Partial melting of secondary pyroxenite at 1 and 1.5 GPa, and its role in upwelling heterogeneous mantle

Abstract

We performed partial melting experiments at 1 and 1.5 GPa, and 1180–1400 °C, to investigate the melting under mantle conditions of an olivine-websterite (GV10), which represents a natural proxy of secondary (or stage 2) pyroxenite. Its subsolidus mineralogy consists of clinopyroxene, orthopyroxene, olivine and spinel (+garnet at 1.5 GPa). Solidus temperature is located between 1180 and 1200 °C at 1 GPa, and between 1230 and 1250 °C at 1.5 GPa. Orthopyroxene (±garnet), spinel and clinopyroxene are progressively consumed by melting reactions to produce olivine and melt. High coefficient of orthopyroxene in the melting reaction results in relatively high SiO₂ content of low melt fractions. After orthopyroxene exhaustion, melt composition is controlled by the composition of coexisting clinopyroxene. At increasing melt fraction, CaO content of melt increases, whereas Na₂O, Al₂O₃ and TiO₂ behave as incompatible elements. Low Na₂O contents reflect high partition coefficient of Na between clinopyroxene and melt ($D_{{{\text{Na}}_{ 2} {\text{O}}}}^{{{\text{cpx}}/{\text{liquid}}}}$). Melting of GV10 produces Quartz- to Hyperstene-normative basaltic melts that differ from peridotitic melts only in terms of lower Na₂O and higher CaO contents. We model the partial melting of mantle sources made of different mixing of secondary pyroxenite and fertile lherzolite in the context of adiabatic oceanic mantle upwelling. At low potential temperatures (T _P < 1310 °C), low-degree melt fractions from secondary pyroxenite react with surrounding peridotite producing orthopyroxene-rich reaction zones (or refertilized peridotite) and refractory clinopyroxene-rich residues. At higher T _P (1310–1430 °C), simultaneous melting of pyroxenite and peridotite produces mixed melts with major element compositions matching those of primitive MORBs. This reinforces the notion that secondary pyroxenite may be potential hidden components in MORB mantle source.

Introduction

Several geochemical and petrological observations support the idea that the source of oceanic basalts consists of heterogeneous mantle including peridotite mixed to pyroxene-rich olivine-poor mafic to ultramafic lithologies, such as eclogite and pyroxenite (e.g., Allègre and Turcotte 1986; Hirschmann and Stolper 1996; Kogiso et al. 2004a; Hofmann 2007; Sobolev et al. 2007; Lambart et al. 2013 and references therein). In particular, the contribution of pyroxene-rich lithologies has been invoked to explain the large isotopic and trace elements variability of oceanic basalts, although uncertainties still exist on how they affect their major element chemistry. Pyroxenites can either represent recycled crustal material or be introduced in the mantle by various magmatic processes, thus resulting in pyroxene-rich mantle components with a wide range of compositions (e.g., Kogiso et al. 2004a; Herzberg 2011; Lambart et al. 2013).

A large number of experimental studies has been dedicated to partial melting of pyroxenites, and their phase relations and melting behavior, including solidus temperature and melt productivity, have been defined for a wide range of temperature, pressure and chemical compositions (e.g., Yaxley and Green 1998; Kogiso and Hirschmann 2001; Hirschmann et al. 2003; Pertermann and Hirschmann 2003a; Kogiso et al. 2004a; Keshav et al. 2004; Médard et al. 2006; Sobolev et al. 2007; Lambart et al. 2009a, 2013). Profiting of available experimental data, increasingly sophisticated geochemical models have investigated the role of mixed pyroxenite–peridotite mantle sources in the genesis of mid-ocean ridge basalts (MORBs) and ocean island basalts (OIBs) (e.g., Hirschmann and Stolper 1996; Phipps Morgan 2001; Sobolev et al. 2005, 2007; Stracke and Bourdon 2009; Herzberg 2006, 2011; Shorttle and Maclennan 2011; Guerenko et al. 2013; Lambart et al. 2012, 2016; Shorttle et al. 2014; Delavault et al. 2015; Brown and Lesher 2016).

An important issue concerning the melting of pyroxenite-bearing mantle sources is the role of interaction between pyroxenite-derived melts and adjacent peridotite. Low-solidus mantle lithologies, such as crustal-derived rocks (eclogite or low-MgO pyroxenite), start to melt at higher pressure than “dry” peridotites (e.g., Yasuda et al. 1994; Pertermann and Hirschmann 2003a,b; Kogiso et al. 2004a; Kogiso and Hirschmann 2006; Yaxley and Sobolev 2007; Spandler et al. 2008). Their partial melts are expected to react to some extent with the surrounding subsolidus peridotite either modifying the composition of the rising melts (Pilet et al. 2008; Mallik and Dasgupta 2012, 2013, 2014), or creating new hybrid rock-types, called secondary or stage 2 pyroxenites (e.g., Yaxley and Green 1998; Sobolev et al. 2005, 2007; Herzberg 2006, 2011; Lambart et al. 2012). Moreover, heterogeneous upwelling mantle is subject to continuous events of partial melting and melt–rock reaction that potentially generate a large range of new lithological heterogeneities, including metasomatized peridotites and variably residual pyroxenites (e.g., Yaxley and Green 1998; Rosenthal et al. 2014). On this basis, recent models of basalts petrogenesis have proposed that secondary pyroxenites could represent diffuse mantle components in the MORB mantle sources (e.g., Sobolev et al. 2007; Lambart et al. 2012). Remarkably, such hybrid components are expected to produce MgO-rich partial melts that are able to percolate through peridotite with minimum interactions, thus preserving the pyroxenite signature of the original melts (Lambart et al. 2009a, 2012). These pyroxenites represent good candidates to transmit the trace element and isotopic signature of deep mantle heterogeneities without impacting the major element composition of basalts.

In spite of their inferred potential role in oceanic basalt generation, the origin of such hybrid mantle components, as well as the composition of their partial melts, is still not well defined, mostly due to the scarceness of available experimental data (Lambart et al. 2013). Lambart et al. (2012) investigated the interaction between melts produced by silica-deficient pyroxenites and peridotite, concluding that this process leads to the formation of secondary high-MgO pyroxenites, compositionally similar to those documented in the Ronda peridotite (Garrido and Bodinier 1999). Sobolev et al. (2007) investigated the melting behavior at 3.5 GPa of an olivine-free secondary pyroxenite originated by eclogite-derived silica-rich melts reacted with peridotite. However, experimental knowledge of the compositions of melts generated by secondary-type pyroxenites in the pressure range of basalt generation at mid-ocean ridges (i.e. 1–1.5 GPa; e.g., Presnall et al. 2002; Falloon et al. 2008) is currently not available.

Natural examples of secondary pyroxenites, originated by melt-peridotite reaction processes, have been documented in a number of studies on ultramafic massifs (i.e. Garrido and Bodinier 1999; Bodinier et al. 2008; Gysi et al. 2011; Marchesi et al. 2013; Borghini et al. 2013, 2016; Montanini and Tribuzio 2015). Specifically, pyroxenite layers in External Liguride (EL) mantle sequences (Northern Apennines, Italy) have been recently interpreted as the product of the reaction between deep infiltrated melts, likely derived from a hybrid eclogite-peridotite source, and the host peridotite (Borghini et al. 2016). Hence, they represent an excellent natural example of secondary (or stage 2) pyroxenite, which forms together with their host peridotite a potential source of E-MORBs (Borghini et al. 2013).

In this paper, we present the results of partial melting experiments performed on a selected natural olivine-websterite, whose origin by deep melt-rock reaction has been well supported on the basis of microstructural and geochemical observations (Borghini et al. 2013, 2016). The major aims of our work are to derive the extent and composition of melts produced by moderate to high degrees of melting of a secondary-type pyroxenite at 1 and 1.5 GPa and discuss the contribution of such melts to the composition of oceanic basalts.

