The Southwest Indian Ridge is characterized by frequent outcrops of mantle rocks in a very slow spreading context. In situ measurements of trace element concentrations in pyroxenes of these rocks, and associated petrogenetic modeling, are reported. Overall, the measured compositions cover the whole range typically observed for abyssal peridotites. The greatest subkilometer-scale compositional variability is observed in the region east of the Melville fracture zone. The best explanation for the observed variability is given by concurrent melting and migration of melts strongly enriched in the most incompatible rare earth elements, such as those produced by a garnet-bearing source, or by refertilization with mixed garnet- and spinel-derived partially aggregated melts. Because the regionally associated basalts bear no “garnet signature” in their chemical compositions, we conclude that the residual mantle preserves the signature of a mantle source component that does not appear in the erupted magmas. Comparison between along-axis variations of basalt isotopic compositions and peridotite chemical compositions suggests that local isotopic enrichments displayed by some basalts can be associated with the “garnet signature” in the peridotite and that our sampling represents only a fraction of the global variability of the subaxial mantle. To the west of the Melville fracture zone, samples are more depleted and homogeneous at dredge scale. In addition to containing enriched components, petrologic modeling indicates that the peridotitic mantle beneath the entire section underwent (previous?) partial melting in the garnet stability field before melting at lower pressures.
In addition to compositional heterogeneities generated by partial melting and melt extraction processes beneath the present-day ridge, abyssal peridotites sampled at the SWIR are isotopically variable at all length scales down to the grain scale, even in the sections not influenced by hot spot activity [Snow et al., 1994; Lee, 1997; Salters and Dick, 2002; Warren et al., 2009]. Mantle heterogeneity along the SWIR is also reflected in the basalt compositions that are mixtures of chemically and isotopically enriched and depleted melts [Le Roex et al., 1992; Salters and Dick, 2002; Meyzen et al., 2003, 2005; Standish et al., 2008]. Isotopic compositions of SWIR peridotites not only encompass those of the spatially associated basalts [Snow et al., 1994], but more recently peridotites have been discovered that extend to more depleted compositions than the basalts, suggesting a pyroxenite component in the mantle source [Salters and Dick, 2002] or at least refertilization of the asthenospheric mantle [Meyzen et al., 2003, 2005; Warren et al., 2009]. This observation strengthens the idea that normal MORB mantle coexists with more refractory mantle resulting from older depletion events, in agreement with other abyssal peridotite isotopic compositions [Brandon et al., 2000; Cipriani et al., 2004; Harvey et al., 2006; Liu et al., 2008]. Although relatively rare, these examples demonstrate the ability of abyssal peridotites to record isotopically but also chemically enriched melts [Seyler et al., 2004; Warren et al., 2009; Brunelli and Seyler, 2010], providing insight into source heterogeneity. This conclusion is supported by recent numerical experiments which show that residual peridotites are powerful markers of diverse source contributions [Liang, 2008; Stracke and Bourdon, 2009]. In a different setting, Mazzucchelli et al.  observed and modeled examples of direct interaction between pyroxenite melts and peridotite in the Balmuccia peridotite massif (northern Italy). They demonstrate the efficiency of melt-peridotite interaction and melt transport to generate melts with E-MORB composition, such as those sampled along ultraslow spreading ridges.
In this paper we investigate local to regional trace element heterogeneities in a suite of abyssal spinel peridotites dredged in the eastern section of the SWIR, to gain some insight into the length scale of the melting heterogeneities and assess the respective effects of melting-induced and source-inherited heterogeneities. This region has the deepest axial valley of the oceanic ridge system and a very thick thermal lithosphere [Cannat et al., 1999]. In such an unusually cold thermal regime small volumes of deep-seated, poorly aggregated partial melts have reasonable chances to be retained in the partially molten mantle [Rabinowicz and Toplis, 2009]. Moreover, residual compositions have greater chance to be modified, as the contrast of compositions (e.g., the degree of disequilibrium) between melt and peridotite is greater, enhancing small-scale heterogeneities in the peridotites. We expect this to be the case when alkaline or garnet-equilibrated melt generated by melting of low solidus lithologies interacts with a depleted peridotite equilibrated in the spinel stability field.
The area between 52°E and 68°E was closely sampled along axis by dredging during the EDUL cruise [Mével et al., 1997]. Out of 69 dredges, 13 delivered mantle-derived lherzolites and harzburgites which provided the peridotite sample set of this study (Figure 1). Based on mean axial depth, the style of segmentation and ridge obliquity, this section of the SWIR has been subdivided into two major regions separated by the Melville fracture zone at ∼61°E. Between the Gallieni and Melville fracture zones (52°E–61°E), the mean axial depth is ∼4330 m [Cannat et al., 1999] and basalt geochemistry and seismic data suggest a crustal thickness of ∼4 km [Muller et al., 1997]. The ridge is slightly oblique spreading, but the obliquity is partially accommodated by transform faults with modest offsets (<50 km) and second-order bathymetric segmentation that has persisted through time since the onset of SWIR spreading [Phipps Morgan and Parmentier, 1995; Patriat et al., 1997; Sandwell and Smith, 1997; Baines et al., 2007]. EDUL Dr69 is located at ∼53°E between the Gallieni and Gauss fracture zones. The dredge recovered serpentinized harzburgites along with minor dunites, pyroxenites and gabbros and a few old basalts. These rocks form a small massif in the axial rift valley, which is the deepest of the western region (∼4500 m). The region between the Gauss and Atlantis II fracture zones comprises two short, near-amagmatic, oblique spreading segments [Baines et al., 2007]. Dredges Dr62 and Dr64 at ∼54°E–55°E and dredges Dr56 and Dr59 at 55°E–56°E yield serpentinized peridotites and gabbros with minor basalts; a few plagioclase-bearing peridotites occur in dredge Dr62. In contrast, between the Atlantis II and Melville fracture zones, a single dredge (dredge Dr42 at ∼59°7′E) has recovered serpentinized peridotites with basalt as the dominant lithology. Between the Gauss and Melville fracture zones bathymetric segmentation correlates with gravimetric segmentation defined by the Mantle Bouger Anomaly (MBA) [Cannat et al., 1999; Rommevaux-Jestin et al., 1997]. There is also a very good correlation between increasing regional axial depth and increasing Na8.0 in basaltic glasses, which overlaps the global Na8.0-depth MORB array of Klein and Langmuir  and Langmuir et al. , suggesting decreasing degree of melting toward the Melville fracture zone.
Between the Melville fracture zone and the Rodriguez Triple Junction (61°E–70°E), the SWIR axis is devoid of transform zones. This section is characterized by a very deep axial valley (mean axial depth ∼4730 m) and strong variability in axial depths and morphology. Based on basalt geochemistry, the mean crustal thickness (∼2.7 km) is thinner than for the region west of the Melville fracture zone [Cannat et al., 1999], but seismic data suggest highly variable crustal thickness (2–6 km) at segment scales with an average of ∼4 km [Muller et al., 1999]. Two changes in ridge trend at 63°67′E and 66°13′E separate a region with nearly orthogonal spreading (ESR of about 7 mm/yr) from two adjacent regions that have the deepest axial valley of the SWIR (∼4740 and 4800 m [Cannat et al., 2008]) and oblique spreading (ESR of about 6 mm/yr). In each region, along-axis, long melt-poor or avolcanic ridge sections separate prominent volcanic centers. Gravity-derived crustal thickness [Cannat et al., 2006], tied to seismic data [Minshull et al., 2006; Muller et al., 1999], reveal large expanses of seafloor characterized by a smooth and rounded topography, called “smooth seafloor,” separated by smaller areas characterized by bathymetric highs and negative MBAs, interpreted as large volcanic centers [Cannat et al., 2003, 2006]. Between the Melville fracture zone and 63°67′E, smooth seafloor constitutes about 50% of the mapped area and provides the main set of serpentinized mantle-derived peridotites dredged along axis (EDUL dredges Dr14, Dr21, Dr23, Dr24, Dr25). The collected samples are dominated by spinel peridotites but plagioclase-bearing peridotites represent ∼30% of the mantle derived samples in dredges Dr21 and Dr23 and the near totality of those of Dr14. Minor dunites, gabbros and basalts were also recovered. Between 63°67′ and 66°13′E, smooth seafloor forms about 28% of mapped area, from which no peridotite samples have been recovered. Although very deep, the region east of 66°13′E provided very few peridotites in two dredges. Dredge Dr3 (67°59′E) recovered a single large piece of harzburgite with no gabbro and abundant basalts, while dredge Dr6 (∼67°13′E) recovered six hand-sized pieces of lherzolite and five pieces of dunite, some of them being strongly mylonitized.
