The volcanoes of Maui Nui (West Moloka'i, East Moloka'i, Lana'i, West Maui, Haleakala, and Kaho'olawe) record Hawaiian magmatism at ∼1–2 Ma. Lavas from these volcanoes nearly span the compositional range erupted from all the Hawaiian volcanoes over the past 5 Myr and represent both the Kea and Ko'olau compositional end-members of Hawaiian lavas. Many aspects of major and trace element and isotope compositions of Hawaiian shield-stage lavas are consistent with ancient, recycled oceanic lithosphere in the plume sources of Kea- and Ko'olau-type magmas (Lassiter and Hauri, 1998; Blichert-Toft et al., 1999). Hypotheses that describe the compositional range of Hawaiian lavas as originating from ancient oceanic lithosphere in the Hawaiian plume implicitly or explicitly infer lithologic heterogeneity in the plume. We present trace element models for the origin of these end-members that explicitly address the petrologic complexities of melting eclogite (derived from ancient oceanic lithosphere) in the plume. Trace element (La/Nb, Sm/Yb, Sm/Hf, and Sm/Nb), major element, and isotope compositions of Lana'i, which erupts dominantly Ko'olau-type lavas, are consistent with the origin of these lavas in large-degree (∼60–70%) melts of ancient upper oceanic crust (basalt + sediment) that mix with plume-derived Haleakala-type melts. Trace element (Sm/Yb, Hf/Zr, and Hf/Nb) and isotope compositions of West Maui and East Moloka'i, which erupt dominantly Kea-type magmas, are consistent with an origin in ancient depleted oceanic lithosphere that has been refertilized with moderate-degree melts (10–30%) of associated crustal gabbro. The physical mechanisms (melt-melt versus melt-solid mixing) through which the oceanic crustal components melt and mix within the plume lead to the generation of isotopically homogeneous Kea-type lavas and isotopically heterogeneous Ko'olau-type lavas. The volcanoes of Maui Nui record the exhaustion of the Ko'olau component and the initiation of the Kea component as dominant compositional end-members in the Hawaiian plume.
 Isotopic variability in Hawaiian shield-stage lavas typically is modeled as mixing between at least two, sometimes up to five or six, distinct compositional components or end-members that originate in the Hawaiian plume, Pacific lithosphere, or local asthenosphere [e.g., Chen and Frey, 1985; Hauri, 1996; Eisele et al., 2003]. Although these end-members are commonly described on the basis of their radiogenic isotope composition, some studies also ascribed to them distinctive major or trace element or stable isotope characteristics as well [e.g., Frey et al., 1994; Eiler et al., 1996; Hauri, 1996]. These compositional end-members, their sources and their relative contributions to shield-stage magmas have been defined to best describe the range of Hawaiian lavas erupted over the past 5 Myr, the time frame sampled by the best-studied and youngest Hawaiian islands and seamounts: Kaua'i at the oldest end and Lo'ihi at the youngest (Figure 1).
 Hypotheses that describe the compositional range of Hawaiian lavas as originating from ancient oceanic lithosphere in the Hawaiian plume implicitly or explicitly infer lithologic heterogeneity in the plume [e.g., Hauri, 1996; Lassiter and Hauri, 1998]. Eclogite derived from ancient oceanic lithosphere may contribute unique chemical signals to plume-derived melts that stem from the oceanic crust source of eclogite as well as mineralogic control by the dominantly clinopyroxene-garnet composition of eclogite, mineral composition-sensitive trace element partition coefficients, and major-element composition-dependent thermodynamic properties [e.g., Hirschmann and Stolper, 1996].
 The distribution of lava compositions in space and time provides an indirect record of the temporal structure of the Hawaiian plume. These data have been used to describe the general sense of compositional variability among the Hawaiian volcanoes over the past 5 Myr, the detailed variability exhibited by individual volcanoes, and the structure of the Hawaiian plume within the past 1 Myr (through the volcanoes of the Big Island). However, there does not yet exist an analogous comprehensive study, utilizing isotope, major and trace element compositions, of plume structure prior to ∼1 Ma. The volcanoes of Maui Nui (West and East Moloka'i, Lana'i, West Maui, Kaho'olawe and Haleakala; also known as the Maui Volcano Complex) provide an opportunity to explore the detailed structure of the Hawaiian plume at ∼1–2 Ma. The compositions of the shield-stage lavas of Maui Nui volcanoes span nearly the range of compositions erupted across the whole Hawaiian chain and there now exists a data set that is comprehensive enough to warrant a more detailed analysis of the Hawaiian plume as it was sampled by the Maui Nui volcanoes.
 This study complements the current detailed understanding of Hawaiian plume activity and structure over the past 1 Myr, as expressed on the Big Island of Hawai'i [DePaolo et al., 2001; Blichert-Toft et al., 2003; Eisele et al., 2003]. Maui Nui includes the youngest Hawaiian volcano to erupt extremely enriched isotope compositions (Lana'i), as well as the oldest volcano to erupt extremely depleted isotope compositions (East Moloka'i). This suggests that plume structure, composition and perhaps magma generation processes during Maui Nui time may have important contrasts with the Big Island.
 Only a few dates exist for shield-stage lavas from the Maui Nui volcanoes. In general, shield-stage lavas erupted from 2 to 0.7 Ma [McDougall, 1964; Naughton et al., 1980] (Figure 1). Several of these volcanoes overlap in age, and it is likely that each volcano erupted contemporaneously with its neighbor(s) for at least part of its growth. By analogy to the Big Island, where Mauna Loa, Kilauea and Hualalai all have had historic eruptions, contemporary shield-building activity of two or more volcanoes should be a common process. The volcanoes of Maui Nui span a greater compositional diversity than observed in the Big Island volcanoes, and may have sampled sources that are not reflected in these younger volcanoes. However, only the youngest shield-stage lavas of any volcano are exposed at the surface, and with few exceptions (e.g., Hawai'i Scientific Drilling Project (HSDP), submarine landslide blocks, distal submarine rift-zone segments), it is possible to obtain samples only of the latest-shield stage lavas of each volcano. Thus generally one can compare the same episode of relative shield growth for each volcano: the waning shield-stage magmatism.
 In this paper, we (1) describe the compositional range of lavas erupted from the Maui Nui volcanoes, including summaries of new chemical data (Appendices A and B), (2) identify mixing relationships among compositional end-members that are consistent with observed variability in Maui Nui, (3) synthesize trace and major element compositional data with recent experimental results to model petrologic and compositional variability in the mantle source and the related controls on partial melting processes, and (4) discuss heterogeneity in the Hawaiian plume evident on intravolcanic and intervolcanic time and length scales.
2. Hawaiian End-Members, Their Compositions, and Sources: A Review
 Most studies of isotopic variability in shield-stage lavas of Hawaiian volcanoes conclude that three primary isotopic end-members are necessary to account for the compositional range of shield-building stage lavas across the Hawaiian chain [e.g., Stille et al., 1986; West and Leeman, 1987; Hauri, 1996; Eiler et al., 1996; Hauri, 1997; Mukhopadhyay et al., 2003]. These are commonly labeled the Kea (sampled dominantly at Mauna Kea, Kilauea, West Maui and East Moloka'i volcanoes), Ko'olau (sampled dominantly at Ko'olau, Lana'i and Kaho'olawe volcanoes) and Lo'ihi (sampled dominantly at Lo'ihi seamount) components [Hauri, 1996; Eiler et al., 1996, 1998]. The Kea component is defined by relatively depleted 87Sr/86Sr, 143Nd/144Nd and 176Hf/177Hf, and higher 206Pb/204Pb, whereas the Ko'olau component is defined by relatively enriched 87Sr/86Sr, 143Nd/144Nd and 176Hf/177Hf and lower 206Pb/204Pb [Lassiter et al., 1996; Eiler et al., 1996; Blichert-Toft et al., 1999]. The Lo'ihi component is intermediate to Kea and Ko'olau in all radiogenic isotope compositions except 3He/4He and 208Pb/204Pb, for which it is higher than what is possible to attain by mixing Kea and Ko'olau components [Staudigel et al., 1984; Eiler et al., 1998]. Principal component analysis has shown that the Kea and Ko'olau end-members can account for 89% of the compositional variability in shield-stage lavas [Eiler et al., 1996]. The Lo'ihi component accounts for an additional 4% of the variability [Eiler et al., 1996], and only makes a significant contribution at Lo'ihi, Mauna Loa, Haleakala and Kaua'i volcanoes.
 It is commonly inferred that the Hawaiian plume is approximately concentrically zoned, and that the Loa-trend volcanoes tap sources in the core of the plume and the Kea-trend volcanoes tap sources on the edge of the plume [Lassiter et al., 1996; Hauri, 1996; DePaolo et al., 2001]. One primary support for this axisymmetric plume model is the observation that all volcanoes erupting extreme Ko'olau-type compositions lie on the Loa trend, and volcanoes erupting extreme Kea-type lavas lie on the Kea trend (Figure 1). Both Loa and Kea trends contain volcanoes of intermediate composition. DePaolo et al.  mapped the isotopic composition of the Hawaiian plume on the basis of estimated magma capture areas, plume upwelling rates and Pacific plate motion and also concluded that the plume is approximately concentrically zoned. Contrasting models propose that the Hawaiian plume is characterized by axial asymmetry. Blichert-Toft et al.  use time-series analysis of Hf and Pb isotopes to argue that, because of the high upwelling velocity of the plume relative to the velocity of the Pacific plate, vertical heterogeneity (plug-flow) in the plume also makes a significant contribution to the chemical variability of the erupted magmas. Eisele et al.  and Abouchami et al.  conclude that, on the million-year time-scale, melting samples long, narrow heterogeneities that are strung out vertically in the plume. Although plume structure such as described in the latter two models could episodically produce lavas that appear to result from a concentric zonation in the plume, this would not reflect a fundamental aspect of plume structure. In the axial asymmetry models, the heterogeneities in the plume are inherited from the plume source region in the mantle, whereas in the concentric-zoning model, heterogeneities can be entrained along the periphery of the upwelling plume [Hauri et al., 1994].