Experimental and analytical techniques

Choice and preparation of starting material

The investigated bulk composition is an olivine websterite (sample GV10 of Borghini et al. 2016), representing a well-preserved cm-width pyroxenite layer from the External Liguride (EL) ophiolitic mantle sequences (Northern Apennines, Italy). Although the EL peridotites are considered exposed subcontinental lithospheric mantle sectors (e.g., Rampone et al. 1995), their MORB-like isotopic composition and chemical features suggest that they were formed by lithospheric accretion of ordinary asthenospheric mantle (Rampone et al. 1995). Therefore, they represent good proxies of an MORB-type fertile mantle source that is modified by deep melt infiltration and melt–rock reaction processes before it undergoes partial melting and basaltic melt extraction (Borghini et al. 2013, 2016), as envisaged by recent MORB petrogenesis models (e.g., Lambart et al. 2012).

Olivine-websterite GV10 belongs to a group of pyroxenites having a geochemical signature inherited by a primary assemblage dominated by garnet and pyroxenes (mostly clinopyroxene and minor orthopyroxene). Originally, they represented clinopyroxene-rich garnet pyroxenites with orthopyroxene-rich borders against the host peridotite, indicative of an origin by silica-rich melt infiltration and crystallization within the peridotite at pressure >1.5 GPa (Borghini et al. 2016). Microstructural and geochemical observations on pyroxenite GV10 suggest that it was originated by deep melt-rock reaction, likely involving peridotite and a pyroxenite (or eclogite)-derived melt, thus supporting its secondary origin (Borghini et al. 2016). Moreover, the emplacement of these pyroxenite layers introduced short-length scale isotopic heterogeneity in the host peridotite, producing veined mantle domains with enriched Nd-isotopic signature that could represent a source of E-MORBs (Borghini et al. 2013). The selected sample GV10 thus constitutes an excellent analog of a secondary pyroxenite component in a fertile MORB mantle source.

Olivine-websterite GV10 has high X _Mg [molar Mg/(Mg + Fe^tot)] (0.82–0.88) coupled with high CaO, moderate Al₂O₃ and low Na₂O contents (Fig. 1). Similar bulk compositions have been documented in Group C pyroxenites from Ronda (Spain; Garrido and Bodinier 1999) (Fig. 1). They have been inferred to result from interaction between peridotite and melts derived from garnet pyroxenites and have been recently interpreted as secondary pyroxenite (Lambart et al. 2012). Compared to bulk compositions of nominally “dry” pyroxenites previously investigated at 1 and 1.5 GPa, GV10 has higher CaO and lower Na₂O contents with respect to pyroxenites investigated by Lambart et al. (2009a) (Table 1; Fig. 1), and higher Al₂O₃ and slightly lower SiO₂ than clinopyroxenites studied by Kogiso and Hirschmann (2001) at 1.0 GPa (Table 1; Fig. 1). In spite of similar X _Mg, GV10 significantly differs in composition from model secondary pyroxenite PX-1 studied by Sobolev et al. (2007), for much higher CaO and lower SiO₂ and Na₂O contents (Fig. 1). Such difference in bulk rock composition suggests that pyroxenite GV10 did not simply originate by eclogite melt–peridotite reaction (as for model secondary pyroxenite PX-1, Sobolev et al. 2005), but more presumably resulted by interaction between a melt produced by an eclogite-peridotite mixed source (e.g., Mallik and Dasgupta 2012) and a peridotite, as argued by Borghini et al. (2016).

Fig. 1

X_Mg value vs. Al₂O₃, CaO, Na₂O and SiO₂ (wt%) diagrams showing the composition of the starting material used in this study (GV10), together with the composition of nominally “dry” pyroxenite and some peridotite bulks investigated in previous experimental studies at 1 and 1.5 GPa (see also Table 1). They are the silica-deficient pyroxenites M5-103, M5-40 and M7-16 studied by Lambart et al. (2009a); clinopyroxenites OLCPX1, OLCPX2 and Pyrox2B by Kogiso and Hirschmann (2001); garnet pyroxenite MIX1G (Hirschmann et al. 2003); lherzolite MM3 (Baker and Stolper 1994; Falloon et al. 2008); depleted lherzolite DMM1 (Wasylenki et al. 2003); MORB-pyrolite, MPY, and Tinaquillo lherzolite, TQ (Robinson et al. 1998). Purple star is the composition of the model secondary (silica-excess) pyroxenites experimentally investigated at 3.5 GPa by Sobolev et al. (2007). Also reported are the compositional fields of worldwide pyroxenites (black line) and External Liguride (EL) pyroxenites (dotted line) from Borghini et al. (2016), and group-C pyroxenites from Ronda (orange line) from Garrido and Bodinier (1999)

Table 1 Starting compositions experimentally investigated at 1 and 1.5 GPa

Pyroxenite GV10 has been crushed in an agate mortar, and pulverized to 2–4 μm using an agate-micronizing mill. Two pellets obtained pressing about 0.5 g of rock powder each have been wrapped in platinum foils, previously preconditioned for Fe, and mounted at the end of a shaft in a vertical furnace. A first Fe preconditioning of platinum foil was performed at 1500 °C for about 12 h at FMQ fO₂ conditions, by covering it with GV10 rock powder. Glass GV10 was obtained by heating the pellets up to 1500 °C at FMQ fO₂ conditions through a night and quenched in dry ice. Electron microprobe analysis on the GV10 glass indicated its homogeneous composition and the lack of any residual mineral phase. Fe loss was lower than 4% with respect to the starting rock powder composition (Table 1).

Experimental and analytical techniques

Experiments were conducted at 1.0 and 1.5 GPa, and temperatures from 1180 to 1400 °C, at the Laboratorio di Petrologia Sperimentale, Dipartimento di Scienze della Terra, University of Milano. Experiments at 1 GPa were carried out in a single-stage piston cylinder and experiments at 1.5 GPa in end-loaded piston cylinder using MgO-Pyrex-Salt assemblies. Run lasted from 50 to 240 h (Table 2). As we did not perform time series experiments, run durations of this work have been chosen following previous experiments on pyroxenite melting (i.e. Kogiso and Hirschmann 2001; Pertermann and Hirschmann 2003a, b). Accordingly, we performed super-solidus experiments with run durations ranging from 50 to 155 h. Subsolidus experiments lasted up to 240 h, in order to approach equilibrium of the run products and favor the development of textures with grain size suitable for microprobe analyses (Table 2).

Table 2 Summary of experimental conditions and run products

Approximately, 20 mg of starting material was loaded into a graphite inner capsule (outer diameter 2.8 mm), and then welded into an outer Pt capsule (outer diameter 3.0 mm, length about 7–8 mm). Graphite is used to isolate the sample from the Pt capsule and avoid Fe-loss (e.g., Kinzler 1997; Walter 1998). Furthermore, the graphite-Pt assembly keeps the oxygen fugacity below the graphite–C–O vapor buffer (e.g., Ulmer and Luth 1991; Médard et al. 2008). We added a layer of vitreous carbon spheres (80–125 μm diameter) at the top of the capsule to enhance the storage and segregation of melt (e.g., Pickering-Witter and Johnson 2000; Wasylenki et al. 2003; Médard et al. 2006). Carbon spheres may also absorb water if present in the charge (Robinson et al. 1998). The mass of the vitreous spheres layer never exceeded 15% of the starting glass. To maintain anhydrous conditions, the platinum-graphite capsule loaded with the starting material was dried overnight in an oven at 250 °C before being rapidly welded shut. The thermocouple tip was separated from the platinum capsule by a 0.5-mm thick hard corundum disc. Assemblies were kept in oven at about 200 °C for several hours before running the experiments. Temperature was measured by K-type and S-type thermocouples and is considered to be accurate to ±5 °C. An initial pressure of 0.25 GPa was applied, then the sample was first heated to 400 °C for 10 min in order to soften the Pyrex; pressure was raised to the experimental value before reaching the desired temperature. Runs were terminated by turning off the power. Capsules were enclosed in epoxy, sectioned lengthwise, polished and carbon-coated.

