Sodium-chromium covariation in residual clinopyroxenes from abyssal peridotites sampled in the 43 – 46 ° E region of the Southwest Indian Ridge

Mantle-derived peridotites sampled at three dredge sites between the Discovery and Indomed fracture zones on the Southwest Indian Ridge axis are analyzed for petrography and major and trace element mineral compositions. While textures and microstructures are those typical of normal residual peridotites these rocks display a large compositional variation encompassing the whole spectrum of abyssal peridotites even at the scale of a single dredge site (≤ 1km). Particularly, clinopyroxenes in peridotites dredged at 44.03° E show a huge variation in sodium contents positively correlated with chromium concentrations. Observed Na-Cr enrichments exceed the commonly reported contents of the spinel abyssal peridotites. Similar values are only found in very few peridotite samples collected at ultra-slow spreading ridges. Major ACCEPTED MANUSCRIPT


A C C E P T E D M A N U S C R I P T
1

Introduction
Abyssal peridotites and Mid-Ocean Ridge Basalts (MORB) are considered as complementary, residual and magmatic, products resulting from adiabatic partial melting of the asthenospheric mantle beneath mid-ocean ridges. The increasing number of observations of natural samples, experiments and numerical modeling provide growing evidences for a lithologically non-uniform MORB mantle source (e.g., Hanson, 1977;Hirschmann and Stolper, 1996;Wood, 1979). Even those ridge sections not influenced by hot spot activity likely contain small amounts of mafic components such as eclogite or pyroxenite (e.g., Hirschmann and Stolper, 1996;Lambart et al., 2012Lambart et al., , 2013 A C C E P T E D M A N U S C R I P T 3 Pertermann and Hirschmann, 2003). Isotopic compositional variability of abyssal peridotites does not systematically overlap with that of the associated basalts and extends to more depleted compositions in several localities (Cipriani et al., 2004;Harvey et al., 2006;Lassiter et al., 2014;Liu et al., 2008;Mallick et al., 2014Mallick et al., , 2015Salters and Dick, 2002;Stracke et al., 2011;Warren et al., 2009;Warren and Shirey, 2012). The idea of a low-solidus component with enriched isotopic signature in the MORB mantle source has been invoked to explain the lack of the most enriched isotopic compositions in residual peridotites with respect to associated MORBs (Salters and Dick, 2002;Liu et al., 2008;Mallick et al., 2014Mallick et al., , 2015. As shown by petrological and numerical experiments, melting of a low-solidus component may considerably modify the melting process of the surrounding peridotite (Morgan, 2001;Katz and Weatherley, 2012;Lambart et al., 2016;Spiegelman and Kelemen, 2003;Weatherley and Katz, 2012). In addition, the small scale (< 1 km) modal and compositional variations created in the residual peridotite during melt extraction, segregation and transport processes, are significantly enhanced when melt-peridotite interaction is combined with a lithologically or mineralogically heterogeneous source (Brunelli et al., 2014;Lambart et al., 2012;Liang and Parmentier, 2010;Liang and Peng, 2010;Stracke and Bourdon, 2009). As a rule of thumb, residual compositions are more efficiently modified when greater is the compositional contrast (e.g., the degree of disequilibrium) between transient melts and partially molten peridotite. The largest variations are observed when alkaline or garnet equilibrated melts, possibly generated by melting of a low-solidus lithology, interact with a depleted peridotite equilibrated in the spinel stability field (Brunelli et al., 2014). Examples from a few geographic areas have demonstrated the ability of abyssal peridotites to record such chemically enriched melts, providing insight into source compositional heterogeneity (Brunelli and Seyler, 2010;Cipriani et al., 2009;D'Errico et al, 2016; A C C E P T E D M A N U S C R I P T 4 Hellebrand et al., 2002b;Mallick et al., 2015;Seyler et al., 2004Seyler et al., , 2011Warren et al., 2009).
In this study we present petrographic and chemical data of abyssal peridotites dredged in a section of the Southwest Indian Ridge (SWIR) whose basaltic crust is characterized by large isotopic and chemical variations. The peridotites sampled at the dredge 26 site during the SWIFT cruise (SWF-26) contain clinopyroxene showing enrichments both in sodium and chromium, up to very elevated concentrations, correlated with an increase in the apparent degree of melting. A review of global compositional variations in abyssal peridotites  reveals that similar depletion/enrichment trend is not unique but very rare, and found no satisfactory explanation. In an attempt to understand this enigmatic trend, we have characterized the major substitutions governing the evolution of the pyroxene compositions and modelled the Na and Cr compositions by combining partial melting and percolation of a Na-rich melt. Our results suggest possible relationships between the Na-Cr covariation and enriched, high-pressure melts, derived from a heterogeneous mantle source.

