Storage and Evolution of Mafic and Intermediate Alkaline Magmas beneath Ross Island, Antarctica

We present the results of phase equilibrium experiments carried out on basanite and phonotephrite lavas from Ross Island, Antarctica. Experiments were designed to reproduce the P–T–X–f O 2 conditions of deep and intermediate magma storage and to place constraints on the differentiation of each of the two predominant lava suites on the island, which are thought to be derived from a common parent melt. The Erebus Lineage (EL) consists of lava erupted from the Erebus summit and the Dry Valley Drilling Project (DVDP) lineage is represented by lavas sampled by drill core on Hut Point Peninsula. Experiments were performed in internally heated pressure vessels over a range of temperatures (1000–1150 (cid:2) C) and pressures (200–400MPa), under oxidized conditions (NNO to NNO þ 3, where NNO is the nickel–nickel oxide buffer), with X H2O of the H 2 O–CO 2 mixture added to the experimental capsule varying between zero and unity. The overall mineralogy and mineral compositions of the natural lavas were reproduced, suggesting oxidizing conditions for the deep magma plumbing system, in marked contrast to the reducing conditions (QFM to QFM – 1, where QFM is the quartz–fayalite–magnetite buffer) in the Erebus lava lake. In basanite, crystallization of spinel is followed by olivine and clinopyroxene; olivine is replaced by kaersutitic amphibole below (cid:3) 1050 (cid:2) C at intermediate water contents. In phonotephrite, the liquidus phase is kaersutite except in runs with low water content ( X H 2 O ﬂuid < 0 (cid:4) 2) where it is replaced by clinopyroxene. Experimental kaersutite compositions suggest that the amphibole-bearing DVDP lavas differentiated below 1050 (cid:2) C at 200–400MPa and NNO þ 1 (cid:4) 5 to NNO þ 2. Olivine- and clinopyrox-ene-bearing EL lavas are consistent with experiments performed above 1050 (cid:2) C and pressures around 200MPa. The plagioclase liquidus at < 1–2wt % H 2 O suggests extremely dry conditions for both lineages ( X H 2 O ﬂuid approaching zero for EL, (cid:3) 0 (cid:4) 25 for DVDP), probably facilitated by dehydration induced by a CO 2 -rich ﬂuid phase. Our results agree with previous studies that suggest a single plumbing system beneath Ross Island in which DVDP lavas (and probably other peripheral volcanic products) were erupted through radial fractures associated with the ascent of parental magma into the lower crust. The longer travel time of the DVDP lavas through the crust owing to lateral move-ment along fractures and the lack of a direct, sustained connection to the continuous CO 2 -rich gas ﬂux that characterizes the main central Erebus conduit is probably responsible for the lower temperatures and slightly wetter conditions and hence the change in mineralogy observed.


INTRODUCTION
Ross Island, Antarctica (77 32'S, 167 10'E) is part of the Erebus Volcanic Province (EVP) and comprises five predominant volcanic centers. The currently active and volumetrically dominant Erebus volcano is surrounded by the now extinct subsidiary volcanic centers of Mts Bird, Terror, Terra Nova, and Hut Point Peninsula (Fig. 1). Erebus is known for its persistently active phonolitic lava lake, which has been at the focus of recent active monitoring and petrological studies as it represents a window into the shallow regions of the alkaline intraplate volcanic plumbing system. Recent investigations employing experimental petrological work (Moussallam et al., 2013) and seismic tomography (Zandomeneghi et al., 2013) have placed tight constraints on the physical and petrological properties of the shallow plumbing system and phonolitic lava lake including its geometry, temperature, pressure, and oxygen fugacity. These studies have the potential to refine the results of numerical models developed for the shallow Erebus system, such as those of Burgisser et al. (2012) and Molina et al. (2012). The intermediate and deep plumbing system beneath Erebus (>8 km depth) has been investigated via study of a suite of melt inclusions (Oppenheimer et al., 2011) and mixed volatile solubility experiments (Iacovino et al., 2013), which have placed constraints on the depths of degassing and the equilibrium fluid contents of primitive and intermediate composition magmas. Little is known, however, about the specific P-T-X-fO 2 conditions of magma storage and differentiation processes at these depths. Understanding the deep system at Erebus is crucial because the constant supply of deeply derived CO 2 -rich gases combined with a sustained energy and mass input into the lava lake suggests a direct link between the phonolitic lava lake and the mantle source of the parental magmas via a deep mafic plumbing system (Kyle et al., 1992;Oppenheimer & Kyle, 2008;Oppenheimer et al., 2011).
Gas emissions from the Erebus lava lake are rich in CO 2 ($30 mol %) and comparatively low in H 2 O ($65 mol %; Oppenheimer et al., 2011;Burgisser et al., 2012), consistent with the volatile compositions deduced from melt inclusions for the deeper system. The longevity of the lava lake itself is hypothesized to reflect, in part, the constant fluxing of heat and CO 2 -rich fluid (Oppenheimer et al., 2011) that originates from depths greater than those recorded by the most primitive melt inclusions (Iacovino et al., 2013a). Indeed, many separate studies have concluded that various aspects of the Erebus system are controlled by the influx of heat and CO 2 , including polybaric fractional crystallization throughout the magmatic system (Oppenheimer et al., 2011), the bimodal eruptive behavior in the lava lake (Oppenheimer & Kyle, 2008), shallow magma chamber processes (Moussallam et al., 2013), and fumarolic degassing and the formation of ice caves and ice towers (Curtis & Kyle, 2011). CO 2 flushing has been invoked to explain the characteristics of degassing and melt inclusions at other volcanoes including Stromboli (Allard, 2010), Vesuvius  (Marianelli et al., 2005), Etna (Spilliaert et al., 2006;Shinohara et al., 2008;Tagliani et al., 2012) and volcanoes in southeastern Guatemala (Walker et al., 2003) that sit atop large sedimentary carbonate sequences, the devolatilization of which could provide the necessary excess carbon (Iacono-Marziano et al., 2009). As Erebus has no such shallow carbon reservoir and probably derives all of its volatiles from the same mantle source responsible for magma production, we propose Erebus as the archetype for mantle-sourced carbon-controlled volcanism on Earth.
Subsidiary extinct volcanic centers and cones on Ross Island are distributed radially around Erebus at c. 120 from each other. This symmetrical distribution has been hypothesized to be the result of radial fractures created by the pressure of updoming mantle material on the crust directly beneath Mount Erebus (Kyle & Cole, 1974;Kyle et al., 1992). The rare earth element patterns in primary Ross Island basanitoids indicate the generation of a basanitic parent magma through lowdegree partial melting of the mantle source (Sun & Hanson, 1975, 1976. Major and trace element modeling suggests that lavas from Erebus [the Erebus Lineage (EL)] and the peripheral Hut Point Peninsula [the Dry Valley Drilling Project (DVDP) lineage] both evolved from a common basanite parent melt, which underwent two distinct fractional crystallization trends (Kyle, 1981;Kyle et al., 1992). So far experimental work has focused on the evolved magmas (Moussallam et al., 2013) and no experimental work is available to constrain storage and differentiation conditions of mafic lavas at Erebus (or other subsidiary volcanoes).
In this experimental study, one lava from each of the two Ross Island lava lineages (EL phonotephrite AW-82038 and DVDP basanite KI-04) were chosen as starting materials for equilibrium crystallization experiments to establish the P-T-X-fO 2 conditions necessary to produce the observed natural rock series. Phase equilibrium experiments are commonly used to constrain the pre-eruptive storage conditions of magmas, as well as their liquid line of descent (Sack et al., 1987;Moore & Carmichael, 1998;Freise et al., 2003;Martí et al., 2008;Andú jar et al., 2010;Di Carlo et al., 2010). Although nonequilibrium processes such as magma mixing and repeated replenishment of the system are important to consider, high pressure-temperature experimentation nevertheless illuminates the fundamental processes that control magma evolution from mantle to surface, allowing us to evaluate to what extent crystal-liquid equilibrium has been attained or maintained at depth.
Experiments were performed over a range of temperatures (1000-1150 C) and at pressures between 200 and 400 MPa to reproduce the intermediate region of the magma plumbing system where much early crystallization (differentiation of a basanitic parent to form intermediate lavas) is thought to occur. A temperature of 1000 C was chosen as a thermal lower bound, as the temperature of the surficial lava lake, which should be the coolest region of the plumbing system, has been constrained to $1000 C by several studies (Kyle, 1977;Caldwell & Kyle, 1994;Dunbar et al., 1994). The resulting experimental charges thus span the entire range of compositions seen in the natural lavas, allowing elucidation of the evolutionary processes occurring throughout the Ross Island magmatic plumbing system. With this approach, our aim has been two-fold: (1) to experimentally establish the conditions necessary to derive the distinct Erebus and DVDP lineages from a hypothesized single basanite parent; (2) to place constraints on the intermediate Erebus and DVDP plumbing systems with respect to pressure, temperature, oxygen fugacity, and dominant volatiles (H 2 O and CO 2 ).
Experiments with H 2 O-CO 2 fluid compositions in the range of XH 2 O ¼ 0-1 not only help elucidate the role of H 2 O in controlling phase equilibria, but also constrain the effect of CO 2 on crystallization processes via melt dehydration. Given that carbon dioxide is known to play an important role in the genesis of alkaline magmas, in both the mantle (Spera, 1984) and crust (Kaszuba & Wendlandt, 2000), it is important to understand its role at Erebus throughout the magmatic system (from mantle to surface).