Basaltic glasses collected in the region east of the Melville fracture zone have unusual compositions [Meyzen et al., 2003]. They are highly enriched in Na2O but do not plot on the Na8.0-Fe8.0 and Na8.0-regional axial depth global MORB trends of Klein and Langmuir  and Langmuir et al.  and show poor correlation with gravimetry [Cannat et al., 1999]. These features suggest higher mean pressure of mantle melting and lower mantle temperature and/or more depleted mantle east of the Melville fracture zone. Major element compositions of the mantle-derived peridotites [Seyler et al., 2003, 2004] also show large heterogeneities of the axial lithospheric mantle at regional and local scales, east and west of the Melville fracture zone, and these heterogeneities are stronger in the easternmost region. Both the basalt and peridotite compositions suggest strong chemical interaction between upwelling upper mantle and percolating melts in a context of thick axial lithosphere and low melt productivity, particularly prevalent east of the Melville fracture zone [Meyzen et al., 2003; Seyler et al., 2003, 2004]. This is possibly related to a heterogeneous MORB mantle source, as shown by the isotopic compositions of basalts [Meyzen et al., 2005].
3. Sample Selection and Characteristics
Twenty-four samples were selected from the thirteen EDUL dredges which collected peridotites. They are moderately altered spinel perodotites, with 60–75% serpentinization, which preserved primary coarse granular textures showing little plastic deformation and dynamic recrystallization. Dredge Dr6 peridotites are exceptions in that they display porphyroclastic textures defined by centimeter-sized Opx and Cpx porphyroclats in a recrystallized olivine matrix. All samples are devoid of magmatic veins, lack plagioclase or its alteration products and contain low-Ti spinels (TiO2 < 0.2%). Major element compositions of the primary minerals are generally restricted in range from sample to sample in individual dredges, even in those characterized by large modal variations. Based on these features, Seyler et al.  concluded that the samples are texturally mantle residues fully equilibrated in the spinel peridotite field. For detailed microstructure study, modal data and major element compositions of the primary mineral phases the reader is referred to Seyler et al. . Additional modal analyses (Table 1) and determinations of major element compositions of the Cpx and Opx are reported in Tables S1 and S2 in the auxiliary material.
Table 1. Reconstructed Primary Modal Compositions of the EDUL Peridotites of This Studya
Five dredges brought up large amounts (300–1000 kg) of peridotites, allowing us to study the chemical variability at a subkilometer scale and possible correlations between major and trace element compositions. Among them, dredges Dr21 and Dr23 east of the Melville fracture zone contain spinel peridotites characterized by huge variations in modal composition from sample to sample; they range from lherzolites (5–10% modal Cpx) to olivine-rich harzburgites (2- < 5% modal Cpx) and virtually Cpx-free, olivine-poor harzburgites [Seyler et al., 2003, 2004]. We selected four samples from dredge Dr21 (Dr21-5-1, Dr21-5-2, Dr21-5-3, Dr21-7-2) and three samples from dredge Dr23 (Dr23-2-5, Dr23-2-8, Dr23-3-2) as representative of the whole modal range. In contrast, spinel peridotites from dredges Dr56, Dr62 and Dr64 west of the Melville fracture zone display homogeneous modal distributions falling at the lherzolite-harzburgite boundary (5 ± 1% modal Cpx). Three samples from dredge Dr56 (Dr56-3-5, Dr56-3-8, Dr56-3-11) and two other samples (Dr62-4-2 and Dr64-1-5) were selected to represent compositionally homogeneous dredges. Dredges Dr25 and Dr59 also contained large amounts of peridotite but samples are highly altered with most pyroxenes replaced by low-temperature amphibole and/or too strongly serpentinized to recognize the primary textures. These two dredges are thus represented by only one sample each, Dr25-3-1 and Dr59-2-1 which are free of amphibole and among the least serpentinized.
In spite of the paucity of available material, Seyler et al.  have distinguished two groups among Dr6 lherzolites, one containing Cr-poor pyroxenes and Cr-poor spinels, the other containing Cr-rich pyroxenes and Cr-rich (low-Ti) spinels. The three selected samples are representative of this bimodality. Dr6-1-3 represents the low-Cr group and Dr6-1-1 and Dr6-1-2 represent the high-Cr group. Samples Dr3-1, Dr14-1, Dr24-1 and Dr42-4-3 are from the other dredges which collected very few ultramafics.
Dredge Dr69 harburgites differ from typical abyssal peridotites. They are olivine-poor, Opx-rich harzburgites (27–35% modal Opx), containing ∼2% of small-sized (<0.5 mm), interstitial Cpx. Their mineral chemistry reflect unusually high whole rock Cr/Al and Mg/Fe, indicative of strong depletion by melt extraction [Seyler et al., 2003]. Sample Dr69-1-10 is a representative sample while sample Dr69-1-1 contains patches of Cpx a few mm in size scattered in the serpentinized matrix. These coarse Cpx do not differ from the interstitial Cpx grains in terms of major element composition. A clinopyroxenite sample (Dr69-8-2) has been included in this study. It consists of a hand-sized fragment containing millimeter-sized Cpx grains, with subordinate amounts of Opx, olivine and spinel. Its spatial relationships with the surrounding harzburgite are unknown, but a cumulate origin related to the gabbroic intrusions is suggested by the mineralogic continuity from spinel pyroxenite through plagioclase-bearing pyroxenite to olivine gabbro, and by major element compositions of their Cpx that are distinct from those of the residual peridotites.
In situ trace element analyses were measured by secondary ion microprobe spectrometry (SIMS) with the Cameca IMS5f at the university of Montpellier 2. 47Ti, 88Sr, 89Y, 90Zr, 137Ba, 139La, 140Ce, 143Nd, 147Sm, 151Eu, 160Gd, 163Dy, 167Er and 174Yb were measured in all sessions; 141Pr, 159Tb and 175Lu were measured in some sessions. Each spot was measured using 10 cycles with counting times of 2 s for Ti, Sr, Y and Zr and of 40 s for Ba and the rare earths. Concentrations were calibrated against the SRM NIST 610 glass using the values given by Rocholl et al. . Repeated analyses show a precision of 5% (1 sigma). 30Si were used as an internal standard to correct for differences between standard and samples, based on the EMP analyses. The effects of most molecular interferences are eliminated in following the energy filtering technique described by Shimizu et al. . More specifically, due to the low concentrations measured in pyroxenes of abyssal peridotites, eight REE oxides produce significant interferences 135BaO+ with 151Eu+, 143NdO+ with 159Tb+, 144NdO+ and 144SmO+ with 160Gd+, 147SmO+ with 163Dy+, 151EuO+ with Er+, 160GdO+ with 174Yb+, 159TbO+ with 175Lu+. The MO+/M+ ratios used for the corrections are 0.0047, 0.14, 0.27, 0.02, 0.08, 0.07, 0.33 and 0.28, respectively (with errors <10%). Corrections are <2% of the measured element/Si ratios for Eu, Gd, Dy and Er, and <5% for Yb. These corrections and treatment of other analytical results follow the procedures described by Zinner and Crozaz , Bottazzi et al. , Fahey  and Plá Cid et al. . Overall precision taking into account all the sources of uncertainty is estimated to be 10% for each oxide/element ratio.
A subset of pyroxenes was also analyzed on polished thick sections by inductively coupled plasma–mass spectrometry (ICP-MS) coupled with laser ablation (LA) at the university of Montpellier 2. The LA system is an automated platform Microlas (Geolas Q+) equipped with a 193 nm laser Excimer CompEx 102 from LambdaPhysik. Data were obtained in three sessions. The first series of measurements (samples Dr23-2-5 and Dr23-2-8) was performed using a 1991 vintage VG Plasmaquad II turbo ICP-MS with a sensitivity of ca.100 million cps/ppm measured on 115In. The other two series were analyzed on a ThermoFinnigan Element 2 High Resolution ICP-MS (HR-ICP-MS) using a single collector double-focusing field Element XR (eXtended Range) with an improved sensitivity of ca. 1 billion cps/ppm on 115In. Analyses were performed using an in-house modified ablation cell of 30 cm3 in a helium atmosphere [Günther and Heinrich, 1999]. A laser energy density of 15 Jcm−2 was used at a frequency of 5 Hz and spot sizes of 120 μm for Opx and 100 μm for Cpx. Signals were acquired in Time Resolved Acquisition mode, for 120 s for the blank and 60 s for measurement. The resulting sensitivity is approximately 4000 cps/ppm on 115In. Concentrations were calibrated against the SRM NIST 612 rhyolitic glass using the values given by Pearce et al. . 29Si was used as the internal standard and 43Ca was also measured. Data were reduced using the GLITTER software [Van Achterberg et al., 2001]. Precision of the analyses, assessed by repeated analyses of reference NIST SRM 612 and basalt BR, varies from 5 to 10%. Average compositions of the standards and 1 sigma error are given in Table S3.
