A Geochemical Investigation of Oxygen Fugacity in the Marianas Subduction Factory

Oxygen fugacity is a fundamental thermodynamic property that describes reduction-oxidation (redox) equilibria in the solid Earth. It controls material transfer from the interior to the exterior of the planet by dictating the speciation of multi-valent elements (e.g., Fe, V, S). Oceanic crust ages and oxidizes as it moves from spreading centers to subduction zones, where it returns to the mantle in modified form. Subducting slabs release H2O-rich fluids and SiO2-rich melts to the mantle wedge in subduction zones, contributing significantly to the isotopic and major, trace, and volatile element composition of arc and back-arc magmas, however the effect that the oxidized nature of subducting slabs on arc basalts remains unclear. Arc basalts have a higher proportion of oxidized (Fe) relative to reduced (Fe) iron, expressed as the Fe/ΣFe ratio, than do mid-ocean ridge basalts (MORB) but there is disagreement as to whether this arises due to shallow level differentiation processes (e.g., crystal fractionation, crustal assimilation, degassing) in the arc crust or to differences in the fO2 of the mantle source. This thesis addresses this problem by examining the oxidation state of Fe and other transition element proxies for fO2 in (1) altered oceanic crust prior to subduction, (2) modern eruptive products from the active Mariana arc and back-arc, and (3) eruptive products representative of subduction initiation and margin evolution in the Marianas. Melt inclusions and submarine glasses record variable magmatic compositions that have the potential to record changes in magma chemistry during crystal fractionation and volcanic degassing. Recent innovations in synchrotron technologies have made studies of Fe redox possible in situ, on small scales (>10 microns), allowing direct observation of changes in Fe redox during shallow level differentiation processes in arc and back-arc magmas. This study reports observations of Fe redox variation from several Mariana arc volcanic centers as well as from the Mariana trough, demonstrating that shallow level differentiation processes are not responsible for generating the oxidized nature of arc basalts. Constraints for mantle source fO2 show that the mantle wedge is more oxidized than MORB source mantle and link this oxidation to influence from recycling slab fluids. Additionally, we explore other transition row element proxies for fO2 and show that these proxies need not preclude oxidized mantle wedge conditions in the Marianas. We examine changes in Fe redox in samples that record subduction initiation and margin evolution, demonstrating that in zones of melt generation within the mantle wedge, oxidation occurs immediately upon subduction initiation and that the mantle wedge remains oxidized for the majority of a subduction zone’s lifetime. We further constrain the fluxes of Fe in to and out of the Mariana convergent margin, demonstrating that the Pacific plate is very oxidized prior to subduction and that 50-70% of this oxidized signature survives the recycling process to be subducted into the deep upper mantle.


INTRODUCTION
Oxygen fugacity (fO 2 ) is a fundamental thermodynamic property that governs reduction-oxidation (redox) equilibria in solid Earth systems. It controls material transfer from the interior to the exterior of the Earth by setting the speciation of multivalent elements (e.g., Fe, S, V, C), which in turn controls their crystal/melt partitioning behaviors (e.g., , their physical state and mobility in the mantle (e.g., , and their solubility in silicate melts (e.g., . Despite its power in dictating chemical exchange in the Earth however, the fO 2 of the upper mantle and whether it varies through space and geologic time is widely debated (e.g., , Trail et al., 2011.
Oceanic crust ages and oxidizes as it moves from spreading centers to subduction zones, where it is recycled into the mantle, and material from the downgoing slab contributes chemically to the mantle source of arc magmas (e.g., , Plank & Langmuir, 1993. Arc basalts have a higher proportion of oxidized (Fe 3+ ) relative to reduced (Fe 2+ ) iron, expressed as the Fe 3+ /ΣFe ratio (i.e., Fe 3+ /[Fe 2+ +Fe 3+ ]), than do MORB . There is disagreement as to whether this arises due to differentiation processes (e.g., crystal fractionation, crustal assimilation, degassing) in the arc crust or to differences in the fO 2 of the mantle source. Experimentally calibrated trace element proxies for mantle fO 2 , which are potentially more immune to differentiation processes in the arc crust, suggest that the fO 2 of arc mantle is similar to the MORB primary magmas . Magmatic oxidation may perhaps be influenced by later stage crustal processes, such as the extensive fractionation of Fe 2+ -bearing minerals (e.g., olivine) or by the assimilation of oxidized crustal material, although such relationships have not yet been observed or quantitatively modeled.
Yet, a global study of basaltic glasses shows that those magmas most heavily influenced by subduction have higher Fe 3+ /ΣFe ratios than MORB ). Moreover, olivine-hosted melt inclusions from a single eruptive event from Agrigan volcano in the Marianas show that the least differentiated melts have the highest Fe 3+ /ΣFe ratios, and the Fe 3+ /ΣFe ratios of reconstructed primary melts correspond to a source mantle that is oxidized 1 -1.6 orders of magnitude over the MORB source . In addition, a paired study of whole rock Fe 3+ /ΣFe ratios determined by wet chemical methods and fO 2 calculated from magnetite-ilmenite mineral pairs demonstrates that andesites from the Mexican volcanic belt experienced no net change in bulk Fe 3+ /ΣFe ratios despite significant changes in volatile content and extent of crystal fractionation . These observations suggest that low-pressure crystallization and degassing do not significantly oxidize arc magmas and instead indicate that high Fe 3+ /ΣFe ratios recorded by arc magmas reflect a mantle source that has higher fO 2 than MORB source mantle.
Outside of mid-ocean ridge settings, Fe redox studies that specifically address the effects of differentiation on Fe speciation have thus far been limited. For example, elevated magmatic water contents, derived from the subducting plate, may suppress plagioclase saturation and decrease the temperature difference between the appearance of silicates and magnetite on the liquidus , potentially influencing whether a basaltic magma follows a calc-alkaline (Fe-depleted) or tholeiitic (Fe-enriched) differentiation path (e.g., ). Yet, magmatic H 2 O and Fe 3+ /ΣFe ratios are strongly correlated , and high magmatic fO 2 also enhances the appearance of oxides relative to silicates on the basalt liquidus (Botcharnikov et al., 2008. The effects of fO 2 and H 2 O on magmatic differentiation may thus be difficult to segregate. Magnetite fractionation in a system closed to oxygen is also expected to reduce magmatic Fe 3+ /ΣFe ratios, but this phenomenon has not been observed directly in the natural rock record. If source mantle fO 2 at convergent margins is elevated over MORB, the cause of this oxidation and the extent to which it varies are central to developing models for the structure and growth of arc crust, and of the oxygen evolution of Earth through time. Does primary fO 2 change as subduction influence varies or diminishes? What effect do variable extents of fluid or sediment melt infiltration have on primary fO 2 ? To answer these questions, we examine the relationships between crystal fractionation, degassing, mantle source composition, subduction influence, and magmatic or mantle fO 2 along the entire Mariana subduction zone. With this work, we investigate a variety of crystal fractionation and degassing processes recorded by arc and back-arc basaltic glasses and examine the relationships between these processes and magmatic Fe redox. We present new major, trace, and volatile element concentrations as well as Fe 3+ /ΣFe ratios in olivine-hosted melt inclusions from single eruptive events at five sub-aerial volcanic centers along the Mariana arc (Sarigan, Guguan, Alamagan, Pagan, and Agrigan), in addition to submarine glasses from NW Rota-1 and Pagan volcanoes  and the Mariana trough back arc spreading center (Fig. 1). After assessing the effects of differentiation on magmatic redox, we use major element trends defined by the data to reconstruct primary melt compositions and mantle source fO 2 conditions. We then pair these with key trace element ratios (Ba/La, Th/La, and Zr/Y) to assess the extent to which different slab derived materials may influence the fO 2 of the mantle wedge.

GEOLOGIC SETTING
The Mariana subduction system is a well-studied ocean-ocean convergent margin with an active sub-aerial and submarine arc made up of ~40 volcanic centers and the Mariana trough, an actively extending back-arc basin ( Fig. 1; Bloomer et al., 1989. The arc is split into three distinct segments, the Northern Seamount Province, the Central Island Province, and the Southern Seamount Province.
The Central Island and Southern Seamount Provinces are both built on oceanic lithosphere previously rifted by the opening of the Mariana trough and the Parece-Vela basin . The composition of erupted products along these arc volcanic centers are well studied and are primarily basaltic in composition (Bloomer et al., 1989, Martindale et al., 2013, Meijer & Reagan, 1981. The northern to central Mariana trough, here termed collectively the northern Mariana trough, is opening asymmetrically in an east-west direction and generally mimics the arcuate shape of the volcanic front . The volcanic arc follows the strike of the Mariana trench north of ~13ºN. South of this latitude, the trench curves sharply to an east-west orientation. In this area, both arc and back-arc volcanism approach the trench and the subducting Pacific plate is shallower beneath this magmatically active area . Taken together, the oceanic upper plate, mafic magmatism, and the presence of a mature back-arc spreading center make the Mariana arc an ideal setting for studying the competing effects of source fO 2 and shallow crustal processes on the Fe 3+ /ΣFe ratios of arc and back-arc basalts.

Mariana arc tephra samples
Olivine hosted melt inclusions were targeted for this study for several reasons.  Figure 1). Each tephra sample was washed in de-ionized water and sieved, taking care to avoid any samples with clasts larger than two centimeters to ensure that all material had a short cooling history upon eruption (e.g., Lloyd et al., 2012). Olivine crystals were either handpicked from sieved size fractions or separated using lithium poly-tungstate heavy liquid separation, using modified techniques from Luhr (2001). Large (0.5 -1 mm), euhedral olivines or olivine fragments were immersed in mineral oil to identify glass inclusions, which were selected for analysis if they were >50 µm in diameter, completely glassy, without daughter or co-entrapped minerals, fully contained by the host olivine, and contained no more than one vapor bubble. Representative photomicrographs are shown in Figure 2. Photomicrographs of every inclusion are shown in electronic appendix K.

Submarine glass samples
Glassy pillow lavas from the southernmost Mariana trough (Malaguana-Gadao ridge) were dredged from the seafloor between 12.5° -13.2°N, during expedition TN273 of the R.V Thomas G. Thompson in 2011. Glassy pillow lavas from submarine volcanic exposures at Pagan and NW Rota-1 volcanoes were provided by Yoshi Tamura .
Glass chips were chiseled and hand picked from the freshest pillow lavas in each dredge and washed in de-ionized water prior to preparation for analysis. We also incorporate previously published data for submarine glass samples from the northern Mariana trough (18.1° -20.9°N;Fig. 1).

Electron Microprobe Analysis
Submarine glass chips and glass inclusions were exposed on a single side and polished for electron microprobe analyzer (EMPA) analysis on a JEOL-8900 5 spectrometer microprobe at the Smithsonian Institution. During major element analysis, the beam was operated at 10nA, an accelerating voltage of 15 kV and 10 µm beam diameter. Sodium and potassium were measured first with 20 second peak count times to minimize alkali loss. Subsequently, Si, Ti, Al, Fe*, Mn, Ca and P were measured with 30-40 second peak count times. All data were subject to ZAF correction procedures. Primary calibration standards include VG-2 glass, Kakanui hornblende, anorthite, microcline, ilmenite, and apatite . The VG-2 and VG-A99 glasses were monitored as secondary standards during each run . Sulfur and chlorine were measured separately using a beam operated at 80 nA, an accelerating voltage of 15 kV and 10 µm beam diameter.
The major element compositions of the olivine hosts were measured adjacent to the glass inclusions as well as at the rims of the olivines to eliminate zoned hosts that reflect potentially complex magmatic histories. A focused electron beam was operated at 10 nA and an accelerating voltage of 15 kV. San Carlos olivine and fayalite were used as primary calibration standards, San Carlos olivine and Springwater olivine were used as secondary standards during each run . Significant olivine zoning was not observed for any samples in this study and the olivine compositions reported in electronic appendix E are average values of all three to six analysis spots on each olivine.

FTIR Analysis
After EMPA analysis of melt inclusions, all sample pits were polished away, being careful to account for possible electronic damage within the activation volume of each EMPA spot. Melt inclusions were then polished from the opposite side until doubly exposed, and submarine glasses were wafered to a nominal thickness of 80 µm (though some were as thin as 20 µm) to create wafers with analyzable pools of optically clear glass. All wafered samples were washed gently with acetone to remove all epoxy residues. Dissolved H 2 O and CO 2 concentrations in glasses and glass inclusions were analyzed by Fourier-transform infrared (FTIR) spectroscopy at the Smithsonian Institution. All samples were analyzed using either a Bio-Rad MA-500 microscope attached to a Bio-Rad Excalibur FTS 3000 FTIR spectrometer or a Continuum microscope coupled with a Thermo-Nicolet 6700 FTIR spectrometer.
Spectra for all samples were collected between 1000-6000 cm -1 using a tungstenhalogen source, KBr beamsplitter and a liquid-nitrogen cooled MCT-A detector. The bench, microscope, and samples were continuously purged by air free of water and carbon dioxide using a Whatman purge-gas generator. Aperture dimensions were selected for each sample depending on the geometry of free glass pathways, ranging in size from 12 µm x 12 µm to as large as 60 µm x 60 µm. concentrations calculated from the 4500 + 1630/5200 cm -1 bands agree within error (<10% relative) with those calculated from the 3530 cm -1 band. Dissolved CO 3 2-concentrations were determined by using the 1515 and 1435 cm -1 absorption bands . Thicknesses of each sample were measured using a piezometric digimatic indicator (σ ± 1 µm). Glass densities and absorption coefficients relevant to each absorption band were calculated using methods from  and Luhr (2001).

XANES analysis
All samples were analyzed in situ for Fe 3+ /∑Fe ratios via micro X-ray absorption near edge structure spectroscopy (µ-XANES) following the methods and techniques of  at beamline X26A, National Synchrotron Light Source, Brookhaven National Laboratory. Spectra were collected in fluorescence mode from 7020 eV and 7220 eV using a Si [311] monochromator and a nominal beam size of 9x5 µm. A beryllium window over the detector was used to attenuate high count rates above the main Fe Kα fluorescence peak. Reference glass LW-0 was monitored continuously during each experimental session to correct for instrument drift. Further details related to this correction can be found in .
Spectra were scrutinized for any influence from host olivines, phenocrysts, or micro phenocrysts in the glass chips and inclusions. If crystal interference was found, these spectra were eliminated from further study. Examples of the influence of crystal interference on Fe-XANES spectra are provided in Electronic Appendix A (Fig. A1).
Determination of Fe 3+ /∑Fe ratios in basaltic glasses following the methods of  have an associated precision of ±0.005.

LA-ICP-MS analysis
Abundances of 33 trace elements (Sc, V, Cr, Co, Ni, Cu, Zn, Rb, Sr, Y, Zr, Nb, Cs, Ba, La, Ce, Pr, Nd, Sm, Eu, Gd, Tb, Dy, Ho, Er, Tm, Yb, Lu, Hf, Ta, Pb, Th, U) were determined in submarine glasses and glass inclusions by laser-ablation inductively-coupled plasma mass spectrometry (LA-ICP-MS) at the Graduate School of Oceanography, University of Rhode Island on a Thermo X-Series II quadrupole ICP-MS coupled with a New Wave UP 213 Nd-YAG laser ablation system following techniques outlined by  and Lytle et al. (2012), normalizing to 43 Ca as the internal standard. The laser energy was 0.20-0.30 mJ at the sample surface for a reference spot (60 µm, 10 Hz) on NIST 612 glass and the repeat rate was decreased to 5 Hz in melt inclusions and thin glass wafers, to achieve a slow drilling rate of ~1 µm/s through thin samples. Spot sizes ranged from 20-80 µm. United States Geological Survey glass standards BCR-2g, BHVO-2g, BIR-1g, and Max Planck Institute glass standards GOR-132-G, StHls-G, T1-G, ML3B-G and KL2-G were used to create linear calibration curves (R 2 >0.990) for each analytical session (Jochum et al., 2006. Melt inclusions were analyzed in single spot analyses.
Counting statistics were examined carefully for each element and those elements that did not return strong signals for the entire length of the laser ablation period were discarded. Submarine glasses were analyzed in triplicate and concentrations were reproducible to within 4% rsd for all elements.