Mineralogy and petrology of Ross Island lavas
The majority of lavas on Ross Island are represented by two strongly undersaturated sodic differentiation lineages: the Erebus Lineage (EL), made up of lavas erupted from Erebus volcano, and the Dry Valley Drilling Project (DVDP) lineage, made up of lavas on the Hut Point Peninsula and sampled via drill coring. The rarer Enriched-Iron Series (EFS) lavas are less silicaundersaturated and constitute only a small volume of Ross Island rocks. The EFS will not be discussed in this study. The DVDP and Erebus lineages are similar in terms of whole-rock chemistry as demonstrated by the similar variation in major element evolution of the DVDP and EL lavas with the exception of P 2 O 5 (Fig. 2), the variation of which is probably being controlled by apatite crystallization. Major element mass balance and trace element modeling of DVDP and EL rocks (Kyle, 1981;Kyle et al., 1992) suggests fractional crystallization from a common basanitic parent melt, with slight differences in differentiation conditions resulting in the two similar yet distinct lineages.
The mineralogy of the EL and DVDP lineage rocks from Ross Island has been extensively documented by Kyle et al. (1992) and Kyle (1981), respectively, whose observations are summarized below. These lineages share the same assemblage with two notable exceptions. First, kaersutitic amphibole is present in almost all DVDP lavas (except in basanites where it is rare) but is absent in EL lavas. Second, olivine is present in all EL lavas and in DVDP basanites, but occurs only as xenocrysts in more evolved DVDP lavas. In the EL rocks, olivine ranges in composition from Fo 88 in basanites to Fo 51 in phonolites. Olivine xenocrysts are rare and occur only in the most primitive lavas. In the DVDP lavas, olivine occurs only in basanites as euhedral to subhedral phenocrysts and in the groundmass.
Clinopyroxene is ubiquitous in all EL and DVDP lavas and occurs as euhedral to subhedral phenocrysts. Pyroxenes with green cores, which show irregular rims, possibly owing to resorption, are occasionally found in DVDP lavas. In the EL lavas, clinopyroxenes show oscillatory zoning in basanites and intermediate lavas, but no zoning in phonolites.
Kaersutite is absent in EL lavas but is an important phase in all DVDP lavas except the basanites where it is rare. In some phonotephrites, kaersutite ranges from euhedral grains to those showing strong resorption and oxidation. It was suggested by Kyle (1981), based on previous experimental work, that temperature and PH 2 O probably exert a strong control on the appearance or absence of kaersutite in DVDP and Erebus lineages, respectively.  Kyle, 1981) and Erebus Lineage (EL; blue triangles; Kyle et al., 1992). For whole-rock compositions the major element totals are normalized to 100%. One lava sample from each of these two lineages (red star: DVDP KI-04; blue star: EL AW-82038) was used as the starting material for the experimental work.
Feldspar is the modally dominant phenocryst phase in EL lavas (up to 40% of the mode in anorthoclasebearing phonolites) with compositions ranging from An 72 to An 59 in basanites, becoming increasingly albitic in the more evolved lavas. The feldspar in the EL phonolite is primarily anorthoclase. In the DVDP lava, feldspar occurs mostly as microphenocrysts or in the groundmass and is mainly labradorite (An 50-60 ) in basanites, and andesine to oligoclase in all other lava types.
Feldspathoids occur in the groundmass of the EL lavas, as nepheline, which contains up to 24% normative Ne. Microphenocrysts of apatite are common in most rocks of both lineages and are often found as inclusions in kaersutite and magnetite in the DVDP rocks. Pyrrhotite is present in all rocks from both lineages and occurs as small round blebs, suggesting the presence of an immiscible sulfide liquid in the Ross Island magmas. Opaque Fe-Ti oxides are common in both lava suites and are typically titanomagnetite as phenocrysts and in the groundmass, plus rare ilmenite phenocrysts.

Starting materials
To investigate the evolution of the Ross Island magmas and the two dominant lava lineages, two samples were chosen, one from each lineage (Table 1): AW-82038 is an Erebus Lineage phonotephrite lava (SiO 2 ¼ 48Á01 wt %; MgO ¼ 3Á18 wt %) from Turk's Head (Kyle et al., 1992) representative of intermediate EL lavas; KI-04 is a basanite (SiO 2 ¼ 43Á06 wt %; MgO ¼ 8Á92 wt %) collected from Hut Point Peninsula for this study and represents the less evolved magmas of the DVDP lineage. Figure 3 illustrates our starting material compositions relative to those of a suite of melt inclusions in Erebus Lineage rocks, which provides a representative cross-section through the evolutionary chemical trend of lavas on Ross Island (Oppenheimer et al., 2011).
The lavas were crushed to a powder and melted in a Pt crucible in a 1 atm furnace at 1300 C. Sample AW-82038 was held in the furnace for $30 min under a stream of nearly pure CO 2 (corresponding to an oxidation state near the Ni-NiO buffer, NNO) and then dropquenched into cold water [sample prepared by Iacovino et al. (2013a)]. Sample KI-04 was held in furnace for $1 h in air (i.e. extremely oxidizing conditions) and then removed from the furnace and immediately placed into cold water to quench. The resulting glassy samples were extracted from their crucibles, re-crushed, and remelted under identical conditions to ensure a homogeneous and crystal-free glass. A portion of each starting material was polished and analyzed by electron microprobe to verify that no compositional changes had occurred during melting (see Table 1).

