Super-reduced mineral assemblages in “ophiolitic” chromitites and peridotites: the view from Mount Carmel

Ophiolites are fossil lithospheric fragments, which can be generated during different stages of a Wilson cycle (e.g., Bodinier & Godard, 2014). Their mantle sections are composed mainly of peridotite residues after melt extraction and other ultramafic rocks (e.g., dunite, chromitite) related to melt-peridotite interaction. The redox states of these lithospheric-mantle fragments range from ∼1 log unit above the quartz–fayalite–magnetite (QFM) buffer to ∼4 log units below QFM (e.g., Stagno et al., 2013). The normal oxygen fugacity (fO₂) of the lithospheric mantle, as sampled in xenoliths, decreases with depth, due to the effect of pressure (P) on Fe³⁺–Fe²⁺ equilibria in upper-mantle minerals (e.g., Frost & McCammon, 2008; Luth & Stachel, 2014). All these estimates are more oxidizing than the iron–wüstite (IW) buffer, which marks the saturation of Fe metal in the upper mantle (e.g., Frost & McCammon, 2008).

Ultrahigh-pressure (UHP) phases and assemblages (summarized in Table S1) also have been separated from or observed in situ in chromitites and peridotites from the Tibetan and Polar-Urals ophiolites, which also contain the SuR assemblages (e.g., Bai et al., 1993; Robinson et al., 2004; Yang et al., 2007, 2015a and b; Dobrzhinetskaya et al., 2009, 2014; Yamamoto et al., 2009; Xu et al., 2013, 2015; Howell et al., 2015; McGowan et al., 2015; Griffin et al., 2016a;,Zhang et al., 2016a). These UHP phases and textural evidence suggest derivation from depths of at least ∼300 km, and probably from the Mantle Transition Zone (e.g., Satsukawa et al., 2015; Griffin et al., 2016a;,Zhang et al., 2017).

Previous studies have treated the SuR phases together with the UHP materials, as being all derived from the very reduced Mantle Transition Zone or the deep levels of the upper mantle (Robinson et al., 2004; Yang et al., 2007, 2014, 2015a and b; Dobrzhinetskaya et al., 2009; Zhang et al., 2016a). However, there are only two unambiguous cases showing that UHP phases enclose SuR phases: one is the association of coesite in stishovite pseudomorphs, cubic BN, and a high-P polymorph of TiO₂ with aPbO₂ structure, with a SuR suite of native Fe, Ti, TiN and boron carbide (Yang et al., 2007; Dobrzhinetskaya et al., 2009, 2014); the other is the inclusions of Ni–Mn–Co alloys in diamond (e.g., Howell et al., 2015). Other SuR phases, such as moissanite, native metals, carbides, nitrides and silicides (Table S1), are found in mineral separates and have no confirmed textural connection with the UHP materials (e.g., Robinson et al., 2004; Yang et al., 2014, 2015a and b; Xu et al., 2015).

Recently, an assemblage of SuR phases, including the first terrestrial occurrence of tistarite (Ti₂O₃) and a suite of native metals, carbides, nitrides, silicides and amorphous carbon (Tables S1, S2), has been found in corundum xenocrysts ejected from Cretaceous volcanoes in Mount Carmel, northern Israel (Griffin et al., 2016b). These SuR assemblages show clear similarities with the SuR suite from the Tibet and Polar Urals ophiolites. In this study, we use the Mt Carmel SuR assemblages to explore the genesis of the ophiolitic analogs in Tibet and the Polar Urals from a different perspective, and propose that the SuR phases in the ophiolites may not be collectively formed together with the UHP materials in the Mantle Transition Zone or the deep upper mantle.

The UHP–SuR mineral associations have been found mainly in the chromitites and peridotites of the Luobusa ophiolite from the Yarlung Zangbo Suture Zone (south Tibet, China; e.g., Yang et al., 2014; Zhang et al., 2016a) and the Ray-Iz ophiolite from the Ural orogenic belt (Russia; e.g., Yang et al., 2015a). The Luobusa ophiolite (Fig. 1) is one of many in the Yarlung Zangbo Suture Zone, which extends ∼2000 km from west to east and separates Greater India from the Lhasa Block (Asia) in the north. The Yarlung Zangbo ophiolites represent “fossil” lithospheric fragments from the Neo-Tethyan oceanic lithosphere, which was mainly consumed during the India–Asia convergence in the Mesozoic/Cenozoic (e.g., Hébert et al., 2012). The Ray-Iz ophiolite (Fig. 1) is one of the ultramafic bodies in the early Paleozoic Uralian orogen, a ∼3000 km long, N–S-trending collision belt that separates Europe from Asia (e.g., Chemenda et al., 1997). The Uralian ophiolites represent relict fragments of the Western Uralian oceanic lithosphere, thrust onto land during the early Paleozoic collision of the European and Asian continents (e.g., Batanova et al., 2011; Shmelev, 2011).