3. Data Selection, Normalization, Filtering, and Assumptions
 For the following discussion, we have compiled from literature sources the available data for shield-stage tholeiites from Maui Nui volcanoes (Appendix A). We also report some new compositional data for Lana'i and East Moloka'i (Appendix B). We do not include or discuss any West Moloka'i data because very few data exist, and most rock exposures are severely weathered, so collection and analysis of additional samples is not feasible. We use all isotope data as they were reported in the original publications. Most of the Pb isotope data have been normalized to NIST SRM 981 values of Todt et al. , and some of the 87Sr/86Sr have been normalized to the Eimer and Amend SrCO3 standard (87Sr/86Sr = 0.70800) or to NIST SRM 987 (87Sr/86Sr = 0.710250). Normalization procedures and measured standard values are not given consistently in the literature from which these data come, thus we are not able to make a standard normalization for all the isotope data that we compare in this paper. While all of the West Maui samples, most of the Lana'i samples and some of the East Moloka'i and Kaho'olawe samples were leached in 6N HCl prior to Pb and Sr isotope analysis, it is unknown whether samples for the other volcanoes were treated likewise.
 We normalized all major element compositions to 100% totals for the oxides, with all Fe reported as FeOt. We corrected major element compositions for variable olivine fractionation or accumulation by adding or subtracting 1% increments of equilibrium olivine until the bulk rock composition is in equilibrium with Fo90 olivine. We also tested this correction using 0.1% increments of olivine, and found no significant difference in the results. For this correction, we assumed that Fe2+ = 0.9*FeOt, and that Kol-liqFe-Mg = 0.3 [Roeder and Emslie, 1970]. This correction is dependent on Mg/(Fe2+ + Mg) rather than only MgO composition of the melt, and thus preserves MgO variability inherent in the primary magmas. We excluded lavas with MgO < 7 wt. % to filter out whole rock compositions that may have been affected by fractionation of clinopyroxene in addition to olivine [Helz and Wright, 1992; Montierth et al., 1995]. To calculate average compositions for each volcano, we included all samples in the average, regardless of whether the sample has complete or partial elemental and isotopic data. This does not yield significantly different results from those obtained including only completely characterized samples.
 Subaerial, low-temperature alteration and weathering may affect major and trace element and isotopic compositions of Hawaiian lavas. K2O/P2O5 is commonly used as an index of alteration, as this ratio decreases during alteration. Samples with K2O/P2O5 > 1–1.25 are generally considered unaltered, whereas samples with lower K2O/P2O5 may have undergone some degree of subaerial weathering [Frey et al., 1994]. In the Maui Nui samples, none of the volcanoes shows a significant correlation between K2O/P2O5 and 87Sr/86Sr, 206Pb/204Pb or 208Pb/204Pb, suggesting that these isotopic compositions have not been significantly modified by alteration. Decrease in SiO2 is also an effect of low-temperature alteration. Of all the volcanoes, only Lana'i has samples (two samples) that show anomalously low SiO2 at low K2O/P2O5. For the following discussion, we do not include these two Lana'i samples. It is difficult to evaluate effects of alteration on isotopic composition in samples for which no major element data are available. However, for all volcanoes except Lana'i, the range and pattern of variability in isotopic composition of the samples without major element data is approximately the same as the samples with major element data, so we include all isotopic data for these volcanoes in the following analysis and discussion. The isotopic range of Lana'i samples for which no major element data exist is greater than the isotopic range of samples for which we can evaluate alteration. However, the strong correlation of 87Sr/86Sr with 143Nd/144Nd in all Lana'i samples is not consistent with any amount of subaerial weathering or seawater contamination, which would destroy this correlation. With the exception of the two Lana'i samples mentioned above, we include all data for shield-stage lavas from the sources cited in Appendix A in our analysis and discussion.
 Our olivine-fractionation correction implies that magmas formed in equilibrium with Fo90 olivine, typical of fertile mantle peridotite, corresponding to some of the most Mg-rich olivine phenocrysts observed in Hawaiian shield-stage lavas. If any of the magmas have a different source (e.g., eclogite, depleted peridotite), then this assumption may require reconsideration. An MgO content of 14–18 wt. % is commonly assumed for primary Hawaiian tholeiites [Chen, 1993; Clague et al., 1995; Norman et al., 2002]. These estimates are based on both glass compositions and Fo content of olivine phenocrysts. For the first part of the following discussion, we assume that the parental magmas for each volcano formed in equilibrium with Fo90 olivine, and then revisit the issue of the magma source lithology at the end of the discussion.
4. Compositions and Observations
 The Kea, Ko'olau and Lo'ihi end-members are all apparent in the isotopic compositions of Maui Nui volcanoes. The Kea component dominates both West Maui and East Moloka'i compositions. Lana'i is dominated by the Ko'olau component, and Kaho'olawe shows contributions from the Ko'olau component as well, but to a lesser degree than Lana'i. Haleakala has the elevated 208Pb/204Pb and 3He/4He unique to the Lo'ihi component, but is not as extreme in its composition as Lo'ihi lavas. Because the focus of this study is the composition and structure of the Hawaiian plume at 1–2 Ma, as sampled by the volcanoes of Maui Nui, we will define and discuss end-members for “Lana'i,” “Haleakala” and “West Maui/East Moloka'i.” These end-members are analogous to, but not always the same as, the Ko'olau, Lo'ihi and Kea components, respectively, as defined to describe the all-Hawai'i variation. We intentionally utilize this end-member definition specific to Maui Nui time to emphasize that there is no a priori reason to presume that mantle compositions available to sampling by Hawaiian volcanism are either homogeneous or constant with time. In the discussion, we address the distinctions and similarities of the Maui Nui versus all-Hawai'i end-members.
 Major element trends in Maui Nui volcanoes, for example as reflected in MgO versus SiO2 and MgO versus TiO2 (Figure 2), also show intervolcano correlation with isotopic and trace element compositions when corrected for effects of olivine crystallization (Figures 3 and 4). Lana'i, the lavas of which make up one compositional end of the Maui Nui isotopic array, has the highest average SiO2c, and lowest average MgOc, FeOc and TiO2c (all fractionation corrected; Figure 3). Haleakala, which is not extreme in any of its isotopic compositions (except perhaps for 208Pb/204Pb), has the lowest average SiO2c and highest average MgOc, FeOc and TiO2c (Figure 3). West Maui and East Moloka'i, which lie at an extreme end of the isotopic array, and Kaho'olawe, which does not, all have intermediate values of these major element compositions. West Maui and East Moloka'i have the lowest average Al2O3c (not shown), whereas Lana'i, Kaho'olawe and Haleakala all have similar Al2O3c. The Maui Nui volcanoes show no distinction in average CaOc or K2Oc + Na2Oc compositions (not shown). Although the major element differences among these volcanoes could result from distinct source melting histories (melt fraction or mean pressure of magma segregation), the isotopic compositions require that three distinct source components contribute to these lavas. Correlation of SiO2c and TiO2c with 206Pb/204Pb among Lana'i, Kaho'olawe and Haleakala show that mixing between at least two of these source components, or magmas derived from two source components, contributes to both major element and isotopic variability (Figure 4).
 Rare earth element (REE) patterns, as quantified by La/Yb and Sm/Yb, show modest variation among the volcanoes (Figure 5). West Maui, East Moloka'i and Haleakala have the steepest slopes (La/Yb ∼8, Sm/Yb ∼3.3) and Lana'i and Kaho'olawe have the shallowest slopes (La/Yb ∼4, Sm/Yb ∼2). However, the range of REE slopes for each volcano spans nearly the entire range for all Maui Nui volcanoes. Among the Maui Nui volcanoes as a whole, TiO2c is positively correlated and SiO2c is negatively correlated with both La/Yb and Sm/Yb.
 West Maui and East Moloka'i have 87Sr/86Sr, 143Nd/144Nd and 176Hf/177Hf at the depleted end of the Maui Nui array (Figures 6 and 7). The 87Sr/86Sr, 143Nd/144Nd and 176Hf/177Hf of these volcanoes show little variability (% variation in 87Sr/86Sr = 0.011% and 0.0071%, 143Nd/144Nd = 0.0035% and 0.0022%, and 176Hf/177Hf = 0.0047% and 0.0047% for West Maui and East Moloka'i, respectively; % variability defined as 100*standard deviation/average). 87Sr/86Sr-143Nd/144Nd in both West Maui and East Moloka'i, and 176Hf/177Hf-143Nd/144Nd in East Moloka'i are not correlated at the 99% confidence interval, and the 143Nd/144Nd-176Hf/177Hf correlation in West Maui, though significant, is poor (Table 1). In contrast, Lana'i and Kaho'olawe, which lie at the more enriched end of the Maui Nui array, have well-correlated isotopic compositions and show a larger degree of variability in 87Sr/86Sr, 143Nd/144Nd and 176Hf/177Hf (variation in 87Sr/86Sr = 0.030% and 0.010%, 143Nd/144Nd = 0.013% and 0.014%, and 176Hf/177Hf = 0.018% and 0.014%, respectively). 87Sr/86Sr and 143Nd/144Nd in Haleakala lavas do not show significant correlation, and they show approximately the same absolute range of variation in 87Sr/86Sr and 143Nd/144Ndas do West Maui and East Moloka'i (variation in 87Sr/86Sr = 0.010%, and 143Nd/144Nd = 0.045%). Only three 176Hf/177Hf analyses for Haleakala are available, and they show less variability (0.0034%) than for West Maui or East Moloka'i; this likely is in part an artifact of the small sample set. The 87Sr/86Sr, 143Nd/144Nd and 176Hf/177Hf of Haleakala fall close to the center of the Maui Nui range, displaced slightly toward the depleted end of the array.
