our experimental results are compared withnatural melt inclusions and gas emissions from both volca-noes to provide a link between high-P evidence from meltinclusions (MI) and the geochemistry of surface gasdischarges.Stromboli is the northernmost volcano of the Aeolian

archipelago, renowned for its persistent activity (Rosiet al., 2000), erupting shoshonitic to high-K basalts(Francalanci et al., 2004). It is an exceptionally well-studiedvolcano for which there exists an extensive database ofwhole-rock and mineral analyses as well as volatile elementanalyses of olivine-hosted melt inclusions (e.g. Me¤ trichet al., 2001; Bertagnini et al., 2003; Francalanci et al., 2004)and gas emission data (Allard et al., 1994; Burton et al.,2007a, 2007b; Aiuppa et al., 2009, 2010).

Masaya, on the Central American volcanic front(Burton et al., 2000) is one of the few basaltic volcanoesknown for its Plinian activity (Williams, 1983). Gas emis-sion crises have recurred in the historical period at�25 year intervals (Stoiber et al., 1986). The geochemicaland mineralogical database for Masaya is much less com-plete than that for Stromboli. However, a few papers (e.g.Sadofsky et al., 2008) give the composition of bulk-rocks,and the gas composition measured on Masaya through dif-ferent activity periods (e.g. Duffell et al., 2003). Melt inclu-sion data are available from Horrocks (2001) and Atlas &Dixon (2006). The passive, but persistent, degassing activ-ity of these two volcanoes provides an excellent exampleof a volcanic system where magmatic fluids are thought tohave equilibrated with silicate melts and rocks in themagma chamber and conduit. Thus, the concentrations ofvolatiles measured in melt inclusions and volcanic gasesfrom these volcanoes can provide quantitative referencedata that can be directly interpreted by experimentalsimulations of equilibrated systems.

EXPER IMENTAL TECHNIQUESTwo synthetic basalt compositions were chosen for theexperiments. The first (St8.1) is based on a crystal-poor,high-K basaltic ‘golden pumice’ from Stromboli [sampleSt8.1 of Bertagnini et al. (2003)], analogous to pumicesemitted during both paroxysms and major explosions. Thesecond (MAS.1) is a based on a basaltic lapilli samplefrom Masaya [sample P2-47 of Sadofsky et al. (2008)]. Thetwo starting compositions are considered to be broadlyrepresentative of the parental magma at depth beneatheach volcano.The starting material consisted of a mechanical mixture

of synthetic oxides (SiO2, TiO2, Al2O3, Fe2O3, MgO) andcarbonates (CaCO3, K2CO3, Na2CO3). Oxides and car-bonates were stored in a drying oven at 1208C prior tomixing in appropriate proportions, and homogenized bygrinding in an agate mortar. The mixture was decarbo-nated at 700^10008C for 6 h in alumina crucibles. Pressedpellets of the starting materials were reduced in agas-mixing furnace, at 10008C for 2 h at an fO2 close tothe intrinsic fO2 of the internally heated pressure vessel(IHPV) of Leibniz University of Hannover.This step is de-signed to minimize gradients in hydrogen fugacity (fH2)during the experiments and consequent H2O loss or gainby redox reactions of iron-bearing components.Structurally bound volatiles were added to the oxide

mixture after the decarbonation step as follows: H2O wasadded as brucite [Mg(OH)2], CO2 as CaCO3, sulphur asgypsum (CaSO4.2H2O) and chlorine as NaCl. Volatileaddition was performed in such a way as to achieve thedesired volatile-free bulk composition. Initial volatile con-tents in both mixtures are representative of the highestvolatile contents measured in melt inclusions from basalts

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 9 SEPTEMBER 2011

1738Downloaded from https://academic.oup.com/petrology/article-abstract/52/9/1737/1437269/Experimental-Simulation-of-Closed-System-Degassingby gueston 16 September 2017

from the two volcanoes: Stromboli data are fromBertagnini et al. (2003) and Me¤ trich et al. (2010); Masayadata from Sadofsky et al. (2008). As no CO2 data forMasaya were presented by Sadofsky et al. (2008), we usedthe highest values (�7000 ppm) reported by Atlas &Dixon (2006). For each volcano two mixtures were pre-pared with different initial sulphur contents to better inves-tigate the behaviour of sulphur and its potential influenceon the behaviour of other volatiles. These mixes arelabelled A (low-sulphur) and B (high-sulphur) and con-tain 1930 and 3560 ppm for St8.1 and 590 and 1400 ppmfor MAS.1, respectively. The sulphur and carbon contentsof the starting materials were measured using an ELTRACS 800 analyzer at Leibniz University, Hannover; theirH2O contents were measured by Karl-Fischer titration(see below). Initial chlorine contents were determined bymelting the starting materials in a g inch piston-cylinderat the University of Bristol at 1.2GPa, 13758C for 1·5 h,with subsequent measurements by electron microprobeanalysis (EMPA) on the resultant glasses. These conditionswere chosen to ensure that the entire volatile budget re-mained in solution. All the above measurements of volatileconcentrations are preferred to the gravimetric estimates

for the starting proportions of reagents, although agree-ment is generally good. Major elements and initial volatilescompositions of all starting material are reported inTable 1.Between 30 and 50mg of the starting material were

loaded into Au80Pd20 capsules of 15mm length, 2·5mminner diameter and 2·9mm outer diameter, and weldedshut. Prior to loading, the capsules were annealed, cleanedin HCl in a heated bath for 1h, rinsed with distilledwater, and then cleaned again in heated distilled H2O for1h. The capsules were weighed before and after weldingand then placed for 2 h in an oven at 1208C to check forleakage.All equilibrium experiments were performed at 11508C;

that is, at a temperature corresponding to super-liquidusconditions in the investigated systems from 25 to400MPa, a pressure range consistent with the entrapmentpressures for melt inclusions from Stromboli (Me¤ trichet al., 2001; Bertagnini et al., 2003; Di Carlo et al., 2006;Pichavant et al., 2009). Experiments were performed in avertically run IHPV at Leibniz University, Hannover,using pure argon as the pressure medium. Four to six cap-sules were run simultaneously, hanging from a Pt wire

Table 1: Experimental starting compositions (in wt %, except for CO2, S and Cl, expressed in ppm)

Stromboli Masaya

St8.1.A St8.1.B MAS.1.A MAS.1.B

Start. mat. Av. glass s Start. mat. Av. glass s Start. mat. Av. glass s Start. mat. Av. glass s

n¼ 9 n¼ 9 n¼ 11 n¼ 8

SiO2 50·02 51·64 0·31 50·07 51·86 0·28 49·39 50·86 0·30 49·42 50·84 0·24

TiO2 0·86 0·81 0·05 0·86 0·84 0·05 1·24 1·17 0·06 1·24 1·18 0·06

Al2O3 19·06 18·57 0·15 19·13 18·63 0·16 19·16 18·91 0·28 19·18 18·76 0·20

Fe2O3 8·37 8·15 0·12 8·37 8·12 0·14 13·26 12·42 0·17 13·27 12·44 0·19

MgO 6·70 6·45 0·11 6·24 6·02 0·12 3·58 3·41 0·08 3·34 3·17 0·07

CaO 10·50 10·90 0·12 10·85 11·11 0·12 8·53 9·23 0·09 8·70 9·41 0·11

Na2O 2·71 2·05 0·07 2·67 1·98 0·09 3·31 2·77 0·09 3·34 2·90 0·11

K2O 1·80 1·44 0·08 1·80 1·44 0·05 1·51 1·23 0·04 1·51 1·29 0·05

Fe2þ/Fetot1 0·022 0·022 0·025 0·025

H2O2 2·70 2·92 1·54 1·66

ppm CO23 4600 4890 7410 6850

ppm S3 1930 3560 595 1400

ppm Cl4 2040 1570 1700 1420

All analyses are normalized to 100%. Start. mat., composition of starting material; major elements from mass of reagents,volatiles as described in footnotes 1–4. Av. glass, average composition of experimental run product glasses. s, standarddeviation. n, number of analyses averaged.1Fe2þ/Fetot measured by colorimetric wet-chemistry on initial powder using the technique of Schuessler et al. (2008).2Initial H2O determined by KFT on initial powder.3Determined by CS analyzer on initial powder.4Determined by EMPA on fused glass in piston-cylinder apparatus.

LESNE et al. CLOSED-SYSTEM DEGASSING


near the top of a double-wound molybdenum wire furnace.Temperature was measured by four S-type thermocouples,placed along the 3 cm hot zone. The temperature differ-ence between the top and bottom of the capsules wasalways less than 108C. Pressure was monitored by a digitalpressure transducer with an uncertainty of about 1MPa.The variation of pressure during the experiments was�5MPa. Experiments lasted between 6 and 12 h, andwere ended by drop-quenching, with a cooling rate of�108C s�1 (Berndt et al., 2002). These run times werechosen to ensure attainment of equilibrium in the system,but minimize possible S and Fe loss to the capsule walls.After quenching, the capsules were weighed to verify thatthey had remained sealed during the experiment. Thevapour-saturated conditions of the experimental chargeswere confirmed by the presence of water hissing out fromthe capsules during opening, together with the wet appear-ance of the recovered glass chips. Quenched glasses weremounted in epoxy resin and polished with diamond pastefor subsequent electron microprobe analysis. The pressure,temperature and duration of all experiments performed inthis study are reported inTable 2.

ANALYT ICAL TECHNIQUESScanning electron microscope (SEM)All charges were examined using a Hitachi S-3500N SEMat the University of Bristol in back-scattered electronmode to check for the occurrence of quench crystallizationor bubble formation.

Electron microprobe analysis (EMPA)Experimental run products were analyzed by EMPA atBristol University using a Cameca SX100 electron micro-probe. Major elements were analysed separately from

sulphur and chlorine. Analytical conditions applied were15 kV accelerating voltage, 6 nA sample current, beamdiameter of 10 mm and peak counting time of 10 s. For Cl,an acceleration voltage of 15 kV, a 10 nA beam current, abeam diameter of 10 mm and counting time on peak of60 s were applied. NaCl was used as a standard for chlorineand pyrite was used as a standard for sulphur.To minimizealkali migration during the analysis of hydrous, alkali-richglasses, the analytical conditions were 15 kV, 4 nA and 5 stotal counting time for Na and K, which were analyzedfirst. A ZAF correction procedure was applied. Majorelement calibration utilized wollastonite (Ca), hematite(Fe), albite (Na, Si), corundum (Al), olivine (Mg) andorthoclase (K) standards. Multiple measurements weremade for each sample to check for homogeneity.Some samples were also analysed for their sulphur oxi-

dation state. Measurements of l(SKa) were performed ontwo spectrometers of the Cameca SX-100 microprobe atthe University of Bristol following the method reported byCarroll & Rutherford (1988), Wallace & Carmichael(1994), Me¤ trich & Clocchiatti (1996) and Jugo et al. (2005).PET crystals (2d¼ 8·742—) were used on each spectrom-eter to simultaneously obtain two independent measure-ments of the SKa wavelength shift. Standards used wereFeS (sulphide, S2�) and BaSO4 (sulphate, S6þ). EMPAoperating conditions were 20 kV and 25 nA with a spotdiameter of 15 mm. Each spectrometer was moved 0·00004sin � units for 100 steps over the range of 0·61198^0·61594sin � during a single spot analysis. For each step, countingtime varied between 100 and 1600ms, which results in amaximum beam exposure time of 10^160 s. Sixteen waves-can spectra were collected for counting times of 100msfor each step, eight spectra for 200ms, four for 400ms,two for 800ms and one for 1600ms. Each summed spec-trum was fitted with a Gaussian function to obtain thepeak position. The resulting SKa wavelength shifts foreach spectrometer are calculated as the difference of theFeS standard value (reported as —�103):

�lðSKaÞsample ¼ lðSKaÞsample� lðSKaÞsulphide ð1Þ

�lðSKaÞstandard ¼ lðSKaÞanhydrite� lðSKaÞsulphide:

ð2Þ

The proportion of sulphur that is sulphate (S6þ/�S) in asample can be calculated relative to the peak shift of thebarite standard:

S6þ=X

S ¼ �lðSKaÞsample=�lðSKaÞstandard: ð3Þ

Analyses of samples and standards showed different re-sults for the two crystal spectrometers used; a betterGaussian fit was obtained with spectrometer 1 than spec-trometer 2, hence the former was used for all analyses re-ported here.

