Near-Primary Mantle Melts and Their Implications for the Mechanism of Island Arc Basalt Oxidation

Near-primary melt compositions (i.e., in equilibrium with >Fo88 olivine) are rare in arc systems. Yet, such melts provide essential views of mantle-derived melts, without further modification by fractional crystallization or other crustal processes, and reveal the diversity of melt compositions that exist in the arc mantle wedge. This study presents new measurements of naturally glassy, near-primary olivine-hosted melt inclusions from one dredge of Evita seamount (SS07/2008 NLD-02) in the southern Vanuatu arc system. Two distinct basalt types were identified in hand sample upon collection, based on contrasting phenocryst assemblage (Type 1: 1% phenocrysts; Type 2: 15% phenocrysts). We selected melt inclusions from each type and determined major elements and sulfur by EMP, H2O and CO2 by FTIR, trace elements by LA-ICP-MS, and Fe/∑Fe ratios by XANES. Melt inclusions from both lava types show equilibrium with ≥Fo88 olivine, consistent with host olivine compositions, and thus are near-primary melt compositions. Both have high Mg# (>65), and are basalt to basaltic andesite (49-55 wt% SiO2). Samples from Type 1 show relatively flat REE patterns, classic high Ba/Th ratios, and positive anomalies in Pb and Sr. In contrast, samples from Type 2 exhibit steeply sloped REE patterns with strong depletions in the HREE that suggest garnet in the source lithology for these magmas. Moreover, the Type 2 samples have low Ba/Th ratios and high La/Yb (29.5-43) and Sr/Y (50-58), which are classically attributed to partial melting of the basaltic slab. The slab surface temperature (SST) was calculated from H2O/Ce data; Type 1 SST shows temperatures comparable to global arcs (~767°C), while Type 2 SST is the hottest yet constrained by this method (~1041°C). Volatile analysis reveals that both lava types have had some degassing of H2O with CO2, and give minimum H2O contents of each magma: ~3 wt.% for Type 1, ~2.5 wt.% for Type 2. XANES analysis shows that Type 1 samples have Fe/∑Fe ratios similar to global arc basalts (~0.23), while Type 2 samples have Fe/∑Fe ratios that are among the highest measured in natural terrestrial glasses (~0.34), and have much higher concentrations of S. Mixing calculations suggest that Type 2 is not a simple mixture of the Type 1 basalt with an end-member slab melt. Alternate explanations include the possibility that Type 1 and Type 2 are instead the results of a mantle melt component mixing with either slab fluid or slab melt (but not both), or that they are distinct melts from different parts of the wedge that have migrated into the same volcanic system. A global correlation between H2O and Fe/∑Fe ratio suggests an oxidized, H2O-rich component is common to most arcs. The Type 1 magma conforms to this global trend, but Type 2 does not. Despite its highly oxidized condition and high sulfur content, Type 2 is too dry to be the end-member component that appears to be delivering oxidation to most global arc magmas.