Experimental techniques
Between 30 and 60 mg of glassy starting material powder was added to Au 80 Pd 20 capsules with lengths of $20 mm, an outer diameter of 2Á9 mm, and an inner diameter of 2Á5 mm. Liquid H 2 O was added via a microsyringe ($0-6 mg), and CO 2 was added as silver oxalate powder (Ag 2 C 2 O 4 , $0-18 mg). The X H2O in the H 2 O-CO 2 mixture of pre-run charges was varied at steps between zero and unity (i.e. X H2O ¼ 0, 0Á25, 0Á5, 0Á75, and 1), and the fluid proportion in the charge was always $10% to ensure fluid saturation (see Supplementary Data Appendix 1; supplementary data are available for downloading at http://www.petrology.oxfordjournals.org). Once filled, capsules were weighed, crimped at their open end, arc welded shut with a graphite electrode, and then subsequently re-weighed to check for any mass loss or gain during welding. After welding, capsules were placed in an oven at 110 C for several minutes and then re-weighed to check for any leaks in the capsule. As a third method to check for leaks, some capsules were held in a beaker of oil at 300 C for a few seconds. If bubbles were observed escaping from the capsule, it was considered to have leaks and was thus discarded. After completion of the experiment, capsules were weighed to check that no fluid loss occurred during the run. Any capsules that showed a change in mass >0Á2 mg either after welding or after heating were considered to have leaked at least part of their fluid phase and were discarded.
Experiments were run in vertically working internally heated pressure vessels (IHPVs; see Holloway, 1971) at the Institut des Sciences de la Terre Orlé ans (ISTO) at isothermal and isobaric conditions between 1000 and 1150 C and between 200 and 400 MPa with the oxygen fugacity of the vessel at about one log unit above the Ni-NiO buffer (NNO þ 1; Table 2). Each run contained 5-10 sample capsules loaded into a vertically working IHPV equipped with a double-wound Kanthal furnace. An Ar-H 2 gas mixture loaded sequentially at room temperature was used as the pressure medium with the Ar/H 2 ratio fixed so as to achieve an experimental fO 2 in the vessel of NNO þ 1 (Scaillet et al., 1992), although the fO 2 within each experimental charge varied slightly depending on its specific H 2 O/CO 2 fluid mixture (see below). It is noteworthy that, because the experimental fO 2 is lower than that imposed during melting of the sample KI-04 starting material, some additional H 2 O was created in the capsule during KI runs via the reduction of Fe 3þ . Thus, even nominally 'dry' runs will contain some H 2 O (typically < 1-2 wt %), and this has been taken into account for the calculation of experimental oxygen fugacity. Pressure was measured with a transducer calibrated against a Heise-Bourdon tube gauge with an accuracy of620 bars. Temperature was measured with two chromel-alumel (Type K) thermocouples, which allow for precise control over the length of the hotspot ($4 cm) to within 65 C. Experiments were run for 17-24 h and then quenched by dropping the sample into the lower, cold part of the pressure vessel (<40 C), which gives a quench rate of >100 C s -1 . The very low viscosities of our mafic alkalic glasses facilitated the attainment of equilibrium even given these relatively short run times. In similar studies on mafic melts, the attainment of equilibrium was observed in runs of similar duration (Di Carlo et al., 2006;Feig et al., 2010;Pichavant et al., 2014).

Fluid phase composition
For capsules that did not leak, the conventional weightloss method was employed as a first-order measure of the composition of the equilibrium H 2 O-CO 2 fluid phase (see Behrens et al., 2009). After the first post-run weighing, the capsule was submerged in a liquid nitrogen bath, thus freezing any liquid water or water vapor but allowing the CO 2 to remain as a gas. The capsule was then punctured to allow for the escape of the CO 2 gas, and a hissing noise and bubbles at the puncture site Whole-rock composition. AW-82038 analyzed by Kyle et al. (1992); KI-04 analyzed in this study, using X-ray fluorescence (XRF).
2 Electron microprobe (EMP) analyses of glass composition after vitrification in 1 atm furnace, but before run in pressure vessel. AW-82038 vitrified at an oxygen fugacity of NNO þ 1 [analysis from Iacovino et al. (2013)]; KI-04 vitrified in air (analysis from this study).
were typically observed. After warming at room temperature for up to 1 min, the capsule was weighed to determine the mass of CO 2 lost after puncture. This was then recorded as the mass of CO 2 in the fluid phase.
The capsule was then placed in a drying oven at 110 C for several minutes to evaporate the remaining H 2 O and then weighed to determine the mass of the H 2 O fluid. During this step, capsule weight was checked periodically until its mass remained constant. It should be noted that this method is an indirect measure of the composition of the fluid phase in our runs and can be prone to significant error owing to uncertainties in the weighing of the capsules before and after opening. Behrens et al. (2009)

SEM and electron microprobe
Charges were initially studied using a JEOL WINSET JSM 6400 scanning electron microscope (SEM) at ISTO for preliminary phase identification, modal abundance determination and textural analysis. Most phases could be identified by visual inspection, and several crystals from each phase were analysed using energy-dispersive spectrometery (EDS). The EDS analyses are not reported here as they were used only for mineral identification.
Representative SEM images from each sample were digitally processed to measure total crystallinity and the modal abundance of all phases. Images were converted to grayscale and a threshold was applied and adjusted to isolate each phase. The total area of each phase was then calculated as a percentage of the total image. Any areas of the image that contained no sample (holes, cracks, etc.) were not included in the total sample area.
Electron microprobe (EMP) analyses of the major element compositions of the glass and crystal phases in the natural samples, starting glasses and experimental run products were made primarily using a Cameca SX-100 electron microprobe at the University of Cambridge. Some samples were analyzed on a Cameca SX-100 at the University of Arizona or on a Cameca SX-50 at the Università degli Studi di Padova. Glass was measured using an accelerating voltage of 15 keV, a beam current of 3 nA, a defocused beam size of 10 lm, and a peak counting time of 10 s, analyzing Na first to minimize any measurement error owing to Na migration. Crystal phases were analyzed with an accelerating voltage of 15 keV, a beam current of 10 nA, a focused beam with a diameter of 1 lm, and a peak counting time of 20 s. In each experimental charge, typically 20-30 analyses were performed on glass and 8-12 analyses were performed on each crystal phase.