Geological setting
The SWIR is a major plate boundary separating Africa and Antarctica with an almost constant ultra-slow spreading rate of about 16 mm/yr (DeMets et al., 1990) from the Andrew Bain Fracture Zone (FZ) (32°E) to the Rodrigues triple junction (70°E). The basalts are normal and enriched MORB (N-and E-MORB), characterized by a large isotopic compositional diversity (Bezos et al., 2005;Chauvel and Blichert-Toft, 2001;Gautheron et al., 2015;Hamelin and Allègre, 1985;LeRoex et al., 1989;Mahoney et al., A C C E P T E D M A N U S C R I P T 5 1989; Meyzen et al., 2003Meyzen et al., , 2005, dominated by the DUPAL anomaly, whereas the peridotites dredged along-axis appear highly heterogeneous in modal, chemical and isotopic composition at all length-scales (Mallick et al., 2014(Mallick et al., , 2015Seyler et al., 2003Seyler et al., , 2004Seyler et al., , 2011Warren et al. 2009;Warren and Shimizu, 2010).
In the continuity of the systematic on-axis sampling of the 49-70°E region performed during the MD107 EDUL cruise (Mével, 1997) (Aslanian et al., 2002). The 43-46°E region is centered on a large positive residual geoid anomaly extending from 32°E to 55°E beneath the SWIR (LeRoex et al., 1989), at midway between the influence of the Marion (~38°E) and Crozet (~52°E) Islands, two hot spots presently located at ~250 km (Marion) and ~1000 km (Crozet) south of the ridge axis (Müller et al., 1993). No peridotite but few basalt compositions from four dredge sites ( Fig. 1), including a glass retrieved with the SWF-26 peridotites, have been published for the region of interest. Their compositions vary from N-MORB in the 43°-44° E segment to E-MORB and alkali-basalt close to the Indomed FZ (Hamelin and Allègre, 1985;LeRoex et al., 1989;Mahoney et al., 1989). In the 32°-47° E region isotopic compositions are typical of the regional Indian ocean mantle (Chauvel and Blichert-Toft, 2001;Hamelin and Allègre, 1985;Mahoney et al., 1989), which might locally contain fragments of lower continental crust (Gautheron et al., 2015).

Sample selection and petrography
Dredges SWF-26, SWF-27 and SC-08 each recovered about 150-160 kg of strongly serpentinized and weathered peridotites occurring as rounded blocks of ~15-30 cm in diameter. Rare dunites and ~40 kg of altered basalt were also recovered in dredge SWF-26, while dredge SC-08 included olivine-plagioclase-phyric basalts. Glassy fragments were present in all the three dredges but gabbroic and pyroxenitic materials were lacking. Visual inspection on hand samples shows that dredges SWF-26 and SC-08 are dominated by spinel-lherzolites while dredge SWF-27 contains spinel-harzburgites.