Consistency between SIMS and LA-ICP-MS measurements was checked by analyzing several samples with both techniques. Elemental concentrations >2 ppm tend to deviate slightly from the 1:1 line, as standard deviations increase (Figure S1); in general, SIMS concentrations are slightly more scattered. Although the clearest and freshest domains of pyroxene grains were selected, submicron sized cracks and alteration veinlets following the crystallographic directions could not be avoided. In agreement with Dijkstra et al. , anomalously high counts of Pb, U, Ba, and to a lesser extent Sr, are attributed to microheterogeneities. Opx is particularly sensitive to alteration effects due to its extremely low trace element content and to the fact that it is more alterable than Cpx. Nevertheless, LA-ICP-MS and ion probe concentrations are within analytical errors for the REEs and differences are not significant in individual samples for all the elements which have been tested. Therefore, LA-ICP-MS and SIMS measurements for each sample have been averaged when their differences are within analytical errors (Tables 2 and 3). Complete trace element data of Opx and Cpx obtained with the two techniques are available in the auxiliary material (Tables S4–S7).
Table 2. Trace Element Concentrations in ppm for Clinopyroxenes of EDUL Peridotite Samples Analyzed in This Studya
Compositions average several pyroxene textural types when variations are within analytical errors. N, number of analyses; SD, standard deviation; bdl, below detection limit. Empty values mean not measured.
Exsolution lamellae in orthopyroxene.
Only one measure.
Rim that differs in composition from other grain rims.
Rare earth element (REE) and other significant incompatible trace element (Sr, Zr, Ti, Y) concentrations have been normalized to C1 chondrite of Anders and Grevesse  and plotted in Figure 2. East of the Melville fracture zone, Cpx grains are unzoned and show no significant differences in trace element contents from grain to grain at thin section and sample scales. All the samples but Dr6-1-2 have Cpx depleted in LREEs relative to the MREEs and HREEs (Figure 2). Based on the shapes of their REE patterns, the eastern samples have been divided into four groups. Group E1 (six samples out of fourteen: Dr21-5-1, Dr21-5-2, Dr21-5-3, Dr21-7-2, Dr23-3-2, Dr24-1) characterized by near-flat HREE contents at YbN = 8.9 ± 1.4 and moderate depletions in MREEs (SmN/YbN = 0.68 ± 0.15) and LREEs (CeN/YbN = 0.05 ± 0.03). These compositions are situated at the very fertile end of the spinel peridotite array sampled at mid-ocean ridges, and they fall in the compositional range characteristic of ultraslow spreading/cold ridges, such as the Vulcan fracture zone in the western SWIR [Johnson et al., 1990] and the Lena Trough in the Arctic Ocean [Hellebrand and Snow, 2003]. The slopes defined by the most incompatible REEs are regularly decreasing toward lighter REEs, except in Dr21-5-1 Cpx that shows a small, preferential enrichment in LREEs. Group E2 (four samples out of fourteen: Dr6-1-3, Dr14-2-1, Dr23-2-5, 23-2-8) is distinguished by slightly lower HREE contents (YbN = 7.6 ± 0.7) and flatter MREE to LREE slopes, being less depleted in LREE/HREE (CeN/YbN = 0.15 ± 0.06) for similar depletion in MREE/HREE (SmN/YbN = 0.67 ± 0.10). Group E3 samples (two samples out of fourteen: Dr6-1-1, Dr6-1-2) have REE patterns that deviate from patterns typical of melting residues. In both samples Cpx have low HREE contents (YbN ∼3–4). Cpx from Dr6-1-2 are characterized by convex-upward REE patterns with a maximum at NdN (∼ 9) and (CeN/YbN) ∼2. Dr6-1-1 is intermediate between Dr6-1-2 and Dr6-1-3. Group E4 (two samples out of fourteen: Dr3-1, Dr25-3-1) are the most depleted of the eastern sample set, with flat HREE contents starting at YbN = 5–6 and almost linear depletion in MREEs to LREEs (SmN/YbN ∼0.25; CeN/YbN = 0.04–0.07; CeN/SmN = 0.22–0.26).
West of the Melville fracture zone, the majority of the Cpx grains are also unzoned, with LREE-depleted compositions generally homogeneous within individual sample. However, a local variability is marked by preferential enrichments in La ± Ce that vary in amplitude within individual grains, a feature frequently reported in depleted Cpx from abyssal peridotites, and the occurrence of a few LREE-enriched compositions. LREE-depleted Cpx are more depleted than eastern samples and display a larger range of depletion, with YbN varying from 3 to 9 (Figure 2). REE patterns are typical of abyssal peridotites, with near-flat HREEs and strong depletion in LREE/HREE (CeN/YbN = 0.09 ± 0.07). Cpx compositions from dredge Dr42 harzburgite are the least depleted in REEs. Going westward, Cpx compositions of dredges Dr56 to Dr64 appear more depleted, comparable with the compositions of the Cpx from the Bullard fracture zone peridotites in the western SWIR [Johnson et al., 1990]. The most depleted Cpx of our sample set are from dredge Dr69; they are as depleted as the Cpx of the Bouvet fracture zone peridotites, which have been shown to represent the most refractory samples of the entire SWIR [Johnson et al., 1990; Brunelli et al., 2003]. In contrast, the Cpx analyzed in the pyroxenite Dr69-8-2 have typical MORB-like, LREE-depleted pattern, with (Ce/Yb)N ∼0.10 for YbN ∼3–4.
The enriched compositions have distinct preferential enrichments that start in the MREEs (Eu) and increase toward La, thus defining spoon-shaped REE patterns (Figure 2). These LREE-enriched compositions, observed in samples Dr56-3-8, Dr62-4-2, Dr64-1-5, Dr69-1-1 and Dr69-1-10, coexist with LREE-depleted compositions in the same thin section and even in the same grain. Spoon-shaped REE patterns may be observed on the rim but not in the associated core of some porphyroclasts. However, there is no true zoning because such Cpx grains are enriched asymmetrically, with spoon-shaped REE patterns characterizing only one side of the rim. In Dr69-1-1, spoon-shaped REE patterns characterize Cpx grains showing magmatic twins that grew replacing Opx [Seyler et al., 2003, Figure 4G], whereas in Dr62-4-2 this feature is present in Cpx forming a veinlet that cuts through a coarse Opx grain. Despite their LREE enrichments, the Cpx having spoon-shaped REE patterns have the same HREE contents as the LREE-depleted Cpx in the same samples, suggesting that the process responsible for LREE enrichment did not affect the less incompatible elements.
In all samples, both west and east of the Melville fracture zone, the concentration of the most incompatible elements (Sr, Nb, Zr) increases with increasing LREEs in Cpx. West of the Melville fracture zone, Sr contents of Cpx increase from group E1 through group E2 to group E3, whereas Sr contents in group E4 Cpx are as low as in group E1 (Figure 3a). The Cpx west of the Melville fracture zone, which have no or very low LREE re-enrichments, also show Sr contents (0.5–3.8 ppm) that increase with Ce contents, but the correlation is shifted to lower Ce/Sr for similar Sr contents (Figure 3a). Cpx from the pyroxenite Dr69-8-2 falls within Group E1 compositions. Although enriched in LREEs, Cpx with spoon-shaped REE patterns have Sr contents similar to those measured in the LREE-depleted Cpx in the same sample, and are thus characterized by higher Ce/Sr ratios and lower, generally negative, Sr/Sr* (where Sr* = (CeN.NdN)1/2); they plot within, or close to, the compositions of the eastern samples (Figure 3a). Dr42-4-3 Cpx show a weak preferential enrichment in LREEs, hence they are strongly depleted in Sr, with Sr/Sr* as low as 0.1, and plot at the most depleted end-member in Figure 3a. To discard a plagioclase effect on the lowest Sr contents we carried out a careful examination with optical and scanning electron microscopes, and found no traces of plagioclase or its alteration products. These strong negative Sr anomalies are observed in Cpx with no Eu anomaly and occur in plagioclase-free peridotites. Such strong negative Sr anomalies associated with anomalous LREE/HREE enrichments were also observed in some ODP Hole 1274 spinel harzburgites [Seyler et al., 2007] and in peridotites from Saint Paul's Rocks in the Atlantic Ocean [Brunelli and Seyler, 2010]. Parkinson et al.  and Rampone et al.  have suggested that the incompatibility of Sr relative to the REEs might change depending on the melting (or enrichment) conditions, resulting in apparent negative or positive Sr anomalies when plotted using diagrams based on standard orders of incompatibility.