Run products were inspected by back-scattered electron images (BSE) and microanalyses were performed using a JEOL JXA 8200 Superprobe equipped with five WDS-wavelength-dispersive spectrometers and one energy dispersive spectrometer (EDS) at the Dipartimento di Scienze della Terra, University of Milano. Both images and X-ray element maps were extremely useful in textural examination of the experimental charges. Analyses on mineral phases were performed using 1 µm beam size and beam conditions of 15 kV and 5 nA. Counting time was 30 s for peak and 10 s for background. Whenever possible, in order to prevent alkali loss, we analyzed glass using a beam size of 5 µm, or 2–3 µm in thin interstitial films in experiments with very low melt fractions.

Results

Approach to equilibrium

The approach to equilibrium in experiments of this study is assessed by a number of observations, first of all the growth of compositionally homogeneous chemically unzoned minerals, as revealed by X-ray element maps (Supplementary Figure 1). A solid proof of approach to equilibrium is the systematic and consistent variations in mineral chemistry at different pressure and temperature conditions, as well as coherent element partitioning (see below). Phase abundances have been calculated by mass balance between the composition of coexisting phases and the bulk composition, using a weighted least squares method (Table 2). Residual sums of squares ranging from 0.09 and 0.63 are perfectly comparable to other experimental studies (i.e. Kogiso and Hirschmann 2001; Pertermann and Hirschmann 2003a, b; Lambart et al. 2009a). Moreover, results of mass balance calculations indicate that Fe loss, possibly due to fractures in graphite inner capsule to the Pt outer capsule, was negligible in these experiments (Table 2).

As a further test, we applied the method proposed by Toplis (2005). The values of distribution coefficient K _D between olivine and liquid measured in our experiments differ from those computed using thermodynamic calculation (Toplis 2005) within the range of ±0.02 (Table 2), thus supporting the attainment of equilibrium. As only exception, run GV10-83-1 at very low degree of melting (F = 1.8%, see below) yielded an olivine/liquid Kd_Fe-Mg measured value much lower than the calculated one. On this experiment, we did not adopt specific technique to account for the effect of quench crystal growth on the very small spots of melt, as applied in previous dedicated studies (i.e. Hirose and Kushiro 1993; Wasylenki et al. 2003; Laporte et al. 2004; Dasgupta and Hirschmann 2007; Falloon et al. 2008). Accordingly, the composition of glasses produced at very low melting degrees (F < 5%) in our experiments must be considered with caution.

Phase assemblages and textures

Run products are summarized in Table 2, together with experimental details.

Crystal size ranges from 1 to 10 μm in subsolidus experiments and increases with increasing temperature, around 5–15 μm in experiments with low melt fractions (<20 wt%), and up to 80 μm in near-liquidus runs.

At 1 GPa and 1180 °C, the stable mineral assemblage is made of prismatic grains of clinopyroxene, and orthopyroxene (5–10 μm in size), minor rounded olivine and tiny spinel (mostly 2–5 μm in size). At 1200 °C, few tiny interstitial patches of glass coexist with clinopyroxene, orthopyroxene, olivine and spinel. At 1250 °C, orthopyroxene is absent; residual clinopyroxene and olivine reach 15 μm in size (Fig. 2a), and interstitial glass blebs (mostly confined at the top of the capsule) are larger as compared to lower temperature experiment. At 1280 °C, glass is still in equilibrium with clinopyroxene, olivine and spinel and it occurs as large pools (about 120–150 μm) at the rim of the capsule. At 1300 °C, spinel is absent and a homogeneous glass coexists with coarse residual clinopyroxene and olivine (10–30 μm). At 1330 °C, olivine is the only residual phase and it consists of coarse polygonal grains confined at the bottom portion of the capsule (Fig. 2b).

Fig. 2

Representative textures in melting experiments on pyroxenite GV10. a Back-scattered electron (BSE) image of melting experiment GV10-83-2 (1 GPa; 1250 °C): glass is mostly confined at the top of the capsule and occurs as interstitial blebs between residual clinopyroxene and olivine (around 15 μm in size) and smaller spinel (up to 5 μm). b BSE image of run GV10-83-8 (1 GPa; 1330 °C): residual olivine as coarse polygonal grains floating in the glass at the bottom of the capsule. c BSE image of experiment GV10-83-21 (1.5 GPa; 1370 °C): coarse clinopyroxene and olivine coexist with glass is either contained in the trap layer or as a pool towards the top of capsule

In experiments at 1.5 GPa, a subsolidus fine-grained (>10 μm) mineral assemblage made of clinopyroxene, olivine, orthopyroxene, spinel and garnet is stable at 1230 °C. Garnet is the first phase to disappear at 1250 °C, and orthopyroxene is no more present at 1280 °C. In runs at temperatures from 1280 to 1350 °C, glass proportion increases progressively with temperature and the residual phases are clinopyroxene, olivine and spinel (Table 2). Grain size is almost constant with clinopyroxene and olivine ranging 10–30 μm and spinel always smaller than 5 μm. At 1370 °C, spinel is entirely consumed and coarse clinopyroxene and olivine coexist with glass; in such experiments with high degrees of melting, glass is either contained in the trap layer or as a pool towards the top of the capsule (Fig. 2c). At 1400 °C, rounded large olivines float in a homogeneous glass, thus indicating that olivine is the liquidus phase also at 1.5 GPa.

Liquid compositions as a function of pressure and temperature

The glass compositions are reported in Table 3 and in Fig. 3, where major and minor element contents are plotted against melt fraction. Melt compositions are basaltic and vary significantly as a function of melt fraction (F) and only slightly with pressure.

Table 3 Average compositions (wt%) of liquids and residual phases

Fig. 3

Oxide concentrations in experimental melts plotted as a function of melt fractions (wt%). Data plotted at F = 100% correspond to the bulk-rock composition. Also reported for comparison are trends (black lines) extrapolated by melting experiments at 1 and 1.5 GPa on three silica-deficient pyroxenites by Lambart et al. (2009a). Grey symbols linked by dotted lines are experimental results at 1 GPa on clinopyroxenite bulks (Kogiso and Hirschmann 2001). Comparison between the compositions of melts produced by GV10 at 1 and 1.5 GPa is illustrated in Supplementary Figure  2

At increasing melt fraction, MgO and CaO abundances progressively increase and Al₂O₃, Na₂O and TiO₂ contents decrease at both 1 and 1.5 GPa (Fig. 3). Na₂O and TiO₂ contents are lower than about 2 wt% in all experiments, and they decrease at increasing melt fraction at both pressures (Fig. 3). K₂O behaves as highly incompatible element, reaching about 0.4 wt% in experiments with the lowest melt fractions at 1 and 1.5 GPa, and drastically decreasing at increasing melting degree (Table 3).

The highest SiO₂ contents are observed in the experiments with the lowest melt fraction (F < 8%); SiO₂ abundance significantly decreases as long as orthopyroxene is a melting phase and it remains almost constant once orthopyroxene is exhausted, i.e. at F = 2–12% at 1 GPa and F > 7% at 1.5 GPa. The FeO compositional trend is rather flat at 1 GPa, ranging from 7.1 to 8.3 wt%; at 1.5 GPa it has a convex shape, increasing up to 9.7 wt% with orthopyroxene consumption (corresponding to 20% melt fraction), and slightly decreasing after orthopyroxene disappearance at higher melt fraction (Fig. 3). Cr₂O₃ melt contents are very low as long as spinel is present as residual mineral, and increases at high melt fraction (F > 60%) after spinel exhaustion, at both 1 and 1.5 GPa (Fig. 3). The X _Mg value [X _Mg = Mg/(Mg + Fe_tot)] of melts increases with the melt fraction, from 0.53 to 0.79 in runs at 1 GPa, and from 0.63 to 0.80 at 1.5 GPa (Table 3). It is positively correlated with X _Mg value of olivine and clinopyroxene (Fig. 4), supporting approach to equilibrium.