INTRODUCTION
It is important to understand how deep-Earth processes influence the availability of oxygen in solid Earth systems, and vice versa. We know that different materials from the subducted slab give arc magmas distinct geochemical signatures. We also know that the oxidation conditions of arc magmas have been broadly linked to slab signatures (e.g., Kelley and Cottrell, 2009;Brounce et al., 2015). However, we still lack detailed links between specific types of slab-derived materials (e.g., slab fluids, sediment melts, and slab melts) and the redox conditions of arc magmas. We need these links for a complete picture of how subduction modifies the mantle wedge on both local and global scales, and also of the consequences of these deep processes for the construction of island arc crust.
The majority of island arc basalts are presently considered partial melts of mantlewedge peridotite due to fluxing from water brought down by the subducting slab (Kushiro et al., 1968;Stolper & Newman, 1994;Kelley et al., 2006Kelley et al., , 2010Langmuir et al., 2006). However, some studies support the idea that slab dehydration is not the predominant mechanism for some island arc magma formation (e.g., Kelemen, 2003;Kay, 1978;Yogodzinski andKelemen, 1998, 2007). Melts of the lithospheric slab (i.e. the basaltic layer) may be more prevalent than previously recognized, and may in fact be an ubiquitous component of arc magmatism (e.g., Moyen, 2009;Sajona 1993). This is particularly evident in some locations where the slab surface is heated enough by the surrounding mantle, such as locations where a torn plate edge is suspended in the mantle (e.g., Defant and Drummond, 1990;Yogodzinski et al., 2001;Plank and Cooper, 2009).
Studies also have shown that oceanic crust ages and its minerals oxidize over time as it moves from spreading centers and is recycled into subduction zones (e.g., Alt et al., 1986). Fluids or melts from the subducting slab contribute a chemical signature to the mantle source for island arc magmas. Resulting island arc lavas have a higher proportion of oxidized iron (Fe 3+ ) relative to reduced iron (Fe 2+ ), expressed as Fe 3+ /ΣFe ratio, than MORBs or OIBs (Carmichael, 1991), and appears to be a result of a more oxidized magma source in the mantle wedge (e.g. Brounce et al., 2015;Kelley and Cottrell, 2009).
The composition of arc basalts can be heavily influenced by fluids and sediment or slab melts that infiltrate the mantle wedge in subduction zones (e.g. Plank and Langmuir, 1993) and slab fluids may be capable of driving mantle wedge oxidation (Ballhaus, 1993;Kelley and Cottrell, 2009;Kelley and Cottrell, 2012;Brounce et al., 2015). Magmatic Fe 3+ /∑Fe ratios increase toward subduction zones and correlate linearly with H 2 O content (e.g., Fig. 1; Kelley & Cottrell, 2009) and element tracers of slab-derived fluids (Brounce, et al., 2014). These observations indicate a direct link between geochemical signatures of subduction and oxidation of the mantle wedge.
Using this direct link, we may be able to see how different subduction signatures influence mantle wedge oxidation, and investigate how the mantle becomes oxidized. This thesis presents new measurements of major, volatile, and trace elements, in addition to Fe 3+ /∑Fe ratios, in naturally glassy melt inclusions hosted by high-Fo olivine from the Evita seamount in the southern Vanuatu arc system. This study allows for a closer look at the igneous geochemistry of the Evita seamount in order to (1) assess the relative contributions of aqueous fluids, sediment melts, and slab melts to magmas in the southern Vanuatu island arc system, (2) determine the primary oxygen fugacity (fO 2 ) of the sub-arc mantle, and (3) evaluate how oxidation and slab signatures may be linked.

Geologic Setting of Vanuatu
An ideal place to address the location of and processes related to slab melting and oxidation is the Vanuatu island arc system. It is an intriguing place to study for several reasons. Vanuatu is an intra-oceanic arc, which minimizes the likelihood of magma-crust interactions. The island arc lavas are dominantly basaltic, which makes inferring mantle source easier (Peate et al., 1997). It also has a complex tectonic history that may have allowed different mantle sources and subduction components to be sampled at different locations and times within the arc. The Vitiaz lineament marks a fossil trench where the Pacific Plate once subducted beneath the Indo-Australian Plate to form the Vitiaz arc.
Vanuatu was part of this arc before the North Fiji back-arc basin began spreading about 12 Ma (Auzende et al., 1995). At present, a section of the Indo-Australian plate has been subducting eastward at 70° under the Pacific Plate for 7-4Ma (Mitchell and Warden;1971;Peate et al., 1997), with active volcanism beginning about 6 Ma.
In the southern part of the Vanuatu Arc, an adjacent section of the Indo-Australian plate continues to move horizontally at the surface to the northeast, forming a transform fault ( Figure 2a) and the Hunter Fracture Zone (HFZ). This leaves a subducting slab-edge hanging in the mantle, which may set the scene for high temperature on the slab surface and slab melting in a subduction zone (Cooper et al., 2009;Kincaid et al., 2004). Just northeast of the HFZ, at the Hunter Ridge, near-primary arc tholeiites have been reported (Sigurdsson, 1993), along with a wide variety of lithologies including rhyolites, MORBs, and peridotites that reflect the complex tectonic setting of the region.
evidence for sediment involvement that varies systematically, decreasing from north to south (Peate 1997 (Peate, 1997). In the southern part of this arc, subducted oceanic crust may find favorable conditions to melt and enter the mantle wedge as well, due to the torn slab edge in this region.