RESULTS
In total, 113 olivine-hosted glass inclusions were prepared for analysis. Thirtyfour of these inclusions were lost during various stages of preparation or did not return glassy XANES spectra, suggesting that they had an unfavorable geometry for XANES analysis or were otherwise devitrified. The remaining 79 inclusions were subject to data filtering procedures outlined below.

Inclusion/olivine equilibrium and post-entrapment crystallization
Melt inclusions are trapped in olivine phenocrysts at high temperatures. As inclusion and olivine cool during magmatic ascent and eruption, olivine may precipitate along the wall of the inclusion during post-entrapment crystallization (PEC). To screen for the effects of PEC, the predicted equilibrium olivine composition was calculated for each melt inclusion using Fe 2+ /Mg K ! !"#/!"# = 0.3 , and compared to the measured host forsterite contents (Fo; Mg/[Fe 2+ + Mg]) of the olivine host of each inclusion. If the predicted equilibrium Fo inclusion matched the measured Fo host , equilibrium between inclusion and host was assumed and no action was taken. If the inclusion composition has been modified by PEC, the predicted equilibrium Fo inclusion should be lower than the measured Fo host (e.g., . In these cases, calculated equilibrium olivine was added to the inclusion in 0.1% increments until the inclusion and host reached equilibrium. If the predicted equilibrium Fo inclusion was higher than the Fo host (indicative of Fe loss, e.g., Danyushevsky et al., 2000) no action was taken. Those inclusions that required >2% PEC correction or whose Fo host -Fo inclusion disagreed by more than 2% were excluded from further consideration or modeling, although their compositions are reported in electronic appendix B. Furthermore, for each melt inclusion suite, the inclusion compositions were compared to published whole-rock and melt inclusion data for each volcano, with particular attention to variation in FeO* vs MgO (Electronic Appendix A; Danyushevsky et al., 2000). Inclusion compositions that fell outside of the data field defined by the published whole-rock data were excluded. This rigorous data filtering is meant to avoid melt inclusion compositions with complicated magmatic histories that may cloud the discussion of magmatic redox variations and source fO 2 .
The remaining discussion considers only the 48 melt inclusions that satisfy the requirements outlined here.

Compositions of Mariana arc and back-arc melt inclusions and glasses
In order to constrain the effects of fractional crystallization and volcanic degassing on Fe 3+ /∑Fe ratios in natural basaltic magmas, we first identify the mineral and volatile phases that have fractionated, degassed, or diffused to create the variations in major element and volatile concentrations observed in melt inclusions and submarine glasses from the Mariana arc and trough. Fractional crystallization models that aimed to match the observed major element variations were generated using Petrolog3 (Fig. 3; for all sample suites in this study. A single crystallization model was chosen for each geographic location and is compared with published natural sample compositions and data from this study in electronic appendix A. Each model follows the general pattern of olivine ± clinopyroxene ± plagioclase ± magnetite fractionation. Individual model parameters, including mineral-melt models and pressure conditions that were chosen to generate each liquid line of descent (LLD) can be found in electronic appendix A. Volatile element behavior was investigated for each LLD by examining the behavior of volatile species (e.g., H 2 O, CO 2 , S) relative to other volatiles (e.g., CO 2 vs H 2 O). For all samples, we also compare major and volatile element variations to previously published melt inclusion data from the same island (electronic appendix A, Fig. 4).
Here, we outline the magmatic processes that are captured by these samples and the magnitude and variations of their Fe 3+ /∑Fe ratios.  Figure 4a for an ascending, degassing magma . There is also no correlation between S and  (3.87 -7.40 wt%).

Mariana volcanic arc
Alamagan inclusions have a range in CO 2 and H 2 O concentrations that are consistent with CO 2 degassing (Fig. 4a, . They exhibit a range of sulfur concentrations from 664 -1544 ppm, which vary with H 2 O and are broadly consistent with sulfur degassing (Fig. 4b; e.g.,   , , . They are basaltic in composition and range in MgO from 3.87 -7.72 wt%.
Dissolved H 2 O and CO 2 concentrations are consistent with the pressure of the water column at an eruption depth of approximately 4 km.  and  have shown that these glasses are the result of magmas that were variably saturated with a H 2 O-CO 2 rich fluid upon eruption. The range in H 2 O contents of these samples results from the variable influence of the subducting slab in the genesis of northern Mariana trough magmas .
Fe 3+ /∑Fe ratios for these glasses are reported by  and examined here in the context of the

Fractional crystallization
If Fe 3+ behaves simply as an incompatible element and magmatic fO 2 is not buffered, low pressure fractionation of olivine, clinopyroxene, and plagioclase from basaltic magmas should lead to an increase in the Fe 3+ /∑Fe ratio of the magma. Data for MORBs are consistent with this behavior, showing slight oxidation during fractional crystallization ). In such a system that is closed to oxygen exchange with its surroundings, Fe 2+ partitions into olivine and clinopyroxene, while Fe 3+ remains in the melt. As fractionation proceeds, the total volume of liquid For example, EQ 1 has been invoked to explain the stabilization of solid sulfide phases in oxidized, magnetite saturated magmas with relatively low S contents from the Pual Ridge in the Manus Basin . The saturation of magnetite from a basaltic melt will remove a greater proportion of Fe 3+ than Fe 2+ from a melt, provided the magnetite has a higher Fe 3+ /∑Fe ratio than the melt. If the concentrations of both Fe 2+ and Fe 3+ are controlled only by crystal fractionation, when magnetite begins fractionating, EQ 1 will shift to the right to maintain the equilibrium constant, K eq . This will reduce S 6+ , producing S 2and potentially promoting the saturation of a solid sulfide phase at relatively low dissolved sulfur concentrations. Importantly, this shift in equilibrium to the right of EQ 1 will also produce a new equilibrium proportion of  , Liu et al., 2007. The maximum sulfur content observed in these samples is only 553 ppm, suggesting that the Mariana trough magmas are not sulfide saturated. Instead, sulfur may have partitioned into a vapor phase, a process that has been shown to reduce magmas if the vapor phase is SO 2 ( Fig. 4b; , although H 2 S degassing could also oxidize magmas under the right conditions (Métrich et al., 2009). It is difficult to assess the independent importance of sulfur degassing on the Fe 3+ /∑Fe ratios of these samples because of the evidence for simultaneous magnetite fractionation. The relative constancy of Fe 3+ /∑Fe ratios in Mariana trough samples with MgO < 5.5 wt% suggests that the redox equilibria illustrated by EQ 1 may have played an important role in controlling Fe 3+ /∑Fe ratios in these samples, particularly if both magnetite fractionation and S degassing reduce magmas. Further work in quantifying the relative reduction potentials of Fe and S in basaltic magmas will aid in understanding the evolution of Fe 3+ /∑Fe ratios during magmatic processes.

The effect of fO 2 on tholeiitic index
Magmatic differentiation may influence the Fe 3+ /∑Fe ratio, as discussed above, but magmatic fO 2 also plays an important role in determining the differentiation path of basaltic magmas, which may follow variably tholeiitic or calc-alkaline trends, depending on the extent of FeO* enrichment or depletion that occurs in the early stages of crystallization , Miyashiro, 1974. Predominant models for generating these trends involve the interplay of plagioclase and magnetite fractionation and the petrological factors that control these phases. The generation of calc-alkaline magmas (i.e., FeO* depleted) in subduction settings may be related to the high pre-eruptive water contents of arc magmas, which suppress plagioclase, but not magnetite, crystallization (Botcharnikov et al., 2008 at MgO = 8.0 ± 1.0 wt% (FeO* 4.0 /FeO* 8.0 ), such that a tholeiitic magma has a THI>1.0, and a calc-alkaline magma has a THI<1.0 . Samples from both the Mariana arc and trough from this study are consistent with observations from , where Mariana arc volcanoes display slightly calc-alkaline trends and also have higher pre-eruptive water contents (>1.5 wt%) than the Mariana trough, which falls distinctly in the tholeiitic field (THI>1.2) and has lower preeruptive water contents (Fig. 7). Globally, the THI is also well correlated with magmatic fO 2 , where arc samples are more oxidized than back-arc and MORB (Fig.   7). This suggests that the oxidized nature of arc basalts may also play an important role in the generation of calc-alkaline differentiation trends by promoting magnetite saturation over silicates   4). This demonstrates that H 2 O degassing is not an oxidizing process in basaltic magmas, rather it is redox neutral   . Using H 2 O as a proxy for slab-derived influence on the arc and backarc mantle sources, subduction influence has also been linked to magmatic and mantle source oxidation (e.g., Fig 4d; Kelley   Fig. 6) as a possible explanation for magmatic reduction and associated S loss. Figure 6 shows the relationship between sulfur concentrations and Fe 3+ /∑Fe ratios, as observed in melt inclusion suites from Sarigan, Alamagan, Agrigan, and Guguan volcanoes. Sulfur degassing is likely recorded by Alamagan, Agrigan, and Guguan melt inclusions. Agrigan inclusions show a slight reduction in Fe 3+ /∑Fe ratios with decreasing sulfur concentration that is consistent with the melt inclusion suite studied by   . In the case of cooling, the pressure of the inclusion decreases and hydrogen fugacity in the inclusion may be lower than in the surrounding melt. This would result in hydrogen diffusion into the inclusion and potentially, the reduction of iron. Alternatively, in the case of degassing a host magma, hydrogen fugacity in the inclusion may be higher than in the surrounding melt, causing hydrogen to diffuse out of the melt inclusion and potentially oxidizing iron contained in the inclusion . Experimental observations by   Fig. A10), supporting the notion that the heterogeneity recorded in melt inclusion suites in this study reflect true magmatic heterogeneity rather than post-entrapment melt inclusion processes.
Why are arc basalts more oxidized than MORB?

Primary melt compositions and fO 2
Arc basalts are shown here to be more oxidized than MORB, even at comparable MgO concentrations. Additionally, we show that fractional crystallization and degassing processes are capable of both oxidizing and reducing Fe 3+ /∑Fe ratios in arc basalts, although these effects are minor, and neither can explain the oxidation of Fe in arc basalts over MORB. This suggests a fundamental difference between the Fe 3+ /∑Fe ratios of mantle-derived primary arc magmas and primary MORB magmas.
To test this, primary melt compositions (i.e., in equilibrium with mantle olivine at Fo 90 ) were reconstructed using methods modified after Klein and Langmuir (1987), where modeled or data-defined LLDs provide constraints to account for the effects of fractional crystallization on magmatic composition. The fractionation models generated for each geographic location were used to calculate the primary magma composition for each sample with MgO > 5 wt%. Each composition was projected back to MgO = 7.0 wt% using the slope of the fractionation model between 5 and 7 wt% MgO for all major elements (excluding Fe). In some cases, the fractionation model was poorly fit to the most incompatible major elements (e.g., P 2 O 5 ) and the slope of a line defined by the natural data was used for the calculation instead (Electronic Appendix E). For Fe, FeO (actual) and Fe 2 O 3(actual) concentrations were plotted versus MgO concentration and projected back along the slopes of lines defined by the natural data for each sample suite. The point of MgO = 7.0 wt% was chosen because data for the arc basaltic glasses above 7 wt% MgO are sparse, and selecting a higher MgO limit would be arbitrary. At or above 7 wt% MgO, both the data-defined and modeled LLDs suggest that olivine or olivine + clinopyroxene are the only phases on the liquidus, due mostly to the suppression of plagioclase saturation in water rich magmas (Gaetani et al., 1993. Once at MgO = 7 wt%, each composition was subject to addition of equilibrium composition olivine in 0.1% increments until in equilibrium with Fo 90 olivine (Electronic Appendices A, E-G). Alternatively, clinopyroxene could be included along with olivine as a liquidus phase to higher MgO concentrations, although it is difficult to know when olivine becomes the only liquidus phase. Because of this uncertainty, we also used PetroLog to add clinopyroxene and olivine simultaneously back until achieving equilibrium with Fo 90 olivine. When applied to a suite of 20 melt inclusions from Sarigan volcano, this method returned nearly identical average primary melt Fe 3+ /∑Fe ratios as the olivine only method (0.217 using PetroLog, 0.220 using the olivine only method) and because of differences in major element composition, a primary melt fO 2 ~0.28 log units below that of the method described above. This is within the ± 0.5 log unit uncertainty of the Fe 3+ /Fe 2+ proxy for fO 2 , so we chose to use the olivineonly addition method described above because it can be applied consistently to all samples discussed below (MORB, BABB, and arc samples). An important conclusion drawn from this test is that the choice of correction method does not significantly impact the reconstructed primary oxygen fugacity. Temperatures and pressures of last equilibration with peridotite for each calculated primary melt composition were determined using the melt thermobarometer of Tables 1, 2). Primary fO 2 was calculated using the algorithm of Tables 1, 2) relative to the QFM buffer calculated at pressure and temperature according to .
Mid-ocean ridge primary magmas have fO 2 similar to that of the QFM buffer (Fig. 8; There are several trace element proxies (e.g., V-based, Cu and Zn/Fe* ratios) for modeling mantle source fO 2 from the compositions of erupted basalts that can provide important additional constraints on the fO 2 of the mantle wedge in the Marianas. The application of these models requires knowledge of the composition and mineral mode of the mantle source, the mechanisms for melt generation, and constraints on LLDs on a volcano-to-volcano basis . These parameters are not likely to be uniform from mid-ocean ridge settings to subduction zones, and all are challenging to constrain. A comparison between Fe redox and trace element proxies for mantle source fO 2 from this dataset is currently in progress.

The source of elevated fO 2 in the mantle wedge
Tracking sediment melt and slab fluid influences; Trace element and isotopic compositions of subduction related lavas are influenced by contributions from the downgoing slab that may include sediment melts, aqueous fluids, and slab melts. Key trace element ratios (e.g., Th/La, Ba/La) have been shown to record the contributions from these sources in lavas erupted within the arc and back-arc system of a convergent margin (e.g., , Plank, 2005, Plank & Langmuir, 1993. Thorium is enriched in subducted terrigenous sediment and will become mobile when the sediments cross their solidi and begin to melt. Sediment melts move into the mantle wedge and contribute to the production of arc and back-arc magmas that have elevated Th/La ratios relative to MORBs. Barium, on the other hand, is mobilized preferentially over melt-mobile La via aqueous fluids that escape the subducting slab as it descends into the mantle, generating aqueous slab-derived fluids that are expected to have elevated Ba/La ratios (Johnson & Plank, 1999, Kessel et al., 2005. In the Marianas in particular, the Ba/La ratio of the bulk subducting sediment package is low (Ba/La ~ 15, Plank & Langmuir, 1998), and thus sediment melts that move from the slab into the mantle wedge likely also have low Ba/La ratios. The high Ba/La ratios of Mariana arc lavas (commonly >20) require the presence of aqueous fluids to transport Ba preferentially over La into the mantle wedge. The Marianas is a special case among global subduction settings in this respect, because trends between key trace element ratios require that separate sediment melts and aqueous fluids contribute to the composition of arc lavas , Plank, 2005, Plank & Langmuir, 1998 Fig. 9a; Plank et al., 2007). Back arc samples deviate from the MORB array in Figure 9b towards a fluid with high Ba/La ratio, though lower Ba/La ratio than the fluid influencing the arc samples . Mariana arc samples display a large range in Ba/La ratios that reflect significant slab fluid influence that varies in magnitude along the Mariana margin. NW Rota-1 is the least influenced by slab fluids, with Ba/La ratios similar to southern Mariana trough samples. This is an interesting observation considering its position relative to the trench, where it sits approximately 50 kilometers farther from the trench than the main subaerial arc, where the slab depth to slab is 50-100 km deeper . NW Rota-1 may thus receive a different style of slab fluid (or less of it) than the subaerial arc volcanic centers.
Sarigan and Alamagan melt inclusions span a large portion of the entire range of Ba/La ratios observed for the arc, suggesting that there is trace element heterogeneity in parental magma compositions at Sarigan and Alamagan volcanoes.
Mantle wedge composition; In addition to containing geochemical signatures of the subducting slab, melts generated in the mantle wedge at the back-arc spreading center and under the volcanic arc may reflect variations in mantle source composition that is inherent to the mantle  or generated by prior melt extraction (e.g., Kincaid, 2003. Here, we use the Zr/Y ratio to characterize mantle source composition. Zirconium and yttrium are not significantly fractionated during low-pressure crystal fractionation and, are expected to be relatively absent from slab-derived materials in the Marianas (Pearce & Parkinson, 1993. The Zr/Y ratios of arc and back-arc lavas are thus assumed to reflect the mantle source composition, independent of slabderived additions to the mantle wedge. The Zr/Y ratio of the mantle source is, however, fractionated as the result of prior melting episodes because Zr is more incompatible than Y during mantle melting. Therefore, melts of fertile mantle will have high Zr/Y ratios, but progressive melting of the same parcel of mantle will generate subsequent melts with lower Zr/Y ratios. Mantle entering the wedge in the Marianas passes through the back-arc melting triangle and experiences melt extraction there before moving under the volcanic arc, such that the mantle under the arc is more depleted in trace elements due to melt extraction at the back-arc . The Zr/Y ratios of Mariana trough and arc magmas reflect this contrast, with high Zr/Y ratios of Mariana trough basalts reflecting a relatively enriched MORB-type mantle ( Fig. 9c; Table 1, 2;Langmuir et al., 2006). The Zr/Y ratios of Mariana arc basalts are significantly lower, however, reflecting the more depleted arc mantle source, consistent with the predicted effects of back-arc spreading in mantle circulation and prior geochemical studies of the Mariana arc . The Zr/Y ratios in the Marianas do not follow the mixing relationships between the estimated source composition and the sediment materials (orange and green lines, Fig. 9c), supportive of the notion that Zr and Y do not travel with slab-derived materials into the mantle source.
Despite observed heterogeneity in trace element compositions discussed here, we emphasize that major element compositions are relatively narrow and uncorrelated with trace element ratios like La/Yb, which is greater than can be expected from simple fractionation of La from Yb during differentiation (Fig. A13). This suggests that while trace elements record the presence and mixing of several different parent magmas, mixing does not control the major element relationships in these magmas and major elements can reasonably be described by simple crystal fractionation (electronic appendix A).