5.2. Correlations Between Major and Trace Element Compositions in Clinopyroxenes
Na2O contents in Cpx are positively correlated with concentrations in incompatible elements, but two groups of Cpx show different behaviors. First of all, Cpx with spoon-shaped REE patterns are enriched in Ce but not in Na2O with respect to the LREE-depleted Cpx in the same samples (Figure 3b). Second, group E3 Cpx have low Yb contents (Figure 4) although they are strongly enriched in Na2O (with ∼1.2% Na2O, they are among the Na2O-richest Cpx known in abyssal peridotites). Group E4 Cpx and Cpx from dredge Dr69 also appear to be depleted in Yb and enriched in Na2O.
The major element characteristics of minerals, such as Cr# [ = Cr/(Cr + Al)] of spinel, have been found to correlate negatively with Yb content in Cpx of abyssal peridotites [Hellebrand et al., 2001]. A similar negative correlation between Cr# in Cpx, which has been shown to correlate with Cr# in spinel [Seyler et al., 2003], and Yb is observed in EDUL samples (Figure 5a). Samples east of the Melville fracture zone define a global trend of decreasing Yb contents and increasing Cr# from group E1 through group E2 to group E3 samples. Group 4 samples plot on the same trend. Cpx west of the Melville fracture zone follow a similar trend but shifted to slightly lower Yb contents, with the exceptions of Dr42-4-3 and Dr69-1-10 Cpx that plot well above this trend. Figure 5a shows that the most refractory sample (harzburgite Dr69-1-1) and the most fertile sample in terms of cpx mode and enrichments in highly incompatible elements (lherzolite Dr6-1-2) have similar Cpx Yb content and Cr#. Figure 5b shows that both Na2O and Cr# of Cpx increase from Group E1 through Group E2 to Group E3. Group E4 Cpx are also significantly enriched in Na2O with respect to their Cr#, and plot on the trend defined by the groups E2–E3. Similar behavior can also be observed within dredge Dr69. In Figure 5b, only three dredges contain peridotites that fall on the trend of residual abyssal peridotites in general; they are samples from dredges Dr56, Dr62 and Dr64. Na2O and Cr# compositions of the Cpx with spoon-shaped REE patterns are similar to the compositions of the LREE-depleted Cpx in the same samples.
5.3. Trace Element Compositions of Orthopyroxenes
Trace element concentrations measured in porphyroclast cores are constant, within analytical accuracy, on grain and sample scales for the HREEs (Dy to Lu), Y and Ti, but dispersed for the most incompatible elements (LREEs and Sr) particularly in the most depleted samples (Dr3-1 and Dr25-3-1, Dr56 and Dr69 peridotites), for which concentrations approach the detection limit. C1-normalized REE patterns are typical of Opx from abyssal peridotites, i.e., characterized by linear decreases from HREEs to LREEs (Figure 2).
East of the Melville fracture zone, Opx REE concentrations (average sample compositions) behave the same way as those of coexisting Cpx. HREE contents range between 1.7 and 2.4 in groups E1 and E2 Opx, but REE slopes are steeper in group E1 Opx [(Ce/Yb)N = 0.003–0.006)] than in group E2 Opx [(Ce/Yb)N = 0.009–0.017)]. Group E3 Opx have the lowest HREE contents (YbN ∼0.9) with relatively flat REE patterns [(Ce/Yb)N = 0.053–0.100)]. Group E4 Opx have lower YbN (1.3–1.4) than Group E1 and E2 Opx but the REE slope is steeper in Dr3 (not measurable in Dr25-3-1), consistent with the more depleted compositions of their Cpx. West of the Melville fracture zone Opx are in general more depleted than Opx from the eastern samples, with YbN ranging from 0.8 to 1.9. They also have steeper Yb-Lu slopes whereas MREE and LREE concentrations are below or close to detection limits.
5.4. Orthopyroxene-Clinopyroxene Trace Element Partitioning
REE partitioning between Opx and Cpx (Dopx/cpx) in mantle-derived peridotites have been shown to decrease with decreasing temperature [Witt-Eickschen and O'Neill, 2005; Lee et al., 2007]. As a consequence, trace elements are expected to be redistributed among the peridotite minerals as the residual mantle crosses the thermal boundary layer. Because the trivalent REEs are likely to replace Ca2+ in the pyroxene M2 site, the Ca-Mg exchange reaction during subsolidus cooling exerts a control over REE partitioning between Opx and Cpx [Witt-Eickschen and O'Neill, 2005; Lee et al., 2007]. Pyroxenes in abyssal peridotites show abundant exsolutions of the other pyroxene phase, indicating disequilibrium that may lead to incorrect estimates of the REE contents [Agranier and Lee, 2007]. EMPA CaO contents analyzed with a defocused beam (averages of several spots of 10 μm diameter) are slightly higher than those analyzed with LA-ICP-MS (single spot of 120 μm diameter) in the same grain (22.0 ± 0.8 wt% and 20.0 ± 1.7 wt%, respectively, for the 10 samples analyzed with both techniques). However, we have shown above that the trace element concentrations in the Cpx measured with SIMS and LA-ICP-MS are similar within analytical accuracy, although the two techniques analyze different volumes of material, thus integrating different amounts of Opx exsolutions. As shown by Hellebrand et al. , the REE concentrations in Cpx are thus high enough to not be significantly affected by the exsolution process. In Opx, EMPA CaO contents average 1.4 ± 0.3 wt% while LA-ICP-MS concentrations average 1.7 ± 0.3 wt%. Sample Dr23-2-8 is exceptional in that the CaO content of the grain analyzed by LA-ICP-MS (1.45 wt%) differs considerably from the CaO content of the grain analyzed by SIMS (1.05 wt%). As for Cpx, Opx trace element compositions measured with both techniques in the same sample do not differ significantly, although the SIMS analyses tend to be more scattered, with slightly lower REE concentrations. Average temperatures calculated with the Ca-in-Opx geothermometer of Brey and Köhler  indicate that the compositions measured in the porphyroclast cores represent subsolidus compositions. The calculated temperatures are 1118 ± 56°C (EMPA) and 1185 ± 55°C (LA-ICP-MS) at P = 1 GPa and 1145 ± 57°C (EMPA) and 1214 ± 56°C (LA-ICP-MS) at P = 1.5 GPa.
Apparent subsolidus Dopx/cpx were calculated by dividing REE concentrations in Opx by REE concentrations in Cpx (Copx/Ccpx). Following Witt-Eickschen and O'Neill  and Lee et al. , results are plotted as a function of effective cationic radii in Cpx, assuming all REEs are in eightfold coordination (Figure 6). The resulting trends vary from near-straight lines with negative slopes from Lu to La, to U-shaped curves, caused by progressive divergence in more incompatible REEs. Scattering also increases for MREE and LREE ratios in more depleted samples whereas Copx/Ccpx in less depleted samples show relatively smooth variations with progressively varying cationic radii. In the whole set, Copx/Ccpx (porphyroclasts) show 12–14% variations for Lu to Tm, 21–22% for Er to Dy, 27–32% for Tb and Gd, 41–46% for Pr to Eu, except a 100% variation for Nd, and 58–68% for Ce and La. In conclusion, all the LREEs are probably disturbed and will not be discussed further. When the MREEs and HREEs are considered, the lowest Copx/Ccpx ratio (Dr 56-3-5) corresponds to the lowest Opx CaO content, while the highest Copx/Ccpx ratios (Dr56-3-11 and Dr62-3-9) correspond to the highest Opx CaO contents. No difference is observed when comparing the different sample groups.