Preliminary Study and Sampling of Southern Vanuatu Seamounts
In 2008 assess the relationships between these two distinct signatures, the processes they represent, and the conditions of their source regions (e.g., temperature, redox, volatile content) at the same location. Doing so at a single volcano removes key variables (e.g., crustal thickness and structure, composition of mantle source) because both magmas presumably have similar ascent paths from source to surface.
At island arc subduction zones, magmas that come to the surface through volcanism may be subject to crustal differentiation processes. Fractional crystallization and degassing can alter the original magma composition, making it difficult to infer the characteristics of the mantle-derived magma at depth (e.g., P-T conditions of the mantle origin, major elements, original volatile content, magmatic redox; e.g. Kelley and Cottrell, 2012;Marsh and Carmichael, 1974;Métrich, 2009;Marsh, 1976;Kelemen, 2003, Kushiro, 1968Kushiro, 1972;Stolper and Newman, 1994). We thus lose the ability to link deep processes to these key factors that reflect the characteristics of the magma source.
The best possible way to interrogate primary arc magmas is to sample them directly. Melt inclusions trapped within early-forming phenocrysts may preserve both pre-eruptive volatiles and near-primary melts that have had minimal crystallization relative to their mafic host lavas. Thus, they provide a uniquely pristine view of the mantle. Such melts may reveal the diversity of melt compositions that exist in the arc mantle wedge.
Olivine-hosted melt inclusions can give us a rare glimpse into mantle conditions beneath island arc subduction zones. These glass inclusions, usually 10-200 µm in size, form when basaltic magmas crystallize and solidify, becoming trapped in a crystal. If trapped early enough in a magma's crystallization history, high-forsterite olivine (>Fo 88 ) may preserve near-primary melts, which are effectively unmodified from their origin in the mantle. Olivine crystals are the first to form on the hydrous basalt liquidus (Bowen, 1922) and thus can trap the most primitive of mantle melts, and preserve dissolved magmatic volatiles such as dissolved water, carbon dioxide, and sulfur.

Sample Preparation
Four representative pillow basalt samples from Evita seamount, collected at a depth of 1323m on the 2008 cruise SS07/2008, and stored at the Marine Geological Samples Laboratory (MGSL) at the Graduate School of Oceanography at URI, were chosen for melt inclusion analysis (two rocks each from NLD 02-01 and NLD 02-02).
Hand samples were cut and made into thin sections for detailed petrographic analysis. Melt inclusions were prepared as double-polished wafers, exposing the melt inclusion on both sides, to create a clear optical path through the glass with no interference from the host olivine, which is required for some analytical methods described below. Melt inclusions were examined under a petrographic microscope in both plane-polarized light and through crossed polars to identify crystal-free regions and to ensure clean glass was available for analysis ( Figure 3). Each melt inclusion was checked to be sure it had paths of clear glass, contained no more than one vapor bubble, was fully enclosed by the host olivine crystal, showed no cracks that could have led to postentrapment degassing, and contained no visible secondary crystals.
Some samples consisted of clear glass from both a melt inclusion and external, matrix glass still adhered to the olivine crystal. Where possible, this extermal glass was analyzed, and labelled in figures as either "Type 1 xlg" or "Type 2 xlg," marked with a circle instead of the diamond shape used for the melt inclusions.

EMP Analysis
Melt inclusions and host olivines were analyzed for major elements and S concentrations by electron microprobe using the JEOL-8900 5 spectrometer microprobe at the Smithsonian Institution, operating at 10nA, 15kV and with a 10 micron beam diameter (Table 1; Table 2a). Na and K were analyzed with 20 second peak count times to minimize alkali loss in hydrous glasses, followed by Si, Ti, Al, Fe, Mn, Mg, Ca, and P with 30-40 second peak count times (Luhr, 2001). The glasses were analyzed in a second round for S at 80nA and 15kV also using a 10 micron beam. The S concentrations were referenced to the VG-2 standard with 1340 ppm sulfur. Adjacent olivine was analyzed with a point beam, and primary and secondary standards were those used by Luhr (2001).

LA-ICP-MS Analysis
Melt inclusions, host olivines, and some exterior matrix glass were also analyzed for trace element abundances by laser-ablation inductively-coupled plasma mass spectrometry (LA-ICP-MS) at the Graduate School of Oceanography, University of Rhode Island. Analyses were conducted using a Thermo X-Series II quadrupole ICP-MS coupled with a New Wave UP 213 Nd-YAG laser ablation system, using spot sizes ranging from 20-60 µm and 5 Hz repeat rate to maximize ablation time in thin, wafered samples. Beam energy measured at the sample surface for a reference spot of 60 µm and 10 Hz was 0.35-0.41 mJ. Typical ablation duration in the melt inclusions, host olivines, and matrix glass lasted from 20-60 seconds, depending on sample thickness. An example LA-ICP-MS ablation spectrum is shown in Figure 4. Glasses and olivines were analyzed for 34 trace elements (Table 4); for olivines, we report only 24 minor and trace elements because these were consistently above the detection limit (Table 2b).
Procedures for reducing LA-ICP-MS data follow those outlined by Kelley et al. (2003), using 43 Ca as the internal standard for glasses and 26 Mg as the internal standard for olivine. Calibration curves were generated from eight natural-composition glasses from the USGS (BIR-1G, BCR-2G, BHVO-2G) and the Max Planck Institute (KL2-G, et al., 2006). A crystal of San Carlos olivine (Fo 88 ) was also analyzed as an in-house reference.

