Abstract

The valence of carbon is governed by the oxidation state of the host system. The subducted oceanic lithosphere contains considerable amounts of iron so that Fe3+/Fe2+ equilibria in mineral assemblages are able to buffer the (intensive) fO2 and the valence of carbon. Alternatively, carbon itself can be a carrier of (extensive) ‘excess oxygen’ when transferred from the slab to the mantle, prompting the oxidation of the sub-arc mantle. Therefore, the correct use of intensive and extensive variables to define the slab-to-mantle oxidation by C-bearing fluids is of primary importance when considering different fluid/rock ratios. Fluid-mediated processes at the slab–mantle interface can also be investigated experimentally. The presence of CO2 (or CH4 at highly reduced conditions) in aqueous COH fluids in peridotitic systems affects the positions of carbonation or decarbonation reactions and of the solidus. Some methods to produce and analyse COH fluid-saturated experiments in model systems are introduced, together with the measurement of experimental COH fluids composition in terms of volatiles and dissolved solutes. The role of COH fluids in the stability of hydrous and carbonate minerals is discussed comparing experimental results with thermodynamic models and the message of nature.

The investigation of redox processes and the role of volatiles especially at the slab–mantle interface are crucial for depicting the framework of Earth carbon cycling. During the early stages of the Earth's history, the abundances of carbon and other volatiles in the different reservoirs were determined by the coupled evolution of the terrestrial magma ocean and the primitive atmosphere (Hier-Majumder & Hirschmann 2017). Since the Archean, the efficient deep subduction of organic carbon produced by photosynthesis could have promoted carbon burial in the mantle and an increase of atmospheric levels of oxygen through time (Duncan & Dasgupta 2017). In this picture the oxidation state of the mantle, mainly governed by Fe0 and Fe2+/Fe3+ ratios, and its dynamics played a key role in modulating the fO2 of the Earth's surface, leading to the precipitation of minerals in their oxidized forms such as carbonates (Sverjensky & Lee 2010; Andrault et al. 2018).

Carbon at the modern terrestrial surface is largely divided between carbonates and organic deposits, with a total budget of 1 × 1023 g C, corresponding to about 100 ppm in the upper mantle (Porcelli & Pepin 2014). Recent estimates from volcanic degassing suggest that the carbon content in the modern deep mantle is even higher (c. 263 ppm; Anderson & Poland 2017; Barry 2017). In fact, the fate of carbonates and organic carbon in modern subduction zones is still largely unconstrained, although recent studies suggest that most of the subducted carbon, in the form of carbonates and organic matter, could be recycled back to the surface (Kelemen & Manning 2015).

In this contribution we aim to introduce some basic principles regarding the importance of the use of intensive and extensive variables to define the ‘oxidation transfer’ from the slab to the overlying mantle by C-bearing fluids, and the buffering capacity of the mantle in the carbon speciation at subduction zones, both from the natural and experimental point of view.

Slab-to-mantle carbon transport: the message from nature

The slab–mantle interface is a key location where fluid-mediated element transfer and redox processes occur (e.g. Hayes & Waldbauer 2006; Manning et al. 2013; Debret & Sverjensky 2017). At pressures up to 3 GPa the slab–mantle interface is composed by the mixing of slab and suprasubduction mantle slices in a metasedimentary or ultramafic matrix, to form mélange zones (Bebout 2007; Konrad-Schmolke et al. 2011; Marschall & Schumacher 2012; Bostock 2013; Guillot et al. 2015; Bebout & Penniston-Dorland 2016). The key point of this geodynamic environment is the presence of high fluid fluxes, which allow the chemical exchange within the mélange materials forming hydrated and low-viscosity layers atop the subducting plate (Schmidt & Poli 2014, and references therein). A different scenario is proposed at higher pressures and temperatures, where the formation of silicate supercritical liquids from dehydration and/or melting of the subducted oceanic crust and metasediments induce percolation by porous flow processes. This process leads to the formation of several hydrous phases, such as amphibole and phlogopite or, occasionally, forms near-monomineralic metasomatic rocks such as orthopyroxenite or phlogopitite layers, acting as a filter for aqueous fluid percolation in the suprasubduction mantle, close to the slab–mantle interface (Manning 2004; Hermann et al. 2006; Malaspina et al. 2006; Scambelluri et al. 2006; Spandler & Pirard 2013).

