Geochemical Constraints on Mantle Sources and Melting Conditions in Pacific Back-Arc Basins

The effect of water in the mantle has been well studied and has well-known effects on the behavior and properties of magmas and the mantle including an increase in the extent of melting, suppression of plagioclase during crystallization, and a general reduction of seismic parameters and viscosity. At mid-ocean ridges, magmatic and mantle H2O contents are relatively low, and a reasonable understanding of the behavior of H2O has been obtained (Dixon and Stolper, 1995; Dixon et al., 1995; Dixon et al., 2002; Asimow and Langmuir, 2003). However, at back-arc basins, H2O can also be added to the mantle source by the subducting slab, changing the melting behavior and the mantle source composition (Stolper and Newman, 1994; Taylor and Martinez, 2003). Quantitative constraints on these factors lag behind, and this thesis will test hypotheses related to the role of volatiles in these three major processes at back arcs: (1) tracing mantle source compositions and flow vectors, (2) refining mantle melting models, and (3) constraining the origin of back-arc slab-derived fluids. Tracing mantle source compositions is done best in places where mantles of starkly contrasting compositions are juxtaposed, as in the case of plume-ridge interaction. The NW Lau Basin, a back-arc with little influence from the slab, provides an ideal setting to address mantle flow where the mantle source contrast is potentially well made with the interaction of relatively depleted mantle with the Samoan plume. Geochemical tests of the interaction between the Samoan Plume and the Lau Basin mantle have relied on one tracer ( 3 He/ 4 He), but the addition of volatiles (H2O, CO2), trace elements (e.g., La, Nb), and other radiogenic isotopes (Sr, Nd, Pb, Hf) provides further constraints on tracing the enriched Samoan mantle composition. Our new data suggest two-component mixing of MORB-like mantle with an enriched mantle source, similar to Samoa, although consideration of a complete regional data set suggests there may be other sources of heterogeneity in the mantle beneath NW Lau. Aside from tracing mantle flow, volatiles and trace elements provide constraints on mantle melting as H2O has an effect on where and how much melt can be made in the mantle, recorded in incompatible trace element signatures (e.g., Ti, Nb). Observations based on geochemical data suggest two possibilities: mixing of end-member melts or a continuous melting regime, but most models of mantle melting are restricted to isobaric-isothermal conditions and offer unrealistic tests of the competing hypotheses of back-arc magma generation. We developed a robust adiabatic, hydrous melting model and combined with a well-constrained mantle source composition, we model back-arc magma generation. The release of slab fluids is also an important part of subduction systems, as the fluid composition and the extent of its addition to back-arc mantle sources affects enrichment of resultant basalts. The composition of slab fluids reaching back-arc basins will differ from arc fluids depending on the pathway traveled by the fluid/melt, and the conditions of their release from the plate. Combining geochemical data with recent geochemical models of slab conditions (e.g., H2O/Ce) and geodynamic models for slab surface temperatures (SST) at each subduction zone provides a robust test of the depth origin of back-arc slab-derived fluids. Average SSTs for these global back arc basin spreading segments, referenced to 4 GPa, range from ~775-1000°C, hotter on average than global arc SSTs (730-850°C), suggesting that back-arc basin fluids originate at warmer temperatures than their respective arcs.


LIST OF TABLES
.S1 New volatile and trace element data for samples

INTRODUCTION
Water is the central component that distinguishes back-arc spreading ridges from normal mid-ocean ridges. Water is known to have a strong influence on melting within the mantle, yet the role that water plays is dependent on the tectonic setting (e.g. Danyushevksy et al., 1993). Specifically, the addition of water from the subducted plate to back-arc basin magma sources strongly influences mantle melting, magmatic crystallization processes, and mantle source composition (Stolper and Newman, 1994;Taylor andMartinez, 2003, Sisson andGrove, 1993). This thesis will develop a comprehensive picture of the specific roles of water and other volatiles in (1) identifying mantle heterogeneity as a tracer of mantle flow beneath back-arc spreading centers, (2) influencing mantle melting along realistic adiabatic ascent paths, and (3)  environments. Northeast of the Tonga-Kermadec subduction system is the Samoan plume, which is hypothesized to be migrating into the NW Lau basin through a tear in the subducting plate. Previous studies, using He isotopes as a tracer of the distinctive Samoan mantle composition, show the extent to which the plume may have infiltrated into the back-arc basin Hilton et al., 1993;Turner and Hawkesworth, 1998;Lupton et al., 2009). MORBs have a relatively homogeneous He isotope ratio (~8 R a ; where R a is the isotope ratio normalized to the atmospheric ratio), but Samoa is one of the global highs in He isotopes (up to 33 R a ). Elevated 3 He/ 4 He ratios in the NW Lau Basin are classically interpreted as evidence of the Samoan plume leaking through the plate boundary Hilton et al., 1993;Turner and Hawkesworth, 1998;Lupton et al., 2009). If the elevated 3 He/ 4 He signatures are indicative of the Samoan plume influencing the NW Lau basin, there should be additional indicators such as: 1) clear mixing and correlation between 3 He/ 4 He and trace elements and other isotopes and 2) elevated mantle temperature.  first looked at He and Sr isotopes in a very limited set of samples from the northern Lau Basin. Samples in the Rochambeau Bank region were found to have both higher He isotope ratios and enriched Sr isotope signatures, suggestive of mixing between enriched mantle in Samoa and depleted mantle in Rochambeau Bank. The Samoan plume is proposed to be drawn into the Lau Basin by mantle flow influenced by crustal extension at the spreading centers. Turner and Hawkesworth (1998) investigated this hypothesis following previous work Hilton et al., 1993;Turner and Hawkesworth, 1997), by compiling a greater, but still limited, data set. Samples were analyzed for He, Sr, Nd, Pb isotopes and the enriched He isotopic signature of Lau Basin lavas was taken as conclusive evidence for the presence of the Samoan plume down to the Peggy Ridge. Lupton et al., (2009)   There are many models concerning the relationship between H 2 O and melt fraction (F) at back-arc basins. One model from  suggests that back-arc lavas record mixing trends between a dry, MORB-like melt and a wet, arc-like melt, while another model propose that basalts record a continuum between wet and dry melting . These two processes have different consequences for melt compositions, but studies attempting to resolve between these models have been limited by poor data coverage and the lack of a realistic hydrous melting model to constrain the melting process. While the Mariana Trough and East Scotia Ridge have been well studied (Stolper and Newman, 1994;, there are three back-arc basins (N. Fiji, Lau, Manus) in the Western Pacific that are less well investigated. Although these basins have been well-sampled, the central role of volatiles has never been comprehensively addressed in these regions (e.g. Perfit et al., 1987;Johnson et al., 1987. Using glass from BABBs from N. Fiji, Lau, and Manus Basins, the collection of major, trace element and volatile data, provides the first comprehensive data set for each of the three basins. With these new data, the role that the subducting slab and fluid components play in back-arc basin melting can be constrained. Determination of the influence of water on the liquid lines of descent (LLD), specifically plagioclase and clinopyroxene fractionation, allows for more accurate corrections of the data back to equilibrium with the mantle at Forsterite 90 (Fo 90 ), which is essential for constraining mantle melting. Combining the data sets with modeled LLDs (Petrolog3; Danyushevsky and Plechov, 2011) provides necessary additional constraints on the influence of water on melt composition during crystallization. Using the model-and data-based LLDs to project compositions back to equilibrium with the mantle, constraints on the influence of volatiles on mantle melting were determined.
Using the tighter constrains on mantle equilibrium (Fo 90 ) composition provided by the LLDs, a more accurate constraint on source composition and extent of melting was obtained from trace elements. Well constrained values for titanium source concentrations ( ) are important because is often used to estimate the melt fraction (F) for lavas . Better estimates for and F reduce source concentration errors and better constrain an accurate value for F, an important aspect for modeling mantle melting. Langmuir et al. (1992) developed an adiabatic melting model for a typical MOR scenario, but does not account for a more hydrous mantle as found in back-arc basin settings. A hydrous back-arc melting model was developed by Kelley et al., (2010), and while providing estimates for hydrous melting, the models are isobaric and isothermal, which does not account for changing water concentrations or realistic adiabatic melting paths. Combining the approaches of Langmuir et al. (1992) with Kelley et al., (2010) produces a well constrained, realistic polybaric hydrous back-arc basin melting model against which the inversion of natural melt compositions are tested to resolve the competing hypotheses of mixing vs. melting processes to generate back-arc basin basalt compositions.
There remain many questions about the relationship between water and trace elements in back-arc basin basalts, especially concerning the compositions of the subducted inputs, the effect of dehydration of the subducting slab, and fluid pathways through the mantle wedge. Chapter 3 focuses on how these factors combine to create the fluids that modify back-arc mantle sources, by taking an integrated modeling approach, combining new geochemical data (e.g., H 2 O/Ce; ) with petrological modeling (e.g., Stolper and Newman, 1994) and geodynamic models for each back-arc basin (subduction zone geometry and thermal structure Syracuse and Abers, 2006;Syracuse et al., 2010). Using the global back-arc data set generated as described in Chapter 2, the pressure and temperature conditions of slab dehydration are constrained using the H 2 O/Ce ratio of the fluid, which is a sensor of slab surface temperature. Back-arc lavas have higher H 2 O/Ce ratios than the average MORB or plume H 2 O/Ce ratio of ~200 ) because of additions from the subducting plate to the back-arc source, but the relative contributions of H 2 O and Ce from the mantle and the slab must be resolved in order to apply the H 2 O/Ce thermometer. The mantle contribution to each basalt were separated out from the fluid component using Nb-Ce systematics (Cooper et al., 2012), and the H 2 O/Ce ratio of the fluids were translated into slab surface temperature using the model of Cooper et al., 2012. Armed with temperature constraints, these are related to the depth of origin of the back-arc fluid by comparison with computational models of slab thermal structure (e.g. Syracuse et al., 2010).
However, these models do not provide absolute constraints on exact positions of the fluid origin, instead providing a guide with enough resolution to test the first-order question of whether the fluids come from deep or shallow, and the relative contrast of conditions among the various back-arcs.

Introduction
Tracing long-term movements of the Earth's mantle through direct observation of the Earth itself is tremendously difficult, given the enormous contrast in time scales over which tectonic/dynamic motions occur relative to the brief periods over which observations may be made. Studies of shear wave splitting provide characterizations of mantle anisotropy and flow, as olivine a-axes align with mantle flow vectors [e.g., Silver and Chan, 1991;Zhang and Karato, 1995]. Beneath volcanic arcs, shear wave splitting studies suggest a range of potential vectors, from arc-normal to arc-parallel [e.g., Russo and Silver, 1994;Smith et al., 2001;Conder and Wiens, 2007], challenging conventional ideas about the coupling of plate and mantle wedge flow.
Geochemistry provides another tool to image mantle flow, particularly in places where mantle sources of contrasting geochemical characteristics are juxtaposed.
Using geochemistry to trace mantle flow has been successfully applied in the Mariana subduction system, where four primary contributions to the mantle source have been identified using Nb/Ta and Ta/Yb ratios, including a depleted mantle asthenosphere and enriched lithosphere that are modified by at least two distinct subduction-derived components . Additionally, while Pb isotopes distinguish two different mantle domains (Pacific and Indian) in the southwest Pacific, Pb mobility during subduction makes tracing the flow of these distinct mantle domains challenging [see Heyworth et al., 2011]. The use of less mobile Hf-Nd isotopes provides alternative geochemical tracers of the two mantle domains, showing the influx of the Indian mantle into the Fiji Islands and the N. Fiji and Lau Basins [Pearce et al., 2007].
The Lau Basin has been a particular focal point for efforts to trace mantle domains and movements using geochemistry. Extensive work at the Eastern Lau Spreading Center (ELSC), has shown a regional connection between decreasing subduction influence in the back-arc with increasing distance from the Tonga Arc ]. The ELSC also shows evidence of mixing between a number of mantle components, including Indian-like mantle and an enriched mid-ocean ridge basalt (MORB) mantle, with materials derived from the subducted slab [Bezos et al., 2009;Escrig et al., 2009]. The NW Lau Basin, on the other hand, comprises several spreading centers and rift zones that are located far west of the active subduction zone and are not likely to be influenced by modern subduction. The elevated He isotopic signature of the nearby Samoan Plume has long been viewed as a diagnostic tracer of enriched Samoan mantle, and following this rationale, basalts erupted in the NW Lau Basin with 3 He/ 4 He ratios higher than MORB suggest influence from a Samoa-like source in the mantle beneath NW Lau [e.g., Hawkins and Melchior, 1985;Poreda, 1985;Wright and White, 1987;Farley et al., 1992;Lupton et al., 2009]. Reliance on a singular geochemical tracer as evidence of the regional movement of Samoan mantle however, leaves room for doubt about the origin of the elevated 3 He/ 4 He isotope ratios in the Lau Basin, particularly since other western Pacific back-arc basins have similarly elevated 3 He/ 4 He in the absence of a known nearby mantle plume [e.g., Manus basin; . Do the Lau helium isotopes correlate with other geochemical signatures of Samoan mantle? Is there corroborative evidence, such as elevated mantle temperature, of hot spot influence beneath NW Lau? These questions must be addressed before the presence or scale of Samoan mantle migration into the NW Lau Basin can be clearly resolved.
With this work, we aim to address these outstanding questions with new geochemical data for major and trace elements, dissolved volatiles, and radiogenic isotopes for a suite of glasses from a high-density sampling of spreading centers and rift zones in the NW Lau Basin. These samples have been previously analyzed for 3 He/ 4 He ratios [Lupton et al., 2009], and our work thus adds essential geochemical constraints on the genesis of NW Lau magmas that we use to test the identity of the enriched He signature, both through trace element/isotopic systematics and their relationships (or lack thereof) to He isotopes, and through petrologic modeling of mantle melting conditions beneath NW Lau spreading centers. We show that, surprisingly, He isotopes do not correlate with trace element or isotopic signatures in NW Lau, and that a portion of the mantle signature in this region, though elevated in 3 He/ 4 He, is otherwise uncharacteristic of Samoa in its trace element and isotopic composition. Moreover, mantle temperature in the region, although high, does not decrease towards the south as it would if hot mantle were infiltrating from the north.
We explore the consequences of these observations for models of mantle flow in this complex region, and present alternative hypotheses that could explain the data.