FTIR Analysis
The H 2 O and CO 2 contents of the wafered glass samples were determined by Fourier Transform Infrared Spectroscopy (FTIR) at the Graduate School of Oceanography. Infrared spectroscopy has been used for decades to gather transmission spectra of volatiles in glasses (e.g. Stolper, 1982;Dixon, 1995). FTIR spectroscopy has many advantages for measuring volatiles, including excellent analytical sensitivity for H 2 O (~10 ppm), fine spatial resolution (≥11 µm), and the ability to determine CO 2 and H 2 O species (CO 2 , CO 3 2-, H 2 O, OH -) in sample glasses (Devine, 1995).
We followed methods of Kelley & Cottrell (2012) for reduction of FTIR spectra.
Spectra were collected from 400-6000 cm -1 using a Thermo Nicolet iS50 bench FTIR spectrometer coupled with a Continuum microscope equipped with a liquid nitrogencooled MCT-A detector, KBr beam splitter, and a tungsten-halogen source. The bench, microscope, and samples were continuously purged with dry air. Thicknesses of the samples were directly measured with a digital piezometric micrometer and ranged between 20-150µm (Table 5). Clear, glassy regions of the wafered glass samples were selected optically and the size of the aperture customized to maximize coverage of melt inclusion samples.
Background spectra were collected as a reference before each new spectrum of a sample was collected. Transmission FTIR spectra were read as absorption spectra, and were collected on three different spots in each sample, in addition to a spectral map of the entire inclusion. Dissolved CO 3 2was quantified using the antisymmetric stretching absorptions at 1515 and 1435 cm -1 ; the background for fitting the carbonate peaks was done using an empirical least-squares fitting routine developed by Sally Newman. Total peak heights above background for H species were determined by fitting the spectral background with a spline function using OMNIC software. OHwas quantified using the absorption at 4500 cm -1 , and molecular water at the absorption bands of 1630 and 5200 cm -1 . Peak heights were analyzed at absorption bands at 3530 cm -1 for total H 2 O at low H 2 O concentrations, whereas 1630, 5200, and 4500 cm -1 absorption bands were used for glasses with higher H 2 O content ( Figure 5; Table 5).

Petrography
In thin section, Type 1 lava shows olivine as the only phenocryst phase ( suggesting they are genetically related. Vesicularity of Type 2 lavas is ~50%, and total phenocryst abundance is ~5-15%.

Assessment of Post-Entrapment Crystallization of Olivine
Post-Entrapment corrections are sometimes necessary in order to add back any host crystal that has been lost from the melt inclusion. As melt inclusions cool with the host olivine crystal, it is possible to precipitate a bit of the host mineral from the melt composition onto the inclusion walls, which modifies the glass composition relative to when the inclusion was first trapped. In order to reconstruct the melt inclusion composition at the moment of trapping, we must assess and apply post-entrapment crystallization (PEC) corrections.

equilibrium between melt inclusions and their olivine hosts was assessed by separating
FeO* in each melt composition into Fe 2 O 3 and FeO using the measured Fe 3+ /∑Fe ratio, and then calculating the olivine in equilibrium with each melt composition. The calculated equilibrium olivine was compared with measurements of the host olivine composition for each sample, and any deviations from equilibrium were attributed to post-entrapment processes (Danyushevsky, 2000). (Roeder and Emslie, 1970) to calculate the olivine composition in equilibrium with each melt inclusion, and compare this with the measured composition of its olivine host. If a melt inclusion is in equilibrium with an olivine of lower Fo than its host, then post-entrapment olivine crystallization has occurred. In this case, we add equilibrium olivine back to the melt composition in 0.01% increments until the Fo value of the melt indicates equilibrium with the host. If a melt inclusion appears to be in equilibrium with olivine of higher Fo than its host, then post-entrapment Fe 2+ diffusion out of the inclusion may have occurred. To correct this, the FeO concentration of the inclusion is increased, and concentrations of the other major elements are proportionally diluted, until the melt indicates equilibrium with its host (Figure 8). Using these corrections, we can account for any post-entrapment processes.
Both Type 1 and Type 2 samples contained olivine hosts that were Fo 87 -Fo 92 .
PEC corrections were necessary in 11 out of 12 melt inclusions to equilibrate with the olivine host, and ranged from 2% to 15% olivine added (Table 3). The corrected PEC values for major elements are used for all subsequent analysis and discussion.