Correlations that are statistically significant at the 99% confidence level are bold. Limits on the correlation coefficient above which the correlations are accepted as statistically significant at the 99% confidence level depend upon sample size, which ranges from n = 3 to n = 70 for the data we discuss [Sachs, 1984]. If a correlation is determined to be significant at the 99% confidence level, but the calculated Pearson's correlation coefficient differs by less than 0.1 from the acceptance limit, we consider the correlation to be significant but poor. These values are bold and italic.
 Although the 87Sr/86Sr, 143Nd/144Nd and 176Hf/177Hf correlations among Maui Nui volcanoes can be described by mixing between two end-members, the Pb isotope compositions require a third component. 206Pb/204Pb-208Pb/204Pb in lavas from West Maui, East Moloka'i and Kaho'olawe are significantly correlated, but not in Lana'i or Haleakala lavas (Figure 8, Table 1). Intravolcano variation in 206Pb/204Pb is ∼0.23% for West Maui, East Moloka'i and Haleakala, 0.4% for Lana'i and 0.8% for Kaho'olawe, whereas 208Pb/204Pb variability is ∼0.1% for all volcanoes except Kaho'olawe (0.24%).
 In contrast to 87Sr/86Sr, 143Nd/144Nd and 176Hf/177Hf, in Pb isotope space the composition of Haleakala is not intermediate to the other volcanoes. Its 208Pb/204Pb is elevated relative to 206Pb/204Pb. 206Pb/204Pb-87Sr/86Sr (not shown) and 206Pb/204Pb-176Hf/177Hf (Figure 9) variation in the Maui Nui lavas tend to define curved, rather than linear arrays. In 206Pb/204Pb-207Pb/204Pb (Figure 8) and 206Pb/204Pb-176Hf/177Hf (Figure 9) compositional space, Haleakala lies along the trend defined by other volcanoes, suggesting that it is 208Pb/204Pb rather than 206Pb/204Pb which is distinct at Haleakala.
 On the basis of small intravolcano variability in 87Sr/86Sr, 143Nd/144Nd and 176Hf/177Hf, West Maui, East Moloka'i and Haleakala appear either to have sampled distinct, homogeneous sources, or consistently and reproducibly sampled a mixture of sources in constant proportions. In contrast, the more linear arrays of Lana'i, and to a lesser extent Kaho'olawe, are consistent with the lavas from these volcanoes resulting from variable extents of mixing between two components. Further consideration of Pb isotope data (Figure 8) indicates that the West Maui/East Moloka'i end-member may be composed of subcomponents that are variable in their Pb isotope composition. West Maui lavas define a single linear array in both 206Pb/204Pb-208Pb/204Pb and 206Pb/204Pb-207Pb/204Pb space. East Moloka'i lavas define two arrays. One is coincident with the West Maui array, and the other is subparallel to West Maui but is slightly offset to higher 206Pb/204Pb. However, the East Moloka'i isotope data were analyzed in two different labs [Sawyer, 1999; Xu et al., 2005], so alternatively, the two arrays might be an analytical artifact. The intravolcano correlations of West Maui and East Moloka'i 206Pb/204Pb-208Pb/204Pb and 206Pb/204Pb-207Pb/204Pb (Table 1) suggest that these magmas result from mixing between two or three compositional end-members for each volcano. The noncorrelation in Pb isotope variation in Lana'i and Kaho'olawe indicates that these magmas sample an end-member that is heterogeneous in Pb isotope composition, or are mixtures of two end-members that are each heterogeneous in Pb isotope composition.
 In Pb isotope space, as well as in several major element compositions, Haleakala is not intermediate between the Lana'i and West Maui/East Moloka'i end-members, and thus requires a third, distinct mantle source. In Maui Nui 206Pb/204Pb-208Pb/204Pb variation, as well as Pb isotope-major element variation (e.g., 206Pb/204Pb-SiO2, 206Pb/204Pb-TiO2), Haleakala lavas, but not West Maui or East Moloka'i lavas, lie on an extension of the trend defined by Lana'i and Kaho'olawe. It appears that the Lana'i and Kaho'olawe lava compositions are the result of mixing between the Haleakala primary magmas or source and a more enriched component.
 Using this construction of end-members, in the simplest scenario, West Maui/East Moloka'i and Haleakala each sample primarily a single end-member (which, in the case of West Maui/East Moloka'i, comprises two subcomponents defined by Pb isotope composition but not resolvable in 87Sr/86Sr, 143Nd/144Nd or 176Hf/177Hf), whereas Lana'i and Kaho'olawe sample the enriched end-member only in combination or mixture with the Haleakala end-member. Although West Maui/East Moloka'i and Haleakala lavas could result from mixing multiple mantle sources in constant proportions, the isotopic homogeneity within each volcano would require a consistently reproducible process that operates on time scales that are significant relative to the life span of the volcano. Thus we favor the homogeneous end-member interpretation and we treat West Maui/East Moloka'i and Haleakala sources as two individual plume components.
5. Geochemical Evaluation of Ancient Oceanic Lithosphere in Maui Nui Mantle Sources
 Many aspects of major and trace element and isotope compositions of Hawaiian shield-stage lavas are consistent with ancient, recycled oceanic crust in the plume sources of Kea- and Ko'olau-type magmas. The negative δ18O-187Os/188Os correlation among Hawaiian shield-stage lavas, with δ18O < 5.0‰ and unradiogenic 187Os/188Os in the Kea-type lavas, and δ18O > 5.5‰ and radiogenic 187Os/188Os in the Ko'olau-type lavas, is compelling evidence for the incorporation of hydrothermally altered oceanic lithosphere in sources of both Kea- and Ko'olau-type magmas [Lassiter and Hauri, 1998]. Conversely, the normal-mantle-like 187Os/188Os and δ18O and elevated 3He/4He of Haleakala lavas [Kurz et al., 1987; Martin et al., 1994; Eiler et al., 1996; Lassiter and Hauri, 1998] do not seem to implicate recycled crustal material in the Haleakala magma source. The hyperbolic correlation of 206Pb/204Pb and 176Hf/177Hf (Figure 9) is consistent with the presence of pelagic sediment in the Ko'olau plume source [Blichert-Toft et al., 1999]. These models that use isotopic relations to infer ancient oceanic lithosphere in the plume source generally have not investigated the phase petrology of melting this type of source and the implications that this has for both major and trace element evolution of the magmas. Specifically, recycled oceanic crust in the mantle would be present as eclogite embedded in peridotite. In addition to isotopic and trace element contrasts with mantle peridotite, an eclogitic mantle source would have strongly contrasting thermodynamic properties and melting behavior that must be incorporated into chemical models of Hawaiian sources.
 The physical mechanisms for incorporation of ancient oceanic lithosphere into the plume, its interaction with other plume materials, and the process of melting and transport of eclogite partial melts to the surface, would have effects on lava compositions that may be both predictable and distinctive. In the model we present, we incorporate available experimental observations and thermodynamic modeling of melting interactions between eclogite and peridotite, and connect this to geochemical (major and trace elements and isotopes) predictions. This model complements previous studies that argue for the existence of recycled oceanic lithosphere in OIB sources on the basis of geochemical arguments, but which do not discuss a physical and phase petrological framework [Lassiter and Hauri, 1998; Chauvel and Hémond, 2000], and petrologic studies that address the physical and chemical aspects of eclogite and peridotite interactions in OIB genesis, but do not discuss the isotope and trace element implications [Yaxley and Green, 1998; Takahashi and Nakajima, 2002; Pertermann and Hirschmann, 2003a; Kogiso et al., 2004b].
 In the first case, partial melts of eclogitic lithologies are insulated from surrounding peridotite, and thus maintain geochemical signatures that reflect their equilibration with an eclogite, not peridotite, residue. A fertilized peridotite reaction zone between eclogite-derived melt and nonfertilized peridotite can effectively armor either a solid eclogite pod or a conduit through which eclogite-derived melts may travel without equilibrating with surrounding peridotite [Takahashi and Nakajima, 2002]. Alternatively, a subsolidus orthopyroxene-rich reaction zone that provides a chemical barrier between the lithologies may develop at the interface of eclogite and peridotite [Pertermann and Hirschmann, 2003b].
 As a second end-member case, eclogite melt infiltrates and reacts with surrounding solid peridotite to form a solid fertilized peridotite. This has been observed in eclogite-peridotite “sandwich” experiments [Yaxley and Green, 1998; Takahashi and Nakajima, 2002], and is simulated in calculations using the pMELTS algorithm of Ghiorso et al. . Eclogite melt reacts incongruently with olivine in peridotite to form orthopyroxene, and upon subsequent melting of this fertilized peridotite, the major element composition of the melt will reflect a peridotite source [Hirschmann et al., 2003]. Although these melts may have enriched isotope and trace element compositions reflecting the eclogite contributions, they will produce lower SiO2, higher MgO melts indicative of a peridotite, not eclogite, source [Yaxley and Green, 1998; Hirschmann et al., 2003].