The Luobusa ophiolite (including the Luobusa, Xiangkashan and Kangjinla areas, ∼70 km²) is tectonically sandwiched between Triassic flysch sediments in the south and the Gangdese Arc batholiths or the Tertiary Luobusa conglomerate belt in the north. From north to south, it is comprised of (1) a serpentinite mélange zone with gabbroic and basaltic blocks or dykes (zircon U–Pb ages of ∼130–120 Ma; Zhang et al., 2016b), (2) a dunite-facies zone with mainly high-Cr# [molar Cr³⁺/(Cr³⁺ +Al³⁺)] dunite and chromitite in a harzburgite framework, (3) a clinopyroxene-poor harzburgite zone with high-Cr# dunite and chromitite lenses and gabbro dykes (zircon U–Pb ages of ∼160–150 Ma and ∼130–120 Ma; Zhong et al., 2006; Chan et al., 2015; Zhang et al., 2016b), and (4) a clinopyroxene-rich harzburgite or lherzolite zone with rare lenses of low-Cr# dunite and disseminated chromitite. Previous studies have proposed that the Luobusa ophiolite initially formed in a mid-ocean-ridge spreading center, and was then modified in a supra-subduction-zone (SSZ) environment (e.g., Zhou et al., 1996, 2005).

The Ray-Iz ophiolite is a block of ∼400 km², set in a mélange belt sandwiched between sediments from the European continent and the eastern early Paleozoic arc complexes. From north to south, it comprises (1) a clinopyroxene-rich harzburgite or lherzolite zone, (2) a dunite-bearing harzburgite zone, which also contains economic chromitite deposits, and (3) a dunite–clinopyroxenite–gabbro complex (e.g., Shmelev, 2011). A wide range of Neo-Proterozoic to Devonian ages has been reported for the ophiolite (e.g., Batanova et al., 2011). Previous studies have suggested that the clinopyroxene-rich harzburgite or lherzolite zone in the Ray-Iz ophiolite was generated at a mid-ocean ridge, and the dunite- and chromitite-bearing harzburgite zone in a forearc spreading region (e.g., Batanova et al., 2011; Shmelev, 2011).

The UHP–SuR mineral assemblages in the Luobusa ophiolite have been separated from, or observed in situ, in the high-Cr# chromitites and harzburgites from the dunite-facies zone and the Cpx-poor harzburgite zone, and have been studied over ∼30 years (see references above). In the early studies, tons of chromitite, harzburgite and dunite were crushed for mineral separation, resulting in uncertainty about the relationships between individual UHP–SuR phases and with their host rocks, and raising the possibility of contamination during processing. In the last 10 years, in situ observations of UHP–SuR phases with clear spatial contexts with chromite, olivine, PGE alloys or Fe–Ti alloys have confirmed their natural origin (e.g., Yang et al., 2007, 2014; Dobrzhinetskaya et al., 2009; Yamamoto et al., 2009; Satsukawa et al., 2015; Zhang et al., 2016a).

In the Ray-Iz ophiolite, Yang et al. (2015a) have reported diamond and CF-type chromitite similar to those in the Luobusa ophiolite, and coesite in the diamond. The SuR phases are similar in nearly all respects to those from the Luobusa ophiolite (Table S1), suggesting that the two ophiolites may have experienced comparable tectonic processes and mantle-cycling dynamics.

Two basic models have been proposed to explain the genesis of the UHP–SuR association in the ophiolitic chromitites and peridotites. One proposes that the UHP and SuR phases were encapsulated in chromite grains that crystallized from high-Cr magmas, in the Mantle Transition Zone or the lower part of upper mantle, and then migrated upwards through a mantle plume (Yang et al., 2007, 2014), or in upwelling mantle driven by the rollback and breakoff of a subducting slab (Zhou et al., 2014; Zhang et al., 2016a). The other model proposes that both the UHP chromitites and their enclosing peridotites originally formed at shallow depths, but were subducted into the Mantle Transition Zone, and later exhumed back to shallow depths (Arai, 2013). McGowan et al. (2015) and Griffin et al. (2016a) showed how slab rollback would develop a focused upwelling channel that would bring the harzburgite massifs and their chromitites to the surface on timescales of 6–10 m.y.