Tectonic Setting
The Tonga-Lau system is an oceanic subduction zone in the southwest Pacific, where the Pacific Plate subducts beneath the Indo-Australian Plate (Figure 1.1).
Behind the Tonga Arc, back-arc spreading initiated at ~6 Ma in the Lau Basin , which is a V-shaped basin with several actively spreading segments that impinge on the Tonga Arc towards the south. The rates of both plate convergence and back-arc spreading are highest at the north end of the subduction zone , which exhibits the fastest back-arc opening on Earth, spreading at a rate of 160 mm/yr, decreasing southwards to rates of 60 mm/yr [e.g., ; see Figure 1.1]. At the northern end of the Tonga Arc, the plate boundary bends 90° and the Pacific Plate ceases to subduct. This northern boundary of the system is the Vitiaz Lineament, interpreted to be a paleo-subduction zone that is now a transform boundary separating the Pacific and Indo-Australian plates , and is likely the locus of a tear in the Pacific Plate at depth [Millen and Hamburger, 1998]. To the northeast of Tonga lies Samoa (Figure 1.1), an ocean island hot spot with an age-progressive volcanic chain on the Pacific Plate [e.g., Hart et al., 2004;Koppers et al., 2008] that may be the surface expression of a deep-rooted mantle plume [e.g., Montelli et al., 2004;Jackson et al., 2007a].

Prior Work
Several previous studies have attempted to assess the nature and extent of interactions between Samoan mantle and the Lau Basin through a tear in the Pacific Plate at the northern end of the Tonga/Lau system. Gill and Whelan [1989] first used Nd isotopes to show that enriched Ocean Island basalt (OIB) source mantle had reached Fiji by 3 Ma.  first analyzed He and Sr isotopes in a small population of samples from the Rochambeau Bank (RB), a shallow submarine volcanic center near the Vitiaz Lineament (note that RB is distinct from the region of rifting, Rochambeau Rifts [RR], immediately to the east; Figure 1.1b-c). They showed that some samples from RB have elevated 3 He/ 4 He ratios (up to 22 R a ; where R = 3 He/ 4 He and R a = R air = 1.39 * 10 -6 ) and enriched Sr isotopic signatures relative to lavas from further south in the Lau Basin that are more typical of normal mid-ocean ridges ( 3 He/ 4 He = 8-10 R a ; Peggy Ridge [PR], Central Lau Spreading Center [CLSC]).
Because Samoan lavas can have very high 3 He/ 4 He ratios [up to 33.8 Ra;Jackson et al., 2007b] and very radiogenic 87 Sr/ 86 Sr, the elevated He and Sr isotope ratios at RB were used as evidence of mixing between enriched Samoan mantle, which was proposed to be drawn into the Lau Basin by mantle flow influenced by crustal extension at the spreading centers, and "ambient" depleted mantle beneath RB .
Further work [e.g. Hilton et al., 1993;Turner and Hawkesworth, 1998;Lupton et al., 2009] has refined these initial hypotheses, although until very recently, none have involved a focused regional survey and comprehensive sampling of spreading centers in northwestern Lau. Turner and Hawkesworth [1998] reviewed data for the northern Lau Basin, including Niuafo'ou island (NF) and the Mangatolu Triple Junction (MTJ; Figure 1b), concluding that elevated He isotope ratios were sufficient evidence for the presence of Samoan mantle beneath the Lau Basin despite radiogenic isotope signatures that were not conclusively related to Samoa. Lupton et al., [2009] analyzed 3 He/ 4 He ratios of 41 new glass samples from a comprehensive survey and sampling of spreading centers in the NW Lau Basin (a subset of the samples in this study). Their study found that all lavas north of the PR had 3 He/ 4 He ratios higher than MORB but, surprisingly, found no clear correlation between 3 He/ 4 He ratios and latitude or ridge morphology. Recent work by Tian et al. [2011] reported trace element and Sr-Nd radiogenic isotope data for northern Lau Basin samples from RB, PR, MTJ, and NF, including the RB samples from . Their results suggested a geochemically heterogeneous mantle source with both a subduction signature (in the east) and an enriched mantle source signature (in the west) from two Samoa-like end-members.  built upon this data set by analyzing volatiles and noble gases in these samples. Their results suggest hotspot influence in the NW Lau magmas. Moreover, further analyses of samples from the Lupton et al.
[2009] study were conducted for Ne isotopes [Lupton et al., 2012] and chalcophile elements . The Ne isotopes correlate with 3 He/ 4 He, providing additional evidence for influence from the Samoan Plume [Lupton et al., 2012]. The chalcophile element data showed Cu and Ag enrichment uncharacteristic of MORB or Samoan sources, suggesting the presence of an additional high-Cu mantle source in the region. The lack of a simple trend of decreasing 3 He/ 4 He in NW Lau with distance from the northern plate boundary [Lupton et al., 2009], coupled with a scarcity of supportive data for these samples (e.g., major, trace, volatile elements or radiogenic isotope ratios), raises questions about the simple hypothesis of plume migration into the Lau Basin mantle. With the present study, we provide new data to accompany the He isotopes for these samples and use these data to assess the identities of mantle sources beneath the Lau Basin and the consequences for interpretations of mantle flow in this region based on lava geochemistry.

Samples and Methods
The basaltic glass samples reported here are new samples from the Northern Lau Basin that were collected from 63 bottom dredges during voyage SS07/2008 of the R/V Southern Surveyor. Sample locations and rock descriptions including phenocryst information are given in Table 1.S1. The samples were analyzed for major elements by electron microprobe, trace elements by laser ablation and solution inductively coupled plasma mass spectrometry, volatiles by secondary ionization mass spectrometry, and radiogenic isotopes by multicollector inductively coupled plasma mass spectrometry (see supplementary information for more detailed information on sample selection and analysis methods).

Effects of Degassing
As magma ascends from the mantle to the surface, dissolved volatiles will exsolve into a vapor phase at low pressures, resulting in volatile loss from the melt.
Major volatile species (e.g., H 2 O, CO 2 ) have different vapor/melt solubilites, enabling an assessment of volatile loss from each glass. Carbon dioxide has lower solubility in silicate melt at low pressure and is expected to begin degassing before H 2 O [Dixon and Stolper, 1995], and the mixed CO 2   found to be vapor-oversaturated or saturated at the pressure of collection, which is typical of mid-ocean ridge basalts and reflects relatively fast transport and eruption of magma from mid-crustal depths .

Effects of Crystallization
Assessing the extent of H 2 O degassing is important because H 2 O influences magmatic crystallization and the liquid line of descent (LLD). Water suppresses plagioclase and clinopyroxene crystallization [e.g., Sisson and Grove, 1993a;Sisson and Grove, 1993b] and its effects can be seen in the major element systematics of the NW Lau basalts. The basalts can be segregated into two major groups on the basis of H 2 O content, which we reference here to H 2 O (8.0) (i.e., glass H 2 O concentration corrected for fractional crystallization to the equivalent concentration at 8 wt.% MgO; Table 1.S2), calculated using the expression of Taylor and Martinez [2003]. The differences in major element composition of the two end-member parental magmas likely reflect differences in melting processes and/or source composition.
Specifically, the wetter parent magma has higher Na 2 O and Al 2 O 3 , lower CaO, and a lower CaO/Al 2 O 3 ratio than the dry parent magma. Higher concentrations of incompatible elements (e.g., Na 2 O, Al 2 O 3 , H 2 O) could reflect lower extents of melting of more hydrous mantle (e.g., Langmuir et al., 1992;, although all incompatible elements should be affected similarly if this were the case, and no difference is required in TiO 2 , K 2 O, or P 2 O 5 for the parental magmas. The difference may instead reflect variation in the modal cpx content of the mantle source, which will tend to drive melt compositions to higher Na 2 O and Al 2 O 3 , and lower CaO/Al 2 O 3 ratio with increasing source fertility (i.e., higher modal cpx; Klein and Langmuir, 1987). It is important to note, however, that the parent magmas were chosen to bracket the majority of the data, which represent more of a compositional continuum rather than two discrete populations of melts.

Trace Element Variability
The NW Lau Basin basalts span a wide range of trace element enrichment, as shown on Figure 1

Isotope Variability
The Sr and Nd isotopes of two RR samples, NLD-07-01 and NLD-20-01, overlap the least radiogenic part of the Samoan field (Figure 1.5c). Tian et al. [2011] report Sr-Nd isotopes for two other similar samples from RB. These are the only basalts in the NW Lau Basin with radiogenic enough Sr to match Samoan basalts. Our samples, especially NLD-20-01, also overlap the Samoan field in Hf-Nd and Pb isotope space (Figure 1.5d and supplementary information). The Sr, Nd, Hf, and Pb isotopes of all other samples from both the RR and NWLSC are much more depleted than Samoa, with MORB-like ratios similar to the CLSC, although 3 He/ 4 He ratios in all of these samples range from 12-28 and show no correlation with any of the radiogenic isotopes ( Figure S5d-e).

Discussion
Trace element abundances and 3 He/ 4 He ratios show that the mantle beneath the NW Lau Basin is geochemically enriched relative to normal MORB or the proximal CLSC. Here, we explore the geospatial patterns of mantle enrichment, and the relationships between 3 He/ 4 He and these new data, in order to assess the identities and locations of distinct mantle components contributing to magmatism in the NW Lau Basin. In addition, we model the pressure and temperature conditions of mantle melting beneath NW Lau in order to test for a regional thermal gradation possibly associated with the Samoan Plume.

The spatial distribution of enrichment in NW Lau
The NW Lau samples span a wide range of enrichment, as shown on Figure   1.4, with a general trend of enrichment in incompatible trace elements broadly decreasing from the north (RR) to the south (CLSC). If Samoan material were simply migrating into the Lau Basin, the expectation would be that "Samoa-like" affinity or enrichment should decrease with latitude towards the south. Looking at the main tracer for Samoan affinity, the 3 He/ 4 He ratio, Lupton et al. [2009]

Constraining the influence of subduction on NW Lau mantle sources
One source of enrichment that could perturb correlations of trace elements with He isotopes is subduction, which is not known to significantly modify 3 He/ 4 He ratios of magmas [Poreda and Craig 1989], but can enrich light REE and Ba/La ratios by way of fluid and sediment melt additions to the mantle source. Although the NW Lau Basin is located ~530km from the active Tonga subduction zone, the basin is opening in a region of paleo-subduction and the possibility for lingering influence from the past subduction along the Vitiaz Lineament remains. We test for effects of subduction  [Miller et al. 1994]. The NW Lau basalts from RR and NWLSC have no significant negative Nb anomalies (Nb/Nb* ≥ 1) and Ce/Pb ratios ≥ 15, consistent with the range for normal MORB [Miller et al. 1994] and inconsistent with the trend among subduction-influenced basalts from the ELSC, which have both Nb/Nb* ≤ 1 and Ce/Pb in most samples < 15 ( Figure 1.7b). Based on these observations, we find no evidence for subduction influence on the composition of the mantle beneath the NW Lau Basin.

Geochemical characteristics of NW Lau mantle sources
Since we have ruled out subduction as a possible process for adding contaminants to the mantle source, and Figure 1  beneath NW Lau, with trace element characteristics similar to regional basalts erupted at the CLSC (both enriched and depleted), Samoa, and NFB.
The number of possible mantle components may be further refined using He, Sr, Nd, Pb, and Hf isotopes, which are particularly important geochemical tracers of mantle sources, as isotopes are insensitive to fractionation by melting and crystallization processes. Figure 1.5a [data from

Modeling Mantle Melting Conditions beneath the NW Lau Basin
Hotspots such as Samoa are characterized by elevated mantle potential temperature, [e.g., Putirka, 2008] resulting in higher extents of melting of an adiabatically upwelling mantle. Subduction influence within the mantle source of NW Lau appears largely absent, thus magma production in this region should therefore be a simple function of mantle potential temperature coupled with the inherent composition of the mantle source. Melting beneath mid-ocean spreading ridges is driven by adiabatic upwelling of the mantle over a range of mantle potential temperatures Langmuir, 1987, 1989]. If a hot plume were infiltrating the Lau Basin, however, we may also expect elevated mantle temperature in the region, possibly cooling off with distance from the plate boundary. Recent estimates of the Samoan mantle potential temperature place it at ~1720°C [Putirka et al., 2007], whereas previous constraints from the Northern Lau Basin indicated a regional mantle potential temperature of ~1460°C [Falloon et al., 1999] to ~1400°C ]. The seismic structure beneath the CLSC shows low velocity zones (V p and V p /V s ), interpreted as melt regions, located 30-50 km beneath the CLSC, consistent with a regional T p of ~ 1400°C [Conder and Wiens, 2006].
Accurate petrological constraints on mantle temperatures and melting conditions require that basalt compositions be corrected for the effects of fractional crystallization and referenced to a common point along the liquid line of descent (LLD). The LLD modeling done here (see section S.5) allows reconstruction of melts to their primary compositions by using the modeled fractionation slopes to project melt compositions back to the points of cpx-in and plag-in (Table 1.S6, Figure S4) before adding equilibrium olivine to each melt until it is in equilibrium with Fo 90 ( This is the classically-hypothesized component contributing elevated 3 He/ 4 He in the Lau basin. Our detailed sampling and analysis of regional volcanism in the NW Lau Basin reveals the influence of a minimum of two mantle sources beneath the NW Lau Basin, as originally proposed by Turner and Hawkesworth [1998], to explain the majority of the trace element and isotopic data reported here. Lead, strontium, and neodymium isotopes (Figure 1.5d) require two end-members: a Samoa-like enriched mantle source and a depleted "Indian" MORB-like mantle source. The first mantle source is similar to Samoa with respect to both trace elements and radiogenic isotopes, with elevated 3 He/ 4 He and radiogenic isotopic compositions similar to the Vai trend in Samoa (e.g., Figure 1. 5, 6, 8, and S5). The second mantle source is a depleted, CLSC-type MORB source, with LREE depletion and MORB-like radiogenic isotopes indicative of an origin in the Indian mantle domain (Figure 1.5, 6, S5).
Following the same approach for calculating (see section 4.4 above and supplement), we also calculate the source trace element compositions of the NW Lau lavas ( Figure S6) to test whether two component mixing is evident in the source compositions for the NW Lau samples. The samples from NW Lau show evidence for binary mixing between an enriched source and DMM [Workman and Hart, 2005]. The modeled theoretical enriched source, determined from the most enriched RR samples, is enriched in certain trace elements relative to Ta'u source (e.g., Ba, Nb, La, Th) but has Samoan-like isotopes ( Figure 1.5, Figure S6).

