The Galápagos Spreading Center (GSC) between 90.5°W and 98°W manifests its interaction with the nearby Galápagos plume by way of variations in lava geochemistry, crustal thickness, and morphology along the ridge axis. Natural glasses from stations with ∼9 km average spacing were analyzed for major and minor elements, H2O, and CO2. Samples can be classified as enriched mid-ocean ridge basalts (E-MORB), transitional MORB (T-MORB), or normal MORB (N-MORB) on the basis of K/Ti ratios. E-MORB dominate the GSC east of 92.6°W. T-MORB are mainly found between 92.6°W and 95.5°W. West of the propagating rift tip at 95.5°W, N-MORB dominate. High K/Ti E-MORB also have higher H2O, Al2O3, and Na2O and lower FeO*, SiO2, and CaO/Al2O3 relative to N-MORB at similar values of MgO, characteristics consistent with lower mean extents of partial melting relative to N-MORB. We examine the melting process along this section of the GSC with a set of equations that simulate a deep zone of hydrous melting related to the depression of the mantle solidus by H2O. This model constrains the range of mantle source compositions, the depth of the additional hydrous melting zone, the melt productivity in the hydrous region, and the ratio of mantle flow rate through the hydrous zone relative to the anhydrous zone (Uw/U0) that can explain the measured crustal thickness as well as the fractionation-corrected concentrations of K, Na2O, H2O, and Ti along the GSC. Far from the hot spot, the measured crustal thickness and N-MORB compositions are explained by passive mantle upwelling (Uw/U0 = 1), mean melt fraction () ∼ 0.06, and a source with ∼35 ppm K, 130 ppm H2O, 2300 ppm Na2O, and 1050 ppm Ti. The transitional zone has a source enriched in K and could have a slight excess plume-driven flow through the hydrous melting zone (Uw/U0 ≤ 1.5). The crustal thickness and glass compositions in the “enriched” region of the GSC nearest the hot spot are best explained by only a slight increase in the temperature of the mantle (<∼20°C), coupled with a mantle source moderately enriched (relative to N-MORB source) and plume-driven flow through the hydrous zone of Uw/U0 = 1.5–3.5.
 The global mid-ocean ridge system is affected by near- and on-axis hot spots, commonly thought to be caused by mantle plumes. Mantle plumes can impose large physical and chemical anomalies on otherwise normal ridges. These anomalies provide insight into processes that often cannot be studied as well from a “normal” mid-ocean ridge such as intermediate-wavelength variations in mantle flow and melting, mantle source composition and mixing, and the effects of variable magma supply on axial morphology, basalt chemistry, and crustal accretion. The Galápagos Spreading Center (GSC), with its intermediate spreading rate and off-axis plume, provides an excellent setting for studying these processes as they relate to plume-ridge interaction.
 The nearly east-west trending GSC separates the Cocos and Nazca plates in the eastern equatorial Pacific at a full opening rate of ∼45–57 mm/yr in our study region [DeMets et al., 1994] (Figure 1). At 91°W the GSC lies ∼200 km north of the Galápagos Archipelago, the western end of which marks the probable center of the Galápagos mantle plume [Geist, 1992; White et al., 1993]. Effects of the nearby hot spot are manifest in the regional bathymetric swell that extends ∼1300 km along the ridge and peaks near 91°W, where the axial depth is more than 1 km shallower than portions of the ridge far from the hot spot, and in a regional mantle-Bouguer gravity anomaly that reaches its minimum (−70 mGal) near 91°W [Canales et al., 2002]. The hot spot effect can also be seen in variations in axial morphology along the ridge [Canales et al., 1997; Detrick et al., 2002; Sinton et al., 2003]. Within ∼350 km of the Galápagos hot spot the GSC has an axial high morphology; with increased distance from the hot spot it changes to a transitional morphology, ultimately becoming a Mid-Atlantic-Ridge-like rift valley farthest from the hot spot (Figure 2).
 A major objective of the Galápagos Plume Ridge Interaction Multidisciplinary Experiment (G-PRIME) [Detrick et al., 2002] is to better understand the processes of magma genesis along this hot spot-influenced ridge. Geophysical constraints on the rate of magma production are provided by measurements of crustal thickness from wide-angle seismic refraction experiments along three different portions of the GSC, as well as multichannel seismic reflection data gathered along and parallel to the ridge axis in the region 91.25°–95°W [Canales et al., 2002]. Our geochemical sampling program obtained dredge and wax core samples at 91 stations between 90.5°–98°W; one station is within the transform zone near 90.5°W, all other sampling stations are along the ridge axis, with an average spacing between stations of ∼9 km (Table 1, Figure 2). Sampling locations were chosen on the basis of multibeam bathymetric data collected during the cruise. In this paper we report major and minor element, H2O and CO2 data for G-PRIME (G′) glasses (Table 2). In addition, we reanalyzed selected glasses from the 95.5°W area [Christie and Sinton, 1981, 1986; Yonover, 1989; Hey et al., 1992] in order to ensure uniform data quality and to augment the G′ data set. We incorporate these data, along with our seismic constraints on crustal thickness, into a simple, inverse model to examine the source compositions and conditions of melting along this hot spot-influenced section of the GSC.
Table 1. Location and Estimated Recovery of Sampling Stationsa
Stn, dredge or wax core station number; osc, Overlapping Spreading Center; ol, olivine; pl, plagioclase.
All values in wt%, except CO2 (ppm). Major and minor elements by University of Hawai'i electron microprobe; H2O and CO2 by University of Miami FTIR. FeO* is total Fe as FeO. Compositions for major and minor elements are averages for all individual samples within a group; n indicates the number of individual samples averaged in each group.
Sample group names are the station number followed by the sampling method: D for dredge or C for wax core. Alvin and Atlantis samples [Hey et al., 1992] are named according to the four-digit dive number or dredge number (preceded by A). Multiple groups per station are designated as a, b, or c.
MORB types (see text).
Individual sample number analyzed by FTIR.
Precision for the major and minor element analyses based on standard deviations of groups containing nine or more individual samples.
 Glass compositions were measured using the University of Hawaii Cameca SX-50, five-spectrometer electron microprobe. Major and minor element analyses were obtained on glass chips from ∼200 individual samples. Alvin samples and Atlantis dredge samples [Hey et al., 1992] were reanalyzed using the same procedures. Reported analyses are averages of ten spots collected from three to six glass chips per sample, using an accelerating voltage of 15 kV, 10 nA beam current, and 10 μm beam diameter. Peak counting times were 110 seconds for P; 60 seconds for K; 50 seconds for Mn; 40 seconds for Fe; 30 seconds for Mg, Al, Si, Ca, and Ti; and 20 seconds for Na. Background counting times were 90 seconds for P; 30 seconds for K; 20 seconds for Mn; 10 seconds for Na, Ca, and Fe; and 5 seconds for Mg, Al, Si, and Ti. Samples were calibrated using Makaopuhi glass standard A-99 (Mg, Si, Ti, Fe), Juan de Fuca glass standard VG-2 (Na, Al, Ca), and mineral standards apatite (P) and orthoclase (K). A PAP matrix correction was applied to all analyses.