Variations in fO 2 with mantle source-related variables;
To assess the source of elevated fO 2 in the mantle wedge, we now examine relationships between tracers of mantle wedge composition (Zr/Y) and subduction influence (Th/La, Ba/La), and primary fO 2 . There are no systematic relationships between primary fO 2 and Zr/Y ratio within the arc or back-arc data (Fig. 10a), beyond a first-order contrast between the arc (higher fO 2 , lower Zr/Y) and the back-arc (lower fO 2 , high Zr/Y). Mariana trough Zr/Y ratios on average are higher than for mid-ocean ridge or arc samples, suggesting (a) that the Mariana trough taps an enriched mantle source , and (b) that sub-arc mantle is more depleted than back-arc mantle by virtue of previous melt extraction in the back-arc melting regime. It is possible that melt extraction varies the activity of Fe 3+ in mantle spinel phases. If it increases, then melting could contribute to the oxidized nature of arc basalts relative to back-arc primary magmas (i.e., . Arc primary melts are offset towards more oxidized fO 2 s than the Mariana trough, and also record a more depleted mantle source, however there is no relationship within either the Mariana trough or arc samples between extent of depletion and mantle source fO 2 . Moreover, MORBs encompass the full range in Zr/Y ratios observed at the Mariana arc and back-arc, with no coincident variation in fO 2 (Fig. 10a), suggesting that the observed oxidation in the Marianas is not solely a result of variable mantle source composition. We conclude that the mantle composition alone (as recorded by Zr/Y ratio) is not responsible for the elevated fO 2 of arc and back-arc basalts.
There is a weak relationship between primary fO 2 and Th/La ratio in the northern Mariana trough, but not among Mariana arc samples (Fig. 10b), and the Mariana arc and trough largely overlap in Th/La ratios. It is clear that the oxidized nature of the mantle source under the arc is unrelated to sediment melt influence, as indicated by Th/La ratio, because all of the arc samples have higher fO 2 than the northern Mariana trough samples within the same range of Th/La ratios (Fig. 10b).
These observations indicate that melts of the sediments at this arc either (a) are not significantly oxidized relative to MOR source mantle, and/or (b) make up too small a proportion of total primary melt to impact mantle fO 2 .  Taken together, these results suggest that slab-derived, Ba-rich fluids are significantly more oxidized than the upper mantle. There is strong evidence that the slab lithosphere is highly serpentinized prior to subduction , Van Avendonk et al., 2011 and deserpentinization reactions (dehydrating serpentinite assemblages) in the subducting slab may contribute a significant proportion of the fluids that lead to the formation of arc magmas , van Keken et al., 2011. Serpentinization reactions that occur on the seafloor prior to subduction involve the infiltration of fluids (e.g., seawater) into peridotite. Iron is oxidized, transforming Fe 2+ contained in olivine into Fe 3+ to form magnetite at the expense of oxygen contained in H 2 O molecules. This reaction generates magnetite, brucite, and serpentine coexisting with reduced aqueous fluids  that are ultimately lost from the system, resulting in a net oxidation of the rock. When the slab travels along a prograde P-T path during subduction, serpentinite minerals become unstable at ~600º C and release aqueous fluids (e.g., , Spandler et al., 2014, which happens at 150-180 km depth for the southern Mariana subducting slab geotherm (van Keken et al., 2011). The deserpentinization reactions over the P-T path of subducting slabs are complicated, but these may potentially consume magnetite and serpentine minerals to form olivine, which would reduce Fe and create oxidized fluids (e.g.,  or fluids carrying oxidized species (e.g., sulfate or SO 2 ; . These oxidized fluids must percolate through the subducting slab, where they may scavenge Ba from the altered oceanic crust and overlying sediment package, before ultimately entering the the mantle wedge where they lower the peridotite solidus and generate oxidized hydrous melts with high Ba/La ratios under the volcanic arc. Additionally, there is a distinct contrast with the fO 2 of primary melts under the back-arc, where subduction influence is lower than the arc and the fluid composition is fundamentally different. This may be because 1) the influence from oxidized slab fluids is less and thus the oxidizing power of the slab fluids is diminished beneath the back-arc, and/or 2) the fluids that are generated by the dehydrating slab that reach the back-arc have different sources from those fluids that contribute to arc volcanism and perhaps are not as oxidized. It is unlikely that fluids percolate through the entire volume of the mantle wedge. Rather, they may concentrate in rising diapirs or along interconnected networks (e.g., , Marchall & Schumacher, 2012. This limits the proportion of mantle wedge that interacts with slab fluids significantly, and may make it possible for slab fluids to create oxidized primary melts that relate linearly to the extent of slab fluid influence. Fingerprinting the source of fluids from within the slab is a major challenge in subduction zone studies, and requires further investigation to test explicitly. It is also possible that the mantle wedge contains important buffering assemblages that control the mantle source fO 2 for arc basalts, although the relationship between slab fluid influence (Ba/La) and primary fO 2 in Figure 10c indicates that there are no buffering species present in the mantle wedge in this range of fO 2 . Sulfur speciation is shown to be highly sensitive to changes in fO 2 between QFM and QFM+2.0 (Jugo, 2009. If a solid sulfide phase exists in the mantle wedge and is not exhausted during melting under the volcanic arc, the solid sulfide-sulfate phase boundary may serve to buffer the mantle wedge during melting, such that increasing the influence of oxidized slab fluids cannot increase primary fO 2 until either the sulfur phase is exhausted or another, more oxidized phase is added (e.g., Fe 2 O 3 ;       Figure 1. Black line in panels e -h show the trajectory of olv±plag±cpx±mgt fractionation, generated using Petrolog3  fractionating the composition of 80-1-3 (VG10498) at 1.5 kbar, using mineral melt models of , , and  in a system closed to oxygen. FeO* is total Fe expressed as FeO. Error bars are shown in the lower left hand corner of each plot, in panels (a) through (d).     and Kelley et al. (2010), shown for comparison. (a) Plot of H 2 O vs. CO 2 variation. Isobars and open system degassing curve were calculated for a basalt at 1200°C using VolatileCalc . Error associated with H 2 O concentrations are ~0.25 wt% and ~75 ppm for CO 2 concentrations.  Wade et al. (2006, small cross) and Sisson and Layne (1993, black square). Dashed line represents the extension of the degassing trajectory to high sulfur concentrations. Black star represents an approximately degassed magma. (c) Fe 3+ /ΣFe ratios vs. H 2 O concentrations for Mariana arc melt inclusions from this study as well as from Kelley and Cottrell (2012, light grey diamonds), and back-arc submarine glasses. Grey MORB field are data taken from . Additional melt inclusions in this panel are from    Kelley and Cottrell (2012, light grey diamonds). Grey field represents global MORB glass data from . Black line is the trajectory of olv±plag±cpx±mgt fractionation, as in Figure 2. Dashed purple line is the trajectory of olv±cpx±plag fractionation from a starting composition similar to the composition of melt inclusion Agri04-05, generated using Petrolog3 ) at 1 kb, using mineral-melt models of  and  in a system closed to oxygen. (b) Magmatic fO 2 , plotted relative to the QFM buffer vs. MgO for the same samples. The fO 2 s and position of the QFM buffer  are calculated at the pressures and temperatures of melt inclusion entrapment using the algorithm of . Pressures for each melt inclusion suite are taken as the average pressure of entrapment recorded by CO 2 -H 2 O contents of the inclusions in the suite . Temperatures for each melt inclusion suite are taken as the average olivine-liquid temperature of the inclusions in the suite . The dashed black line marks the position of QFM. Error bars are shown in the lower left hand corner of each plot.  . The grey line marks the boundary between tholeiitic and calc-alkaline differentiation trends, as defined by . Magmatic fO 2 s are calculated as in figure 5. Brown square represents the average of MORB data from . Grey triangle represents both northern and southern Mariana trough samples from this study. Error bars for H 2 O and magmatic fO 2 represent the standard deviation of the sample population from the average. Error bars for THI represent the minimum and maximum estimates of THI, based on the available data for each location.  . Pressures and temperatures of melt generation were calculated using the Si-thermobarometer of . Oxygen fugacities were calculated from Fe 3+ /ΣFe ratios after . Gray bars with dashes represent primary fO 2 s for Agrigan volcano, calculated by . The dashed line marks the position of QFM, which is equal to the approximate primary fO 2 for MORB source mantle   The grey field is Pacific MORB data from , shown for comparison. Data for discrete materials from the Pacific plate taken from ODP sites 800 and 801 are shown as small black and white stars, respectively, and their calculated bulk compositions are shown as large black and white stars (Plank & Langmuir, 1998). Orange and green dashed lines show approximate mixing lines between the suggested arc mantle source composition (black circle; Plank, 2005) and various discrete components of the sediments at ODP Sites 800 and 801. The solid black line is an approximate mixing line between the suggested back-arc mantle source composition (grey circle;    . In (c), large symbols represent the average calculated primary fO 2 and measured Ba/La ratio for each volcano. The error bars on these symbols represent the standard deviation of the population from the average value. The bold dashed line marks the position of QFM, which is equal to the approximate primary fO 2 for MORB source mantle . Individual primary melt compositions for each location are shown as small symbols. The short-dashed, thin black line is a standard linear regression through the average fO 2 and Ba/La ratio for the MORB, all of the northern Mariana trough data, and the average fO 2 and Ba/La ratio for the southern Mariana trough and each of the volcanic centers.  "nd" indicates that insufficient data were available to perform the calculation 1 Highest MgO concentration measured for a melt inclusion or submarine glass sample at a given geographic location. 2 Fe 3+ /ΣFe ratio of the sample with the highest MgO concentration. 3 Pressure (GPa) from PetroLog fractionation model. 4 Average magmatic temperature (°C) recorded by melt inclusion/olivine thermobarometry, calculed after Putirka et. al., 2007. 5 Average magmatic log(fO 2 ) relative to the quartz-fayalite magnetite oxygen buffer (ΔQFM) at the pressures and temperatures recorded by melt inclusions. Calculated using the algorithm of . 6 Pre-eruptive water content determined by averaging the H 2 O contents of inclusions that form vertical paths on a CO 2 -H 2 O diagram (Fig. 4b). For Guguan, this is the H 2 O concentration of the single inclusion for which we have data. The uncertainty in these values is +/-0.5-1.0 wt%. 7 calculated after Zimmer et. al., 2010. 8 Average Fe 3+ /ΣFe ratio calculated for a primary magma at a given geographic location. 9 Pressure of melt generation (GPa), calculated using the Si-thermobarometer of   10 Temperature of melt generation (°C), calculated using the Si-thermobarometer of   11 log(fO 2 ) relative to the quartz-fayalite-magnetite oxygen buffer (ΔQFM) at pressure and temperature of melt generation. Calculated using the algorithm of . 12 Estimated from melt inclusion data from other suites in this study.

Screening for olivine interference in Fe-µ-XANES spectra
When collecting Fe-µ-XANES spectra on olivine-hosted melt inclusions, it is important to avoid hitting the olivine crystal with the beam during analysis. Olivine contains several weight percent of Fe 2+ and even a very small amount of olivine interference will "contaminate" the pre-edge structure of Fe-µ-XANES spectra collected for melt inclusions and bias the result towards more reduced values. The region of XANES spectra at higher energies than the Fe-Kα absorption edge contains information related to Fe-coordination and can be used to distinguish glass structure (random and on average, uncoordinated) from olivine signal (strong coordination, Fig.   A1). All melt inclusion and seafloor glass spectra were visually inspected and compared to spectra taken on San Carlos olivine and standard glasses from  in order to screen for crystal contamination. Any spectra demonstrating signs of spectral features similar to those observed in San Carlos olivine were not considered in this study and additional spectra were collected to accommodate for this elimination.

Model liquid lines of descent
To constrain the effects of fractional crystallization on magmatic Fe 3+ /∑Fe ratios, model liquid lines of descent that match the observed major element variations were generated using PetroLog3 . The mineral-melt models that most closely replicate the natural data were chosen for each location, resulting in some variation in the models used from volcano to volcano. Individual model parameters chosen to generate each LLD are provided in the figure captions to If crystal fractionation is the main control on magmatic Fe redox, each model is meant to assess the extent to which Fe redox ratios may vary as a result of the relative incompatibility of Fe 3+ in each fractionating phase. Implicit in this assessment is the assumption that fractional crystallization proceeds in a system closed to oxygen.
Evidence from a global study of MORB glass suggests that closed system fractionation of olivine ± plagioclase ± clinopyroxene explains variations in Fe 3+ /∑Fe ratios ). In our present study, each modeled LLD assumes a D Fe3+ = 0 and a D Fe2+ that varies according to the mineral-melt models chosen, for olivine ± clinopyroxene ± plagioclase fractionation. Magnetite mineral-melt models use a non-zero D Fe3+ that depend upon the composition of the melt. Each model was generated using a starting composition from a measured sample in this study.

Agrigan
Agrigan tephra samples Agri7, Agri04, and Agri05 all contain crystals of olivine, plagioclase, clinopyroxene, and magnetite. The clasts are red-brown, rounded and slightly weathered. In tephra sample Agri04, there are occasional clay fragments present. In tephra sample Agri05, there are occasional clasts of country rock that are red in color. The largest size fraction observed in these tephra samples is ~0.5 cm in diameter.
The modeled LLD begins with sample Agri-04-05 as the parental composition.

Guguan
Tephra samples Gug11 and Gug23-02 are black in color and are fresh. The largest size fraction in both tephras is ~0.5 cm in diameter. Both tephra samples contain crystals of olivine, plagioclase, and clinopyroxene.
The modeled LLD begins with sample SD46-1-1 as the parental composition

Sarigan
Tephra sample Sari15-04 is brown in color and show signs of slight alteration.
The largest size fraction observed in this tephra sample is ~0.25 cm in diameter. This tephra sample contains crystals of olivine, clinopyroxene, plagioclase, and magnetite.