HREE and MREE distribution curves for EDUL samples plot close to the high-temperature end of Witt-Eickschen and O'Neill's  data set for xenoliths. This result is not surprising, since CaO content of Opx's in EDUL peridotites, and temperatures calculated with the same geothermometer, are within the range of those characterizing the high-T group studied by Witt-Eickschen and O'Neill : CaO = 1.49–1.76 wt%; T = 1150–1250°C. The EDUL distribution curves for the HREEs (Lu to Gd) are also similar to those calculated by Lee et al.  and Agranier and Lee  at high temperatures (1300°C–1400°C) based on the elastic strain theory and peridotite partial melting experiments. These estimated temperature suggest that Copx/Ccpx ratios for the MREEs and HREEs measured in EDUL peridotites are close to REE partitioning at the solidus-subsolidus transition in the thermal boundary layer. As shown by Agranier and Lee , Copx/Ccpx for the LREEs indicate that equilibrium on the mineral scale is not achieved for the most incompatible trace elements in abyssal peridotites with respect to peridotite xenoliths, even when only mineral cores are considered. The sample closest to equilibrium is also the most enriched sample, Dr6-1-2, in agreement with observations of Agranier and Lee .
6.1. Degree of Melting
All peridotites but two (Dr6-1-1; Dr6-1-2) have Cpx trace element compositions within the range of residual mantle peridotites, characterized by LREE-depleted patterns. The relatively good overall correlation of Yb versus Cr# in Cpx (Figure 5a) along with the strong correlation of Cpx and spinel Cr# shown by EDUL rocks [Seyler et al., 2003], allows the use of Cpx composition to estimate the degree of melting [Hellebrand et al., 2001]. In this way we have estimated the apparent degree of melting of each sample based on the HREE contents in Cpx, assuming a homogeneous depleted MORB mantle source (DMM), and a near-fractional melting process following the experimental geometry and melting parameters of Hellebrand et al. [2002a] and Brunelli et al. . Results (Table 4) show that, although strongly variable, the samples west and east of the Melville fracture zone define two different degrees of melt extraction: 15–23% total melting and 5–17% total melting, respectively. Estimated degrees of partial melting are broadly anticorrelated with the modal abundance of Cpx in the peridotites (Figure 7), supporting the assumption of a first-order correlation due to variable extent of melt extraction [Jaques and Green, 1980; Dick and Bullen, 1984; Johnson et al., 1990].
Table 4. Extents of Partial Melting Estimated From Heavy Rare Earths (Eu-Yb) Contents in Clinopyroxenes and Coexisting Spinel Compositionsa
Following the approach of Hellebrand et al. [2002a] the role of garnet during partial melting has been derived from fractionation of MREEs and HREEs in Cpx. The compositions of EDUL Cpx plot on two major trends in a (Sm/Yb)N versus YbN diagram (Figure 8). A main broad trend (trend 1) starts with 4–5% melting in the garnet stability field, continuing with variable degrees of melting in the spinel stability field. I is defined by the eastern groups E1 pro parte, E4 and the western samples. This near-fractional melting trend is within the range previously reported for residual abyssal peridotites [Hellebrand et al., 2002a; Hellebrand and Snow, 2003; Brunelli et al., 2006; Seyler et al., 2007; Brunelli and Seyler, 2010]. The second trend (trend 2) takes root in group E1 peridotites and shows an increase in (Sm/Yb)N at decreasing YbN, through group E2 to group E3 peridotites. This evolution is paralleled by the progressive enrichment in Na2O, LREEs, Sr (Figure 4) and other highly incompatible elements (Nb, Zr) and by progressive increase in Cr# (Figure 5). The three peridotite groups are randomly distributed in a restricted ridge region, between 62°E and 67°E, having no specific relation to exact locations. For example, groups E1 and E2 can be found in dredge Dr23, and groups E2 and E3 in dredge Dr6, that gives evidence for processes acting at short length scales (<1 km).
6.2. East of the Melville Fracture Zone
6.2.1. Group E1-E2-E3 Peridotites
Hump-shaped REE patterns close to that of Dr6-1-2 Cpx are scarce in abyssal peridotites. They have been observed in the Central Indian Ridge (sample CIRCE 93-7 [Hellebrand et al., 2002a]) and in the Arctic Ocean [Hellebrand and Snow, 2003]. In both occurrences, the peridotites are harzburgites very depleted in Cpx (0.1% modal cpx in CIRCE 93-7) and their peculiar REE patterns have been explained by refertilization with a depleted melt. Modeling by Hellebrand et al. [2002a] shows that entrapment and crystallization of a small amount of refractory melt in depleted harzburgites has the effect of significantly increasing Sm/Yb while keeping Yb very low and almost constant. With increasing refertilization, Sm/Yb starts to decrease whereas Yb increases. This effect produces a trend of inverse correlation between Sm/Yb and Yb, resembling EDUL trend 2, with the most refertilized samples having the highest Yb content and the lowest Sm/Yb ratio [Hellebrand et al., 2002a, Figure 9]. Such behavior results from the low modal abundance of Cpx relative to Opx, the latter being the major phase controlling bulk distribution coefficients [Hellebrand et al., 2002a]. Instead, EDUL peridotites are lherzolites (sample Dr6-1-2 contains ∼10% modal Cpx) and EDUL cpx with the highest Yb content and the lowest Sm/Yb ratio plot on the global trend of residual compositions (trend 1). Modeling shows that refertilization of lherzolites with a depleted melt has the effect of increasing both Yb and Sm/Yb [Hellebrand et al., 2002a], thus defining a trend opposite to that of EDUL trend 2. Finally, EDUL Dr6-1-2 has a (Sm/Yb)N twice that of CIRCE 93-7 for similar YbN, excluding refertilization by entrapment of depleted melt. A model of refertilization of residual lherzolites by entrapment of a LREE-enriched alkali melt could explain the positive correlation between the strong enrichments in LREEs and Na2O, but in this case Yb is also expected to increase [Hellebrand and Snow, 2003]. In addition, refertilization by melt retention will also have the effect to increase the Al/Cr ratio (due to the aluminous character of the aggregated melts relative to the residue) and the other incompatible element abundances in the peridotite. As a consequence, while it will reproduce the negative correlation between Cr# and HREEs (see Figure 5), the negative correlation between strongly and moderately incompatible elements and the positive correlation between Na2O and Cr# (Figures 4 and 8) cannot be reproduced. Therefore, compositional variations observed among the peridotites sampled along the 62°E–67°E ridge section must reflect small-scale heterogeneous premelting mantle and/or an effect of the melting process itself.
In order to clarify the exact nature of the enrichment processes we have calculated the compositions of the melts in equilibrium with Cpx and compared them to the compositions of the basalts recovered along the same EDUL section [Meyzen et al., 2003]. Calculated melts in equilibrium with group E1 Cpx (Figure 9a) are too depleted in LREEs to be in direct equilibrium with the regional basalts, showing that these peridotites are residual after melting. In contrast, melts in equilibrium with group E2 Cpx (Figure 9b) fall within the compositional range of the regional basalts. Whereas the HREE concentrations of group E2 Cpx can be explained by moderate degrees of melting in the spinel stability field (8–11%; Table 4), their LREEs are too enriched to be fitted by near-fractional melting models, not even accounting for a large fraction of melt being retained in the residue, up to 2% in a dynamic melting scenario. An additional 0.5–1.2% re-enrichment with a fully aggregated melt is then required to match the entire REE patterns. This simple melting and melt retention scenario may also explain the apparent equilibrium with the regional basalts, as well the high fertility in terms of major elements. However, this scenario cannot link group 2 to group 1 Cpx nor either to group 3 Cpx as discussed in the previous paragraph.
The convex-upward REE pattern observed in Dr6-1-2 Cpx is typical of Cpx in equilibrium with an alkaline magma [Irving and Frey, 1984; Downes et al., 2004] and has no equivalent among the known EDUL basaltic compositions (Figure 9c), not even among all other basalts sampled along the SWIR. The composition of the equilibrated melt could be that of a melt in equilibrium with a garnet-bearing residue (similar to melt generated by 1% melting of a garnet equilibrated DMM), suggesting that a garnet-bearing lithology was present in the partially molten mantle beneath the SWIR. Such melt may have never reached the surface, being trapped in the peridotite, or representing a minor component of the regional MORBs.
The equilibrium of the group E1 peridotites with depleted melts favors the interpretation of these rocks as being representative of mantle parcels that experienced “normal” melting, i.e., near-fractional melting in which melt extraction prevails. Therefore all compositional parameters such as the concentration of incompatible (REE), or compatible (Cr), elements are directly related to the total degree of melting as extensively discussed in the literature, even in the case of a small amount of melt retention in the residue. In fact the depleted nature of these melts cannot significantly modify the overall compositional fingerprint of the whole rock [Brunelli et al., 2006]. For this reason group E1 can be assumed to be the closest to pure residual compositions and hence an acceptable starting point to test possible partial melting models for the region east of the Melville fracture zone.