 The number of individual samples analyzed per dredge varied between 1 and 12, depending on the size of the dredge haul. Dredges were dragged over ∼200–500 meters of seafloor, so it is possible that rocks in a dredge haul include samples from more than one eruptive unit. Individual samples from each station with compositions that agreed within ∼5–10% in the low-abundance oxides K2O, P2O5, TiO2, and Na2O were designated as a single group. High-abundance element concentrations were compared for consistency of the groupings; in all cases some variance between groups is observed in all elements. Most dredges contain only one group; the maximum number of groups in a dredge is three (dredges 41, 55). Once groups were determined, all analyses for samples from that group were averaged (Table 2). Each group probably represents a single lava flow or group of closely related flows. An estimate of the precision of the microprobe data is reported as one standard deviation from the mean for groups containing 9 or more individual samples (Table 2).
 On the basis of major element composition and along-axis locations, one glass sample from each of 42 of the groups was analyzed for dissolved H2O and CO2 using infrared spectroscopy at the University of Miami following the procedures of Dixon and Clague . Precision of the analyses is about ±2% for total water, and ±7 to 10% for molecular water and carbonate. Because of the larger uncertainty in the compositional dependence of the molar absorptivity for water dissolved as molecular water and carbon dissolved as carbonate in silicate glasses, the accuracy of the molecular water and carbonate analyses is estimated to be about ±20%.
3.1. Classification of MORB Types
 To assess potential differences in parental magma compositions, we use K/Ti ratios to divide our samples (Figure 3) [Schilling et al., 1983; Hekinian et al., 1989; Sinton et al., 1991; Langmuir et al., 1992], because K/Ti is relatively unaffected by fractionation processes at MgO values greater than that at which FeTi oxide appears as a fractionating phase (∼4.5 wt% MgO in G′ samples; see Figures 4c and 4e). G′ samples can be divided into three general types: enriched (E-) MORB, with K/Ti ratios >0.15 and K2O contents >0.20% (all percentages are given in weight percent); transitional (T-) MORB, with K/Ti ratios mainly between 0.09 and 0.15; and normal (N-) MORB, with K/Ti ratios <0.09 (Figure 3). K/Ti ratios alone effectively discriminate the three sample groups. Most MORB with K/Ti > 0.15 have K2O abundances >0.20%, and are therefore “enriched” in the more incompatible element, K. However, two sample groups (15D and 39D-b) have K/Ti > 0.15 but K2O < 0.20%. On most plots, including those involving trace element data not reported here, these samples follow trends associated with T-MORB. Thus we restrict the definition of GSC E-MORB in this paper to sample groups with K/Ti > 0.15 and K2O > 0.20%.
 Although the threefold division accounts for most of the compositional diversity, it is apparent that significant variability at constant MgO, especially in SiO2 and Al2O3, is present within the E- and T-MORB types (Figures 4e–4i). We have therefore subdivided these types as follows: Relative to most E-MORB (hereafter E1-MORB), E2-MORB have anomalously low SiO2 and high CaO/Al2O3, the latter because of both lower CaO and higher Al2O3 than E1-MORB. T2-MORB have slightly higher incompatible element values, particularly K2O, relative to “regular” T1-MORB. T3-MORB have low CaO/Al2O3 ratios, and low SiO2, as well as higher Na2O and K2O than T1-MORB.
3.2. Compositional Variations
 Variations in glass compositions with MgO (Figure 4) indicate large variations in fractional crystallization along the GSC. The least differentiated of all G′ samples are N-MORB, which have MgO values that range from 6.9% to almost 10%. Slight changes in the slopes of MgO versus CaO, SiO2, and CaO/Al2O3 for N-MORB show that plagioclase joins olivine in the fractionating assemblage between ∼8.0 and 8.5% MgO. N-MORB are characterized by low concentrations of elements that are incompatible during mantle melting: ≤0.08% K2O, 1.67–2.56% Na2O, 0.77–1.68% TiO2, 0.05–0.12% P2O5. N-MORB have the lowest H2O concentrations of all G′ samples, with no values greater than 0.22% (Figure 5a). These “normal” MORB values are typical of, or even slightly lower than, H2O values found along other “normal” portions of the global mid-ocean ridge system [Michael, 1988, 1995; Dixon et al., 1988; Danyushevsky et al., 2000].
 As a group, T-MORB are slightly more differentiated than N-MORB, with MgO contents ranging from 6.1 to 9.3%, and with the bulk of the samples having MgO < 8.0% (Figure 4). T-MORB are most distinct from N-MORB in K2O, giving rise to their higher K/Ti ratios. T-MORB have average values of incompatible oxides Na2O, P2O5, TiO2, and FeO* (total Fe reported as FeO) higher than in N-MORB because the N-MORB tend to be less fractionated, but the differentiation trends are collinear. Differentiation trends for the T-MORB also are collinear with those for N-MORB for CaO, Al2O3, and SiO2. As with the N-MORB, changes in the slope of CaO and SiO2 trends for the T-MORB indicate plagioclase fractionation beginning between ∼8.0 and 8.5% MgO.
 T2-MORB differ from T1-MORB in CaO/Al2O3, mostly because of lower CaO. T2-MORB also are enriched slightly in TiO2, K2O, and H2O relative to T1-MORB. T3- MORB have much lower CaO/Al2O3 than T1-MORB, as a result of both lower CaO and higher Al2O3 at a given value of MgO. T3-MORB have significantly lower SiO2 than T1-MORB, and are slightly enriched relative to T1-MORB in Na2O, K2O, TiO2, P2O5, and H2O.
 E-MORB are more differentiated than N- and T-MORB, with MgO contents ranging to values as low as 3.3% (Figure 4). E-MORB are enriched relative to N- and T-MORB in K2O, Na2O, P2O5, and H2O. TiO2 values are slightly higher at a given MgO value in E-MORB than in T-MORB. Lower relative FeO* and SiO2, and higher Al2O3 values also characterize E-MORB. E1-MORB Al2O3 and CaO/Al2O3 variations indicate that clinopyroxene joins the fractionating assemblage near ∼5.5–6.0% MgO. FeO and TiO2 plots show that oxides fractionated from E-MORB magmas with less than ∼4.5% MgO.
 E2-MORB show the same relationships to E1-MORB as do T3-MORB relative to T1-MORB: they have lower CaO/Al2O3 as a result of both lower CaO and higher Al2O3 at a given MgO value, and significantly lower SiO2. E2-MORB are among the least differentiated E-MORB. E2-MORB also tend to have slightly higher concentrations of incompatible oxides Na2O, K2O, TiO2, and P2O5 than E1-MORB at the same MgO value.
 Samples from group 78D are scoriaceous: all glass with >60% vesicularity. Group 78D, with 6.15% MgO, is classified as an E2-MORB on the basis of its low SiO2 content and high values of Na2O, K2O, TiO2, and P2O5 relative to E1-MORB (Figure 4). This sample group has CaO and Al2O3 compositions that are more similar to E1-MORB, but H2O content (0.18%) that is much lower than E1- or E2-MORB, indicating that it is degassed, consistent with its high vesicle content.
