Aqueous fluids are effective oxidizing agents of the mantle in subduction zones


Aqueous fluids produced by dehydration of the downgoing slab facilitate chemical exchange in subduction zones, but the efficiency of fluid-mediated redox transfer as a mechanism to deliver oxidized material from the slab to the sub-arc mantle remains hotly debated. Here we report the first direct measurements of the oxidation state of experimentally produced slab fluids using in situ redox sensors. Our experiments show that the dehydration of natural antigorite serpentinite at shallow subduction zone conditions (1 GPa, 800 °C) produces moderately oxidizing fluids (QFM + 2) with elevated concentrations of Na, K, Ca, and Mg. The composition and redox of the experimental fluids are then used to parameterize a thermodynamic reactive transport model to investigate the interaction of slab fluid with the sub-arc mantle from 1–4 GPa and 700–900 °C. Recently determined equation of state parameters for aqueous fluids at high pressures now enables thermodynamic modeling of aqueous fluid–rock interactions at conditions relevant to deep subduction zones for the first time. Our thermodynamic modeling demonstrates that aqueous fluid can efficiently oxidize Fe in mantle minerals via the reduction of H+ to H2 in the fluid. We estimate that < 1–3 kg of serpentinite-derived fluid at 850–900 °C is required to increase the Fe3+/ΣFe in 1 kg of sub-arc mantle from MORB-like values (0.15) to those of primitive arc basalts (0.2–0.3). We calculate that a slab fluid flux of 1.4 × 109–1.4 × 1014 kg year−1 is required to oxidize sufficient sub-arc mantle to produce the average annual flux of magmas at arcs, which overlaps with the estimated range of H2O flux in subduction zones.


Subduction zones are the principle conduits through which oxidized materials are recycled to the deep Earth and provide the largest global flux of volatiles from the Earth’s interior to the atmosphere. Chemical exchange within subduction zones is mediated by the migration of aqueous fluids initially sourced from Earth’s oceans and sequestered in hydrous minerals within oceanic lithosphere. During subduction, elevated temperatures and pressures cause the breakdown of hydrous minerals and the continuous release of aqueous fluids into the overlying mantle wedge up to ~ 2 to 4 GPa and 800 °C (Fig. 1) (e.g., Ulmer and Trommsdorff 1999; Poli and Schmidt 2002). Released “slab fluids” react with sub-arc mantle and are the driving force behind hydrous mantle melting via the lowering of the mantle solidus temperature (e.g., Grove et al. 2012).

Fig. 1

Schematic drawing of a subduction zone illustrating serpentinite and chlorite stability after Till et al. (2012). Isotherms are taken from the geodynamic models of Grove et al. (2009) using a dip of 30° and a convergence rate of 40 mm year−1. Light blue dashed region indicates the continuous dehydration of serpentinite in the downgoing slab and porous flow of fluid leading to alteration in the wedge. Dark blue arrows indicate region of largest fluid release and transport via channelized flow to regions of arc mantle melting (orange region). Magenta boxes contain redox values for relevant lithologies and fluids: serpentinites (Debret et al. 2014, 2015; this work), slab fluids (this work), and arc magmas (Kelley and Cottrell 2009). Hexagons indicate modeling depths and are color coded to match Fig. 4

Within the last 10–20 years, the scientific community has become increasingly interested in the role of oxygen (or more precisely, redox potential) in geologic processes. In particular, there has been a major focus on quantifying the transfer of redox potential within subduction zones because of the strong link between life and atmospheric O2 and the observation that lavas and gases erupted at subduction-related arc volcanoes are the most oxidized on Earth. Primitive arc lavas have Fe3+/ΣFe from 0.18–0.32 (Kelley and Cottrell 2009), analogous to fO2 ca. 1–2 log units above the quartz-fayalite-magnetite oxygen fugacity buffer (~ QFM + 1–2), whereas MORBs, which are sourced from dry mantle melting, have Fe3+/ΣFe ~ 0.15, analogous to an fO2 of ~ QFM (Fig. 1). In contrast, the oxidation state of volcanic gases may be more directly linked to their emission temperature rather than the oxygen fugacity of their source (e.g., Moussallam et al. 2019).