FTIR spectroscopy
Whenever possible, the H 2 O and CO 2 concentrations in the glass phase of each experimental charge were measured using a Thermo Scientific Nicolet iN10 MX mapping FTIR system at the US Geological Survey in Menlo Park, California. Transmission IR spectra of doubly polished glass chips ranging in thickness from $20 to 150 lm were obtained in the 6000-1000 cm -1 wavenumber range using a KBr beamsplitter with a variable aperture size between 25 and 500 lm 2 . The sample stage was continuously purged with dry air, and a background measurement was made before the collection of each spectrum to minimize and correct for the contribution of atmospheric H 2 O and CO 2 . Moderately to highly crystal-rich charges were characterized with FTIR mapping to identify crystal-free regions for the analysis of dissolved volatiles.
The concentrations of OH, H 2 O, CO 2 , and CO 3 species were calculated with the Beer-Lambert Law (Ihinger et al., 1994), using near-IR (4500 and 5200 cm -1 ), total water (3500 cm -1 ), molecular H 2 O (1640 cm -1 ), CO 2  (2345 cm -1 ), and carbonate (1430 and 1525 cm -1 ) bands. The total water 3500 cm -1 band was always preferred for retrieving H 2 O concentrations; in five samples the peak at this band was unusable and so near-IR bands were used. The H 2 O concentrations in samples measured with both the total water and near-IR bands agree well, with an average standard deviation of 0Á57 wt % and an R 2 value of 0Á83. Sample thicknesses were measured to an accuracy of 62 lm using a digital tabletop micrometer equipped with a fine tip that allowed for measurement at multiple spots on the sample surface. Additional sample thicknesses for some samples were also measured using an ADE Phaseshift Microxam surface-mapping microscope (with white light interferometer) to an accuracy of 60Á5-2 lm. In very thin samples, the presence of interference fringes in reflectance IR spectra allowed for the measurement of the sample thickness in the precise spot that was analyzed with transmitted light, following the method of Nichols & Wysoczanski (2007, and references therein; see Supplementary Data Appendix 2 for details).
Glass densities were calculated for each sample using the model of Lange & Carmichael (1987), assuming an H 2 O molar volume of 22Á9 cm 3 mol -1 (Ochs & Lange, 1999). Molar absorption coefficients strongly depend on glass composition (Pandya et al., 1992). Because of variation in the crystallinity of our samples, and therefore significant variation in glass chemistry, an attempt was made to use absorption coefficients appropriate for each sample. All absorption coefficients used in this study are listed in Table 3. Where possible, the model of Dixon & Pan (1995) was used to calculate e 1525 and e 1430 (carbonate absorption) for single samples. This calculation can only be applied to glasses with an Na/(Na þ Ca) ratio between 0Á25 and 0Á5 (in our case, most glasses classified as phonotephrite and tephriphonolite fall in this range). For glasses with Na/(Na þ Ca) < 0Á25 or > 0Á5, e 1525 and e 1430 values were selected from the literature (see Table 3). (See Supplementary Data Appendix 2 for details of FTIR spectral background fitting, absorbance measurements, sample thicknesses, and glass density calculations.) Dissolved volatile concentrations could not be measured in nine crystalline samples (29-75 vol. % glass) where glass pools were too small to analyze by FTIR spectroscopy. A further six glassy samples (95-99 vol. % glass) could not be successfully prepared for FTIR analysis. For these 15 samples, H 2 O and CO 2 contents were calculated using the mass-balance method (see Supplementary Data Appendices 3 and 4 for details), and the 35 samples for which FTIR measurements were obtained were used as internal standards. The standard error for these calculations is large at 61Á4 wt % and 61000 ppm for calculated H 2 O and CO 2 values, respectively.

Oxygen fugacity
The fO 2 of all IHPV runs was controlled by loading the pressure vessel with sufficient H 2 to achieve an apparatus fO 2 of $NNO þ 1. The fO 2 inside each experimental capsule will vary depending on the H 2 O/CO 2 ratio of the fluid phase contained therein. Assuming that the fH 2 of the pressure vessel will be the same as the fH 2 inside an experimental capsule owing to the very fast diffusion rates of hydrogen, we can calculate the fO 2 inside each capsule following the method of Iacovino et al. (2013a). First, because the partial pressure of H 2 is fixed for each experiment, we can calculate the fH 2 of the pressure vessel for each run using the equation where / is the fugacity coefficient of hydrogen for the P and T of a given experiment (Shaw & Wones, 1964), and P H2 is the partial pressure of hydrogen gas added to the IHPV at the start of the run. If the XH 2 O fluid is known, we can also then calculate the fH 2 O inside the capsule as where fH 2 O pure is the fugacity of pure H 2 O vapor at the experimental P and T conditions, assuming ideal mixing between H 2 O and CO 2 in the fluid [calculated with the modified Redlich-Kwong equation of state of Holloway (1977) as modified by Flowers (1979)]. Now, it is possible to calculate the oxygen fugacity for each experiment as follows: where K F is the equilibrium constant of formation for H 2 O (Robie et al., 1979). Oxygen fugacities in our experiments are reported in Table 2 and range from a DNNO of þ0Á3 to þ3Á2.

Textural and mineralogical observations
Optical and electron microscope images revealed homogeneously distributed euhedral crystals in most experimental charges, with crystals ranging in size from <5 mm to $80 mm in length. Phases identified include olivine, clinopyroxene, plagioclase, amphibole, Fe-Ti oxides, apatite, glass and fluid inclusions. In the most highly crystalline charges, glass tends to collect in pools and small subhedral crystals ('quench' texture) are ubiquitous. Quench texture is especially common in KI samples with high H 2 O contents, probably owing to the extremely low-viscosity melt allowing for the rapid crystallization upon quench. In samples where crystals were too small to analyze by electron microprobe (<25 mm 2 ), phases were analyzed quantitatively for identification purposes only. Those analyses are not reported here.

Phase relationships
The In AW, kaersutite is the liquidus phase at median to high water contents, crystallizing below about 1025 C. Amphibole is common in the AW experiments and is stable to melt H 2 O contents as low as $1 wt %. A reaction relationship between kaersutite and clinopyroxene is inferred to exist around XH 2 O fluid ¼ 0Á1-0Á2, below which clinopyroxene is the dominant calcic ferromagnesian phase (experiments AW-19B and AW-43). As with kaersutite, clinopyroxene is restricted to below about 1025 C. Plagioclase feldspar is the dry liquidus phase and is restricted to low water contents, with its stability field increasing with decreasing temperature. Phase equilibria at 300 MPa are similar to those at 200 MPa [the stability field for clinopyroxene at 300 MPa ( Fig. 4b) is inferred based on that at 200 MPa (Fig. 4a)]. Although apatite was not always detected, it is inferred to crystallize below 1100 C at both pressures. Its stability is mostly constrained by water-saturated experiments, but apatite is probably stable but undetectable in drier runs (see Harrison & Watson, 1984). This hypothesis is supported by the steady decrease in the P 2 O 5 content of residual liquids in phonotephrite experiments. The polybaric section at 1000 C (Fig. 4c) shows that in the pressure range 200-400 MPa amphibole is always the liquidus phase, followed by Pl and Cpx, in that order.
In KI experiments at 200 MPa (Fig. 5a), olivine is the liquidus phase followed by clinopyroxene and kaersutite. Although loosely constrained, the liquidus is located above 1100 C except at very high H 2 O contents (XH 2 O in approaching unity), where it drops to $1075 C. A peritectic-like relationship exists at intermediate melt water contents and 1050 C at 200 MPa where olivine is replaced by kaersutite at low temperatures. At these P-T conditions, clinopyroxene replaces kaersutite as the dominant ferromagnesian phase at low and high melt water contents where it coexists with olivine, except at very low melt water contents (XH 2 O in approaching zero) where clinopyroxene is the only stable silicate phase. The phase relationships observed at 400 MPa (Fig. 5b) are broadly similar to those at 200 MPa except that the kaersutite stability field is enlarged towards H 2 O-rich conditions, this mineral being in reaction relationships with both olivine and clinopyroxene as temperature decreases. The polybaric section at 1100 C (Fig. 5c) illustrates that olivine is the liquidus mineral in the pressure range 200-400 MPa, for a large range of fluid compositions (melt water content), followed by clinopyroxene at XH 2 O in < 0Á25.  Isothermal (a, b) and isobaric (c) phase diagrams illustrating the stability fields of different phases in KI experiments. Numbers next to points indicate wt % H 2 O dissolved in the glass of the experimental charge as measured by FTIR spectroscopy. Although not shown, titanomagnetite is ubiquitous in all charges. Apatite was not always detected but is inferred to be stable in runs below $1100 C. A dashed line indicates an inferred phase boundary. Fig. 4. Isothermal (a, b) and isobaric (c) phase diagrams illustrating the stability fields of different phases in AW experiments. Numbers next to points indicate wt % H 2 O dissolved in the glass of the experimental charge as measured by FTIR spectroscopy. Although not shown, titanomagnetite is ubiquitous in all charges. Apatite was not always detected but is inferred to be stable in runs below $1100 C. A dashed line indicates an inferred phase boundary.