 H2O contents of GSC samples correlate positively with K2O contents (Figure 5b), indicating that H2O behaves incompatibly, with H2O slightly more compatible than K2O (i.e., bulk distribution coefficient (D) for H2O > D for K2O). This result is consistent with the conclusions of Michael [1988, 1995], Dixon et al. , and Danyushevsky et al. , who argued that water in basalts behaves incompatibly during melting and crystallization, with a D of ∼0.01. The strong positive correlation between K2O and H2O suggests that the same processes that affect abundances of other incompatible elements in MORB also control H2O. Additionally, this excellent correlation is consistent with negligible contamination of the magmas by seawater, which should elevate H2O relative to K2O in a nonsystematic way [Danyushevsky et al., 2000].
3.3. MgO8.0 Calculations
 In order to determine parental magma compositions for each magma type, we have adjusted our data for the effects of low-pressure differentiation by calculating oxide values at MgO = 8.0% [Klein and Langmuir, 1987]. The adjusted values are indicated hereafter with a subscript of 8.0. This adjustment allows for comparisons among lava types as well as with global MORB data sets. We used least squares regressions to quantify the oxide variations versus MgO for each type and adjusted all data using the slope of the regression. Samples with MgO contents >8.5% or <3.5% were excluded; thus the empirical lines along which we adjusted the data probably represent cotectic crystallization of olivine and plagioclase ± pyroxene only. Separate slopes were calculated for each of the three general magma types (i.e., N-, T1- and E1-MORB). Straight lines were fitted through SiO2, Al2O3, FeO*, and CaO data. Observed trends for the minor elements K2O, Na2O, TiO2, P2O5, and H2O are curved, as expected for incompatible elements approximating Henry's Law behavior; accordingly, best-fit power law curves were used for adjusting these data.
 Regression of the E2-MORB data alone would not produce a statistically valid regression due to the small number of data points. We therefore adjusted E2-MORB using the same slope as for E1-MORB, assuming that they evolved by similar processes, but from a different parental magma. Similarly, T2- and T3-MORB were excluded from the regressions, but oxide values were adjusted using the same slope as used for T1-MORB. Slopes determined in this study are given in Table 3.
Table 3. Empirical Slopes Calculated for MgO(8.0) Adjustmentsa
Power Law Regressions
Linear regressions were applied to SiO2, Al2O3, FeO*, and CaO; power law regressions (y = a*xm, where the shown numbers are values of m) were applied to incompatible elements TiO2, Na2O, K2O, P2O5, and H2O. Samples with MgO < 3.5 wt% or > 8.5 wt% were excluded from the regressions, with the exception that regression of CaO, Na2O, K2O, P2O5, and H2O in E-MORB includes samples with <3.5 wt.% MgO; E-MORB trends for these oxides are not affected by the appearance of new fractionating phases at this composition. No T2, T3, or E2 samples were used to calculate slopes of lines or curves, with the exception of H2O, for which T2 and T3 data were included for curve fitting. However, all E2, T2, and T3 data were corrected using the equations and slopes as used for the “regular” groups (i.e., T1 and E1).
3.4. Along-Axis Variations
 E-MORB dominate the GSC east of 92.6°W; T-MORB are mainly found between 92.6°W and 95.5°W (shaded area in Figures 6a–6j); only N-MORB (except for the 78D scoria E2-MORB) are found west of 95.7°W. The limit of E-MORB occurrence is consistent with the “plume-influenced” “type B” basalts of Fisk et al . Fisk et al.  and Schilling et al.  also noted petrological differences between basalts from within the “H-zone” (high magnetic amplitude zone [Anderson et al., 1975]), which is confined to the region of the GSC between the tips of the propagating rifts at 95.5°W and 85°W, and those outside of it. The boundary separating N-MORB- and T-MORB-dominated provinces in our data coincides with the 95.5°W propagating rift tip (Figure 6a).
 The western limit of the E-MORB-dominated province near 92.7°W, coincides with the transition from an axial high morphology (to the east) and the rapid doubling of the depth to the seismically imaged axial magma lens and base of layer 2A (from east to west) [Detrick et al., 2002]. The westernmost extent of T-MORB, 95.7°W, corresponds with the propagating rift (PR) system near 95.5°W. This PR tip marks the boundary between axial rift-valley morphology, to the west, and the region between 95.5°W and 92.7°W, dominated by transitional morphology.
 Within the N-MORB region, FeO*(8.0) and TiO2(8.0) decrease slightly from west to east, while SiO2(8.0) increases. Between ∼95.1°W and 95.5°W, FeO*(8.0) and TiO2(8.0) are elevated relative to the surrounding regions [Christie and Sinton, 1981, 1986; Sinton et al., 1983]; H2O(8.0) also is high in this region. From ∼94.2°W to the propagating rift tip near 95.5°W, Mg# (molar MgO/(MgO + FeO*)) shows a well-defined trend, decreasing to the west. Between ∼94.2°W and the eastern edge of the overlapping spreading center system at ∼93.2°W, FeO*(8.0) and SiO2(8.0) are nearly constant, while Mg# is highly variable compared to its narrow range at any given location farther to the west. East of ∼93.1°W, T-MORB FeO*(8.0) and SiO2(8.0) values decrease. These trends continue beyond the first occurrence of E-MORB at 92.7°W until they reach their minima near 91.7°W.
 Along-axis variations in elements adjusted to 8.0% MgO show that the most extreme values are between ∼91.7°W and 92.4°W (Figures 6a–6j). K2O(8.0), TiO2(8.0), Al2O3(8.0), Na2O(8.0), P2O5(8.0), and H2O(8.0) peak in this region, while FeO*(8.0), SiO2 (8.0), CaO(8.0), and CaO(8.0)/Al2O3(8.0) are at their lowest values. We infer that these geochemical features characterize the most plume-influenced rocks.
 Samples from dredge 17, at 91.8°W, have the highest TiO2(8.0), Al2O3(8.0), Na2O(8.0), and P2O5(8.0), and the lowest FeO*(8.0), SiO2(8.0), and CaO(8.0)/Al2O3(8.0), thus defining the location of greatest plume influence along the western GSC. We note that the apparent peaks in K/Ti near 91.4°W (samples 11D-a and 11D-b, Figure 6f) probably reflect peaks in fractionation (MgO contents of 3.33 and 3.98%, respectively) rather than peaks in parental magma. The parental magma peak in K/Ti is therefore farther west, which is in agreement with the extremes in the other oxides. The greatest geochemical signature of plume influence also coincides with the greatest crustal thickness (∼8 km) [Canales et al., 2002].
4. Modeling Hydrous Melting
 Compared to N-MORB, E-MORB are enriched in oxides that are incompatible during melting of mantle peridotite (e.g., K2O, Na2O, TiO2, Al2O3, H2O). As shown by Schilling et al. , the relative enrichment is approximately inversely proportional to the bulk distribution coefficient (D) for melting of mantle peridotite, i.e., the most incompatible elements are the most enriched. This relationship suggests that variable melting processes are important in controlling the observed distribution along axis [Fisk et al., 1982; Schilling et al., 1982]. Verma and Schilling  and Verma et al.  showed that Sr and Nd isotope ratios also vary along axis, with the highest 87Sr/86Sr and lowest 143Nd/144Nd occurring at 91°–92°W. As argued by these authors, these results also require variation in the contribution of at least two mantle source components along axis, with the higher 87Sr/86Sr and lower 143Nd/144Nd source components increasing to the east.