Diffusive re-equilibration of melt inclusions
To assess whether melt inclusion suites in this study are significantly changed by diffusive re-equilibration, we compare the trace element variability recorded in suites of melt inclusions from Sarigan, Agrigan, and Alamagan volcanoes to submarine glasses from the southern Mariana trough ( Fig. A10; i.e., Cottrell et al., 2002, Kent, 2008. There is a simple relationship between the variability in trace element compositions within a population of basaltic glasses and the bulk partition We also examine the relationship between the sizes of melt inclusion and their major element composition and Fe redox, to test whether smaller diameter melt inclusions have been re-equilibrated and larger diameter melt inclusions have not. There is no relationship between the size of melt inclusion and the major element composition, measured Fe 3+ /∑Fe ratios in the inclusions, or the amount of disequilibrium between melt inclusion and olivine host (Fig. A11). The exception to this is the apparent disequilibrium between four Agrigan melt inclusions and their olivine hosts, where the smallest diameter melt inclusions appear to have undergone more post-entrapment crystallization than larger inclusions. These melt inclusions require more than 2% post-entrapment crystallization correction to reach equilibrium with its olivine host and are not considered in this study. These results suggest either that diffusive re-equilibration (a) has not occurred in melt inclusions used in this study, or (b) has gone to completion such that even slow diffusing elements in olivine (e.g., Ca) have completely re-equilibrated in melt inclusions with 50-300 µm diameter.

Trace element heterogeneity and magma mixing
Melt inclusions and submarine glasses from a single arc volcano commonly record significant heterogeneity in trace element compositions that potentially reflect contributions and mixing between several parental magmas (e.g., Figure 9 in the main text, Fig. A13). If the parent magmas are also heterogeneous in the major element compositions there is some concern that the major element relationships reflect magma mixing rather than crystal fractionation. This potentially introduces error in the calculations described above, of the compositions and fO 2 of primary magmas in the Marianas. Figure  suggesting that this observation is not simply a melt inclusion phenomenon. We conclude that the major element relationships observed in Figure 3 in the main text, as well as in Figures A2-A7 in the electronic appendix can be reasonably described by simple crystal fractionation.

Sensitivity test of primary melt model results
The method used to calculate primary melt compositions and fO 2 has many sources of error, uncertainty in the inputs, and calculation constraints. Here, we present the results of sensitivity tests of the model results for a subset of samples.
The calculations in the main text reference all primary melt compositions to a mantle olivine at Fo90. If the mantle is more depleted by melt extraction, as might be the case under the volcanic arc, or more fertile, the final endpoint of the calculation may reasonably be Fo91 or Fo89, respectively. Figure A12, panel a, demonstrates the sensitivity of the final calculation to mantle olivine Fo#. The difference between primary fO 2 calculated for arc and back-arc primary melts in equilibrium with Fo89 and Fo90 olivine, or Fo90 and Fo91 olivine, is equal to ~0.07 log units relative to the QFM oxygen buffer. This is much smaller than the observed offset between MORB, Mariana trough, and Mariana arc magmas.
The calculations in the main text correct the major element compositions of samples with MgO between 5 and 7 wt% to a reference MgO = 7.0 wt%, after which the calculation assumes that olivine is the only liquidus phase back to a melt composition in equilibrium with Fo90 olivine. For the arc samples, it is possible that clinopyroxene remains a liquidus phase to higher MgO concentrations. Figure (2012), Kelley et al., (2010), and , respectively. The white circles are whole rock data for tephras collected at Agrigan volcano (Plank, unpublished data). Black circles are whole rock data for lavas collected at Agrigan volcano, compiled using GeoROC (Electronic appendix J). The black line is the Petrolog3 fractionation model, using a starting composition equal to Agri04-05 and fractionating olivine, clinopyroxene, and plagioclase at 1 kbar, using mineral melt models of  and , treating Fe 2+ /Fe 3+ as a closed system .  , respectively. Black circles are whole rock data for lavas collected at Pagan volcano, compiled using GeoROC (Electronic appendix J). The black line is the Petrolog3 fractionation model, using a starting composition equal to HDP1147-R06 and fractionating olivine, plagioclase, clinopyroxene, and magnetite at 3 kbar and decompressing at a rate of 5 bar/°C. Mineral melt models of , , and  were used, treating Fe 2+ /Fe 3+ as a closed system . for Alamagan melt inclusions from tephras Ala02 and Ala03, compared to the available literature data for Alamagan melt inclusions and lavas. Large circles are melt inclusions used in this study. Small circles are olivine hosted melt inclusions that have been eliminated from this study due to Fo host -Fo melt inclusion disequilibrium (light green) and to elevated FeO* relative to the literature data (dark green). The open, thick lined circles are olivine hosted melt inclusions from . The open, thin lined circles are whole rock data for tephras collected at Alamagan volcano (Plank, unpublished data). Black circles are whole rock data for lavas collected at Alamagan volcano, compiled using GeoROC (Electronic appendix J). The black line is the Petrolog3 fractionation model, using a starting composition equal to Ala02-01 and fractionating olivine, plagioclase, and clinopyroxene at 2 kbar and decompressing at a rate of 15 bar/°C. Mineral melt models of  and  were used, treating Fe 2+ /Fe 3+ as a closed system .  , respectively. The white circles are whole rock data for tephras collected at Guguan volcano (Plank, unpublished data). Black circles are whole rock data for lavas collected at Guguan volcano, compiled using GeoROC (Electronic appendix J). The black line is the Petrolog3 fractionation model, using a starting composition equal to SD46-1-1 and fractionating olivine, plagioclase, clinopyroxene, and magnetite at 2 kbar and decompressing at a rate of 5 bar/°C. Mineral melt models of , , and  were used, treating Fe 2+ /Fe 3+ as a closed system .   Kelley et al. (2010) and , respectively. The white circles are whole rock data for tephras collected at Alamagan volcano (Plank, unpublished data). Black circles are whole rock data for lavas collected at Sarigan volcano, compiled using GeoROC (Electronic appendix J). The black line is the Petrolog3 fractionation model, using a starting composition equal to Sari15-04-27 and fractionating olivine, plagioclase, and clinopyroxene at 1.8 kbar and decompressing at a rate of 9 bar/°C. Mineral melt models of , , and (Nielsen, 1988) were used, treating Fe 2+ /Fe 3+ as a closed system .   , and Danyushevsky (2001) were used, treating Fe 2+ /Fe 3+ as a closed system .   vs. the measured Fo# of the olivine host. Solid black line is a 1:1 relationship. Dashed black lines represent the error envelope of the equilibrium Fo# calculation. This error envelope is calculated by propagating average analytical error for FeO* and MgO through the equations for calculating equilibrium Fo#. Any melt inclusion that falls below the 1:1 line is subject to the PEC correction described in the main text, until the melt inclusion composition is in equilibrium with the composition of its olivine host. Any melt inclusion composition that falls above the 1:1 line, but lies within the error envelope, is left uncorrected.         Oxygen fugacity (fO 2 ) is an intrinsic thermodynamic property that records the chemical activity of oxygen and controls the speciation of multi-valent elements in the solid Earth. The ratios of oxidized to total Fe (expressed as Fe 3+ /ΣFe) in unaltered, basaltic samples from the modern Mariana arc reflect mantle wedge fO 2 that is up to 1.8 orders of magnitude higher than for MORBs 5,6,8 , which is linked to the influence of aqueous fluids released from the oxidized subducting oceanic lithosphere on the mantle wedge beneath the arc volcanoes [2][3][4][5][6][7] . The timescales and material flux required to oxidize the wedge have been modeled 9,10 , but lack observational constraints and thus the results of these models have big uncertainties. Here, we present the Fe 3+ /ΣFe ratios of pristine submarine glass from basaltic pillow lavas that record the initiation and evolution of subduction along the Izu Bonin-Mariana (IBM) convergent margin, from 52 Ma to the modern Mariana arc. We use these data to determine the timescales over which the Mariana mantle wedge fO 2 became more oxidized than MORB source mantle. We also report the Fe 3+ /ΣFe ratios of pristine, Jurassic age Pacific MORB glasses recovered from ODP Site 801C (Fig. 1a)  As the lavas become younger, they reflect increasing influence of aqueous slab fluids in the mantle wedge 12 . The youngest group of lavas (<43 Ma) record the transition to normal arc lavas, similar in composition to the modern IBM volcanic arc 12 . These lavas were erupted in the submarine environment and samples included in this study have pristine glass along quench margins (Fig. 1b). We measured high precision Fe 3+ /ΣFe ratios (±0.005) of five FAB, three boninite, and two Site 801 pillow glasses by micro-x-ray absorption near-edge structure (µ-XANES) spectroscopy (see supplement) 15 . Four early transitional pillow glasses from DSDP Site 458 were too micro-crystalline to obtain µ-XANES spectra free from crystal interference. For these samples, Fe 2+ O determinations were done using micro-colorimetric procedures and combined with bulk glass FeO* concentrations to calculate Fe 3+ /ΣFe ratios (1σ±0.02, see supplement) 16,17 . Although less precise, the results from micro-colorimetry are comparable to the results from µ-XANES (see supplement).

Arc basalts have a higher proportion of oxidized Fe (Fe 3+ ) relative to reduced Fe (Fe 2+ ) compared to mid-ocean ridge basalts (MORB), likely
Jurassic-aged MORB glasses (Site 801; ~170 Ma) have Fe 3+ /ΣFe ratios of 0.167 and fall within the modern MORB field in both major element composition and Fe redox (Fig. 2). The FAB glasses have Fe 3+ /ΣFe ratios that range from 0.165 (overlapping with MORB) to 0.195 (slightly more oxidized than MORB) and span a wide range in compositions from 7.56-2.75 wt% MgO (Fig. 2). Early transitional glasses are more oxidized, with Fe 3+ /ΣFe ratios that range from 0.202-0.249 at 4.56 -7.56 wt% MgO, which overlap entirely with modern Mariana arc basalts and are significantly more oxidized than MORB or Mariana trough glasses (Fig. 2). The boninite glasses have slightly lower Fe 3+ /ΣFe ratios (0.210-0.220), consistent with the transitional glass with the lowest MgO content. The black and gray lines in figure 2 are modeled liquid lines of descent that show the expected change in Fe 3+ /ΣFe ratios during crystal fractionation in a system closed to oxygen (unbuffered) for modern back-arc and arc magmas (see supplement). It is apparent from Fig. 2 that although fractionation undoubtedly played a small role in modifying the Fe 3+ /ΣFe ratios of these glasses (Fig. 2, see supplement), the elevated Fe 3+ /ΣFe ratios of boninite and transitional glasses are unrelated to the Fe 3+ /ΣFe ratios of FAB or MORB, and cannot be generated by shallow level crystal fractionation of a reduced primary melt. This is similar to the case of modern Mariana back-arc and arc samples 6 and suggests that there were fundamental differences in the fO 2 of the mantle sources that produced these magmas.
The major element relationships for these samples indicate that a simple linear correction is valid to account for variations during differentiation down to relatively low MgO concentrations, so we calculated primary melt compositions and Fe 3+ /ΣFe ratios for all samples with MgO>4.5 wt% (see supplement). Using the corrected major element compositions, we calculate primary melt fO 2 relative to the quartz-fayalitemagnetite (QFM) buffer at pressure and temperature according to  using the algorithm of . We pair the calculated primary melt Primary Jurassic-aged MORBs from ODP Site 801C have Ba/La ratios that range from 3.8 to 4.5 and fO 2 from QFM to +0.08, similar to modern MORB primary melts. This indicates that there has been no change in the fO 2 of MORB source mantle from Jurassic to present day (Fig. 3a). Primary FAB melts have Ba/La ratios that range from 4.3-10.3 and fO 2 from QFM+0.05 to +0.4, which overlaps MORB primary melts at the low end but extends to slightly higher Ba/La ratios and fO 2 . This suggests that there are small additions from the subducted slab during melt generation processes at the immediate onset of subduction in the Marianas (Fig 3a, b).

Methods Summary
We determined Fe 3+ /ΣFe ratios of pillow glass by micro-x-ray absorption nearedge structure (µ-XANES) spectroscopy at beamline X26A, National Synchrotron Light Source, Brookhaven National Laboratory 1 . Spectra were collected in fluorescence mode from 7020 eV to 7220 eV using a Si [311] monochromator and a nominal beam size of 9x5 µm. LW-0 was monitored continuously during each experimental session to correct for instrument drift. Spectra were scrutinized for any influence from phenocrysts or micro-phenocrysts in the glass chips and inclusions.
Determinations of Fe 3+ /ΣFe ratios following these methods have an associated precision of ±0.005 1 .
In some cases, pillow glasses were too micro-crystalline to obtain optically

Subduction influence
Basaltic lavas from the Izu-Bonin-Mariana forearc record subduction initiation and the evolution of melt generation, from primarily decompression melting and minor traces of slab fluid influence to significant slab fluid influence and flux melting 6 . This is evident in their trace element compositions (S. Fig. 1). The trace element signatures of FAB glasses are similar to NMORB, with enrichments in elements typical of influence from fluids released from subducting slabs ( Fig. 1a;  Each composition was projected back to MgO = 7.0 wt% using the slope defined by the natural data between 5 and 7 wt% MgO for all major elements (S. Fig. 2   proxies, from QFM to QFM+2 for the V/Yb and Cu proxies, and ≥QFM for the V/Sc proxy. To improve these estimates for arc mantle, more work on the mobility of transition row elements during subduction is necessary.

INTRODUCTION
Oxygen fugacity (fO 2 ) is a fundamental thermodynamic property that represents the availability of oxygen to govern reduction-oxidation (redox) equilibria in the solid Earth. For example, Fe in silicate melts exists as a mixture of reduced (Fe 2+ ) and oxidized (Fe 3+ ) species (expressed as Fe 3+ /ΣFe), where the proportion of each species present in the melt can be related to fO 2 using the simple equilibrium, By governing the proportions of multi-valent species in magmatic systems, fO 2 can influence elemental partitioning behaviors during melting, crystallization, and degassing .
Despite its key importance in petrogenetic processes, fO 2 has been difficult to constrain in natural igneous systems because it cannot be directly measured. Instead, we rely on calibrated proxies, such as the direct speciation or crystal/liquid partitioning behavior of multi-valent elements, as records of magmatic and mantle fO 2 .
Classically, Fe redox ratios (i.e., Fe 3+ /ΣFe) recorded in erupted basaltic lavas have been used to constrain mantle source fO 2 in mid-ocean ridge and arc settings (e.g., , showing that the mantle source for mid-ocean ridge basalts (MORB) have fO 2 ~ one to two log units more reduced than the quartzfayalite-magnetite oxygen buffer (QFM-1 to QFM-2) and QFM+1 to QFM+4 for the mantle source of arc basalts. These Fe redox ratios were obtained using wet chemical titration techniques on whole rock erupted lavas, which present significant analytical challenges (e.g.,  and natural bulk samples analyzed using wet chemical titrations may not yield magmatic Fe 3+ /ΣFe ratios due to crystal accumulation or other analytical interferences (e.g., . Given the potential for composition changes in crustal magma chambers, the relevance of whole-rock Fe 3+ /ΣFe ratios to the fO 2 of the mantle source has also been questioned (e.g., . More recently, trace element proxies (e.g., V-based, Cu, and Zn/Fe*) have been developed to avoid the challenges of magmatic redox changes during differentiation processes and alteration of Fe 3+ /ΣFe ratios during weathering, and provide ease of use by exploiting commonly-measured trace elements (Jackson et al., 2010;. These proxies are designed to avoid the effects of differentiation, such that the measured trace element composition of the lava reflects only information related to the mantle source. They require knowledge of the mantle source composition, mineral mode, and mechanism for melting, all of which can be difficult to constrain in some tectonic settings (e.g., subduction zones). Considering all of these proxies, the current view of fO 2 in the upper mantle varies widely. The V/Sc, Zn/Fe*, and Cu-based proxies all suggest that the fO 2 of the upper mantle is homogeneous and independent of tectonic setting, though the specific value for upper mantle fO 2 varies with each proxy. The V/Sc ratio predicts fO 2 for the upper mantle between QFM-1 and QFM+1 , Zn/Fe* ratio predicts fO 2 between QFM-2 and QFM+2 , and the Cu-based proxy predicts fO 2 between QFM-2 and QFM Fig. 1a). Like the Fe-based proxy, the V/Yb ratio and D V olv/melt proxies suggest instead that the fO 2 of the upper mantle is heterogeneous, though each of these proxies presents a wide range in values for fO 2, particularly in subduction settings. According to the V/Yb proxy, MORB mantle has fO 2 ~QFM-1 and arc mantle is between QFM-1 and QFM+3 . Much less data exists for assessing the D V olv/melt proxy in natural settings, but the few existing data suggest that MORB mantle has fO 2 between QFM and QFM+1, while arc mantle is between QFM+1.5 and QFM+3 Fig. 1a).
Across these various proxies, we find an uncertainty of four orders of magnitude for the fO 2 of MORB source mantle and uncertainty of five or more orders of magnitude for arc source mantle. A fundamental motivating question thus remains, what is the fO 2 of the upper mantle, and is it heterogeneous with respect to tectonic setting? Here, we test several proxies for constraining mantle fO 2 in basaltic systems by applying the proxies to the same set of samples. We compare the fO 2 derived via these proxies on basaltic samples from MORB, the Mariana trough back-arc spreading center, and Mariana arc volcanoes in order to trace the origins of discrepancies between transition element proxies for fO 2 . We show that careful accounting of model uncertainties related to element mobility in subduction zones and variations in partition coefficients in concert with examination of large datasets on a volcano-tovolcano basis, may reconcile the perceived discrepancies between many of these proxies. Our assessment taken as sum points towards an upper mantle that is heterogeneous in fO 2 as a function of tectonic setting. Guguan, Alamagan, Agrigan, Pagan). We also include submarine glasses from Pagan and NW Rota-1 volcanoes (Brounce et al., revised;. We use submarine glasses from the Malaguana-Gadao segment of the southern Mariana trough (Brounce et al., revised) and from the northern Mariana trough (18.1° -20.9°N; for the back-arc basin basalt dataset. For MORB, we use a global distribution of samples (Cottrell and Kelley, 2011 and references therein;. The Fe 3+ /∑Fe ratios and trace element concentrations were all determined using the same analytical procedures in the same laboratories, thus producing an internally consistent compiled data set.