Post-Entrapment Crystallization of Other Phases
Although melt inclusions selected for preparation had no visible daughter crystals, in three cases, daughter crystals were evident only after preparation and analysis.
Daughter crystals were found in 3 of the melt inclusions from the Type 2 lava only ( Figure 9). Scanning electron microscope [SEM] analysis revealed these crystals to be cpx in all cases. Only the Type 2 samples had daughter crystals. There was no evidence of systematic modification of the major element compositions of these inclusions relative to those with no daughter crystals (Figure10), and so we choose to show the data. It is important to note that in all subsequent figures, these inclusions with daughter crystals are given different symbols -red stars -and are labeled as "dxtal" for daughter-crystal, where they have been plotted.

Crystallization of the Evita Magmas
The MgO concentration is a commonly used proxy for extent of crystallization, because mafic minerals crystallizing out of the melt, beginning with olivine (Bowen, 1922), will take MgO out of the melt. Thus, a higher MgO content indicates less crystallization and more primitive magma compositions. Type 1 samples have generally higher MgO concentrations than Type 2 samples (Type 1: 6.6-13.0 wt.% MgO; Type 2: 6.4-7.8 wt.% MgO; Figure 11), and on this basis, one might easily infer that the Type 2 magma has experienced more fractional crystallization than Type 1. Yet, our analyses of both magmas include melt inclusions from >Fo 88 olivine (Figure 8) given MgO content for the suite of magmas studied in the region (Sigurdsson, 1993 decreases, which is also consistent with the petrographic data.
The highest Mg# samples from both magma types show some other differences.
From Figure 10, Type 1 has similar Fe, higher Na and Al, and lower Ca than Type 2.
Additionally, the data show that Type 2 has higher K 2 O than Type 1. Arcs have higher K 2 O than MORB due to its incompatible, fluid-mobile nature. Altered oceanic crust has more potassium than MORB to begin with before subduction (Kelley, 2003). Type 1 and Type 2 have similar TiO 2 , which is lower in arcs relative to MORB due to high extents of melting and more depleted mantle sources, a signal seen especially in arcs with back-arc spreading (Kelley, 2010).

Trace Element Signatures of the Evita Magmas
While major elements can give us a view into the primitive nature and crystallization of the Evita magmas, minor and trace elements provide specific views of the magma source in the mantle wedge. They can reveal incompatible elements, fluidmobile elements, and melt-mobile elements that have implications for the composition of magma sources. Some studies (e.g., Kay, 1978;Yogodzinski andKelemen, 1998, 2007) have found enriched, primitive magmas that are heavily influenced by slab melt trace element signatures, i.e., steeply-sloped rare earth element (REE) patterns, depleted in heavy relative to light rare earth elements, with high Sr/Y and La/Yb ratios, and low Y and Yb content. These signatures indicate garnet in the source of the melt, which is hosted in the thermally altered lithospheric slab (Moyen, 2009;Sajona, 1993).
Type 1 samples show relatively flat to slightly LREE enriched REE patterns that are typical of island arc basalts (e.g., the Marianas, Fig. 11). In contrast, Type 2 samples show strong enrichments in the LREE and steeply sloped REE patterns, suggestive of an incompatible-element enriched source and the presence of a residual phase capable of fractionating the REE (e.g., garnet). The Spider Diagram reveals differing anomalies between the magma types ( Figure 12). Type 1 samples have a slight positive anomaly in Sr, which is typical of island arc basalts and suggestive of the addition of an aqueous fluid to the mantle source; however, Type 2 samples show negative Pb anomalies and slightly negative Sr anomalies, suggestive of a source that has lost fluid-mobile elements such as Pb and Sr. Both sample types show negative Nb-Ta and Ti anomalies, indicative of a source material with residual rutile or another phase that partitions the HFSE. Type 2, however, has much larger Nb-Ta and Ti anomalies than does Type 1, and Type 2 also displays a negative Hf-Zr anomaly that is absent in Type 1. A negative Hf-Zr anomaly may signify a source material that also contains residual zircon. Type 1 samples also have higher Ba/Th ratios than do Type 2, while Type 2 have higher La/Yb (29.5-43) and Sr/Y (50-58) ratios than Type 1.