Experimental phase compositions
Compositions of the experimental mineral phases and residual glasses are given in Table 4. Chemical analyses for all natural and experimental phases are given in Supplementary Data Appendix 5. Titanomagnetite, apatite, and some clinopyroxene crystals were too small to be analyzed by electron microprobe (<25 mm 2 ). In six charges, clinopyroxenes were large enough to be analyzed, and their compositions are reported in Table 4 (Fig. 6), reflecting the increase in the melt fraction of the charge and a change in oxygen fugacity caused by an increase in the water activity in the melt (e.g. Feig et al., 2010). Crystal-melt Fe-Mg partition coefficients (K D ) for most experimental olivines range from 0Á33 to 0Á39, with an overall average value (including outliers) of 0Á33. Our olivines are expected to have a higher K D value than the canonical 0Á3 owing to the alkalinity and redox state of the charges. Using the model of Carmichael & Ghiorso (1990), which accounts for the effect of melt composition on K D , we calculate that our experimental K D values should range between 0Á34 and 0Á36, suggesting that the majority of the olivine crystals in our experiments are approaching equilibrium with their host melts. Four outliers, which do not appear to have attained equilibrium, are samples KI-06 (K D ¼ 0Á22), KI-08 (0Á22), KI-10 (0Á52), and KI-36 (0Á43). In samples KI-06 and KI-10, olivines were rare and small, suggesting that they grew late during the experiment, possibly during quench. Some olivines may have been affected by Fe loss to the experimental capsule, which is difficult to avoid at high T even given the relatively short run times of 24 h.
Amphibole is classified as sodic-kaersutite on the basis of 23 oxygens and crystallized in both phonotephrite and basanite experiments (compositions reported in Table 4). Al contents in kaersutites show a positive correlation with pressure, as documented for other systems (e.g. Prouteau & Scaillet, 2003) but in contrast to the results of experiments with Erebus phonolite (Moussallam et al., 2013). Kaersutites synthesized at the lowest fO 2 conditions have markedly lower Mg# and higher Ti contents. At all pressures, the modal proportion of amphibole in AW runs increases as the water content of the melt decreases.
Plagioclase feldspar was observed only in the driest AW experiments and ranges from An 57 to An 65 . It was not found in any of the KI experiments.

Glass compositions
The compositions of residual experimental glasses highlight the evolution of the melts from the starting composition to intermediate and evolved compositions. FeO, MgO, CaO, and TiO 2 all show continuously decreasing trends with melt evolution reflecting the crystallization of pyroxene or kaersutite, olivine, plagioclase, and titanomagnetite. In contrast, Al 2 O 3 , and K 2 O show continuously increasing trends. In detail, Harkertype plots clearly show the effect of the fractionating assemblage on the liquid lines of descent of each starting material (Figs 7 and 8).
KI-related melts in charges held above 1050 C show limited compositional evolution relative to the starting composition, except for FeO and MgO contents, which are affected by olivine and titanomagnetite crystallization to some extent. In contrast, charges held at or below 1050 C show the prominent roles of either kaersutite or Cpx-Ol-dominated assemblages. In both cases, derivative liquids produced by crystallization at these temperatures reach SiO 2 contents of 50-52 wt % (i.e. are more evolved than the AW starting composition), trending toward the phonotephritic to phonolitic fields. Charges in which kaersutite is stable are characterized by higher TiO 2 , FeO, and CaO melt contents relative to those in which Cpx and Ol are stable (at the same SiO 2 content). In contrast, the crystallization of a Cpx 6 Ol assemblage produces liquids with higher Al 2 O 3 , Na 2 O, and K 2 O contents relative to those in which kaersutite predominates. It is worth noting that the role of kaersutite increases at high pressure; charges crystallized at 400 MPa have up to 60 wt % amphibole, whereas at 200 MPa, amphibole does not exceed 37 wt % of the crystal cumulate ( Table 2).
The relationship between the evolution of CaO and MgO in KI liquids is illustrated in Fig. 9. Such a plot is useful for illuminating the evolution of basaltic compositions sensu lato in which olivine and Ca-pyroxene (or amphibole) play a dominant role (e.g. Pichavant et al., 2009Pichavant et al., , 2014. The KI trend can be divided into two segments: the first runs broadly parallel to the MgO axis, primarily reflecting depletion in MgO of the liquid owing to olivine crystallization. The second segment corresponds to liquids in which either kaersutite or Cpx 6 Ol crystallize, defining either a kaersutite control line or the Cpx-Ol cotectic (at 1050 C), along which both CaO and MgO contents decrease. It is apparent that liquids lying on the Cpx-Ol cotectic are slightly less CaOrich than those driven by kaersutite crystallization. Also shown in Fig. 9 are the Cpx-Ol cotectics experimentally determined for Stromboli and Vesuvius primitive basalts (Di Carlo et al., 2006;Pichavant et al., 2009Pichavant et al., , 2014, which, like the magma compositions investigated here, are mafic and alkali-rich. It is clear that basaltic liquids    saturated in Cpx 6 Ol and kaersutite define a common general and relatively narrow trend in such a diagram, despite the compositional differences existing between the three bulk-rock compositions used to establish the cotectics. Significant departure of a Cpx 6 Ol-bearing mafic rock composition from this trend might indicate that accumulation or contamination processes of some sort have occurred. AW-related charges produced liquids compositionally different from the starting material mostly at 1000 C. Liquids produced at higher temperatures cluster broadly around the bulk-rock AW composition. As observed for KI, two distinct trends appear depending on whether kaersutite or Cpx 6 Pl is the dominant fractionating silicate. In both cases, the most evolved compositions reach the phonolitic field with SiO 2 in excess of 52-53 wt %. Kaersutite-saturated liquids in particular have higher MgO and CaO contents, whereas liquids crystallizing Cpx 6 Pl have higher K 2 O (at the same SiO 2 content). For most major elements in AW experiments, there is no apparent effect of pressure, liquid compositions being governed mostly by variations in T and H 2 O. The sole exception is Al 2 O 3 : liquids produced at 400 MPa are distinctly richer in Al 2 O 3 relative to those at 200 MPa, the former having >20 wt % Al 2 O 3 whereas at 200 MPa the majority of liquids have less than 20 wt %. This reflects in part the fact that, for the AW composition, kaersutite crystallization depletes the liquid less in Al 2 O 3 relative to Cpx and Pl.