 In addition to the incompatible trace element and isotopic variations are variations in SiO2, Al2O3, FeO and CaO, all of which occupy principal lattice sites in the major mantle minerals olivine, pyroxenes, spinel, and garnet. As such, variations in these oxides are more likely to be controlled by stoichiometry during melting than by source enrichment. It is notable that many of the chemical characteristics of the most plume-influenced portion of the GSC are consistent with relatively low extents of partial melting. For example, E-MORB are enriched in incompatible elements, and this enrichment is coupled to high Al2O3 and low SiO2 and CaO/Al2O3. The E-MORB lavas, which carry the strongest signature of relatively low extents of melting, are, however, found in the region of shallowest depths and greatest crustal thickness [Detrick et al., 2002] and are nearest to the Galapagos “hot spot.” Recent studies [e.g., Plank and Langmuir, 1992; Detrick et al., 2002; Asimow and Langmuir, 2003; Asimow et al., 2004] suggest that a key to understanding the apparent paradox (maximum melt production (thickest crust) coinciding with low mean extents of melting) is in the role of water on the melting of upwelling mantle peridotite. For example, Asimow and Langmuir  showed that the observed increase in crustal thickness close to the Galápagos hot spot can be associated with ∼40% reduction in mean degree of melting when water is incorporated into mantle melting models.
 The recognition of the potential importance of water to enhancing melt production was first noted for the Galápagos region by Schilling et al.  and Fisk et al.  and for the region around the Azores hot spot [Schilling et al., 1980; Bonatti, 1990; Asimow et al., 2004]]. Asimow and Langmuir  developed quantitative models that relate source composition, including water content, to melting processes along the GSC using previously published data as constraints. This work indicated that hydrous melting can explain many of the compositional and crustal thickness variations of the region and that potential temperatures along the hot spot-affected ridge are less than those predicted from anhydrous models. In the following sections, we develop an alternative hydrous melting model that allows for variable extents of active upwelling in the hydrous and anhydrous melting regions. The results of this modeling support the general conclusions of Asimow and Langmuir , with important differences in the required compositional and temperature variations along axis.
Plank and Langmuir  (hereafter PL92) evaluated the effects of melting and mixing melts on crustal composition by examining the residual melting column (RMC, Figure 7a), the conceptual “net result” of melting of the mantle [see Langmuir et al., 1992; PL92]. The RMC is useful for calculating the volume and composition of the aggregate melt that makes up the oceanic crust, somewhat independent of the physical shape of the melting region.
 When the rate of mantle flowing through and exiting the melting region is constant over the depth of the melting zone, the width of the RMC is constant and the mean concentration () of a given element in the melt arising out of the RMC is
where F is melt fraction. The quantities and F are related by a melting function, CL(F). PL92, for example, used the accumulated fractional melting (AFM) equation [Shaw, 1970],
where D is the bulk solid/liquid partition coefficient of the element being evaluated, C0 is the initial source concentration of that element, and fractional melts are pooled between melt fractions 0 and F.
 The mean fraction of melting, , is the total quantity of melt divided by the height of the melting zone,
where z is depth below the top of the melting zone, dz is an infinitely small depth interval, and melting begins at z = hD. Melt productivity, B, is dF/dz. If it is uniform with depth, F(z) = Bz, B = Fmax/hD, and
where Fmax is the maximum extent of melting at the top of the RMC. Crustal thickness, Zcr, which is proportional to the total volume of melt extracted from the melting region, is calculated simply by multiplying by hD, the depth to the solidus.
Equations (3) and (5) correspond with PL92 equations (7) and (8), respectively. These equations are based on four assumptions: flow rate through and out of the melting zone that is uniform with depth (i.e., RMC of uniform thickness), adiabatic melting, perfect fractional melting, and constant melt productivity.
 In Figure 7b, we relax the assumptions of uniform mantle flow and constant melt productivity (Figure 7b). In this case, the RMC width is controlled by U(z), the rate of flow through and out of the melting region. The mass flux of melt removed at a given depth is proportional to U(z)F(z)dz, the mass flux of a given element is proportional to CL(z)U(z)F(z)dz, and therefore adding all melt concentrations from all depths in the RMC yields
We greatly simplify the rather complex change in melting behavior as a function of pressure by separating the melting zone into two regions: an anhydrous melting zone spanning depths of z = 0 − hD, and a hydrous zone with z = hD − hw. We approximate the change in melting behavior between the two zones by specifying constant but distinct average productivities, Bw and BD, for the hydrous and dry melting zones, respectively (Figure 7).
Equation (6) also allows us to treat a more complex mantle flow function U(z) associated with plume-ridge interaction. For a normal mid-ocean ridge where mantle flow is passively driven by plate separation, U(z) is nearly constant (i.e., approximately the half spreading rate of the ridge) and therefore this term drops out of (6). The flow rate of a mantle plume beneath the lithosphere, however, is likely to vary as a strong function of depth [e.g., Ito and Mahoney, 2002; Ribe et al., 1995; Ito et al., 1999]. Deep in the upper mantle, the buoyant mantle plume stem rises rapidly, possibly an order of magnitude or more times typical plate motions [e.g., Ito et al., 1999]. As it begins to interact with the lithosphere, the material ascends at a slower rate and is diverted sideways. Mantle plume buoyancy can thus push mantle rapidly into and out of the deep (hydrous) portion of the melting zone many times more rapidly than plate spreading can pull material in and out of the shallow (anhydrous) portion of the melting zone. Again, we simplify the continuous function of U(z) by specifying an average velocity Uw through the hydrous zone and an average velocity U0 through the anhydrous melting zone. If U0 is approximately the half spreading rate of the two plates, then the fundamental parameter Uw/U0 describes the speed of deep mantle plume flow relative to plate motion, with Uw/U0 = 1 representing the condition beneath a normal mid-ocean ridge, and Uw/U0 > 1 representing plume influence. Plume-driven flow (Uw/U0 > 1) also contrasts with “active” mantle flow beneath mid-ocean ridges, which is predicted to be driven by the more shallow buoyancy associated with melting. This shallower, melt-related buoyancy enhances the flux of mantle though the shallow portion of the melting zone [Scott and Stevenson, 1989; Turcotte and Morgan, 1992] and thus could be simulated by Uw/U0 < 1, but in this study, we examine only Uw/U0 ≥ 1.
 With these assumptions, an equation describing pooled melt concentrations,
has two parts. The first integral in the numerator represents melting in the hydrous zone to a maximum fraction Fw and the second integral in the numerator represents melting in the anhydrous region, which continues to Fmax. With
and using the AFM function (2) to relate CL to F, we integrate equation (7) and get our “hydrous melting equation,”
 The denominator of equation (10) is proportional to crustal thickness (Zcr), with the first and second terms representing contributions to the crust by hydrous melting and anhydrous melting, respectively. Seismically constrained values of Zcr [Canales et al., 2002] can be incorporated into (10) by substituting
where the ratio of crustal density (ρc) to mantle density (ρm) converts weight fraction to volume fraction.