Table 2: Experimental run conditions

Experiment no. P (MPa) T (8C) Time (h)

1 200 1150 6

2 100 1150 6

3 250 1150 12

4 300 1150 12

5 150 1150 6

6 200 1150 12

7 50 1150 6

8 50 1150 6

9 25 1150 15

10 400 1150 20

All four starting materials were run at each condition,hence St8.1.A10 denotes starting composition St8.1.Arun in experiment 10.



Fourier transform infra-redspectroscopy (FTIR)IR spectra were recorded with a Bruker IFS 88 FTIRspectrometer coupled with an IR-scopeII microscope atLeibniz University, Hannover, with the following operat-ing conditions: MCT narrow range detector for bothmid-infrared (MIR) and near-infrared (NIR); globarlight source and KBr beamsplitter for MIR; tungstenlamp and CaF2 beamsplitter for NIR. Spectral resolutionwas 1cm�1 in MIR and 4 cm�1 in NIR.To minimize the ef-fects of varying atmospheric CO2, the sample stage of theIR microscope was shielded and purged with dry air.Fifty and 100 scans, respectively, were used to obtain MIRand NIR spectra with good signal/noise ratio.Water dissolved in glasses was analyzed by using the

bands at 5200 and 4500 cm�1. The spectra were obtainedon doubly polished glass chips (thickness �200 mm); 2^4spots were analyzed from each sample. To calculate H2Odissolved in the melt, both as molecular H2O and hydroxylgroups, we referred to the work of Shishkina et al. (2010),who calibrated molar absorption coefficients (e) fortholeiitic basaltic melt (eH2O¼ 0·69� 0·07 andeOH¼ 0·69� 0·07 Lcm�1mol�1). A tangential baseline cor-rection was drawn to measure the height of the peaks. Theaccuracy of our FTIR analysis is estimated to be betterthan �20% for H2O. Behrens et al. (2009) stated thatFTIR reproduces Karl-Fischer titration (KFT) datawithin �6% relative.The concentrations of CO2 dissolved in quenched glasses

were determined by measuring the heights of peaks ofCO2�

3 at 1430 cm�1 (to avoid overlap of the H2O peak at1630 cm�1 with the CO2�

3 peak at 1520 cm�1; seeBotcharnikov et al., 2006) after subtracting a carbonate-freespectrum obtained from a volatile-free basaltic sample, ad-justed to the same sample thickness. For the molar absorp-tion coefficient at 1430 cm�1, we used the value of Fine &Stolper (1986): 375�20 Lcm�1mol�1, for consistency withprevious FTIR studies on similar melt compositions(Pichavant et al., 2009; Lesne et al., 2011b).Density measurements, required for determination of

volatile concentrations in glasses, are reliable only if meas-urements are made using a single piece of glass more than15mg in mass. However, owing to the low initial amountof powder loaded (30^50mg), and breakage during capsuleopening, no glass fragments 45mg could be recovered.Hence, density was estimated using the linear relationshipgiven by Ohlhorst et al. (2001) for a basaltic glass, assuminga partial molar volume of H2O in the glass of12·0� 0·5 cm3mol�1 (Richet et al., 2000):

r ¼ ð � 20 � 8� 6 � 6ÞcH2O þ 2819� 13 � 5 ð4Þ

where r is the density of the sample in kgm�3 and cH2O isthe concentration of water in the glass in wt %. The dens-ity for each quenched glass was calculated by iteration.

Sample thicknesses were determined by micrometer to aprecision of �3 mm. The run-product glasses are not ves-icular (�1 vol. % bubbles) and the presence of so few bub-bles does not influence the FTIR analyses.

Karl-Fischer titration (KFT)The initial H2O contents of the starting materials weremeasured by Karl-Fischer titration at the LeibnizUniversity, Hannover, using a procedure similar to that de-tailed by Behrens (1995). Starting powders were loaded ina Pt crucible and heated to 13008C with an induction fur-nace to extract dissolved water. All water released was con-ducted by an Ar flux to a CuO furnace to convert anyH-bearing species present into H2O, which was analyzedin the titration cell by a coulometric method (Behrens,1995). It is known that about 0·10wt % of unextractedH2O is found in samples containing initially more than1·5wt % H2O (Behrens, 1995). However, the compositionsfor which unextracted water was observed are more silicicthan our basaltic melts (Behrens et al., 2004), hence we didnot make any correction for unextracted water in ourKFT analyses. Instead, an additional uncertainty of�0·10wt %, originating from possible incomplete degas-sing, has been incorporated into the error propagation cal-culations. Although KFT results are highly reproducible(Behrens, 1995), KFT is a destructive technique that re-quires a large sample, and therefore could not be used todetermine total water dissolved in experimental quenchedglasses because of insufficient material.

Secondary ion mass spectrometry (SIMS)SIMS was applied to experimental glasses whose H2O andCO2 concentrations were below detection by FTIR. Inaddition we analysed several glasses by SIMS that hadalso been analysed by FTIR to evaluate the consistency ofthe two techniques. SIMS analyses were carried out onAu-coated grain mounts using a Cameca IMS-4f ionmicroprobe at the University of Edinburgh. Single glassfragments were polished and multiply mounted in indiumblocks to minimize the carbon background from epoxy.Background was further minimized by pumping down toa vacuum of �10�9 Torr in a custom-built airlock attachedto the sample chamber. Operating conditions were 10 kVO� primary-beam with 5 nA current at the sample sur-face. Positive secondary ions were extracted at 4·5 kV withan offset of 50V (for C) and 75V (for H) to reduce trans-mission of molecular ions. A mass resolving power of 1500was used to eliminate interferences of 24Mg2þ on 12Cþ. Tocalibrate H2O and CO2 we followed the method ofBlundy & Cashman (2008), by analyzing basaltic glassstandards with known H2O and CO2 contents, and bybuilding working curves of 1H/30Si vs H2O and 12C/30Si vsCO2. A total of eight basaltic glass standards were run,covering a range of 0·15^4·6wt % H2O and 0^1000 ppmCO2. Calibration was performed afresh on each analysis



day and uncertainties on the working curve were propa-gated through to calculate uncertainty on dissolved H2Oand CO2.For glasses analysed by both SIMS and FTIR we

observed a consistent offset in H2O contents, with FTIRgiving values 18·5% relative higher than SIMS (Fig. 1).Weattribute this difference to uncertainty in the FTIR ab-sorption coefficients, which are very sensitive to bulk com-position. This affects accuracy rather than precision. AsSIMS analysis of H2O is less compositionally sensitivethan FTIR, and given that SIMS analyses of the mostH2O-rich glasses give better overall agreement with theinitial H2O content as determined by KFT, we havereduced all FTIR H2O data by a factor of 1·185. Both theoriginal and corrected H2O data are reported in Table 3,but only the latter are used in the figures and discussion.For CO2 agreement between SIMS and FTIR showed nosuch offset, suggesting that the absorption coefficients ofFine & Stolper (1986) are well suited to our glass compos-itions. Consequently no correction was made to the TIRCO2 data.

Iron redox state and implications forwater gain and lossIn an iron-bearing experimental charge open to hydrogenexchange through the capsule walls, reduction of ferric

iron can produce water via the equilibrium

Fe2O3ðmeltÞ þH2ðvapÞ ¼ 2FeOðmeltÞ þH2OðvapÞ:

ð5Þ

Where the hydrogen fugacity during the experimentalrun differs from that of the starting material synthesis, H2

diffusion will significantly modify the original H2O con-tent of the capsule. As we need to know precisely the totalvolatile content of the system (meltþ fluid) to calculatethe fluid composition by mass balance (see below) we mea-sured the Fe2þ/Fetot ratios in the starting material and theexperimental charges. By comparing the Fe2þ/Fetot of therun products with that of the starting materials it is pos-sible to calculate any H2O added to (or lost from) thestarting material via reaction (5). The Fe2þ/Fetot ratio ofour quenched melts was determined by colorimetricwet-chemical analysis, following the method of Schuessleret al. (2008). They estimated the uncertainty on Fe2þ/Fetotratios to be �0·03. Fe2þ/Fetot ratios of our starting mater-ials, measured using the same technique, are reported inTable 1.

RESULTS AND DI SCUSSIONRun productsAll quenched glasses from the St8.1 starting compositionare brown and free of crystals. A few bubbles are present

y = 1.1854x R2 = 0.8845

0.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50

4.00

4.50

0.00 1.00 2.00 3.00 4.00

H2O (wt%) measured by SIMS

H2O

(w

t%)

mea

sure

d by

FT

IR

St8.1.A

St8.1.BMAS.1.A

MAS.1.B

Fig. 1. Comparison of H2O contents of experimental glasses measured by both FTIR and SIMS. Deviation from a 1:1 line is attributed to uncer-tainty in the extinction coefficient used for processing the FTIR data. A correction factor of 1·1854 has been applied to all of the FTIR data tobring them into line with SIMS data. No such correction was required for CO2 data.



Table 3: Experimental results

P (MPa) Run �NNO Fe2þ/Fetot H2O

added

by iron

reduction

Volatiles dissolved in the melt Volatiles dissolved in the fluid (mol %)

H2O1

(wt %)

s H2O2

(wt %)

CO2

(ppm)

s S3

(ppm)

s Cl3

(ppm)