As will be discussed below, carbon plays a major role in the initiation of metasomatic-related redox processes at the slab–mantle interface. It may release large ‘excess oxygen’ during redox–decarbonation reactions, if they occur; also, its speciation in metamorphic fluids may govern fluid-related redox reactions. Carbon may be transported at mantle depth by subduction in different forms and oxidation states. Relevant examples of carbonate-bearing metasediments equilibrated at HP and UHP are provided by eclogitic rocks from the Italian Alps. In his study of impure marbles from the internal Sesia–Lanzo Zone, Castelli (1991) reported a foliation of calcite and dolomite parallel to the eclogitic foliation developed by phengite, quartz, omphacite, grossular-rich garnet, zoisite and Al-rich titanite equilibrated at P > 1.5 GPa and T ∼600°C. Another occurrence, from the Dora–Maira UHP terrane, is represented by impure calcite- and dolomite-bearing marbles from Costa Monforte (Castelli et al. 2007; Ferrando et al. 2017), which show a foliation dominated by carbonates and partial dissolution of dolomite. The best evidence of the dissolution process promoting the carbon transport at depths is provided by the discovery of fluid inclusions in the metamorphosed seafloor Mn-rich sediments from Lago di Cignana (Zermatt–Saas, Italy), containing bicarbonate and carbonate ions, together with crystals of dolomite, magnesite and diamond (Frezzotti et al. 2011). In these occurrences carbon is present ubiquitously as C4+ and C0.

Carbon in its oxidized form is also transported during the subduction of carbonates and elemental carbon in meta-ophiolites equilibrated at HP conditions. One example is provided by the work of Ague & Nicolescu (2014), who reported fluid-mediated decarbonation reactions of metacarbonate layers in contact with serpentinite mélange from the islands of Syros and Tinos (Greece). In these rocks carbonate-rich fluid inclusions occur in HP minerals such as omphacite and glaucophane in altered layers surrounding carbonate veins. Graphite associated with calcite has been detected in the serpentinites of Cogne (Italian Western Alps) (Carbonin et al. 2015). These phases formed by seafloor metamorphism of mantle peridotites, triggered by carbon-bearing fluids during an advanced stage of the opening of the Alpine Tethys in the Late Jurassic (Toffolo et al. 2017). Also, the interaction between metacarbonate rocks and serpentinized mantle occurring at HP has strong implications for the deep transport of oxidized carbon. As shown by Scambelluri et al. (2016), dolomite marbles and serpentinites may react during deserpentinization to form carbonate + olivine hybrid rocks, such as those cropping out in the ophicarbonate unit of the Ligurian Western Alps (Italy). A lower degree of carbonate dissolution in serpentinite-derived fluids at more reducing conditions may lead to precipitation of graphite. One example is shown in the Malaspina outcrop (Alpine Corsica) where the sediments in contact with serpentinites develop a reaction zone of graphitic carbonaceous material in the blueschist facies during the Alpine orogeny (Malvoisin et al. 2012; Galvez et al. 2013). Similar occurrences have been reported by Vitale Brovarone et al. (2017) in samples from the Sesia–Lanzo zone in the Western Alps. In this case the reduction of carbonates in ophicarbonates occurring within serpentinized peridotites induces the precipitation of graphite and the formation of methane- and hydrogen-bearing fluid, preserved as inclusions within the matrix calcite. Finally, nearly pure methane or methane-rich fluid inclusions have been found also in UHP eclogites from Dabie-Shan (Fu et al. 2003) and in jadeitite from Myanmar, in the Himalayan border of China (Shi et al. 2005).

Direct evidence of the transport of carbon-bearing fluids in the suprasubduction mantle and in the mantle wedge are represented by fluid inclusions in peridotites, which are mostly CO2 bearing (Andersen & Neumann 2001; Frezzotti & Touret 2014; Seo et al. 2016). Rarely, fluid inclusions contain methane (Song et al. 2009), suggesting that strongly reduced conditions are not a common feature in the lithospheric mantle. CO2 has been often detected up to 3000 ppm in melt inclusions in olivine from subduction-related igneous rock. However, it is widely accepted that the concentration of CO2 in these inclusions is lower than in their source magmas because of the partial degassing before melt entrapping (e.g. Metrich & Wallace 2008).