DISCUSSION
Pre-eruptive conditions of KI and AW magmas and the production of the phonolite lava lake Overall, the experimental mineral phases reproduce those observed in the natural rocks, showing that our experiments correctly capture the natural system in terms of fO 2 -T-H 2 O conditions of evolution. In detail, experiments best reproducing the Ol composition of KI (Fo 78-80 ) are at around 1100 C and with less than 1Á5 wt % dissolved H 2 O (corresponding to XH 2 O fluid < 0Á2; charges KI-10, KI-14 and KI-23, Table 4), in the pressure range 200-300 MPa. Higher melt water contents would yield olivine that is too magnesian. The natural DVDP clinopyroxene composition was best reproduced at 200 MPa and 1050 C as shown in Fig. 10 (sample KI-12; En 42 Wo 47 Fs 11 ). Other experimental pyroxenes are more iron-rich (Fs 14-17 ) than those in the natural samples, although the most Fe-rich clinopyroxenes measured (KI-23) are possibly quench crystals, as indicated by high variability in Ca and Na, and high K. It should be noted that the experimental clinopyroxenes are in general rich in Al 2 O 3 and Na 2 O, consistent with those in the natural rocks (Kyle, 1981). This Al-and Na-rich character of Cpx does not therefore reflect crystallization at higher pressure (e.g. Kyle, 1981) but instead the Al-and Na-rich character of the melt from which the pyroxenes grew. Experiments between NNO þ 1Á5 and NNO þ 2 produced kaersutite that most closely resembles the natural amphiboles in terms of Ti content ($5Á5-6Á5 wt % TiO 2 ) and average Mg# (65), although the natural amphiboles have a significant spread in Mg# (56-72) (Fig. 11). Most natural DVDP amphiboles contain 12Á5-14 wt % Al 2 O 3 (Kyle, 1981), resembling the compositional range of experimental AW amphiboles. Given the correlation between kaersutite Al contents and pressure (Fig. 12), this suggests that significant DVDP differentiation [in which the role of amphibole is important (Kyle, 1981)] occurred at 200-400 MPa. The An contents in experimental phonotephrite feldspars range from An 57 to An 65 and closely mimic the compositions of natural phonotephrite feldspars (Fig. 13). The fact that the Or contents of experimental plagioclases are higher than those of the natural rocks is probably due to contamination of the microprobe analyses by surrounding glass.
The experiments allow us to constrain possible petrogenetic paths for magma production at Erebus. We first consider the production of intermediate, AWlike, magmas from a KI-type parent. To this end, the AW bulk-rock has been plotted in Fig. 8 displaying KI-related liquids. As can be seen here, an AW-type liquid (48 wt % SiO 2 ) is bracketed for several oxides by charges run at 1050 C and 200 MPa (between KI-09 and KI-10), in particular for Na 2 O, K 2 O and CaO, and to a lesser extent Al 2 O 3 and TiO 2 . FeO, which is too low in the experiments (8 wt % vs 9Á4 wt %), does not fit these trends, suggesting therefore that the experimental fO 2 is slightly too oxidized compared with nature. The   7. Harker variation diagrams for AW experimental glasses illustrating the production of a phonolite magma resembling that in the Erebus lava lake [the composition used by Moussallam et al. (2013); black dot] by crystallization of an AW-type magma. The liquids that best approximate the lava lake composition are the driest ones at 1000 C (<1Á7 wt %) for all pressures. The closest experimental analogues are charges  MPa. Lower than natural FeO concentrations in these two runs could indicate that the experiments were too oxidized relative to the natural magmas. High experimental CaO contents may indicate that the production of lava lake phonolite requires lower temperatures, possibly around 975 C. The roles of kaersutite and Cpx 6 Pl crystallization are clearly defined by two distinct evolutionary trends. Fig. 8. Harker variation diagrams for KI experimental glasses illustrating the production of intermediate AW-like magmas (represented here by the AW whole-rock; blue dot) from a KI-type parent. At low T, two evolutionary trends are apparent in which either Cpx 6 Pl or kaersutite amphibole is the dominant fractionating phase. An AW-type daughter liquid is bracketed for several oxides by experiments KI-09 and KI-10, thus constraining the melt water content reached at the stage of AW magma production to between 0Á7 and 1Á5 wt %. FeO is too low in the experimental charges relative to natural AW whole-rocks, indicating that the experiments were more oxidized than the natural magmas.
corresponding melt water content reached at the stage of AW magma production is thus bracketed by that of those two charges, between 0Á7 and 1Á5 wt % (Table 2). We next evaluate the conditions for the production of a phonolite magma similar to that of the lava lake by crystallization of an AW-type magma (Fig. 7). The experimental AW liquids define a trend toward the lava lake composition [taken as the composition used by Moussallam et al. (2013)], although that end-member is not bracketed by the experimental results. At all pressures, the liquids best approaching the lava lake are the driest ones at 1000 C (<1Á7 wt %). In detail, the liquids obtained at 400 MPa are too Al 2 O 3 -rich relative to that of the lava lake, suggesting this is too high a pressure for phonolite production in the Erebus plumbing system. In contrast, the two driest liquids at 200 MPa (AW-19B and AW-20B) are close synthetic analogues for the lava lake, charge AW-19B having the observed mineralogy of the lava lake. Notably, the TiO 2 trend falls short, with experimental liquids being 0Á5 wt % too rich in TiO 2 relative to the lava lake, suggesting a lower temperature for the natural system. FeO concentrations in those two driest 200 MPa charges are slightly below that of the lava lake, which could be taken to indicate that the experimental fO 2 is slightly too high, as inferred above for the production of AW magmas. The MgO and CaO trends clearly distinguish the role of kaersutite versus that of Cpx 6 Pl, suggesting again the latter assemblage as the dominant one during differentiation of the EL. It is in CaO content that the difference between the . KI runs are shown with different symbols in terms of P-T conditions. The Cpx-Ol cotectic at 1050 C is shown for KI melts. Also plotted are experimentally determined Cpx-Ol cotectics for melts from Stromboli  and Vesuvius (Pichavant et al., 2014). Despite compositional differences in the starting materials used in these three studies, mafic alkali-rich liquids saturated in Cpx 6 Ol and kaersutite define a relatively narrow evolutionary trend in this diagram. KI melts are marked by initial Ol crystallization, followed by either Cpx 6 Pl or Kaers. experimental liquids and the lava lake is greatest. Inspection of the experimental CaO trends suggests that the production of a magma with a CaO content similar to that of the lava lake requires temperatures below 1000 C, possibly around 975 C. This is in agreement with the work of Moussallam et al. (2013), who experimentally determined the liquidus of the phonolite lava lake under nearly dry conditions at 975-1000 C.
In summary, KI near-liquidus conditions are inferred to be around 1100 C and 1 6 0Á5 wt % H 2 O at 200-300 MPa. Production of intermediate magmas at Erebus (AW-type) from a KI-type parent requires a temperature drop to around 1050 C at a pressure of 200 MPa, reaching slightly higher melt water contents but still below 1Á5 wt %. Redox conditions appear to be moderately oxidized at this stage, possibly between those defined by the quartz-fayalite-magnetite (QFM) and NNO solid buffers, or just slightly above NNO. AW-type magma may in turn crystallize down to 975-1000 C, possibly in the pressure range 100-200 MPa, yielding a phonolitic liquid similar to that sustaining the lava lake, which is characterized by very low water contents and reduced conditions. As shown by Moussallam et al. (2014), the transition between moderately oxidized conditions in the crustal magma reservoir and the more reduced conditions typifying the lava lake may result from degassing processes involving melt-COHS fluid equilibrium (see below). Although the temperature of the lava lake has been inferred to lie somewhere between 950 and 1080 C (see Introduction), the present study, in combination with that of Moussallam et al (2013), consistently favors the lower bound of that temperature range.  Kyle (1981). Error bars represent 1SD.   Kyle et al. (1992)]. It should be noted that all experimental plagioclase crystals were very small and so their chemical analyses via EMP may be contaminated by the surrounding glass.