 Case one is the basis for our model of the Lana'i plume source. We model the Lana'i source as eclogite derived from a mixture of recycled hydrothermally altered oceanic crust and pelagic sediment. These eclogite melts form and separate from their source, and mix in variable proportions with melts derived from the Haleakala source prior to eruption. Case two is the basis for our model of the West Maui/East Moloka'i source. We postulate melting of eclogite derived from recycled lower oceanic crust gabbro. This melt infiltrates, reacts with and freezes in to the associated recycled depleted oceanic lithosphere, thereby creating a hybrid refertilized peridotite source. This source subsequently melts to produce the West Maui/East Moloka'i lavas. In these models, we do not specify a physical interpretation for the Haleakala source.
 In the following sections, we model the trace element parameters that are most diagnostic of the physical materials and processes we postulate as the sources of the Maui Nui lavas. Pelagic sediment is characterized by high La/Nb relative to Hawaiian magmas, so we model La/Nb to constrain sediment contributions to Lana'i lavas. Lana'i lavas have the lowest TiO2 of all Maui Nui lavas (Figure 2); thus we explain this distinctive major element composition in our Lana'i model. For the Lana'i and East Moloka'i/West Maui models, we predict very different amounts of melting of garnet-bearing sources. Therefore we model Sm/Yb for both Lana'i and East Moloka'i/West Maui. Our East Moloka'i/West Maui model involves stages of eclogite melting and peridotite melting. To help constrain melt fraction (F) for each stage of melting, as well as eclogite melt:depleted peridotite mixing proportions, we model trace element ratios (Hf/Zr, Hf/Nb) for which partition coefficients and therefore fractionation during melting are particularly sensitive to the major element compositional differences of garnet and clinopyroxene in eclogite versus peridotite bulk rock compositions [Pertermann et al., 2004].
 For the models, we use partition coefficients that have been determined experimentally at pressures (∼3 GPa) and for bulk rock and mineral compositions appropriate to the processes we model. We prefer to use sets of partition coefficients that were determined for each phase during the same experiment or sets of experiments. This criterion significantly limited the available possibilities for partition coefficients. Within the range of partition coefficients that met our criteria, we found that the choice of partition coefficient did not significantly affect our conclusions for peridotite melting. For eclogite melting, we tested the model using partition coefficients from both Klemme et al.  and Pertermann et al. . Although the two sets of partition coefficients result in slightly different fractionation trajectories and mixing arrays for the trace element ratios we model (Sm/Yb, Hf/Zr, Hf/Nb), we nonetheless draw consistent conclusions with both sets of partition coefficients. In the following discussion we show results with the Pertermann et al.  partition coefficients.
5.1. Lana'i Source Model
 We model the Lana'i and Kaho'olawe magma compositions as the product of mixing between partial melts of ancient upper oceanic crust + pelagic sediment-derived eclogite and partial melts of the Haleakala plume component. Similar models for the origin of the Ko'olau component as dacitic partial melts of quartz eclogite have been proposed by Hauri  and Takahashi and Nakajima . In our model, we incorporate experimentally determined effects of subduction-zone dehydration on the trace element composition of pelagic sediment and hydrothermally altered upper oceanic crust, and evaluate the sensitivity to variable proportions of pelagic sediment:upper oceanic crust, and variable amounts of melting of this eclogite component in the plume. For this model, we assume that after dehydration of the sediment and crust, the pelagic sediment is incorporated completely, as a solid, into the upper oceanic crust, and that this mixture undergoes all subsequent melting or mixing as a single compositional unit. For the trace element composition of the dehydrated pelagic sediment, we use analyses of residues from dehydration experiments on pelagic clay done by Johnson and Plank . For the trace element composition of dehydrated hydrothermally altered upper oceanic crust, we use the experimental products of oceanic crust amphibolite high-pressure dehydration experiments done by Kogiso et al.  and low-MgO eclogite xenoliths inferred to be fragments of Archean dehydrated oceanic lithosphere from west Africa [Barth et al., 2001] (Table 2). We assume that the 87Sr/86Sr of this dehydrated composite upper crust is 0.7048, consistent with literature estimates of the composition of the Ko'olau component [e.g., Eiler et al., 1998], and corresponding to the 206Pb/204Pb of the Ko'olau component (17.85) inferred by Blichert-Toft et al. . For calculating the composition of melts derived from this eclogite source, we assume batch melting and use the partition coefficients of Pertermann et al.  (Table 3), which were determined specifically for anhydrous, high pressure (3 GPa) systems and garnet and clinopyroxene compositions that are consistent with an eclogite derived from mid-ocean ridge basalt.
Table 2. Oceanic Lithosphere Trace Element Model Compositions
 Most previous work on Ko'olau volcano, which is geochemically similar to Lana'i, has identified La/Nb higher than other Hawaiian shield-stage lavas [Frey et al., 1994; Roden et al., 1994; Huang and Frey, 2003]. Jackson et al.  discuss the possible association of this anomalously high ratio with a crustal source in the plume. Pelagic sediments are characterized by very high REE concentrations relative to high field strength elements (HFSE), so (light, middle REE)/HFSE is a sensitive indicator for presence and proportion of pelagic sediment in an oceanic crust-derived magma source. In Figure 10, we show calculated compositions for melts of upper oceanic crust-derived eclogite containing 4–7% pelagic sediment. The calculated La/Nb for this melt is not sensitive to clinopyroxene/garnet (cpx/gt) in the residue, nor to F (melt fraction) as long as F > ∼0.2. For F > 0.2, the observed La/Nb is consistent with 5–7% sediment in the Lana'i source. Mixing of 30–60% of this eclogite-melt with the average Haleakala primary magma composition generates the range of compositions that we see in the lavas erupted at Lana'i (Figure 10).
 La/Nb, Sm/Hf and Sm/Nb are higher in pelagic sediment than in Maui Nui lavas, and are lower in dehydrated upper oceanic crust than in Maui Nui lavas. Thus these ratios are useful for modeling the proportion of pelagic sediment in the Lana'i magma source. As with La/Nb, Sm/Hf and Sm/Nb are more sensitive to a difference of 1% in the amount of pelagic sediment in the Lana'i source than they are to reasonable variation of F (0.5–0.9, as predicted by pMELTS for the Lana'i source composition, and T and P conditions appropriate for the Hawaiian plume) or cpx/gt (1–9) in the residue. Thus this model shows a unique sensitivity to proportion of sediment in the source. The Sm/Nb of the Lana'i lavas is consistent with 5–10% sediment in the Lana'i end-member source. Sm/Hf is best modeled with addition of 1% sediment to the upper oceanic crust. However, Sm/Hf of the data set that we use for the eclogite starting composition of this model [Barth et al., 2001] has a standard deviation of about 20%, which encompasses the range of Sm/Hf predicted for 1–3% sediment addition. The small amount (<10%) of sediment in the Lana'i source predicted by these calculations is consistent with previous work that uses isotope compositions of Ko'olau lavas to conclude that the sediment proportion in the Ko'olau source is 3–20% [Eiler et al., 1996; Lassiter and Hauri, 1998].
 Sm/Yb is far more sensitive to residual cpx/gt and F than to sediment proportion, and therefore is useful for constraining the amount of melting of the Lana'i source to produce compositions that are consistent with observation. To identify the dependencies of cpx/gt and F on the mantle thermal and melting regime, we performed a series of experiments with the pMELTS algorithm [Ghiorso et al., 2002]. We used a MORB major element starting composition and dehydrated eclogite +5% dehydrated sediment as the trace element starting composition (Tables 2 and 4). Using pMELTS, we calculated the amount of melting (F) and cpx/gt proportion for isentropic (adiabatic) melting, from 3 to 2 GPa, for starting temperatures ranging from 1500° to 1550°C. These pressure and temperature ranges are consistent with experimental and thermodynamic estimates for P-T conditions in the Hawaiian plume [e.g., Eggins, 1992; Herzberg and O'Hara, 2002]. Also, pMELTS models of peridotite melting at these P-T conditions predict melt fractions consistent with independent evaluations of F for the Hawaiian plume [Hofmann et al., 1984; Eggins, 1992; Sims et al., 1995, 1999; Herzberg and O'Hara, 2002; Feigenson et al., 2003]. The pMELTS calibration uses experiments run primarily at P ≤ 3 GPa, thus it is not appropriate to use pMELTS to model melting at P > 3 GPa. Melting in the Hawaiian plume ceases when the plume intersects the base of the Pacific lithosphere, thus we generously estimate melting to cease at 2 GPa. We also ran pMELTS simulations for isobaric melting at 2 and 3 GPa, for comparison with isobaric melting experiments.
Table 4. Oceanic Lithosphere Major Element Model Compositions
 Using F and mineralogical modes from pMELTS, and partition coefficients determined for high-P melting of eclogite [Pertermann et al., 2004], we calculated Sm/Yb in the melt as a function of F (Figure 11). Sm/Yb is very sensitive to the extent of eclogite partial melting. Thus, for any particular melting curve, even at rather high F (>0.55), there is only a very narrow range (∼<5%) in F over which the observed Lana'i Sm/Yb (2.3–3.0) is obtained. For adiabatic melting, the absolute F ranges from about 0.55 to 0.80, depending upon which adiabat the eclogite parcel follows. Regardless of which adiabat the eclogite is on, as long as there is garnet in the residue, there is a very small range in F which will produce Sm/Yb typical of Lana'i (2.3–3.0).