 Finally, the mean degree of melting, , is the weighted average of the mean degree of melting in the hydrous and anhydrous regions,
4.2. General Predictions of the Hydrous Melting Model
Figure 8 demonstrates the effects of hydrous melting and plume-driven flow (equation (10)) compared to “dry” melting and uniform flow (equation (3)). As hw increases with increasing concentrations of H2O in the mantle source, decreases (Figure 8a), reflecting the contribution from a relatively large volume of mantle melting to only small degrees. The plume-driven flow parameter, Uw/U0 also controls the flux of material melting in the hydrous region. Increasing Uw/U0 leads to decreases in at the same value of hw (Figure 8a).
Figure 8b shows the effects of hw and Uw/U0 on the enrichment of an incompatible element (D = 0.01 shown) in a melt, compared to that predicted by the equation for anhydrous, uniform mantle flow (equation (3)). When melting is only occurring in the hydrous region, Fmax ≤ Fw and CL/C0 follows the same curve for all conditions, including anhydrous melting. As soon as productivity increases at the onset of dry melting, the predicted curves diverge. For hydrous plus anhydrous melting, a proportionally smaller flux of depleted melts is diluting the hydrous melts and CL/C0 remains higher than the curve for anhydrous melting only. Melt enrichment (CL/C0) increases with increasing hydrous melting interval hw and flow parameter Uw/U0 because both of these parameters increase the relative flux of hydrous (low-degree) melting. For example, compared to reference values of hw = 30 km and Uw/U0 = 1, approximately doubling hw (hw = 50 km, Uw/U0 = 1) has approximately the same effect of doubling Uw/U0 (hw = 30 km, Uw/U0 = 2), both increasing CL/C0 by ∼15% at Fmax = 0.2. Increasing Uw/U0 to 5 approximately doubles CL/C0 at Fmax = 0.2. These general calculations illustrate the importance of hydrous melting and plume-driven flow on the amount melts are enriched relative to the initial source. As we demonstrate next, it is thus necessary to consider these factors when using observed variations in magma compositions to constrain source variations.
4.3. Application to G′ Data
 Although isotopic data [e.g., Schilling et al., 2003] suggest that there may be quasi-continuous variation in source composition along axis, major and minor element glass data indicate that the western GSC can be divided into three broad regions respectively dominated by N-, T-, and E-MORB. In this study we use compositional data that are the average fractionation-corrected compositions for lavas from these three regions. Because the processes controlling major element oxides such as SiO2, CaO, Al2O3, and FeO* are not easily accounted for by simple distribution coefficients, we focus only on the incompatible components K, Na2O, Ti, and H2O. The behavior of these elements during melting is moderately well known [e.g., Dixon et al., 1988; Johnson et al., 1990; Langmuir et al., 1992; Michael, 1995; Niu et al., 1996]. Although partitioning relations can be expected to vary with changing temperature, pressure and composition, there are too few detailed data to allow for a fully quantitative treatment of these effects. We have therefore adopted the common approach of using constant D values throughout the melting process (Table 4).
Table 4. Variables Used in Model
Grid Search Step Size
The computational procedure involves searching through possible values of C0Na. Search through C0Na is arbitrary; searching through the source composition of any other element yields identical solutions.
maximum degree of melting in zone of hydrous melting
 In the hydrous melting equation (equation (10)), CL is controlled by the independent variables D, Fmax, BD, hD, Bw, hw, C0, Uw/U0, and Zcr (with equation (11)). The average values of Zcr. for each portion of the ridge, and the average values of K(8.0), H2O(8.0), Na2O(8.0), and Ti (8.0) are input into our model as constant, singular variables for each general MORB type (Table 4). For melt productivities, we consider BD values of 0.26–0.53%/km (0.8–1.6%/kbar). This range of productivities fully encompasses the values used by Langmuir et al.  and McKenzie and Bickle , and is consistent with average productivities in the anhydrous region shown by Asimow et al. . We considered a range of Bw values between 0.03–0.05%/km (0.09–0.15%/kbar), which we estimated from curves of Asimow et al. [2001, 2004].
 Water concentration in the source (C0H2O) and the height of the wet melting column (hw) are linked, but the precise relationship is not yet well understood. Hirth and Kohlstedt  schematically related the depression of the hydrous solidus to depth (and pressure) on the basis of calculated contours of the activity of water in olivine. Bell et al.  asserted that due to errors in calibration, all estimations of water solubility in olivine to date are underestimates and require upward revision by a factor of 2–4. Katz et al. , Asimow and Langmuir , and Asimow et al.  incorporate nonlinear dependences of hw on C0H2O. The parameterization of Katz et al.  predicts an increase of hw due to an increase in water content of ∼0.10 ± 0.03 km/ppm at C0H2O = 550 ppm, whereas the dependence along the adiabats of Asimow et al.  decreases from approximately 0.42 km/ppm at C0H2O = 50 ppm to ∼0.19 km/ppm at C0H2O = 300 ppm. In a later section we will show that G′ data are consistent with C0H2O values <∼250 ppm. Thus a simplified linear relation
with hw in km and C0H2O in ppm, is a reasonable approximation to the solidus depression at relatively low water contents.
 For each MORB type, we have seven equations: equations (8), (9), (11), and (10) for K8.0, H2O8.0, Na2O8.0, and Ti (8.0), plus constraints on an eighth equation (equation (13)). It is thus possible to place bounds on eight unknowns: Fmax, hD, hw, Uw/U0, and four C0s. We use a grid search method to invert for the range of solutions that satisfy data on CL and Zcr, given ranges of BD, Bw, and hw/C0H2O.
4.4. Relation to Previous Hydrous Melting Models
 Various methods have been developed that use geochemical data to understand the partial melting process. Like us, McKenzie and O'Nions  use an inverse method, but their method differs in that it assumes a source composition and uniform mantle flow (Uw/U0 = 1) to invert for F as a function of depth (i.e., effectively inverting for Fmax, hD, and BD). Maclennan et al.  built upon McKenzie and O'Nions  to also invert for U(z). Both McKenzie and O'Nions  and Maclennan et al.  only considered anhydrous melting and neither uses crustal thickness directly in the inversion. In contrast, the forward methods of Katz et al.  and Asimow and Langmuir  explicitly incorporate the effects of water, consistent with thermodynamic constraints calibrated against available experimental data on peridotite melting, solubility relations and hydrous equilibria. These models also allow for continuous variation in productivity with depth and water content as well as for variations in modal mineralogy of the source as melting proceeds. As such, these models represent the most realistic models of peridotite melting in the presence of water that are currently available. Such models present many advantages. The principal disadvantages are that many of the dependencies in these models are poorly constrained by currently available experimental data and the models are computationally complex.
 Our more rudimentary treatment of hydrous melting is based on analytic solutions of basic principles of mass balance and phase equilibria; it is relatively simple computationally and extremely flexible. It allows several variables to vary independently, even when we suspect that there should be dependencies among some of them (e.g., Bw and C0H2O [Asimow et al., 2004]). This allows us to produce a broad range of solutions that can then be evaluated for reasonableness as our understanding of the different dependencies improves. The most important weakness of our method is that we must prescribe a range of productivities and hw/C0H2O, and thus can only place approximate bounds on possible solutions. One obvious strength is that we can place bounds on all the unknowns, including source composition (i.e., C0s). A unique feature of our model is that it allows for different upwelling rates in different parts of the melting regime.