s H2O s CO2 s S s Cl s Min. of

S6þ/Stot

St8.1.A

26 St8.1.A9 n.d. 0·744 1·73 0·15 1·47 455 15 115 41 2026 82 87·2 16·2 8·25 0·92 4·53 0·53 0·03 0·35

52 St8·1.A8a n.d. 0·744 2·44 0·15 2·08 n.d. 751 57 1994 60 n.d. n.d. n.d. n.d.

52 St8·1.A8b n.d. 0·744 2·53 0·15 2·15 555 4 925 66 2430 179 84·1 23·8 12·2 2·0 3·71 0·68 n.c.

150 St8·1.A5 2·4 0·60 0·69 3·03 0·15 2·57 3001 30 1719 155 1979 134 81·0 34·4 17·5 4·0 1·19 0·95 0·31 0·95 0·84

200 St8·1.A1 2·0 0·64 0·73 3·18 0·22 2·70 5151 43 1733 13 1730 138 78·9 42·4 18·1 5·1 1·21 0·45 1·71 1·16 0·76

200 St8·1.A6 1·8 0·744 3·18 0·15 2·70 4851 48 1782 142 2071 142 80·7 37·8 18·4 4·6 0·92 0·95 n.c. 0·84

250 St8·1.A3 1·6 0·69 0·77 3·37 0·15 2·86 5751 57 1595 156 1963 162 76·4 42·4 20·7 5·8 2·38 1·33 0·50 1·34 0·82

300 St8·1.A4 1·7 0·67 0·76 3·59 0·15 3·05 9601 96 1721 179 1961 142 71·1 57·3 26·1 9·6 2·08 1·97 0·71 1·74 0·76

406 St8·1.A10 n.d. 0·68 0·744 n.d. 3·17 15405 48 1564 142 2141 79 64·9 77·8 30·1 14·8 4·98 3·19 n.c. 0·82

St8·1.B

26 St8·1.B9 n.d. 0·754 1·21 0·15 1·03 405 2 36 41 1485 50 86·8 7·1 6·5 0·3 6·52 0·36 0·14 0·25

50 St8·1.B7 1·7 0·68 0·77 2·20 0·16 1·87 605 5 1607 314 1476 136 85·4 10·7 9·3 0·7 5·14 0·94 0·23 0·47

52 St8·1.B8 n.d. 0·754 2·40 0·15 2·04 655 6 1189 73 1533 61 83·0 10·7 10·1 0·7 6·81 0·59 0·10 0·40

100 St8·1.B2 1·8 0·66 0·75 2·83 0·15 2·40 601 6 3043 201 1537 337 84·7 14·2 13·2 1·3 1·95 0·84 0·12 1·24

150 St8·1.B5 2·0 0·63 0·73 3·12 0·15 2·65 2301 23 3170 197 1570 133 82·4 17·1 15·8 1·8 1·81 1·01 0·01 0·82

200 St8·1.B6 n.d. 0·754 3·24 0·15 2·76 4151 41 3230 187 1582 148 81·9 18·5 16·4 2·1 1·65 1·0 n.c.

250 St8·1.B3 1·8 0·66 0·75 3·54 0·15 3·01 6351 64 3240 203 1985 167 77·4 24·4 20·5 3·4 2·13 1·5 n.c.

300 St8·1.B4 1·6 0·68 0·77 3·66 0·15 3·11 9201 92 3140 187 1560 173 75·7 25·9 21·2 3·7 3·07 1·6 0·07 1·48 0·79

406 St8·1.B10 n.d. 0·754 3·98 0·15 3·38 14901 149 3090 67 1635 63 63·5 41·7 30·7 8·1 5·81 2·0 n.c. 0·88

MAS.1.A

25 MAS.1.A9 n.d. 1·174 1·86 0·15 1·58 355 9 30 20 1320 360 76·2 18·2 20·3 2·5 2·14 0·27 1·30 1·58

50 MAS.1.A7 1·0 0·75 1·26 2·39 0·12 2·03 905 4 270 118 1260 159 69·4 21·3 26·9 3·7 1·63 0·64 2·03 1·52

52 MAS.1.A8 n.d. 1·174 2·35 0·15 1·99 855 13 205 32 1265 67 67·6 24·2 28·2 4·3 2·07 0·37 2·10 1·44

52 MAS.1.A8b n.d. 1·174 2·39 0·15 2·03 705 12 170 48 1170 55 66·0 24·8 29·1 4·5 2·32 0·45 2·61 1·49

100 MAS.1.A2 1·7 0·67 1·13 2·51 0·15 2·13 1201 12 395 127 1425 213 62·3 29·2 34·7 6·2 1·31 0·87 1·65 2·13

150 MAS.1.A5 1·7 0·66 1·13 2·61 0·15 2·22 2251 22 410 137 1645 145 59·6 32·7 38·6 7·6 1·36 1·05 0·38 2·14

200 MAS.1.A1 1·5 0·68 1·14 3·03 0·15 2·58 4901 49 480 125 1150 125 24·5 55·6 67·3 18·3 1·52 1·73 6·65 4·17

200 MAS1.A6 n.d. 1·174 2·81 0·15 2·39 3451 34 375 104 1640 178 51·6 38·6 45·9 10·2 1·97 1·04 0·49 2·72

250 MAS.1.A3 1·4 0·70 1·17 3·09 0·15 2·62 6101 61 370 111 1480 153 22·0 60·2 71·8 20·8 3·30 1·88 2·90 4·32

300 MAS.1.A4 1·2 0·72 1·20 2·99 0·15 2·54 7501 75 310 108 1330 244 39·3 46·9 53·8 13·6 3·17 1·46 3·76 3·89

406 MAS.1.A10a n.d. 1·174 3·37 0·15 2·86 14451 145 165 36 1315 79 n.d. n.d. n.d. n.d.

406 MAS.1.A10b n.d. 1·174 3·23 0·09 2·75 13901 139 40 21 1590 61 n.d. n.d. n.d. n.d.

MAS.1.B

26 MAS.1.B9 n.d. 1·204 1·93 0·03 1·64 355 6 60 25 1220 63 77·0 6·7 17·6 0·8 4·76 0·82 0·65 0·50

50 MAS.1.B7 n.d. 1·204 2·15 0·14 1·83 755 13 505 126 1225 143 75·3 14·0 20·3 1·9 3·68 1·10 0·73 0·75 0·67

52 MAS.1.B8 n.d. 1·204 2·03 0·15 1·73 905 13 525 73 1510 179 77·7 13·9 18·9 1·8 3·37 0·95 n.c.

100 MAS.1.B2 1·4 0·69 1·204 2·59 0·15 2·20 651 6 870 119 1470 217 68·2 19·9 28·8 3·7 3·09 1·51 n.c.

150 MAS.1.B5 1·5 0·68 1·16 2·63 0·15 2·24 1951 20 935 192 1245 117 65·5 21·3 30·5 4·1 2·92 1·89 1·01 1·05 0·62

200 MAS.1.B6 n.d. 1·15 2·84 0·15 2·41 n.d. 950 112 1400 163 n.d. n.d. n.d. n.d.

250 MAS.1.B3 1·5 0·68 1·14 2·87 0·15 2·44 4501 45 835 243 1440 140 55·4 27·9 39·9 6·7 4·8 2·9 n.c. 0·70

300 MAS.1.B4 0·1 0·82 1·36 3·00 0·15 2·55 7051 71 605 179 1220 112 60·4 24·0 32·5 4·9 5·8 2·2 1·31 1·20 0·65

1Determined by FTIR.2H2O recalculated from SIMS analyses.3Determined by EMPA.4Fe2þ/Fetot not measured: average value is taken.5Determined by SIMS.Min. of S6þ/Stot is a minimum of oxidation state of sulphur was measured by EMPA (see text for explanation). n.d., notdetermined. n.c., not calculable. Volatiles dissolved in the fluid are expressed in mol %.



(less than 1 vol. %), mostly at the rim of the sample.Quenched glasses from Masaya are dark, owing to theirhigher iron contents, again with a few bubbles at the rimsof the samples, but occasionally within the glass. In mostMAS.1 glasses we noticed the presence of few tiny ironoxide crystals (around 1 vol. %). Melt compositions forvolatile elements in each run are reported in Table 3.Major element analyses are given as Supplementary Data(available for downloading at http://www.petrology.oxfordjournals.org/). Based on comparison of starting ma-terials and experimental glasses (Table 1), iron loss to thecapsule is calculated to be �6% relative for MAS.1 and�3% relative for St8.1, which is at the low end of therange reported by Di Carlo et al. (2006) for similar startingmaterials and run conditions.

Fe2þ/Fetot: iron reduction, H2O production, redox state

Fe2þ/Fetot analyses are reported in Table 1 for the startingmaterials and inTable 3 for Fe2þ/Fetot measured in the ex-perimental glasses. Results obtained from the starting ma-terials showed that the Fe2þ/Fetot ratios are 0·022 for St8.1and 0·025 for MAS.1. Experimental charges show signifi-cantly higher values of Fe2þ/Fetot as a result of H2 ingressthrough the capsule walls: from 0·596 for St8.1.A5 to0·681 for St8.1.B7 and St8.1.B4; and from 0·662 forMAS.1.A5 to 0·818 for MAS.1.B4. This is a consequence ofsynthesizing our starting materials at an fO2 higher thanthat prevailing in the IHPV apparatus during theexperiments.The change of Fe2þ/Fetot ratios between the starting ma-

terials and the experimental charges reveals significantwater production through iron reduction, via reaction (5).This additional water must be added to that measured byKFT in the starting materials to constrain the proportionsof volatile components via mass balance. Thus in the St8.1samples, an average of 0·75wt % of H2O is producedduring the experiments, whereas in the MAS.1 samples,an average of 1·17 wt % H2O is produced in MAS.1.Aand an average of 1·2wt % H2O in MAS.1.B. The differ-ence in water production between St8.1 and the MAS.1starting materials may be attributable to the higher ironcontent of the latter (Table 1). The amount of H2O addedby reaction (5) in each run where Fe2þ/Fetot was measuredis given inTable 3.Where no Fe2þ/Fetot value was measuredwe took an average of the Fe2þ/Fetot measured for the dif-ferent compositions.The measured Fe2þ/Fetot ratios can be used to calculate

the experimental redox conditions following empirical re-lationships (Sack et al., 1980; Kilinc et al., 1983; Kress &Carmichael, 1988, 1991) of the form

ln½XFe2O3=XFeO� ¼ a ln fO2 þ b=T þ cþX

Xidi ð6Þ

where a, b, c and d are constants found by regression of alarge number and variety of silicate melts equilibrated

from air to almost the iron^wu« stite oxygen buffer, over arange of temperatures at 1bar. Although this equationwas calibrated for dry systems, the effect of dissolvedwater on the Fe2þ/Fetot ratio at a given fO2 is negligible(Botcharnikov et al., 2005b). Calculated fO2 varies betweenNNOþ1·5 and NNOþ 2·4 for the St8.1 experiments andNNOþ1·0 and NNOþ1·7 for MAS.1 (where NNO is thenickel^nickel oxide buffer). Kress & Carmichael (1991)gave a standard error of 0·21 for the ln[XFe2O3/XFeO] calcu-lated, giving uncertainties in measured fO2 of �0·86 logunits. Calculated fO2 values are given in Table 3. Not allof the experimental charges could be analyzed, because ofthe paucity of product, but results show that the St8.1.Asamples are always more oxidized than St8.1.B, withthe exception of the experiments performed at 250 MPa.The same observation is made for the MAS.1 A and Bsamples.

Volatiles dissolved in the meltThe amount of H2O dissolved in the melt is pressure de-pendent, as shown in Figs 2a and 3a. No systematic differ-ences in water content were measured betweensulphur-rich (St8.1.B and MAS.1.B) and sulphur-poor(St8.1.A and MAS.1.A) starting materials, indicating thatS has a negligible effect on H2O solubility. At pressuresabove 100MPa, H2O contents in St8.1 are 20% higherthan in MAS.1, whereas at 50MPa H2O contents are simi-lar in both compositions (around 2·6wt %). At 25MPaMAS.1 dissolves 15^38% relative more H2O than St8.1.We attribute these differences to melt compositional fac-tors. A noteworthy feature is the good agreement betweenH2O measured in the highest pressure (400MPa) glassesby SIMS and the bulk H2O content as determined by acombination of KFT in the starting material and H2Ofrom iron reduction. This vindicates our decision to usethe SIMS values of H2O to correct the FTIR data (seeabove).CO2 dissolved in experimental glasses also shows a

strong pressure dependence (Figs 2b and 3b). For bothbulk compositions, dissolved CO2 concentrations are simi-lar, again irrespective of the initial sulphur contents. CO2

contents are very similar for both St8.1 and MAS.1 overthe entire pressure range, with the exception of the experi-ment at 300MPa, where St8.1 contains about 20% moreCO2.Sulphur concentration (reported as S total) measured in

the St8.1 experiments remains constant (or increases veryslightly) from 400MPa to 100MPa (Fig. 2d) and then de-creases sharply from 100MPa to 25MPa for both initialsulphur concentrations. The maximum sulphur contentsdissolved in the melt were measured in samples synthe-sized at �200MPa for both the S-rich (St8.1.B6 andSt8.1.B3, �3200 ppm) and the S-poor (St8.1.A1 andSt8.1.A6, �1800 ppm) starting materials. The maximummeasured sulphur matched the initial sulphur added to



the system, suggesting that the S loss to the capsule is negli-gible. Results for MAS.1 (Fig. 3d) show a more pronouncedincrease in dissolved S from 400 to 150MPa, although thedata are less precise. Dissolved S decreases sharply below

100MPa. As for St8.1, the highest sulphur contents dis-solved were measured in experiments performed at�200MPa. The single MAS.1 run at 400MPa shows thesame sulphur content as at the lowest pressure (25MPa).