These models have treated the SuR phases as forming together with the UHP materials in the deep, reduced upper mantle or Mantle Transition Zone. However, most of the UHP phases do not require low fO₂ conditions, and only a few of the SuR phases require high P (see above and Table S1). Are the UHP and SuR phases necessarily related? Can other mechanisms introduce such low-fO₂ assemblages into the ophiolitic chromitites and peridotites? How could these SuR assemblages survive in the more oxidizing ophiolitic environment? These questions are still enigmatic. Here we use a remarkable, recently recognized mineral system from Mt Carmel, Israel, to suggest an alternative genesis for the ophiolitic SuR phases.

Cretaceous pyroclastic vents (volcanics) and related alluvial deposits in the Mt Carmel-Kishon River area of northern Israel (southern Galilee; Fig. S1 in Supplementary Material) have recently yielded abundant xenocrysts of corundum, moissanite, native Fe and a range of highly reduced phases (Griffin et al., 2016b and c). This area lies within a complex system of faults, related to the Africa–Arabia plate boundary that evolved into the Dead Sea Transform, active since Miocene time. The early development of the Jurassic–Early Cretaceous Levant magmatic province coincided with major rifting across Africa, but no major events of rifting (or subduction) are correlated with the sparse Early Cretaceous continental magmatism in Israel. This magmatism (137–139 Ma; e.g., Segev, 2009) comprises subalkaline to alkaline basalts, nephelinites, and basanites with hotspot/mantle plume geochemical and isotopic signatures (Garfunkel, 1989; Stein & Hofmann, 1992, 1994; Laws & Wilson, 1997). The Cretaceous volcanics were covered by Eocene and younger formations, and were exposed by uplift and block-faulting, with subsequent erosion.

Late Cretaceous (94–98 Ma, Turonian-Cenomanian) volcanic activity in northern Israel took place in the Mt Carmel–Umm El Fahm area (Fig. S1). The dominant volcanic rocks are pyroclastics of mafic to ultramafic composition, erupted in a shallow marine environment (Sass, 1980). Black (fresh) pyroclastics occur in eruptive vents, whereas variegated pyroclastics occur as layers interbedded with carbonates. Lapilli consist both of volcanic glass (now cryptocrystalline) and microporphyritic mafic to ultramafic lavas. Analyses of the latter show that the magmas were tholeiitic picrites (Griffin et al., 2016b). The pyroclastic deposits carry mantle-derived xenoliths and xenocrysts (spinel peridotites, garnet ± spinel pyroxenites, garnet websterites), crustal garnet– ilmenite granulites, and megacrystic Mg-ilmenite, amphibole and clinopyroxene; P–T estimates on garnet websterites suggest maximum conditions of ca. 2.3 GPa and 1030 °C (Esperanca & Garfunkel, 1986; Mittlefehldt, 1986; Apter, 2014; our unpublished data).

Over the last decade, the Shefa Yamim exploration project, focused on placer gem deposits, has recovered xenocrysts of corundum (sapphire, ruby and “non-gem” corundum) and moissanite (SiC). Aggregates of “non-gem corundum” (“Carmel Sapphire”; hereafter NGC) up to >24 cts in weight have been recovered. Crystals of moissanite can be over 4 mm long. Sampling of the Cretaceous pyroclastic rocks on Mt Carmel has yielded moissanite, all types of corundum, and the suite of minerals described below. The distribution of NGC in the drainage system suggests that it, and other minerals discussed here, are derived mainly from the Cretaceous pyroclastics. Analyses of rocks and minerals (with crystallographic data) are given by Griffin et al. (2016b and c); analytical methods are briefly described in the Supplementary Material.

The CL images (Fig. 6) reveal that most NGC xenocrysts are aggregates of corundum crystals. The distribution of melt pockets and the related dark CL show that the crystals are skeletal, with intricate, crystallographically controlled internal cavities and the hollow faces typical of hopper growth (Fig. 6), which has provided a structure in which abundant melt pockets could be trapped. Electron backscatter diffraction (EBSD) mapping reveals that the aggregates consist of domains in which multiple individual crystals have very similar crystallographic orientation, while individual domains are misoriented to various degrees (Griffin et al., 2016b). This clustering of similarly oriented crystals also is typical of hopper growth.