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T 7 clinopyroxene) aggregates (Fig. 3). In the lherzolites SWF-26-2-9, SWF-26-2-11 and the harzburgite SC-08-03, the orthopyroxene assemblages are not elongated and of the same size as olivine, coexisting with smaller-sized (<1 mm) clinopyroxene. In contrast, the lherzolite SWF-26-2-5 contains smaller orthopyroxene crystals but abundant , up to 15 mm, elongated aggregates of clinopyroxenes defining a foliation (Fig. 3A). In all samples pyroxene aggregates are characterized by jagged, strongly irregular outlines with deep olivine-filled embayments (  Fig. S2F). Similar propagation textures, but involving clinopyroxene and plagioclase, also occur in oceanic gabbros (e.g., Agar and Loyd, 1997;Dick et al., 2002); they have been reproduced experimentally and explained as the result of a partial melting reaction triggered by water-rich fluids, that starts on the primocryst boundaries and proceeds by dissolution/precipitation of the primocrysts (Koepke et al., 2005). In abyssal peridotites, orthopyroxene -olivine textural relationships are generally interpreted as resulting from pyroxenes incongruent melting in the upwelling melting mantle (Ceuleneer et al.,1988;Nicolas,1986). Similar textures also develop where melts, generated at deeper levels, migrate upward and react with residual peridotites (Daines and Kohlstedt, 1994;Kelemen et al., 1992;Seyler et al., 2007), leading to pyroxene dissolution and olivine precipitation (Kelemen et al., 1992). Based on this interpretation and on Koepke et al. (2005)' work, we suggest that extreme resorption and propagation textures in clinopyroxene and spinel of abyssal peridotites also form by partial melting and/or melt-rock reactions during the final stage of melting.
Alternatively, thin clinopyroxene and spinel material, interstitial or filling cracks, may represent frozen melts injected into fine-scale porosity structure, as observed in some primitive cumulates (Natland and Dick, 2001) or in a partially molten rock still in deformation (Seyler et al., 2001 and references therein). It is noteworthy that intergranular extensions of coarse clinopyroxenes are often interconnected, with preferential crystallization of clinopyroxene + spinel intergrowths at the junctions, suggesting interconnection of a former intergranular melt over cm-scale distances . Plastic deformation and dynamic recrystallization are restricted in intensity and space, affecting preferentially the borders of the pyroxene crystals or aggregates. These grains appear twisted or broken (Fig. 3D); the rare recrystallized grains are anhedral, with occasional 120° triple junctions and lack of exsolutions. Spinel intergrowths with pyroxene neoblasts are locally observed where spinel arrays merge with the recrystallized borders.
In conclusion the studied peridotites display textures similar to those considered as residual upper mantle following melt extraction (e.g. Ceuleneer and Cannat, 1997;Dick et al., 2010;Seyler et al., 2007). Crystal deformation is limited probably because intergranular melt reacted with the peridotite during equilibration at high temperature in the asthenosphere -lithosphere boundary layer (Dick et al., 2010). Clinopyroxene and spinel are commonly associated in similar microstructures and appear to be the last

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T 9 phases to have been mobilized before freezing. There is no evidence for modal metasomatism, i.e., the orthopyroxene phase contains no mineral inclusions such as those described in some peridotites (Luguet et al., 2003;Seyler et al., 2004), magmatic sulfides are lacking and late-stage crystallization from trapped melts is modest compared to some abyssal peridotites (Seyler et al., 2001(Seyler et al., , 2007Suhr et al., 2008).

Analytical methods
Olivine, pyroxenes and spinel major elements were analyzed using the automated CAMECA-CAMEBAX electron microprobe of the CAMPARIS micro-analysis center  Table 3. Complete trace element data of orthopyroxenes and clinopyroxenes are available in Supplementary Data Table S1.