Fig. 4

Fe-Mg partitioning of glass vs. olivine and clinopyroxene. X _Mg = Mg/(Mg + Fe^tot)

Differences in melt composition as a function of pressure are generally modest, except for higher SiO₂ and CaO abundances in experiments at 1 GPa with respect to those at 1.5 GPa (Supplementary Figure 2).

In Fig. 5, the compositions of melts produced by GV10 pyroxenite are projected from diopside (Di) in the pseudo-ternary diagram forsterite (Fo)–Ca-Tschermak’s pyroxene (CaTs)–quartz (Qz), together with the compositions of partial melts obtained in previous pyroxenite melting experiments at 1–1.5 GPa (Kogiso and Hirschmann 2001; Lambart et al. 2009a). Lambart et al. (2009a) demonstrated that the majority of melts produced by partial melting of silica-deficient pyroxenites plot on the right side of the CaTs-En join (Fig. 5), thus attesting the absence of a thermal divide at P < 2 GPa. Partial melts from GV10 pyroxenite, which is silica-deficient too, also plot on the right side of the CaTs-En join (Fig. 5), further supporting that the thermal divide is not relevant at such low pressures (1–1.5 GPa).

Fig. 5

Molar projections from diopside [Di] onto the pseudo-ternary diagram forsterite (Fo)—calcium tschermak (CaTs)—quartz (Qz) (O’Hara 1972) of melts produced by experiments on GV10 at 1 and 1.5 GPa. The small yellow circles represent the composition of solid residues calculated using the modes and composition of residual minerals in experimental charges. The melt compositions reported in previous experimental studies on other pyroxenites are also reported (Lambart et al. 2009a; Kogiso and Hirschmann 2001). Symbols are as in Fig. 1. Larger symbols represent the starting materials

Mineral compositions as a function of pressure and temperature

Average compositions of minerals from each experimental run are listed in Table 3.

X _Mg value of olivine is 0.83 in the subsolidus runs at 1 and 1.5 GPa, and it increases with increasing degree of melting up to 0.92 (Table 3; Fig. 4). CaO content in olivine is rather high, spanning 0.3–0.5 wt% at both 1 and 1.5 GPa (Table 3), and it is comparable to previous experimental studies (e.g., Libourel 1999; Kogiso and Hirschmann 2001).

Clinopyroxene at subsolidus conditions has X _Mg values equal to 0.84, and rather high Na content (recalculated on the basis of 6 oxygens and all Fe = Fe²; Na = 0.046 and 0.052 a.p.f.u. at 1 and 1.5 GPa, respectively), suggesting an increase of the jadeitic component with pressure (Table 3). At increasing melt fraction, residual clinopyroxene shows progressively higher X _Mg values, ranging from 0.84 to 0.9 (Fig. 4), and systematically lower jadeite (from about 5 to 1%) and TiO₂ concentrations, at both 1 and 1.5 GPa (Table 3). The Al₂O₃ content does not vary significantly along the melting interval, but it tends to be higher in residual clinopyroxene of melting experiments at 1.5 GPa (Table 3). The CaO abundance tends to increase with melt fraction in experiments at 1 GPa, whereas it remains rather constant at 1.5 GPa (Table 3). The Cr₂O₃ content slightly increases at higher melt fraction (Table 3).

Spinels are Al-rich and have X _Mg (0.68–0.77 at 1 GPa, 0.66–0.82 at 1.5 GPa) and X _Cr (0.021–0.036 at 1 GPa, 0.019–0.038 at 1.5 GPa) that increase with melt fraction. Their TiO₂ content stays low at both 1 and 1.5 GPa (TiO₂ ≤0.26 wt%) without any correlation with pressure or melt fraction (Table 3).

Orthopyroxenes have X _Mg ranging 0.84–0.85, and Al₂O₃ contents around 7–8 wt% in subsolidus runs, as well as in near-solidus experiments. The CaO contents are rather high, with values from 1.9 to 2.1 wt% at 1 GPa and from 1.8 to 2.1 wt% at 1.5 GPa (Table 3).

Average composition of garnet in subsolidus experiments at 1.5 GPa and 1230 °C consists of 0.62 pyrope, 0.20 almandine, 0.14 grossular molar fractions, and it shows very low Cr₂O₃, TiO₂, Na₂O contents (Table 3).

Solidus temperature and melt productivity

The solidus of pyroxenite GV10 is between 1180 and 1200 °C at 1 GPa, and between 1230 and 1250 °C at 1.5 GPa (Fig. 6). At pressure of 1 and 1.5 GPa, pyroxenite GV10 starts melting at lower temperature, compared to all fertile lherzolites experimentally investigated (Fig. 6).

Fig. 6

Experimentally determined solidus temperature at 1 and 1.5 for pyroxenite GV10 compared to other pyroxenites and peridotites. Pyroxenites are: M5-103, M5-40 and M7-16 studied by Lambart et al. (2009a); OLCPX1, OLCPX2 and pyrox2B by Kogiso and Hirschmann (2001). Peridotites are: lherzolite MM3 (Falloon et al. 2008), fertile lherzolite MPY and depleted lherzolite TQ (Robinson et al. 1998), depleted lherzolite DMM1 (Wasylenki et al. 2003). Symbols are as in Fig. 1. Temperature interval named “Fertile peridotite” at 1.0 GPa includes experiments of variably fertile lherzolites from Pickering-Witter and Johnson (2000) and Schwab and Johnston (2001)

Melt fraction vs. temperature trends at both 1 and 1.5 GPa are concave upward, suggesting that isobaric melt productivity (dF/dT)_P increases with temperature (Fig. 7), in agreement with previous experimental results on pyroxenite or eclogite compositions (e.g., Pickering-Witter and Johnson 2000; Kogiso and Hirschmann 2001; Pertermann and Hirschmann 2003b; Lambart et al. 2009a), and thermodynamic models of peridotite partial melting (i.e. Hirschmann et al. 1999; Asimow et al. 2001). At both pressures, the isobaric melt productivity at low melting degrees (F < 12.6%) is about 0.2%/°C (Fig. 7). At higher melting degrees (F > 12.6%), the mean isobaric productivity is 0.9%/°C at 1 GPa and 0.6%/°C at 1.5 GPa (Fig. 7).

Fig. 7

Temperature-melt fraction curves determined at 1 GPa (a) and 1.5 GPa (b) for pyroxenite GV10 (yellow squares) compared to those extrapolated by previous experimental work on dry pyroxenite and peridotite: (1) pyroxenites M5-103, M5-40 and M7-16 from Lambart et al. (2009a); (2) clinopyroxenites Pyrox2B, OLCPX1 and OLCPX2 from Kogiso and Hirschmann (2001); (3) lherzolite MM3 from Baker et al. (1995) (green squares) and Falloon et al. (2008) (green circles)

Discussion

Melting and phase relations

Phase abundances and glass proportions have been calculated by weighted least squares mass balance that included nine oxides, SiO₂, TiO₂, Al₂O₃, Cr₂O₃, FeO, MgO, CaO, Na₂O and K₂O. Detailed explanation of mass balance calculations is reported in Borghini et al. (2010).