Volatiles and Degassing of the Evita Magmas
We We also look at sulfur because it commonly degasses with H 2 O (Kelley & Cottrell, 2012). S and H 2 O are positively correlated in both magma types, suggesting that they degassed together (Sisson and Layne, 1993;Wade et al., 2006;Kelley and Cottrell, 2012). The highest measured concentrations of H 2 O, CO 2 , and S are therefore minima for the primary melts. Type 2 samples also have much higher S concentrations than Type 1 samples ( Figure 13b). Like iron, sulfur can exist in different valence states, and its solubility is sensitive to redox conditions, increasing with increasing oxidation. Sulfur is more soluble in oxidized magmas because S 6+ has higher solubility than S 2- (Carroll & Rutherford, 1988). Thus, S content gives us some initial indication of the oxidation of each magma type, and we can infer that Type 2 magma may be more oxidized because it has higher S.

Oxidation of the Evita Magmas
The Fe 3+ /∑Fe ratio (which is a proxy for fO 2 ) is useful to assess in these inclusion samples because it gives us a view into the oxidation conditions of their respective source magmas ( Figure 14;  Figure 15a). Type 2 basalts have high average Fe 3+ /∑Fe ratios at 0.297 (ΔQFM+1.5), with a maximum of 0.328 (ΔQFM +1.9). The uncorrected PEC ratio for Type 2 is the highest Fe 3+ /∑Fe ratio yet measured in a natural terrestrial basalt. Type 2 basalts are far more oxidized than Type 1 basalts of similar Mg# (Figure 15b), except for the samples at the highest Mg#, which are similar in terms of oxygen fugacity relative to the QFM buffer.

Redox Changes During Magmatic Differentiation
The highest Mg# samples from each magma type are actually fairly similar in oxidation, perhaps suggesting that Type 2 might have become more oxidized by some differentiation process. However, the highest Mg# sample from Type 2 is a daughtercrystallized inclusion; this casts doubt on the fidelity of the Fe 3+ /∑Fe ratio in this melt inclusion, which may have changed during post-entrapment crystallization and cooling.
We thus cannot confidently use this data and propose that its composition is possibly not indicative of the unmodified source. Moreover, for all other melt inclusions, we show that at a given Mg#, Type 2 inclusions are uniformly more oxidized than Type 1 inclusions ( Figure 15) which suggests a fundamental difference in the oxidation conditions of their respective sources. Thus, we conclude that the similar fO 2 is not real, and that there are no apparent significant changes in magmatic redox that go along with indices of crystallization and differentiation (e.g., MgO or Mg#; Figure 15).
There is some apparent relationship in the Type 1 magma that may relate its oxidation to S degassing. Type 1 magmas show that the lowest S samples have the most reduced Fe 3+ /∑Fe ratios (Figure 16), which may relate to the loss of S 2from the melt as SO 2 gas (Kelley & Cottrell, 2012). Although Type 2 magmas have far higher S abundance and range in S concentration than Type 1, S does not correlate with a change in Fe 3+ /∑Fe ratios in this magma. This is likely because the magma is too oxidized, and thus all of the sulfur is S 6+ , which is more soluble in oxidized magmas. The degassingdriven reduction modeled by Kelley & Cottrell (2012) is only possible if S 2is preferentially degassed as SO 2 , leaving electrons behind in the melt that may then reduce Fe. We thus conclude that sulfur degassing is not responsible for the oxidation conditions of these magmas.

Conditions of the Mantle Source
If the varying oxidation of these magmas is not changed by differentiation processes, then the differences must come from the mantle source. Both mantle sources have more oxidized conditions than MORB, or average upper mantle conditions. The magma source for Type 1 is consistent with island arc basalts documented in Kelley & Cottrell (2012). Estimates for primary melts in that study were on the order of QFM+1, whereas the Type 1 averaged QFM+0.8. However, Type 2 magmas are not consistent with this previous study, and have the highest oxidation values yet measured for a basalt.
The Type 2 samples are far more oxidized than the Type 1 samples, and contain the highest yet recorded Fe 3+ /∑Fe values for terrestrial glasses (Figure 14). Type 2 averages QFM+1.5 (maximum QFM+1.9), which is highly oxidized for a primary magma; one value from Kelley & Cottrell (2012) of QFM+1.8 was a differentiated melt.