Major element compositions
Olivine Al2O3 from sample to sample, as expected in a set of residual peridotites that

A C C E P T E D M A N U S C R I P T
experienced variable degree of melt depletion. The clinopyroxene-rich lherzolite SWF-26-2-5 has orthopyroxenes with high Al2O3 contents (5.4 ± 0.2 wt.%), while Al2O3 in its clinopyroxenes range from ~7.5 to ~4.0 wt.%, averaging 6.7 ± 0.5 wt.% in the core of the coarser (≥ 0.5 mm) grains. In this sample, both pyroxenes have low Cr2O3 contents Averaged Cr# in coexisting orthopyroxene, clinopyroxene and spinel are well correlated in the studied sample set, with KD within the usual range of abyssal peridotites (Dick and Bullen, 1984). Mineral Cr# is a function of whole rock Cr#, which reflects the degree of depletion in magmatic components and thus widely used as a proxy for the extent of partial melting in residual mantle peridotites (Hellebrand et al., 2001). Fig. 7 shows that spinel Cr# in the studied samples increases with increasing modal olivine and decreasing modal clinopyroxene as expected for residues of increasing degrees of melt depletion by partial melting. It also appears that the correlation between the modes and the spinel Cr# is not linear, the latter increases more significantly as the rocks become more refractory. Using the relation between spinel Cr# and extent of partial melting established by Hellebrand et al. (2001), SWF-26 peridotites appear to have undergone varying amounts of melt extraction, corresponding to ~3% for the clinopyroxene-rich lherzolite, ~7% in the lherzolites SWF-26-2-9 and SWF-26-2-11 and ~16% in the harzburgite SWF-26-2-7. Samples SWF-27-1-12 and SC-08-03 would have experienced ~12% and ~9% melting, respectively. These results do not take in account possible degrees of melting in the garnet stability field (Hellebrand et al., 2001).

Compositional variations of clinopyroxenes of dredge SWF-26 peridotites
Clinopyroxenes of the four SWF-26 samples define a positive trend in the Cr2O3 -Na2O diagram (Fig. 5B), with the clinopyroxene-rich lherzolite SWF-26-2-5 at the lowest Na2O contents and the harzburgite SWF-26-2-7 at the highest values. Such a feature is inconsistent with the opposite Cr and Na behaviour resulting from magmatic processes.
In addition, the high enrichments both in Cr2O3 (1.3-1.6 wt%) and Na2O (1.5-2.1 wt%) observed in SWF-26-2-7 clinopyroxenes are clearly in excess with respect to the concentrations commonly measured in clinopyroxenes of abyssal peridotites. Similar unusual clinopyroxene compositions have only been recognized in a few dredges from ultra-slow spreading ridges in the SWIR (Seyler et al., , 2011 and Arctic ocean (D'Errico et al., 2016;Hellebrand et al., 2005;Hellebrand and Snow, 2003;Lassiter et al., 2014).  8F). The curvature of the trends reflects the higher incompatibility of Na relative to Al during partial melting. In contrast samples from the 'NaCr' dredges show a negative Na-Al trend and no correlation between Na and Al VI . In all sample sets, both Al IV and Al VI decrease with decreasing Al, but in the 'NaCr' dredges, Al IV decreases faster (Fig. 8G) and Al VI decreases slower (Fig. 8H). A consequence is that the Al VI /Al IV ratio increases with decreasing Al in the 'NaCr' dredges, whereas it slightly decreases in the global set ( Fig.   8I).  Mollo et al. (2013).  (Table 4). In all cases, clinopyroxene evolution from SWF-26-2-5 to SWF-2-2-7 involves a strong increase in the sodic components and decrease in the Tschermak's molecules, whereas the sum of the Ca-Mg-Fe components remains nearly stable (or decreases very slightly). The good

A C C E P T E D M A N U S C R I P T
16 negative correlation between Na + Cr and Ca +Mg (Fig. 8K) may indicate that the coupled substitution, Na + Cr = Ca + Mg (i.e., Ko = Di) was active in the clinopyroxenes of the 'NaCr' dredges (Ikehata and Arai, 2004). However, calculation (1)  . It thus appears that in 'NaCr' dredges, the more incompatible elements (e.g., Ce, Sm) are decoupled from the less incompatible elements (e.g., Yb) and their ratios tend to increase with increasing Cr#.