A Third Mantle Component?
Although the majority of our trace element and isotopic data suggest only two mantle sources, a distinct, third mantle source could be present, following several other lines of evidence. In particular, the highly enriched samples from NWLSC, the coupled He-Ne isotope systematics, the chalcophile element contents, and some radiogenic isotope data point to alternate sources of compositional diversity in the mantle beneath NW Lau. Here, we discuss the data and reasoning behind these additional constraints, and consider the implications for mantle source models in the NW Lau region.
We first explore whether variable degrees of melting of a binary mantle source can explain the observations that suggest more than two mantle sources for our samples (i.e., the most Nb-enriched NWLSC samples, and the differences between RR and NWLSC trends in Figure 1 Figure S5). Samples from RB  plot with or below the trend defined by Samoa [Jackson et al., 2009], whereas samples from RR and NWLSC [Lupton et al., 2012] Tian et al., [2011], is present beneath the NW Lau Basin, our data indicate that it must be localized to the area beneath RB and is not widely dispersed through the regional mantle.
We also note that the Manus Basin is another Pacific back-arc basin with elevated 3 He/ 4 He signatures, that is associated with locally elevated mantle temperature [e.g.,  and with a diffuse regional mantle plume [e.g., . The Manus Basin provides a regional example of elevated 3 He/ 4 He signatures without the presence of a clearly identified mantle plume, thus suggesting that the Samoan mantle is not the only possible source of high 3 He/ 4 He in the NW Lau Basin, and its proximity to the enrichments in NW Lau could simply be coincidence. In this view, the enriched component beneath the NW Lau Basin could be sourced by a diffuse local plume (e.g., Figure 1.10c), as has been suggested for the Manus Basin. Some basalts from the NW Lau Basin do, however, have trace element and isotopic characteristics in common with Samoan sources, making this hypothesis less likely.
Models involving just two mantle sources cannot explain the lack of correlation between noble gases and other isotopes (    and the noble gas and radiogenic isotope constraints Lupton et al., 2012] there may be inherent mantle heterogeneity beneath NW Lau Basin ( Figure   1.10b). The Manus Basin model, where the basalts have a geochemical plume signature without an identified plume in the region , would permit a third hypothesis (Figure 1.10c), which invokes a previously unidentified regional hotspot with high 3 He/ 4 He located to the west of NW Lau Basin, to explain the similarities in trace element and isotopic signatures between some NW Lau basalts and NFB basalts.  Lupton et al., 2012]. Processes such as degassing of the enriched melts  or diffusion of the noble gases [He and Ne; Lux, 1987] could explain the lack of correlation between He and other radiogenic isotopes and trace elements. Helium and Ne degas faster than the other noble gases, which may result in an observed lack of correlations in enriched melts between He (and Ne) and

Independent Behavior of Helium
other isotopes or noble gases [Lux, 1987;Niedermann et al., 1997]. Another process that could result in He moving independently in the mantle is a "leaky lower mantle" scenario induced by rapid spreading rates [Shen et al., 1995;Niedermann et al., 1997].
This model was developed for fast spreading ridges (e.g., the East Pacific Rise), where fast spreading is associated with deep induced mantle flow that could pick up material introduced into the upper mantle by miniature plumes leaking upward from the lower mantle at the 670 km boundary. The NW Lau Basin is opening with extremely fast spreading rates (160 mm/yr in the north), and this process could thus also apply at NW Lau, causing elevated noble gas signatures, with normal MORB Sr-Nd-Pb isotopes.
Another possibility is the Samoan Plume could be entering the NW Lau Basin in a chaotic manner due to the complex tectonics (e.g, fast subduction, tear in the subducting plate, and rapid upwelling of the mantle beneath the back-arc ;Hawkins 1995, Millen andHamburger 1998). If the Samoan Plume is not entering the back-arc basin smoothly then a clean latitudinal decrease of the geochemical tracers might not be expected [Spiegelman 1996].

Conclusions
Our  , the open circles and triangles are from Tian et al. [2011], and the crossed symbols are from . The dashed lines outline the regions shown on panels C-F. Basemaps in panels A-B were created using GeoMapApp [http://www.geomapapp.org; Ryan et al., 2009]. C-F) Swath bathymetry maps and sample locations for the four main regions of this study: RR, NWLSC, PR/LETZ, and CLSC. High-resolution bathymetry was collected in 2008 by R/V Southern Surveyor using an EM300 multibeam bathymetry system. A 1:1 line is plotted with an uncertainty in collection pressure of 50 bars, to account for the possibility that lava flowed downhill from the initial eruption site.   (Taylor and Martinez, 2003 Workman et al., 2006;Jackson et al., 2007b, Jackson et al., 2010 is shown by the solid shaded region for reference. A) Select Rochambeau Rifts glasses are shown by the blue lines, and a field for the rejuvenated-stage Samoan lavas [Jackson et al., 2010] is textured with carat symbols. The heavy black line shows the composition of a 6% fractional melt of the Ta'u Samoan mantle source of Jackson et al. [2007a] using bulk D's from . This composition also is shown by the 6% tick mark along the dashed gray line in Figure 8b that tracks variable degree melts of the Ta'u source. B) Select NWLSC glasses are shown by the orange lines. The most Nb-enriched NWLSC glass ) is highlighted as a heavy line. Lavas from the N160 segment of the North Fiji Basin are encompassed by the field textured with crosses. The heavy black line is the same 6% fractional melt of the Ta'u as in 4a. C) Select glasses from PR/LETZ (gray-purple lines) and glasses from CLSC (red lines).  , the crossed symbols are from , and the filled crosses are from the North Fiji Basin and open hexagons are from Niuafo'ou [Pearce et al., 2007]. The field for the Vai trend of Samoan shieldstage volcanics Workman et al., 2006;Jackson et al., 2007b, Jackson et al., 2010 is shown by the solid shaded region for reference. An average composition for Vailulu'u is indicated by the white star within the Samoan field. Shaded fields for Manus Basin Sinton et al., 2003], Iceland [Condomines et al., 1983;, Loihi [Craig and Lupton, 1976;Rison and Craig, 1983;Kurz et al., 1983], depleted mantle , and Primitive Helium Mantle [PHEM; Farley et al., 1992] are shown for reference. The thin line is the average MORB 3 He/ 4 He ratio of 8. B) Plot of 87 Sr/ 86 Sr vs. La/Sm N . The field for MORB [Tables 1.S5; compiled from PetDB; http://www.petdb.org], is shown by the solid gray shaded region for reference. C) Plot of 87 Sr/ 86 Sr vs. 143 Nd/ 144 Nd. The solid black line shows mixing between a depleted CLSC-like end member and an enriched Samoan-like end member (e.g., Vailulu'u). D) Plot of 206 Pb/ 204 Pb vs. 208 Pb/ 204 Pb. The solid black line shows mixing between a depleted CLSC-like end member and an enriched Samoan-like end member (e.g., Vailulu'u). The thick solid black line is the Northern Hemisphere Reference Line . Black squares represent the fields for the enriched mantle end-member EMII [e.g., Workman et al., 2004], primitive helium mantle [PHEM; Farley et al., 1992], and global hotspot mantle component ["FOZO";Hart et al., 1992] Hirschmann [2000], the two wet solidi are for of 0.05 wt.% and 0.1 wt.% [Kelley et al., 2010]. The solid black line is the adiabat for a mantle potential temperature of ~1400°C. The two dashed lines are melt paths (slope = 2°C/kbar) for the two wet solidi, generated as the adiabat crosses each solidi [Asimow et al., 2004], generating low amounts of F due to melting and the continuing dehydration of the surrounding mantle which reduces water-influenced melt productivity. The three thin lines are melt paths (slope = 7°C/kbar) above the dry solidus, a more productive melt regime, and the thin dashed line is the projected path of the adiabat ]. The outlined region shows the inset of the pressure-temperature diagram. Error bars are error associated with using the thermobarometer [Lee et al., 2009].  -19 -18 -17 -16 -15 -14 RR NWLSC PR/LETZ CLSC C Figure 1.10: Three regional cartoons illustrating possible hypotheses for the origin of enriched mantle beneath the RR and NWLSC. A) The first scenario invokes a contribution from an isotopically recognizable component of Samoan mantle and a separate, noble-gas-only component that causes enrichment in He isotopes without a proportional contribution from Samoan trace elements. B) The second scenario invokes inherent mantle heterogeneity in the NW Lau mantle, unrelated to (or linked to, as in the case of RB) Samoa. C) A third scenario, in which the N. Lau basin is infiltrated by a small quantity of Samoan mantle and is also impinged upon by a previously-unidentified regional hotspot located to the west of the spreading centers and rift zone.   . High resolution bathymetry shows that the NWLSC region is not a single spreading center,  Table 1.S1. Figure 1.1 shows the regional location and tectonic activity for the Northern Lau Basin, while Figure 1.1 shows a summary of the new high resolution 30 kHz multibeam bathymetric data acquired on the voyage and the subsequent sample locations from which this study's glass samples were chosen.

S1.2 Sample Preparation
A total of 62 glass samples from SS07/2008 dredges were selected for geochemical analysis. Fresh glass was chiseled from pillow rims and glass chips were hand-picked for geochemical analysis using a binocular microscope to ensure that only the freshest, crystal-poor glass was used for analysis. The glass data reported in this study are from the same dredges, though not the exact pillows, as those analyzed for 3 He/ 4 He by Lupton et al. [2009] and trace elements by . The glass samples reported in these previous studies originated largely from aggregate glass collected by pipe dredge. In one instance, intra-dredge heterogeneity of glass was identified in the pipe dredge, but all other dredges were comparatively homogeneous (see supplementary section S4). Twelve additional samples from the CLSC were obtained from the Smithsonian Institution's seafloor glass collection and analyzed for volatiles and trace elements.

S1.3 Electron Microprobe Analysis (EMPA)
Glass chips were mounted in 1 inch round epoxy mounts for electron microprobe analysis (EMPA). Major element concentrations (Table 1.S2) were measured on the Brown University CAMECA SX-100 electron microprobe using a 15 kV accelerating voltage, 10 nA beam current and 10 µm defocused beam following the methods of . Calibration was checked against basaltic glass references VG-2 and A99B compositions every 10 glass chips and a series of 5 spots were analyzed and averaged for each glass chip. Accuracy was typically ≤2% RSD [Jarosewich, 2002;.

S1.4 Secondary Ionization Mass Spectrometry (SIMS)
Separate glass chips from the same pillows were mounted in indium for

S.2 Comparison of LA-ICP-MS and solution ICP-MS data
Glasses from the Northwestern Lau basin were analyzed by both solution and laser ICP-MS techniques to assess the accuracy of the laser calibrations and trace element data relative to the more conventional solution-based analysis. Figure 1.S1 shows comparisons of trace element concentrations determined by LA-ICP-MS and solution ICP-MS, which shows excellent agreement (±≤5%) between the two methods for a variety of trace elements (e.g., Rb, Nb, Ba, La, U; Figure S1a-e). Three samples from the NW Lau basin are andesites (NLD-40-02, NLD-41-02, NLD-41-03). When the andesites were removed from the comparison, agreement between the two methods improved slightly for many elements. We hypothesize that the improvement is a result of using a basalt-based calibration for solution-ICP-MS, which may not have properly bracketed the higher concentrations of many trace elements in the andesite unknowns.  Kelley et al., 2003]. The bias in Zr cannot be explained this way because Zr is largely unaffected by hydrothermal alteration processes and no included phase could significantly bias Zr concentrations in bulk vs. glass composition. Instead, we believe the offset for Zr is an analytical bias in the LA-ICP-MS calibration, relative to the solution calibration. We thus increase the LA-ICP-MS concentrations for Zr by 7% to ensure close agreement between the two methods.

S.3 Comparison of LA-ICP-MS and electron microprobe data
Minor elements (K 2 O, TiO 2 , MnO, P 2 O 5 ) are not always routinely analyzed by LA-ICP-MS, but are most commonly measured by electron microprobe (EMP). Figure   S2 shows comparison of minor element concentrations measured by LA-ICP-MS at URI/GSO with those determined by EMP. It is important to note that, for this comparison, we assume that there is no inter-laboratory bias in CaO, the element to which the LA-ICP-MS raw data were normalized for data reduction. Heterogeneity in dredge NLD-44 was identified while conducting LA-ICP-MS analyses of multiple chips from the pipe dredge, from which He isotope samples were taken [see Lupton et al., 2009]. Glass chips were chosen from four pillows and 20 chips were selected from the pipe dredge. As shown in Figure 1.S3, the samples fall into two clearly distinct geochemical groups: LREE enriched and slightly LREE depleted. All glass chips from the pillow basalts, which provided the source material for radiogenic isotope analysis of this sample, and represent the geochemical type that is reported for the major and trace element composition of the glass in this paper, were found to be LREE enriched, while the glass chips from the pipe dredge, which provided the source material for He isotope analysis from this sample [Lupton et al., 2009], fell into both geochemical groups.

S.6 Melting Inversion Process and Ti/Y Source Composition Model
The water content of the NW Lau mantle sources was calculated using the melting inversion process from , which uses the batch melting equation and the Ti/Y source composition model. Provided here is a brief summary of the methodology behind using TiO 2(Fo90) concentration as a proxy for melt fraction (F), while further details can be found in           [Pearce et al., 2007]. The field for the Vai trend of Samoan shield-stage volcanics Workman et al., 2006;Jackson et al., 2007b, Jackson et al., 2010 is shown by the solid shaded region for reference. Shaded field for Manus Basin Sinton et al., 2003] is shown for reference. The field for MORB, encompassing East Pacific Rise MORB glasses [ Table 1.S8; compiled from PetDB; http://www.petdb.org], is shown by the solid gray shaded region for reference. "Indian" and "Pacific" type mantle designations from [Pearce et al., 2007]. The solid black line is the Northern Hemisphere Reference Line . B) Plot of 143 Nd/ 144 Nd vs. 206 Pb/ 204 Pb. The field for the Vai trend of Samoan shield-stage volcanics Workman et al., 2006;Jackson et al., 2007b, Jackson et al., 2010 is shown by the solid shaded region for reference. C) Plot of 143 Nd/ 144 Nd vs. 176 Hf/ 177 Hf. The field for the Vai trend of Samoan shield-stage volcanic [Salters et al., 2011] is shown by the solid shaded region for reference. D) Plot of 3 He/ 4 He vs. 206 Pb/ 204 Pb. The field for the Vai trend of Samoan shield-stage volcanics Workman et al., 2006;Jackson et al., 2007b, Jackson et al., 2010 is shown by the solid shaded region for reference. E) Plot of 3 He/ 4 He vs. 143 Nd/ 144 Nd. The field for the Vai trend of Samoan shield-stage volcanics Workman et al., 2006;Jackson et al., 2007b, Jackson et al., 2010 is shown by the solid shaded region for reference. F) Plot of 21 [Workman et al., 2005] and the solid diamond is Ta'u source [Jackson et al., 2007a]. The solid circle is a theoretical source determined from the highly enriched RR samples and the solid black line is a mixing line between DMM and Ta'u. B) Plot of C 0 vs. C 0 .