Figure 9a shows the compositional range of sources that can be melted to create our observed average fractionation-adjusted N-MORB concentrations of K, H2O, Na2O, and Ti and the observed crustal thickness of 5.7 km in the GSC “N-MORB” region (near 97°W). (See Table 5 for parameters associated with these curves.)
Table 5. Range of Model Inputs That Combine to Match G′ Dataa
Solutions listed produce magma compositions that match G′ crustal thickness and composition values from Table 4. These maximum and minimum values, normalized to the N-MORB midpoint value, were used to plot Figure 9. Symbols are explained in Table 5 and text.
Dry solutions are the minimum (Dry1) and maximum (Dry2) of all solutions predicted by the anhydrous pooled melting equation (equation (3)) that fit our inputs and constraints. All variables that were used to create these minimum and maximum solutions are listed, as well as corresponding model outputs. By definition, anhydrous solutions have Fw, Bw, and hw = 0.
Minimum (min) and maximum (max) source concentrations predicted by the hydrous melting equation (equation (10)).
Midpoint source concentration values are used as normalizing values for all other solutions when plotting Figure 9. The input variables that created these solutions are closest to the midpoint of the range allowed for each input variable, with the exception of Uw/U0, for which only passive upwelling (Uw/U0 = 1) was allowed.
 First, we examine predictions of dry melting only. In this scenario we treat H2O as an incompatible oxide but do not consider the effects of water on melting, i.e., equation (3), hw = Fw = 0, and Uw/U0 = 1. Because we consider a range of BD values, there are a number of possible source compositions (dashed and dotted lines in Figure 9). Dry melting requires 41–53 ppm K in the source, 136–175 ppm H2O, 2100–2600 ppm Na2O, and 991–1136 ppm Ti. The value of for dry melting ranges from 0.08 to 0.104, encompassing those of Asimow and Langmuir , but extending to slightly lower values. Lower BD values combined with a deep dry solidus (high hD) require a less enriched source and lower than high productivity over a shorter dry melting column (Table 5).
 In comparison to the dry melting solutions, consideration of a hydrous melting region reproduces our N-MORB data at lower incompatible element concentrations in the source at lower values of . Using the hydrous melting equation (equation (10)), N-MORB compositions can be produced from a source with as little as 26 ppm K, 107 ppm H2O, 1900 ppm Na2O, and 955 ppm Ti, and as low as 0.046. This low solution is predicted with hw = 26 km, Bw = 0.04%/km, Uw/U0 = 2, hD = 54 km, and BD = 0.27%/km. Passive hydrous melting (Uw/U0 = 1) with hw = 36 km and Bw = 0.03%/km plus anhydrous melting with hD = 41.5 km and BD = 0.49%/km requires the maximum source concentration: 35 ppm K, 141 ppm H2O, 2400 ppm Na2O, and 1113 ppm Ti. Concentrations between these extreme values are also viable, but created by different combinations of the above variables.
 These N-MORB results elucidate the importance of source H2O and variable flow in the mantle. As emphasized by Asimow and Langmuir , hydrous melting produces N-MORB solutions with lower average degrees of melting than the anhydrous pooled melting equation, thus reducing the required concentrations of the incompatible elements in the mantle source and the required temperature of the mantle.
5.2. E-MORB and T-MORB
 In evaluating the E-MORB and T-MORB data, we consider several important questions. Is an enriched mantle in the E-MORB region required to explain major and minor element basalt compositions that are incompatible-element enriched relative to N-MORB? Is plume-driven upwelling required? What are the effects of variable productivities within the hydrous melting region? What is the additional depth to the hydrous solidus? Must E-MORB be produced by elevated mantle temperatures relative to N-MORB?
Figure 9 compares the maximum and minimum source compositions required for each MORB type for solutions to both the hydrous (equation (10)) and anhydrous equations (equation (3)). All solutions are normalized to the source values for the N-MORB “midpoint.” E-MORB solutions span a large range of possible source concentrations, which encompasses nearly the entire range of N-MORB solutions. Many, but not all, of the E-MORB solutions require a more enriched source than for N-MORB. T-MORB solutions span a range that includes solutions with both higher and lower source concentrations than the maximum and minimum N-MORB solutions, respectively, although T-MORB maximum values for H2O, Na2O, and Ti are only very slightly greater than N-MORB maximum values. Parameters corresponding with each of the lines plotted on Figure 9 are reported in Table 5.
 The required source concentration depends strongly on the mantle flow parameter Uw/U0 (Figures 9 and 10). Greater Uw/U0 requires less source enrichment. The dependence on Uw/U0 is strongest for the most incompatible element, K; K values of E-MORB can be produced from N-MORB source compositions only at Uw/U0 ≥ 10. T-MORB require a source enriched in K relative to that for N-MORB only at Uw/U0 ≤ 3, but not for greater values of Uw/U0. Neither T-MORB nor E-MORB require Na-enrichment, although at low Uw/U0 some T- and E-MORB solutions have slightly higher Na contents than the N-MORB sources. At Uw/U0 ≥ 5, some E- and T-MORB solutions require sources with less Na than N-MORB. For Ti, the E-MORB compositions require at least some source enrichment for all values of Uw/U0. The Ti content of the T-MORB source does not vary significantly from that of the N-MORB source. The required source concentration also depends, although less significantly, upon Bw. In general, the more “productive” the hydrous melting region, the less source enrichment is required (Figure 10).
 The above results pertain only to the K, Na, Ti and H2O variability of mantle sources contributing to average N-, T-, and E-MORB magmas along the western GSC. Geochemical data not considered in this model obviously can be used to further constrain source components along the GSC. Although our models do not require the sources for T-MORB and E-MORB to be enriched in Na relative to that for N-MORB, other data suggest that mantle sources contributing to the production of T-MORB and E-MORB along the western GSC have radiogenic isotope and highly incompatible-element ratios that are distinct from those involved in the production of N-MORB [e.g., Schilling et al., 1982, 2003; Verma and Schilling, 1982; Verma et al., 1983].
 The relationship between the source concentration of H2O and is shown in Figure 11a. The model allows E-MORB mantle to contain as little as ∼110 ppm H2O or as much as ∼240 ppm, compared to between 110 and 140 ppm H2O in our N-MORB mantle and 80–150 ppm in our T-MORB mantle. Mean degree of melting for E-MORB may be 0.019 to 0.067, depending on Uw/U0 and productivity in the hydrous region (Bw). By comparison, mean extent of melting for the N-MORB could range from 0.046 to 0.062, and for T-MORB could range from 0.021 to 0.069. is dominantly controlled by Uw/U0 with increasing Uw/U0 producing lower (Figure 11b). For the range of Uw/U0's (≤2) examined for the N-MORB data, models cannot resolve a difference in mean between the N-MORB and E-MORB. Thus, if indeed E-MORB are created by lower than N-MORB, as inferred from SiO2, and CaO/Al2O3 values [see also Asimow and Langmuir, 2003], there must be more plume-driven excess flow in the E-MORB region than in the N-MORB region.