The melt pockets in the NGC can be divided into four types: Type S (silicate glass + oxides), Type A (alloy), Type N (nitrides + borides of Ti), and Type DF (SiO₂-free assemblages with fluorides). Types S, A and N appear to have been mutually immiscible during much of the corundum growth; they are found together in some melt pockets, or in separate pockets, which may be linked in three dimensions (Fig. 4b). They contain a remarkable array of crystalline phases, many requiring very low fO₂; more than half have not previously been reported as minerals, or are known only from meteorites (Table S2).

Type-A pockets are found both along grain boundaries and within corundum crystals. The two most abundant phases are TiC (khamrabaevite) and Fe₃Si (gupeiite) (Fig. 8), but the pockets also contain a range of Fe–Ti–Si– P alloy phases (Table S2). In many cases these are intergrown in patterns that suggest the unmixing or crystallization of originally homogeneous alloy melts. The detailed characterization of these phases is in progress.

The Mt Carmel occurrences are notable for their abundance of SiC, which occurs as crystals up to 4.14 mm long; this is much larger than moissanite reported from kimberlites, ophiolites (typically ≤200 μm) or xenoliths (DiPierro et al., 2003;,Di Pierro & Gnos, 2016; Trumbull et al., 2009; Shiryaev et al., 2011; Liu et al., 2015; Xu et al., 2015). Moissanite has been observed together with TiC, TiSi₂ and other phases in two melt pockets in the NGC, providing a physical link between these highly reduced mineral assemblages. The occurrence of Ti–V alloys and Ca–Al–Fe silicides as inclusions in the Mt Carmel SiC links them to the Type-DF assemblage described above (Table S2), and distinguishes them from the moissanite from kimberlites (Shiryaev et al., 2011) and ophiolites.

Some larger NGC fragments are breccias, with angular clasts of the NGC embedded in a matrix composed dominantly of amorphous carbon (Figs. 5, S2). It is mixed with finely dispersed corundum, various phases common in the melt pockets, SiC and Fe sulfides. Some samples contain abundant grains of cryolite (Na₃AlF₆) and an unnamed K–Al fluoride. Transmission electron microscope (TEM) images show that some corundum grains and melt pockets also are cut by parallel-sided veins of amorphous carbon, ranging from several tens of micrometers to ≤50 nm thick (Fig. S4). This observation, and the lack of detectable Cl (a component of epoxy resin), make it unlikely that the amorphous carbon is an artifact of sample preparation. The microstructures suggest brittle fracture, turbulent mixing and injection of carbon-rich fluids that also contained high levels of S and F; these processes may be linked to the explosive eruption of the host magma.

Perhaps the best constraint on P and T during corundum growth is provided by the occurrence of anorthite as a quench phase (Fig. 7), and more rarely as a euhedral phase, in the glass pockets. Anorthite melts congruently up to 9 kbar; at higher P, it melts incongruently to corundum + melt, and is unstable above 30 kbar (Goldsmith, 1980). The appearance of anorthite as a late phase in the residual melts is consistent with the peritectic reaction corundum + melt → anorthite; this suggests that most corundum crystallization occurred at 9–30 kbar (ca. 30–100 km depth) and ∼1450–1550 °C.

Other constraints on T are broadly similar. The large spinel grains in the melt pockets commonly have excess Al₂O₃, consistent with high-T crystallization in the presence of corundum. The maximum Al₂O₃ in Ti-free spinels is 77.5 wt%, implying crystallization at ∼1400 °C at 1 atm (Hallstedt, 1992). The peritectic reaction corundum + liquid → hibonite (Fig. 10) occurs at ∼1790 °C in the CaO–Al₂O₃ system, but extends down to 1475 °C in the 1-atm CaO–Al₂O₃–SiO₂ system.

Gupeiite (Fe₃Si), a common phase in the Type-A melt pockets, crystallizes from reduced melts at 1200–1300 °C at 1 atm (Kubaschewski, 1982). Tistarite (Ti₂O₃) crystallizes from Ti–O melts at 1700–1870 °C, and Fe₂P from 1166– 1370 °C in the binary system; FeTiSi is on the liquidus in the ternary system from ∼1530°C to ∼1200 °C (Weitzer et al., 2008). Most of the experimental data come from the ceramics and metallurgy literature, and were obtained at low P; they probably give minimum Testimates. In addition, the presence of volatiles (C, N, F) may have lowered liquidus temperatures. However, it seems clear that the corundum, and the associated moissanite, crystallized at high (magmatic) T. The moissanite associated with these phases clearly is not a product of low-T processes, as argued by Schmidt et al. (2014).