(f)

0.00

0.50

1.00

1.50

2.00

2.50

0 100 200 300 400 500

Ptot (MPa)

S/C

l mel

t

(e)

0

500

1000

1500

2000

2500

3000

0 100 200 300 400 500

Ptot (MPa)

Cl (

ppm

)

(d)

0

500

1000

1500

2000

2500

3000

3500

4000

0 100 200 300 400 500

Ptot (MPa)

S (p

pm)

H2O (ppm)

CO

2 (pp

m)

(c)

0200400600800

100012001400160018002000

0.00 1.00 2.00 3.00 4.00

0200400600800

100012001400160018002000

0 100 200 300 400 500

Ptot (MPa)

CO

2 (pp

m)

(b)

0 100 200 300 400 500

Ptot (MPa)

H2O

(w

t%)

(a)

STROMBOLI

0.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50

4.00

St8.1.ASt8.1.B

Fig. 2. Volatiles dissolved in quenched glasses from experiments on Stromboli basalts St8.1.A (filled squares) and in St8.1.B (open squares) as afunction of pressure. The data are fromTable 3. (a) H2O, measured by FTIR and corrected by SIMS measurements (see Fig. 1); (b) CO2, mea-sured by FTIR or SIMS; (c) CO2 vs H2O, defining a degassing trend; (d) S, determined by EMPA; (e) Cl determined by EMPA; (f) S/Clratios in melts. Continuous (St8.1.A) and dashed (St8.1.B) horizontal lines represent the initial volatiles added, with their uncertainties(Table 1). Shaded grey fields bracket the range of initial S, Cl and S/Cl.



Sulphur concentration does not appear to show any sys-tematic variation with redox conditions (Fig. 4), partly be-cause of the large analytical error on the calculated fO2.That aside, the highest values of sulphur dissolved in themelt were measured in the more oxidized experimentalcharges.

In terms of sulphur speciation, expressed as S6þ/�S, ourresults show that under oxidized conditions(fO24NNOþ 2) 80% of the total sulphur occurs as sul-phate (compare Carroll & Rutherford, 1988; Jugo et al.,2010), whereas at slightly lower fO2, (NNOþ1·5) �65%of the sulphur is sulphate (Table 3). Figure 5 shows the

0.000

Ptot (MPa)

S/C

l mel

t

0.000.100.200.300.400.500.600.700.800.901.00

0 100 200 300 400 500

(f)

(e)

0200400600800

100012001400160018002000

0 100 200 300 400 500

Ptot (MPa)

Cl

(ppm

)

(d)

0

200

400

600

800

1000

1200

1400

0 100 200 300 400 500

Ptot (MPa)

S (

ppm

)

H2O (wt%)

CO

2 (p

pm)

(c)

0200400600800

100012001400160018002000

0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50

0 100 200 300 400 500

Ptot (MPa)

CO

2 (p

pm)

0200400600800

100012001400160018002000

(b)

(a)

Ptot (MPa)

H2O

(w

t%)

0 100 200 300 400 5000.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50

MASAYA

MAS.1.AMAS.1.B

Fig. 3. Volatiles dissolved in quenched glasses from experiments on Masaya basalts MAS.1.A (filled diamonds) and in MAS.1.B (open dia-monds) as a function of pressure. Data from Table 3. (a) H2O, measured with FTIR and corrected by SIMS measurements (see Fig. 1);(b) CO2, measured with FTIR or SIMS; (c) CO2 vs H2O, defining a degassing trend: (d) S, determined by EMPA; (e) Cl determined byEMPA; (f) S/Cl ratio in melts. Continuous (MAS-1.A) and dashed (MAS.1.B) lines represent the initial volatiles added, with uncertainties;shaded grey field bracket the range.



speciation obtained for some samples of St8.1.A and MAS.1B. It appears that the shorter the analytical duration, thebetter the results are. Longer duration analyses result inbeam damage leading to large discrepancies in S6þ/�S.Therefore, the data obtained with 16 wavescans of 100msare the most reliable. None the less, even for reduced dur-ation analyses it is likely that S6þ/�S is underestimated,and the values given in Table 3 should be considered asminima.For both St8.1 and MAS.1, nearly all the initial chlorine

added to the charges remains in the melt from 400MPa to25MPa (Figs 2e and 3e). In a single St8.1 experiment at50MPa (St8.1.A8b) Cl in the glass is slightly higher thanthe bulk Cl, as determined by EMPA.We attribute this toeither analytical error or a small amount of accidentalNaCl contamination of the starting material. The consist-ent values of dissolved Cl indicate minimal loss to thevapour over a wide pressure range. Consequently, S/Clratios in the glass decrease markedly at low pressure(Figs 2f and 3f).

FLU ID -PHASE COMPOSIT IONSAND VOLAT ILE PARTIT IONINGBETWEEN MELT AND FLU IDThe fluid phase composition could not be analyzed direct-ly, so mass-balance calculations were used instead. This re-quires accurate knowledge of the initial amounts of

volatiles loaded, plus the additional H2O content intro-duced via reaction (5), and of the concentrations of vola-tiles dissolved in the melt. As detailed above, the mostreliable estimates were used to constrain initial volatilecontents, including the effects of H2-mediated ferric ironreduction. We assume that there is no volatile loss duringthe experiments, and that there is no reaction between thevolatiles and the capsule. The principal uncertainty relatesto sulphur, because it is known that S easily reacts withthe noble metals of the capsules, and in particular withPd. No specific measurement on the S content of the cap-sules was made. As it is impossible to calculate theamount of S dissolved in the capsule, we considered theloss of S to be minimal, as our experiment duration wasrelatively short and the experiments were run at relativelyoxidized conditions (Table 2). At such conditions most S ispresent in the melt as sulphate, limiting the possible reac-tions with the metal capsules. Confirmation of the min-imal S loss to the capsules comes from the observationthat for the St8.1 experiments, the sulphur dissolved in themelt remains almost constant between 100 and 400MPa,whatever the run duration and fO2. Nevertheless, we em-phasize that our calculated S concentrations in the fluidphase provide maximum values.The concentration of each volatile in the fluid phase can

be calculated via mass balance as described by the follow-ing equation, assuming that no SiO2 or other major oxideis dissolved in the fluid phase (as no change in major

: error on ΔNNO

0

500

1000

1500

2000

2500

3000

3500

4000

0.00 0.50 1.00 1.50 2.00 2.50

ΔNNO

S mel

t (pp

m)

St8.1.A St8.1.BMAS.1.A MAS.1.B

Fig. 4. Variation of the sulphur dissolved in experimental glasses with fO2 (log units relative to NNO buffer) calculated from the measuredFe2þ/�Fe ratio. Data fromTable 3.



oxide concentrations appears between each experimentalrun):

X fluidi ¼

ðminitiali � mmelt

i Þ

Mi

100P ðminitial

i�mmelt

iÞ

Mi

ð7Þ

where Xi is the mole per cent of species, minitiali and mmelt

i

are the weight per cent of volatile species added initiallyand measured in the melt, respectively, Mi is the molar

mass of the volatile species considered and � denotes thesum over Cl, S, H2O and CO2.

H2O^CO2

There are several potential sources of uncertainty on theinitial and final water contents, and the fraction of waterin the fluid phase. (1) Errors on the total amount of watermeasured in the melt are propagated through the uncer-tainty on the water measured by FTIR (10% relative)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0 500 1000 1500 2000

S6+/Σ

S

measurement time (ms)

B3B4B5B6B7

(b)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 500 1000 1500 2000

S6+/Σ

S

measurement time (ms)

A10

A1A3A4A5A6

(a)

Fig. 5. Sulphur speciation, as S6þ/�S, measured for some (a) St8.1.A and (b) MAS.1.B samples obtained for different measurement times usingspectrometer 1 of the Cameca-SX 100 electron microprobe. Numbers denote different experimental runs, as listed inTable 2. The convergenceat low measurement times should be noted. These values of S6þ/�S represent minimum estimates of the true S6þ/�S.



and/or SIMS (58%), and the SIMS^FTIR correctionfactor (2·6%). (2) Additional water could have been ad-sorbed onto powdered starting materials. However, agood agreement between H2O in the highest pressureruns and calculated initial water (KFTanalysis plus ironreduction) suggests that the effect of this process is negli-gible. (3) H2O could be produced by H2-mediated sulphatereduction within the charge. The lack of a correlation be-tween dissolved S and fO2, and the relatively oxidizednature of our runs suggest that such reduction process isalso not significant. Uncertainties on the calculated fluidcomposition increase as pressure increases, owing to thegreater solubility and smaller volume of exsolved vapour.Calculated fluid phase compositions for the St8.1 andMAS.1 experiments, with fully propagated calculateduncertainties, are shown in Fig. 6 and Fig. 7, respectively.Calculated fluids in all samples are dominantly H2O^

CO2 mixtures with510mol % S and Cl. XH2O decreaseswith increasing experimental pressure from 25 to400MPa, consistent with greater solubility of H2O in themelt relative to CO2. XH2O in the fluid phase fromboth the St8.1 and MAS.1 experiments increases from

�60mol % at 300^400MPa to 80mol % at the lowestpressures. There is no difference (within error) in H2O/CO2 ratios in the S-rich and S-poor compositions.We could not identify the presence of any phase contain-

ing S as a major element (e.g. sulphate mineral in themelt) in the experimental products using SEM analyses.The low Cl concentrations in the fluid (see below) suggestthat the basaltic melts coexisted with a single fluid phasein all our systems. These observations imply that themass-balance approach was not compromised by partition-ing of volatiles into any additional phase.

SulphurThe sulphur content of the fluid is systematically higher forthe S-rich starting compositions, consistent with observeddifferences in the S content of the glasses (Figs 2d and3d). For St8.1, the sulphur concentration in the fluid phase(XS) in both starting materials decreases strongly from400MPa to 250MPa, remains approximately constantuntil 100MPa, then increases sharply from 100MPa to25MPa (Fig. 6b and d). A similar minimum in XS around

0102030405060708090

100

0 100 200 300 400 500

)%lo

m( diulf

P (MPa)

012345

9876

0102030405060708090

100

0 100 200 300 400 500

)%lo

m( diulf

P (MPa)

0 100 200 300 400 500P (MPa)

0123456789

0 100 200 300 400 500

)%lo

m( diulf

P (MPa)

)%lo

m( diulf

(a)

(c)(d)

(b)

XCl

XS

XCO2

XH2O

XCl

XS

XCO2

XH2O

St8.1.A St8.1.B

Xi

Xi

Xi

Xi

Fig. 6. Composition of the fluid phase (mol %), calculated from mass balance, as a function of pressure in experiments on Stromboli basalt:St8.1.A [(a) H2O and CO2; (b) S and Cl]; St8.1.B [(c) H2O and CO2; (d) S and Cl]. Pressures have been slightly offset for clarity. Error barshave been omitted; they are probably smaller than those given inTable 3 because of the systematic offset to all data points arising from uncer-tainty in the initial volatile content of the starting materials.



100MPa is observed for MAS.1 (Fig. 7b and d), althoughthe data are more scattered. The behaviour of sulphur inthe fluid contrasts with that of sulphur in the melt, whichremains constant from 400MPa to 100MPa for St8.1(Fig. 2d) and which exhibits a bell-shaped curve forMAS.1 (Fig. 3d).Our data allow us to calculate Dfl=melt

S (Fig. 8).Variationsin D

fl=meltS may result from changes in P, melt or fluid com-

position. Fluid composition varies with pressure in all setsof experiments, evolving from CO2-rich to H2O-rich withdecreasing pressure. In contrast, the melt composition isdifferent between the MAS and St8.1 experiments, but isunaffected by pressure. Figure 8 shows that Dfl=melt

S is de-pendent on pressure, with a pronounced minimum at�150 Ma, for both St8.1 and MAS.1 compositions, irre-spective of whether the starting material is S-poor orS-rich (Fig. 8). This is indicative of strong fluid com-positional and pressure controls on D

fl=meltS . Conversely,

Dfl=meltS is higher for MAS.1 than St8.1 at any given pressure

(Fig. 8), consistent with the well-known compositional

dependence of S activity coefficients in basaltic melts(Scaillet & Pichavant, 2005).