Slab Melt Signatures and Mixing
The mantle sources of each magma type have very different oxidation states; what could be causing this? Trace elements tell us that Type 1 has a classic island arc basalt signature of a mantle melt modified by aqueous fluid addition from the dehydrating slab. where they may have been retained by rutile or zircon in a melting slab residue. We can expect to see this in a sediment or slab melt signature, but not in a classic mantle melt.
The negative Pb and Sr anomalies also suggest these elements have been removed from the source by previous dehydration. The large Sr/Y ratios suggest deep slab melting; as pressures increase, plagioclase becomes unstable and releases Sr, whereas garnet becomes stable and traps Y, increasing the Sr/Y ratio dramatically with increasing pressure (Moyen, 2009). The La/Yb ratio has a similar trend, which is also high in Type 2.
This slab melt signature can also be identified when we model the slab surface temperature using H 2 O/Ce (  (Figure 17). This is cooler than the proposed threshold for basaltic slab melting at ~900°C, but is close to the average for global SSTs (~805°C) (Cooper et al., 2012), and is above the H 2 O-saturated sediment solidus, permitting the application of this model for these samples. Type 1 magmas thus likely reflect contributions from slab dehydration and sediment melt components. Type 2, however, has a calculated slab surface temperature of 1041±50°C. This is the hottest slab yet constrained by this method (Cooper et al., 2012), and is well above the threshold for melting of the basaltic slab. Importantly, this new temperature is within the model uncertainty of the temperature determined from the uncorrected H 2 O/Ce ratio. We implement the same procedure for the Type 2 magma, but using an EMORB mantle source (as indicated by the trace element patterns; Fig. 12), but because Type 2 has a Nb/Ce ratio of 0.02, we assume that essentially all of the Nb and Ce in this lava originate from the slab component, and Type 2 thus does not require a correction.
This contrast in slab surface temperatures recorded by the two lavas at Evita presents the problem of how the slab surface can be at two different conditions beneath this volcano, and what it may imply for melt and fluid transport in the wedge. At a slab edge, as is present in the location of seamount Evita, relatively hot mantle material may be flowing around the slab edge and heating up the surface of the subducting slab (Kincaid, 2004;Figure 19a). It is possible that flow regimes in the Vanuatu subduction zone are creating two different melting paths near the slab surface beneath the seamount Evita, and thus creating the two different magma types we see in this study. Type 2 could possibly be a mixture of Type 1, which would be the mantle-melt end-member, with a slab-melt end-member ( Figure 19b).
We attempted to model how much of a slab melt end-member component may be mixing with the Type 1 magma by using an average major element composition of andesitic slab melts (e.g., Klimm, et al., 2008), and the highest Mg# samples from Type 1 and Type 2. Many of the major elements in Table 8 are explained by ~25-50% mixing of an andesitic slab melt with the Type 1 basalt. The slab melt end-member composition is relatively unknown, so we do not necessarily expect a perfect match. SiO 2 is the most robust major element to look at because its composition is much less sensitive to the phase assemblage of the melting slab. On its own, SiO 2 conservatively suggests a mixture of 23.5% andesitic slab melt end member with 76.5% Type 1 basalt. If we assume these proportions, then the Fe 3+ /∑Fe ratio of the slab melt would be 0.456.
However, we found that the results varied widely ( explain what may be happening to produce the differences in these magma types. One possibility is that Type 1 and Type 2 magmas may not be related strictly to each other (e.g., Type 1 is the wrong thing to mix with to make Type 2), which would mean they need to be separated within the mantle wedge flow regime present at this slab edge. Perhaps Type 1 is the mixing result of a mantle melt with slab fluid added, and Type 2 is a mixture of mantle melt with slab melt added; this would leave the mantle melt, without any additions, as an end-member with which to mix (Figure 19c). We lack sufficient data, however, to reconstruct the mantle melt composition, minus any slab additions, to assess this hypothesis, and note that the differences in HFSE abundances between the two types suggests that the mantle sources are different ( Figure 12). It may also be possible, due to complex toroidal flow around the slab edge, that Type 1 and Type 2 magmas are melts of different mantle from different parts of the wedge that are fortuitously delivered to the same ascent path beneath Evita. The magmas could, for example, originate from similar depths, but at different horizontal locations along the slab ( Figure 19d). The slab melt signature would originate closer to the hotter slab edge, while the slab fluid signature would be from farther north, away from the slab edge flow.