Introduction
Back-arc basins (BAB) are similar to mid-ocean ridges in many ways, including decompression melting beneath the spreading center. At mid-ocean ridges (MOR), magmatic and mantle water contents are relatively low, and recent studies have developed a reasonable understanding of the behavior of water in these settings Dixon and Stolper, 1995;Dixon et al., 2002;. At back-arc basin spreading centers, however, H 2 O plays a more complex role because it can be added to the mantle source by the subducting slab, which changes the melting behavior and the mantle source composition (Stolper and Newman, 1994;Taylor and Martinez, 2003). Water from the subducting plate influences magma formation and evolution in many ways, including depression of the solidus temperature (Kushiro et al., 1968;Gaetani and Grove, 1998) and the suppression of plagioclase crystallization (Sisson and Grove, 1993a;Sisson and Grove, 1993b;Danyushevsky, 2001). Both of these processes result in basalts of fundamentally different bulk composition than those from relatively anhydrous tectonic settings.  (Stolper and Newman, 1994;, there are three back-arc basins (Lau, Manus, N. Fiji) in the Western Pacific that are less well investigated for magmatic volatile content.
Although models of adiabatic decompression melting of a relatively anhydrous mantle source have been developed for mid-ocean ridge settings , these models do not account for the more hydrous mantle of back-arc basin settings. A parameterization of hydrous, adiabatic melting was developed by Katz et al. (2003), but the treatment of melt fraction (F) vs.
temperature at low melt fraction does not fully capture the variation observed in data.
Back-arc basin melting models were empirically investigated by  and showed a correlation between (i.e., the change in melt fraction with respect to the change in mantle source H 2 O concentration) and mantle potential temperature (T p ). The problem with the model in  is that F vs. H 2 O was assumed to be a linear relationship. Back-arc basin melting models were also investigated by , who, using the melting parameterization of Katz et al. (2003),  , while the continuous melting regime requires a continuum from wet to dry conditions throughout the back-arc mantle sources .
One approach for testing these competing hypotheses of back-arc magma generation is a forward melting model (e.g., Katz et al., 2003;Kelley et al., 2010), as described above, while another approach is an inverse melting model (e.g., Stolper and Newman, 1994;. An inverse melting model uses the melt composition to constrain the conditions of melting. Stolper and Newman (1994)   ).
Using the model-and data-based LLDs to project compositions back to equilibrium with the mantle, we provide more accurate constraints on source composition and extent of melting in each back-arc basin. Well constrained values for titanium source concentrations ( ) are important because is often used to estimate the melt fraction (F) for lavas . Revisiting previous inverse melting models with more data, we will attempt to determine the shape of the melting function, which will help discern between competing models for trends in H 2 O vs. F in BABBs. The well constrained estimates on melt fraction from the newly calculated values will be compared with a newly developed, hydrous, adiabatic melting model. We will take the advances made since the Katz et al. (2003) model for isobaric/isothermal melting, and incorporate these with an adiabatic melting model  to provide a new forward model of hydrous, adiabatic melting beneath back arcs. We will then test this new, polybaric hydrous back-arc basin melting model against the inversion of natural melt compositions to resolve the competing hypotheses of mixing vs. melting processes of back-arc basin basalt generation.

Mariana Trough
The Mariana Trough (Figure 2.1a), a crescent-shaped back-arc basin opening behind the Mariana Arc, formed from the subduction of the Pacific Plate beneath the Philippine Sea plate . Deep Sea Drilling Project (DSDP) Leg 60 determined that spreading, at a rate of 2.15 cm/yr, in the back-arc began about 6.5 Ma . The Mariana Trough can be divided into three sections, Northern, Central, and Southern, based on spreading characteristics.
The Northern Mariana Trough (NMT) is a region of rifting shown by block-faulted terrain and localized volcanism, where the areas of volcanism are located close to the arc . The Central Mariana Trough (CMT) is the mature spreading center with volcanism restricted to the main spreading axis (Hussong and Fryer, 1983;. The Southern Mariana Trough (SMT) is a shallow region where the end of the back-arc spreading ridge intersects with the volcanic arc, forming a complex region known as the Southeast Mariana Forearc Rift .

East Scotia Ridge
The South American Plate is subducting beneath the Sandwich Plate at a rate of 70-85 km/Myr , forming the South Sandwich Islands and Trench. Located to the west of the South Sandwich Islands, the East Scotia Ridge Back-arc Basin (Figure 2.1b) consists of nine spreading segments (E1-E9) with spreading rates of 60-70 km/Myr .
Spreading segment E1 is a trough that intersects the South Sandwich trench, spreading segments E2 and E9 are axial volcanic ridges, and segments E3-E8 are faulted median valleys similar to the Mid-Atlantic Ridge Bruguier and Livermore, 2001). Spreading along the East Scotia Ridge began ca. 11 Ma, initiating in the north and moving southward over time, with an average basin-wide spreading rate of 65 mm/yr over the last 1.7 Ma .

Manus Basin
Located behind the New Britain arc in the Bismarck Sea is the complex, rapidly opening Manus Back-arc Basin (Figure 2.1c). About 10 Ma, the subduction direction changed as a result of a collision of the Ontong Java Plateau with New Ireland and the North Solomon Arc (Cooper and Taylor, 1987; (Taylor, 1979;Martinez and Taylor, 1996).

Lau Basin
The Tonga-Lau system is an oceanic subduction zone in the southwest Pacific, where the Pacific Plate subducts beneath the Indo-Australian Plate. Behind the Tonga Arc, back-arc spreading initiated at ~6 Ma in the Lau Basin (Figure 2.1d; Taylor et al. 1996), which is a V-shaped basin with several actively spreading segments that impinge upon the Tonga Arc towards the south. The rates of both plate convergence and back-arc spreading are highest at the north end of the subduction zone , which exhibits the fastest back-arc opening on Earth, spreading at a rate of 160 mm/yr, decreasing southwards to rates of 60 mm/yr (e.g., . At the northern end of the Tonga Arc, the plate boundary bends 90° and the Pacific Plate ceases to subduct, and this northern boundary is the Vitiaz Lineament, which is interpreted as a paleo-subduction zone that is now a transform boundary separating the Pacific and Indo-Australian Plates

North Fiji Basin
The North Fiji Basin (NFB; Figure 2.1e) is located to the west of the Lau Basin, opening behind the Vanuatu Arc. The New Hebrides subduction zone, the western boundary of the NFB, experiences subduction at a rate of 9-12 cm/yr , and evolves into the Hunter Fracture Zone, a transform fault, starting at the southern end of the NFB. The NFB contains several regions of spreading, including the Central Ridge System (CR), Eastern Ridge (ER), and Fiji Fracture Zone (FFZ) (Price et al., 1990;Eissen et al., 1994). The CR consists of the N-S segment, N15 segment, and the N160 segment , which are E-W trending enechelon grabens on top of a ridge-like region of elevated topography . The CR is a slow spreading ridge with a spreading rate of 2 cm/yr (Price and Kroenke, 1991;Kroenke and Eade, 1990).

Back-arc Basin Basalt Samples
Complied here is a global back-arc basin data set of major element, trace element, and volatile measurements of 327 basaltic glass samples from the Mariana Trough Stolper and Newman, 1994;, East Scotia Ridge , Manus Basin Sinton et al., 2003;, Lau Basin Danyushevsky et al., 1993;Sinton et al., 1993;Pearce et al., 1995;Bézos et al., 2009;Escrig et al., 2009;Tian et al., 2011;, and North Fiji Basin Danyushevsky et al., 1993;Eissen et al., 1991;Sinton et al., 1993;Eissen et al., 1994;. All samples reported here are literature data, with the exception of select volatiles and trace elements from Manus Basin, Lau Basin, and NFB (M. Lytle, unpublished data; see Manuscript III) and major elements for five samples from NFB (Table 2.S1), three of which are referenced in Eissen et al., 1991. Two others are from the R/V Southern Surveyor SS07/2008 cruise .

Analytical Methods
The five glass chips from NFB that required major element analysis (see section 2.2) were mounted in 1 inch round epoxy mounts for electron microprobe analysis (EMPA). Major element concentrations (Table 2.S1) were measured on the Brown University CAMECA SX-100 electron microprobe using a 15 kV accelerating voltage, 10 nA beam current and 10 µm defocused beam following the methods of . Calibration was checked against basaltic glass references VG-2 and A99B compositions  every 10 glass chips and a series of 5 spots were analyzed and averaged for each glass chip. Precision was typically ≤ 2% RSD (Jarosewich, 2002;.

Interlaboratory Bias
Before using compiled major element data from multiple laboratories, we must consider bias introduced into the data set from analysis of major elements and volatiles in different laboratories. The major element concentrations of all samples were corrected for interlaboratory bias following the procedure discussed in Langmuir et al.  . Therefore, all non-Smithsonian major element analyses were first corrected to the LA-ICP-MS major element values and then corrected back to Smithsonian values (reported as bulk correction factors in Table   2.1). The corrected major elements are reported in Table 2.S2. Major elements from the East Scotia Ridge samples  were collected using the Lamont Doherty EMP glass standard JDF-D2, and as such, the correction factor applied for these samples is the same as the one for Lamont Doherty.

Effect of Degassing
Volatile loss from magma results from exsolution of dissolved gases from the magma during depressurization upon ascent from the mantle to the surface.
Assessment of volatile loss from each glass is possible as major volatile species (e.g.,  (Dixon and Stolper, 1995;  found to be vapor-oversaturated or saturated at the pressure of collection, which is typical of mid-ocean ridge basalts and reflects relatively fast transport and eruption of magma from mid-crustal depths . Samples were filtered for degassing and included in further modeling if the samples either a) appeared undersaturated if there was no CO 2 data or b) lay along or below the 1:1 line ( Figure   2.2b), and therefore are considered saturated or undersaturated (n = 246).

Effects of Crystallization
Water suppresses plagioclase and clinopyroxene crystallization (e.g., Sisson and Grove, 1993a;Sisson and Grove, 1993b) and its effects can be seen in the major element systematics.  (Taylor and Martinez, 2003).

Results
Using dry BABBs (< 0.5 wt.% H 2 O), we can treat these basalts like MORBs and investigate the melting relationships of BABBs with respect to well-characterized adiabatic decompression melting behaviors of MORBs (Klein and Langmuir, 1987;Langmuir et al., 1992).  .
Another well-characterized aspect of MORBs is the relationship between axial depth and the melt fraction. Deeper axial depths (i.e., 5000 -6000m), resulting from decreased crustal thicknesses, which is caused by smaller extents of melting in a region with cooler mantle temperatures and therefore, have higher Na 2 O contents in the melts (Klein and Langmuir, 1987). The dry BABBs follow the well-characterized relationship found in MORBs, where the increasing axial depth (more negative depths) for BABBs correlates with the increasing Na Fo90 of the melt (Figure 2.4a).
Another well understood relationship in MORBs is FeO systematics, in which FeO is sensitive to the pressure and temperature of melting (Klein and Langmuir, 1987;Langmuir et al., 1992). Higher FeO contents in the lava are indicative of melting at deeper pressures and warmer temperatures, where olivine is more MgO-rich and therefore the melt becomes more enriched in FeO.
Higher pressures and warmer temperatures of melting result in higher melt fractions, therefore Na 2 O and FeO are anticorrelated at global spreading centers . Using the dry, adiabatic melting model from Langmuir et al. (1992), the MORB Na 2 O and FeO values were fit with a potential temperature (T p ) model curve and resultant T p parameterizations from Na 2 O and FeO . Figure 2.4b shows constraints on the mantle potential temperature for the dry BABB lavas using Fe Fo90 vs. Na Fo90 , which has been previously shown in many studies (e.g., Stolper and Newman, 1994;Kelley et al., 2010). Na 2 O (Fo90) generally decreases with increasing H 2 O (Fo90) (Figure 2.5b), although this trend is less well defined, because although Na is an incompatible element, Na can also be added to the mantle source region by a slab-derived fluid. Na2O has been identified as a primary component of slab-derived fluids (e.g., Stolper and Newman, 1994;Eiler et al., 2005), making it challenging to separate out mantle and slab-derived contributions to the Na2O content of basalts. Therefore, the addition of Na via subduction to the back-arc mantle source will affect the relationship between H 2 O (Fo90) and Na 2 O (Fo90) , decreasing the strength of correlation observed in Figure 2.5b.
The negative relationship between H 2 O (Fo90) and FeO (Fo90) (Figure 2.5c) shows the effect of water on the MgO and FeO abundances in hydrous melts (Gaetani and Grove, 1998). The addition of H 2 O to the mantle expands the olivine-liquidus boundary, inducing melting at cooler temperatures than in an anhydrous case (Kushiro et al., 1968;Gaetani and Grove, 1998). The MgO and FeO contents of the hydrous melt record the temperature, and therefore, pressure of melting. The composition of a melt produced at cooler temperatures will reflect equilibrium with a Fe-rich olivine, yielding a melt with lower FeO concentration than the anhydrous, higher temperature scenario (Roeder and Emslie, 1970). Therefore, at a similar depth of melting, a melt with higher H 2 O contents is expected to have lower FeO* because it will have lower temperature, resulting in the negative relationship observed between H 2 O (Fo90) and FeO (Fo90) in Figure 2.5c.

Discussion
Previous attempts at modeling back-arc melting were limited by the availability of high-quality trace element data, as a means of assessing the mantle source composition, and by a lack of realistic melting models. Here we improve on the inverse melting model by using trace element data to constrain the mantle source composition for each sample, and we improve on the forward melting models by developing a hydrous, adiabatic melting model that uses the most current constraints on how melting proceeds at low F and allows melting to occur over a range of pressures. These two modeling approaches are then shown to be in fairly good agreement.

Constraining the source composition of the BABB lavas
The source composition of the mantle is an important factor to consider when estimating the melt fraction (F), especially when using TiO 2 as a single-element proxy for F. The inherent assumption in using TiO 2 concentration to constrain melt fraction is that Ti behaves conservatively. Specifically, because Ti is a high field strength element with low mobility in aqueous fluids (Pearce and Parkinson, 1993), the concentration of Ti in a primary mantle melt ( ) should reflect the TiO 2 concentration of the mantle source and not any additions from the slab-derived fluid.
Although there are questions about whether Ti behaves in a completely conservative manner or behaves in a slightly fluid mobile manner, we continue to use Ti to estimate melt fraction, as the study by Stolper and Newman (1994) where (TiO 2 /Y) sample is the TiO 2 /Y ratio of the glass, (TiO 2 /Y) MORB is 0.04, DMM is depleted MORB mantle, and is 0.133 .