Analytical methods
We We also present new wet chemical titrations for Fe 2+ O (actual) using microcolorimetric techniques outlined by  and modified by Carmichael (2014) (Plank, unpublished ICP-AES data) to calculate Fe 3+ /ΣFe ratios.

Determining inclusion/olivine equilibrium and post-entrapment crystallization
We use only olivine host-melt inclusion pairs that satisfy the screening tests of Brounce et al. (revised), the results of which we summarize briefly here. Further details can be found in the supplementary material. Equilibrium host forsterite contents (Fo) were calculated for melt inclusions and compared to the measured Fo for the olivine host of each inclusion using a FeO/MgO K ! !"#/!"# = 0.3

Arc and back-arc LLDs
Variations in major element concentrations due to fractional crystallization for melt inclusion and submarine glass suites were constrained either using Petrolog3  or empirical trends in the natural data and presented by Brounce et al. (revised). Melt compositions in equilibrium with mantle olivine at Fo 90, which we refer to here as primary melt compositions, were reconstructed for each sample with MgO > 5 wt%, using methods modified after Klein and Langmuir (1987). Each composition was projected back in composition to MgO = 7.0 wt% using the slope of the fractionation model or the natural data between 5 and 7 wt% MgO for all major elements. Where the fractionation model was poorly fit to the most incompatible elements (e.g., P 2 O 5 ), and FeO (actual) and Fe 2 O 3(actual) , the slope of a line defined by the natural data between 5 and 7 wt% MgO was used instead. Once at MgO = 7 wt%, each composition was subject to the addition of equilibrium composition olivine in 0.1% increments until in equilibrium with Fo 90 olivine. Details of these models and primary melt composition calculations can be found in Brounce et al. (revised).
Trace elements V, Sc, Yb, and Zn were not corrected for the effects of fractional crystallization in arc and back-arc basalts. Their respective proxies for fO 2 are designed to be unaffected by low-pressure crystal fractionation and volcanic degassing and, accordingly, there is little systematic variation in V/Sc and Zn/Fe* ratios with decreasing MgO (Fig. 2b,

MORB LLDs
To calculate primary MORB melts, we correct the major element compositions of MORB glasses that have only olivine on the liquidus (i.e., glasses with MgO > 8.5 wt%) by adding equilibrium olivine compositions back until the melt composition is in equilibrium with Fo 90 olivine .
Trace elements V, Sc, Yb, and Zn were not corrected for crystal fractionation for the reasons described above. Copper is compatible in sulfide phases, however, and MORB are likely saturated with a sulfide phase from the onset of crystallization ( Fig. 2e;    (Fig. 3). Although there is scatter in the Cu contents of the global MORB data set, Figure 3 shows an individual segment of the East Pacific Rise from 8.3°N -13.7°N that shows coherent fractionation trends in major elements , as well as a clear liquid line of descent for Cu.
We use the slope of this EPR data set to correct all MORB Cu concentrations back to equilibrium with Fo 90 .

Conditions of melt generation
Temperatures and pressures of last equilibration of primary melt with peridotite were calculated using the thermobarometer of . Melt fractions were calculated from primary melt TiO 2 concentrations (TiO 2 Fo90 ) using the expression from Kelley et al. (2006) and Kelley et al. (2010): where C o Ti is the concentration of TiO 2 in the mantle source (0.133), C l Ti is the calculated concentration of TiO 2 in primary melts, and D Ti is the bulk distribution coefficient for Ti during mantle melting (0.04; Kelley et al., 2006).

RESULTS
In order to assess each proxy on our dataset, first we apply each proxy for fO 2 as developed in the published literature on our sample set, using the new and compiled data presented here. In this section we report these initial model outcomes, and in the Discussion section that follows we explore sources of model uncertainty and explanations for disagreements among the various proxies.

Fe 3+ /∑Fe ratios
The relationship between Fe speciation and fO 2 has been experimentally calibrated for a wide range of natural basaltic melt compositions, temperatures, and pressures (Kilinc et al., 1983; (Brounce et al., revised;. These studies have shown that the Fe 3+ /∑Fe ratios of MORB glasses range from 0.13-0.17, with an average value of 0.16±0.01 (Fig. 2a).
Mariana trough glasses overlap with the most oxidized MORB and extend to more oxidized values, with Fe 3+ /∑Fe ratios that range from 0.15-0.21 (Fig. 2a). Mariana arc samples have significantly higher Fe 3+ /∑Fe ratios (0.20-0.34) than Mariana trough or MORB glasses (Fig. 2a). This is consistent with previous observations from the whole rock record, that arc basalts erupt with a higher proportion of Fe 3+ than do MORB . We calculate magmatic fO 2 for each sample (i.e., the fO 2 of the magma as recorded by naturally quenched volcanic glasses) relative to the quartzfayalite-magnetite oxygen buffer (QFM;  using the major element compositions and Fe 3+ /∑Fe ratios measured for each sample, with the algorithm of

Vanadium
Vanadium is a multi-valent trace element in the range of fO 2 relevant to the solid Earth. As fO 2 increases, the oxidation state of V increases and V becomes more incompatible during melting and crystallization. The relationships between D V mineral/melt and fO 2 have been calibrated experimentally for olivine, orthopyroxene, clinopyroxene, and spinel , providing the foundation of three proxies for calculating mantle source fO 2 -V/Sc ratios in basalts and komatiites ), V/Yb ratios in basalts , and D V olv/melt .
The V/Sc proxy assumes similar behavior of V and Sc early in shallow level differentiation processes (i.e., at MgO > 8.0 wt%) to create an fO 2 -sensitive ratio that should not be affected by volcanic degassing or olivine crystallization . Because D V mantle/melt is sensitive to fO 2 , the measured V/Sc ratio in a primitive basalt may be related directly to the fO 2 of its mantle source, provided the mantle source composition and the extent of mantle melting are known. For fixed source concentrations of V and Sc, and melt fraction (F), a mantle with high fO 2 should yield erupted magmas with high V/Sc ratios relative to a mantle with low fO 2 ( Fig. 4; . It is uncommon to find volcanic glass with MgO > 8.0 wt% in subduction settings, so we extend our consideration of samples to 5.0 wt% MgO, taking care to examine each sample suite for the effects of crystal fractionation on V/Sc ratios. The distribution of measured V/Sc ratios in MORB, Mariana trough, and Mariana arc basalts with MgO > 5.0 wt% are similar, ranging from ~5 to 12 (Fig. 2b). This is consistent with previous observations using whole rock data from literature databases GEOROC and PetDB  and supports the notion that V/Sc ratios are not significantly affected by shallow level differentiation (Fig. 4a). Following the procedure of  for calculating mantle source fO 2 , we use uniform mantle abundance for V = 83 ppm and Sc = 16.5 ppm (equivalent to primitive mantle composition, McDonough and Sun, 1995), bulk D Sc = 0.458 at 1350ºC, and parameterizations of D V for mantle minerals described by . The model assumes that basalts erupted in all three tectonic regions are best represented as accumulated fractional melts. Using these conditions, the measured V/Sc ratios in MORB, Mariana trough, and Mariana arc basaltic glasses correspond to mantle source fO 2 between QFM and QFM+2 (Fig. 4a).  proposed the V/Yb ratio proxy with the goal of minimizing the impact of crystal fractionation on the ratio used to calculate mantle source fO 2 because unlike Sc, which is compatible in clinopyroxene, Yb is incompatible in all silicate mineral phases that are likely to fractionate from basaltic magmas (e.g., olivine, plagioclase, clinopyroxene; . The V/Yb ratios of samples in this study are clearly affected by magnetite saturation in samples with MgO < 5.0 wt%, which causes a sharp decrease in V melt The D V olv/liq proxy for magmatic fO 2 requires knowledge of the magmatic temperature at the time of olivine fractionation. For consistency with the published model, we used the Sc/Y thermometer of , who show that Sc/Y exchange between olivine and melt is temperature sensitive in the same experimental samples against which the D V olv/liq fO 2 proxy is calibrated. This thermometer is also shown to be relatively consistent with MgO olivine thermometers . The log D V olv/liq calculated for each melt inclusion-olivine host pair range from -1.68 to -1.97, corresponding to a range in magmatic fO 2 at the time of the melt inclusion entrapment from QFM+2.3 to QFM+3.0 for arc magmas (Table 3). We did not examine olivine phenocrysts in submarine glass samples, so we cannot apply this proxy to Mariana trough or MORB magmas. Three MORB olivines were included in , yielding magmatic fO 2 between QFM-0.5 and QFM.

Zn/Fe* ratios
Zinc occurs only as Zn 2+ in the solid Earth and may have similar partitioning behavior as Fe 2+ during mantle melting . When fO 2 is low and most of the Fe in a system exists as Fe 2+ (i.e., Fe 3+ /ΣFe ratios are low), Zn may not be significantly fractionated from Fe during mantle melting. As mantle source fO 2 increases, the proportion of Fe 3+ in the source increases, and Zn and Fe* (i.e., Fe* is total Fe, regardless of valence state) may be fractionated from one another as a result of the greater incompatibility of Fe 3+ in silicate mineral phases. A mantle with a higher Fe 3+ /ΣFe ratio may produce melts with lower Zn/Fe* ratios than that of a mantle with low Fe 3+ /ΣFe ratios. The Zn/Fe* proxy was developed to calculate directly the Fe 3+ /ΣFe ratio of primary melts, for a mantle with given Zn/Fe* and Fe 3+ /ΣFe ratio .
The Zn/Fe* ratios in MORB, Mariana trough, and Mariana arc basalts with MgO > 5.0 wt% are similar, ranging from ~8 to 14 (Fig. 2d, Tables 1, 2), consistent with observations from published whole rock data . Above 5 wt% MgO, there is no variation in Zn/Fe* ratios with MgO concentrations for MORB, Mariana trough and Mariana arc glasses (Fig. 2d). The total variation at any one MgO value is large for all samples (MORB, Mariana trough, and Mariana arc glasses), however, which may potentially reflect differences in source composition and Fe 3+ /∑Fe ratios, and/or differences in the temperature of melting that may lead to variation in Fe* concentrations. Nonetheless, we conclude that the effects of crystal fractionation on the Zn/Fe* ratio in samples in this study are small enough to be negligible. Because Zn/Fe* ratios of magmas are expected to decrease as Fe 3+ concentration in the source increases, the maximum Zn/Fe* ratio in a mantle melt is tied to the Zn/Fe* ratio of a solid mantle assemblage that has no Fe 3+ present. The higher the fO 2 of the mantle source, the more Fe 3+ is present, and upon melting, the lower the Zn/Fe* ratio of the melt. Thus, an undifferentiated mantle melt cannot have a Zn/Fe* ratio that is greater than the Zn/Fe* ratio of the source mantle. Using a mantle source Zn/Fe* of 9-10, mantle source Fe 3+ /∑Fe ratio = 0.03 , and a K d(perid/melt) Zn/Fe2+ of 0.84 , there are only a few MORB samples to which the Zn/Fe* proxy can be applied (gray box, Fig. 7a given Zn/Fe* ratio, using a lower K d(perid/melt) Zn/Fe2+ produces calcualted Fe 3+ /∑Fe ratios that are higher than when K d(perid/melt) Zn/Fe2+ = 1 is used.

Cu concentration
Like Zn, Cu exists in only one oxidation state (Cu 1+ ) in the range of fO 2 s typical of igneous rocks. As a chalcophile element, Cu partitioning during mantle melting can be dramatically altered by the presence of sulfide phases in the mantle.
The speciation of sulfur is very sensitive to changes in fO 2 in the range of QFM to QFM+2, where sulfide (S 2-) is the dominant species of S at fO 2 s lower than QFM, and as fO 2 increases from QFM to QFM+2, S species change to sulfate. A change in mantle fO 2 from QFM to QFM+2 and the consequent disappearance of sulfides will affect bulk Cu partitioning during mantle melting, making Cu more incompatible in an oxidized mantle with no sulfide phase present. For a mantle with a given Cu content and sulfide mode, a mantle above QFM+2 will produce melts with higher primary melt Cu concentrations than a mantle below QFM+2 . It is important to note that this is only true if sulfide phases are not exhausted during melting. The effect of sulfides on Cu partitioning is erased if the sulfide mode is small and melt fraction is large.

Measured Cu concentrations in MORB and northern Mariana trough lavas
decrease with decreasing MgO, likely because they reflect saturation of the magma with a sulfide phase (Fig. 2e). Agrigan melt inclusions and southern Mariana trough lavas also have decreasing Cu concentrations with decreasing MgO (Fig. 2e), although these samples show no other evidence of sulfide saturation; for example, there are no sulfides present either as included phases in olivine phenocrysts or as separate phases present in the erupted tephra or in the pillow lava (for southern Mariana trough samples; Brounce et al., revised). The sulfur concentrations of these samples are much lower than experimentally calibrated estimates of the sulfur content at sulfide saturation, at the fO 2 indicated by measured Fe 3+ /∑Fe ratios of the glasses , and there is evidence that sulfur has partitioned into a vapor phase (Brounce et al., revised;. Copper has been shown to partition into vapor phases during shallow level volcanic degassing in both mafic and felsic magmas   A small subset of Mariana arc primary melts have elevated Cu concentrations that are more consistent with mantle fO 2 of ~QFM+1.3 (gray shaded region, Fig. 8).

Comparison
Each proxy yields slightly different results when applied to samples in this study. The  (Table 2). However, the log D V olivine/melt calculated directly from trace element measurements on the melt inclusions and their olivine hosts, paired with their magmatic fO 2 calculated from measured Fe 3+ /∑Fe ratios (Fe-XANES) are consistent with the experimental dataset upon which the D V olivine/melt model is calibrated (Fig. 6). The Zn/Fe* proxy yields primary melt Fe 3+ /∑Fe ratios (0. 15-0.19) that are broadly consistent with Fe 3+ /∑Fe ratios reconstructed from Fe-XANES measurements (0.12-0.16) on the few MORB samples to which the proxy can be applied. The Cu-based proxy is also consistent with Fe-, V/Sc-, and Zn/Fe-based proxies when applied to MORB and Mariana trough glasses, predicting a mantle source fO 2 of ~QFM and ~QFM+0.5 respectively. For Mariana arc samples, the Cu proxy predicts mantle fO 2 between ~QFM and QFM+1.3.