Four samples of group E1 peridotites were collected in the same dredge (Dr21); they show small compositional variations, with Cpx REE patterns varying in two different ways. First, Cpx of Dr21-5-1 and Dr21-5-3, on one hand, and Dr21-5-2 and Dr21-7-2, on the other, have near-parallel REE patterns, which cannot be attributed to different extents of melting (REEs fractionate during melting). The observed shift can be qualitatively explained by slightly different amounts of melt entrapment at the end of the partial melting event [Hellebrand et al., 2002a; Brunelli et al., 2006; Brunelli and Seyler, 2010]. Second, the Cpx REE patterns of the first pair of samples (Dr21-5-1 and Dr21-5-3) are slightly more depleted in HREEs and more enriched in LREEs than the second pair (Dr21-5-2 and Dr21-7-2), implying rotation of the REE patterns around the intermediate REEs (Sm-Gd). This rotation which links dredge Dr21 samples reproduces the rotation which links dredge Dr6 samples (Figure 2), indicating that a similar process acts in both sites but with different amplitude. We have shown above that late-stage refertilization by entrapment of partially to fully aggregated melt cannot be responsible for the observed rotation.
Melting of a lithological heterogeneity may severely influence the melting of the host mantle [Kogiso et al., 2004; Mazzucchelli et al., 2009; Liang, 2008]. However, to substantially affect the melting process it is necessary that the liquids produced by the melting heterogeneities are significantly different from those produced (or potentially produced) by the host mantle. For instance undersatured or oversatured silica melts may enhance pyroxene dissolution or force pyroxene crystallization. To address the origin of the E1-E2-E3 trend, a possible interaction to be considered is infiltration of garnet field–derived melt in a spinel field equilibrated mantle during melting. This scenario can be modeled by an open system melting approach where material influx can be accounted for.
Open system melting with melt influx has been modeled by several authors (see Shaw  and Liang  for reviews). We base our work on the formulation of Zou , who rewrote the equations proposed by Ozawa and Shimizu  and the set of partition coefficients and melting modes of Brunelli et al. . The critical point is to define the mode and element composition of the mantle, and of the influxing melt. The simplest case is to consider that in a portion of mantle undergoing “normal” near-fractional partial melting the process converts to open system melting due to influx of a compositionally different melt. Upward movement of an enriched melt can be accompanied by focusing, thus affecting the percolated mantle to a very variable extent and ultimately resulting in juxtaposition of normal residual mantle with partially to strongly percolated rocks at all scales. This behavior is well illustrated at the outcrop scale where mantle rocks are exposed [e.g., Mazzucchelli et al., 2009]. In our numerical experiment we have thus considered the starting point as the least HREE-depleted sample lying on a garnet-spinel melting path (sample Dr21-7-2). This sample can be generated by 5% melting in the garnet field followed by 5% melting in the spinel field and can be assumed as representative of the mantle prior to melt percolation. The best fit to the observed geochemical trends is then obtained by considering an enriched melt percolating through the rock in an open system setting with a finite residual porosity (0.5%) where 75% of the liquid filling the rock porosity is represented by the influxing melt that replaces the residual melt in equilibrium with the host rock. An additional 3% melting of the rock has been generated incrementally in this way. The behavior of the incompatible elements changes dramatically with respect to fractional melting with no melt influx because of the large influence of the incoming melt [Ozawa and Shimizu, 1995; Liang, 2008]. As reported in Figure 10a, a garnet field–derived partial melt induces a rotation around the position of Sm, resulting in the simultaneous decrease of the HREEs and increase of the LREEs (Figure 10b). The exact position of the rotation point is controlled by the composition of the infiltrating melt and to a lesser extent on the amount of melt infiltrating the rock. The LREE enrichment also depends on the geometry of the system, i.e., whether the addition of the incoming melt is continuous along the considered column or occurs as single pulses at the base of it [Liang, 2008]. This will result in a different modulation of the chromatographic efficiency of the melting column. Chromatography is in fact a component of any open system melting because of mass transfer through the cell sequence [Liang, 2008; Liang and Peng, 2010]. A complete discussion of this behavior is however beyond the scope of this paper and will be presented in a forthcoming contribution. The important thing to retain here is that rotation around a single REE may easily result from open system melting where the infiltrating melt is produced in the garnet stability field, as shown by Ozawa and Shimizu . Model results for a spinel field–derived partial melt influxing the melting peridotite are also reported in Figures 10c and 10d. It is noticeable that under the same conditions, the effect on the Sm/Yb ratio is much weaker, while HREEs are buffered at a constant value.
This open system melting model with influx of garnet-derived melt satisfactorily explains the continuity of compositions of Cpx from group E1 through group E2 to group E3. This also explains that, in the same dredge, compositions apparently requiring initiation of melting in the garnet stability field coexist with compositions that can be modeled in the spinel stability field alone. In this model, the trace element composition of Dr6-1-2 is fully equilibrated with the influxing melt, which has for effect to imprint an OIB-like chemical signature untypical of abyssal peridotite but similar to those found in many xenoliths and massif lherzolites. Continuous flux of melt in the rock during open system melting is accompanied by enhanced local melting, which significantly increase Cr# in pyroxenes and spinel. Because compatible elements are retained and incompatible elements are leached according to their incompatibility and relative enrichment with respect to the influxing melt and the rock content, the negative correlation between Cr# and HREE is preserved (or enhanced) (Figure 5a) and concomitantly a negative correlation between the least and the most incompatible elements is generated (Figures 4 and 8). Na2O (but not TiO2 [Seyler et al., 2003]) is also enriched because of the high diffusivity from the Na-rich melt and stability of jadeitic component at high pressure [Blundy et al., 1995; Lundstrom, 2000], leading to a positive correlation between Cr# and Na2O (Figure 5b).
Our modeling suggests that the rate of melt influx is high compared to the rate of partial melting. Thus, local melt production is low and possibly only induced by the chemical instability due to the percolation of a compositionally different melt. Low melting and a high external melt fraction suggest that there is not enough heat to sustain robust local (spinel field) partial melting. In the context of very slow/cold mantle upwelling, a thicker than normal conductive layer may develop lowering the upper extension of the melting region, inhibiting melt extraction and favoring melt stagnation. In the peridotites that started melting in the garnet stability field, this process has the paradoxical effect of erasing the residual garnet signature, by decreasing Yb but increasing MREE and LREE/HREE ratios. Therefore, calculated residual compositions plot along trends that crosscut polybaric fractional melting trends.
At a sample scale, refertilization by late-stage entrapment of enriched melts may nevertheless also account for part of the variability of the less incompatible REEs as well as modal variations among the eastern peridotite set, as shown by Hellebrand et al. [2002a] and Brunelli et al. . For example, Cpx REE patterns of Dr21-5-2 and Dr21-7-2 are parallel but the latter is shifted to higher overall REE contents; this sample likely followed a similar melting history to that of Dr21-7-2 before being refertilized by retaining a small amount of enriched melt at the end of the melting process. In a similar way, sample Dr24-1, the most fertile sample of this peridotite set, has a composition close to that of DMM that suggests low degree of melting (≤5%). At such a low depletion, the MREEs are not fractionated enough from the HREEs to assess the role of garnet in the melting process. However, we observe that the LREEs are quite depleted relative to the MREEs and HREEs (Figure 2). The overall shape of the Dr24-1 REE pattern is parallel to that of residual Cpx calculated after ∼6% fractional melting in the spinel stability field, but is enriched in all the REEs. Entrapment of the last partial melt fraction at the end of the melting history is compatible with increased REE contents and low LREE/MREE ratio, but requires quite a large amount of trapped melt (∼3%). This estimate is, however, consistent with the low pyroxene and spinel Cr#, silicate Mg# and high Cpx TiO2 of this sample relative to the other samples (Table S1) [Seyler et al., 2003].
6.2.2. Group E4 Peridotites
Group E4 Cpx (Dr3-1 and Dr25-3-1) are significantly more depleted in REEs than the Cpx of the other samples east of the Melville fracture zone, but plot on the same near-fractional (closed-system) melting trends as Dr21-5-2 and Dr21-7-2 (Figures 8 and 10). Their high LREE contents and the peculiar linear decrease from MREE to LREE cannot be modeled by either pure fractional nor open system melting. Cpx HREE contents can be reproduced by 5–6% near-fractional melting in the garnet stability field followed by 10–11% melting in the spinel field of a DMM source, with residual porosity ≤0.5% (Table 4). To match the observed LREE trend refertilization with a melt generated across the garnet/spinel transition is needed even though its amount is low (up to 0.2%; Figure 11). This value is comparable to those estimated for the Vema fracture zone by Brunelli et al. . This kind of refertilization is different from the open system melting described in the previous section. Here, the case is that of a normal melting/extraction process where local addition of a melt fraction generated in the garnet field occurs during the late stages of melting or soon after while the residual porosity is still open (even though very small: < 0.2%). Again, this result suggests that garnet-derived melts play a major role in defining the composition of the peridotites from this stretch of the SWIR.