ChlorineCalculated chlorine concentrations in fluids from the St8.1and MAS.1 experiments are compromised by the smallamount of initial added chlorine and the fact that most ofthe initial chlorine (99%) remains dissolved in the melt.Consequently, the calculated chlorine concentrations inthe fluid phase are very low with large errors (Figs 6 and7). What is clear, however, is that chlorine does not showany sudden change in fluid^melt partitioning at low pres-sure and is therefore strongly fractionated from sulphurduring degassing at pressures below �100MPa.

Comparison with calculated solubility anddegassing trendsOur experiments reproduce the degassing of an ascendingbasaltic magma under conditions of a closed system (fixedinitial composition) and are therefore amenable to

0102030405060708090

100

0 100 200 300 400

)%lo

m( diulf iX

P (MPa)

012345678

0 100 200 300 400

)%lo

m( diulf iX

P (MPa)

0102030405060708090

100

0 100 200 300 400

)%lo

m( diulf iX

P (MPa)

012345678

0 100 200 300 400

)%lo

m( diulf iX

P (MPa)

(a) (b)

(c) (d)

XCl

XS

XCO2

XH2O

XCl

XS

XCO2

XH2O

MAS.1.A MAS.1.B

Fig. 7. Composition of the fluid phase (mol %), calculated from mass balance, as a function of pressure in experiments on Masaya basalt:MAS1.A [(a) H2O and CO2; (b) S and Cl]; MAS1.B [(c) H2O and CO2; (d) S and Cl]. Pressures have been slightly offset for clarity. Errorbars have been omitted; they are probably smaller than those given inTable 3 because of the systematic offset to all data points arising from un-certainty in the initial volatile content of the starting materials. The cause of the discrepant results for two experiments on MAS.1.A at200MPa is not known. Comparison with data at higher and lower pressures suggests that the true 200MPa value probably lies between thesetwo extremes.



comparison with solubility models, the conventionalmeans of calculating (closed- or open-system) degassingpaths. Here we make the comparison with the widely usedVolatileCalc programme of Newman & Lowenstern(2002) and the model of Papale et al. (2006). Althoughthese models consider CO2 and H2O as the only volatilespecies, our experimental measurements show that thepresence of sulphur and chlorine in the system does not sig-nificantly influence H2O and CO2 partitioning betweenmelts and fluids at the investigated conditions.Calculations were performed at 11508C for basalt with

49 wt % SiO2 (the only compositional variable inVolatileCalc) having the initial H2O and CO2 contents ofSt8.1 and MAS.1. In calculations using the Papale et al.(2006) model we input the actual composition of thequenched glass, as this model takes greater account of com-positional sensitivity. We considered a closed system withno initially exsolved vapour. At these conditions, thesystem is vapour saturated at pressures below 800MPa (ascalculated fromVolatileCalc).In Fig. 9a we compare our measured H2O and CO2 con-

tents of glasses from the St8.1 (A and B) set of experiments

0.1

1

10

100

1000

0 100 200 300 400 500

P (MPa)

DS

tlem/lf

MAS.1.AMAS.1.B

(b)

0.1

1

10

100

1000

0 100 200 300 400 500

P (MPa)

St8.1.ASt8.1.B

DS

tlem/lf

(a)

Fig. 8. Sulphur partition coefficients between fluid and melt (Dfl=meltS ) calculated for Stromboli (a) and Masaya (b), with fully propagated

uncertainties from both melt and fluid compositions inTable 3. The minima in both panels at �150MPa should be noted.



with those calculated using VolatileCalc and Papale(Papale et al., 2006). The CO2 concentrations are verywell described by VolatileCalc over the entire pressurerange. Conversely, Papale’s model reproduces very wellthe dissolved CO2 contents at pressures below 200MPa,but deviates at higher pressures: dissolved CO2 is overesti-mated by450% relative at 400MPa. Both models predictsimilar dissolved H2O contents: at the highest (400MPa)and lowest (�100MPa) ends of the pressure range themeasured and calculated H2O values agree well. At inter-mediate pressures VolatileCalc overestimates dissolvedH2O by 0·5^0·7wt %. As a consequence, the modelledvapour compositions are significantly depleted in H2Ocompared with those that we have calculated from

experimental data. This is apparent from Fig. 10a, wherewe plot our experimental data on a conventional CO2^H2O plot, contoured for pressure and vapour compositionusing VolatileCalc at 11508C. Experimental data at differ-ent pressures are shown with different symbols and havebeen colour-coded for calculated vapour composition forease of comparison. At the lowest pressures, calculatedsolubility and vapour composition are well matched to theexperiments. There is progressively greater deviation aspressure increases, with modelled volatile compositionsconsistently more CO2-rich than we have calculated fromour experiments. This in part reflects the greater uncer-tainty on the calculated vapour compositions at higherpressures, where there is less exsolved vapour present.

0.000.501.001.502.002.503.003.504.00

(a)

0 100 200 300 400 500 0200400600800100012001400160018002000

P (MPa)

)%t

w(O

2H )

mpp(

2O

C

St8.1.A - Papale St8.1.A - VolatileCalc

- A .1.8tS exp. St8.1.B - exp.

H2O

St8.1.A - Papale St8.1.A - VolatileCalc

.pxe - A .1.8tSSt8.1.B - exp.

CO2

Fig. 9. H2O and CO2 dissolved in experimental charges (symbols) and calculated (lines) H2O and CO2 contents in basaltic melts usingVolatileCalc (Newman & Lowenstern, 2002) and the model of Papale et al. (2006) at the pressure and temperature of the experiment using thestarting compositions inTable 1. (a) Stromboli; (b) Masaya.



However, the degassing trend itself (which is based on pre-cise measured dissolved volatile contents) is also at oddswith the modelled trend, in having less curvature(Fig. 10a). The curvature is a function of the relative parti-tion coefficients of H2O and CO2 between melt andvapour; in an extreme case where both H2O and CO2

had the same partition coefficient, the degassing trendwould be a straight line. The less curved experimentaltrend is consistent with a smaller experimental melt^vapour partition coefficient for H2O than is calculatedfromVolatileCalc, as previously deduced from Fig. 9a.A mismatch is also observed between VolatileCalc and

the experimental data of Pichavant et al. (2009) for anotherStromboli bulk composition (PST-9). These data, which in-clude both measured dissolved H2O and CO2 and calcu-lated (mass-balance) vapour compositions, are plottedalongside our data in Fig. 10a. Again, we see a displace-ment to more H2O-rich vapours than would be calculatedfrom VolatileCalc. Moreover, for the Pichavant et al.(2009) data there is a large pressure discrepancy, mostmarked at 400MPa. This can be attributed to the morecalcic, less aluminous composition of PST-9, comparedwith St8.1, which may enhance CO2 solubility, as calcu-lated by Papale et al. (2006).

Finally, we have calculated the vapour compositions inequilibrium with glasses having our measured H2O andCO2 at the experimental conditions using bothVolatileCalc and the Papale et al. (2006) model (Fig. 11aand c, respectively). For both models we see that calculatedXH2O underestimates our experimental values by up to40mol % (VolatileCalc) or 20mol % (Papale et al., 2006)at the highest pressures, but is in good agreement at thelowest pressures.In the case of the Masaya compositions (MAS.1.A and

MAS.1.B) the agreement between experiments and calcu-lations is better (Figs 9b and 10b) and the vapour compos-itions are much closer to the experimental values for bothsolubility models (Fig. 11b and d). As both the calculated(fromVolatileCalc) isopleths and the isobars closely repro-duce our MAS.1 experiments over the entire pressurerange, the calculated degassing trend has a curvature thatclosely matches the experiments (Fig. 10b).We conclude, in agreement with Papale et al. (2006), and

references therein, that there is significant compositionalcontrol on both volatile solubility and melt^vapour parti-tioning that is not fully captured by VolatileCalc. In themost extreme case (here exemplified by St8.1),VolatileCalc predicts vapour that is too CO2-rich, leading

0200400600800

100012001400160018002000

0 100 200 300 400 5000.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50(b)

MAS.1.A - exp.MAS.1.B - exp.MAS.1.A - PapaleMAS.1.A - VolatileCalc

P (MPa)

H2O

(w

t%) C

O2 (ppm

)

H2O CO

2

MAS.1.A - exp.MAS.1.B - exp. MAS.1.A - PapaleMAS.1.A - VolatileCalc

Fig. 9. Continued.



to much more sharply curved degassing trends than areobserved experimentally. This discrepancy can lead toproblems when attempting to interpret melt inclusionvolatile contents in terms of open- or closed-systemdegassing.

Comparison with melt inclusionsMelt inclusion data for volatile elements (H2O, CO2, Cl,S) are available for samples from Stromboli in the studiesby Bertagnini et al. (2003) and Me¤ trich et al. (2001, 2010).These data can be usefully compared with our

XH2Ov=0.2

XH2Ov=0.6

XH2Ov=0.5

XH2Ov=0.4

XH2Ov=0.7

XH2Ov=0.8

St8.1.A St8.1.B

0

500

1000

1500

2000

2500

3000(a)

0.00 1.00 2.00 3.00 4.00 5.00 6.00

CO

2 (pp

m)

H2O (wt%)

400 MPa

50 MPa

100 MPa

200 MPa

300 MPa

25 MPa

Melt inclusions from: Bertagnini et al., 2003Metrich et al., 2001

Pichavant et al., 2009

400 MPa300 MPa250 MPa200 MPa150 MPa100 MPa50 MPa25 MPa

XH2Ov

: 0.9 - 0.80.8 - 0.70.7 - 0.6< 0.6

Fig. 10. Experimental glass CO2 and H2O contents for (a) Stromboli and (b) Masaya, compared with calculated isobars (fine continuouslines), vapour isopleths (dashed lines) and degassing trends (bold continuous line) calculated using VolatileCalc at 11508C and for the experi-mental starting compositions inTable 1. Experimental data are colour coded to denote the range in calculated vapour composition. In (a) wealso show the experimental data of Pichavant et al. (2009) for a ‘golden pumice’ basalt (PST-9) from Stromboli at pressures of 401, 208 and88MPa.The offset of these data from our own at similar pressures can be ascribed to the compositional difference between their starting mater-ial and ours and its influence on CO2 solubility. The shaded region encompasses data from Stromboli melt inclusions measured by Me¤ trichet al. (2001) and Bertagnini et al. (2003).



experimental isobars, isopleths and degassing trends(Fig. 10a). In terms of our experimental CO2^H2O system-atics, melt inclusions are located between the 250 and400MPa isobars at vapour compositions of XH2O be-tween 0·6 and 0·7. Using the more sparse experimentaldata of Pichavant et al. (2009) would yield a slightlylower pressure range (�180^300MPa) and slightly morecarbonic vapour. In contrast, VolatileCalc gives a pres-sure range of 200^400MPa, but XH2O of between 0·2

and 0·4. The trend described by the melt inclusionsshows too little curvature compared with either ourexperimental degassing trends or those calculated byVolatileCalc.Deviation of melt inclusion data arrays from calculated

degassing paths is not uncommon (e.g. Blundy et al., 2010).For example, at Mount Etna, Spilliaert et al. (2006) andMe¤ trich & Wallace (2008) identified a group ofH2O-depleted melt inclusions that they interpreted as

0

500

1000

1500

2000

2500(b)

0 0.5 1 1.5 2 2.5 3 3.5 4

CO

2 (pp

m)

H2O (wt%)

400 MPa

25 MPa50 MPa

100 MPa

200 MPa

300 MPa

XH2Ov=0.2

XH2Ov=0.8

XH2Ov=0.7

XH2Ov=0.6

XH2Ov=0.5

XH2Ov=0.4

MAS.1.A MAS.1.B

400 MPa300 MPa250 MPa200 MPa150 MPa100 MPa50 MPa25 MPa

XH2Ov

: 0.9 - 0.80.8 - 0.70.7 - 0.6< 0.6

: fluid phase not determined.