Links Between Slab Signatures and Magmatic Oxidation
The final question we address in this study is whether slab melting is truly ubiquitous (e.g. Kelemen, 2003) and responsible for the oxidation seen at global arcs.
Type 1 falls on the trend line of previous global data, including MORBS ( Figure 20). to be the end-member to which most arc magmas seem to mix in H 2 O-Fe 3+ /∑Fe ( Figure  19). There must be some other component that delivers both high H 2 O and highly oxidized signatures to arc magma sources in the mantle wedge.
Type 2 has much higher sulfur concentrations than Type 1, so we can consider the volatile sulfur for this mechanism by comparing it to the Fe 3+ /∑Fe ratios for both types ( Figure 16; Kelley & Cottrell, 2012). However, the high sulfur in the oxidized sample may simply be reflecting the S solubility contrast (which, as discussed above, is a strong function of fO 2 ), and thus we cannot resolve whether slab-derived S is the responsible mechanism for mantle wedge oxidation. We may also speculate about other potential oxidants for the Type 2 magma. If this magma is indeed a mix of slab and mantle melts, it is possible that Fe 3+ itself has been added from the slab to the wedge, creating the highly oxidized signature. For this to be the case, however, we would expect the more oxidized melts to have higher primary FeO* than the more reduced melts. As seen in Figure 10, Type 1 and Type 2 samples have comparable FeO* values, so this is probably not the case. A mixing model also shows that the Type 2 magma is not a simple mixture of a slab melt and a classic, fluid-rich arc basalt (as exemplified by the Type 1 magma).

CONCLUSIONS
Complex mantle flow, arising from the setting of this region near a suspended plate edge in the mantle, may have allowed for the unusual diversity of magmas, derived from radically different slab conditions, at this volcano.                     (Kelley, 2009). Data for H 2 O are published FTIR data from the literature or are FTIR or ion microprobe data from that study. The solid line is a least-squares linear regression through all of the data, with equation y = 0.026x + 0.14 [correlation coe cient (r 2 ) = 0.72].

Normalized Intensity
Energy (eV) Figure 6: XANES pre-edge spectra for standard glass LW_0.035, a Type 1 glass (NLD 02-01-02_S4), and a Type 2 glass (NLD-02-02-05_S3). The data was collected at Brookhaven National Laboratory, using methods outlined by Cottrell et al. (2009). Pre-edge spectra were corrected for energy drift and normalized to a value of 7112.3 eV for the LW_0 reference glass. Determination of Fe 3+ /∑Fe ratios for basaltic glass is precise within ±0.005.  (Cottrell & Kelley, 2009) and Marianas arc data (Brounce et al., 2015).  Figure 11: Chondrite-normalized REEs; normalized to chondrite value by Nakamura (1974). Type 1 and Type 2 samples are compared to a typical arc signal from Agrigan samples in the Marianas (Brounce, 2015). Type 1 samples show trace element signatures comparable to Marianas, whileType 2 do not follow this trend, exhibiting REE patterns that suggest garnet in the source and slab melting.

Marianas
Type 2 dxtal Figure 12: Spider Diagram of Type 1 and Type 2 samples, compared with Agrigan samples from the Marianas (Brounce, 2015). Normalizing primitive mantle composition taken from Sun & McDonough (1989), elements ordered after Hofmann (1988).  Figure 13: (a) Degassing trends of Type 1 and Type 2 samples, modeled with VolatileCalc1.1 (Newman & Lowenstern, 2002 Figure 14: Histogram showing the distribution of pre-PEC corrected Fe 3+ /∑Fe ratio results. Type 1 basalts (blue) have much lower Fe 3+ /∑Fe ratios overall than Type 2 basalts (red). MORBs are reported from Cottrell and Kelley (2011). Marianas are reported from Brounce (2015).   Figure 15: (a) Mg# v. Fe 3+ /∑Fe. Types 1 and 2 are compared to each other and to global MORB (Kelley & Cottrell, 2012) and Marianas (Brounce, 2015). (b) Mg# v. ΔQFM. There is a clear trend in the MORBS of higher QFM with di erentiatiation. This trend is absent in Type 1 and Type 2 from Vanuatu.   Figure 19: Mantle wedge mixing models. (a) At Evita's slab edge, relatively hot mantle material may be owing around the slab edge and heating up the surface of the subducting slab (Kincaid, 2004). (b) Type 2 could possibly be a mixture of Type 1, a mantle melt end member, and a slab melt end member, depicted here. Other possibilities include that (c) a mantle meltis mixed with slab uid to form Type 1 and slab melt to form Type 2, and that (d) complex toroidal ow around the slab edge generates Type 1 and Type 2 magmas from similar depths but apart from each other horizontally on the slab surface, creating two di erent SSTs.