6.3. West of the Melville Fracture Zone
6.3.1. Normal LREE-Depleted Peridotites
Western samples are more depleted than the eastern ones and clearly residual after melting. Still their LREE slopes are too flat with respect to the residual patterns of near-fractional (≤0.5% residual melt porosity) melting (Figure 12). The extents of melting calculated with the Cpx HREE contents in dredges Dr56, Dr62 and Dr64 (Table 4) fit reasonably well with the abundance of Cpx (Figure 7). The flat LREE slopes can be explained by addition of a melt enriched in REE before the end of peridotite melting [Hellebrand et al., 2002a; Brunelli et al., 2006; Godard et al., 2008]. This melt may be a partially aggregated melt formed at a deeper level in the melting region or melt derived from a source heterogeneity. The Cpx from dredge Dr56 define a very short trend of decreasing YbN and increasing SmN/YbN (Figure 8) that suggests a possible influx of garnet-derived melt during peridotite melting.
In contrast, the modeling of Cpx HREE contents in peridotites from dredges Dr42, Dr59 and Dr69 leads to inconsistent results. The Sm to Yb section of the Cpx REE pattern in Dr42-4-3 (Figure 12) can be fitted by 15% fractional melting (including 9% in the garnet stability field) leaving a lherzolite residue (4.6% Cpx) which contrasts with the low modal Cpx (<3%) content of this sample and that of the other harzburgites collected at this dredge haul. Increasing the degree of melting in the spinel stability field up to 19% leaves a residue whose modal composition fits the observed value (Figure 7), but the residual sample must have retained up to 1% of the last-degree melts to match the measured Sm-Yb concentrations. Similarly, Cpx modal proportions and relative MREE to HREE concentrations in samples from dredges Dr59 and Dr69 can be reproduced after 20% fractional melting (6% in the garnet stability field) for Dr59-2-2 and 22–23% fractional melting (4% in the garnet stability field) for Dr69-1-1 and Dr69-1-10. However, the calculated REE patterns are again shifted toward lower absolute REE concentrations, suggesting entrapment of aggregated melt in the upper part of the melting column.
6.3.2. Spoon-Shaped REE Patterns
A few Cpx are characterized by overenrichments in the most incompatible REEs superimposed on the LREE-depleted patterns and affecting the REEs more incompatible than Sm (Dr56), Eu (Dr64) or even Gd (Dr69) and Dy (Dr62). The resulting patterns have been described as spoon shaped. These spoon-shaped REE Cpx occur randomly and coexist in the same thin section with unzoned Cpx which show no or only weak inflections in La ± Ce. They are late-stage features, such as veinlets (Dr62) or part of the grain rims, thus reflect secondary processes operating at lower temperature in the lithospheric mantle. Na2O content and other major element composition are not modified. These REE enrichments are easily modeled as cryptic metasomatism caused by chromatographic fractionation during postmelting elemental diffusion at some distance from melt channels transporting a MORB-type magma (e.g., the regional basalts). Applying trace element chromatographic exchange models developed in literature, the strong enrichments observed in the spoon-shaped Cpx can be accounted for by considering long percolation times or short distances from melt transport channels [cf. Navon and Stolper, 1987; Bodinier et al., 1990; Takazawa et al., 1992; Batanova et al., 1998; Godard et al., 2000; Warren and Shimizu, 2010]. Examples of such enhanced small-scale heterogeneities in the SWIR subridge mantle have been modeled by Warren and Shimizu . Based on calculations of the timescales for diffusive equilibration, Warren and Shimizu  show that incomplete homogenization of the Cpx grains implies that metasomatism occurred relatively deep (≥10 km) in the lithosphere at 1000°C–1200°C. Our Cpx with spoon-shaped REE patterns display features similar to those studied by Warren and Shimizu  (sporadic occurrences in vein-free, otherwise LREE-depleted samples; no change in major element chemistry) but differ by a more restricted range of compositional variation. Indeed, REE patterns usually associated with modal plagioclase (enrichments in all REEs including the HREEs and negative anomaly in Eu) are lacking, thus indicating that the metasomatic event(s) occurred before plagioclase saturation in the infiltrating melts or deeper than the spinel-plagioclase transition. Cpx with spoon-shaped REE patterns were not observed in the peridotite east of the Melville fracture zone.
6.4. Multiscale Heterogeneities
Data of this study confirm the strong compositional heterogeneities of the residual lithospheric mantle that were predicted beneath ultraslow spreading/cold ridges and observed in other regions of the SWIR [Johnson et al., 1990; Johnson and Dick, 1992; Warren et al., 2009; Warren and Shimizu, 2010], at the Gakkel Ridge [Hellebrand et al., 2002b] and in the Equatorial Atlantic [Brunelli and Seyler, 2010]. It is worth noting the contrast of compositions with the thick lithospheric mantle sampled in the 15°20′N region of the slow spreading Mid-Atlantic Ridge, also characterized by large outcrops of serpentinized peridotites and thin volcanic crust [Kelemen et al., 2007]. There, percolating melts have stripped out the peridotite of its incompatible elements leaving homogeneous Cr-Mg rich harzburgites [Seyler et al., 2007; Godard et al., 2008]. In the SWIR, several stages of melt-peridotite interaction operating at small length scales mainly result in fertilization processes; heterogeneities are generated at different times and are superimposed on each other. We suspect that the study of a larger number of samples per dredge would have revealed a greater variability of the peridotite composition at this scale.
Within-dredge heterogeneities observed in the EDUL sample set also show striking differences in magnitude and style with the within-dredge variability which characterizes mantle residua underlying normal oceanic crust in regions of the Mid-Atlantic Ridge. For instance, at the Vema fracture zone (MAR 11°N), the compositions of the peridotites sampled in a single dredge define single spinel-projected, near-fractional, partial melting trends starting in a restricted depth interval in the (Sm/Yb)N versus YbN diagram; the observed within-dredge variability can be interpreted in terms of a relatively large degree of decompressional melting in the spinel stability field slightly modified by late-stage refertilization during melt extraction and entrapment [Brunelli et al., 2006]. On the contrary, EDUL samples collected in the same dredge show a stronger scatter in the estimated amount of garnet stability field melting. This would suggest these samples started melting at variable depths at a subkilometer scale, a conclusion inconsistent with mantle temperature distribution at depth and the hypothesis of a compositionally homogeneous source region. Our model points to a short melting column and melt/rock interaction occurring across, or close to, the garnet/spinel transition suggesting that melting may be confined between the thick/cold lithosphere and the spinel/garnet phase transition, possibly limiting the extent of melting itself [Asimow et al., 1995]. Enhanced reaction at magmatic temperature with deep melts in compositional disequilibrium with the peridotite matrix results in REE patterns characterized by gently dipping LREE slopes that contrast with the strong LREE/MREE and LREE/HREE depletions observed in “normal” residual abyssal peridotites, such as those studied by Johnson et al. . The abyssal peridotites dredged along the axis of the central part of the SWIR (9°E–16°E) display large compositional variability similar to the EDUL samples [Warren et al., 2009; Salters and Dick, 2002]. The central SWIR is characterized by long, oblique, amagmatic spreading segments [Dick et al., 2003], with obliquity and thus effective spreading rate even more pronounced than in the region east of the Melville fracture zone, indicating lower melt supply and thick lithospheric mantle [Cannat et al., 2008], in agreement with the basalt compositions [Le Roex et al., 1983, 1992; Standish et al., 2008]. In SmN/YbN versus YbN (Figure 8) and CeN/YbN versus YbN (not shown) diagrams, Cpx compositions of peridotites from two dredges, Van7-85 and Van7-96 [Warren et al., 2009; Salters and Dick, 2002] define within-dredge trends parallel to the EDUL trend, crossing across spinel-projected near-fractional partial melting trends that start at variable depth intervals and extending toward increasing enrichments in more incompatible REEs as Yb decreases. Warren et al.  have shown the extreme isotopic variability of the peridotites of the central SWIR. Similar isotopic variability may be predicted in the abyssal peridotites east of the Melville fracture zone, reflecting a heterogeneous mantle source.