Fe 3+ /∑Fe ratios in basaltic glasses versus whole rock analyses
Iron is a major element. Magmatic differentiation can change its concentration, and differentiation and post-eruption chemical weathering can change its redox ratios.
For these reasons, it can be difficult to be certain that measurements of Fe 3+ /ΣFe ratios on erupted lavas reflect magmatic values, and even more difficult to reconstruct the Fe 3+ /ΣFe ratios of undifferentiated mantle melts. This is apparent from the comparison of Fe 3+ /ΣFe ratios determined from bulk tephra and from melt inclusions from the same tephra (Fig. 2a). Most of the tephras have slightly lower MgO contents than their melt inclusions, representing more evolved magma compositions, and are also more oxidized. The tephra powders are all visibly weathered and their bulk Fe 3+ /ΣFe ratios may not be entirely representative of the magmatic redox state.
The application of Fe-XANES techniques to determining Fe 3+ /∑Fe ratios in basaltic glasses presents the opportunity to address many of the main concerns regarding Fe redox proxies for magmatic and mantle source fO 2 . Although some sample suites display excellent coherence between magmatic and whole rock redox determinations (e.g., , by measuring Fe 3+ /∑Fe ratios in situ in pristine volcanic glass, we can avoid any potential effects of alteration that may occur under the relatively oxidizing conditions at the Earth's surface (e.g., Grocke et al., submitted). We also avoid the large and potentially random error that may occur during the weighing and transfer of powders in micro-colorimetry procedures for samples with relatively low FeO* concentrations (MgO < 5.0 wt%). While microcolorimetry can be precise, for example in the study of MORB glasses (less than 1% relative; , when FeO* contents are low and the samples are more oxidized, the Fe 2+ O actual concentration can be very low and small weighing errors correspond to large errors in the calculated Fe 3+ /ΣFe ratios (see error bars, Fig. 2a).
Finally, by applying Fe-XANES techniques to carefully chosen melt inclusions, we can measure Fe 3+ /∑Fe ratios in samples that potentially record changing magmatic compositions during crystallization and degassing. This information can be used to determine the effects of these processes on magmatic Fe 3+ /∑Fe ratios and if necessary, to correct for these variations. Once this is done, the algorithm of  can be used to calculate magmatic and mantle source fO 2 to ± 0.5 log units.
The range in possible mantle source fO 2 calculated from Fe 3+ /∑Fe ratios determined via Fe-XANES is much smaller than that calculated from Fe 3+ /∑Fe ratios determined via micro-colorimetric techniques (Fig. 1) and indicates that primary arc melts are more oxidized than MORBs (Brounce et al., revised). Additionally, no study has yet demonstrated an oxidizing differentiation process in natural magmas that could create the very oxidized Fe 3+ /∑Fe ratios observed in arc lavas (Brounce et al., revised;Cottrell, 2012, Moussalam, 2014). Thus we conclude that the Fe 3+ /∑Fe ratio proxy strongly suggests that the upper mantle is heterogeneous with respect to fO 2 , where the mantle source for arc volcanism is more oxidized than that of MORB.

Ratios of vanadium to other trace elements
Application of the V/Sc proxy for determining mantle source fO 2 from the composition of erupted basalts involves making several assumptions or inferences about the mechanisms for melt generation and the composition of the source itself.
First, it is necessary to distinguish between mechanisms for melt generation in various tectonic settings. Mantle melts at mid-ocean ridge and back-arc basin settings exist because the mantle is upwelling passively in response to divergent plate motions; these melts are best modeled as accumulated fractional melts . Under an arc however, melt generation takes place due to a shift in the position of the peridotite solidus towards lower temperatures because of the addition of water to the mantle wedge from subducting slabs. The melting column under arc volcanoes integrates melts formed over a relatively focused depth range corresponding to the hot core of the mantle wedge .
Although there remains much to understand about melt generation and transport processes in this part of the mantle wedge, arc lavas may be better modeled as equilibrium batch melts (e.g., Kelley et al., 2010). When modeling the behavior of mildly incompatible elements such as Sc, and V in settings with fO 2 <QFM+2 during melting, at a fixed mantle source composition, the difference between accumulated fractional melts (black lines, Fig. 4a) and batch melts (gray lines, Fig. 4b) for the calculated mantle source fO 2 is small. At fO 2 higher than QMF+2, V becomes increasingly incompatible (D bulk V < 0.1) and the difference between batch and fractional melting processes for calculated mantle source fO 2 increases. In particular, at fO 2 > QFM+2 and F > 0.10, a batch melting model yields a slightly lower source fO 2 than a fractional melting model at a given V/Sc ratio.
The ranges in possible source concentrations for V and Sc are also important to consider.  use estimates for bulk silicate Earth (V o = 83 ppm, Sc o = 16.5 ppm, McDonough and Sun, 1995) for both mid-ocean ridge and arc mantle sources (black lines, Fig. 4a). Here, we used the average compositions for NMORB, EMORB, and DMORB reported by  to calculate the range of V source and Sc source for mantle that feeds the mid-ocean ridge system. We calculated primary The full range of expected V/Sc ratios during melting of a source mantle within the range of V and Sc concentrations calculated for each MORB group at QFM is small (thick gray line, Fig. 4a) but offset to higher V/Sc ratios than when a BSE source is used. Whichever compositions are used, the V/Sc model shows that MORB are consistent with extraction from a source mantle at ~ QFM. The mantle source for Mariana trough lavas unmodified by subduction is NMORB to EMORB mantle in composition , and for southern Mariana trough samples with the highest V/Sc ratio, using the NMORB source composition calculated here, suggest mantle fO 2 > QFM+1.
In the case of the Mariana arc, the mantle source under the arc is the residue of melt extraction at the Mariana trough (Plank, 2005, Stolper and. We model this composition by assuming that 10% fractional melt is extracted from the source composition at the back arc, yielding a source mantle with 14.9 ppm Sc and 86.2 ppm V if melting at the back arc took place at QFM, or 76.5 ppm V if melting at the back arc took place at QFM+1. We take the first composition and use a batch melting model to calculate the range of V/Sc ratios expected during melting beneath the arc at QFM, QFM+1, and QFM+2 (gray solid lines, Fig. 4b), and find that this model is not significantly different from models using MORB type mantle compositions (compare to gray field in Fig. 4b). We find, however, that if we take any of the source Sc concentrations discussed here, to generate the Sc abundances observed in Mariana trough and Mariana arc lavas with the highest MgO concentrations, it would require D Bulk Sc < 0.3, which occurs at mantle temperatures in excess in 1500ºC (calculated from Beattie et al., 1991, This temperature is likely too hot to be reasonable for melting in the mantle wedge . At ~1350ºC, to generate Sc abundances in mafic arc magmas from this study requires either (1) a mantle source with higher Sc concentrations, (2) Sc addition to the mantle by aqueous fluids or melts from the subducting slab, or (3) a lower bulk partition coefficient for Sc during melting than is used in the V/Sc model, or (3) . The mantle under the Mariana arc is more refractory and depleted than mantle under the Mariana trough, so it is unlikely that the mantle source has higher Sc content than outlined here (Plank, 2005, Stolper and. Scandium is not present in abundance in experimental liquids and melts found to be in equilibrium with altered oceanic crust at P and T relevant for subducting slabs, so it is also unlikely that Sc is added to the mantle wedge from the subducting slab. It may be that our D Bulk Sc should reflect more incompatible behavior for Sc during melting in arc settings, relative to MOR settings. Figure 9 shows the relationship between Sc/La, V/La, Zn/La, and Cu/La and Sm/La ratios in MORB (light gray field, Liu, 1997), Mariana trough, and Mariana arc samples. The MORB data define the relationship between each of these elements and Sm during melting of the MOR source upper mantle. Mariana trough samples from this study overlap the MORB field, suggesting that, to a first order, similar models for melt generation can be used to calculate source fO 2 in these regions. However, the Mariana arc data do not plot with MORB data, instead they have higher Sc/La, V/La, Zn/La, and Cu/La ratios than MORB samples with the same Sm/La ratio. This suggests that there is a fundamental difference in the behavior and abundance of these elements during melting between arc and MOR settings. In the case of V, the difference between MORB and Mariana arc samples in Fig. 9b potentially reflects elevated source fO 2 conditions in arc settings relative to MOR. For Sc, Zn, and Cu, it may be that the bulk partition coefficients for these elements are lower during melting in arc settings than in MOR settings. Peridotites found in the forearc of subduction zones, are commonly harzburgites with modal abundances of ~70% olivine, up to 30% orthopyroxene, and 1-3% each of clinopyroxene and spinel (e.g., Pearce et al., 1984). These harzburgites are thought to reflect significant degrees of melt extraction that takes place during hydrous melting of forearc and arc mantle sources, depleting the source rock significantly of clinopyroxene. This has an important effect on the relative partitioning of V and Sc, because D Sc cpx/melt ~1.2 and at fO 2 ≥ QFM, D V cpx/melt < 1. As fO 2 increases, V becomes increasingly more incompatible in clinopyroxene, while Sc remains compatible. At any constant fO 2 ≥ QFM, the bulk D V will increase and bulk D Sc will decrease as the modal abundance of clinopyroxene decreases. The result is that melts of a mantle depleted in trace elements (because of prior melt extraction at the backarc) that is 1% clinopyroxene, 70% olivine, 3% spinel, and 26% orthopyroxene (in the style of forearc peridotites described in Pearce et al., 1984) have lower V/Sc ratios at a given fO 2 than the same depleted trace element composition mantle that has 18% clinopyroxene, 55% olivine, 2% spinel, and 25% orthopyroxene (the modal composition used by . The red dash-dot line in figure 4b illustrates this effect for the depleted mantle composition discussed above, which becomes more pronounced as fO 2 increases above QFM. Additionally, as melt fraction increases above 20%, lines of constant fO 2 converge and it becomes difficult to distinguish between QFM and QFM+1 (Fig. 4b). If clinopyroxene is completely eliminated from the mantle assemblage at 20% melt fraction, a depleted mantle with fO 2 of ~QFM+1 will produce a melt with V/Sc = 6.4. From this, we conclude that the V/Sc ratios of Mariana arc basalts in this study need not preclude oxidized conditions in the mantle wedge if the mantle source is depleted in trace elements and has a lower modal abundance of clinopyroxene than MORB or Mariana trough source mantle.
Normalizing V to another trace element that is not compatible in clinopyroxene or other mantle phases may avoid the potential problems outlined above. Examining variations in V/Yb ratios on a volcano-to-volcano basis and choosing samples that have not clearly fractionated V/Yb ratios via magnetite crystallization limits the range of V/Yb ratios that are used to infer mantle source fO 2 significantly when compared to the range in V/Yb ratios examined by , and thus yield a narrower range for the possible fO 2 of arc mantle (Fig. 1). However, even within samples from a single volcano, we observe significant variation in V/Yb ratios in samples with MgO > 5.0 wt%, suggesting that there is more work to be done in understanding the relationship between V/Yb ratios and mantle source fO 2 .

Vanadium partitioning
The D V olv/melt proxy avoids many potential problems associated with complex subduction zone magmatism because it relies simply on the relationship between magmatic fO 2 and V partitioning between olivine and melt. The calibration of D V olv/melt as function of fO 2 developed by  yields magmatic fO 2 s that are systematically higher than given by other proxies for both MORB and arcs, and are notably offset from Fe 3+ /∑Fe ratios measured on the same inclusions (Fig. 6).
Although  base their calibration on a large experimental dataset, many of these experiments are at 1 atm, anhydrous conditions and use synthetic starting compositions that lack Fe, or use natural compositions (i.e., komatiites) that are unusual in the geologic record and not necessarily appropriate matches for normal MORB and arc basalts. Natural basaltic magmas contain dissolved volatiles and differentiate at pressures between 1 and 10 kbar (e.g.,  and these factors may impact the relationship between V partitioning and fO 2 .
Alternatively, it is possible that the Fe 3+ /∑Fe ratios recorded in arc melt inclusions are reduced during some post-entrapment process and therefore represent a minimum estimate for magmatic fO 2 . However, the fO 2 predicted for MORB olivine-pillow glass pairs from D V olv/melt are also more oxidized than other proxies, suggesting that the bias is not a melt inclusion phenomenon. Additionally, we observe that the calculated D V olv/melt values for melt inclusions and magmatic fO 2 calculated from the melt inclusions' measured Fe 3+ /∑Fe ratios are entirely consistent with those of the experimental conditions upon which the empirical relationship is derived. This suggests that there may be a bias in the empirical parameterization that describes the relationship between D V ol/melt and magmatic fO 2 . Whatever the cause of the offset between the D V olv/melt and Fe 3+ /∑Fe ratio proxies, the two proxies are consistent with one another in that they both suggest that the upper mantle is heterogeneous with respect to fO 2 , specifically that the mantle source at arc volcanoes is more oxidized than MORB.

Broader application of the Zn/Fe* proxy
The Zn/Fe* proxy is consistent with measured Fe 3+ /∑Fe ratios for the four MORB samples with Zn/Fe*≤10, assuming a mantle source Zn/Fe* = 9 ± 1. The remaining MORB, Mariana trough, and Mariana arc samples have Zn/Fe* ratios higher than ~11, a value too high to be the result of melting a mantle with Zn/Fe*=9, even when there is no Fe 3+ present. Zinc and iron may be fractionated from one another as the result of crystal fractionation in the shallow crust, in particular at the point of clinopyroxene saturation, causing the magmatic Zn/Fe* ratio to increase . The Zn/Fe* ratios measured in samples with MgO > 5.0 wt% do not correlate significantly with decreasing MgO (Fig. 2d), however, which suggests that Zn/Fe* = 9 ± 1 does not encompass the full range of possible mantle source compositions relevant to samples in this study. If the highest Zn/Fe* ratios measured in this study (~14; MORB, southern Mariana trough, Pagan) are controlled mostly by melting, it would require a mantle source with Zn/Fe* as high as 14, significantly higher than the recommended Zn/Fe* = 9 ± 1. Allowing for source variation in Zn/Fe* of this magnitude would render the model more applicable. In Fig. 7b, we show the distribution of whole rock measurements of Zn/Fe* ratios in peridotites sampled from xenoliths and exhumed terranes. The blue bars show values measured on xenoliths from the western USA and Tanzanian craton, which are the basis for Zn/Fe* source = 9 ± 1 (Le Roux et al., 2010;, although this data set actually ranges from 7 to 13. We also show the range for abyssal peridotites  and xenoliths erupted in serpentine mud volcanoes from the Mariana forearc . One sample each from the Mariana forearc and western USA/Tanzanian craton have Zn/Fe* ratios as high as 13, and several abyssal peridotites have Zn/Fe* ratios as high as 15, demonstrating that the full range of Zn/Fe* ratios in the upper mantle is greater than Zn/Fe* = 9 ± 1. To explain the full range of Zn/Fe* ratios observed in MORB and arc basalts with MgO > 5 wt% would require variation in mantle source Zn/Fe* ratios from ~8 to 14, which is consistent with observations from figure 7b. If this level of variation is allowed, then the Fe 3+ /∑Fe ratios of primary melts calculated using the Zn/Fe* proxy are entirely consistent with those calculated from direct measurements of Fe 3+ /∑Fe ratios in basaltic glasses (Fig. 1b), although we lack a means to constrain independently the Zn/Fe* ratio of the source based solely on melt composition. The Mariana forearc xenoliths with Zn/Fe* ratios > 9 are among the most heavily serpentinized xenoliths from Conical Seamount , suggesting that slab-derived fluids responsible for serpentinization may have increased the Zn/Fe* ratios of the peridotites. If true, then Zn mobility in slab-derived fluids and melts is also a key factor to consider for accurate application of the Zn/Fe* proxy.

Primary melt Cu contents
In the case of the Cu proxy for mantle source fO 2 , MORB and Mariana trough samples are clearly consistent with a mantle source fO 2 between QFM and QFM+0.5.
Above 15%  These values are either within or just outside of the limit of the confidence interval of our calculation of primary Cu concentrations in Mariana arc samples (± 10ppm).
Despite these uncertainties, the Cu contents of Mariana arc basalts reflect melting of a mantle source with fO 2 between QFM and QFM+1.3 (Fig. 1b). There is a significant impact of varying the Cu source (green dash-dotted line, QFM+1.3) for the case where melt extraction in the back-arc depletes the source composition of Cu by 8 ppm, to a value of 20 ppm. This model also relies heavily upon knowledge of the S content of the source mantle, which is difficult to know in subduction zones because of the possibility for S recycling in slab fluids or sediment melts (e.g., . If sulfur is added to the mantle wedge in slab fluids or sediment melts, increasing the S content of the mantle source and by extension the sulfide mineral mode, sulfide phases in the wedge may persist to higher melt fractions. This would make Cu more compatible in the mantle mineral assemblage at higher melt fractions, decreasing the Cu content of arc magmas at a given fO 2 . More work towards understanding the mobility of sulfur and abundance of Cu in the mantle wedge must be done before the Cu proxy can be used to accurately predict mantle source fO 2 in subduction settings, though we note that like the V/Sc and Zn/Fe* proxies, reasonable constraints can be used with the Cu proxy to yield oxidized mantle conditions at arc volcanoes.