Evidence for syn-melting metasomatism
Besides the effects of incomplete sub-solidus re-equilibration, modal and chemical variations of the sub-ridge lithospheric mantle on a scale < 1 km are attributed to two major late-stage and post-melting processes, namely the entrapment of unextracted melt and reactive porous flow (Brunelli et al., 2006;Elthon, 1992;Hellebrand et al., 2002a;Johnson and Dick, 1992;Warren and Shimizu, 2010). In most cases, these processes are identified by the decoupling between the major elements indicators of melting, such as Cr# in spinel and pyroxenes, and the highly incompatible trace elements, such as LREE, Sr and Zr. In the same way, Na2O contents measured in clinopyroxenes are generally higher than expected for a residual phase, and considered as a marker of a refertilization event (Elthon, 1992 Adding 1-2% of enriched MORB or alkali-basalt (2-3 wt.% Na2O) will also reproduce the Na2O increase. Because melt is strongly enriched in Al2O3 and depleted in Cr2O3 with respect to the residual peridotite, Cr# should decrease in the re-equilibrated peridotite, although calculation shows that Cr# reduction is negligible for melt addition < 2% vol.
Cr# in minerals is proportional to the bulk Cr#, therefore clinopyroxene Cr# will decrease or remain nearly constant while Na2O concentration increases. Entrapment model thus cannot explain the large and concomitant variations in Na2O and Cr# observed in the clinopyroxenes of the 'NaCr' peridotites, although it may work for moderate Na2O enrichment with no significant modification in the major elements indicators of melting.
Reactive porous flow is another process that leads to selective enrichments in incompatible elements (Vernières et al., 1997), triggered by melt infiltration from a

A C C E P T E D M A N U S C R I P T
20 nearby melt channel (Kelemen et al., 1992;Lundstrom, 2000Lundstrom, , 2003 or upward advective transport of intergranular partial melts (Godard et al., 1995;Brunelli et al., 2006). In a melting column, incompatible trace elements can be fractionated in the same way as for the chromatographic columns: during upward melt migration incompatible elements are modulated according to their partition coefficients (Liang and Parmentier, 2010), leading to preferential enrichments in LREE relative to HREE or spoon-shaped REE patterns in bulk rock and clinopyroxene (Hellebrand et al., 2002a;Warren and Shimizu, 2010).

A C C E P T E D M A N U S C R I P T
21 clinopyroxenes increase with increasing extent of melting, we conclude that refertilization and partial melting (or at least part of the partial melting) were two linked processes that have occurred simultaneously.