The low fO₂ of the assemblages in the melt pockets is defined by the presence of SiC, Ti₂O₃, TiO and native V, and the abundance of reduced phases such as TiN and TiC, both with Ti²⁺ (our unpublished EELS data) and Fe₃Si. These assemblages indicate fO₂ between 8 and 11 log units below the IW buffer (Fig. 3). The absence of Fe, Ni, Mn or Cr in the silicate and oxide phases in the melt pockets suggests that fO₂ was at least below the IW buffer, consistent with the removal of Fe as immiscible melts (Fig. 11) by the time the skeletal/hopper NGC aggregates began to grow.

While many aspects of the Mt Carmel volcanic system are not yet understood, the pyroclastic material does at least provide context at a grain (mm–cm) scale, allowing definition of parageneses and sequences. A simplified model of the processes involved has been constructed, linking the observations available at this time.

The growth of corundum as hopper or skeletal crystals implies rapid crystallization from a melt supersaturated in Al₂O₃; the regular arrangement of the crystals in the aggregates suggests growth on a substrate, rather than the random accumulation of phenocrysts from a magma. The quench structures in the melt pockets (Fig. 7) and the metal-oxide melts (Fig. 11) suggest that this process was part of an active magmatic/hydrothermal system that was terminated by the explosive eruption of the Cretaceous volcanic rocks. The key question becomes: how can Al-supersaturation, and the implied disequilibrium, be achieved in a magmatic system within the upper mantle (30–100 km depths)?

We suggest that such conditions might be imposed on magmas in conduits or magma chambers, and perhaps on the immediately adjacent mantle, if rising asthenosphere-derived magmas were accompanied by significant proportions of deep-mantle fluids. Numerous investigators (e.g., Frost & McCammon, 2008; Rohrbach & Schmidt, 2011; Stagno et al., 2013) have argued that the sublithospheric convective mantle (>250 km depth) is metal (Fe)-saturated; if this is correct, then any C–O–H fluids at these depths will be totally dominated by CH₄ ± H₂ (e.g., Zhang & Duan, 2009). The passage of such fluids through the more oxidized peridotitic mantle might be sustained by the reduction of the wall-rocks of fluid conduits, leading to removal of FeO, and/or the plating out of FeO-free phases on the walls of the conduit or magma chamber.

Reactions (2) and (3) are well known from industrial-scale applications, in which the partial oxidation of methane is used to generate hydrogen and “syngas” (e.g., Konnov et al., 2004; Lemke et al., 2005; Lyubovsky et al., 2005).

The progressive reduction and desilication of a mafic–ultramafic magma, leading ultimately to the rapid precipitation of corundum from an Al₂O₃-supersaturated magma, can be accounted for by infiltration of CH₄ ± H₂ fluids into a magma chamber (with initial fO₂ ≥ QFM), or in a magma-conduit situation.

In the first stage of the process, fO₂ in the magma is lowered to the magnetite–wüstite buffer (Fig. 3), leading to precipitation of the abundant, commonly vesicular, wüstite lapilli found in the ejecta. The CO₂ released by the reduction of magnetite would react with the magma to precipitate the observed aggregates of high-Mg dolomite, removing some Ca and Mg; this reaction can proceed only until fO₂ drops below the EMOD buffer (Fig. 3). It is noteworthy that the matrix of the SiC-rich xenolith described by Di Pierro et al. (2003; see below) consists of brucite, calcite and magnesite, which might represent low-T breakdown products of higher-T magmatic carbonates.

The precipitation of SiC would lead to desilication of the magma; if rapid enough, this would drive the magma into supersaturation in Al₂O₃, as required for the rapid hopper/skeletal growth of the corundum (NGC) aggregates. The porous structure of the resulting crystal mass could allow for the passage of volatiles, while helping to isolate the precipitated phases from back-reaction with the magma. The hydrogen produced by the simplified reactions (2) and (3) is immiscible with water under mantle conditions (Bali et al., 2013). Solution of the water in the remaining magma, and flushing of hydrogen through the crystal framework, could help drive down fO₂ within that framework still further, culminating in the precipitation of the highly reduced phases in the melt pockets. The Type-DF and Type-N melts may represent the end-members of this fluid evolution, in terms of both desilication and lowering of fO₂, and the concentration of volatiles such as N, F and B.