Oxidation state of a rock system: some principles

The valence state of carbon and its speciation are governed by the oxidation state of the host system; that is, by the equilibria among mineral assemblages containing redox-sensitive major elements (e.g. Fe, Mn). Alternatively, the carbon species in subducted rocks and in the deep fluids may control the oxidation state of the host system by redox reactions during metamorphic reactions and fluid–rock interactions. As an example, the so-called DCDD decarbonation–redox reaction occurring in subducted metacarbonates, where dolomite reacts with silica to form diopside, carbon and oxygen, has long been considered one of the redox equilibra at the most oxidizing conditions at which diamond can form in eclogites (Luth 1993): 
formula
(1)
A dilemma arises about whether the rock system buffers the redox state of carbon or the carbonate-involved redox reactions define the redox state of the rock. To solve this dilemma one must consider (1) the definition of ‘oxidation state (or redox state) of a rock’ (host system) and the proper use of intensive and extensive variables, (2) the role of ‘perfectly mobile and inert components’ and (3) the degree of fluid–rock interaction at the slab–mantle interface controlling the C-bearing mass transfer.
A redox equilibrium is defined as a process characterized by the flow of electrons from the substance being oxidized (‘reducing medium’) to the substance being reduced (‘oxidizing medium’). For instance, ionic iron in aqueous solutions is present in two valence states, related by the redox equilibrium 
formula
(2)
From left to right we have reduction, from right to left we have oxidation. Because oxygen is the most common electron acceptor in natural systems, in Earth Sciences oxidation and reduction generally mean gain and loss of oxygen, by the exchange with an external medium. To rule this exchange, the variable traditionally considered in Earth Sciences is the oxygen fugacity, fO2 (bar). Because iron is the fourth most abundant element in the Earth's crust, we use oxygen redox buffers considering reactions involving iron-bearing phases (Frost 1991). In the simple system Fe–O–SiO2, as a first approximation the amount of oxygen is the variable that predicts whether iron can be found as native state, as Fe2+ in silicates or as Fe2+ or Fe3+ in oxides: 
formula
(3)
 
formula
(4)
 
formula
(5)
 
formula
(6)
Because in nature solid phases do not usually display pure end-member compositions, most of the redox reactions (including the equilibria (3)–(6)) are not univariant curves, but become multivariant fields. In particular, the replacement of Fe2+ by Mg and Ca, and of Fe3+ by Al and Cr in iron silicates stabilizes the above equilibria to higher fO2 (see fig. 9 of Malaspina et al. 2009). Similarly, the addition of Fe and the consequent dolomite/diopside activities <1, shift the DCDD buffer (equilibrium (1)) to higher fO2 (see fig. 2 of Luth 1993). As already pointed out by Frost (1991), in many rocks it is therefore more appropriate to say that oxygen fugacity is a function of the Fe/Mg ratio (and Ca/Mg) of the rock-forming silicates and carbonates.

Oxidation state of ‘dirty’ rock systems: playing with components

The oxidizing or reducing capacity of a rock is determined by the amount and oxidation state of redox-sensitive major elements, and also by the composition of solid solutions of the mineral assemblages of the rock (Frost 1991). As a consequence, the fO2 is probably very inhomogeneous in a subducting slab, reflecting the different bulk chemical–mineralogical compositions of the slab lithologies. An attempt to picture this concept has been made by Cannaò & Malaspina (2018), who showed a very complex and inhomogeneous fO2 pattern in subduction environments, particularly at the interface between the slab and the overlying mantle. Whereas the subducted oceanic crust records ΔFMQ (= log fO2sample – log fO2FMQ; see equilibrium (5)) between ≤0 and +2.5 up to +4, slices from the slab and from the suprasubduction mantle in the subducted mélange look variously oxidized, from ΔFMQ = −1 to ΔFMQ = +12. In this framework, oxygen fugacity is more probably an intensive variable that is governed by the mineral assemblages in the rock, rather than a variable that is imposed from the environment.