Differences between EL and DVDP magmatic evolution
Although neither experimental series completely reproduced the phase assemblages of the natural Erebus starting lavas (i.e. occurrence of kaersutite in EL experiments whereas it is not present in the corresponding rocks), our results shed light on critical aspects of the Erebus system and elucidate key differences in the formation conditions of the two natural lava suites. Plagioclase is confined to very low H 2 O-high CO 2 conditions, suggesting maximum melt H 2 O concentrations of $1-2 wt % for both EL and DVDP lava suites at the P-T conditions explored, consistent with melt inclusion volatile contents (Oppenheimer et al., 2011). The presence of kaersutite in all except the most primitive natural DVDP lavas suggests higher water activities for the DVDP suite compared with the EL, with XH 2 O fluid around 0Á2, indicated by the co-precipitation of plagioclase and kaersutite in phonotephrite experiments. The peritectic-like relationship inferred between olivine (6 clinopyroxene) and kaersutite allows us to place constraints on the relative temperatures of the EL and DVDP suites. All EL lavas plus DVDP basanites (the most primitive of which probably represents the parent melt for both suites) contain olivine and lack kaersutite. Conversely, all other DVDP lavas lack olivine and contain kaersutite. This is consistent with EL and primitive DVDP lavas being thermally buffered during the early and intermediate stages of differentiation to above 1050 C, with intermediate and evolved DVDP lavas undergoing a drop in equilibration temperature to below 1050 C.
The chemical compositions of both experimental series (squares and diamonds) and natural EL melt inclusions (shaded fields) are shown in variation diagrams of their major oxides versus SiO 2 in Fig. 14. Because of the rarity of melt pools large enough to be measured in the DVDP lavas, no residual melt compositions exist in the literature.
Many of the experimental glass compositions match the natural trends fairly well, but some are clearly offset, indicating a difference in natural versus experimental P-T-H 2 O-fO 2 conditions for those charges.
Mismatch between natural and experimental glass compositions is notable in particular for CaO and MgO, and to a lesser extent for Al 2 O 3 . Higher than observed CaO values suggest insufficient fractionation of plagioclase and/or calcic clinopyroxene in the experiments compared with the natural lavas. The lack of plagioclase in the KI experiments and moderate plagioclase crystallization in the AW experiments may indicate that the liquid line of descent we have established experimentally is either hotter or, more probably, wetter than that of the natural system. Deep and intermediate magmas are not expected to be significantly cooler than the temperature of the surficial lava lake, which provides a thermal lower bound of $975-1000 C (Kyle, 1977;Caldwell & Kyle, 1994;Dunbar et al., 1994;Moussallam et al., 2013). It is thus unlikely that significantly lower temperatures are responsible for enhanced calcic plagioclase crystallization in natural samples. For this reason, we suggest that the Ross Island magmas differentiated under lower water activities in our experiments, significantly constraining the natural XH 2 O fluid conditions of the DVDP and EL series to \ll 0Á25. The elevated MgO contents in the KI experiments relative to natural melt inclusions may reflect the lack of a cardinal phase consuming this element, in particular olivine, which in turn could be controlled by oxygen fugacity. Low experimental Al 2 O 3 contents in KI melts may reflect insufficient plagioclase crystallization in addition to illustrating the dominant role of kaersutite fractionation in the DVDP magmas compared with EL melts, whose evolution is defined instead by Cpx fractionation.
Trends in the K 2 O and P 2 O 5 contents of natural EL glasses are distinct from those of EL and DVDP wholerocks (Figs 2 and 14). Experimental K 2 O melt contents mimic those of the whole-rock samples in that they are depleted relative to EL melt inclusions, probably reflecting the dominant role of kaersutite in our experiments and in natural DVDP rocks. The trends in P 2 O 5 content illustrate well the onset of apatite saturation in experimental charges and natural EL melt inclusions. In EL lavas, apatite saturation occurs when the melts have differentiated such that the silica content is !48-49 wt %, after which P 2 O 5 decreases with increasing SiO 2 (Fig. 14g), in agreement with existing solubility models (Harrison & Watson, 1984). This suggests a hotter temperature of c. >1100 C for the primitive and intermediate EL magmas, and a lower temperature (indicated by the onset of apatite saturation) for the evolved lavas.

Redox constraints on Ross Island magmatism
The presence of kaersutite in our experimental samples is largely independent of H 2 O content. The growth of amphibole in relatively dry experiments indicates that some other component is taking the place of OH À in the crystal structure. The role of fluorine is usually invoked to explain the enhanced stability of amphibole, but this has not been detected in the kaersutite of the DVDP rocks (Kyle, 1981). Unusually low H contents have been measured in other natural kaersutites (notably those from Martian meteorites) suggesting that the amphibole does not require a hydrous environment to grow (Popp & Phillips, 1995) In addition to Cl À and F À occupying O3 site occupancies, oxy substitutions in kaersutite are favored by high Ti and Fe 3þ contents (Popp & Taras Bryndzia, 1992) via the reactions This may explain the growth of kaersutite in our driest experiments, especially given the oxidizing experimental conditions, which will favor oxy substitutions via the reaction in equation (4). The presence of kaersutite in relatively dry natural samples thus may indicate the prevalence of oxidizing conditions for the DVDP lavas. Given that the natural phase assemblages were reproduced under the relatively oxidizing experimental conditions employed, we infer a similarly oxidizing environment for the deep and intermediate Ross Island magmas. This is in direct contrast to the reducing conditions (QFM -1Á22) inferred for Erebus's phonolitic lava lake (based on oxide pairs; Kyle, 1977) but is consistent with a reducing upwards redox trend that can be inferred for the EL. Based on the experimental study of Moussallam et al. (2013), a phonolite reservoir at 200 MPa has an oxygen fugacity of QFM -0Á5 at 1000 C. Intermediate tephriphonolites at 300 MPa have an estimated oxygen fugacity of QFM -0Á44 at 1081 C [based on oxide pairs from sample 97009 from Kyle et al. (1992)]. Coupled chemical-physical modeling of conduit flow by Burgisser & Scaillet (2007) has shown that reducing upwards behavior is expected for some sulfur-bearing volcanic systems. This is confirmed by the systematic analyses of the Fe redox state of a suite of melt inclusions at Erebus, which have conclusively shown a strong reduction trend as pressure decreases (Moussallam et al., 2014). In addition, the large CO 2 flux proposed as the mechanism for dehydrating Erebus magmas might also contribute to the reduction in oxidation state of the shallow Erebus system by reducing H 2 O fugacity.