Reactions (2) and (3) would precipitate abundant carbon, which may be represented in the carbon-rich veins and breccias. These reactions also produce CO₂ and H₂O, which could contribute to volatile overpressure, and thus to explosive eruption, fragmenting and ejecting the crystallized material that grew in the magma chamber or along the walls of the magma conduits.

This scenario is clearly speculative at this stage of the research, but provides several hypotheses that can be tested. The next stage will require the detailed mapping of the different types of mineral assemblages and their 3D and 4D relationships, the characterization of the large number of previously unrecognized phases, and stable-isotope data to constrain the sources of the fluids.

Moissanite is generally regarded as rare in nature, but it may be more abundant than commonly realized. It has been separated from kimberlites and lamproites (Shiryaev et al., 2011) and from kimberlite-borne eclogite xenoliths (our unpublished data), and reported as inclusions in diamonds (e.g., Moore & Gurney, 1989; Leung et al., 1990). A summary of the literature is provided by Di Pierro & Gnos (2016).

Prior to the Mt Carmel discoveries, the largest natural crystals of moissanite (1–2 mm) were reported from a beach cobble found in Turkey (Di Pierro et al., 2003; Di Pierro & Gnos, 2016). The source of the cobble is not known, but it is believed to come from Tertiary volcanics in the region. The moissanite occurs together with ca. 25 vol% of SiO₂, with minor phlogopite and pyroxene, in a fine-grained matrix of brucite, magnesite and calcite. Moissanite crystals contain inclusions of native Si, Fe-silicides and Ca–Al–Si phases, some rich in REE, which the authors interpret as formed by the partial decomposition of high-P phases.

The recent (from 2012) explosive eruptions of the Tolbachik fissure on the Kamchatka peninsula produced mafic pumice carrying abundant microdiamonds, similar in morphology and C-isotope composition to the Tibetan diamonds (Gordeev et al., 2014; Karpov et al., 2014). The diamonds are associated with moissanite, Ti-rich corundum, and “iron oxide globules”, providing a possible link to the Mt Carmel association. Native metals (Fe, Cu and Au) have crystallized in pumice vesicles, suggesting that the gases exsolved from the melt created a fO₂ at or below the IW buffer. Karpov et al. (2014) noted that similar associations of native metals and SiC are frequently found in the effusive products of volcanoes along the Kuril-Kamchatka volcanic arc, and Kaminsky et al. (2016) cataloged occurrences of diamond in Kamchatka lavas and dykes.

The Avacha volcano in Kamchatka carries polycrystalline microdiamond with dispersed SiC (Gorshkov et al., 1995; Kaminsky et al., 2016). These diamond aggregates contain inclusions of Mn–Ni–Fe-silicides and W and B carbides, indicating a very reduced environment. Gorshkov et al. (1995) argue that the diamond grew on a fine-grained matrix of SiC, through the reaction of Si or quartz with “carbon-bearing magmatic gases”. Kaminsky et al. (2016) prefer a substrate of metal carbides and silicides, with SiC and Si appearing at a late stage in the evolution of the diamond aggregates, in a low-P environment analogous to diamond production by chemical vapor deposition.

Spinel-peridotite xenoliths from the same volcano contain native Fe and Ni, and Fe silicides (Ishimaru et al., 2009) indicative of a local reducing environment in the sub-arc mantle. The reduced phases are small and clearly out of equilibrium with their relatively oxidized host, and even with one another. The authors interpret this disequilibrium, and microstructural evidence, as showing that the reduced phases were deposited through interaction between the host and highly reduced C–H fluids injected shortly before eruption. They also note that the native Fe is free of Ni, which is the case in both the ophiolitic peridotites and chromitites, and most of the Mt Carmel “Fe balls”.

Some key similarities and differences in the phase associations reported from the localities are obvious in Table S1. The most striking difference is that no UHP phases have been recognized in the Mt Carmel samples; this is consistent with the presence of anorthite in the Type-S melt pockets, since crystallization of anorthite is limited to depths ≤∼100 km (Goldsmith, 1980).