The choice of describing the oxidation state of rock systems in terms of the intensive variables fO2 (bar) and μO2, which are linked by the relation 
formula
(7)
is entirely valid only if O2 can be considered a ‘perfectly mobile’ component (Korzhinskii 1936). In open systems a component is defined as ‘perfectly mobile’ when, during the exchange between the system (e.g. a mineral) and the external medium (e.g. a fluid), the chemical potentials equalize (i.e. they reach equilibrium) in a short time. A component is ‘inert’ when its exchange between the system and the external medium is difficult and the chemical potentials do not equalize. In the case of perfectly mobile components, the independent parameter must be the chemical potential (e.g. μO2), whereas if a component is inert the mass of that component (e.g. nO2) must be considered as the independent parameter. The O2 molar quantity nO2 is the conjugate extensive variable of μO2, similarly to other couples of intensive and extensive variables such as (P × −V) and (T × S) (see details given by Hillert 2008). As shown in a general example by Tumiati et al. (2015), considering the variable nO2 as molar axis, instead of μO2, the univariant curve separating a phase A and a phase B leaves room for a di-variant two-phase field (A + B). This difference can be easily seen in Figure 1, where a comparison of the variables μO2 and nO2 with T is reported for the simple system Fe–O2. Considering the intensive variable μO2 as independent (Fig. 1a), the equilibria between magnetite and hematite (blue curve, equilibrium (6)) and between iron and magnetite (pink curve) are univariant. If the extensive variable nO2 is considered as independent (Fig. 1b), the univariant curves open to di-variant areas, where the relative proportions between magnetite and hematite and between iron and magnetite (horizontal black lines) are ruled by nO2. As a consequence, oxygen can be added or subtracted in oxygen buffer assemblages without changing μO2 (or fO2). The addition of oxygen to a hematite + magnetite assemblage leads thus to an increase in the abundance of hematite, an increase of Fe3+/ Fetot and an increase of nO2, without changing fO2, as long as both minerals are present. The system is therefore buffered at constant fO2 and μO2.

In geodynamic settings where a high fluid/rock ratio is expected, such as subduction mélanges, oxygen is probably transported along fractures and veins, possibly through mechanisms of dissolution–reprecipitation of O-enriched oxides and silicates (Tumiati et al. 2015), or by advective processes (Marschall & Schumacher 2012; Tumiati et al. 2013; Nielsen & Marschall 2017). On the other hand, fluid percolation at low fluid/rock ratios occurs when the metasomatic fluid phases produced at UHP interact with peridotitic rocks at the slab–mantle interface. In such occurrences O2 cannot be considered a perfectly mobile component, because most of the redox reactions take place between solid oxides and silicates, where O2 is bonded. On this principle, the amount of inert components (e.g. FeO then forming Fe2O3 and vice versa) has a fundamental role and the molar quantity of O2 must be considered as an independent state variable. This concept is well shown in Figure 2, where the redox state of the slab–mantle interface at UHP is portrayed in terms of intensive fO2, normalized to the FMQ buffer (Fig. 2a) and of extensive nO2 (Fig. 2b). Figure 2a shows a patchy inhomogeneous redox state of the slab–mantle interface. In this schematic representation three rock systems are considered: (1) slab eclogite, (2) a Grt + Opx-rich layer forming during the reaction of slab-derived supercritical liquids and (3) a suprasubduction metasomatized mantle peridotite. To quantify the redox budget of these three rock systems in terms of extensive nO2 (Fig. 2b), we must consider Fe3+-bearing minerals (i.e. garnet and clinopyroxene) and their contribution in terms of excess O2 with respect to a reference state where only Fe2+ is present. This approach has been proposed by Tumiati et al. (2015) and applied to natural case studies of HP and UHP slab–mantle interaction by Li et al. (2016) and Malaspina et al. (2017). In the following section we will report step by step how to play with these components.

From fO2 to nO2 in ‘dirty’ rock systems and the role of carbon in the slab-to-mantle oxidation front

We consider the above three rock systems as composed by:

  1. eclogite = 50 mol% garnet + 50 mol% clinopyroxene;

  2. Grt–Opx-rich layer = 10 mol% garnet (+ 90 mol% orthopyroxene);

  3. metasomatized peridotite = 5 mol% garnet + 5 mol% clinopyroxene (+ 90 mol% of olivine and orthopyroxene).