The oxygen fugacity of the upper mantle
We have presented an assessment of several proxies for mantle source fO 2 and tested each proxy on a global set of MORB submarine glasses, submarine glasses from the Mariana trough and Mariana arc, as well as olivine hosted melt inclusions from several volcanic centers along the Mariana arc. After testing a range of different constraints that affect the outcome of each proxy, the results of this study suggest that these proxies converge upon an upper mantle that is heterogeneous with respect to fO 2 , and that this heterogeneity is a function of tectonic setting.
For MORB and Mariana trough samples, results from Fe 3+ /∑Fe ratios, V/Sc ratios, and primary melt Cu concentrations are consistent with a source mantle in these locations of QFM and between QFM and QFM+1, respectively (Fig. 1b). The Zn/Fe* ratios of a small subset of our MORB dataset are entirely consistent with Fe 3+ /∑Fe ratios determined via Fe-XANES if source Zn/Fe* is fixed at a value of 9, suggesting that MORB primary melts have Fe 3+ /∑Fe ratios ~0. Fig. 1b).
In the case of the Mariana arc, it is important to consider the full range in possible modal compositions of the mantle wedge, as decreasing clinopyroxene abundance has a significant impact on the application of the V/Sc ratio proxy to arc basalts. If the mantle source for arc magmas is depleted in bulk V and Sc abundances due to melt extraction at the back-arc and has a low clinopyroxene mode due to the generation of high melt fractions during hydrous melting, the V/Sc ratios of Mariana arc basalts need not preclude oxidized mantle wedge conditions. We emphasize the importance of examining the composition of arc basalts on a volcano-to-volcano basis when assessing the potential for compositional variability, as it can impact the application of certain trace element proxies for fO 2 . For example, we demonstrate that variation in V/Yb ratios in Mariana arc samples with MgO > 5.0 wt% is significant, suggesting that work remains to be done to fully understand the relationship between V/Yb ratios and mantle source fO 2 . In the simplified case of vanadium partitioning between melt and olivine, we show that mafic Mariana arc magmas have significantly elevated magmatic fO 2 (>QFM+2) over MORB, although we note that fO 2 s determined using D V olv/melt are systematically higher than those determined using Fe 3+ /∑Fe ratios (Fig. 1b). Finally, the Cu contents of some Mariana arc primary magmas reflect melting at QFM+1.3, consistent with constraints from Fe 3+ /∑Fe ratios, although we demonstrate that this proxy is also sensitive to starting mantle Cu and S content, which, for S in particular, is not well constrained in subduction zone settings.
Similar to the case of V/Sc ratios, the Cu contents of Mariana arc primary magmas need not preclude oxidized mantle source conditions. In total, the Fe redox, V/Sc, V/Yb, and Cu proxies for mantle source fO 2 in this study are consistent with the fO 2 of the arc mantle source between QFM and QFM+3, and in particular, that each of these proxies yield source fO 2 > QFM+1 for some portion of the arc samples in this study (Fig. 1b).
These results are summarized in Figure 1b. Overall, we recommend that any study seeking to constrain mantle source fO 2 from any proxy discussed here take full and careful consideration of each model before applying. In particular, considering the potential differences between petrogenesis at the mid-ocean ridge, back-arc spreading centers, and arc volcanoes may impact results from each proxy significantly.

CONCLUSIONS
We   Brounce et al., revised;, V/Sc ratios , D V olv/melt (this study; , Zn/Fe* ratios , and Cu abundances . Panel (a) shows results from the application of each model as recommended. Panel (b) summarizes the conclusions of this work.   Kelley, 2011, 2013), Mariana trough , and Mariana arc basalts (Brounce et al., revised).   , Mariana trough , and Mariana arc basalts (Brounce et al., in review).   . The black line is a simple linear regression with slope of 306, along which Cu contents were projected back to Fo90.  . The thick gray line shows the sensitivity of the model to changing the V and Sc mantle source abundance at QFM. In (b), a batch melting model; the black solid lines show the relationship between V/Sc ratios, melt fraction, and mantle source fO 2 using the same mantle source V and Sc compositions as in panel (a) (also at QFM, QFM+1, and QFM+2). The black dashed lines show a batch melting model for the case where melt extraction along the Mariana trough brings a depleted mantle composition into the Mariana arc source, for QFM, QFM+1, and QFM+2 cases. The red dash-dot lines use the same mantle source compositions as in the black dashed lines, but with a mantle mineral mode of 1% cpx, 70% olv, 3% spinel, and 26% opx, also at QFM, QFM+1, and QFM+2. Symbols are as in Figure 2.  , and references therein), three MORB glass-olivine pairs (black circles; , and olivine hosted melt inclusions (colored samples, symbols as in Figure 2). Magmatic fO 2 for olivine hosted melt inclusions are calculated using the Fe 3+ /∑Fe ratios (Fe-XANES) measured directly on the melt inclusions, and the algorithm of  relative to the QFM oxygen buffer at 1200°C, 1 atm . The 1 sigma error bar represents the confidence interval for magmatic fO 2 calculated using the  algorithm. The orange line is an example calibration line for the relationship between log D V ol/liq and magmatic fO 2 using melt inclusion composition Sari15-04-03. Mariana forearc Pearce, 1998 Savov et al., 2005 30 25   c. d.    (Jochum et al., 2006;. San Carlos olivine (Fo 88 ) was analyzed periodically as a check on the determination of unknown olivine compositions. Olivine grains were analyzed in single analyses. Counting statistics were examined carefully for each element and those elements that did not return strong signals for the entire length of the laser ablation period were discarded.

Inclusion/olivine host equilibrium and post-entrapment crystallization corrections
We use only olivine host-melt inclusion pairs that satisfy the screening tests of Brounce et al. (in review), which we summarize briefly here. Equilibrium host forsterite contents (Fo) were calculated for melt inclusions and compared to the measured Fo for the olivine host of each inclusion using a FeO/MgO K ! !"#/!"# = 0.3 . If the predicted equilibrium Fo inclusion matched the Fo host , we assumed equilibrium between inclusion and host. If the Fo inclusion was lower than the Fo host , indicative of crystallization within the inclusion after it was trapped in its olivine host, olivine of the equilibrium major element composition was added to the inclusion major element composition in 0.1% increments until the Fo inclusion and Fo host were equal. If the predicted equilibrium Fo inclusion was higher that the Fo host , no action was taken. Those inclusions that required >2% correction or whose Fo host -Fo inclusion disagreed by more than 2% were excluded. Each melt inclusion suite was compared to published whole rock and melt inclusion data for each volcano. Inclusion compositions that deviate significantly from the major element relationships defined by the available published data were excluded.

Abstract
Oceanic lithosphere is altered and oxidized as it spreads along the seafloor, added to the oceanic crust by alteration on the seafloor is not output by arc or back-arc magmas. If this oxygen is retained in the slab that subducts into the deep mantle, it may contribute to mantle redox heterogeneity.

Introduction
Oceanic crust is created at mid-ocean ridges and is progressively altered and oxidized as it spreads along the seafloor, until it comes to a convergent margin and subducts .
Despite the clear influence that the subducting slab imparts on the trace element and isotopic composition of arc lavas, the ultimate fate of the oxidized signature within the subducting slab is unclear. The oxidized nature of arc basalts is linked to the oxidized condition of the subducting slab [Brounce et al., in revision;, but the efficiency of the transfer of oxygen or oxidized species from the slab to the mantle wedge is poorly constrained.  [Sleep et al., 2012]. A mass balance of Fe 3+ using globally averaged data from oceanic crust, oceanic sediments, arc basalts, and oceanic island basalts suggests that the Earth's mantle has been a sink for oxidized iron through geologic time . However, subduction zones active today vary widely in the composition and flux of sediments being recycled, the age and extent of alteration of oceanic crust being subducted, the extent of serpentinization of the underlying lithosphere, the extent to which sediments and oceanic crust are returned to the arc crust during subduction, and the composition and flux of material erupted at arc volcanoes, such that this balance of Fe 3+ cycling in the Earth has large uncertainties.
Establishing links between the altered, subducting slab and the materials emplaced in the arc and back-arc crust in a single convergent margin system are necessary to determine the efficiency of redox transfer in the subduction cycle. Iron is a useful tracer of the balance of oxygen through subduction zones because it exists as Fe 2+ and Fe 3+ , and their charges are balanced mostly by oxygen. Accurately determining the proportions of Fe 2+ and Fe 3+ in subduction inputs and outputs is a key step towards developing an accurate model of the transfer of oxygen through subduction cycles.
Here, we present measurements of the Fe 2+ O of lithologies representative of inputs to the Mariana convergent margin, determined by whole-rock wet chemistry, in order to constrain the bulk oxidation state of material entering the subduction zone.
We pair these new data with previously collected major element concentrations on the same samples

Methods
Terry Plank donated powdered samples of sediment and altered oceanic crust, as well as composite powders described by  and Plank et al. [2000] ( Table 1). All samples were recovered from ODP Site 801 in the western Pacific during Legs 129  and 185 Fig. 1]. We followed the micro-colorimetric methods described by The direction in which the reaction proceeds will depend on pH, and in solutions above pH ~5, the reaction will move from right to left, reducing Fe 3+ to Fe 2+ . During digestion of rock powders, V 5+ is added to oxidize Fe 2+ from the rock powder at pH < 5. Once the rock powder is completely in solution, the pH of the solution is increased to ~5 and EQ 1 is reversed to regenerate Fe 2+ in the presence of a coloring agent, 2:2' bipyridine. The complexation of Fe 2+ with bipyridine ligands colors the solution red, which can be quantified using a spectrophotometer.

Solution recipes
First, we will outline the recipes for creating the solutions referenced in this procedure. Some of the materials necessary for these solutions are hazardous. Consult the material safety data sheets for each compound before proceeding. To make the

Results
Analytical results are presented in Table 1. Pristine glass chips from ODP Site 801, measured by XANES (see chapter 2, this volume), have the same Fe 3+ /ΣFe ratios as modern MORB glass [~0. 16;Brounce et al.,in prep], indicating that before alteration, the igneous basement here has a similar bulk redox state to modern MORB.
Whole-rock basalts that were petrographically described as "minimally altered" are more oxidized (0.29 -0.46 Fe 3+ /ΣFe; Fig. 3 Plank et al., 2000]. There are five samples in this study that were also analyzed by  for Fe 3+ /ΣFe ratios using a titration method ( Table 1). The analytical procedure is not discussed in , so it is difficult to directly compare the data in this study with the data that has been previously published. Two of those samples are taken from the surface pelagic clay units (801A3R2-145 and 801A5R3-145), and both have Fe 2+ O concentrations below the detection limit of this study and previous work . Sample 801C4R1-72 is a hydrothermal deposit that is very oxidized, and though we report a Fe 3+ /ΣFe ratio for this sample, it is very close to the detection limit of our procedure. Sample 801C5R1-95a, a pervasively altered basalt, agrees with the Fe 3+ /ΣFe ratio calculated from data from  to within 0.06. Sample 801C15R7-31, also a pervasively altered basalt, has a Fe 3+ /ΣFe ratio determined in this study of 0.33 and 0.52 as determined by , which is not in good agreement. Although the powder samples that were used for analysis in this study and in   This is difficult to assess from the brief description of the methods, which do not include any explicit information about the use of rock standards during analysis, provided by .

Discussion
The Pacific plate at ODP Site 801 is clearly oxidized as the result of alteration reactions that take place on the seafloor, prior to subduction ( Figure 3). In order to assess the fate of this oxidized signature, we present a mass balance calculation for Fe 3+ through the Mariana subduction system, using data presented in this work and in previous chapters in this volume. Implicit in such a mass balance is the assumption  . The total input flux of Fe 3+ to the Mariana trench from the subducting sediments and altered oceanic crust can be calculated following the methods outlined by  and Lecuyer and Ricard [1999]. The input flux for Fe 3+ can be calculated using the equation: where z i is the thickness of layer i, ρ i is the density of layer i, v i is the convergence rate  the seafloor between the time the crust was created and the time that the slab will be subducted (Fig. 4).
To calculate the output flux of Fe 3+ to the Mariana subduction system, we consider the magmatic output along the arc and back-arc. For the volcanic arc, we use an average primary melt Fe 3+ /ΣFe ratio of 0.244 and FeO* of 9.1 wt% from Mariana arc basaltic melt inclusions and submarine glasses [Brounce et al., in review]. For the back-arc, we use an average primary melt Fe 3+ /ΣFe ratio of 0.172 and FeO* of 8.13 wt% from Mariana trough submarine glass [Brounce et al., in review;]. The output flux can be calculated then using the equation, where P j is the production rate of crust in setting j , ρ j is the density of crust in setting j, and prior to the addition of slab fluids). For considering the fate of oxidized signatures in slabs through the subduction system, quantifying the contribution solely from the subducted slab is key. To do this, we assume that, outside of the mantle wedge that is influenced by subduction, the oceanic upper mantle is homogeneous with respect to fO 2 and that primary melts of this mantle have an Fe 3+ /ΣFe ratio = 0.14 independent of the extent of melting . We also assume that pressure of melting is the dominant control on the FeO* content of a primary mantle melt , so we assign arc and back-arc primary melts the same FeO* contents as those calculated above for arc (9.1 wt%) and back-arc (8.13 wt%) primary melts. Note that we do not consider Fe 3+ a conserved element, nor do these calculations necessitate that Fe itself is mobile during subduction. Rather, we are taking advantage of the observation that Fe speciation will respond to the transport of any multi-valent elements that may be mobile in slab-derived fluids or melts (e.g., S, C, V) and may thus trace the transport of oxidized slab materials in the subduction system. The mantle contribution for arc magmas then is between 3-4 O 2 (kg/yr/cm arc length) and between 8-10 Fe 3+ (kg/yr/cm arc length), and between 8-18 O 2 (kg/yr/cm arc length) and 20-44 Fe 3+ (kg/yr/cm arc length) for the back-arc, which when combined, represent a significant proportion (~72%) of the total output flux in the Mariana subduction system (Fig. 4). Independent of the mantle contribution, the output flux for arc and back-arc magmatism combined is 4.  ] and fluid seeps . Constraining the mass flux in this area is difficult because of the heterogeneity of materials coming out (fluids, mud, serpentinized xenolith blocks), the lack of redox information for most of these samples, and the uncertainty in how pervasive this activity may be in the forearc, with both depth in the trench and along strike. These uncertainties make it very difficult at this time to constrain an output flux for the forearc.
Second, it is possible that some or all of the excess Fe 3+ or O 2 is transferred from the slab to the mantle wedge and remains there, for example, by serpentinizing mantle wedge peridotite. This is similarly difficult to assess because of the lack of a comprehensive understanding of the mechanisms for material transfer from the slab to the volcanic arc. However, serpentine minerals become unstable at ~600°C  and the wedge can reach temperatures >1300°C , so it is unlikely that a significant portion of the wedge remains serpentinized.
Third, it is possible that some or all of the excess is retained in the subducting slab when it is subducted past the Mariana subduction system and into the deep mantle. In this case, the excess oxidized signature would suffer the same fate(s) of deep subducted slabs, of which there are many possibilities. They may be subducted to the transition zone and/or the core mantle boundary, where they potentially contribute to the sources for mantle plumes. In this scenario, we would predict that EM-I, EM-II, or HIMU style plumes may have some portion of the return flux of the excess oxidized signatures from subducting slabs, because the isotopic signatures of these endmembers are hypothesized to be derived from ancient subducted slabs [e.g., ].