Modeling hydrous, adiabatic melting
A hydrous, adiabatic melting model is developed following the approach and steps applied by Langmuir et al. (1992) in developing a dry, adiabatic melting model for MORB. The dry solidus was identified and then the melt curves were spaced at two constant dT/dF values of 3.5°C/% for < 22% melt fraction (point of cpx-out) and 6.8°C/% for > 22% melt fraction . The first modification is to change the dry solidus to a hydrous solidus, using the hydrous, isobaric, isothermal equation from Kelley et al. (2010). A mantle adiabat is calculated using dT/dP = 1°C/kbar and the mantle potential temperature (T p ) is selected to be a reasonable estimate for the specific back-arc basin from Na-Fe systematics . Calculation of the adiabatic melt path relies on the melt productivity, which is the extent of melting as a function of pressure, referred to as γ in Langmuir et al. (1992).: where dT/dP adiabat is the slope of the adiabat (1°C/kbar), dT/dP solidus is the slope of the solidus (13°C/kbar), H ƒ is the heat of fusion or energy required for melting, and C p is the heat capacity (0.238 cal/g/°C; Katz et al., 2003). Here we use a value of -0.16%/GPa for melt productivity from Ganguly, 2005 and because the melt fraction contours are spaced more closely as pressure increases, γ must include a pressure correction term to account for the change in spacing between melt fraction contours. Langmuir et al. (1992) used a pressure correction term of 1-P/88, where P is in kbar and 88 reflects the decrease of the melting interval by a factor of 2 over ~45 kbar and the melting contours are assumed to decrease proportionally.
Since γ is the extent of melting as a function of pressure, we will calculate F over 1 kbar pressure intervals. However, γ is a negative value and the absolute value of γ must be taken, yielding: and the total melt fraction is calculated using the following equation: where is the final pressure of melting and is the initial pressure of melting. The temperature, per 1 kbar pressure intervals, for the adiabatic melt path then can be calculated with the following equation: in which dT/dF changes with increasing F, and F is the total melt fraction at a specific pressure (Figure 2.7).

Application of models
The adiabatic, hydrous melting model developed above requires a mantle potential temperature that is appropriate for the back-arc basin in question to be selected. As seen in Figure 2.4b, T p is constrained using the dry BABB lavas and the Na and Fe T p parameterizations from . difference, which is within the range reported for global mid-ocean ridges (250°C; Klein and Langmuir, 1987).
The resulting melt fraction curves from the adiabatic model show a near linear shape (Figure 2.8) rather than a more curved shape observed in the isothermal/isobaric melting models (e.g., Katz et al., 2003;Kelley et al., 2010). The shape of the melt fraction curves is important to consider because linear data trends are indicative of mixing rather than melting . Another important feature of the adiabatic melt curves is the shallowing of the slope as a function of increasing T p and pressure, which according to , should not change with increasing temperature. However, the melting models  used to arrive at these conclusions were isothermal/isobaric models for one specific pressure. The relationship of mixing vs. melting will be further examined using the newly developed hydrous, adiabatic melting model.
Using the inverse modeling and the constraints on T p for each basin, we can examine the relationship between F and in a hydrous, adiabatically upwelling mantle, as shown in Figure 2.8. To evaluate the effects of uncertainties on data trends, we used a Monte Carlo error analysis, which allows each parameter to vary simultaneously within its assigned uncertainty during the calculation of F and , on ten select samples from the back-arc basins. The error ellipses around data points represent 90% error confidence, and the elongation of the ellipses emphasizes that errors on this diagram are highly correlated. The error ellipses are accounting for additional error (±50%) associated with estimating the using TiO 2 /Y systematics (see section 4.1), rather than previous constraints on , which result in ±10% error (see Kelley et al., 2010).
The five T p curves shown in Figure 2 Therefore, the assumption that a more linear melting model trend suggests mixing and not melting is not valid for an adiabatic, hydrous melt curve. Samples produced at back-arc spreading centers can result from either an integrated melting process or mixing between a shallow, hydrous, arc-like melt and a drier, MORB-like melt.
Further work, such as constraining the depth of release for the slab-derived fluid, is required to further determine whether back-arc basin basalts are products or melting or mixing processes.

Introduction
Back-arc basin basalts (BABB) have elevated H 2 O contents compared to midocean ridge basalts (MORB), which is often taken as evidence of the incorporation of H 2 O-rich slab derived fluids in the back-arc mantle source (Taylor and Martinez, 2003). Yet, many back-arc spreading centers do not vertically overlie the subducting plate, and the origin and transport pathways of slab-derived fluids to the back-arc are not well constrained. The question thus remains, from where in the subducted plate do these fluids originate, and how are they incorporated into the back-arc mantle source?
Although a broad spectrum of models for fluid transport in the mantle wedge have been proposed (e.g., Davies and Stevenson, 1992;, few constraints have drawn from the geochemistry of back-arc basin lavas themselves.
Here, we present new measurements of H 2 O and trace elements in back-arc basin basalt glasses, to address the questions of where back-arc fluids originate in the subducting plate, and how they are delivered to the mantle sources of back-arc spreading centers.
The question of how slab derived fluids reach the back-arc is an interesting and widely debated topic. Fluid migration in the mantle wedge of a subduction zone is commonly simplified to a vertical pathway, where fluids released from the slab travel directly upwards through the mantle to volcanic centers at the surface (e.g., Tatsumi, 1989). This surely represents an oversimplification, as the subducting slab is not always vertically present beneath back-arc spreading centers (e.g., Mariana Trough; Creager and Jordan, 1986). Several proposed mechanisms for delivering slab derived fluids to the back-arc mantle source derive from both dynamical and geochemical viewpoints. For example, amphibole formation/breakdown reactions are proposed as a mechanism for transporting H 2 O horizontally away from the subducted plate (Davies and Stevenson, 1992). Another concept for fluid migration that has been investigated is diapiric flow where the release of H 2 O, in a fluid fluxed mantle wedge, forms a partially molten melt region from which diapirs form at the top. The diapiric flow can rise to the surface either as isolated diapirs with trailing conduits or networked flow of coalescing diapirs and resultant thick conduits, in which melt diapirs rise rapidly to the surface (~10 4 to 10 6 years; Tatsumi, 1989;Weatherly and Katz, 2012).
Another approach to addressing the question of fluid migration is geochemical modeling, which places fewer constraints on the migration processes or specific pathways, but uses geochemical characteristics of natural samples to constrain where the fluids originate in the slab. Analysis of back-arc basin submarine glasses using trace elements (i.e., Ba/Nb, Th/Nb, Th/Ta, and Nb/Ta) as tracers of total subduction input, mantle depletion, and shallow vs. deep subduction components provide constraints on shallow vs. deep origin of the back-arc fluids .

Measurements of H 2 O in submarine glasses from Lau, North Fiji, Manus, and
Woodlark Basins, in combination with major elements (i.e., K 2 O and TiO 2 ) provided geochemical constraints for determining whether the subduction-related component in BABBs was a fluid or a melt . A focused, comprehensive study of the Mariana Trough investigated the role of water in BABBs and proposed a positive correlation between extent of melting and H 2 O concentration of the mantle (Stolper and Newman, 1994). The Mariana Trough fluids reaching the back-arc source were proposed to go through a sort of chromatographic geochemical exchange with the mantle that partly controls the fluid composition (Stolper and Newman, 1994).
Another model proposes mixing melts from a shallow, arc-like fluid with low-water, fractional melts beneath the back-arc spreading center , based on observations of low FeO* in the wettest BABB melts that are suggestive of shallow hydrous melting. An alternative model proposes a fluid released at depth that enters the back-arc source from below , which expands the melt region beneath back-arc spreading centers to greater depths than observed at mid-ocean ridges. There is a large diversity of geochemical models for the origin of slab-derived fluids, despite obvious limitations caused by few quantitative constraints.
However, before the question of fluid migration can be addressed, constraints on the source location of the slab-derived back-arc basin fluids must first be The new thermometers have yet to be used at back-arcs, but they present great promise for aiding in the resolution of models for the origin of back-arc fluids and how they may be delivered to back-arc mantle sources.
With this work, we will address the origins of elevated H 2 O contents and H 2 O/Ce ratios in back-arc basin basalts with new and existing geochemical data for major and trace elements and dissolved volatiles for a large suite of submarine glasses from five global back-arc spreading centers. Using this comprehensive data suite, we will apply the new thermometers Cooper et al., 2012) to these global back-arc lavas to determine the temperature conditions of slab fluid release. We will show that slab fluids that reach the back-arc basin mantle source reflect ~60-100°C higher slab temperatures than arcs, when referenced at a common pressure, and they therefore originate from hotter parts of the slab than the fluids that reach arc mantle sources. With respect to slab thermal models for these settings, these constraints suggest that the fluids supplying back-arc mantle either derive from much higher pressures (≥8 GPa), or indicate a significant 3-D component to slab thermal structure that current 2-D models do not capture. Additionally, we explore along-strike variations and the consequences of these observations for models of mantle flow in global back-arc basin settings.

Geologic Context
The samples in this global back-arc basin study are selected from five back-arc basins located in both the Pacific and Atlantic Oceans (Figure 3.1). The first of four Pacific back-arc basins is the Mariana Trough (Figure 3.1a), a crescent-shaped backarc spreading center opening behind the Mariana Arc, formed from the subduction of the Pacific Plate beneath the Philippine Sea plate .
Deep Sea Drilling Project (DSDP) Leg 60 determined that spreading in the back-arc began about 6.5 Ma at a rate of 2.15 cm/yr .
The  Taylor et al. 1996), which is a V-shaped basin with several actively spreading segments that impinge upon the Tonga Arc towards the south. The rates of both plate convergence and back-arc spreading are highest at the north end of the subduction zone , which exhibits the fastest back-arc opening on Earth, spreading at a rate of 160 mm/yr, decreasing southwards to rates of 60 mm/yr (e.g., Taylor et al., 1996). The six spreading segments Located to the west of the Lau Basin is the North Fiji Basin (NFB; Figure   3.1e), which is opening behind the Vanuatu Arc. The New Hebrides subduction zone, the western boundary of the NFB, is subducting at a rate of 9-12 cm/yr , and evolves into the Hunter Fracture Zone, a transform fault, at the southern end of the NFB. The two spreading segments focused on in this study are the Triple Junction (TJ) and the N15 spreading center (N15).
The last back-arc basin, East Scotia Ridge (Figure 3.1b), is located in the South Atlantic Ocean. The South American Plate is subducting beneath the Sandwich Plate at a rate of 70-85 km/Myr , forming the South Sandwich Islands and Trench. Located to the west of the South Sandwich Islands, the East Scotia Ridge Back-arc Basin formed ca 11 Ma, with an average basin-wide spreading rate of 65 mm/yr over the last 1.7 Ma Bruguier and Livermore, 2001;. The East Scotia Ridge spreading segments focused on in this study are E2, E3, E4, E5, E6, and E9.

Samples and Preparation
The 327 basaltic glass samples in this study are a combination of complied, previously-published data and newly collected data for submarine glasses from the Mariana Trough, East Scotia, Lau, Manus, and North Fiji back-arc basin spreading centers. We present new trace element and dissolved volatile data from submarine glasses for the Lau, Manus, and North Fiji basins (Table 3.S1; Eissen et al., 1991;Sinton et al., 1993;Eissen et al., 1994;Sinton et al., 2003;. The basaltic glass samples reported here are from the glassy rims of basaltic pillow lavas and flow tops collected by sea floor dredging. Submarine glasses are important for this study because the glass is a representative snapshot of the basaltic liquid upon eruption, recording the volatile and trace element composition of the magma before eruptive degassing removes most volatiles from the lava. Samples analyzed in this study are from the publicly accessible repositories of the Smithsonian Institution Volcanic Glass Collection , the URI/GSO Marine Geological Samples Laboratory Eissen et al., 1994;, or were contributed by John Sinton . Additional samples include previously published volatile and trace element data from the Mariana Trough, East Scotia Ridge, and Lau Basin Volpe et al., 1987;Danyushevsky et al., 1993;Sinton et al., 1993;Pearce et al., 1995;Bézos et al., 2009;Escrig et al., 2009;Tian et al., 2011;Escrig et al., 2012;).

Secondary Ionization Mass Spectrometry (SIMS)
Glass chips from Manus Basin, Lau Basin, and North Fiji Basin lavas were mounted in indium for analysis of dissolved volatiles (H 2 O, CO 2 , S, Cl, F) by secondary ionization mass spectrometry (SIMS) at the Carnegie Institution of Washington. Volatile analysis (Table 3.S1) was done in triplicate using the CAMECA IMS 6f ion microprobe with a 5-10 nA Cs+ primary beam following procedures outlined by  and , using 16  were determined by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS; Table S1) using a New Wave UP 213nm Nd:YAG deep penetration laser coupled with a Thermo XSeriesII quadrupole ICP-MS at the Graduate School of Oceanography, University of Rhode Island following procedures described by Lytle et al. (2012). Analyses were run using an 80µm spot size, 10 Hz repeat rate, and 80-90% energy output. Nine natural glass standards from United States Geological Survey (BIR-1g, BHVO-2g, BCR-2g) and Max Planck Institute  GOR132-G, StHls-G, ATHO-G, T1-G, ML3B-G, KL2-G) were used to produce calibration curves that were linear to r 2 > 0.99 for all elements reported. Analysis of glass chips were done in triplicate with an average reproducibility of 4% RSD for all elements.