 We explore the range of possible values of hD, the depth interval of dry melting, for N-, T-, and E-MORB in Figure 12. From equation (9), the range of permissible hD values is limited largely by our specified range of BDs, with larger hDs resulting from lower values of BD. E-MORB solutions for hD are between 41.5 and 54 km, T-MORB solutions range from 42 to 49.5 km, and N-MORB range from 45 to 51.5 km. Figure 12 also shows that for similar BD values, the difference in dry solidus depth ΔhD, and therefore the temperature difference ΔT between N-MORB and E-MORB, is quite small. At any given BD value, the maximum ΔhD between N-MORB and E-MORB solutions is 9 km. To convert ΔhD to ΔT, we fit a line to the solidus-pressure function of Hirschmann  in the appropriate pressure range to derive the slope,
Our results suggest that the maximum potential temperature difference between N-MORB and E-MORB (ΔT) is ∼34°C. When similar Bw values are compared between N- and E-MORB regions, this difference reduces to as little as ΔhD = 3 km, or ΔT ∼ 11°C. T-MORB depths and temperatures are very similar to those for N-MORB. Thus our model indicates that the Galápagos “hot spot” increases the temperature of the mantle beneath the inflated portion of the GSC by only a few tens of degrees relative to our model N-MORB mantle, and this temperature difference is required in the E-MORB region alone; i.e., for the elements considered here, no mantle thermal anomaly is required west of 92.7°W. The thermal anomaly of 11–34°C predicted by our model is less than the ∼40°C anomaly predicted by Asimow and Langmuir , which reflects the incorporation of higher upwelling rates close to the Galapagos hot spot in our models. The range we predict is similar to the ∼30°C anomaly predicted by Canales et al.  on the basis of modeling of gravity anomalies in the region.
6. Most Likely Solutions for G′ Data
 In order to assess the full range of potential solutions to our G-PRIME data, we have allowed a generous range of variables. However, we may reasonably constrain our results further in an attempt to produce most-likely solutions for the different regions of the GSC. We further limit the solutions using the following arguments.
 1. Crustal thickness, axial morphology, and the N-MORB composition are typical of normal mid-ocean ridge basalts globally. We therefore infer that plume-driven mantle flow is negligible in generating N-MORB, eliminating all N-MORB solutions for Uw/U0 > 1. We allow E-MORB solutions only for Uw/U0 > 1, as required if the inference that E-MORB are the products of lower compared to N-MORB is true, and as argued by Canales et al.  on the basis of gravity modeling.
 2. Expecting a constant average productivity in the anhydrous region along the ridge, we narrow our BD range to between 0.36 and 0.42%/km [Asimow et al., 2001, Figure 4] for all MORB types.
 3. Although some enrichment in the incompatible elements, as well as in radiogenic isotopes, in the E-MORB and T-MORB regions may be necessary to explain the data, it is unlikely that the plume-affected region(s) are depleted in these elements relative to the N-MORB region. Thus we eliminated all T- or E-MORB solutions with source concentrations of Na that are less than N-MORB solutions remaining after steps 1–3.
 4. Finally, we assume that the additional zone of hydrous melting is at least as deep in the T- and E-MORB regions as in the N-MORB region. We eliminate any solutions where hw for the T-MORB or E-MORB source was less than hw for the N-MORB source.
 These additional constraints considerably narrow the range of solutions. They imply that GSC N-MORB were created by melting of a passively upwelling source with ∼35 ppm K, ∼130 ppm H2O, 2300 ppm Na2O, and 1050 ppm Ti. Mean degree of melting is ∼0.06, and the maximum degree of melting is ∼0.20. Hydrous melting spanned a depth interval (hw) of ∼30 km, and dry melting spanned a depth interval (hD) of ∼50 km. Model output uncertainties are typically <10% of the above solutions, but the uncertainties associated with the simplifying assumptions inherent to the methodology are likely to be larger.
 If T-MORB were created by any plume-driven mantle flow, it is less than Uw/U0 = 2. Solutions produced at higher Uw/U0 values are ruled out by restrictions (3) and (4). Of the elements considered in our model, T-MORB parental magma compositions vary from those for N-MORB only in the highly incompatible element, K. This variation can be explained by either slightly higher of a source with only a few ppm more H2O than the N-MORB source (when Uw/U0 = 1), or slightly lower of a source with a few ppm less H2O than for N-MORB (when Uw/U0 = 1.5 for T-MORB) (Figure 13). These results indicate that T-MORB can be created by melting a source very similar to the N-MORB source with only K enriched by ∼50%. Within the range of model uncertainty, values of source Na, Ti and H2O, , Fmax, hw, and hD are indistinguishable from those in the N-MORB region.
 The maximum excess mantle flow in the E-MORB region allowed by restrictions (3) and (4) is Uw/U0 = 3.5. This limits the minimum of E-MORB to 0.03, and requires that E-MORB have at least some K and H2O source enrichment relative to N-MORB and T-MORB. Our results indicate that relative to the N-MORB source, E-MORB represent the products of melting a source enriched in K by ∼150% (75 ppm), H2O by 50% (190 ppm), Na2O by <10% (2400 ppm), and Ti by ∼40% (1450 ppm). Compared to the N-MORB region, the depth interval of hydrous melting is ∼40% greater in the E-MORB region. Differences in anhydrous melting between the two regions are not different within resolution, but given the model uncertainties for hD, this result allows for only a slight temperature anomaly beneath the E-MORB region of <20°C.
7. Global and Local Implications
 Even the most “normal” mid-ocean ridges contain some amount of water [Michael and Chase, 1987; Dixon et al., 1988; Michael, 1995; Danyushevsky et al., 2000]. Our equations that take into account the “extra” region (no matter how small or large) of low-degree partial melts contributed by the presence of water have implications for the concept of “typical” degrees of melting at mid-ocean ridges, and for the composition of incompatible-element-depleted mantle from which MORB are generated. Normal mid-ocean ridges are commonly considered to be produced by mean extents of melting of ∼10% from a melting column on the order of 60 km deep, producing ∼6 km of crust [e.g., Klein et al., 1991; Langmuir et al., 1992; Forsyth, 1993].
 Recent works by several authors, including the present study, indicate that these values should be reevaluated. We estimate that GSC N-MORB were created by ∼ 6%. Forsyth  and Plank et al. , who defined “mean F” as the mean value of F(Fv) for melts pooled from the melting region (as opposed to FB, bulk melt fraction), concluded that average N-MORB are produced by ∼6.67% melting. Asimow et al.  predicted a mean melt fraction of no more than 8% for N-MORB, whereas Asimow and Langmuir , using new (pHMELTS) algorithms, predicted FB as low as 6.5% for normal regions of the Galápagos Spreading Center. There are two important and related results of the explicit consideration of the effects of water on the melting of MORB mantle. Incorporation of a region of hydrous melting reduces significantly the mean extent of melting required to produce normal MORB. A consequence of the reduction in mean extent of melting is that the composition of the mantle beneath even “normal” segments of mid-ocean ridges can be even more depleted in incompatible elements than was previously allowed by dry melting equations [Asimow and Langmuir, 2003; Asimow et al., 2004; this study].