Fig. 10. Continued.



re-equilibration of the magma with a deeper CO2-rich gasphase during ponding and crystallization at �200MPa.Also, melt inclusions from Jorullo and Colima volcanoesin central Mexico have been reported to deviate from thecalculated degassing trend, for open or closed systems,and have been interpreted as influenced by a fluxingCO2-rich gas (Johnson et al., 2009; Blundy et al., 2010).Me¤ trich & Wallace (2008) proposed that gas-fluxing bydeeply derived magmatic CO2 may be a common processat basaltic volcanoes. Such an interpretation is consistentwith our experimental data. An issue is whether thevapour composition responsible for fluxing at Strombolihas XH2O in the range 0·2^0·4 (VolatileCalc) or 0·6^0·7(our experimental data).It is also instructive to compare our St8.1 experimental

data for S and Cl with the melt inclusions from Stromboli(Me¤ trich et al., 2001, 2010; Bertagnini et al., 2003).

In Fig. 12a and b, respectively, we plot S and Cl in melt in-clusions against the calculated H2O^CO2 saturation pres-sure for the same melt inclusion using VolatileCalc [itwould make relatively little difference if we calculatedpressure from our experimental data or used Papale et al.(2006)]. Melt inclusions show a good match to thelow-sulphur series of experiments (St8.1.A), showing littlechange in dissolved S and Cl from 400 to �200MPa, fol-lowed by a sharp decrease in S, but not Cl, atP5150MPa. The matrix glass analyses of Me¤ trich et al.(2001) plot at the low-pressure extremity of this trend.These glasses have negligible H2O and CO2 (indicative ofdegassing to low pressure), very low S (500 ppm) but Clcontents (1200 ppm) only slightly below those of the meltinclusions. This strong fractionation of S from Cl at lowpressures is entirely consistent with our experimental data(Figs 2f and 13c).

0

20

40

60

80

100

0 20 40 60 80 100

Papa

leX

H2O

Exp. XH2O

MAS.1.BMAS.1.A

0

20

40

60

80

100

0 20 40 60 80 100

Papa

leX

H2O

Exp. XH2O

St8.1.A St8.1.B

1:1

1:11:1

1:1

0

10

20

30

40

50

60

70

80

90

100

0 20 40 60 80 100

Vol

atile

Cal

c X

H2O

Exp. XH2O

MAS.1.A MAS.1.B

0

10

20

30

40

50

60

70

80

90

100

0 20 40 60 80 100

Vol

atile

Cal

c X

H2O

Exp. XH2O

St8.1.A

St8.1.B

(a)

(d)(b)

(c)

Fig. 11. Experimentally determined fluid phase compositions (in mol % H2O) for St8.1 and MAS.1 compositions compared with those calcu-lated using (a, b) VolatileCalc (Newman & Lowenstern, 2002) and (c, d) the model of Papale et al. (2006) using the measured experimentalglass CO2 and H2O contents, pressures and temperatures. A 1:1 line is shown for reference. Errors on experimental fluid compositions arefully propagated from uncertainties on starting compositions and glass compositions. Because of some cross-correlation of these uncertainties,the error bars are maxima.



Further support for low-pressure S^Cl fractionation isprovided by basaltic melt inclusion data from Etna(Spillaert et al., 2006). Although Etna basalts are compos-itionally different from those of Stromboli, the melt inclu-sions are a reasonable match to the high-sulphurexperiments (St8.1.B). Moreover, the Etna melt inclusionscover a very wide range of calculated H2O^CO2 satur-ation pressures. Once again we see near-constant S and Clfrom 400 to 150MPa, followed by a rapid decline in Swith little change in Cl (Fig. 12). In fact, the melt inclusionCl contents increase slightly at pressures below 50MPa,probably because of further enrichment in the residualmelt owing to microlite crystallization. Melt inclusiondata from Stromboli and Etna strongly suggest that frac-tionation of S from Cl is diagnostic of degassing at pres-sures below 100MPa.There are rather fewer melt inclusion data for Masaya

with which to make comparisons. Sadofsky et al. (2008)

measured S, Cl and H2O (but not CO2) dissolved in meltinclusions from Masaya. H2O is low, between 1·4 and1·7wt %. On the basis of the water content only, we calcu-lated saturation pressures withVolatileCalc, for a basalt of49 wt % SiO2, at 11508C, of between 20 and 29MPa.However, Atlas & Dixon (2006) measured 6000 ppm ofCO2 in melt inclusions from Masaya, suggesting thatthese pressures are serious underestimates. In the absenceof a Masaya melt inclusion and groundmass dataset withH2O, CO2, S and Cl measurements, it does not seem in-structive to make comparisons with our experimentaldata. It is, however, noteworthy that of the six melt inclu-sion analyses from Masaya given by Sadofsky et al. (2008)the overall variation is S is much greater (1482 to241ppm) than that in Cl (599 to 264 ppm), with thelowest S contents corresponding to the lowest H2O (mostdegassed) melt inclusions. Again, this supports our conclu-sion that S/Cl fractionation is a low-pressure phenomenon.

St8.1.A

St8.1.B

Métrich et al., 2001 - Groundmass

Bertagnini et al., 2003 - MI Spilliaert et al., 2006 - MI

Métrich et al., 2010 - Large scale eruption - MI

Métrich et al., 2010 - Small scale eruption - MI

Experimental data: Stromboli data: Etna data:

0

500

1000

1500

2000

2500

3000

0 100 200 300 400 500

Cl m

elt (

ppm

)

Ptot (MPa)

(d)

0

500

1000

1500

2000

2500

3000

0 100 200 300 400 500

Cl m

elt (

ppm

)

Ptot (MPa)

(c)

0

500

1000

1500

2000

2500

3000

3500

4000

0 100 200 300 400 500

S mel

t (pp

m)

Ptot (MPa)

(b)

0

500

1000

1500

2000

2500

3000

4000

0 100 200 300 400 500

S mel

t (pp

m)

Ptot (MPa)

3500(a)

Fig. 12. Sulphur (a, b) and chlorine (c, d) dissolved in experimental glasses from St8.1 compared with analyses of melt inclusions and matrixglasses from Stromboli (a, c) (Me¤ trich et al., 2001, 2010; Bertagnini et al., 2003) and Etna (b, d) (Spilliaert et al., 2006). The consistent behaviourof S and Cl in experimental and natural glasses over a wide range in pressure should be noted.



Comparison with gas chemistry measuredat the ventGases released from the volcanic vents at Stromboli andMasaya are the complement to the melt inclusion data. Itis therefore instructive to compare gas chemistry with ourcalculated volatile compositions. Magmatic gases offerricher potential to interpret and anticipate subvolcanic

processes as they can be measured in real time, ratherthan a posteriori as is the case with melt inclusions. AtStromboli, petrological studies show that a basalticmagma in equilibrium with a fluid phase in a deep-seatedreservoir at 400MPa (Me¤ trich et al., 2001; Bertagniniet al., 2003; Pichavant et al., 2009) starts to degas duringascent. Under these conditions, according to our experi-mental results, fluids evolve from dominantly CO2-rich at400MPa, to progressively more H2O-rich until 150MPa,and then become dramatically H2O-enriched at lowerpressures. These results are consistent with experimentalresults obtained for golden pumices from Stromboli(Landi et al., 2004) equilibrated with an H2O^CO2 fluidphase (Pichavant et al., 2009).Burton et al. (2007a) and Aiuppa et al. (2010) have mea-

sured the compositions of emitted gases at Stromboliduring ‘quiescent periods’,‘typical explosions’and ‘small ex-plosions’. By making measurements within the crater theywere able to minimize atmospheric contamination, suchthat the measured H2O/CO2 ratios are thought to reflectthose of the magmatic gas. The highest observed values ofmolar H2O/CO2 ratios (�6·1) were measured during qui-escent degassing by Burton et al. (2007a). Aiuppa et al.(2010) obtained values up to 50 for the same period.Burton’s value is reproduced experimentally for theSt8.1 basalt in equilibrium with a fluid phase at a pres-sure of less than 100MPa. Lower ratios have been mea-sured during smaller explosions (4·5) and typicalexplosions (2·3). According to our experimental results,these ratios indicate pressures of 200^300MPa (Fig. 13a).Also, H2O/CO2 ratios reported by Aiuppa et al. (2010)have been measured to be the smallest during strombolianexplosions.Burton et al. (2007a) measured high molar CO2/S

(20·7�2·1) ratios during typical explosions.When activityintensity decreases, these ratios decrease, reaching 7·8during quiescent periods. According to our experimentaldata, elevated CO2/S ratios are characteristic of intermedi-ate pressures, between 100 and 300MPa (Fig. 13b), whenS is at its lowest level in the vapour phase (Fig. 6). TheCO2/S ratio associated with explosions (�20) is matchedexperimentally in the range 150^300MPa, whereas thelower values typical of quiescence would require lower orhigher pressures than this. There would, then, appear tobe a congruency between the gas data and the experimen-tal data, in that explosions involve degassing at pressuresof �200MPa, whereas quiescent degassing occurs atlower pressures �100MPa. Aiuppa et al. (2009, 2010) pro-posed a model of degassing at Stromboli in which they sug-gested that the compositional characteristics of the gasemissions during quiescent and syn-explosive activity atStromboli result from the mixing of gases sourced by (1)the degassing of dissolved volatiles in the shallow partand (2) CO2-rich gas bubbles coming from depth

St8.1.BSt8.1.A

(a)

0

2

4

6

8

10

12

14

16

0 100 200 300 400 500

H2

OC/

O2

P (MPa)

0.1

1.0

10.0

100.0

1000.0

0 100 200 300 400 500

lC/S

P (MPa)

(c)

P (MPa)

Quiescent activity

Typical explosionsTypical exp

0

5

10

15

20

25

30

0 100 200 300 400 500

CO

2/S

(b)

Fig. 13. Molar ratios in experimental fluids from Stromboli (St8.1.Aand St8.1.B), as functions of pressure, showing the sensitivity of someratios to the pressure of last equilibrium between fluid and melt.(a) H2O/CO2; (b) CO2/S; (c) S/Cl. [Note the logarithmic scale in(c).] In (b) the range is indicated in CO2/S for gases measuredduring typical explosions and quiescent activity by Burton et al.(2007a).