 Pyroxenes in equilibrium with eclogite whole rock compositions can contain 5–10% of the super-aluminous Ca-eskolaite (CaEs) component [Pertermann and Hirschmann, 2003b]. However, pMELTS does not account for this CaEs component, and consequently will systematically underestimate the residual clinopyroxene, and thus cpx/gt, in these simulations. Because of this, the actual location of our predicted Sm/Yb versus F curves may shift slightly to lower Sm/Yb at a given F. However, the slope of this curve will not change, and therefore our conclusion holds that these magmas are the result of a specific, reproducible, large degree of melting, which may be controlled by the Pacific lithosphere thickness and thus the pressure at which Lana'i melts segregate from the source. However, because we cannot conclusively determine which adiabat the plume follows, and because pMELTS underestimates cpx/gt in eclogite bulk compositions, it is difficult to constrain the absolute F more specifically than from 0.5 to 0.8.
 Using the 1540°C adiabat as a reference, we calculate mixing curves between average Haleakala primary magmas and predicted Lana'i source melts for eclogite F ranging from F = 0.68 to F = 0.76 (Figure 12). This range in F produces a wide enough range in Sm/Yb to account for the observed Sm/Yb in Lana'i lavas (2.3–3.0) obtained by mixing 30–50% eclogite melt with 50–70% Haleakala melt. This is similar to mixing proportions inferred from the La/Nb variations. Using a warmer (1550°C) or cooler (1500°C) adiabat will result in, respectively, higher (0.75–0.83) or lower (0.52–0.59), predicted F for the Lana'i source (see Figure 11), but will not affect the absolute mixing proportions between melts from the Haleakala and Lana'i sources.
 The low TiO2 content of Lana'i lavas relative to other Maui Nui lavas is one of their most distinctive major element characteristics, and we also use this to estimate relative mixing proportions of Lana'i component melts and Haleakala component melts (Figure 13). Takahashi and Nakajima  experimentally determined the compositions of eclogite partial melts of an Archean MORB. At 2.7 GPa, and 85% melting, the melt had 1.16 wt. % TiO2 and 7.62 wt. % MgO [Takahashi and Nakajima, 2002]. In a similar experiment at 3.5 GPa, Yaxley and Green  reported 85% partial melts of average oceanic basalt with 1.5 wt. % TiO2 and 6.9 wt. % MgO. Eclogite melting experiments show a general trend of decreasing TiO2 with increasing MgO and melt fraction [Yaxley and Green, 1998; Takahashi and Nakajima, 2002; Pertermann and Hirschmann, 2003a], which implies that the low TiO2 component in the Lana'i lavas derives from high-degree, basaltic andesitic melts of eclogite, rather than smaller degree, dacitic melts as proposed by Hauri . Higher degree eclogite melts are also more consistent with the higher-temperature solidus for eclogite compared to peridotite.
 These experiments imply that low-MgO (<10 wt.%) partial melts are produced from eclogite in the Hawaiian plume. However, our trace element calculations and observed MgO compositions indicate that these melts do not erupt without mixing with Mg-rich Haleakala melts. At a given MgO, Lana'i and Kaho'olawe lavas generally have slightly higher Ni than other Maui Nui lavas (data sources from Appendix A), which may indicate that Lana'i and Kaho'olawe lavas have not fractionated as much olivine as the other Maui Nui lavas. This in turn implies that the Lana'i and Kaho'olawe lavas contain a component of low-MgO primary magma, as predicted for an eclogite melt. Taking the Takahashi and Nakajima  and Yaxley and Green  experimental results as the composition of the eclogite-derived melt for the Lana'i source, mixing of this melt with 30–70% of the Haleakala primary melt will produce the range of magmas parental to Lana'i lavas, and mixing of this Lana'i primary melt with 70–80% Haleakala melt will result in the Kaho'olawe parental magmas. Imposing olivine fractionation trajectories on these parental magmas can produce the observed lava compositions (Figure 13). Although these major element trends allow for a broader range in eclogite melt-Haleakala melt mixing proportions than predicted from the trace element compositions, the mixing proportions are consistent for the major and trace element calculations and isotope observations.
5.2. West Maui/East Moloka'i Source Model
 For the West Maui and East Moloka'i magmas, we model the plume source as a gabbro + depleted lithosphere package of ancient oceanic lithosphere [Lassiter and Hauri, 1998]. The eclogite in the ascending plume crosses its solidus at a higher pressure than the associated depleted lithosphere does, such that during ascent in the plume, gabbro-derived eclogite begins to melt first. Its melt percolates into and fertilizes adjacent depleted lithosphere. Various metasomatic processes have been proposed in the source of Hawaiian lavas and xenoliths [Wright, 1984; Chen and Frey, 1985; Sen, 1987; Salters and Zindler, 1995; Norman and Garcia, 1999; Sobolev et al., 2000; Sen et al., 2003], but in most models the proposed metasomatic agent is very small degree melt (F = 0.001–0.03) of a mantle or recycled crustal source. Here, we specifically propose larger (F = 0.1–0.3) degree melt of oceanic crust gabbro-derived eclogite as the metasomatic agent.
 For the gabbro-derived eclogite, we use major and trace element compositions for the Gabal Gerf ophiolite gabbros [Zimmer et al., 1995] (Tables 2 and 4). Calculations for eclogite melting use partition coefficients determined for anhydrous MORB melting at high pressure (3 GPa) [Pertermann et al., 2004] and residual cpx/gt determined by pMELTS of the desired F at 3 GPa. Because pMELTS underestimates cpx/gt in eclogitic bulk compositions, we also model results for a higher residual cpx/gt. We determined the trace element composition of the complementary depleted lithosphere as the residue from MORB-melting, using phase proportions calculated with MELTS [Ghiorso and Sack, 1995] for 18% melting of the MM3 experimental composition [Baker and Stolper, 1994], at 0.5 GPa, and with MORB trace element compositions from Sun and McDonough . We used MELTS, rather than pMELTS, for this particular calculation because the MELTS calibration is more appropriate for the lower pressure melting conditions we model here [M. Ghiorso, personal communication, 2004]. For partition coefficients of peridotite melting in the Hawaiian plume, we chose clinopyroxene and garnet partition coefficients determined experimentally for high pressure (2.8–3 GPa) melting of a moderately depleted lherzolite [Salters and Longhi, 1999] (Tables 3 and 5).
 We use gabbro in our models instead of upper basaltic crust because isotopic evidence (δ18O) points toward lower oceanic crustal gabbro in the source, and also because the oceanic gabbro is in closer physical association with the depleted lithosphere than upper basaltic crust, and thus it is more probable that gabbro melts, rather than upper oceanic crustal melts, will refertilize depleted lithosphere. The lower crustal gabbro does not have a dramatically different major or trace element composition compared to the upper crustal compositions we model. Small major element differences are apparent in TiO2, Al2O3 and CaO (Table 4). Our gabbro composition has slightly higher Sm/Yb and Hf/Zr compared to the dehydrated altered upper basaltic crustal composition we use in the Lana'i model (Table 2). However, small differences in parent-daughter ratios of radiogenic isotopic systems can lead to large differences in radiogenic isotopic compositions of upper and lower crust during aging over the Gyr time scale [Gaffney et al., 2004]. Although using gabbro versus upper basaltic crustal compositions does not result in significantly different conclusions from our model, we use the gabbro compositions in our models for consistency with the physical processes we envision.
 In the West Maui/East Moloka'i model, the parameters for which we aim to identify a mutually consistent set of values are: degree of eclogite melting (FE), mixing proportions of eclogite melt:depleted lithosphere to form the hybrid refertilized peridotite (referred to hereafter as “hybrid peridotite”), and degree of melting of the hybrid peridotite (FH). Partitioning of HFSE into garnets is strongly controlled by major element composition of the garnet, which in turn is dependent upon bulk composition of the rock with which the garnet is in equilibrium. Hf/Zr (and Hf/Nb) show greater extents of fractionation at a given F for melting of an anhydrous MORB-like bulk composition compared to a peridotite bulk composition [Pertermann et al., 2004]. We take advantage of this difference in partitioning behavior to differentiate between FE and FH. Because Sm/Yb in a melt is sensitive to residual garnet and cpx/gt, we also include Sm/Yb in our models.
 The four panels in Figure 14 illustrate the sensitivity of this model to FE, FH and eclogite melt:depleted lithosphere mixing proportions as well as residual cpx/gt of both eclogite and hybrid peridotite. It is apparent that even at large FE (>0.2), there is significant Hf/Zr fractionation, whereas at FH > 0.2, Hf/Zr fractionation is much less. This makes Hf/Zr versus Sm/Yb a good diagnostic tool for large FE. Calculations for mixing of various degrees of eclogite melt with solid depleted lithosphere results in relatively flat arrays (curves marked with diamonds) for low (10:90) eclogite melt:depleted lithosphere. These arrays get steeper as eclogite melt:depleted lithosphere increases. At a given mixing proportion, the composition of the hybrid peridotite is much more dependent on FE than cpx/gt residual to the eclogite melt. Lower cpx/gt in the hybrid peridotite results in larger Sm/Yb fractionations at a given F than a hybrid peridotite with higher cpx/gt. Bulk compositions of mechanical, solid-solid mixtures of eclogite and depleted lithosphere lie to the right (toward higher Hf/Zr) of hybrid peridotite compositions. There is no combination of F and mode with which a mechanical mixture could melt that would result in West Maui/East Moloka'i-type lava compositions.