In the pressure and temperature range investigated in this study, olivine is stable along the whole melting range with low modal abundances (<14%; Table 2; Fig. 8). Clinopyroxene is the dominant phase in GV10 pyroxenite residual assemblage (Table 2; Fig. 8). Spinel persists over than 40% of melt fraction at both pressures (Table 2; Fig. 8). Orthopyroxene disappears from the residue at F higher than 12%, and low modal abundance of garnet is stable only in the subsolidus assemblage at 1.5 GPa (Table 2; Fig. 8). The sequence of mineral phases disappearance at increasing melt fraction is, therefore, orthopyroxene–spinel–clinopyroxene–olivine and garnet–orthopyroxene–spinel–clinopyroxene–olivine, at 1 and 1.5 GPa, respectively.

Fig. 8

Modal abundance calculated by weighted least squares mass balance in experiments at 1 GPa (a) and 1.5 GPa (b) as a function of temperature. The grey dotted lines represent the temperature of mineral phases disappearance. ol olivine, cpx clinopyroxene, opx orthopyroxene, sp spinel, gnt garnet

To constrain the stoichiometry of reactions controlling the consumption of mineral phases along the melting interval of GV10, we performed mass balance calculations (within the same chemical system used for determination of modal abundances, see above), using the mineral and glass compositions at the beginning and the end of each phase exhaustion step with increasing melt fraction. Average melting reactions for pyroxenite GV10 are reported in Table 4, together with those for pyroxenites and peridotites at 1 and 1.5 GPa available in the literature. Melting reactions in pyroxenite GV10 at 1.5 GPa are similar to those at 1 GPa, except for the first melting interval at 1.5 GPa in which garnet is rapidly consumed (reaction 14, Table 4).

Table 4 Melting reactions (wt%) for pyroxenites and peridotites at 1 and 1.5 Gpa

Although the melting relation opx + cpx + sp → liq + ol is the same for pyroxenites and peridotite (Table 4), the contribution of orthopyroxene with respect to clinopyroxene is higher when pyroxenite is considered, as suggested for example by reaction 1 at 1 GPa (Table 4):

$$0. 8 6 {\text{ opx}} + 0. 5 7 {\text{ cpx}} + 0.0 8 {\text{ sp}} = 1 {\text{ liq}} + 0. 5 1 {\text{ ol}}$$

This results in rapid consumption of orthopyroxene during melting, and is likely related to the high CaO content in the bulk rock (Fig. 1), favoring the stabilization of clinopyroxene over orthopyroxene in the residue. Similar melting relation was documented in other pyroxenites and clinopyroxene-rich lherzolites (e.g., FER-B, Pickering-Witter and Johnson 2000; M5-103, Lambart et al. 2009a).

At F higher than 12%, clinopyroxene is the main phase that contributes to melting, with minor contribution of spinel until its exhaustion (Table 4; Fig. 8), following the average reaction (e.g., at 1 GPa, reaction 2, Table 4):

$$0. 9 7 {\text{ cpx}} + 0. 1 {\text{ sp}} = 1 {\text{ liq}} + 0.0 7 {\text{ ol}}$$

The coefficients are comparable to those obtained in previous experimental works on anhydrous pyroxenites and clinopyroxene-rich peridotites (reactions 5 and 6, Table 4). Around F = 90%, clinopyroxene is also consumed and olivine is the liquidus phase (reactions 3 and 17, Table 4; Fig. 8).

As expected, along the whole melting interval, residual phases show progressively higher X _Mg values at increasing melt fraction (Table 3; Fig. 4). At low melting degrees, the high coefficient of orthopyroxene involved in the melting reaction (reactions 1, 14 and 15, Table 4) results in relatively high SiO₂ content in partial melts at both pressures (Fig. 3). After orthopyroxene exhaustion, melt composition is controlled by the composition of coexisting clinopyroxene and, to a lesser extent, spinel. CaO content of partial melts increases with melt fraction until the disappearance of clinopyroxene (Fig. 3).

The most striking compositional feature observed in clinopyroxene is the decrease in jadeite component with increasing melt fraction. As a whole, the low Na₂O contents in melts (see Fig. 3) reflect relatively high partition coefficient of Na between residual clinopyroxene and partial melt ($D_{{{\text{Na}}_{ 2} {\text{O}}}}^{{{\text{cpx}}/{\text{liquid}}}}$). As shown in Fig. 9, $D_{{{\text{Na}}_{ 2} {\text{O}}}}^{{{\text{cpx}}/{\text{liquid}}}}$ calculated from experiments on pyroxenite GV10 are positively correlated with pressure and temperature. The increase of $D_{{{\text{Na}}_{ 2} {\text{O}}}}^{{{\text{cpx}}/{\text{liquid}}}}$ with pressure observed in GV10 experiments is well correlated with the $D_{{{\text{Na}}_{ 2} {\text{O}}}}^{{{\text{cpx}}/{\text{liquid}}}}$ calculated from experiments at 2 and 2.5 GPa on pyroxenites with comparable bulk composition (MIX1G, Hirschmann et al. 2003; M5-40, Lambart et al. 2013) (Fig. 9a). This overall correlation is in good agreement with available quantitative models for Na partitioning between clinopyroxene and silicate melt (Blundy et al. 1995; Villiger et al. 2007), as well as with the trend defined by fitting literature experimental data (Villiger et al. 2007) (Fig. 9a).

Fig. 9

Partitioning of Na₂O between clinopyroxene and glass ($D_{{{\text{Na}}_{ 2} {\text{O}}}}^{{{\text{cpx}}/{\text{liquid}}}}$) as a function of pressure (a) and temperature (b). Our data are compared with $D_{{{\text{Na}}_{ 2} {\text{O}}}}^{{{\text{cpx}}/{\text{liquid}}}}$ from experiments on lherzolites MM3 (Falloon et al. 2008), MPY and TQ (Robinson et al. 1998), fertile lherzolites FER-B (Pickering-Witter and Johnson 2000), INT-A and INT-B (Schwab and Johnston 2001), pyroxenites MIXG1 from Hirschmann et al. (2003), M5-40 and M7-16 from Lambart et al. (2009a), OLCPX1, OLCPX2 and pyrox2B from Kogiso and Hirschmann (2001). Data in the system diopside-albite are from Blundy et al. (1995) and from fractional and equilibrium crystallization experiments are by Villiger et al. (2007). Curves 1 and 2 in (a) refer to quantitative model proposed by Blundy et al. (1995) and fit of literature experimental data by Villiger et al. (2007), respectively. Number between parenthesis in (b) refer to experimental pressure (GPa)

On the other hand, the correlation of Na cpx/liquid partitioning with temperature has been poorly discussed in the literature. Figure 9b reports a compilation of $D_{{{\text{Na}}_{ 2} {\text{O}}}}^{{{\text{cpx}}/{\text{liquid}}}}$, derived from experiments on fertile peridotites and pyroxenites at 0.7–2.5 GPa (see the figure caption for references). Overall, these data point to a positive correlation of Na cpx/liquid partitioning with temperature (Fig. 9b). Blundy et al. (1995) argued that, at 1 GPa, temperature dependence of $D_{{{\text{Na}}_{ 2} {\text{O}}}}^{{{\text{cpx}}/{\text{liquid}}}}$ is greatly subordinate to that of pressure. In fractional crystallization experiments, Villiger et al. (2007) observed that the Na distribution coefficients between clinopyroxene and liquid decrease at decreasing temperature, as observed in our experiments (Fig. 9b), but they concluded that a clear relationship between Na partitioning and temperature cannot be established. Our results corroborate the existence of such correlation, although a proper understanding of factors controlling such relationship would need a more focused study.

Similar to Na₂O, Al₂O₃ and TiO₂ behave as incompatible elements (Fig. 3), whereas Cr₂O₃ is compatible and mostly partitioned in residual spinel (Table 3; Fig. 3).