(P4100MPa). Such a model is consistent with our experi-mental data.It is striking that the trends in gas chemistry observed by

Burton et al. (2007a) and defined by our experimentaldata show a similar pressure dependence. Moretti &

Papale (2004) calculated fluid phase compositions in equi-librium with a shoshonitic melt and showed that CO2 con-tents in the fluid phase are weakly dependent on the redoxconditions, with sulphur exsolution occurring later inmore oxidized systems. Redox conditions in our experi-mental charges are �NNOþ1·5, and we observed thatCO2/S ratios start to decrease below 200MPa. These re-sults are in good agreement with Moretti & Papale’s(2004) calculations. CO2 partitions strongly into thevapour phase at all pressures, whereas sulphur largely re-mains in the melt until relatively low pressures. CO2/S istherefore a powerful indicator of magma ascent and degas-sing below �200MPa (Fig. 13b).The pressure dependence of sulphur degassing makes

S/Cl ratios a sensitive indicator of degassing pressures.The sharp decrease in S/Cl of low-pressure melt inclusionsand groundmass glasses has already been shown (Fig. 12).The corollary is elevated S/Cl ratios in vapours derivedfrom shallow levels. For this reason S/Cl ratios are oftenused to monitor volcanoes. S and Cl are much less abun-dant in the atmosphere than H2O and CO2, and thereforemeasurements give much higher signal to noise ratiosthan for volcanogenic H2O and CO2. Burton et al. (2007a)measured S/Cl molar ratios in the gas phase ranging from1·0^1·5 during quiescent periods to 4·5�0·8 during typicalexplosions. This behaviour appears to disagree with ourprevious interpretations, in that the higher S/Cl ratiosduring explosions would imply lower degassing pressuresthan the lower S/Cl ratios during quiescent periods(Fig. 13c). However, our S/Cl ratios should be viewed withcaution, as Cl in the fluid phase is very small, leading tolarge uncertainties in fluid S/Cl ratios. Additional factorsthat may influence the behaviour of Cl include enrichmentowing to low-pressure crystallization (as shown above forEtna), or separation of a brine phase upon intersection ofthe low-pressure vapour solvus. At Stromboli, Burton et al.(2007a) described a second nucleation event involvingCl-rich bubbles. Alternatively, this may reflect ascentfrom depth of fresh, undegassed magma with a higherS content.For Masaya, the recent work of Martin et al. (2010) pro-

vides data on all major volatile components, as measuredby open-path FTIR across the active vent between 1998and 2009. The molar SO2/Cl ranges from 1·6 to 4·6. Thesevalues are consistent with our experiments on bothMAS.1.A and MAS.1.B (Fig. 14), although they are notdiagnostic of any particular pressure. The observed H2O/CO2 and CO2/S value are in the ranges 10^41 and 1·5^3·5,respectively. These are substantially different from our ex-perimental ratios even at 25MPa. Although the observedvalues could be related to degassing at very low pressures(525MPa), a more likely explanation is that the initialmagma at Masaya is much poorer in CO2 than our experi-mental starting materials (�7000 ppm), which were

0

1

2

3

4

5

6

0 100 200 300 400P (MPa)

MAS.1.A MAS.1.B

(a)H

2O/C

O2

0

10

20

30

40

50

60

0 100 200 300 400P (MPa)

(b)

CO

2/S

0.01

0.1

1

10

0 100 200 300 400

lC/S

P (MPa)

100(c)

Fig. 14. Molar ratios in experimental fluids from Masaya (MAS.1.Aand MAS.1.B), as functions of pressure, showing the sensitivity ofsome ratios to the pressure of last equilibrium between fluid andmelt. (a) H2O/CO2; (b) CO2/S; (c) S/Cl. [Note the logarithmic scalein (c).]



based on the melt inclusion data reported by Atlas& Dixon (2006). Involvement of meteoric water may alsoplay a role, although that would have little effect on CO2/SO2 ratios, which also vary by a factor of two at Masaya.It is interesting to note that the observed H2O/CO2 andCO2/S ratios for the five campaigns (1998^2009) reportedby Martin et al. (2010) are inversely correlated [see alsothe data of Aiuppa et al. (2009) for Stromboli]. This is en-tirely consistent with our experimental data at pressuresbelow 100MPa (Fig. 14).This relationship would character-ize fluctuation in degassing pressures at low pressure evenif the initial CO2 contents were lower than used in ourexperiments.

CONCLUSIONSExperiments have been performed to simulateclosed-system equilibrium degassing of two different bas-altic magmas containing the volatile H2O, CO2, S and Clabundances under oxidized, super-liquidus conditionsover the pressure range 400^25MPa.Volatiles dissolved inthe melt were measured by different techniques and fluidphase composition was calculated through mass-balancecalculations. This is the first time that equilibrium experi-ments have been performed with such a complex fluidphase. Our principal findings are as follows.

(1) Exsolved fluids are predominantly H2O^CO2 mix-tures, with the H2O/CO2 ratio increasing withdecreasing pressure.

(2) Adding different initial amounts of S to the startingcomposition does not affect the behaviour of H2Oand CO2 (for the amounts of volatiles considered inthis study).

(3) S starts to degas significantly at 150MPa in basalticsystems, under oxidized conditions (�NNOþ1·5),whereas Cl remains in the melt. This leads to strongfractionation of S and Cl at low pressures.This behav-iour is consistent with observations on melt inclusionsand matrix glasses from Etna and Stromboli.

(4) Sulphur partitioning between fluid and melt is sensi-tive to pressure, fluid composition and melt compos-ition. The variation in D

fl=meltS with pressure is

strongly non-linear, with a pronounced minimum inthe vicinity of 150MPa.

(5) Experimental results are broadly consistent withavailable melt solubility models, for S- and Cl-free sys-tems. However, the models fail to capture some detailsof the experiments, notably the composition of thefluid.

(6) Changing the pressure at which vapour segregatesfrom its parent magma has a profound influence ongas chemistry. Low-pressure (�100MPa) gas loss ischaracterized by elevated H2O/CO2 and S/Cl ratios,whereas elevated CO2/S ratios seem to be diagnostic

of intermediate pressure (100^300MPa). The Cl con-tent of vapour may be complicated if low-pressurephase separation occurs.

(7) H2O/CO2 and CO2/S ratios measured at Stromboliare in reasonable agreement with our St8.1 experi-ments, suggesting that the initial volatile budget ofour starting materials provides a good match to thatof relatively undegassed magma at depth. At Masayavent gases have much higher H2O/CO2 and lowerCO2/S ratios than in our experiments, suggestingthat relatively undegassed magma at depth beneathMasaya has considerably less than the �7000 ppm inour MAS.1 starting material.

ACKNOWLEDGEMENTSWe acknowledge O. Diedrich for the preparation of sam-ples for analysis, T. Shishkina for H2O^CO2-bearing bas-altic glass samples used to calibrate SIMS, and U. Bauerand A. Wegorzewski at Hannover who helped with KFT,carbon analysis and Fe determination. We acknowledgeC. J. De Hoog, R. Hinton and J. Craven for SIMS analysesat the University of Edinburgh, and S. Kearns for helpwith the Bristol electron microprobe. We acknowledgehelpful discussions with M. Burton, J. Phillips, M. Polacciand H. Mader. The paper was much improved followingthe helpful reviews of C. Martel, R. Moretti andJ.Webster.

FUNDINGThis research was supported by NERC standard grant NE/F004222/1.

SUPPLEMENTARY DATASupplementary data for this paper are available at Journalof Petrology online.

REFERENCESAiuppa, A., Federico, C., Giudice, G., Giuffrida, G., Guida, R.,

Gurrieri, S., Liuzzo, M., Moretti, R. & Papale, P. (2009). The2007 eruption of Stromboli volcano: Insights from real-time meas-urement of the volcanic gas plume CO2/SO2 ratio. Journal of

Volcanology and Geothermal Research 182, 221^230.Aiuppa, A., Bertagnini, A., Me¤ trich, N., Moretti, R., Di Muro, A.,

Liuzzo, M. & Tamburello, G. (2010). A model of degassing forStromboli volcano. Earth and Planetary Science Letters 295, 195^204.

Allard, P., Carbonnelle, J., Me¤ trich, N., Loyer, H. & Zettwoog, P.(1994). Sulfur output and magma degassing budget of Strombolivolcano. Nature 368, 326^330.

Alletti, M., Baker, D. R., Scaillet, B., Aiuppa, A., Moretti, R. &Ottolini, L. (2009). Chlorine partitioning between a basaltic meltand H2O^CO2 fluids at Mount Etna. Chemical Geology 263, 37^50.

Atlas, Z. D. & Dixon, J. E. (2006). Interconnected magmatic conduitsystems as recorded by melt inclusions from Masaya and Apoyo



Calderas, Nicaragua, Eos Trans. AGU 87(52), Fall Meet. Suppl.,Abstract V13D-06.

Behrens, H. (1995). Determination of water solubilities in high viscos-ity melts. An experimental study on NaAlSi3O8 and KAlSi3O8

melts. EuropeanJournal of Mineralogy 7, 905^920.Behrens, H., Ohlhorst, S., Holtz, F. & Champenois, M. (2004). CO2

solubility in dacitic melts equilibrated with H2O^CO2 fluids:Implications for modeling the solubility of CO2 in silicic melts.Geochimica et Cosmochimica Acta 68, 4687^4703.

Behrens, H., Misiti, V., Freda, C., Vetere, F., Botcharnikov, R. E. &Scarlato, P. (2009). Solubility of H2O and CO2 in ultrapotassicmelts at 1200 and 12508C and pressure from 50 to 500 MPa.American Mineralogist 94, 105^120.

Berndt, J., Liebske, C., Holtz, F., Freise, M., Nowak, M.,Ziegenbein, D., Hurkuck, W. & Koepke, J. (2002). A combinedrapid-quench and H2-membrane setup for internally heated pres-sure vessels: Description and application for water solubility in bas-altic melts. American Mineralogist 87, 1717^1726.

Bertagnini, A., Me¤ trich, N., Landi, P. & Rosi, M. (2003). Strombolivolcano (Aeolian Archipelago, Italy): An open window on thedeep-feeding system of a steady state basaltic volcano. Journal ofGeophysical Research 108(B7), 2336, doi:10.1029/2002jb002146.

Blundy, J. & Cashman, K. (2008). Petrologic constraints on magmasystem variables and processes. In: Putirka, K. D. & Tepley, F. J.,III (eds) Minerals, Inclusions and Volcanic Processes. Mineralogical

Society of America and Geochemical Society, Reviews in Mineralogy and

Geochemistry 69, 179^239.Blundy, J., Cashman, K.V., Rust, A. & Witham, F. (2010). A case for

CO2-rich arc magmas. Earth and Planetary Science Letters 290,289^301.

Botcharnikov, R. E., Behrens, H., Holtz, F., Koepke, J. & Sato, H.(2004). Sulfur and chlorine solubility in Mt. Unzen rhyodaciticmelt at 8508C and 200 MPa. Chemical Geology 213, 207^225.

Botcharnikov, R. E., Freise, M., Holtz, F. & Behrens, H. (2005a).Solubility of C^O^H mixtures in natural melts: new experimentaldata and application range of recent models. Annals of Geophysics48, 633^646.

Botcharnikov, R. E., Koepke, J., Holtz, F., McCammon, C. &Wilke, M. (2005b). The effect of water activity on the oxidationand structural state of Fe in a ferro-basaltic melt. Geochimica et

Cosmochimica Acta 69, 5071^5085.Botcharnikov, R. E., Behrens, H. & Holtz, F. (2006). Solubility and

speciation of C^O^H fluids in andesitic melt at T¼1100^13008Cand P¼ 200 and 500 MPa. Chemical Geology 229, 125^143.

Botcharnikov, R. E., Holtz, F. & Behrens, H. (2007).The effect of CO2

on the solubility of H2O^Cl fluids in andesitic melt. European

Journal of Mineralogy 19, 671^680.Burton, M., Oppenheimer, C., Horrocks, L. A. & Francis, P. W.

(2000). Remote sensing of CO2 and H2O emission rates fromMasaya volcano, Nicaragua. Geology 28, 915^918.

Burton, M., Allard, P., Mure¤ , F. & La Spina, A. (2007a). Depth ofslug-driven Strombolian explosive activity magmatic gas compos-ition reveals the source. Science 317, 227^230.

Burton, M. R., Mader, H. M. & Polacci, M. (2007b). The role of gaspercolation in quiescent degassing of persistently active basalticvolcanoes. Earth and Planetary Science Letters 264, 46^60.

Carroll, M. R. & Rutherford, M. J. (1985). Sulfide and sulfate satur-ation in hydrous silicate melts. Journal of Geophysical Research 90,C601^C612.