Large compositional variations overprinted on the residual peridotites make it very difficult to assess the total degree of melting experienced by the peridotite during the last decompressional melting beneath the present-day ridge axis. The degree of melting estimated from crustal thickness is also subject to caution considering that magma can partly be generated by melting of pyroxenite or other heterogeneities. There is also evidence that melt produced over a large mantle region is delivered after focusing toward restricted areas of prominent volcanic centers, resulting in strong along-axis variability in the estimated crustal thickness [Cannat et al., 2008]. Indeed, melting indicators in EDUL basalts and peridotites are poorly correlated [Meyzen et al., 2003; Seyler et al., 2003] (Table 4), as predicted by numerical flux melting models [Liang, 2008]. As shown by Cannat et al. , lack of (or very thin) volcanic crust in the avolcanic oblique segments does not necessarily reflect very low melt supply from the underlying mantle, because the melts can propagate along axis toward adjacent volcanic segments [Cannat et al., 2008; Standish et al., 2008]. Nevertheless, this study indicates that peridotites west of the Melville fracture zone are systematically more depleted in incompatible trace elements than peridotites east of the Melville fracture zone, suggesting higher degrees of partial melting, in agreement with the wealth of data concerning this region of the SWIR. East of the Melville fracture zone, the highest fertility in terms of incompatible trace element contents, but also in terms of major element mineral compositions and Cpx mode [Seyler et al., 2003], characterize the peridotite sampled at ∼62°35′E (dredge Dr24); then the overall fertility decreases to both sides (toward dredge sites Dr3 and Dr25) to approach the depleted compositions west of the Melville fracture zone. However, this dome-like variation in the eastern region could be an artifact due to insufficient sampling. Alternatively, it could reflect a concentration of the heterogeneities in the source region around ∼62°E.
6.5. Source Composition
Nd-Sr isotopes in dredge Dr6 basalt suggest that a small proportion of an enriched component is present in its composition (144Nd/143Nd = 0.51302, 87Sr/86Sr = 0.70312 [Meyzen et al., 2005]), even though the trace element patterns lacks a clear “garnet” fingerprint [Meyzen et al., 2003]. Basalts sampled in dredge Dr11 may help unraveling the correlation between peridotites and basalts in dredge Dr6. The anomalously (relative to other EDUL basalts) isotopically enriched composition of Dr11 basalts (144Nd/143Nd = 0.51288, 87Sr/86Sr = 0.70362) is associated with a REE pattern (Figure 9c) showing a marked downward inflection toward the HREEs [Meyzen et al., 2003, 2005]. This suggests that a phase retaining the HREEs (assumed to be garnet) was present in the mantle source, and that the garnet-bearing source produced melts with an enriched isotopic signature. In this case, however, the ratio between enriched component and depleted “normal” ambient mantle source is greater than for the Dr6 basalt where only the isotopic evidence is measurable. If we assume similar mechanisms for Dr6 and Dr11 (also supported by the character of melts in equilibrium with the Dr6 peridotites) we conclude that an enriched, garnet-bearing component was present in the upper mantle beneath this entire region of the SWIR although its signal is variably diluted in the erupted MORBs. Additionally, this observation demonstrates that deep partial melts can locally reach shallower levels throughout the melting column without significant mixing with more depleted melts produced at lower pressure, and also shows the ability of residual peridotites to preserve traces of these partial melts. This is possible in “a highly channelized melt transport system,” which allows isolation of melts produced at different depths [Spiegelman and Kelemen, 2003]. Melting of enriched lithologies may itself promote the dissolution process ultimately leading to the development of a dunitic channel [Mazzucchelli et al., 2009]. Hence, the apparent proportion of dunites recovered in dredge Dr6 is higher than in the other dredges [Mével et al., 1997] and most of them contain a few insterstitial Cpx grains highly enriched in Na2O (M. Seyler, personal data, 2009). Finally, this observation is consistent with a short melting column (small degree of decompressional melting) and a thick thermal lithosphere, which allows the preservation of the chemical signature of an early melting stage in the residual peridotite.
Dredge Dr3 is located in the EDUL ridge region where basalts have isotopically enriched compositions. Nd-Sr isotopic composition of the Dr3 basalt also suggests the presence of a small proportion of an enriched component (144Nd/143Nd = 0.51307, 87Sr/86Sr = 0.70290 [Meyzen et al., 2005]). In contrast, Dredge Dr25 is located further to the west where the analyzed basalts are isotopically more depleted, extending from moderately enriched to DMM compositions. Dredge Dr25 delivered no fresh basalt but is located very close to dredge Dr26 on the same intersegment. Dr26 basalts have the most enriched compositions of this ridge section (144Nd/143Nd = 0.51306 and 87Sr/86Sr = 0.70287 [Meyzen et al., 2005]), similar to that of Dr3 basalt. Thus, similar LREE enrichments in Dr3-1 and Dr25-3-1 peridotite Cpx would suggest a link with the enriched component in the regional mantle.
The region between the Gauss and Melville fracture zones is represented by basalts with normal MORB isotopic signatures, except at ∼55°4′E where the dredge Dr57 basalt is isotopically similar to most eastern samples (144Nd/143Nd = 0.51307 and 87Sr/86Sr = 0.70291 [Meyzen et al., 2005]); these basalts form a volcano located on the same intersegment as the dredge Dr56. This suggests that the short trend of decreasing YbN and increasing SmN/YbN observed in the Cpx of Dr56 peridotites may also be connected with local isotopic enrichment, although this effect is much less marked in Dr56 peridotites than in the eastern peridotites.
Source heterogeneities can be represented by pyroxenite veins as proposed by Salters and Dick  or metasomatic enrichments of a previously depleted peridotite [Warren et al., 2009; Meyzen et al., 2005]. Adiabatic decompression of asthenospheric peridotite mantle previously contaminated by low-degree melts enriched in H2O, alkalis and other incompatible elements may have triggered an early stage of partial melting in the garnet stability field, which could explain the garnet signature both in the residues and in the metasomatic melts. The discovery in dredge Dr23 of olivine-poor harzburgites enclosing in their Opx metasomatic minerals enriched in volatiles and incompatible elements supports the hypothesis of interaction between peridotite and low-degree melts prior to the generation of the regional MORBs [Seyler et al., 2004].
The easternmost SWIR region located east of the Melville fracture zone is characterized by relatively fertile mantle residues highly heterogeneous from the dredge scale down to the sample scale. Melting modeling suggests that this variability results from spinel field open system melting in presence of garnet-derived melts flowing through the peridotite. Retention of variable amounts of mixed (garnet- and spinel-derived) partially aggregated melts at the top of the melting column results in the observed short length-scale variability. Peridotite REE compositions attest for melt derived from a garnet-bearing lithology to be generated beneath the present-day ridge, the garnet fingerprint appearing as a diluted component in the regional basalts.
West of the Melville fracture zone, the mantle appears more homogeneous at a local scale while effects of melt influx during melting are less important. We interpret this as resulting from a greater degree of melting of the peridotite in the spinel stability field. Enhanced melt production and extraction reduces the possible interaction with percolating melts thus better approaching the fractional end-member melting model.
The observed heterogeneity differs in the two regions: high-frequency, strong heterogeneities in the region east of the Melville fracture zone deriving from strong melt-rock interaction in an open system melting with large influence of the incoming melt. In the western region the mantle residues are homogeneous at a larger, regional, scale while the melting style is more akin to the fractional melting with lower melt influx. These observations agree well with the different thermal setting estimated from bathymetry and geophysics [Cannat et al., 2008] that reveal the eastern region to be characterized by oblique spreading and colder thermal setting.
In both regions east and west of the Melville fracture zone, the peridotite compositions require initiation of melting in the garnet stability field. However, the garnet signature in residual clinopyroxenes was partially masked during the open melting process due to the progressive rotation of the REE patterns.
We are grateful to Bernard Boyer for his assistance with the SIMS analyses and to Olivier Bruguier for technical assistance on the LA-ICP-MS. This paper benefited from constructive reviews by Elisabetta Rampone and an anonymous reviewer. Funding for this research was provided by Centre National de la Recherche Scientifique–Institut National des Sciences de l'Univers programs Dynamique et Evolution de la Terre Interne and Planétologie–Sciences de la Terre. DB has partially been funded by Fondazione Cassa di Risparmio di Modena through CARBRIDGE project. This is IPGP contribution 3196.