Comparison with previous studies

The solidus of GV10 at 1 and 1.5 GPa is located at lower temperature with respect to experimental results on variably fertile lherzolites (Pickering-Witter and Johnson 2000; Schwab and Johnston 2001) and clinopyroxenites investigated by Kogiso and Hirschmann (2001) (Fig. 6). Melting experiments at 1 GPa on these clinopyroxenites have clinopyroxene + olivine as residual minerals (Kogiso and Hirschmann 2001). In our experiments, the stability of residual spinel, in addition to clinopyroxene and olivine, likely resulted from higher Al₂O₃ content of pyroxenite GV10 (Table 1). Lower CaO and higher MgO contents in GV10 could have enhanced minor modal orthopyroxene (Tables 1, 2), never found in clinopyroxenites (Kogiso and Hirschmann 2001).

The solidus curve and the isobaric melt productivity (dF/dT)_P of pyroxenite GV10 are in agreement with the data reported for pyroxenites previously investigated by Lambart et al. (2009a) (Fig. 7). In particular, pyroxenite GV10 displays solidus and melt productivity curves intermediate between M5-40 and M5-103 (Figs. 6, 7).

According to Kogiso et al. (2004a), alkali contents and X _Mg of the bulk composition mostly influence the solidus temperatures of natural anhydrous pyroxenites. On this basis, Lambart et al. (2016) have recently proposed an empirical parameterization to derive the melt fraction versus temperature from 5% melting up to clinopyroxene disappearance for pyroxenites as a function of pressure, temperature and bulk composition. Pyroxenite GV10 has X _Mg and (Na₂O + K₂O) lower than M5-103 (Table 1). The model successfully predicts the relative location of the (dF/dT)_P between GV10 and pyroxenites studied by Lambart et al. (2009a), but locates this curve at slightly higher temperature with respect to the curve extrapolated by the experiments.

The most important differences in the melting relations between GV10 and the pyroxenites M5-40 and M5-103 of Lambart et al. (2009a) are the presence of plagioclase and the abundance of spinel. Both M5-40 and M5-103 have plagioclase in the subsolidus assemblage at 1 GPa, 30 and 4%, respectively, whereas plagioclase is not stable in GV10 pyroxenite. This is due to the much lower Na₂O/CaO ratio of GV10, which tends to stabilize plagioclase at lower pressure (Green and Falloon 1998; Borghini et al. 2010). Spinel is present in very low modal abundance (<1%) or absent in the experiments on M5-40 and M5-103. At 1.5 GPa, spinel is absent in M5-103 but it is present at near-solidus conditions in equilibrium with plagioclase and garnet in the M5-40 (Lambart et al. 2009a).

The variational trends of MgO and FeO abundances in melts at both 1 and 1.5 GPa obtained for pyroxenite GV10 are similar to those of M5-40 and M5-103, with absolute contents mostly varying at intermediate values (Fig. 3), according to the bulk compositions (Table 1). TiO₂ and Cr₂O₃ compositional trends defined by GV10 melts follow those observed for M5-40 pyroxenite, except for the higher TiO₂ contents in GV10 at low F. Progressive Al₂O₃ content decrease with increasing F at both pressures is similar to what observed for M5-103, but with a lower slope, because spinel is a residual phase up to relatively higher F in GV10 pyroxenite. The high CaO contents of GV10 melts mimic those documented in melts produced by clinopyroxenites at 1 GPa (Kogiso and Hirschmann 2001), and by garnet-olivine clinopyroxenite M7-16 at 1.5 GPa (Lambart et al. 2009a) (Fig. 3). At both investigated pressure conditions, the variation of SiO₂ contents in GV10 melts defines a trajectory of rapid decrease at low F and a slight increase at higher F, similar to M5-103 but more accentuated (Fig. 3). This trend results by early exhaustion of orthopyroxene and late disappearance of spinel. Variation trends of Na₂O contents, marked by progressive decrease at increasing F, are similar to those documented for all pyroxenites at 1–1.5 GPa, but partial melts of GV10 display the lowest Na₂O absolute concentrations (Fig. 3), consistent with higher $D_{{{\text{Na}}_{ 2} {\text{O}}}}^{{{\text{cpx}}/{\text{liquid}}}}$ (Fig. 9). Differently from pyroxenites investigated by Lambart et al. (2009a), no partial melts with nepheline-normative compositions are produced by GV10 pyroxenite at 1–1.5 GPa. By contrasts, at low melting degrees GV10 melts have quartz-normative compositions (Table 3).

Major element chemical signature of pyroxenite-derived melts

Lambart et al. (2013) highlighted that the determination of pyroxenite contribution in major element composition of oceanic basalts may be difficult because, at identical pressure and temperature conditions, many pyroxenitic melts are quite similar to peridotitic melts. Although many pyroxenites are expected to have lower X_Mg and higher TiO₂ and alkali contents, these chemical features are diluted in their partial melts by the higher degrees of melting compared to peridotites (e.g., Médard et al. 2006; Lambart et al. 2009a). However, some pyroxenites can produce SiO₂-poor and/or Fe-rich melts that are able to significantly affect the composition of melts produced by a pyroxenite–peridotite mixed mantle source. Accordingly, Lambart et al. (2013) identified a group of primitive MORB that represents the best candidates to preserve this pyroxenite signature, even after the mixing with peridotitic melts (px-MORB; with MgO ≥9 wt%, SiO₂ <48.8 wt% and FeO >8.3). Further major element markers for pyroxenite contribution, namely high CaO/Al₂O₃ ratios and high TiO₂ contents, have been proposed for OIBs (e.g., Prytulak and Elliot 2007; Jackson and Dasgupta 2008; Shorttle and Maclennan 2011). Moreover, primary magma calculations (PRIMELTS2.XLS, Herzberg and Asimow 2008) suggest that low CaO–MgO trends in some OIBs are not consistent with the composition of melts from a pure peridotite and they can discriminate the contribution of melts derived from a low-Ca pyroxenite, as the modeled olivine-free Px-1 (Sobolev et al. 2005, 2007).

To identify the major element signature of secondary pyroxenite GV10, we compare the major element composition of melts produced in our experiments with those of melts from experiments on various peridotite bulks, at 1.0 and 1.5 GPa (Figs. 10, 11, see figure captions for references). We only considered the melt compositions produced at the same temperature interval investigated in this study, i.e. 1200–1330 °C at 1 GPa and 1250–1400 °C at 1.5 GPa.

Fig. 10

Compositions (in wt%) of GV10 partial melts at 1.0 GPa from this study compared with those of experimental melts from variably fertile and depleted lherzolites within the temperature range 1200–1330 °C

Fig. 11

Compositions (in wt%) of GV10 partial melts at 1.5 GPa from this study compared with those of experimental melts from variably fertile and depleted lherzolites within the temperature range 1250–1400 °C

GV10 melts span within almost the same MgO range defined by experimental peridotitic melts (Figs. 10, 11), except for low MgO contents at very low melting degree (F < 2%) and high MgO abundances in melts resulting from F of about 90% at 1 GPa (Fig. 10).

At 1 GPa, SiO₂, TiO₂, FeO and Al₂O₃ abundances of GV10 melts largely match the composition of many peridotite-derived melts. However, at low melting degrees GV10 melts display higher FeO and lower Al₂O₃ contents relative to melts from a normal fertile lherzolite (e.g., MM3, Fig. 10). Peculiar chemical features are the CaO and Na₂O vs MgO trends (Fig. 10). At similar MgO, pyroxenite GV10 tends to produce melts with higher CaO and lower Na₂O contents than peridotite-derived melts (Fig. 10). For high degrees of melting (i.e. the highest MgO abundances), GV10 melts display CaO contents much higher than all fertile lherzolites, but coupled to very low Na₂O contents. Comparably low Na₂O abundances are only observed in melts produced by moderate melting degrees of depleted lherzolites (e.g., DMM1, Wasylenki et al. 2003; Fig. 10).