Effects of Low-Pressure Differentiation
Volatile loss from magma results from exsolution of dissolved gases from the magma during depressurization upon ascent from the mantle to the surface.
Assessment of volatile loss from each glass is possible as major volatile species (e.g., H 2 O, CO 2 ) have different vapor/melt solubilities. Carbon dioxide has lower solubility in silicate melt at low pressure and is expected to begin degassing before H 2 O (Dixon and Stolper, 1995), and the mixed CO 2 -H 2 O content of a glass reflects the minimum pressure of final equilibration of vapor with melt if the latter was volatile-saturated. been removed from the melt (Dixon and Stolper, 1995;. Samples without measured CO 2 data were treated as having no CO 2 in the glass. Based on this analysis, most glasses have likely lost variable amounts of CO 2 , but H 2 O concentrations are relatively unmodified from the original magmatic values. found to be vapor-oversaturated or saturated at the pressure of collection, which is typical of mid-ocean ridge basalts and reflects relatively fast transport and eruption of magma from mid-crustal depths . However, for the samples with no CO 2 data, the samples were only used in this study if they appeared undersaturated on Figure 3.2b. Samples were considered undegassed for H 2 O when they lay along or below the zone of saturation (gray field in Figure 3.2b; n = 246), where the pressure of H 2 O-CO 2 saturation was equal to or less than the pressure at the sample collection depth. Those that are interpreted as having lost some H 2 O were excluded from modeling.
Beyond degassing, H 2 O/Ce ratio is unlikely to be affected by magmatic processes such as crystallization. This lack of fractionation occurs due to a similarity in the partitioning (D H2O and D Ce ) of H 2 O and Ce between mafic silicate minerals and melt Dixon et al., 2002;. During melting or crystallization, mafic components such as olivine and orthopyroxene have D H2O /D Ce > 1 and clinopyroxene and garnet have D H2O /D Ce < 1 but the bulk lithology of spinel lherzolite has D H2O /D Ce ≈ 1, and therefore, there is little overall fractionation of H 2 O and Ce during mantle melting or crystallization .

Water and Trace Element Variations in BABBs
H 2 O/Ce ratio is a useful indicator of subduction influence, as the H 2 O content of the lavas generally increases with decreasing distance to the arc (e.g., , while Ce, a non-fluid mobile trace element, remains approximately constant. An additional ratio that is sensitive to subduction influence is Nb/Nb*, where Nb* is the projected concentration of Nb based on neighboring Th and La abundances. Nb prefers to remain in the slab rather than partitioning into an aqueous fluid or melt due to its compatibility in residual rutile in the slab phase assemblage. As a result, Nb  (Figure 3.3b). MORB samples  plot at low H 2 O and H 2 O/Ce ratios and arc samples (Cooper et al., 2012)

Application of the H 2 O/Ce Slab Surface Thermometer
A number of criteria must be met before the thermometer can be accurately applied. These include determining whether slab fluids have influenced the source of a given basalt, constraining the mantle source composition, constraining the contributions of H 2 O, Ce, and Nb to back-arc slab fluids, and assessing whether the slab mineralogy required for accurate application of this model is appropriate for BABB fluids. Here, we first assess these key criteria, and then use these new and existing data to apply the H 2 O/Ce thermometer to basalts from global back-arc basin settings.

Slab Mineralogy
An important concern in the application of the geothermometer is whether the BABB fluids derive from slabs that are saturated with allanite and/or monazite, which control the Ce budget of the fluids . The point of monazite-out is calculated through mass balance at ~950°C and 40% melting   that monazite has not been exhausted, and that the geothermometer can be applied.

Identifying Slab Contributions to Magmatic H 2 O in BABBs
Determination

Constraints on mantle source composition
The H 2 O and Ce contents of back-arc basin basalts derive from two sources, the MORB-like mantle and the fluid added to it from the subducted slab. In order to isolate the slab fluid composition, the mantle contributions to these element abundances must be determined, and then subtracted, from each basalt. This mantle unmixing process, to remove the effect of the mantle contribution to the H 2 O/Ce ratio, requires each basalt to be referenced to an appropriate mantle source composition. Cooper et al. (2012) used the Nb/Ce ratio to constrain the bulk composition of the mantle sources beneath arc volcanoes. Constraints on the source composition can also be determined using high field strength elements (HFSE), such as Nb and Zr, which are comparatively immobile during subduction. The advantage of using Nb/Zr ratio to determine the source composition is that we can determine the appropriate mantle source for each sample, rather than choosing one source for all samples within the basin (see Cooper et al., 2012). Therefore, following the approach of , we use Nb/Zr systematics to split the BABB samples into three groups of varying source enrichment (Figure 3.4; depleted, Nb/Zr < 0.02; normal, 0.02 > Nb/Zr < 0.03; enriched, Nb/Zr > 0.03). Depleted samples were referenced to the H-DMM source of Workman and Hart (2005), normal samples to the NMORB source composition of Sun and McDonough (1989), and enriched samples to the EMORB source of Sun and McDonough (1989).
Within each basin, we find a range of mantle source compositions, from enriched to depleted. Both the Mariana Trough and East Scotia Ridge have been viewed as having depleted mantle sources, but the Nb/Zr systematics show that the mantle sources of both basins are on average more enriched (e.g., . A hot spot signature has been identified in the basalts from the Manus Basin , and although hot spots are generally considered to be enriched, the Nb/Zr ratios show that the majority of basalts from Manus Basin require a depleted to normal source. The Lau Basin basalts are mainly depleted, consistent with inferences of a highly depleted mantle beneath the Tonga Arc Woodhead et al., 1993;Caulfield et al., 2008), and BABBs from North Fiji Basin show a variety of source enrichment.

Nb/Ce of the Slab Fluid
The Nb/Ce ratio of the arc slab fluid was set at 0.04, the minimum Nb/Ce value observed at arcs (Cooper et al., 2012), but the BABB fluids are likely to be more solute-rich (e.g., as modeled by Stolper and Newman, 1994) and perhaps capable of carrying more Nb. The importance of Nb mobility in slab fluids is significant and must be considered in the case of the BABB slab fluids. For this study, we chose a Nb/Ce ratio of 0.06, the minimum Nb/Ce value observed in back-arc basins (i.e., Lau Basin and Manus Basin samples). The lowest Nb/Ce ratio for BABB is higher than the lowest observed Nb/Ce ratio for arcs, and the Nb/Ce ratio could be significantly higher for the Mariana Trough based on the fluid concentrations of Ta and the light REE from Stolper and Newman (1994). The impact of choosing a higher fluid Nb/Ce ratio than 0.06 would be a decrease in the projected fluid H 2 O/Ce ratio, which will result in higher calculated SSTs.

Slab Surface Temperatures Derived from the H 2 O/Ce Model
After determining an appropriate mantle source for each back-arc basin basalt   Table   3.1).

The Slab Surface Temperatures Recorded by Back-Arc vs. Arc Fluids
The global range of corrected SSTs for the arcs reported by Cooper et al. (2012) is ~730 -900°C. The geothermometer is calibrated at 4 GPa, but a pressure limitation arises as the fluids may have released from the subducting slab at a pressure other than 4 GPa. Therefore, the resultant temperatures from the geothermometer at 4 GPa must be projected to the pressure at which the fluid was released in order to constrain the actual temperature of fluid release from the slab.  For arcs, Cooper et al. (2012) projected SSTs to a depth, h, which is the vertical distance from the volcanic center to the seismically-defined slab surface (Syracuse and Abers, 2006). Comparison between the back-arcs and their partner arcs provides a direct contrast of the origin of arc vs. back-arc fluid sources. However, we cannot project back-arc SSTs to a depth, h, because these constraints are unavailable for back-arcs, and because slabs, at least at shallow depths, are not vertically present beneath some back-arcs. 2D thermal models (i.e., Syracuse et al., 2010) give the P-T geotherms of the slab surfaces for global subduction zones that could be used to project SST 4GPa to the pressure and temperature of intersection with the modeled slab surface. Therefore, back-arcs may be projected to a depth, d, which is the depth in the There are multiple approaches to estimating the slab fluid composition and temperature within the back-arc basins. Taking the sample with the maximum H 2 O/Ce ratio at each back-arc spreading center constrains the minimum SST for each global back-arc basin (~760-1000°C; Table 3.2), but these are all within error (50C) of the arc SSTs and are unlikely to represent the mean characteristics of the slab fluids of these basins. Averaging H 2 O/Ce instead over each spreading segment provides an estimate of SST that likely represents the mean slab P-T conditions of fluid release, rather than the extreme outliers (i.e., minimum and maximum H 2 O/Ce ratios; Figure  3.7). The average slab surface temperatures constrained for each spreading segment within each back-arc basin (Table 3.1), as well as the average SST for each basin as a whole, are generally hotter than the associated arcs (Figure 3.6, 3.7). However, one segment in the East Scotia Ridge, E9, overlaps with the estimated SST of the South Sandwich Arc.
Comparison between the back-arc spreading segment average slab fluid temperatures (Table 3.1) and the arc fluid temperatures generally shows that the temperatures in the back-arcs are hotter than the associated arcs at 4 GPa ( We present three possible scenarios to explain the complication of projecting the back-arc basin segment temperatures to depth: (1) back-arc slab-derived fluids come from deeper than the 8 GPa limit of the 2D thermal models, (2) back-arc fluids come from the same depth as arcs but from hotter edges of the slab, or (3) thermal models predict slab surface geotherms that are too cold. The first scenario projects the back-arc slab-derived fluids to depths greater than 8GPa, as the D80 models stop at 8 GPa, in a region where the SST is modeled to be nearly isothermal. under-predict the slab temperature. Therefore, due to the complications that have arisen with projecting the SST to depth and the inability to constrain pressure, we will discuss the SST variations relative to the 4 GPa reference pressure of the model.

Local Variations in Back-Arc SSTs
The back-arc basin SST 4GPa are warmer than the respective arcs ( Considering the hypothetical scenario of a superadiabatic SST geotherm at depths >8 GPa (Figure 3.8a), we also project the SST for each segment that appears hotter than the D80 SST models of Syracuse et al. along the H 2 O-isopleths to 8 GPa (Figure 3.6). This exercise shows that, were this scenario an accurate reflection of the origin of slab-derived fluids that reach back-arc sources, the SST required would be ~200 -400°C higher than that beneath arcs. The further the spreading segment is from the arc, the H 2 O content has been observed to decrease closer to MORB-like H 2 O values (< 1.0 wt.%). The expected decrease in the presence of slab-derived fluids due to distance from the arc is also observed in the H 2 O/Ce values of BABB (250-3900) compared to arc basalts (400-20,000). The temperatures for the back-arc basin spreading segments, projected to 8 GPa, are ~1100 -1350°C, which are higher than 950°C, the temperature predicted by  where fluids will no longer be saturated with monazite and allanite. Therefore, as the fluids are assumed to be saturated in monazite (see section 3.3.1), the back-arc basin slab-derived fluids do not likely originate at depths greater than 8 GPa.
The effect of the decrease in amount of slab fluids in the back-arc is shown in and North Fiji Basin, might show SST variations, but these variations can't be fully resolved with the models at present (Figure 3.10a, 3.10c, 3.10e). The East Scotia Ridge samples show fluids reaching the mantle source at the southern end of the system are colder than those that reach the spreading center interior (Figure 3.10b).
This temperature relationship is consistent with models of temperature variations in a subducting slab experiencing symmetric rollback .
For a comprehensive examination of the SST variations along the Tonga/Lau subducting slab (Figure 3.10d), we look at the near-arc trend (ELSC, FSC, and MTJ).
MTJ is close to the northern boundary of the subducting slab, and displays a hotter temperature (879°C) than FSC, the next closest spreading region (818°C), suggesting that the subducting plate edge may be warmer than the center (Figure 3.8b, Figure   3.10d). The along-arc trend shown by the ELSC, FSC, and MTJ are consistent with laboratory models of non-symmetrical slab rollback, where the northern most part of the subducting slab experiences stronger flow around the plate edge and thus warmer temperatures (Figure 3.11; . Contrasting the near-arc trend (ELSC, FSC, MTJ) with the distal-arc trend (>150km from the arc; CLSC, ILSC, PR), hotter SSTs are observed for CLSC, ILSC, and PR. This provides strong evidence that the slab fluids increase in temperature, and originate from greater slab depths, as distance from the arc increases.
We may also assess variations in temperatures for the individual back-arc basins as a whole. Taking the maximum and minimum H 2 O/Ce ratios for each back-arc basin provides minimum and maximum temperature constraints on the origin of the slab-derived fluids that reach the back-arc. Table 3.2 shows the maximum and minimum temperature variations within each back-arc basin using the maximum and minimum H 2 O/Ce ratios. In Figure 3.7, at 4 GPa, the slab-fluid temperatures for the arc and the minimum back-arc temperatures overlap within the 50°C uncertainty of the model for all the back-arc basins except NFB. The minimum constraints on temperature thus show that a small proportion of the fluids that reach the back-arc may start at very similar PT conditions as those that reach arcs.
Using the maximum and minimum constraints on temperature, however, biases the perspective towards the extreme outliers, but when the segment weighted averages for each basin are calculated (Table 3.1), the mean temperatures are hotter outside of uncertainty. The large difference between the arc and minimum back-arc temperature in NFB is due to the location of the spreading centers >450 km from the arc. Figure   3.7 shows the comparison between the arcs and several constraints on the back-arc basin temperatures, where both the segment weighted average temperatures and the maximum temperatures for each back-arc are warmer than the arc, outside error (Tables 3.1, 3.2), thus suggesting a deeper depth of origin when projected.

Constraints on back-arc basin slab fluid origins
Observations from Figures 3.6 and 3.9 show that back-arc basins have hotter fluids than their respective arcs, but not how or why the back-arc fluids are hotter.
Here we discuss three possibilities presented by these observations: (1) back-arc slabderived fluids come from deeper, (2) back-arc fluids come from the same depth as arcs but from hotter edges of the slab, or (3) thermal models predict slab surface geotherms that are too cold.
The first scenario considers the possibility that the back-arc slab-derived fluids come from deeper depths than the associated arc. As observed in Figure 3.6, the backarc basin SST 4GPa are hotter than the arcs. When these back-arc segments are projected along H 2 O-isopleths to intersect the slab surface at depth, the BAB SST 4GPa do not intersect the D80 thermal model SSTs. The D80 thermal models stop at 8 GPa in a region where the subducting slab geotherm is adiabatic (Syracuse et al., 2010), and a schematic of the slab surface profile (Figure 3.8a) shows the SST eventually reaching an inflection point and becoming hotter with minimal increase in depth. Using the schematic in Figure 3.8a, the back-arc basin segment SST 4GPa can be projected to depth >8 GPa. In this scenario, back-arc basin slab derived fluids originate significantly deeper then respective arcs and come from deep in the mantle (>8 GPa).
The temperatures projected for the back-arc basin spreading segments at 8 GPa range from ~1100 -1350°C (Figure 3.9b), which are higher than the upper temperature limits on monazite and allanite saturation.  predicts a temperature of 950°C, above which slab-derived fluids will no longer be saturated with monazite and allanite. If the slab-derived fluids are not saturated with monazite and allanite, then the basic premise of the H 2 O/Ce geothermometer is no longer valid and can not be applied to back-arc slab-derived fluids Cooper et al., 2012). However, the  . Models by 2004) show thermal differences between the plate edge and plate center, based on both the rollback and downdip motions of the subduction slab. The transition from downdip motion to rollback motion leads to an increase of 3°C in the SST in the center of the slab, which, when scaled to mantle temperatures, is an excess of 200°C. A schematic of the thermal variations is shown in Figure 3.8b, which simplifies the thermal structure of the subducting slab into two end members. The subducting slab thermal variations explain the observed back-arc segments SST 4GPa variations relative to the arc SST 4GPa .
The third scenario addresses the possibility in which the 2D thermal models predict too cold geotherms for the subducting slabs. While laboratory models have shown that 3D slab subduction has thermal variations within the slab, the laboratory models, when scaled to mantle values, have also been found to have higher temperatures than 2D numerical models . The inclusion of back-arc spreading in numerical modeling increases the mantle wedge temperature , which in turn increases the subducting slab surface temperature through coupling and thermal exchange between the wedge and slab surface particles (Kincaid and Hall, 2004). Therefore, if the current models of slab surface temperatures predict too cold SSTs, warmer SSTs might help with projecting the back-arc basins SSTs to depth. Warmer SSTs will enable the projection of BABB fluids to temperatures that intersect the subducting slab at depths constrained by the thermal models (<8 GPa).
The most likely scenario for the origin of the back-arc basin fluids is a combination of the last two scenarios, in which the thermal models may underpredict SST and the slab-derived fluids reflect thermal variations in the subducting slab (see section 4.2;Figures 3.9 and 3.10). One important consideration is that the thermal models from Syracuse et al. (2010) are 2D models and may not capture the 3D thermal variations, and therefore may also underpredict the SST. Figure 3.11 shows the potential scenario where back-arc basin fluids originate from depth and reflect potential thermal variations due to 3D flow around the edges of the subducting slab.
The motion of the subducting plate plays a significant role in the thermal structure of the mantle wedge, where only down-dip motion of the subducting plate results in warmer plate edges and cooler center, and roll-back motion of the subducting plate results in cooler plate edges and warmer center .