Conclusions
We present the Fe 3+ /ΣFe ratios of sediments and variably altered oceanic crust   The Fe 3+ /ΣFe ratios of materials in this study (brown, green, orange, and blue circles) versus their position in the drill core. The x symbols represent the Fe 3+ /ΣFe ratios determined on materials from the same core by , for comparison. The white diamonds show the Fe 3+ /ΣFe ratios of two pristine MORB glass chips, determined via Fe-XANES [Brounce et al., in prep]. A simplified schematic of materials recovered during drilling at Site 801 during legs 129 and 185 is shown to the right.      6. Choose a file name, type it in the "File Name" field in x26a_flyscan and hit UPDATE. The predicted map time will be shown at the bottom of the x26a flyscan window. If you wish to shorten or lengthen this time, change the value in the "Pixel Size" field and hit enter. This value will be copied to " Step Size" directly underneath the "Pixel Size" field. Do not edit the 2 nd step size window.
Although this may vary, the target map collection time is 1-3.5 minutes.
7. When you are ready to map, hit "SCAN".
Every row creates its own file in your directory, every row should be the same file size. All rows for one sample name will be compiled into one folder within the current beam directory, labeled with the sample name that you chose to enter in x26a_flyscan.

When are you finished mapping and ready to continue with XANES analyses, go to x26a_xmap 12 Detector Control window and change Collection Mode to "MCA spectra".
To view concentration maps: 11. Calculate absolute motor positions for each of the three chosen points by adding the relative x and relative y motor positions of each point to the absolute x and y starting motor position for the concentration map.
12. Move to these absolute motor positions by manually entering the motor positions into the motor widget. Save each point as you move to it by using stage_widget in the normal fashion.

Section D: Trouble shooting
The x-ray beam dumps regularly at 0700 and 1900 EST. You may request to delay the beam dump by up to 5 minutes by calling the control room and asking for more time very politely.
The x-ray beam may also dump unexpectedly at any time of day without warning.
This frequently happens during electrical storms, very windy days, or during winter storm conditions (although it has also happened on beautiful sunny days in May). It may take minutes to days to resolve the problem. Refer to Beam TV for updated information. It is rarely a good idea to call the control room for information during these times.
As with any complicated experimental technique, both software and hardware malfunctions may take place and these may put both manual and automated data collection in jeopardy. In most cases, it is very obvious when a problem has occurred.
Common problems: -one or more IDL sessions may be "frozen" 1. It is likely that this IDL session must be killed and restarted. If this occurs for widgets that are hardware associated (e.g. Motor control XYZ) it may only be necessary to kill that particular session and restart, without concern over altering *.sv file information. If this occurs for stage_widget, be sure to check if previously saved positions are still saved when stage_widget is rebooted. If this occurs for the xanes_scan window, take caution when re-loading *.sv file info. Remember to always hit enter in every field and check all scan parameters for accuracy.
-beam may be available, but no scan activity occurs 1. This commonly results in a wide variety of consequences. All IDL sessions may have to be killed and restarted.
-data quality over one or more detectors may be poor 3. This may be due to poor positioning of saved points.
This is a risk when points are saved near cracks or edges. Move the data point 5 -10 microns further from the crack or edge and re-collect the spectra.
4. This may be due to detector error (all detectors).
Detector errors appear to be random and affect 1-10 data points per 12 hour session. There has been no satisfactory solution proposed for this. It is advised to check all data as it is collected to be sure all samples have three acceptable spectra collected for it.
5. This may be due to detector error (less than 11 detectors). This appears to be random as well but occurs rarely. In mu_beta_plot it is possible to select which detectors to use when stacking spectra. Unselect the detector(s) that are have collected poor quality data.
-collected spectra may be of acceptable quality, but is clearly not the expected "glass-like" shape 1. This may be due to poor positioning of saved points. If this occurs for a glass chip, be sure that the correct photos are being used to choose points. Check the z position of the motor carefully. Move the position 5-10 microns from the original spot and re-collect the spectra.
If this occurs for a melt inclusion, re-map the inclusion and check the line spectra carefully. If the problem persists, the inclusion must be thinned to allow the beam to pass through the glassy part of the inclusion without interacting with the phenocryst host. If the problem continues to persist, the inclusion is too small to be analyzed by XANES.
2. This may be due to a motor position error. Re-check the saved motor positions. If the problem persists, there may be a serious motor malfunction. Call a beam technician for assistance.
-the stage motor may be moving in abnormally small increments or abnormally slowly IDL will restart scanning where it last stopped. This abort scan procedure does not work very consistently in this case as well.
Kill IDL *Note: you should ALWAYS try the "Abort Scan" instructions above before attempting this procedure. This will exit all beam control windows and requires careful attention to detail.
1. Exit the IDL session window for xanes_scan by clicking on the "x" in the upper right hand corner.
2. Exit all other IDL session windows in the same manner.

Restart IDL
*Note: Be very careful to follow all steps in the order that they are listed here.

A. Department of Mineral Sciences, Smithsonian Institution
The instrument is located in room E-424, near the freight elevator and Exhibit Hall exit. The Nicolet 6700 is the newer instrument that sits on the right hand bench. The Excalibur is the older instrument and sits along the back wall. These instructions pertain exclusively to the Nicolet 6700.
Allen wrenches, stage holders, calibration equipment, and spare bulbs are housed in the wooden case next to the instrument or in the drawers of the desk that the instrument sits on.
The instrument is usually kept "on". If there are no green lights on the display panel of the instrument, there may have been some unusual service. Contact Tim Rose before proceeding.

B. Kelley Laboratory, University of Rhode Island
The instrument is located in CACS101. The FTIR is the instrument that sits on the table top to the right hand side of the room.
Allen wrenches, stage holders, calibration equipment, and spare bulbs are housed in drawers of the desk that the computer for this instrument sits on.
The instrument is usually kept "on". If there are no blue lights on the display panel of the instrument, there may have been some unusual event that requires service. Contact Katie Kelley.

The main bench (SI/URI):
The main bench is the large box on the right of the microscope. There are two blue lights, one that will blink while the IR source is on and one that displays that the instrument itself is on. There is an on/off switch to the left of this that will power on/power down the instrument. On the right hand side of the bench on the URI instrument, there is a large knob -this is the ATR crystal where reflection bench work can be done (solids, liquids). There are two doors on the front that can be liftedinside is the sample compartment for the main bench.

The microscope (SI/URI):
The microscope is attached to the left hand side of the instrument. It has motorized stage movement that can be controlled by using the joystick on the desk. The focus knob is also motorized. Rotating the mechanical stop will change the z-position of the stage rapidly, rotating the knob in small degree increments will change the z-position of the stage slowly. Rotating away from your body will move the stage up, rotating towards your body will move the stage down. The SI microscope has both a 15x and 32x objective. These objectives can be changed by simply rotating the objective mount (do NOT grab the objective to rotate, grab the neck that the objectives are mounted to). Be sure that the sub-stage condenser is the appropriate condenser for the objective chosen (e.g. the 15x condenser should be used with the 15x objective The extra condenser should be in the room, likely in a nondescript cardboard box. To change the condenser, move the stage as high as is safe to do so. On the right hand side of the condenser, there is a small set screw. Loosen this screw and the condenser should become removable. Replace the condenser, tighten the set screw, and return the stage to its normal position.
To focus the condenser, there is a knob to the right and below the stage. With the aperture light on, turn this knob to focus the light cone to z-position of the stage.
There is both reflected and transmitted light options on the microscope. On the left hand side, at the base of the microscope, there are two knobs. The front-most knob turns on and controls the intensity of transmitted light; the second knob turns on and controls the intensity of reflected light. The third knob turns on and controls the light passing through the aperture of the instrument. Note: any analysis performed on the instrument is independent of your control of the knob for these light sources (i.e. there is no configuration that you must return the light sources to in order to successfully collect high quality data).
There is an upper and lower polarizer on both the SI and URI microscopes for viewing in cross-polarized light. The upper polarizer is removable and is located just under the oculars, pointing at your body. The lower polarizer is rotatable and also removablelocated on the left hand side of the lower base of the microscope towards the back of the instrument.
The camera that provides the live video feed to the OMNIC/Atlus software is on the top of the microscope-viewing column. If the video feed is out of focus, on the right hand side of the camera there are two allen bolts, one marked "Focus" and one marked "Lock". First loosen the bolt marked "Lock" and then turn the "Focus" bolt until the image on the live video feed is in focus again.
The two buttons at the base of the microscope pointing towards your body control the sampling mode (transmission or reflectance) and which detector (position 1 or position 2) is to be used for the experiment. There is also a means to control this through the OMNIC/Atlus software.
On the right hand base of the microscope there are 3 small knobs in addition to the large knob that controls the aperture window size and rotation angle. There is also a means to control this through the OMNIC/Atlus software. This choice is at the user's discretion, however it is recommended that the user choose the software option.

Purge gas (SI):
On the wall behind the instrument, there are three gas flow meters that turn on/off/control the supply of building supplied gases to the equipment in the room. The two left most meters control gas flow into the older instrument in the room. The meter on the right hand side controls gas flow to the Nicolet 6700.

Purge gas (URI):
To the left of the table that the FTIR sits on there is a purge gas generator. A large on/off switch is located on the front of the instrument. A column with indicator paper is also located on the front of the instrument, which will indicate when the instrument (and the air it is exhausting) is dry. The generator should typically run for ~1 hour before connecting the air exhaust to the FTIR microscope/bench. There is a water vent in the back of the generator with a small length of tube that will drip water (sometimes violently) that is condensed from the air inside of the generator. The small tube should always sit inside of a container, the level of water in the container should be checked regularly. A black tube comes out of "outlet" connector and has a 3 way split. Two tubes come out of this and connect (1) to the "purge in" port on the microscope and (2) to the back of the main bench.

Section 2. Starting a run
For the SI instrument, use the calendar on top of the bookshelf to schedule yourself for a day in the lab to be sure to avoid any conflicts. Underneath of this bar there is a drop down menu that says Experiment: Default -Transmission (default.exp). The next row contains several "quick buttons" that are shortcuts to software navigation options that can also be found in the top menu bar.

Section 3. Experiment Set up (SI/URI)
Click the "Expt Set" quick button, or choose Collect > Experiment Setup from the top menu bar to set the experiment parameters.
The Experiment Set up window houses all of the software controls for the FTIR instrument such as whether your sample will be in the bench or on the microscope stage, which detector you wish to use, ect… The second tab is labelled "Bench". The total interferogram, summed from the interferograms of each wavelength of IR light, as measured by the detector is shown here. If the condenser is focused properly, the Peak-to-Peak value for the interferogram should read ~10-12 V for a ~20x20 aperture through air. In this tab you must also indicate whether you will be using the bench or the microscope detector (the top position in the list is "Position 2" on instrument), beamsplitter (KBr), source (IR or turbo-IR; turbo-IR will increase sensitivity by increasing the intensity of IR light), region of interest, ect… The following figure shows a typical experimental set up for analyzing hydrous basaltic glasses.
Typically, one sample should be analyzed at three distinct locations within the sample.
A background spectra is taken and subtracted from each sample spectra in order to subtract out any spectral contributions from the fluxing gas in the instrument or within the cache.
After the experiment is set up appropriately, navigate to the first sample and focus on the surface in reflected light. Choose an appropriate aperture size either by using the knobs at the base of the microscope on the right hand side or by using the OMNIC/Atlus software. To do this, in the top menu bar choose Atlus > Apertures Dimensions. A window will appear that allows you to set the X and Y dimensions as well as the rotation angle. Always check the microscope and re-focus the condenser every time the aperture dimensions change or the stage position has changed. If there appears to be an offset between what the software is registering and the actual microscope image, you can re-initialize the aperture by going to Atlus > system configuration. Under Microscope, there is a button called "Aperture". Click this. After about 60 seconds, the problem should be fixed.
Without changing the stage focus or aperture dimensions, move the blank space on the sample mount. Refocus the condenser. In Experiment Setup, under the Collect tab, in the section called "Background Handling", click the radio knob for "Collect background before every sample". Click OK.
To collect background spectra, use the quick button "Collect Bkg". The software may/may not prompt you to prepare to collect a background spectrum. If it does, just hit OK. The background will begin to collect. You may stop data collection at any time by clicking "Pause", located in the lower left hand corner. Close the window and choose "No" to the dialogue box asking if you'd like to add the spectra to a window.
During collection, there will be a drop down bar across the top of your actively collecting spectra window, with options of adding the spectra that you are collecting to certain windows. If there are no windows currently open, the only choice will be to "add to a new window". If there are other windows open, you will be given the option to add it to any of those windows. You have the entire length of the data collection to choose which window to add the spectra to but you must choose before collection ends.
When collection ends, a box will appear, confirming that you wish to add this spectra to whichever window you chose during collection from the drop down menu. If you hit "Yes", the spectra will appear in the window that you indicated. If you hit "No", the spectra go away and the data are lost. If this happens, you must recollect the spectra. The addition of these spectra to any window has no implications for the format or way in which the data can be ultimately saved. Even if the window chosen is not your ideal window, it is best to just choose "Yes" so that your data does not disappear.
Save the background spectra by choosing File > Save As > navigate to your folder under C:/My Documents, choose a file name and save as a *.spa. If you have several spectra open in the same window, the drop down menu above the chart will allow you to select the spectra that you wish to save. As a general rule, the spectra currently plotted in the color red is the currently active spectra upon which any action will be taken.
To apply this background file to the subsequent sample spectra, go to Experiment Setup, under the Collect tab, in the section called "Background Handling", click the radio knob for "Use specified background file". The "Browse" button will become active. Click this and navigate to the background *.spa file that you just saved. Click OK in Experiment Setup.
Navigate back to your sample, choose your spot and be sure that the stage and condenser focus as well as the aperture dimensions are all still appropriate. If any of these things are changed at this time, you must recollect a background file. To collect a single point sample spectra, click the quick button "Col Smp" or go Collect > Collect Sample. Hit OK to the title of the spectra (this has no bearing on the name of the file that you will write the data to). The software may or may not prompt you to prepare to begin to collect sample spectra. If it does, hit OK. As with the background collection, a drop down menu will be available for you to choose a window in which to add the spectra that you are currently collecting. When the collection is over, go to File > Save as > navigate to your folder and save two copies of your sample spectra, one as a *.spa and one as a *.csv.

Section 5. Introduction to Atlus software (SI/URI)
To access the Atlus imaging tools, choose Atlus > Show Atlus window from the top menu bar. If nothing happens, Atlus is probably already open. Minimize the experiment window within the larger Omnic window and look around: The Atlus screen will look like this:

321
The left hand image can house a static image or image mosaic that you create. The right hand image is the live video feed. See Section 1 for details on how to adjust the live video feed focus.
In Atlus, you may create a static image by drawing a box in the left hand image and choosing Atlus > capture mosaic. Be aware that large mosaic images take a long time to make. The best approach is to click on the "Full Range View" quick button at the bottom of the Atlus window and to navigate at the microscope to the sample mount.
When you know that you are in the general area, you can click on the square quick button (third button from the left) and draw and small square on the order of the size of the red cross hairs that mark your location. A map of this size will take ~2 minutes to collect. When you are finished with this composite image, go to Atlus > clear You may navigate around the stage by clicking on the "Stage movement tool" quick button and clicking on the right or left hand image. On the right hand image, you may navigate by clicking on the red aperture and dropping it in a new location (with the arrow quick button chosen).
If you manually choose points and analyze in a one-by-one fashion, you may lay markers down on the left hand image by clicking on the flag button and dropping the marker where you would like a marker to appear. This image can be saved by choosing: Edit > Copy Video Image and pasting into the Paint application.
On the right hand image, you may drop individual data points: After changing any numbers in the text fields, hit "Update" and time estimation will appear.

SAME SIZE AS YOUR APERTURE WINDOW.
For instance, a map was made of a glass inclusion ~70 ums in diameter with a 20x20 um aperture and 5x5 um stage motor steps. In the map, clicking on the center pixel retrieves a baseline corrected spectra that yields a total intensity of the 3530 cm-1 peak of 1.501. Then, a single point with a 20x20 um aperture was manually centered on the inclusion and one spectra collected. This baseline corrected spectra yields a total intensity of the 3530 cm-1 peak of 1.507. Whether a mathematical construct can be performed within the Atlus or Omnic software to yield higher resolution has yet to be determined.