Modelling concomitant sodium and chromium enrichments in clinopyroxenes
Here we model the variations of Na2O in clinopyroxenes during partial melting of a peridotite with an incremental open-system melting model, which assumes that a melt enriched in incompatible elements enters the system during the melting process (Ozawa and Shimizu, 1995;Brunelli et al., 2014). The model assumes Na behaving as a dilute element and uses the equations, melting modes and dynamic parameters of Brunelli et al. (2014). Initial mode (spinel-equilibrated) and Na2O concentrations (0.13 wt.%) of the DMM are from Workman and Hart (2005). Sodium partition coefficient between clinopyroxene and melt ( D cpx Na ) has been shown to be strongly pressure-dependent, becoming more compatible with pressure and reaching the unity at around 3 GPa in mafic systems (see the compilation of Bedard, 2014 and references therein). In order to reproduce the behaviour of Na during melting we performed several calculations varying D cpx Na ; this parameter is assumed constant for every calculation thus simulating an isobaric process. Sodium partition into orthopyroxene is shown to be independent from P, T and composition (Kinzler, 1997;Kinzler et al., 1992), averaging 0.05 with little variability. We assumed this value in all our calculations. The composition of the residual clinopyroxene is related to the composition of the residual peridotite by allows calculating the expected Cr# of the clinopyroxene for the model incremental degree of melting (Brunelli et al., in prep). Na2O concentrations can thus be plotted versus Cr# of the clinopyroxene as a proxy of the incremental degree of melting.
Hereafter we assumed a variable critical mass porosity (c). The critical mass porosity represents the threshold value for the porosity interconnection and hence melt extraction (Zou, 1998). Fractional melting corresponds to c =0 while batch melting c =1 (Zou, 1998). Brunelli et al. (2014)  when considering an open melting system. In this model, the notation c/F represents the ratio between the critical porosity and the degree of melting for a given melting cell; this ratio is a proxy of the batch character of the melting process becoming closer to pure fractional at low values (c/F  0) or to batch melting for high values (c/F 1).
In a fractional melting scenario, (c/F = 0; Fig. 11A Melting under dynamic conditions also reveals that the Na content of the residual clinopyroxene can be strongly enriched when the system behaves with a more batch character (high c/F values). However a better match of the observed Na2O-Cr# trend requires lowering the Na content of the source along with increasing the batch character of the process. In Fig. 11C, we observe that the Na2O-Cr# trend is well matched by assuming the source to be more Na-depleted than DMM, down to 0.08 Na2O wt.% in the source. With that value the Na2O-Cr# trend can be matched by increasingly higher c/F values i.e. toward more batch conditions. The higher is the batch character of the process the lower is the required Na partition, hence the inferred pressure of the process. According to the calibration of Bedard (2014), it appears that the inferred pressure values are deeper than 3 GPa, well inside the garnet stability field. High degrees of melting in the garnet field are however not recorded in the trace element patterns of the residual clinopyroxenes ( Fig. 9) nor admitted by the low thermal setting of the region. From this simple modelling it also appears that the depth of the process and its exact nature (c/F) cannot be determined contextually.

A C C E P T E D M A N U S C R I P T
24 We now assume that an exotic melt enters the system during the melting process. The parameter β represents the influx rate of the exotic melt. The composition of the melt cannot be defined a priori hence we run some calculations at variable sodium content. It appears that liquids having MORB composition (Na2O < 2 wt.%) do not modify substantially the trends derived for dynamic systems. When instead considering high Na melts (Na2O ≥ 3 wt.%) the Na2O-Cr# trend appears to be matched at progressively low pressures (lower D cpx Na ) and more fractional conditions. In Fig. 12 we report the model results for melting in open-system with influx of a Na-rich melt (Na2O = 4 wt.%) comparable to the high pressure melts hypothesized by Lundstrom (2000). We fixed the incoming rate at β = 0.2 and vary Modelling under open-system melting conditions suggests therefore that when relatively high Na melts enter the melting system the residual pyroxenes can be progressively enriched in Na when the system tends to near-batch conditions. However, even in case of near-fractional processes, the simple incoming of Na-rich melts can produce the observed scatter in the global set for residual mid-ocean ridge clinopyroxenes.

A C C E P T E D M A N U S C R I P T
25

Discussion
Melts produced by partial melting in the upwelling mantle ascent more rapidly than the surrounding solid matrix; they are channelized in the shallow mantle where partial melts aggregate and are transported to the surface (Kelemen et al., 1992). Melt channelling introduces substantial horizontal variations in the composition of the surrounding, partially molten peridotite, due to mineral reactions triggered by the differences of chemical compositions between the melt within the channel and the intergranular partial melt in equilibrium with the residual peridotite (Liang and Parmentier, 2010). On one hand, high-pressure melts are poorer in SiO2 relative to latestage melts because the olivine stability field expands at the expenses of orthopyroxene with decreasing pressure (Stolper, 1980). Consequently, reactions between highpressure melts undersaturated in orthopyroxene and peridotite equilibrated at low pressure may significantly modify the peridotite mode by orthopyroxene resorption and olivine precipitation (Daines and Kohlstedt, 1994;Kelemen et al., 1992). In another hand, early melts generated at greater depth from a more fertile lithology are also richer in Na2O and incompatible trace elements than interstitial melts in equilibrium with the residual peridotite at lower depth. The infiltration-reaction experiments of Lundstrom (2000,2003) have shown that Na diffuses rapidly in response to gradients in silica activity from the SiO2-poor/Na2O-rich melt into the adjacent partially molten peridotite.
The incongruent nature of the orthopyroxene melting results in an increase of the residual olivine and contextually leads to an increase in the silica content of the produced melt which in turn increases the diffusion rate of Na (Lundstrom, 2000). In Lundstrom's experiments this melting reaction is attested by changes in mineral modes (decrease in orthopyroxene, increase in olivine and melt) and mineral compositions (increasing Cr and decreasing Al VI in clinopyroxene; increasing Cr# and decreasing Mg#