At 1.5 GPa, GV10 melts resemble those of melts produced by very fertile lherzolites (e.g., lherzolite HK-66, Hirose and Kushiro 1993, or refertilized lherzolites KG1 and KG2 from Kogiso et al. 1998), except for much lower Na₂O contents in the entire pyroxenite melting interval, and slightly lower SiO₂ coupled to higher CaO in high-MgO pyroxenitic melts (Fig. 11). These latter also show higher FeO and slightly lower Al₂O₃ and TiO₂ contents compared to melts from a normal fertile lherzolite (e.g., MM3; Fig. 11).

In summary, as previously observed by Lambart et al. (2009a, 2013), for many oxides the compositions of partial melts from secondary pyroxenite GV10 are similar or very close to those of melts produced by depleted and variably fertile peridotites. The most striking geochemical signature of pyroxenite GV10 melts is their very low Na₂O contents. In addition, they display higher CaO and slightly lower Al₂O₃ concentrations, thus resulting in higher CaO/Al₂O₃ ratio than peridotitic melts. In the next section, we discuss to what extent the GV10 chemical signature can impact the compositions of melts produced by decompressional melting of a mixed pyroxenite–peridotite mantle source at MOR settings.

Secondary pyroxenites in heterogeneous upwelling mantle

Experimental results on secondary pyroxenite GV10 can be used to derive information on the composition of melts produced by adiabatic upwelling and melting of heterogeneous mantle sources. In our modeling, we envisage that mantle portions made by different proportions of pyroxenite GV10 and fertile lherzolite rise along mantle adiabats assuming a range of mantle potential temperatures (T _P) consistent with mantle melting at MOR settings (T _P = 1260–1430 °C; e.g., Langmuir et al. 1992; Shen and Forsyth 1995; Green et al. 2001; Presnall et al. 2002; Herzberg et al. 2003). At this purpose, we applied the recent parameterization of Lambart et al. (2016) that provides a method to simulate the decompression melting of mixed pyroxenite–peridotite mantle sources. Specifically, this model predicts the melt productivity of the two melting components at a given T _P of the rising mantle, as a function of bulk composition and mass fraction of pyroxenite (Lambart et al. 2016), using a modified version of the thermodynamic treatment of adiabatic fractional melting by Phipps Morgan (2001).

In the calculations, we assumed the parameterization of the peridotite solidus temperature proposed by Hirschmann (2000), which predicts slightly higher solidus temperature than Katz’s parameterization adopted by Lambart et al. (2016), and is more consistent with experimental results on fertile lherzolite (e.g., Falloon et al. 2008) (see Fig. 6). Using Lambart et al. (2016) modeling, pyroxenite GV10 starts to melt at higher pressure than peridotites, consistent with experimental results (Fig. 6).

We considered two outermost mantle sources with pyroxenite mass fractions of 5 and 50%. Although the pyroxenite proportion into the mantle is commonly assumed to be around 2–10% (e.g., Hirschmann and Stolper 1996; Pertermann and Hirschmann 2003b), field evidence from orogenic and ophiolitic massifs indicates that the extent of pyroxenites in mantle sequences might be locally significantly higher than 10% (e.g., Gysi et al. 2011). This is for example the case of some outcrops of the External Liguride mantle sequences, in which secondary pyroxenite abundance can reach a proportional ratio of 1:1 to the associated peridotites (Borghini et al. 2013, 2016).

We first envisaged a mantle upwelling scenario along an adiabat with potential temperature (T _P) lower than 1310 °C. At such temperature and 1.5 GPa, pyroxenite GV10 generates low to moderate melt amounts (up to 6.7 and 10.7% for pyroxenite mass fractions of 50 and 5%, respectively), while lherzolite is at subsolidus or incipient melting conditions. Partial melts from low to moderate degrees of melting of GV10 (F < 12%) are Qz- or Hy-normative (Table 3). Therefore, they are expected to react with adjacent unmolten peridotite, producing orthopyroxene-rich reaction zone, as described for eclogite melts–peridotite interaction (Yaxley and Green 1998; Yaxley 2000; Sobolev et al. 2005, 2007). As the ambient peridotite has zero or near-zero porosity and rather higher solidus temperature, these melts should freeze and crystallize at the pyroxenite–peridotite boundary. If pyroxenite-derived melts are able to percolate and impregnate the adjacent peridotite, they are expected to form orthopyroxene- and clinopyroxene-enriched peridotites (e.g., Lambart et al. 2012). Similar peridotites, refertilized by interaction with pyroxenite-derived melts, have been documented in various natural case studies (e.g., Pearson et al. 1993; Rivalenti et al. 1995; Varfalvy et al. 1996; Zanetti et al. 1996; Mukasa and Shervais 1999; Bodinier et al. 2004; Borghini et al. 2013).

Along the adiabatic upwelling of such heterogeneous mantle, this metasomatic process is likely to be continuous, as documented by experimental works on peridotite-eclogite reaction (Yaxley and Green 1998; Rosenthal et al. 2014). Multiple events of reaction and freezing processes introduce lithological zonation at the pyroxenite–peridotite boundary. In the case of reaction between silica-rich partial melts and peridotite, the resulting orthopyroxene-rich borders can act as permeability barrier that prevents melt escaping from pyroxenite (e.g., Kogiso et al. 2004b; Rosenthal et al. 2014). If infiltration of melts from pyroxenite into surrounding peridotite is inhibited, large batch of pyroxenitic melts could be retained in equilibrium with solid residues, isolated from further reaction with peridotite (Lambart et al. 2012). Mechanical instabilities caused by melt accumulation can lead to rapid expulsion of these melts into veins or channels (Kogiso et al. 2004b).

Extraction of melt fractions from pyroxenite GV10 would leave a wehrlitic residue made by a clinopyroxene + olivine (±spinel) assemblage, introducing a range of clinopyroxene-rich heterogeneities in the rising mantle. Their bulk compositions, calculated using the compositions of residual minerals in our experiments, are similar to the bulk composition of clinopyroxenites investigated by Kogiso and Hirschmann (2001) (Fig. 5). Such residual components have Mg-rich minerals and low jadeite contents in clinopyroxene, and their solidus temperature can be significantly increased above that of the ambient peridotite (e.g., Rosenthal et al. 2014; Lambart et al. 2016).

For mantle adiabats with potential temperature (T _P) > 1310 °C, along the pressure interval 1–1.5 GPa, pyroxenite and peridotite are melting simultaneously and the melt productivity of the two components will depend on their mass fractions in the source (e.g., Phipps Morgan 2001). To make some inferences on the contribution of secondary pyroxenite GV10 on major element chemistry of basalts, we computed the compositions of melts produced by mixed pyroxenite–peridotite sources at 1 GPa (Fig. 12). We used the pyroxenite and peridotite “instantaneous” melt fractions provided by the isentropic decompression melting model of Lambart et al. (2016), assuming two pyroxenite/peridotite mass fractions in the source as described above (5 and 50% of GV10), and different T _P of the ascending mantle (Suppl. Figure 3). A fundamental assumption of this approach is that pyroxenite and peridotite behave as two closed systems, i.e. in thermal equilibrium but chemically isolated, until melt mixing and extraction, as assumed in previous melting models (e.g., Lambart et al. 2009a; Shorttle and Maclennan 2011; Sorbadere et al. 2013). This also implies that partial melts remain in equilibrium with their solid residues until extraction.