Figures 15 and 16 illustrate the best fit of the model parameters to the Maui Nui data. We show hybrid peridotite fertilized in 8:92 eclogite melt:depleted lithosphere proportions, by eclogite melts of FE = 0.1, 0.2, 0.4, 0.6 and 0.8, and melting trajectories of the hybrid peridotite with residual phase proportions ol:opx:cpx:gt 43:43:12:2. These residual phase proportions, based on pMELTS calculations, are generally consistent with proportions calculated or inferred as residual in the Hawaiian plume [Hofmann et al., 1984; Wagner and Grove, 1998]. Melting of eclogite also generates sizable fractionations in Hf/Nb at larger degrees of melt (Hf/Nb = 0.34–0.47 at F = 0.2–0.4), whereas comparable Hf/Nb fractionation in peridotite melting occurs at lower degrees of melt (Hf/Nb = 0.27–0.39 at F = 0.05–0.20). At larger degrees of peridotite melt, Hf/Nb fractionation is insignificant. The West Maui/East Moloka'i compositions are consistent with 10–20% melting of a hybrid peridotite source that was fertilized by 10–30% melts of oceanic crustal gabbro-derived eclogite.
 East Moloka'i and West Maui are nearly identical in their isotopic compositions, indicating that their sources have formed from the same mixing proportions during hybridization. The simplest explanation for the offset in Hf/Zr, but in no other chemical parameters, of East Moloka'i from West Maui, is that the East Moloka'i eclogite melt formed with a slightly higher residual cpx/gt.
5.3. Discussion of the Source Models
 Our models of the Lana'i and West Maui/East Moloka'i sources address the existence of eclogite heterogeneities derived from ancient oceanic crust, their melting behavior in the plume, and the physical processes for mixing of source components (melt-melt versus melt-solid). Our model naturally explains the relative geochemical heterogeneity of Lana'i lavas compared to West Maui/East Moloka'i lavas. The data are most consistent with the origin of Lana'i lavas in melt-melt mixing between magmas derived independently from the Lana'i and Haleakala plume sources. Solid-solid mixing of the Lana'i (eclogite) and Haleakala (peridotite) plume sources prior to melting, in proportions appropriate to the observed Lana'i lava trace element and isotope compositions, would increase the modal amount of garnet in this mixed source. For extents of melting commonly inferred for peridotite in the Hawaiian plume (∼6–30% [Hofmann et al., 1984; Eggins, 1992; Sims et al., 1995, 1999; Herzberg and O'Hara, 2002; Feigenson et al., 2003]), the increased abundance of garnet in the source would generate much higher Sm/Yb than is observed in Lana'i lavas. However, high degrees of melting (F > 0.6) of the Lana'i source prior to mixing with Haleakala source melts is consistent with the observed Sm/Yb. A melt-melt mixing scenario implies that unmixed melts exist to relatively shallow depths within the plume, and possibly even within the lithosphere and crustal magma chambers. Observations of melt inclusions in individual olivine crystals indicate that melt heterogeneities in Hawaiian magmas exist on scales that can readily be trapped over the course of crystallization of individual phenocrysts in the shallow magma chamber [Sobolev et al., 2000; Norman et al., 2002]. These independent observations support the plausibility of high-level mixing of heterogeneous melts that we propose.
 The fertilization model for the West Maui/East Moloka'i source is a process that can both account for recycled oceanic crust isotopic signatures in the West Maui/East Moloka'i lavas and generate the characteristic isotopic homogeneity of Kea-type lavas. The infiltration/fertilization of the depleted lithosphere by the gabbro melts is a stage of premixing of these two lithologic segments prior to the melting event that generates the erupted lavas. The infiltration of gabbro-eclogite melt is a self-buffering process, as gabbro melt can only “freeze-in” to the peridotite up to proportions of about 40:60 eclogite melt:peridotite, based on pMELTS calculations and experiments designed to address this process in the mantle wedge above subduction zones [Rapp et al., 1999]. Because this model of melt-solid (metasomatic) interaction provides a mechanism for premixing or homogenization in the Kea source, we favor this model over solid-solid or melt-melt mixing scenarios. Although either solid-solid or melt-melt mixing mechanisms could potentially generate homogeneous magmas, the processes and proportions of mixing would need to be consistently reproducible over millions of years. Furthermore, as shown in Figure 14, solid-solid mixing of the model eclogite and peridotite cannot produce a mix which will, upon melting, generate any Maui Nui shield-stage lavas.
 The Lana'i and West Maui/East Moloka'i models present contrasting mechanisms for interaction of eclogite-derived melt with surrounding peridotite. These contrasts may be understood through consideration of the relative eclogite melt:peridotite proportions. At high ratios (>40:60), peridotite is effectively soluble in eclogite melt. Through reaction/infiltration processes (as modeled by Spiegelman et al. ), the eclogite melt would infiltrate and dissolve peridotite along initially higher-permeability zones, and self-organize into steady-state channels that localize and isolate much of the eclogite melt flux. This process can account for the separation of eclogite melt from the residue and its transport through peridotite without much apparent chemical modification. Numerical modeling of flow in porous media involving precipitation instead of dissolution, such as expected for lower eclogite melt:peridotite, shows that incipient melt channels will crystallize and choke, forcing the melt to divert to other regions [Aharonov et al., 1997]. In this case, the flow of eclogite melt into the peridotite is diffuse, and may generate a homogeneous, fertilized peridotite. However, in order to effectively homogenize a large volume of peridotite, the eclogite pods may need to be physically smaller but broadly distributed, so that the surface area for interaction between eclogite melts and peridotite is maximized. This may accomplished through stretching and mixing mechanisms in the mantle or Hawaiian plume [Farnetani et al., 2002]. The resulting melt-rock ratio of the hybrid, as discussed earlier in the geochemical modeling section, is 8:92, so only a small amount of melt needs to be efficiently distributed throughout the rock. Also, these proportions must be an average on the scale of the melting region in the plume that eventually produced the West Maui/East Moloka'i lavas, but the proportions could vary significantly on a local scale, as long as they remain lower than the 40:60 proportions where peridotite will become effectively soluble in eclogite melt.
 Lana'i-source eclogite and hybrid refertilized depleted peridotite (West Maui/East Moloka'i source) will follow different T-P-melting paths in the plume. Using pMELTS [Ghiorso et al., 2002], we simulated the relative melt productivities of these two lithologic units, melting along an adiabat from 3 to 2 GPa, starting at an initial temperature of 1540°C, and for the same bulk compositions as used in the geochemical modeling (Figure 17). At pressures corresponding to the base of the Pacific lithosphere (assuming a 100 km-thick lithosphere), the Lana'i source eclogite is predicted to have melted 72-78%, and the West Maui/East Moloka'i source peridotite is predicted to have melted 11–15%. These pMELTS calculations, based solely on the thermodynamic properties of eclogite and hybrid peridotite, predict the same extents of melting of these sources as we determined in our geochemical models (F = 0.68–0.76 for Lana'i source, assuming a 1540°C adiabat; F = 0.10–0.25 for West Maui/East Moloka'i source, independent of adiabat). For a thinner lithosphere (2 GPa at the base), the melt fractions are higher: ∼20% for West Maui/East Moloka'i and ∼93% for Lana'i. At a lower initial temperature, 1500°C, the melt fractions are slightly lower at 2.8–3 GPa, but show similar relative proportions: ∼7% for the West Maui/East Moloka'i source, and ∼55% for the Lana'i source. Figure 17, showing melting curves for the 1540°C adiabat, has two curves for the West Maui/East Moloka'i source, corresponding to depleted peridotite that has been hybridized (in 10:90 eclogite melt:depleted peridotite proportions) with F = 0.2 versus F = 0.4 eclogite melts. These two starting compositions do not show significant differences in their P-F paths.
 The mixing mechanisms and proportions in the fertilization model are consistent with observed and inferred δ18O and 187Os/188Os for ancient oceanic lithosphere. Both of these isotope systems should have contrasting signatures for oceanic crust and the associated complementary residual oceanic lithosphere, and thus trace lithospheric contribution to the Maui Nui source. Oxygen in oceanic lithosphere is modified by hydrothermal alteration on- and off-axis at the mid-ocean ridge. Upper (basaltic) oceanic crust has δ18O up to 15–20‰, whereas oceanic gabbro has δ18O ∼ 0–3‰ [Gregory and Taylor, 1981; Stakes, 1991; Hart et al., 1999; Miller et al., 2001]. Residual oceanic lithosphere, which may be variably hydrothermally altered [Frueh-Green et al., 2003], has δ18O in olivine of 3–5‰ [Cocker et al., 1982; Hoffman et al., 1986]. Mixing of the gabbro and residual lithosphere components in 8:92 proportions will result in δ18O = 4.5–5‰, as observed in West Maui lavas [Gaffney et al., 2005]. Because of the strong relative compatibility of Os with respect to Re during mantle melting, ancient depleted oceanic lithosphere should be Os-rich and have low 187Os/188Os, whereas ancient oceanic gabbro should have low Os concentrations and high 187Os/188Os. Gabbro has [Os] one to four orders of magnitude lower than abyssal peridotites [Lassiter and Hauri, 1998; Blusztajn et al., 2000; Brandon et al., 2000], and Os is inferred to be highly compatible during melting, so Os from gabbro-derived eclogite melts may not be a significant contributor to the hybrid fertilized lithosphere. Finally, Os (and possibly Re), may be lost from hydrothermally altered crust via fluids that are expelled during dehydration of the subducting slab, further depleting the gabbro in Os (and possibly Re) [Brandon et al., 1999; Borg et al., 2000]. Outside of West Maui, few 187Os/188Os analyses of Maui Nui lavas are available, but at least in West Maui they are consistent with our model for the West Maui/East Moloka'i source [Gaffney et al., 2005].