We consider also the Fe3+-bearing component in garnet as skiagite (Fe2+3Fe3+2Si3O12) or andradite (Ca3Fe3+2Si3O12) and the Fe3+-bearing component in clinopyroxene as aegirine (NaFe3+Si2O6) or esseneite (CaFe3+AlSiO6). If O2 is not regarded as a phase or species but merely as a ‘stoichiometric’ component expressing Fe3+, it can be made explicit in the skiagite ((FeO)0.59(SiO2)0.35(O2)0.06), andradite ((CaO)0.35(FeO)0.24(SiO2)0.35(O2)0.06), aegirine ((Na2O)0.13(FeO)0.27(SiO2)0.53(O2)0.07) and esseneite ((CaO)0.27(FeO)0.27(Al2O3)0.13(SiO2)0.27(O2)0.07) formulae, rewritten in terms of barycentric molar fractions of oxides and O2 (Table 1). This means that 1 mole of pure skiagite or andradite and of pure aegirine or esseneite would produce 0.06 and 0.07 moles of excess nO2 respectively.

Taking into account the garnet (up to 5 mol% of skiagite) and omphacite (up to 6 mol% of aegirine) composition of a mafic eclogite (Proyer et al. 2004), 1 mole of eclogite provides 3.6 mmol of excess O2; that is, 1 m3 of eclogite contributes for 200 moles of excess O2 (Fig. 2b). The same calculation can be done considering andradite and esseneite as Fe3+ end-members of garnet and omphacite, respectively. The general result will slightly vary (in the error range) owing to the different molar volumes of these two minerals with respect to skiagite and aegirine. Following the above simple calculation, 1 m3 of Grt–Opx-rich layer probably occurring at the slab–mantle interface with a garnet composition of up to 1.5 mol% of skiagite (Malaspina et al. 2017) would contribute 11 moles of excess O2; 1 m3 of suprasubduction metasomatized mantle peridotite such as that from the Chinese Sulu UHP belt, which contains garnet with up to 6 mol% of skiagite and clinopyroxene with up to 5 mol% of aegirine (Malaspina et al. 2009, 2012), would contribute 13 moles of excess O2. As a consequence, as shown in Figure 2, at low fluid/rock ratios mass transfer is supported by a gradient in nO2 and a metasomatic oxidation front probably develops from the oxidized slab to the overlying mantle. The comparison between Figure 2a and b clearly indicates that the equilibrium attainment is difficult, as the oxygen chemical potentials of these different lithologies do not equalize.

The contribution of excess O2 owing to reduction of carbonates during the slab-to-mantle metasomatism can be also broadly estimated. As discussed by both thermodynamic and experimental models (e.g. Gorman et al. 2006; Poli et al. 2009), decarbonation owing to metamorphic reactions is not an efficient mechanism to transfer carbon from the slab to the mantle, because carbonates are stable at PT conditions characterizing subduction zones. Let us assume equilibrium (1) as an ‘extreme’ example of possible redox-induced decarbonation reaction occurring in the slab during subduction. We can use the same approach as described above to model equilibrium (1), making explicit the oxygen produced by reduction of the carbonate ion CO32− in dolomite to C (graphite/diamond). One mole of pure dolomite, whose formula can be written as (CaO)0.17(MgO2)0.17(C)0.33(O2)0.33, would be characterized by 0.33 moles of excess oxygen. Applying equilibrium (1) to a siliceous dolomitic limestone composed by 40 mol% of dolomite (+ 60 mol% of quartz) (Bucher & Grapes 2011), 1 m3 of this rock would contribute 6.5 × 104 moles of excess nO2. This means that if we assume this protolith as a proxy for 1 m3 of subducted metasediment, the (unlikely) complete reaction would produce a quantity of excess O2 two orders of magnitude higher compared with that from oceanic subducted rocks, and three orders of magnitude higher compared with that from the suprasubduction mantle wedge. It must be specified that the potential large contribution in excess oxygen of carbonatic metasediments is inhibited by the wide stability of carbonates at oxygen fugacities close to FMQ (Lazar et al. 2014), characterizing the slab–mantle interface at sub-arc conditions (Fig. 1). However, natural evidence of carbonate reduction to graphite has been provided in slab rocks at shallower and colder conditions (0.9–1.5 GPa and 430°C; Galvez et al. 2013; c. 1 GPa and 370°C; Vitale Brovarone et al. 2017), where oxygen fugacity conditions close to FMQ are expected to be low enough to reduce carbonates (see below).