The most striking similarities are in the presence of high-T SuR phases, notably moissanite with metallic melt inclusions (Fe-silicides), in all four localities. This is accompanied in each instance by Ti-rich corundum, a key feature of the Mt Carmel association. TiC (khamrabaevite), TiN (osbornite) and possibly the TAZ phase are reported as inclusions in Ti-rich corundum from both the Tibetan and Polar-Urals localities, as well as the Mt Carmel samples. Ti-rich corundum is also reported from the Tolbachik lavas (Karpov et al., 2014), but the inclusion data are not yet available. Ishimaru et al. (2009) report “possible native Ti” in an Avacha xenolith, but the analytical sum is 72.5 wt%, suggesting that this phase may be TiC. Tistarite has not been recognized from Kamchatka, nor from the Tibetan or Uralian localities. However, Xu et al. (2013) give analyses of “TiO₂” inclusions in corundum, with levels of MgO and other minor elements similar to those seen in the Mt Carmel (and Allende) tistarite (Griffin et al., 2016b). Since oxygen apparently was not analyzed separately by Xu et al. (2013), these analyses may in fact represent tistarite.

The Fe balls with wüstite rims described from Tibet and the Polar Urals (Fig. 2d) are similar in structure to those seen in the Mt Carmel samples, although the oxide rims on Fe balls at Mt Carmel apparently represent more complex oxide melts, rather than simply wüstite (Fig. 11). Fe-oxide balls are reported from the Tolbachik volcano, but there are only surface analyses and no information is available on the internal structure.

Breccia veins with amorphous carbon host the in situ diamonds and SiC reported from Tibet and the Polar Urals (Fig. 2a, b; Yang et al., 2014, 2015a and b). Such breccia veins are common in the Mt Carmel corundum aggregates (Fig. 5), and polycrystalline microdiamond with amorphous carbon is reported from Avacha (Kaminsky et al., 2016). Carbon appears to have played an important role in the development of the SuR associations in all four localities.

However, the carbides of Fe and W, and the native Ni, Co and PGEs, found in the chromitites and peridotites of both Tibet and the Polar Urals, and the volcanic rocks of Kamchatka (cohenite, (Fe,Ni,Co)₃C; Gordeev et al., 2014) are not recognized in the Mt Carmel material. This may reflect a key difference in their respective situations: in the former localities, the reducing fluids appear to have interacted with wall-rock peridotites, as well as with the more evolved mafic melts proposed for the Mt Carmel system.

One common factor among these various localities is the presence of deep-seated mantle rocks or volcanism, related either to deep subduction and slab rollback (Kamchatka, Tibet, Polar Urals), or to the rise of “hot-spot” magmas (Mt Carmel; Stein & Hofmann, 1992, 1994). Another, related commonality is their tectonic setting on plate boundaries (Pacific subduction, continental-collision zones; the Dead Sea Transform). These major translithospheric structures provide the channels along which deep-seated rocks, magmas and deep-mantle fluids (CH₄ ± H₂) may rise and mix. If this is the case, further investigation of the trace minerals carried by explosive magmas in such situations should turn up evidence that such processes are more widespread than currently recognized.

Based on the inspiration of the Mt Carmel example, we speculate that the SuR phases observed in the Tibetan and Polar-Urals ophiolites probably did not experience a long residence time in the convective upper mantle or the Mantle Transition Zone, although some UHP materials may have done so. The un-aggregated nitrogen in the Tibetan ophiolitic diamonds (Howell et al., 2015; Yang et al., 2015b), the coexistence of SiC and FeO-bearing olivine and chromite (e.g., Zhang et al., 2016a), the intergrowth of nitrides, native Fe and Ti with oxides and silicates in a reaction rim around a partially unmixed Fe–Ti alloy (Yang et al., 2007; Dobrzhinetskaya et al., 2009), and the transition-metal alloys with complex unmixing textures (Xu et al., 2015; Yang et al., 2015a and b; Fig. 2d), all suggest that these phases were produced and then quenched during a rapid dynamic process, which could freeze-in the disequilibrium assemblages and textures and isolate them from the ambient convective mantle.

These features are consistent with the exhumation model described by McGowan et al. (2015) and Griffin et al. (2016a), in which the UHP chromitites and peridotites were brought to the surface in a focused upwelling channel, rooted in the Mantle Transition Zone and induced by slab rollback. Continuous influx of reduced C–H–O fluids (CH₄ ± H₂) buffered by the metal-saturated convective mantle (Rohrbach & Schmidt, 2011) could react with chromitites and peridotites and drive down the local fO₂ to at least the IW buffer. The formation of SuR phases could begin deep in the convective mantle (Dobrzhinetskaya et al., 2009), continue through the diamond-stability field (Howell et al., 2015), and climax with ascent into the shallow lithosphere. The short time scale for the exhumation of the deep-seated chromitites and peridotites is supported by numerical modeling, which shows an ascent time of less than a few million years from the Mantle Transition Zone to the lithosphere (McGowan et al., 2015; Griffin et al., 2016a). The implied rapid cooling at shallow depths would facilitate the preservation of SuR–UHP associations in the shallow lithospheric mantle.