Conclusions
The geochemical data, combined with the application of the H back-arc fluids come from the same depth as arcs but from hotter edges of the slab, or (3) thermal models predict slab surface geotherms that are too cold. The first scenario is unlikely as fluids projected to depths greater than 8 GPa are likely to not be saturated in monazite, and therefore the H 2 O/Ce geothermometer is not applicable.
However, the second two scenarios are not mutually exclusive and therefore, the origin of the back-arc basin fluids is a combination of thermal variations caused by 3D flow and 2D thermal models underpredicting the slab surface temperatures.   . Data from MTJ is whole rock trace element data (Falloon et al., 1992) and glass H 2 O data . Basin averages are calculated weighted average from individual spreading segment averages.                                     Cooper et al. (2012) and the recalculated temperature using Agrigan data from , projected to a depth of h, the depth from volcano to slab surface as determine by Syracuse et al. (2010). B) East Scotia Ridge. The arc PT value (black cross) is from the D80 thermal models (Syracuse et al., 2010). The blue cross is the projected SST for the E9 segment, the only segment cool enough to intersect the SST at depth. C) Manus Basin. The arc PT value (black circle) is from the D80 thermal models (Syracuse et al., 2010). The gray circle is the projected temperature to a depth of h, the depth from volcano to slab surface as determine by Syracuse et al. (2010). D) Lau Basin. The arc PT value (black diamond) is from Cooper et al., 2012. E) North Fiji Basin. The arc PT value (black elongated square) is from the D80 thermal models (Syracuse et al., 2010). The gray elongated square is the projected temperature to a depth of h, the depth from volcano to slab surface as determine by Syracuse et al. (2010).  Cooper et al., 2012). The black symbols are the arc temperatures from Cooper et al., 2012 and the purple symbol for South Sandwich Arc is the temperature from the D80 thermal model (Syracuse et al., 2010) and the pink symbols for New Britain Arc and Vanuatu Arc are calculated using whole rock trace element data and assume H 2 O content from 4.0 wt.% (Woodhead et al., 1998;Raos and Crawford, 2004). Data from the back-arc basins constrains the temperature range, where the blue symbols are the minimum temperature constraints from the sample with the highest H 2 O/Ce value, the purple symbols are the averages calculated from the segment averages, and the green symbols are the maximum temperature constraints from the sample with the lowest H 2 O/Ce value.  During the process of sample analysis by laser ablation ICP-MS (LA-ICP-MS), three spots are typically collected on unknown glass sample chips. Upon analysis of this collected laser data, a mismatch between spots on unknown samples was observed. The first spot (spot 1) always had lower concentrations for rare earth elements (REE) and the difference between concentrations for spot 1 and spots 2 and 3 was greater for the light REE (LREE). The observed lack of correlation between spots was found to be present only in the first analyzed sample after sample mounts were switched and the sample chamber was exposed to the ambient atmosphere. Several tests were performed to determine both the cause of the spot disparity and arrive at an appropriate solution to avoid the lack of correlation between spots on future analyzes.
As noted, the first spot also has lower REE concentrations, and preventing the lower concentrations observed in spot 1 will also be significantly important for samples such as melt inclusions, in which only one spot can be analyzed.
The first test involved multiple steps starting with three runs of the standards in the "normal" order and one run of the standards in the "reverse" order, as the standards are typically analyzed in a set order and could be influencing the data analysis. The next step was three separate tests to test three different variables including (1) the number of times the laser was purged, (2) the wait time between sample mount changes, and (3) the "warmness" of the laser. Each test had a normal run that analyzed the standards and 3 spots each on four unknowns. Test 1 consisted of a normal run then a sample mount switch followed by one purge of the laser system, a 15 minute wait time, and reanalysis of the four unknowns and then another sample mount switch followed by one purge, no wait time, and reanalysis of the four unknowns ( Figure A1). Test 2 consisted of a normal run then a sample mount switch followed by two purges of the laser system, a 15 minute wait time, and reanalysis of the four unknowns and then another sample mount switch followed by two purges, no wait time, and reanalysis of the four unknowns ( Figure A1). Test 3, shown in Figure   A2, consisted of a normal run then a sample mount switch followed by two purges of the laser system, a 15 minute wait time, 5 minute laser warm-up time, and reanalysis of the four unknowns and then another sample mount switch followed by two purges, a 15 minute wait time, 10 minute laser warm-up time, and reanalysis of the four unknowns.
Upon analysis of these tests, the analysis order of the standards had no impact on the observed concentrations or sample spot disparity. The number of purges performed on the laser system and the length of time for laser warm-up appeared to be insignificant compared to the amount of time spent waiting between sample mount changes post-purge. The samples analyzed following a 15 minute wait time had no disparity between spots on the first sample analyzed, while the samples analyzed following no wait time had a noticeable disparity between spot 1 and spots 2 and 3.
Further tests were performed to determine the amount of time required to wait before continuing sample analysis after changing sample mounts. All tests had two purges of the laser system following sample mount change before analysis continued and the first test had a two minute wait time, while the second test had a five minute wait time and the third test had a ten minute wait time. The analysis of these tests (see Figure A3) showed that the ideal sample wait time was 10 minutes ( Figure A3c) as both the two minute ( Figure A3a) and five minute ( Figure A3b) wait times had noticeable differences between spot 1 and spots 2 and 3, although spot 1 was closer to spots 2 and 3 after a five minute wait time rather than a two minute wait time.
Part 2: Laser Tuning Tests: A complication that arises during tuning is that NIST 612 is heterogeneous and does not ablate evenly. Six additional natural glass samples (NLD-65-01-01, RC2806 16D-1g, EN026 10D-3g, RC2806 4D-3g, RC2806 40D-9g, Kilauea glass) were analyzed for comparison of count rates to the known tuning benchmark count rates on NIST 612. The six natural glass samples, while ablating consistently, have the problem of crystal inclusions, not exposed at the surface. The results of three analyzes are shown in Table A1 and RC2806 40D-9g has the highest sensitivity on 139 La out of all the natural glass samples. The counts on 139 La for RC2806 40D-9g are similar to the counts observed on NIST 612, providing a similar tuning reference range with similar sensitivity but a more reliable ablation pattern.

Part 3: Laser Conditions Tests:
The three main laser properties which affect the count rate and sensitivity of laser ablation ICP-MS are spot size (µm), repeat rate (Hz), and laser energy (%). Spot size ranges from 5 -160 µm, repeat rate ranges from 1 -20 Hz, and laser energy ranges from 50 -100%. The three laser properties affect count rate and can vary between different standards. While the count rate on the standards will change as properties change, ideal laser settings will maintain a constant relationship between standards. Providing constraints on the behavior of each of the laser properties will determine the range of conditions under which the laser can be used, while maintaining accuracy as the ratio between samples remains constant. During data reduction, samples are normalized to CaO, which removes the variation in signal, providing a constant reference in which to analyze the effects of the three laser properties.
Each of the three laser properties were tested individually and were focused on the three USGS standards (BIR-1g, BHVO-2g, BCR-2g). The first test ( Figure A4a) analyzed each USGS standard with the following spot sizes: 5 µm, 8 µm, 12 µm, 16 µm, 20 µm, 30 µm, 40 µm, 60 µm, 80 µm, 100 µm, and 120 µm. Looking at a low abundant element such as U normalized to CaO, the ratio between BIR-1g, BHVO-2g, and BCR-2g is found to be largely unchanged above 20 µm. The second test analyzed the USGS glasses with the following settings for repeat rate: 1 Hz, 2 Hz, 4 Hz, 5 Hz, 10 Hz, and 20 Hz. As shown in Figure A4b, a repeat rate of 5 Hz or greater is found to provide a constant ratio between BIR-1g, BHVO-2g, and BCR-2g for U normalized to CaO. The third test also analyzed the USGS standards while increasing the laser energy from 0% to 100% in 5% increments. Again, looking at U normalized to CaO in Figure A4c, it is determined that U is not detected until the laser energy reaches 50%, after which the ratio is largely unchanged above 50%.
Analysis of the three USGS glass standards provided constraints on the accuracy of results under varying conditions. This provides good constraints for typical glass analyzes, but for analysis of materials such as melt inclusions, where there is little material available for ablation, the length of time the material is ablated must be maximized. However, that requires a better understanding of how the laser properties interact and jointly affect count rate, sensitivity, and laser drilling rate (i.e., the length of time it takes to drill through a sample). This round of testing was conducted in two parts focusing on interaction between spot size, repeat rate, and laser energy and finally the effect these properties had drilling rate. Table A2 and Figure A5 show the results from three tests in which (1) the spot size was changed while keeping repeat rate and laser energy constant, (2) the repeat rate was changed while keeping spot size and laser energy constant, and (3) the laser energy was changed while keeping spot size and repeat rate constant. These tests showed that the laser energy output (mJ) increased with both increasing spot size (µm) and repeat rate (Hz). When changing the laser energy (%) from 50% to 80%, the laser energy output (mJ) increased, but further increases in laser energy (%) had little change on the laser energy output (mJ). Overall, increasing each of the settings will increase the laser energy output (mJ), thus increasing the amount of material ablated and increased sensitivity.
The second part of the testing focused on what effect each of the three laser properties had on the laser drilling rate (µm/s), shown in Table A3. The first test determined the effect of a changing spot size on drilling rate, in which it was observed, in Figure A6a, that the drilling rate remained relatively constant at ~2 µm/s over the entire range of spot sizes (5 -160 µm). The second test looked at the effect of a changing repeat rate on drilling rate and found that the drilling rate increased from ~1 325 µm/s at 5 Hz to ~4 µm/s at 20 Hz, shown in Figure A6b. Finally, Figure A6c shows the third test, which looked at the effect of increasing the laser energy (%) on the drilling rate, and found a very minimal change from ~2 µm/s at 55% energy to ~3 µm/s at 80% energy. Therefore, to decrease the laser drilling rate, the best option is to decrease the repeat rate from the default of 10 Hz to 5 Hz, which decreased the drilling rate by 50% or from ~2 µm/s to ~1 µm/s.