Carroll, M. R. & Rutherford, M. J. (1988). Sulfur speciation in hy-drous experimental glasses of varying oxidation state: results frommeasured wavelength shifts of sulfur X-rays. American Mineralogist

73, 845^849.

Carroll, M. R. & Webster, J. D. (1994). Solubilities of sulfur, noblegases, nitrogen, chlorine, and fluorine in magmas. In:Carroll, M. R. & Holloway, J. R. (eds) Volatiles in Magmas.

Mineralogical Society of America, Reviews in Mineralogy 30, 231^279.Clemente, B., Scaillet, B. & Pichavant, M. (2004). The solubility of

sulfur in hydrous rhyolitic melts. Journal of Petrology 45, 2171^2196.Di Carlo, I., Pichavant, M., Rotolo, S. G. & Scaillet, B. (2006).

Experimental crystallization of a high-K arc basalt: The goldenpumice, Stromboli volcano (Italy). Journal of Petrology 47, 1317^1343.

Dixon, J. E. (1997). Degassing of alkalic basalts. American Mineralogist

82, 368^378.Dixon, J. E., Stolper, E. M. & Holloway, J. R. (1995). An experimental

study of water and carbon dioxide solubilities in mid-ocean ridgebasaltic liquids. 1. Calibration and solubility models. Journal of

Petrology 36, 1607^1631.Duffell, H. J., Oppenheimer, C., Pyle, D. M., Galle, B.,

McGonigle, A. J. S. & Burton, M. R. (2003). Changes in gas com-position prior to a minor explosive eruption at Masaya volcano,Nicaragua. Journal of Volcanology and Geothermal Research 126,327^339.

Fine, G. & Stolper, E. (1986). Dissolved carbon dioxide in basalticglasses. Concentrations and speciation. Earth and Planetary Science

Letters 76, 263^278.Francalanci, L., Tommasini, S. & Conticelli, S. (2004). The volcanic

activity of Stromboli in the 1906^1998 AD period: mineralogical,geochemical and isotope data relevant to the understanding of theplumbing system. Journal of Volcanology and Geothermal Research 131,179^211.

Horrocks, L. A. (2001). Infrared spectroscopy of volcanic gases atMasaya, Nicaragua, Ph.D. thesis, The Open University, MiltonKeynes.

Johnson, E. R., Wallace, P. J., Granados, H. D., Manea, V. C.,Kent, A. J. R., Bindeman, I. N. & Donegan, C. S. (2009).Subduction-related volatile recycling and magma generation be-neath central Mexico: insights from melt inclusions, oxygen iso-topes and geodynamic models. Journal of Petrology 50, 1729^1764.

Jugo, P. J., Luth, R.W. & Richards, J. P. (2005). Experimental data onthe speciation of sulfur as a function of oxygen fugacity in basalticmelts. Geochimica et Cosmochimica Acta 69, 497^503.

Jugo, P. J., Wilke, M. & Botcharnikov, R. E. (2010). Sulfur K-edgeXANES analysis of natural and synthetic basaltic glasses:Implications for S speciation and S content as function of oxygenfugacity. Geochimica et Cosmochimica Acta 74, 5926^5938.

Kilinc, A., Carmichael, I. S. E., Rivers, M. L. & Sack, R. O. (1983).The ferric^ferrous ratio of natural silicate liquids equilibrated inair. Contributions to Mineralogy and Petrology 83, 136^140.

Kress, V. C. & Carmichael, I. S. E. (1988). Stoichiometry of the ironoxidation reaction in silicate melts. American Mineralogist 73,1267^1274.

Kress, V. C. & Carmichael, I. S. E. (1991). The compressibility of sili-cate liquids containing Fe2O3 and the effect of composition, tem-perature, oxygen fugacity and pressure on their redox states.Contributions to Mineralogy and Petrology 108, 82^92.

Landi, P., Me¤ trich, N., Bertagnini, A. & Rosi, M. (2004). Dynamics ofmagma mixing and degassing recorded in plagioclase atStromboli (Aeolian Archipelago, Italy). Contributions to Mineralogy

and Petrology 147, 213^227.Lesne,P., Scaillet, B., Pichavant,M., Iacono-Marziano,G.&Beny, J.M.

(2011a).The H2O solubility of alkali basaltic melts: an experimentalstudy. Contributions toMineralogyandPetrology162,133^151.

Lesne, P., Scaillet, B., Pichavant, M. & Beny, J. M. (2011b).The carbondioxide solubility in alkali basalts: an experimental study.Contributions to Mineralogy and Petrology 162, 153^168.



Luhr, J. F. (1990). Experimental phase relations of water- andS-saturated arc magmas and the 1982 eruption of El Chichon vol-cano. Journal of Petrology 31, 1071^1114.

Martin, R. S., Sawyer, G. L., Spampinato, L., Salerno, G. G.,Ramirez, C., Ilyinskaya, E.,Witts, M. L. I., Mather, T.,Watson, I.M., Phillips, J. C. & Oppenheimer, C. (2010). A total volatile inven-tory for Masaya volcano, Nicaragua. Journal of Geophysical Research115, B09215, doi:10.1029/2010JB007480.

Mavrogenes, J. A. & O’Neill, H. S. C. (1999). The relative effects ofpressure, temperature and oxygen fugacity on the solubility of sul-fide in mafic magmas. Geochimica et Cosmochimica Acta 63, 1173^1180.

Me¤ trich, N. & Clocchiatti, R. (1996). Sulfur abundance and its speci-ation in oxidized alkaline melts. Geochimica et Cosmochimica Acta 60,4151^4160.

Me¤ trich, N. & Wallace, P. (2008). Volatile abundances in basalticmagmas and their degassing paths tracked by melt inclusions. In:Putirka, K. D. & Tepley, F. J., III (eds) Minerals, Inclusions and

Volcanic Processes. Mineralogical Society of America and Geochemical

Society, Reviews in Mineralogy and Geochemistry 69, 363^402.Me¤ trich, N., Bertagnini, A., Landi, P. & Rosi, M. (2001).

Crystallization driven by decompression and water loss atStromboli volcano (Aeolian Islands, Italy). Journal of Petrology 42,1471^1790.

Me¤ trich, N., Bertagnini, A. & Di Muro, A. (2010). Conditions ofmagma storage, degassing and ascent at Stromboli: new insightsinto the volcano plumbing system with inferences on the eruptivedynamics. Journal of Petrology 51, 603^626.

Moretti, R. & Papale, P. (2004). On the oxidation state and volatilebehavior in multicomponent gas^melt equilibria. Chemical Geology213, 265^280.

Moune, S., Holtz, F. & Botcharnikov, R. E. (2009). Sulphur solubilityin andesitic to basaltic melts: implications for Hekla volcano.Contributions to Mineralogy and Petrology 157, 691^707.

Newman, S. & Lowenstern, J. B. (2002).VOLATILECALC: a silicatemelt^H2O^CO2 solution model written in Visual Basic for excel.Computers and Geosciences 28, 597^604.

Ohlhorst, S., Behrens, H. & Holtz, F. (2001). Compositional depend-ence of molar absorptivities of near-infrared OH and H2O bandsin rhyolitic to basaltic glasses. Chemical Geology 174, 5^20.

Papale, P. (1999). Modeling of the solubility of a two-componentH2OþCO2 fluid in silicate liquids. American Mineralogist 84,477^492.

Papale, P., Moretti, R. & Barbato, D. (2006). The compositional de-pendence of the saturation surface of H2OþCO2 fluids in silicatemelts. Chemical Geology 229, 78^95.

Pichavant, M., Di Carlo, I., Le Gac, Y., Rotolo, S. G. & Scaillet, B.(2009). Experimental constraints on the deep magma feedingsystem at Stromboli Volcano, Italy. Journal of Petrology 50, 601^624.

Richet, P., Whittington, A., Holtz, F., Behrens, H., Ohlhorst, S. &Wilke, M. (2000). Water and the density of silicate glasses.Contributions to Mineralogy and Petrology 138, 337^347.

Rosi, M., Bertagnini, A. & Landi, P. (2000). Onset of the persistentactivity at Stromboli volcano (Italy). Bulletin of Volcanology 62,294^300.

Sack, R. O., Carmichael, I. S. E., Rivers, M. & Ghiorso, M. S. (1980).Ferric^ferrous equilibria in natural silicate liquids at 1 bar.Contributions to Mineralogy and Petrology 75, 369^376.

Sadofsky, S. J., Portnyagin, M., Hoernle, K. & Van Den Bogaard, P.(2008). Subduction cycling of volatiles and trace elements throughthe Central American volcanic arc: evidence from melt inclusions.Contributions to Mineralogy and Petrology 155, 433^456.

Scaillet, B. & Pichavant, M. (2005). A model of sulphur solubility forhydrous mafic melts: application to the determination of magmaticfluid compositions of Italian volcanoes. Annals of Geophysics 48,671^698.

Schuessler, J. A., Botcharnikov, R. E., Behrens, H., Misiti, V. &Freda, C. (2008). Amorphous materials: properties, structure, anddurability: oxidation state of iron in hydrous phono-tephriticmelts. American Mineralogist 93, 1493^1504.

Shishkina, T., Botcharnikov, R., Holtz, F., Almeev, R. &Portnyagin, M. (2010). M. Solubility of H2O and CO2-bearingfluids in tholeiitic basalts at pressures up to 500 MPa. ChemicalGeology 277, 115^125.

Signorelli, S. & Carroll, M. R. (2000). Solubility and fluid^melt parti-tioning of Cl in hydrous phonolitic melts. Geochimica et

Cosmochimica Acta 64, 2851^2862.Spilliaert, N., Allard, P., Me¤ trich, N. & Sobolev, A. (2006). Melt inclu-

sion record of the conditions of ascent, degassing and extrusion ofvolatile-rich alkali basalt during the powerful 2002 flank eruptionof Mount Etna (Italy). Journal of Geophysical Research 111, B04203,doi:10.1029/2005/JB003934.

Stelling, J., Botcharnikov, R. E., Beermann, O. & Nowak, M. (2008).Solubility of H2O- and chlorine-bearing fluids in basaltic melt ofMount Etna at T¼1050^12508C and P¼ 200 MPa. Chemical

Geology 256, 102^110.Stoiber, R. E., Williams, S. N. & Huebert, B. I. (1986). Sulfur and

halogen gases at Masaya Caldera Complex, Nicaraguaçtotal fluxand variations with time. Journal of Geophysical Research: Solid Earthand Planets 91, 2215^2231.

Symonds, R. B., Rose, W. I., Bluth, G. J. S. & Gerlach, T. M. (1994).Volcanic-gas studies: methods, results and applications. In:Carroll, M. R. & Holloway, J. R. (eds) Volatiles in Magmas.

Mineralogical Society of America, Reviews in Mineralogy 30, 1^66.Wallace, P. J. & Carmichael, I. S. E. (1994). S-speciation in submarine

basaltic glasses as determined by measurements of SK-alphaX-ray wavelength shifts. American Mineralogist 79, 161^167.

Webster, J. D., Kinzler, R. J. & Mathez, E. A. (1999). Chloride andwater solubility in basalt and andesite melts and implications formagmatic degassing. Geochimica et Cosmochimica Acta 63, 729^738.

Webster, J. D., Sintoni, M. F. & De Vivo, B. (2009). The partitioningbehavior of Cl, S, and H2O in aqueous vapor-� saline-liquid satu-rated phonolitic and trachytic melts at 200 MPa. Chemical Geology263, 19^36.

Williams, S. N. (1983). Plinian airfall deposits of basaltic composition.Geology 11, 211^214.

Wilson, A. D. (1960). The micro-determination of ferrous iron in sili-cate minerals by a volumetric and colorimetric method. Analyst 85,823^827.



Experimental Simulation of Closed-System Degassing inthe ... · Experimental Simulation of...

Documents

Transcript of Experimental Simulation of Closed-System Degassing inthe ... · Experimental Simulation of...