Effects of slab-derived COH fluids in the metasomatism of mantle rocks

Although, as discussed above, metamorphic decarbonation is not predicted to be an efficient process at pressures and temperatures typical of subduction zones (see also Kerrick & Connolly 2001; Gorman et al. 2006), aqueous dissolution of carbonates (Caciagli & Manning 2003; Pan et al. 2013; Ague & Nicolescu 2014; Facq et al. 2014) and the upwelling of carbonatitic melts originating from the carbonated oceanic crust (Poli 2015) are considered the most promising vehicles to remobilize carbon towards the mantle wedge, along with diapirism of carbonated subduction mélange (Marschall & Schumacher 2012). The oxidation of organic matter and graphite is another efficient mechanism to produce carbon-rich fluids, especially in the presence of silicate and the consequent decrease in water activity (Tumiati et al. 2017), although the total amount of carbon released by this process is poorly constrained owing to the uncertainties associated with the estimated abundance of these phases in subducted rocks (Plank & Langmuir 1998).

The interaction of mantle rocks with carbon-bearing fluids is evidenced by metasomatic assemblages containing carbonates and by carbon-bearing fluid inclusions in mantle minerals. Case studies of slices of metasomatized suprasubduction mantle dragged by the continental crust during subduction and/or exhumation are rare. Among them, we can mention orogenic garnet peridotites and websterites from the Ulten Zone (Italian Alps), Donghai County (Sulu, China) and from Bardane and Ugelvik (Western Gneiss Region, Norway), which experienced metasomatism by ±C-bearing subduction fluid phases up to 200 km depth. Ulten garnet peridotites were metasomatized by slab-derived fluids, which enhanced the crystallization of pargasitic amphibole and dolomite (Sapienza et al. 2009; Malaspina & Tumiati 2012; Förster et al. 2017). Sulu peridotites record a multistage metasomatism by alkali-rich silicate melt, and a subsequent influx of a slab-derived incompatible element and silicate-rich fluid, which crystallized phlogopite and magnesite during the Triassic UHP metamorphism (Zhang et al. 2007; Malaspina et al. 2009). Websterites from Bardane preserve remnants of crust-derived fluids, which precipitated graphite/diamond + dolomite/magnesite + Cr-spinel + phlogopite/K-amphibole in multiphase inclusions hosted by majoritic garnet (Van Roermund et al. 2002; Scambelluri et al. 2008; Malaspina et al. 2010). Peridotites from Ugelvik are interlayered with metre-thick lenses of garnet pyroxenites parallel to the compositional banding. These pyroxenites have been described by Carswell (1968, 1973) and interpreted as crystallized at very high temperatures from mantle-derived melts. They are made of porphyroblastic majoritic garnets in equilibrium with clinopyroxene and K-rich amphibole. Bardane websterites and Ugelvik peridotite–clinopyroxenites experienced the prograde subduction at UHP and early retrograde exhumation history of the Baltica plate margin (Spengler et al. 2009)

The growth of carbonates in mantle rocks can occur only at relatively low-temperature and high-pressure conditions, above the so-called carbonation surface, represented by the reactions 
formula
in the Ca-free systems MS + CO2 (Koziol & Newton 1998), and 
formula
(8)
in the Ca-bearing CMS + CO2 (Wyllie et al. 1983). Tumiati et al. (2013) constrained the dolomite carbonation surface in the more complex hydrous system NKCFMAS + COH at P > 1.5 GPa for T ≈ 950°C, and suggested the following reaction: 
formula
(9)
confirming that dolomite is produced at the expense of clinopyroxene and olivine in the mantle wedge metasomatized by CO2-bearing aqueous fluids.
These reactions are sensitive to the XCO2 (= CO2/(H2O + CO2)) of the fluid phase, as the lower the XCO2, the higher are the pressures required to accomplish it. Moreover, the position of these reactions in the PT field varies as a function of the chemical complexity of the thermodynamic system (e.g. Olafsson & Eggler 1983; Wallace & Green 1988). This is consistent, for example, with the occurrence of clinopyroxene-free, dolomite + amphibole peridotites in the Ulten Zone peridotites (Sapienza et al. 2009), for which the growth of dolomite in equilibrium with orthopyroxene has been estimated at about 2 GPa and 850°C (Malaspina & Tumiati 2012). In fact, the upper stability of dolomite was defined by Brey et al. (1983) in the simple CaO–MgO–SiO2 system by the reaction 
formula
(10)
with magnesite being the only stable carbonate (P > 2.5 GPa at T = 900°C; Tumiati et al. 2013).