Stracke et al.  and Kogiso et al.  present quantitative models that specifically address the role of subduction zone dehydration on the isotopic evolution of recycled oceanic crust, and show that Rb/Sr and 87Sr/86Sr are particularly sensitive to alteration and dehydration processes (87Sr/86Sr of 2 Ga crust: 0.7015–0.7045 [Kogiso et al., 1997]; >0.708 [Stracke et al., 2003]). Eiler et al. , Hauri , Lassiter and Hauri , Blichert-Toft et al. , and Gaffney et al.  have modeled or inferred radiogenic isotope compositions as well as δ18O for postulated recycled oceanic lithosphere components specific to the Hawaiian plume. These Hawai'i studies make variable assumptions about perturbation of parent-daughter ratios during hydrothermal alteration at the mid-ocean ridge, and during dehydration of the slab in the subduction zone. Although the predicted radiogenic isotope compositions of recycled oceanic crust may vary greatly, the models we propose for the Lana'i (larger degree melts of eclogite derived from basalt + sediment) and West Maui/East Moloka'i (depleted lithosphere fertilized by smaller degree melts of gabbro-eclogite) sources are broadly consistent with this previous work. Furthermore, the 87Sr/86Sr composition for Lana'i that we use in our modeling was determined empirically, and thus is not dependent upon assumptions of the 87Sr/86Sr of ancient crust.
 Several arguments have been made against eclogite in the source of Hawaiian magmas. Herzberg and O'Hara , Feigenson et al. , and Niu and O'Hara  argue that the inferred high-MgO primary compositions of Hawaiian magmas are inconsistent with contributions from a mafic source in the plume. However, these studies did not discuss apparently primary low-MgO magmas from Ko'olau and Lana'i. High-MgO glasses and lavas are observed for several Hawaiian volcanoes, and this is strong support for high-MgO primary magmas in these volcanoes. However, high-MgO glasses or high Fo olivine have not been reported for either Lana'i or Kaho'olawe, and very few lavas with MgO > 12 wt. % are observed for either of these volcanoes. Thus, if these volcanoes did have initial MgO as high as proposed for other volcanoes, then they are consistently fractionating large amounts of olivine prior to eruption, so that those primitive compositions are never sampled at these two volcanoes. Norman et al.  use melt inclusions from Ko'olau, which is isotopically similar to Lana'i, to determine that 14 wt. % MgO is a reasonable primary magma composition for this volcano. This is similar to the primary magma that we infer for Lana'i and Kaho'olawe, which forms from mixing between eclogite-derived melts and primary Haleakala magmas. Trace element and U-series isotope systematics have also been used to argue against eclogite in the Hawaiian plume [Stracke et al., 1999; Feigenson et al., 2003], but these studies have all focused on volcanoes for which we do not propose eclogite in the source, and thus cannot be used to rule out eclogite in the Lana'i/Kaho'olawe magma source.
6. Implications for Structure and Processes in the Hawaiian Plume Beneath Maui Nui
 The relative ages of late shield-stage magmatism for Maui Nui volcanoes are W. Moloka'i > East Moloka'i ≥ West Maui ≥ Lana'i > Kaho'olawe > Haleakala [McDougall, 1964; Naughton et al., 1980; Chen et al., 1991]. Probably two, three, or possibly even four of these volcanoes erupted contemporaneously during their shield-building periods. The presumed contemporaneity of lavas that span nearly the entire range of isotopic compositions seen in the Hawaiian archipelago indicates that multiple compositionally extreme components coexisted in the plume and were sampled without apparent significant communication between the magmas (Lana'i and East Moloka'i/West Maui).
 The contemporaneous volcanoes of Maui Nui show much greater isotopic compositional contrast than contemporaneous volcanoes on the Big Island. Hawaiian compositional end-members have typically been defined in such a way as to account for the whole range of mantle variability sampled by Hawaiian volcanoes over a 5 Myr time span. Although this is an appropriate way to describe the system on a large scale, this treatment does not address compositional subtleties apparent on a shorter temporal or spatial scale. Furthermore, the end-members have not always been sampled regularly over the past 5 Myr. For example, of all the volcanoes that show extreme Ko'olau-type compositions (Ko'olau, Lana'i, Kaho'olawe), none are younger than Kaho'olawe. Conversely, of all the volcanoes that erupted extreme Kea-type compositions (East Moloka'i, West Maui, Mauna Kea, Kilauea), none are older than East Moloka'i. The Lo'ihi component is the only end-member defined for shield-stage lavas that apparently has been sampled regularly throughout the last 5 Myr (in Kaua'i, Haleakala, Mauna Loa and Lo'ihi), which suggests that it is a pervasive component of the Hawaiian plume. Recent work by Frey et al. , Huang et al. , and Schafer et al.  has identified an additional, very depleted component sampled by rejuvenated lavas from the Hawaiian volcanoes as well as older volcanoes in the Emperor Seamount chain. Although this component is not represented in Maui Nui shield-stage lavas, it apparently exists as another long-lived component in the Hawaiian plume over the past 80 Myr.
 The East Moloka'i and West Maui lavas are nearly indistinguishable chemically, and thus both tapped the same or nearly identical mantle sources. The plume component they tapped may have been a single, very large domain in the plume that was effectively homogenized through the infiltration and fertilization processes proposed here. This contrasts with the interpretation that the basalt+sediment - derived eclogite (Lana'i component) occupied much smaller domains in the plume. Furthermore, these volumes occupied separate places in the plume, such that they were not both sampled at the same volcano, but rather at neighboring volcanoes, at the same time or in close succession. This may reflect the stretching and mixing mechanisms in the mantle and plume [e.g., Farnetani et al., 2002], or the material properties of the different lithologies, and their response to mechanical homogenization in the mantle. These stretching mechanisms may be effective at separating the basalt+sediment and gabbro+peridotite oceanic lithosphere layers in the plume. As discussed above, although the Ko'olau and Kea-type lavas erupted contemporaneously in the Maui Nui volcanoes, both components are individually present in older and younger volcanoes. This indicates that the separation of the oceanic lithosphere into its upper and lower parts may occur during mixing and stretching in the mantle, and that the upper oceanic crust segment can be incorporated into the Hawaiian plume independently of the lower lithosphere segment, and vice versa. Alternatively, the recycled oceanic lithosphere may remain as a single physical unit during recycling. In this case, the location of the lithospheric package relative to the melting zone and plume periphery would determine whether the upper or lower lithospheric segment is sampled by melting. For example, the lower lithosphere segment may lie within the melting zone beneath West Maui, but the upper lithosphere part of the segment may lie close to the edge of the plume, and therefore will not melt (or will make only a very minor contribution to the erupted magma). It may be coincidental that both upper and lower lithosphere segments are sampled in the plume during the time of Maui Nui, rather than a fundamental indication of how oceanic lithosphere physically behaves in the mantle during storage, transport and incorporation into the plume.
 The source of Haleakala lavas is a relatively primitive plume component. Its elevated 3He/4He (13.1–16.8 RA [Kurz et al., 1987]), high time-integrated Th/U (high 208Pb/204Pb), and normal-mantle-like δ18O are also consistent with the relatively primitive nature of this component [Eiler et al., 1998]. The high MgO and low SiO2/FeO of the Haleakala primary magmas imply high pressures of segregation of the magma from its source. The intermediate 87Sr/86Sr, 143Nd/144Nd, 176Hf/177Hf and 206Pb/204Pb of Haleakala make it difficult to identify this as a separate component, but the elevated 208Pb/204Pb requires a distinct source that is not a mixture between Lana'i and West Maui/East Moloka'i.
 This interpretation of the components and the way that they are sampled does not require a concentrically zoned plume model during Maui Nui time as is inferred for recent plume structure from analysis of Big Island magmatism [DePaolo et al., 2001]. The Haleakala component is sampled at volcanoes that lie on both the Kea and Loa trends. The location of the Ko'olau volcanoes on the Loa trend and the Kea volcanoes on the Kea trend may instead represent axial asymmetry as would be predicted by the Farnetani et al.  model for vertical stretching of heterogeneities in the plume. Time-series analyses of HSDP samples suggest that vertical heterogeneity [Blichert-Toft et al., 2003] as well as lateral zonation [Eisele et al., 2003] characterize the Hawaiian plume. Furthermore, the Lana'i component heterogeneities must be of a small enough size that their melts generally mix with the Haleakala component, whereas the West Maui/East Moloka'i component is of a large enough extent that its melts typically erupt without apparent mixing with Haleakala. However, Haleakala is close enough in composition to West Maui/East Moloka'i that smaller degrees of mixing between West Maui/East Moloka'i and Haleakala may not be resolved.
7. Summary and Relevance to the Past 5 Myr of Hawaiian Shield-Stage Magmatism
 The enriched Ko'olau component sampled in Maui Nui by Lana'i and Kaho'olawe volcanoes is also sampled at the older Ko'olau volcano. However, the Pb-Pb and Sr-Nd isotope correlations of Ko'olau, Lana'i and Kaho'olawe are slightly offset from one another, indicating that this enriched component is temporally heterogeneous. Lana'i and Ko'olau are unique relative to the rest of the Hawaiian shield-stage lavas, as compositions this enriched are not observed at any older or younger volcanoes. Thus the Ko'olau/Lana'i component may represent an episodic recycling process, or possibly even a unique event related to the preservation in the plume of an unusually large domain of ancient subducted pelagic sediment. Furthermore, heterogeneity in this enriched end-member indicates that compositional heterogeneities in the oceanic crust are preserved or perhaps enhanced through the recycling and storage process. For example, Rb/Sr or U/Pb may be variably modified during hydrothermal alteration [Hart et al., 1999], which can lead to divergent 87Sr/86Sr or 206Pb/204Pb during aging of a single segment of oceanic crust.