A C C E P T E D M A N U S C R I P T
26 in spinel). In addition, the bulk content of Na2O is increased in the diffusively infiltrated peridotite, so that subsolidus re-equilibration of the peridotite with some amount of unextracted melt will result in a peridotite significantly enriched in Na2O. This process may be an alternative to the refertilization by addition of melt (Elthon, 1992) to explain Na enrichment in 'normal' abyssal peridotites. However it is insufficient to produce extreme enrichments in both Na and Cr.
From these observations we infer that high Na2O concentrations can be reached in the infiltrated peridotite when a Na-rich or alkaline melt is reinjected in the melting peridotite after (partial) extraction of the preceding melt batch. Increased Na content lowers the silica activity coefficient in the melt (Lundstrom, 2003  3-The positive correlation between Na and Cr does not necessary reflect a kosmochlor component, but may result from the fact that the Al-Tschermak's molecule decreases more rapidly than the sodic components (Jd ± Ko), leading to inverse correlation between octahedral and tetrahedral Al. Assuming that Al depletion and increasing Cr# in pyroxenes and spinel reflect increasing melt depletion in the peridotites, our modelling

A C C E P T E D M A N U S C R I P T
28 suggests that similar clinopyroxene compositional variation can be reproduced when Na is added during partial melting of the peridotite. This may involve upward percolation of a Na-rich melt derived from melting of low-solidus heterogeneities, or, alternatively, diffusion of Na from nearby melt channels into the partially molten peridotite during mantle upwelling, as suggested by Lundstrom (2000). Both explanations are consistent with the compositions of the associated basalts, which vary from normal-MORB to enriched MORB and alkali-basalts.
4-In clinopyroxenes of the global abyssal peridotite set, Na2O is positively correlated with the abundances of incompatible trace elements. However, in the dredges in which Na and Cr in clinopyroxenes are positively correlated, Na2O is correlated only with highly incompatible trace elements (e.g., Sr, Zr, Nb, LREE), whereas the less incompatible elements (e.g., Ti, Y, HREE) tend to remain nearly constant or to decrease.
We can predict that the study of a larger sample set and consideration of trace element behaviour will add more complexities to the present model. In this study, clinopyroxene major elements define uniform compositional trends because they are also controlled by the crystal structure, while trace element ratios will reflect different melt composition, pre-existing mantle heterogeneities and/or different melting parameters.

A C C E P T E D M A N U S C R I P T
29 ICPMS analyses and to Bernard Boyer for his technical assistance with the SIMS analyses. Philippe Recourt is thanked for the SEM-EDS assistance and Michel Fialin for the EMPA assistance. The manuscript has been substantially improved after thorough reviews by an anonymous reviewer.

A C C E P T E D M A N U S C R I P T
A C C E P T E D M A N U S C R I P T    Seyler et al., 2011). Normalization values are from Anders and Grevesse (1989). DMM clinopyroxene is from Workman and Hart (2005).

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T 44 wt.%) is from Workman and Hart (2005) in panels A and B., and Na2O = 0.08 wt.% in panel C. Peridotite samples are represented as in Fig. 8. 4-Numerical modeling suggests that the peridotite interacted with a sodium-rich melt during partial melting in the upwelling mantle.