To relate the oxidation state of metasomatized suprasubduction mantle and the role of COH and alkali components of the crust-derived metasomatic agents, we report in Table 2 and Figure 3 the Fe3+ measurements of garnets and the calculated fO2 of the above peridotites. An apparent correlation between the composition of the metasomatic agent (C- and alkali-bearing) and the fluid-induced oxidation of the peridotite mineral assemblage probably exists. Where metasomatism was induced by C-free metasomatic fluids, peridotites do not provide evidence of oxidation and garnets have Fe3+/ΣFe ≅ 0.

The composition of COH fluids in equilibrium with elemental carbon (i.e. graphite, diamond) is, on the other hand, dependent on the redox state of the system and on the PT conditions and can be predicted by conventional thermodynamic models, which rely on equations of state that consider molecular species only (e.g. Connolly & Cesare 1993; Zhang & Duan 2009). On the basis of these models, at fixed P and T aqueous fluids become enriched in CO2 in oxidized systems, and in CH4 in reduced systems, passing through intermediate redox conditions where the abundance of dissolved carbon species is minimized and the activity of water is therefore maximized (Fig. 4). The other COH species (CO, hydrocarbons, H2, free O2) have been considered only as minor species at conditions relevant to the upper mantle. However, recent more complex thermodynamic models highlighted the importance of charged carbon species, such as carbonate or bicarbonate ions in relatively oxidized systems and organic dissolved compounds (acetates, formates, propionates) in relatively reduced systems (Sverjensky et al. 2014; Pan & Galli 2016; Tiraboschi et al. 2018).

Whereas the occurrence of carbonate and bicarbonate species has been demonstrated in experiments (Facq et al. 2014), in first-principles molecular dynamic simulations (Pan & Galli 2016) and in natural samples (Frezzotti et al. 2011), organic compounds have not yet been detected in experimental fluids. Moreover, the presence of small Raman peaks ascribable to aromatic species have been interpreted as quench products (Li 2016).