The interaction of CH₄ ± H₂ fluids with the solid phases of the chromitites and peridotites during ascent can account for the coexistence of some UHP and SuR phases in the ophiolites. It can explain the differences between the Tibet and Israel suites, but does not obviously account for their similarities; specifically, the association of Ti-rich corundum, related SuR phases and SiC were preserved both in the Mt Carmel and the ophiolitic samples. To complete this story, it is required that the same types of fluids (CH₄ ± H₂ from the deep convective mantle) interact with mafic magmas under relatively low-P conditions, leading to desilication, reduction, supersaturation in Al₂O₃, and the final production of SiC and amorphous carbon. These melts might have been generated by decompression melting in the asthenospheric component of the upwelling channel, and migrated upwards with the exhumation of buoyant harzburgites (including chromitites). A Mt Carmel-type evolution could then generate and emplace the main suite of SuR phases into the recycled chromitites and harzburgites, which subsequently accreted as part of the new oceanic lithosphere and eventually became the ophiolites in the Tibetan and Polar-Urals collision zones.

The analogies to the Kamchatka volcanism described above offer another possibility: the derivation of CH₄ ± H₂ fluids from the subducting slab (e.g., Ishimaru et al., 2009), as rollback initiated the ascent of the harzburgites and chromitites in Tibet and the Polar Urals. Analyses of fluid inclusions in mantle-wedge peridotites (e.g., Song et al., 2009) have been ascribed to this mechanism. The serpentinization of slab and mantle-wedge peridotites could also supply CH₄ ± H₂ fluids to the recycled chromitites and harzburgites at shallow depths, and facilitate the formation and stability of SuR phases in the protoliths of the Tibetan and Polar-Urals ophiolites.

The low-P (<3 GPa) assemblage of SuR phases found as xenocrysts in Cretaceous volcanic rocks on Mt Carmel, Israel provides unique insights into the development of localized highly reduced volumes in the uppermost mantle. The mineral system is interpreted to reflect the interaction of mantle-derived CH₄ ± H₂ fluids with the associated hot-spot basaltic magmas. These interactions led to desilication of the magma, supersaturation in Al₂O₃ with rapid growth of abundant Ti-rich corundum, and phase assemblages requiring fO₂ as low as IW-11. There are strong similarities between this system and the SuR phases and associated Ti-rich corundum in ophiolitic peridotites and chromitites in Tibet and the Polar Urals, and with less well-known SiC–diamond–corundum assemblages in volcanoes on the Kamchatka peninsula. We suggest that the ophiolitic SuR mineral suites probably reflect interaction between CH₄ ± H₂ fluids and previously subducted peridotites (including chromitites) during their exhumation from the deep upper mantle or Mantle Transition Zone, overprinted by a shallow magmatic system similar to that observed at Mt Carmel. While the Tibet and Polar-Urals occurrences and those on Kamchatka involve large-scale deep subduction systems, there is no evidence of subduction associated with the Mt Carmel volcanics. The common factor among these occurrences appears to be the presence of CH₄ ± H₂ fluids. We suggest that this type of fluids may commonly accompany deep-seated volcanism or magmatism, especially near plate boundaries and other lithosphere-scale discontinuities, and that SuR conditions may be more common than generally considered (though local and probably transient) in the upper mantle. This flux of CH₄ ± H₂ fluids represents a previously unrecognized aspect of the global carbon cycle.

We thank Steve Craven and Manal Bebbington for help with difficult sample preparations, Will Powell for guidance with LAM-ICPMS analysis, Laure Martin (UWA) and Xian-Hua Li (IGG-CAS, Beijing) for isotopic analyses by SIMS, Martin Saunders with the TEM images, and Shi-Lin Li for preparation of Fig. 1 in this manuscript. We also thank Brian Marker (Chairman of the IUGS Publications Committee) and Hui Chen (Executive Editor of Acta Petrologica Sinica) for their permission to reprint Fig. 2b and c, respectively. Dorrit Jacob and Steven Foley contributed insightful discussions about oxygen fugacity and mineral growth. This study used instrumentation funded by ARC LIEF and DEST Systemic Infrastructure Grants, Macquarie University and industry. This is contribution 968 from the ARC Centre of Excellence for Core to Crust Fluid Systems (www.ccfs.mq.edu.au) and 1153 from the GEMOC Key Centre (www.gemoc.mq.edu.au).