The case for CO 2 -dominated volcanism at Erebus
The results of this study add to the compelling evidence that Erebus volcano is a remarkably CO 2 -rich, H 2 O-poor system. Melt inclusion volatile contents from Erebus maintain markedly high C/H ratios, even in intermediate and evolved melts. Conceptually, this can be explained by the presence of a CO 2 -rich gas buffering the melt to higher dissolved CO 2 concentrations, an interpretation consistent with previous modeling (Oppenheimer et al., 2011) and experimental work (Iacovino et al., 2013). As discussed above, the natural phase assemblages of the starting lavas in this study could be approximated only under extremely H 2 O-poor conditions. Even for KI-type lavas, which are slightly wetter than the AW lavas, our experiments suggest that low-H 2 O conditions dominate (XH 2 O fluid $ 0Á2). More evolved kaersutite-bearing lavas of the DVDP lineage were best reproduced in terms of modal abundance of amphibole (<3-5%) only in experiments with higher H 2 O contents (low-H 2 O runs containing 15-19% amphibole), indicating a possible increase in water activity in evolved DVDP magmas. Given that the Erebus system is thought to be fluid-saturated, our results agree with the conclusions of Oppenheimer et al. (2011), who suggested that carbon-induced dehydration of Erebus magmas at various depths promotes crystal fractionation and magmatic differentiation.
Even though our driest experimental melts contain only 1-2 wt % H 2 O, insufficient plagioclase crystallization and higher than observed CaO contents in our experimental glasses indicate that our experimentally established liquid line of descent is more water rich than the natural system. This suggests a maximum H 2 O content of 1-2 wt % in Ross Island basanites and a maximum of $1 wt % in EL phonotephrites. Moussallam et al. (2013) demonstrated that evolved Erebus melts (i.e. lava lake phonolite) must have extremely low water contents, with a maximum of $0Á5 wt % (but probably even below 0Á2 wt %). The very tight constraints that these two studies place on the maximum water contents of Erebus magmas very closely match the trends in H 2 O contents of Erebus melt inclusions (Oppenheimer et al., 2011), which range from 1-2 wt % in basanite to 0Á5 wt % in all other lavas, phonolites having 0Á2 wt %.
Our results are consistent with the presence of a CO 2 -rich fluid in the EL plumbing system but not in the DVDP. Given that both lineages are thought to be derived from the same parental melt (Kyle et al., 1992), we suggest that a sustained connection of the EL plumbing system to its deep mantle source and a subsequent CO 2 -rich fluid flux that characterizes the EL (and lack of such in the DVDP) is responsible for the differences in storage conditions and thus mineralogy of the two lava suites. The decoupled ascent of CO 2 -rich vapor through the magmatic system of Erebus may explain the extreme dehydration of the EL melts, as CO 2 will degas early and deep and leach H 2 O from the melt and into the vapor phase upon ascent (Mé trich & Wallace, 2008). CO 2 fluxing through the EL system can also explain the sustained high-temperature regime of EL magmas inferred from phase equilibria (i.e. olivine-kaersutite stability) and would suggest a more reducing environment relative to the DVDP, consistent with enhanced kaersutite stability under oxidizing conditions.
One source, two lava lineages: pre-eruptive histories of EL and DVDP magmas Based on our experimental results and the observations presented here, we propose two differentiation pathways from a single basanitic parent magma for the Ross Island lavas: a high-temperature pathway, followed by EL lavas, and a low-temperature pathway, followed by DVDP lavas. The parental magma beneath Erebus is a high-temperature (>1050 C) basanite that accumulated at high pressure, perhaps in the upper mantle (Kyle, 1981) or lower crust (Oppenheimer et al., 2011). Both EL and DVDP basanites must have ascended rapidly (with little to no differentiation during ascent; Kyle, 1981), sourced directly from a high-temperature region near the crust-mantle boundary, as they both contain olivine phenocrysts. Rapid magma ascent is supported by the observation of mantle xenoliths within some basanites (see, e.g. Moore & Gullikson, 2013). The sustained high-temperature regime inferred for the EL (>1050 C) could be facilitated by the direct connection of the EL conduit to the hot parental magma storage region beneath Erebus. The sustained open-vent nature of Erebus volcano implies a continual resupply of heat and gas from a deep source region (Oppenheimer & Kyle, 2008;Oppenheimer et al., 2009Oppenheimer et al., , 2011Iacovino et al., 2013a;Moussallam et al., 2013). The difference in high-and low-temperature regimes, coupled with the fact that EL lavas have erupted continuously whereas DVDP lavas have stopped erupting, suggests that the DVDP plumbing system was a shorter-lived offshoot of the main Erebus conduit with little resupply of heat from the source region. This is consistent with the model proposed for the formation of the subsidiary Ross Island volcanic centers in which magma followed radial fractures created by the updoming of mantle material into the crust directly beneath Erebus (Kyle & Cole, 1974;Kyle et al., 1992). More extensive dating of Hut Point Peninsula rocks would be necessary to precisely constrain the timing and longevity of the activity associated with DVDP lavas.
The DVDP magmas travelled along an offshoot of the main Erebus conduit, culminating at the surface on Hut Point Peninsula. During their rapid ascent, some olivine was removed from the DVDP basanites and injected into subsequently produced intermediate magmas as xenocrysts. Temperature decreased to below $1050 C during differentiation of the intermediate and evolved magmas, evidenced by the lack of olivine phenocrysts and the presence of kaersutite in all subsequent lava types. This drop in temperature may be facilitated by the lack of continuous heat input from the source region coupled with long travel times through the shallow crust compared with the EL lavas.

CONCLUSIONS
Experiments were performed to constrain and compare the conditions of deep and intermediate magma storage and differentiation of EL and DVDP lavas. Our results indicate relatively dry conditions (<1-2 wt % H 2 O) for both lineages, established by the plagioclase liquidus. The appearance of kaersutitic amphibole in natural intermediate and evolved DVDP lavas indicates an elevated water activity for the DVDP plumbing system compared with the EL, which does not contain kaersutite. The co-precipitation of kaersutite and plagioclase in natural DVDP lavas places very tight constraints on the H 2 O content of the intermediate DVDP magmas, as phonotephrite experiments produce this assemblage only between XH 2 O fluid c. 0Á15 and 0Á35. Previous studies on kaersutite-bearing rocks have shown that kaersutite can be stabilized at such low water activities by the replacement of OHin the crystal structure by Fand Cl -, and via oxy-substitutions (see above). This, combined with the approximate reproduction of the natural phase assemblages in our experiments at NNO þ 1Á5 to NNO þ 2Á0, indicates that similarly oxidizing conditions may be present during deep and intermediate magma storage. We can also constrain temperature conditions for both lineages based on the presence or lack of olivine and kaersutite, for which our experiments indicate a peritectic-like relationship at intermediate melt water contents and $1050 C. Olivine is present in all EL lavas and in DVDP basanites, but is absent in intermediate and evolved DVDP rocks. The opposite is true of kaersutite, which appears only in intermediate and evolved DVDP rocks. Because kaersutite is stable to very low melt water contents (with its stability field increasing with evolution of the melt), we infer that the lack of kaersutite in EL lavas and DVDP basanites is the result of sustained high temperatures during deep and intermediate magmatic differentiation (>1050 C) and that kaersutite is stabilized in DVDP magmas owing to a drop in equilibration temperature from above 1050 C in basanites to below 1050 C in the intermediate magmas.
The differences in T, XH 2 O fluid , and possibly fO 2 between DVDP and EL magmas can be explained by the EL plumbing system being dominated by the flux of a CO 2 -rich fluid (as hypothesized in previous studies) and the lack of such a fluid in the DVDP. This is consistent with the model of the formation of Erebus and its subsidiary volcanic centers.