Redox buffered fluid–rock interaction: the experimental point of view

Experiments on mantle rocks interacting with carbon-bearing slab-derived fluids have been conducted for decades to retrieve phase stability and melting relations in metasomatized bulk compositions (e.g. Olafsson & Eggler 1983; Wallace & Green 1988). Recent experiments focused on the composition of the equilibrated metasomatizing fluid in terms of dissolved volatile and solute species (Tumiati et al. 2017; Tiraboschi et al. 2018). Challenging issues of these last experiments are related to the difficulty in extracting fluids avoiding back-reactions in the fluid phase during quench. Further complexities are related to the low amount of fluid present in millimetre-sized experimental capsules, and the tendency of water to condense on the tubing systems, preventing quantitative analyses. In situ experiments, carried out in hydrothermal anvil cells, could bypass most of these problems, but are to date limited to relatively low temperatures and very simple chemical systems (e.g. Facq et al. 2014; McCubbin et al. 2014). Ex situ experiments are more versatile, because a wide range of PTX conditions can be explored using, for instance, piston cylinders and multi-anvil apparatuses. Moreover, in these experiments the fugacity of oxygen can be controlled using the double-capsule technique (Eugster & Skippen 1967), and the volatiles in the fluid phase quenched to room T and P (which, however, do not necessarily represent the species in the fluid at the experimental run conditions) can be analysed by mass spectrometry, a technique that provides great sensitiveness to volatile COH species. The first analyses of experimental COH fluids were provided by Eggler et al. (1979), who investigated the solubility of CO and CO2 in various silicate melts at 3 GPa and 1700°C. Recent attempts to measure volatiles in experimental capsules have been described by Dvir et al. (2013), who used an IR gas analyser suitable for oxidized H2O–CO2 fluids, and by Tiraboschi et al. (2016), who used quadrupole mass spectrometry to analyse quantitatively small amount of fluids down to 1 μmol with uncertainties of about 1 mol% for molecular H2O, CO2, CO, CH4, H2 and O2. With this technique, Tumiati et al. (2017) demonstrated that whereas the composition of COH fluids in equilibrium with graphite at 1–3 GPa and 800°C is consistent with conventional thermodynamic models (Connolly & Cesare 1993; Zhang & Duan 2009), the addition of either quartz or magnesium silicates (forsterite, enstatite) enhances the CO2 content in high-pressure fluids by 10–30 mol%. In relatively complex systems, therefore, C–O–H models are not fully adequate to predict the composition of slab-derived fluids in terms of dissolved volatiles, and more complex models including dissolved charged species are needed (Sverjensky et al. 2014; Galvez et al. 2015; Pan & Galli 2016). However, these thermodynamic models still rely on a very limited experimental dataset, so more data on the solubility of minerals in COH fluids would be extremely useful. To date, only few papers have been published on this topic. Newton & Manning (2000) employed the weight-loss technique (Anderson & Burnham 1965) to retrieve the quartz solubility in H2O–CO2 fluids, suggesting that it decreases by increasing the CO2 fraction in the fluid (Newton & Manning 2009). Tumiati et al. (2017) and Tiraboschi et al. (2018) used a modified version of the so-called cryogenic (or freezing) laser ablation inductively coupled plasma mass spectrometry technique (Kessel et al. 2004, 2005) to retrieve the solubility of quartz, forsterite + enstatite and enstatite + magnesite in H2O–CO2 fluids in equilibrium with graphite. In contrast to what is observed in graphite-free fluids, the solubility of minerals in the presence of graphite is largely enhanced, presumably because organic Si- and Si–Mg–bearing compounds can form at relatively reduced conditions (Tiraboschi et al. 2018).

The fluid phase is therefore somewhat elusive in experiments. In contrast to solid phases, which can be analysed routinely with well-established techniques (e.g. electron microprobe, X-ray diffraction), the analysis of experimental fluids is particularly challenging. Nevertheless, as for any other phase, gaining the composition (and the speciation) of fluids is of fundamental importance and it can be addressed only by combining different experimental approaches and analytical techniques.

Conclusions

The correct use of extensive and intensive parameters in the definition of the redox state of the system depends on whether oxygen can be considered a perfectly mobile component. In natural systems, at low fluid/rock ratios, oxygen is inert and its molar quantity (‘excess oxygen’ or the ‘redox budget’ of Evans 2012) must be considered. Alternatively, at high fluid/rock ratios such as in channelized flow regimes and in fluid-saturated experiments, the independent parameter is fO2. Awareness of this distinction helps to solve the dilemma about whether the valence state of carbon and the speciation of its compounds is governed by the equilibria among mineral assemblages containing redox-sensitive major elements (e.g. Fe, Mn), or the carbon species in subducted rocks and deep fluids control the oxidation state of the host system by redox reactions during fluid–rock interactions.

At redox states fixed by the FMQ buffer, the composition of C-bearing fluids following conventional thermodynamic models is predicted to be a H2O + CH4 mixture at low T and low P, nearly pure water at low T and high P, and H2O + CO2 at high-T conditions. The occurrence of carbon species in different oxidation states may not be directly related to a drastic change in fO2 (i.e. CH4 reduced conditions–CO2 oxidized conditions), but all the variables P, T and fO2 must be considered simultaneously. As shown in Figure 4, graphite/diamond-saturated COH fluids can be methane-bearing at FMQ conditions (‘oxidized’ condition in the mantle) at UHP and relatively low temperature.

Finally, the dissolution of silicates controls the composition of deep COH fluids in equilibrium with graphite, even in the absence of carbonates, boosting the dissolution of graphite in subduction environments at high fluid/rock ratios (mélange) in the form of volatile CO2.

Acknowledgements

We thank R. W. Luth and an anonymous reviewer for their constructive comments. Discussion with S. Poli in the early stage of the paper and the editorial assistance of I. M. Villa significantly helped.

Scientific editing by Igor Villa

This is an Open Access article distributed under the terms of the Creative Commons Attribution 3.0 License (http://creativecommons.org/licenses/by/3.0/)