Fig. 12

SiO₂ versus FeO; Al₂O₃, CaO and Na₂O versus MgO; CaO/Al₂O₃ and TiO₂ versus X _Mg of partial melts produced by decompression melting of two-lithologies mantle source at 1.0 GPa. Melt compositions resulting from the model are compared to the compositions of the most primitive MORB (with X _Mg ≥0.67; Gale et al. 2013). In SiO₂ versus FeO diagram, the field “1” (black dashed line) represent the compositional variability of MORB with MgO ≥9 wt%. Colored numbers refer to the potential Temperature (T _P) adopted in the isentropic decompression melting model (Lambart et al. 2016), to derive the “instantaneous” melt fractions produced simultaneously by secondary pyroxenite GV10 and a fertile lherzolite (see Supplementary Figure 3). Colored dashed lines represent the approximated trend of melt composition variability from mixed sources with 5 to 50% pyroxenite at a given T _P. For the fertile lherzolite we adopted the solidus temperature parameterization proposed by Hirschmann (2000) (see the text for further explanation). Calculations include two heterogeneous mantle sources made by 5% (secondary pyroxenite) GV10 + 95% (fertile lherzolite) MM3 and 50% GV10 + 50% MM3. We derive the mass fraction of pyroxenite-derived melt in aggregated liquids (LF_Py) by the equation LF_Py = Ψ × F _Py/[(Ψ × F _Py) + (1 − Ψ) × F _Per], where Ψ is the mass fraction of pyroxenite in the source, F _Py and F _Per are the melt fractions produced by pyroxenite and peridotite, respectively (Lambart et al. 2009a). The composition of aggregated “instantaneous” melts is provided by the equation: X _mix = (X _Py × LF_Py) + (X _Per × LF_Per), where X _mix, X _Py and X _Per are the weight percentages of a given element in the aggregated melt, pyroxenite melt and peridotite melt, respectively, and LF_Per is the liquid fraction from the peridotite (1 − LF_Py) (Sorbadere et al. 2013). Melt compositions at the given F have been derived from experiments on GV10 (this study) and MM3 (Falloon et al. 1999, 2008)

Major element compositions of melts used in the calculations were derived from experimental results on pyroxenite GV10 (this study) and lherzolite MM3 (Baker and Stolper 1994; Falloon et al. 2008). This latter resembles the composition of External Liguride lherzolites hosting secondary pyroxenite GV10, which were chemically and isotopically identified as a potential MORB-type mantle source (Rampone et al. 1995; Borghini et al. 2013). For each pyroxenite–peridotite melt fraction pair, obtained at 1 GPa for different T _P using Lambart et al. (2016) modeling, we mixed the composition of pyroxenite GV10 and lherzolite MM3 melts using the method proposed by Sorbadere et al. (2013) (see details in caption of Fig. 12). In Fig. 12, the compositions of mixed melts resulting from this model are plotted together with the compositional field defined by the most primitive MORBs. These latter were selected from a broad database recently compiled by Gale et al. (2013), considering only glass compositions with X _Mg ≥ 0.67, i.e. liquids close to equilibrium with mantle olivine. This selection reduced the original dataset of 12,527 compositions to only 184 glass compositions.

The dashed colored lines in Fig. 12 approximate the range of composition of melts generated for different fixed T _P by varying the mass fraction of secondary pyroxenite in the source from 5 to 50%. The increase of secondary pyroxenite GV10 component in the source results in mixed melts with higher FeO, CaO and lower Al₂O₃, Na₂O contents, whereas it only weakly influences the SiO₂ and TiO₂ abundances (Fig. 12). SiO₂ and FeO contents plot far from the composition of px-MORBs (Lambart et al. 2013), and the CaO/Al₂O₃ ratio of mixed melts still varies within the range of MORB (CaO/Al₂O₃ <0.9). MgO contents of mixed melts remain rather constant for low T _P but it tends to significantly decrease, at increasing pyroxenite mass fraction, assuming T _P > 1330 °C (Fig. 12). As expected, at fixed pyroxenite fraction in the source, MgO of mixed melts is positively correlated with T _P, reflecting higher melting degrees of both mantle components (peridotite and secondary pyroxenite).

Melt compositions derived by the model match the variability defined by the most primitive MORB for most of the major element oxides. Melt compositions best fitting the compositional variability of MORBs are those calculated assuming T _P in the range 1280–1370 °C. The major discrepancy is given by the higher X _Mg of mixed melts produced at higher T (>1370 °C), mostly from 5%-pyroxenite source (Fig. 12), which more closely resemble the X _Mg values of peridotite-derived melts. As expected, this mismatch tends to be higher at higher temperature (i.e. at higher melting degrees). However, the X _Mg of melts may be rapidly homogenized to MORB values by early olivine fractionation (e.g., Langmuir et al. 1992), melt-rock reaction within dunite channels (e.g., Kelemen et al. 1995; Asimow and Stolper 1999; Lambart et al. 2009b) and/or reactive crystallization (e.g., Collier and Kelemen 2010). Therefore, the results of our modeling do not allow discriminating the T _P of MORB generation.

To sum up, the results of this model support the postulation that secondary pyroxenite, as the one investigated in this study, potentially represents a hidden component in the MORB mantle source (e.g., Shorttle and Maclennan 2011; Lambart et al. 2013; Rosenthal et al. 2014). A remarkable implication is that, if secondary pyroxenites inherited trace element and/or isotopic enriched signature during their formation by melt-rock reaction processes, their presence in the mantle source could explain the decoupling between “enriched” trace element/isotope and “normal” major element compositions sometimes documented in oceanic basalts (e.g., Salters and Dick 2002; le Roux et al. 2002; Paulick et al. 2010; Zhang et al. 2012). In this context, mantle portions made by embedded secondary pyroxenite and fertile lherzolite as documented in ultramafic massifs (e.g., Borghini et al. 2013), eventually delaminated and incorporated into the asthenosphere, could constitute the source of trace element and isotopic enriched signature in MORBs.

Summary and concluding remarks

We have experimentally determined phase relations, melting conditions and melt and minerals compositions of an anhydrous natural olivine-websterite (GV10), representing a secondary-type pyroxenite, at 1 and 1.5 GPa and temperatures from 1180 to 1400 °C. Residual mineral assemblages are dominated by clinopyroxene and minor olivine (and spinel). Relatively high X _Mg and low alkali content of our starting composition result in solidus temperature only slightly lower than fertile lherzolite. At both pressure conditions, higher modal orthopyroxene relative to clinopyroxene is involved in the melting reaction, thus resulting in the early disappearance of orthopyroxene, and high SiO₂ contents in melts produced by low- to moderate-melting degree (F < 13%). Although for many oxides the compositions of partial melts from secondary pyroxenite GV10 are quite similar to those of melts produced by variably fertile lherzolites, they differ for higher CaO contents, resulting in slightly higher CaO/Al₂O₃ ratios, and very low Na₂O abundances. These latter reflect relatively high partition coefficient of Na between residual clinopyroxene and partial melt, together with the low initial bulk content. $D_{{{\text{Na}}_{ 2} {\text{O}}}}^{{{\text{cpx}}/{\text{liquid}}}}$ values obtained in this study, together with other experimental, define a positive correlation with pressure and temperature.

Profiting by the recent parameterization of Lambart et al. (2016), we have simulated the partial melting of mixed pyroxenite–peridotite sources in the context of adiabatic mantle upwelling. At relatively low potential temperatures (T _P < 1310 °C), low fractions of secondary pyroxenite melts are expected to react and freeze within the subsolidus adjacent peridotite, resulting in orthopyroxene-rich reaction zones (or refertilized peridotite) and high-X _Mg clinopyroxene-rich residues. At higher T _P, secondary pyroxenite and peridotite melt simultaneously. Melt compositions computed by varying the pyroxenite mass fractions in the source (5–50%) largely encompass the chemical variability defined by most primitive MORBs (mostly in the T _P range 1280–1370 °C), supporting the idea that secondary pyroxenites potentially represent a hidden component in MOR mantle sources. Therefore, fertile mantle peridotite veined by secondary pyroxenite may act as a carrier of trace element and isotopic enriched signature in MORBs.