 The depleted Kea component has been a dominant component in the Hawaiian plume for the past 2 Myr. Although we and others argue that this component is also created by recycling oceanic crust and lithosphere, it is characterized by a high degree of homogeneity. This contrasts with the Ko'olau component, where the recycled component exhibits a relatively high degree of heterogeneity. We attribute the homogeneity of the Kea component to the homogenizing and natural buffering by infiltration on a large scale of the gabbro and peridotitic segments of oceanic lithosphere. Some minor heterogeneity is preserved in the Kea component, however, as evident in the linear and distinct Pb isotope arrays observed in East Moloka'i and West Maui, as well as in Mauna Kea [Eisele et al., 2003].
 Our interpretation that the Haleakala component is the pervasive plume component at Maui Nui is also applicable to volcanoes at the older and younger ends of the chain. Haleakala lavas in Maui Nui are analogous to the Lo'ihi component, which is characterized by even higher 3He/4He than observed at Haleakala [Kurz et al., 1983]. The Lo'ihi component is sampled on Kaua'i at the oldest end of the chain, at Lo'ihi at the youngest end of the chain, and by Mauna Loa on the Big Island [Kurz et al., 1983, 1987, 1995; Mukhopadhyay et al., 2003]. The relatively continuous presence of this component in lavas erupted over the past 5 Myr indicates that this is a long-lived plume component and may constitute the matrix of the plume itself. The deepest lavas of Mauna Kea, sampled through HSDP-2, also trend to the lower SiO2, and higher 208Pb/204Pb and 3He/4He characteristic of the Lo'ihi component [Feigenson et al., 2003; Kurz et al., 2004]. The oldest lavas in the Ko'olau Scientific Drilling Project also show indications of the Lo'ihi component [Huang and Frey, 2003]. Thus this component has been a long-lived contributor to Hawaiian shield-stage magmas, and was sampled by magmas during the preshield and throughout the shield-building stages of magmatism.
 Although deep stratigraphic sections, accessed either through drilling or submersible, in volcanoes are rare, none yet observed show the transition between Kea and Ko'olau within the lavas of a single volcano. Individual volcanoes transition from Lo'ihi at depth to either Kea or Ko'olau compositions in the younger lavas [Feigenson et al., 2003; Huang and Frey, 2003; Kurz et al., 2004]. From this, we may surmise that the extreme compositions of both Ko'olau and Kea-type lavas are late-stage phenomena, and it is only during the final stages of shield-building magmatism that their signals are preserved.
 Interaction of Kea or Ko'olau component magmas with each other, Lo'ihi-like plume matrix or the Pacific lithosphere as the magmas travel to the surface could dilute or erase their distinctive geochemical signatures. Thus contemporaneous magma pathways from the plume to the surface must maintain enough physical separation that chemically resolvable contamination does not occur. At shallow depths, the spacing of fractures through which magmas travel must be small enough to isolate chemically distinct plume-derived magmas from each other. Deeper within the Pacific lithosphere, channelized networks may develop through which magmas can pass by reaction of magma with the solid lithosphere matrix [Kelemen et al., 1995; Spiegelman et al., 2001]. Once the channels are established, this could allow magmas to preserve their distinctive geochemical signatures. This may be one of the fundamental mechanisms controlling the time-composition trends in Hawaiian volcanoes. Early generated Ko'olau or Kea magmas do not erupt at the surface, but instead are exhausted in forming reaction-armored melt-conduit channels that at a later stage of volcanism allow the passage of these melts to the surface with little modification by ambient mantle or Pacific lithosphere. Such processes have been modeled for adiabatically upwelling mid-ocean ridge systems [Kelemen et al., 1995], and may be applicable to upwelling in the Hawaiian plume as well. During the latest stages of shield-stage magmatism, the magma flux decreases to the point that reaction-armored conduits are no longer maintained in the oceanic crust, and erupted magma compositions reflect small amounts of chemical interaction with crustal materials [Gaffney et al., 2004].
 The late shield-building stage lavas of Maui Nui span nearly the entire range of compositions observed across the Hawaiian chain. These lavas record the exhaustion of the enriched Ko'olau component, and the initiation of the depleted Kea component as dominant compositional end-members.
 Isotope compositions are consistent with a component of ancient, recycled oceanic lithosphere in the plume sources of both Kea and Ko'olau type magmas. We propose physical models for melting this component in the plume. Kea-type magmas derive from melting of homogeneously hybridized gabbro + depleted lithosphere segments of recycled lower oceanic lithosphere. Hybridization and homogenization occur when 10–30% melts of the gabbro-derived eclogite infiltrate and reactively freeze into the depleted lithosphere peridotite in 8:92 eclogite melt:depleted lithosphere proportions. Our model contrasts with earlier metasomatic models, in that we propose that the refertilizing metasomatic fluid originates in moderate-degree (10–30%), rather than small-degree (<5%), melts. Ko'olau-type magmas originate from high-degree (F ∼ 0.6–0.7) melts of recycled upper basaltic crust and sediment that mix with Haleakala-type melts, in proportions that range from 30–50% eclogite melt. Independent lines of evidence, including isotopic compositions, trace element compositions, experimental constraints and thermodynamic modeling, all point toward this physical model.
 Physical mechanisms of melt-solid interaction and melt transport are important in the generation of both the homogeneity and heterogeneity that we observe in Maui Nui lavas. Infiltration and “freezing” of gabbro-derived eclogite melts in the depleted oceanic peridotite leads to homogenization of the lower oceanic lithosphere-derived recycled segment in the plume. The inherent homogeneity of the Kea-type magmas results from the self-buffering nature of this refertilization process. Organization of upper crust-derived melt into isolated melt channels preserves the Ko'olau-type compositions from modification by interaction with peridotite matrix.
Appendix B:: Analytical Methods and Lana'i and East Moloka'i Data
 New isotope data for Lana'i and East Moloka'i are presented in Table B1. We completed sample preparation for Sr and Nd isotope analyses at the University of Washington (UW), Seattle, and for Pb isotope analyses at Ecole Normale Supérieure in Lyon (ENSL), according to procedures described by Nelson . We prepared Hf separations at ENSL, following the procedures described by Blichert-Toft et al. . TIMS (thermal ionization mass spectrometer) isotope analyses of Sr and Nd were carried out on the VG Sector at UW. MC-ICP-MS (multi-collector ICP-MS) isotope analyses of Pb and Hf were done on the VG Plasma 54 at ENSL [Blichert-Toft et al., 1997; White et al., 2000]. Pb isotope analyses used the Tl doping technique. Reported Pb isotope compositions are normalized to NIST SRM-981 206Pb/204Pb = 16.9356, 207Pb/204Pb = 15.4891, and 208Pb/204Pb = 36.7006 [Todt et al., 1996]. Two-sigma external reproducibility for Pb isotopic compositions is determined from repeat analyses of the 98B internal standard (n = 28 over two analytical sessions) and is 0.043%, 0.048% and 0.056% for 206Pb/204Pb, 207Pb/204Pb and 208Pb/204Pb, respectively. 176Hf/177Hf was normalized for mass fractionation relative to 179Hf/177Hf = 0.7325. The JMC 475 Hf standard gave 176Hf/177Hf = 0.282160 ± 0.000010 (2-sigma) and was measured between every second or third sample. Precision for 176Hf/177Hf is determined from replicate aliquots (n = 10) and analyses of our UWBCR-1 internal standard powder (176Hf/177Hf = 0.282865 ± 0.000016). Although our standard powder has been well-homogenized, the quoted precision represents a maximum value, given that it incorporates all procedural sources of error and any sample heterogeneity that may exist. Strontium NIST SRM-987 standards (n = 47) have an average 87Sr/86Sr = 0.710258 ± 0.000046 (2-sigma). Average composition of Nd La Jolla standards (n = 25) is 143Nd/144Nd = 0.511845 ± 0.000020 (2-sigma). This precision for the standard is similar to that obtained for multiple aliquots (n = 10) and analyses of our internal UWBCR-1 basalt powder (143Nd/144Nd = 0.512617 ± 0.000024, 2-sigma). The in-run precisions of all the Pb, Hf, Nd, and Sr isotope data presented here were better than the external reproducibilities reported above.
Table B1. Lana'i and East Moloka'i Isotope Compositions
RTH86 - K61
RTH86 - K64
RTH86 - K200
RTH86 - K206
RTH86 - K207
RTH88 - K1
RTH88 - K2
RTH94 - K13
RTH94 - K14
 New major and trace element data for Lana'i are presented in Table B2. Major element (XRF) and trace element (ICP-MS) analyses were completed at the Washington State University GeoAnalytical Laboratory, according to procedures described by Johnson et al. . Analytical precision is described in detail by Gaffney et al. .
Table B2. Lana'i Major and Trace Element Compositions
Major Elements, wt. %
Trace Elements, ppm
 We thank Mark Ghiorso for comments and discussion on an earlier version of this manuscript. We thank Gervais Hinn at UW and Philippe Télouk at ENS-Lyon for technical support and patient guidance. Marc Hirschmann and Anthony Koppers provided helpful reviews, and Robert Duncan and Bill White provided editorial comments. This work was supported by a Geological Society of America Harold T. Stearns grant, a DOSECC Internship grant and a UW Graduate Research grant to A.M.G., and supplemental NSF funding to B.K.N. to support analyses in Lyon. J.B.T. acknowledges financial support from the French Institut National des Sciences de l'Univers.