Part 4: Data Normalization Tests:
Data reduction in LasyBoy requires normalizing the laser data to major element data, typically either CaO or TiO 2 . Comparison of data reductions by CaO and TiO 2 showed that CaO content is the better normalizing value. Ca is a major element and as such is consistent within the mantle and Ti varies in the mantle and as such may not be an ideal consistent measurement for calibration. Using CaO normalized laser data, the experimental major element laser data (specifically SiO 2 ) was found to be closer to accepted microprobe data and the LREE pattern is smoother (see Figure A7 for REE comparison between Ti and Ca normalization for three samples).       Figure A5: Plots of laser properties vs. laser energy output (mJ). A) Plot of spot size (μm) vs. laser energy output. Increasing spot size from 5 -160 μm corresponded to an exponential increase in laser energy output. B) Plot of repeat rate (Hz) vs. laser energy output, where repeat rate ranged from 5 -20 Hz. The increase in repeat rate corresponded to a small increase in laser energy output. C) Plot of laser energy (%) vs. laser energy output, where laser energy ranged from 50 to 100% energy. Laser energy output increased to increases in laser energy from 50 -70% and then leveled out from 70 -100% laser energy. Figure A6: Plots of laser properties vs. laser drilling rate (μm/s). A) Plot of spot size (μ m) vs. drilling rate, where spot size ranged from 12 -60 μm and showed a consistent drilling rate of ~2 μm/s. B) Plot of repeat rate (Hz) vs. drilling rate. Repeat rate ranged from 5 -20 Hz, which corresponded to an increase in drilling rate from ~1 μm/s to ~4 μm/s. C) Plot of laser energy (%) vs. drilling rate, where laser energy ranged from 55 to 80% energy. Drilling rate showed a slight increase from ~2 μm/s to ~3 μ m/s.  18. To increase the signal, start with the XSII stage alignment wizard under the Music Note symbol, which will optimize the horizontal and vertical settings for the torch box, as opening the ICP-MS up to check/change cones could jiggle the torch box position.
19. Next run the XSII XS-1 ppb KT. This will optimize the settings for the lenses and the nebulizer gas flow. 20. After you have ensured that the counts are maximized on 115 In, save the final settings.
21. Select "Template" (1) then "Solution Startup" (2)  Add the experiment to the queue by clicking the "Q" button (4) by the green "On" and red "Off" buttons. Save the experiment in the "Solution Startup Scans" folder, with the name of "Startup MMDDYY." Hit append and then okay to the warning message. Once the ICP-MS software has a box that says "Place sample probe into sample. Tune A 1ppb", click continue.
22. Go to the "Results" tab and then the "Mass Uncorrected" Tab. Check the average sensitivity on 115 In and 238 U for all 5 runs ( > 250,000 counts) and calculate the CeO/Ce ratio (< 2%), Ba ++ /Ba ratio (< 5%) and check that mass 220 background is less than 0.5. Element stability (% RSD) should be less than 2. If the oxide ratio is too high, retune and turn down the nebulizer gas or increase the sampling depth. Once you have a successful startup scan, print the results by going to the "Reports" tab. Select "Instrument Configuration" and "Mass Uncorrected ICPS" then hit the refresh button and then print. 23. After a successful tuning, wipe the probe off and place in the 250mL Tune F bottle to perform the detector cross calibration.
on the autosampler. 4. Under "Setup" then "Timings" and under "Maximum delay" make sure uptake time is 60s and washout time is 120s.
5. Go to "Setup" then "Analyte" and make sure all the elements you wish to analyze are selected. Typically analyzed elements are Li, Be, K 2 O, Sc, TiO 2 , V, Cr, Co, Ni, Cu, 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 for basaltic rock samples. 6. If you modify the sample list, go to "Setup" then "Acquisition Parameters" tab and make sure to change the number of scans to equal 60s and that the newly added elements have 3 channels selected, not the default of 1. 7. Under "Sample List" will be the appropriate setup for a 10 unknown sample run. The samples are run in blocks of 4 samples bracketed by a drift. The layout (see table below) is a 10 sample run, in which each unknown sample is analyzed twice and within each run of every sample, it is run three times and an average is reported for that sample run. 10. After you have placed all the samples in the rack and on the autosampler, confirm sample location with the information you have filled out on your experiment sample list. 11. The column height for all the samples should be 140mm not 144mm. and then okay.
2. If there is no increase in signal or counts displayed in the Real Time Display once placed in 1ppb Tune A, check that the tubes are tighten down enough on the peristaltic pump. Also check that the spray chamber has a good seal with the gray connector tube by slowly rotating the spray chamber angle and see if the observed counts increase. 3. When changing pump tubing, make sure to carefully heat the pump tubing enough to insert the autosampler probe, spray chamber, or drain tubing into the peristaltic pump tubing without too much trouble. The peristaltic pump rotates clockwise. The colored tubing should have the yellow clip side going to the autosampler probe and the orange clip side going to the spray chamber. The drain tubing should go from the spray chamber to the peristaltic pump and then to the drain. 4. If running extremely sensitive samples, such as Ti, change the pump tubing before every run. 5. Make sure there are no visible bubbles entering the spray chamber. 6. Run a mass calibration () if the sensitivity is lower than normal and suspect the ICP-MS is not sampling the plateau of the peak. a. Use Tune A solution. b. High resolution peak width: 0.3 -0.4. c. Standard resolution peak width: around 0.7. d. Error should be around zero. 7. If the nebulizer is clogged and the nebulizer spray is pulsing.
a. Find a small, never used plastic syringe. b. Affix a short length of stretchy plastic tubing to the nozzle, of a size that will fit snugly over the end of the nozzle and over the ends of the openings on the nebulizer. c. Mix up a small volume (~125mL maximum) of 50% nitric acid and set up a secondary containment tray for the procedure.
Connect to pump and select play Accessory window d. Draw up some acid into the syringe and connect it to the sample introduction end of the nebulizer. Gently push acid through until it drips from the spray tip. e. Do the same for the leg that connects to the nebulizer gas, pushing acid through until it drips from the tip. f. Let the nebulizer sit for at least 30 minutes. g. Connect the syringe to the spray trip of the nebulizer and gently push more nitric acid through the nebulizer in the reverse direction and confirm that acid drips from both openings on the nebulizer. h. If acid does not flow through, try increasing the pressure, or switch the syringe to the other openings and try in the normal flow direction. i. When unclogged, flush the nebulizer several times using milli-Q water, following the same approach you did with the nitric acid. 8. If when running a solution, the autosampler has an error saying "Invalid Tray Type", stop the experiment and make sure that you have selected the appropriate tray in both the ESI software and in PlasmaLab. To change the tray type in ESI, click on the appropriate tray rack and then select tray size (5 x 12 for a 10 unknown sample run).
To change the tray type in PlasmaLab, select the accessories window, connect to the autosampler, and then under the appropriate tray number, select the tray size. Hit okay and then restart the experiment.
11. Switch back to the 'Major" tab and make sure the nebulizer gas level is set between 0.90 and 1.05.
12. Put the ICP-MS into operate mode by hitting the green "On" button. 13. Once in operate mode, slowly turn up the helium gas to full (read out of 760). Click the right arrow for the helium gas, which will increase the level in helium in approximately intervals of three. This should be about one click per every couple seconds and take about two minutes to reach full. 14. After the helium gas is at full, give the ICP-MS 30 minutes to warm up. 15. While the ICP-MS is warming up, go back to the laser software and: a. Turn up the co-axial light until you can see the samples in the viewing window of the software b. Focus the optics using the arrows in the top left of the viewing window c. Set the repeat rate at 10 Hz, the laser energy at 65%, and the spot size at 60µm d. Using the sample map button, make a map of your samples. Change the number of images and then select "make new map." For a 1" round mount, an 18 image height x length map covers majority of the sample mount. e. If the map does not show up in the sample map window, move around on the 1" mount and the map will fill in on the sample map window.
Make sure gray box is in this position once you have moved to NIST 612.
17. Focus on NIST 612, then place a line on NIST 612 in a region that has not been previously lasered.
18. Select the line and check that the properties of the line match the properties you have set in the laser software (repeat rate, laser energy, and spot size). The line ablates at 1µm/s.

Select Standards
Previously lasered region Navigation controls 24. Click OK on the "Place sample probe into sample. NIST 612" and then immediately click "Run Scans." The two boxes must be clicked right after each other to tell both the ICP-MS to start collecting data and the laser to start ablating the sample. 25. After the second run of data is collected, check the sensitivity on 139 La by going to "Results" then "Numerical Results" then "Mass Uncorrected ICPS" Save the final settings 1 2 4 3 tab and calculate the ThO/Th ratio (< 2%). Mass 220 measures the background. If the oxide ratio is too high, retune and turn down the nebulizer gas or increase the sampling depth.
Once you have a successful startup scan, print the results by going to the "Reports" tab. Select "Instrument Configuration" and "Mass Uncorrected ICPS" then hit the refresh button and then print. 26. After a successful tuning, in the laser software, use the position button to move to NIST 610 to do a cross calibration. Make sure the laser is focused then set the spot size to 160µm. Do not place an actual spot on NIST 610. 27. Back in PlasmaLab, go to the  and select "set up the detector." Select X-Cal which determines the proportionality between analogue and pulse count detectors as analogue measures the elements at the half-way point (high concentration elements) and pulse measures the elements reaching the detector 5. Under "Sample List" the first ~8 samples should be standards. Then you can type in your sample list. Record information such as repeat rate and spot size either in sample name on the sample list or in a notebook. 6. Once you are ready to start, add the experiment to the queue. Save the file in the folder with your name and give the file a name that identifies the nature of the samples and the current date. Click append and then okay to the warning message. 7. Once the ICP-MS software has a box that says "Place sample probe into sample. Sample Name", switch to the laser software and use the position button to navigate to the standard or use the sample map to navigate to your sample. Focus the laser then place a spot. Check the properties on the spot. Make sure they match the laser properties. 8. If you lasering a glass wafer or a melt inclusion, change the repeat rate to 5Hz. 9. Ensure that there is only one spot listed in the scan patterns box. 10. Make sure you have the run scans box open in the laser software. Click OK on the "Place sample probe into sample. Sample Name" and then immediately click "Run Scans." The two boxes must be clicked right after each other to tell both the ICP-MS to start collecting data and the laser to start ablating the sample. 11. Switching 1" round sample mounts a. Put laser in bypass b. Change out mount, make sure it is flat in the sample holder, and use isopropanol to make sure sample surface is clean c. Once the sample holder is reloaded, purge the laser twice and wait 10 minutes before starting analysis.
12. Check the laser results as you go. Go to "Tools" (1) then "Calculate Results". Then within the experiment go to "Setup" then "Acquisition Parameters" (2) and remember to log the y axis (3). Make sure you see data here before moving on to the next sample! a c: 2X "Instrument Configurations" tab.
3. If you are adding sample names as you go, and then if the experiment comes up with a message stating "Experiment is no longer in the queue" and you still have samples to analysis in that experiment, add the sample names to the list and then click on queue (Q) and hit append then okay. The experiment will now be back in the queue. 4. If you have put the laser into bypass and did not purge the laser before running a spot, the laser will automatically start purging before firing the sample. 5. If you hit "Run Scans" and the laser does not fire, make sure the laser has been enabled. 6. If you have focused the laser and then when you go to fire, the focus changes make sure you place your spot after you focus. The sample spot will reference whichever focus level you have the optics on when you place the spot. 7. If the properties for the spot size are different than the ones set in the laser, the laser will rotate to the appropriate spot size while the ICP-MS is collecting data, taking away from data analysis time. 8. If the lights do not turn on (light interlock), check that the door on the right side of the laser base is fully closed. 9. If there is extremely low sensitivity while tuning, make sure the sample holder is fully seated and flush with the laser. 10. If the helium runs out while analyzing, the torch will be knocked out. This will require a simple change of helium tanks and restart of the ICP-MS. 11. If the argon runs out while you are running, the torch will be knocked out and cause an argon interlock. Take the following steps to fix the problem. a. Abort the experiment and make sure the ICP-MS is properly in vacuum mode. b. PlasmaLab will not show the Real Time Display (RTD). c. Stop the PlasmaLab services. d. Turn off the PC in the ICP-MS.

Accessory Window
Connect to pump and select stop e. Unplug the yellow cable and the red/black cable which controls the lens module. f. Wait 30 seconds then reconnect the red/black cable and the yellow cable. g. Turn the PC back on. h. Restart PlasmaLab services. i. Under the "Advanced" tab, turn the lens back to on from standby. j. RTD should start now. k. If there is a problem with data collection and the experiment not starting, put the ICP-MS into vacuum mode, then turn the ICP-MS back on (operate mode), and the experiment should start and collect data now.
Maintenance 1. Changing the Laser Water Filter yearly a. From the task bar at the top of the NWR menu, choose LASER and uncheck the first option, Enable Laser Power Supply. This turns the water pump off. Disconnect the RED hose on the back of the Power Supply. Some water that is already in the lines will drip out but not too much. b. Place this RED hose into a bucket, only about 1 liter of water will come out total. c. Now, recheck the Enable Laser Power Supply and the water pump will turn on for about 1 or 2 seconds and will pump some water into the bucket. The interlock screen will come on because water is not going back into the system. d. Un-check Enable Laser Power Supply and then Re-check it. Some more water will come out. Do this about 10 to 12 times, eventually you will see only air coming out of the hose. Un-check the Enable Laser Power Supply. 95% of the water is now drained. e. The DI Cartridge is inside the Power Supply. Remove the cover (12)(13)(14)(15)(16) total screws, 3-4 on each side along the bottom, 3-4 on each side on the back). Remove two screws which hold the cartridge bracket in place. Replace the DI Cartridge. f. Use Teflon thread tape to sealcareful not to over-tighten! g. Plug the RED hose back into the Power Supply 10. After you have placed all the samples in the rack and on the autosampler, confirm sample location with the information you have filled out on your experiment sample list. 11. The column height for all the samples should be 140mm not 144mm. 12. Once you are ready to start, add the experiment in the queue. Click append and then okay to the warning message. Make sure the experiment starts correctly and the autosampler probe goes to the correct sample. 13. Once finished with the entire experiment file, go to "Results", and copy the entire "Mass Uncorrected" data file, including the headers, to an excel file.
2. If there is no increase in signal or counts displayed in the Real Time Display once placed in 1ppb Tune A, check that the tubes are tighten down enough on the peristaltic pump. Also check that the spray chamber has a good seal with the gray connector tube by slowly rotating the spray chamber angle and see if the observed counts increase. 3. When changing pump tubing, make sure to carefully heat the pump tubing enough to insert the autosampler probe, spray chamber, or drain tubing into the peristaltic pump tubing without too much trouble. The peristaltic pump rotates clockwise. The colored tubing should have the yellow clip side going to the autosampler probe and the orange clip side going to the spray chamber. The drain tubing should go from the spray chamber to the peristaltic pump and then to the drain. 4. If running extremely sensitive samples, such as Ti, change the pump tubing before every run. 5. Make sure there are no visible bubbles entering the spray chamber. 6. Run a mass calibration () if the sensitivity is lower than normal and suspect the ICP-MS is not sampling the plateau of the peak. a. Use Tune A solution. b. High resolution peak width: 0.3 -0.4. c. Standard resolution peak width: around 0.7. d. Error should be around zero. 7. If the nebulizer is clogged and the nebulizer spray is pulsing.
a. Find a small, never used plastic syringe. b. Affix a short length of stretchy plastic tubing to the nozzle, of a size that will fit snugly over the end of the nozzle and over the ends of the openings on the nebulizer. c. Mix up a small volume (~125mL maximum) of 50% nitric acid and set up a secondary containment tray for the procedure. d. Draw up some acid into the syringe and connect it to the sample introduction end of the nebulizer. Gently push acid through until it drips from the spray tip. e. Do the same for the leg that connects to the nebulizer gas, pushing acid through until it drips from the tip. f. Let the nebulizer sit for at least 30 minutes. g. Connect the syringe to the spray trip of the nebulizer and gently push more nitric acid through the nebulizer in the reverse direction and confirm that acid drips from both openings on the nebulizer. h. If acid does not flow through, try increasing the pressure, or switch the syringe to the other openings and try in the normal flow direction.
i. When unclogged, flush the nebulizer several times using milli-Q water, following the same approach you did with the nitric acid. 8. If when running a solution, the autosampler has an error saying "Invalid Tray Type", stop the experiment and make sure that you have selected the appropriate tray in both the ESI software and in PlasmaLab. To change the tray type in ESI, click on the appropriate tray rack and then select tray size (5 x 12 for a 10 unknown sample run).
To change the tray type in PlasmaLab, select the accessories window, connect to the autosampler, and then under the appropriate tray number, select the tray size. Hit okay and then restart the experiment. not open, just a screwdriver as a lever on one of the lid catches to separate the lid from the base. e. The detector is located on the far left of the high pressure vacuum chamber. Confirm orientation before removing the detector. f. Wearing gloves replace the new detector and ensure the "feet" are in the grooves. g. Close lid and fasten the three metal clamps. h. In PlasmaLab click "On" and say yes to the message asking to put the instrument into "Vacuum Mode." i. PlasmaLab will ask if a new detector has been fitted. Click yes and then enter the serial number and model number, provided with the new detector. j. The detector will now outgas overnight (~7 hours).