 The explicit incorporation of plume-driven, enhanced mantle flow through the melting zone has important consequences for a number of melting parameters close to the Galápagos hot spot. For example, we require only about 50% enrichment in mantle H2O in the E-MORB region, compared to ∼250% increases in the models of Asimow and Langmuir  (Figure 13). The thermal anomaly suggested by our model (<20°C) is about half that predicted by Asimow and Langmuir . In general, incorporation of enhanced flow greatly reduces the magnitude of the thermal and compositional anomaly required to explain the Galápagos bathymetric and chemical anomaly. If enhanced upwelling is a direct consequence of compositional or thermal variations in the mantle, then models that neglect this important process will tend to over-estimate the magnitude of the anomaly in the mantle.
 A geodynamic model of mantle flow and melting that is consistent with a small temperature anomaly and plume-driven, excess mantle flow was presented by Ito et al. . This model simulated a mantle plume stem with a broad radius (200 km) and small excess temperature (≤50°C) beneath the Galápagos archipelago. The model predicted this “warm” plume material to flow north to the GSC melting zone and generate along-axis crustal thickness variations, consistent with the results of G′. Recent seismic studies including tomographic inversions of P and S body waves [Toomey et al., 2001] and receiver functions analyses [Hooft et al., 2003] suggest the presence of a mantle plume beneath the Galápagos archipelago, but a plume stem of higher excess temperature and smaller radius than that simulated by Ito et al. . To reconcile these results with ours, it is possible that the seismic tomography is imaging a narrow region of excess melting and melt retention rather than a region of narrow and high excess temperature. The receiver function study, which imaged a thinning of the mantle transition zone beneath the archipelago, however, would not be sensitive to melt in the upper mantle. Thus another possibility is that a narrow, high temperature anomaly at the center of the Galápagos plume is sheathed by a broader region of low excess temperature. If so, this “warm” material will likely have the most influence on the volume and composition of melts erupting along the GSC, which is consistent with our results. Further analyses of existing seismic data, as well as more complete seismic data coverage of the area is needed to test these possibilities. Also, the constraints on the nature of the mantle source and conditions of melting that we have placed in this study equip future geochemical studies to better address issues regarding the mechanisms of mass transport and chemical mixing of Galápagos plume material along the GSC.
 1. Samples collected from the GSC can be classified as E-, T-, or N-MORB on the basis of K/Ti ratios >0.15, 0.15 to 0.09, and <0.09, respectively. High K/Ti E-MORB also have higher H2O, Al2O3, and Na2O, and lower FeO*, SiO2, and CaO/Al2O3 relative to N-MORB at similar values of MgO. T- and E-MORB may be further subdivided into T1, T2, T3, E1, and E2 on the basis of CaO/Al2O3 ratios, SiO2 content, and subtle variations in incompatible elements.
 2. E-MORB dominate the GSC east of 92.6°W, where the crust is thickest (6.5–8 km). T-MORB are mainly found between 92.6°W and 95.5°W, where crustal thickness is 6–7 km. West of the propagating rift tip at 95.5°W, where crustal thickness is <6 km, N-MORB dominate. E-MORB incompatible element concentrations, including fractionation-corrected values for H2O, peak near 91.8°W and decrease with increasing distance from the hot spot. Fe8.0, Si8.0, and Ca8.0/Al8.0 all show their lowest values near 91.8°W.
 3. Geochemical boundaries correlate with geophysical and morphological characteristics. The transition from N-MORB to T-MORB occurs at the 95.5°W propagating rift tip, which also marks the boundary between axial rift-valley morphology and transitional morphology. Near 92.7°W, morphology changes from transitional to an axial high, the axial magma chamber seismic reflector shoals by >1 km, the thickness of seismic layer 2A diminishes by half, and lavas become dominated by E-MORB.
 4. Our new melting model considers the effects of an “additional” zone of hydrous melting that is created by the depression of the mantle solidus in the presence of water in the mantle. Variables in the equation include depth interval of the additional hydrous melting (hw), the fraction of melt liberated per unit of depth of decompression in the hydrous region (productivity, Bw), source concentration (C0) of incompatible elements, including H2O, and the flow rate of mantle passing through the hydrous region relative to the anhydrous region (Uw/U0). Incompatible element concentrations in pooled magmas are predicted to increase with height hw and Uw/U0. Of these variables, Uw/U0 has the strongest effect on , with higher Uw/U0 values corresponding to lower .
 5. We use this hydrous melting equation to model the variables that may combine to match the crustal thickness and average values of fractionation-corrected K2O, Na2O, H2O, and TiO2 in lavas measured along the GSC. We estimate that GSC N-MORB were created by ∼0.06 from a source with ∼35 ppm K, 130 ppm H2O, 2300 ppm Na2O, and 1050 ppm Ti. The absolute value of depends on the effect of a small amount of water on the position of the solidus, but is estimated to be ∼0.06. We estimate hw ∼ 30–40 km and hD ∼ 40–50 km.
 6. The extreme bounds that we consider in our model parameters suggest that the E-MORB region must be enriched in K unless the upwelling rate is >10. E-MORB may be as low as 0.02 if Uw/U0 = 10, or as high as 0.065 if Uw/U0 = 1. The E-MORB source may have as little as 110 ppm H2O or as much as 240 ppm H2O with a hydrous melting depth interval in the range of hw = 30–60 km (assuming hw/C0H2O of 0.25 km/ppm) and a depth interval of dry melting of hD = 45–60 km. We estimate a maximum temperature anomaly ΔT of ∼34°C in the E-MORB region compared to the N-MORB region. With the above conditions of hydrous melting we can explain the geochemical evidence for lower as well as geophysical observations suggesting greater magma production nearest the Galapagos hot spot. The direct effect of water on the major element composition of these magmas may be an additional factor contributing to decreasing SiO2, FeO, MgO, and CaO near the hot spot [Gaetani and Grove, 1998].
 7. Our preferred conditions required to explain the total variation in crustal thickness and glass compositions along the western GSC are only a slight temperature increase (<20°C), coupled with a moderately enriched mantle source and plume-driven, deep mantle flow of Uw/U0 = 1.5–3.5.
 8. Incorporation of enhanced deep mantle flow close to the hot spot further reduces the required magnitude of the thermal and compositional anomaly required to explain the gradients in chemical composition and crustal thickness along the GSC.
 We are grateful to the captain, crew, and shipboard scientific party of R/V Maurice Ewing Cruise EW0004 for their help with data acquisition at sea. The Ecuadorian government and the Parque Nacional Galápagos permitted work in their territorial waters. JoAnn Sinton put excessive care and extreme hours into sample preparation. Discussions with Mark Behn, Pablo Canales, Bob Detrick, Dave Graham, Charlie Langmuir, John Mahoney, and Brian Taylor were helpful in various aspects of this study. We are grateful to Charlie Langmuir and Peter Michael for thorough and detailed reviews and to Dave Graham and Bill White for additional comments and suggestions, which led to substantial improvement. Many figures in this paper were prepared using the GMT software package [Wessel and Smith, 1998]. This work was supported by NSF grants OCE98-18632 to the University of Hawai'i, OCE-0002189 to the UC Davis, and OCE99-0000478 to the University of Miami. This is SOEST contribution 6396.