The question of how arc lavas obtain this oxidized signature is a topic of extended debate. Is this signature inherited from an oxidized mantle source (e.g., Wood et al. 1990; Parkinson and Arculus 1999; Kelley and Cottrell 2009; Brounce et al. 2014) or internally generated during hydrous magmatic differentiation in the lithosphere (e.g., Lee et al. 2005)? Obtaining direct measurements of sub-arc mantle oxygen fugacity has proven elusive, and estimates from mantle xenoliths vary widely, from QFM-1 to QFM + 4 (Brandon and Draper 1996; Lee et al. 2005; Kelley and Cottrell 2012; Brounce et al. 2014), which led to the question of whether the sub-arc mantle is oxidized at all relative to MORB-source mantle.

More recently, a plethora of observational, experimental, and model-based studies point toward an oxidized sub-arc mantle, calling upon aqueous fluids or slab sediment melts to carry redox potential from the slab to the center of the mantle wedge, where the majority of sub-arc mantle melting takes place. Measurements of exhumed slab lithologies have demonstrated that serpentinites, key hydrous mineral assemblages in the subducting slab, are continuously dehydrated and reduced during subduction, from Fe3+/ΣFe = 0.5–1 in oceanic serpentinites to < 0.4 in high-grade antigorite serpentinites (Debret et al. 2014, 2015; Groppo and Castelli 2010; this work). The strongest pulses of fluid release occur at key phase transitions, including chrysotile-out, lizardite-out, and antigorite-out (potentially of greatest importance to mantle melting), and chlorite-out (John et al. ; 2011; Alt et al. 2012; Padrón-Navarta et al. 2013). These observations have been confirmed by experimental investigations (Merkulova et al. 2016, 2017), which suggest that derivative aqueous fluids should be highly oxidizing, with precise oxidation potential controlled by Fe3+ (i.e., magnetite abundance) in the protolith. The concentrations and isotope ratios of some fluid-mobile elements such as B, Be, U, and Th have been used to track the movement of fluids from the ocean to the slab, and then to arc magma (e.g., Hawkesworth et al. 1997; Kendrick et al. 2011; Kessel et al. 2005; Scambelluri and Tonarini 2012; Tenthorey and Hermann 2004). In particular, experiments on U and Th solubility in aqueous fluids require oxidizing slab fluids to explain U/Th signatures in arc magmas (Bali et al. 2010). Recent thermodynamic modeling indicates the generation of slab fluids close to the hematite-magnetite oxygen fugacity buffer (< QFM + 5) during the reactive breakdown of antigorite to olivine, enstatite, and chlorite (Debret and Sverjensky 2017).

Experimental investigations into magmatic processes suggest that crustal differentiation and degassing processes do not alter (Waters and Lange 2016), or even tend to reduce rather than oxidize arc magmas (Kelley and Cottrell 2012), highlighting the need for an oxidized mantle source region. Petrologic modeling of V, Sc, and Ti partitioning between mantle minerals and primitive mantle melts (Wang et al. 2019) suggests that the arc mantle should be ~ 10 times more oxidized than MORB-source mantle to produce observed arc magma compositions. All of this evidence implies that slab fluids may be efficient transporters of redox potential into multiple regions within the mantle wedge and suggests that the sub-arc mantle may be a significant subsurface oxygen reservoir.

So far, experimental studies that reproduce slab fluid generation (e.g., Ulmer and Trommsdorff 1995; Bromiley and Pawley 2003; Padrón-Navarta et al. 2010; Merkulova et al. 2016) have focused on the composition and redox of reacted solid products rather than on measurements of fluid directly. In addition, our ability to model the thermodynamics of redox transfer in subduction zone fluids has, until very recently, been stifled by a lack of data at high pressure. Therefore, the identification of precise reactive mechanisms during subduction zone relevant fluid–rock interactions has been impossible. In this study, we use a novel experimental procedure to measure fluid redox state in situ during antigorite dehydration experiments and take advantage of very recent advances in high P–T fluid thermodynamics to model the chemical interaction (including redox reactions; e.g. Debret and Sverjensky 2017) of our experimentally produced slab fluid with mantle lithologies, to test three key questions: (1) What is the redox state of slab fluids? (2) Can slab fluids alter the redox state of mantle minerals at subduction relevant P and T? (3) If so, what is the oxidant?

Experimental methods

Natural serpentinite starting material

For these experiments, it was necessary to obtain relatively pristine (i.e. free of obvious carbonate veins and retrograde alteration) volatile-bearing serpentinite starting material representative of the hydrated ultramafic lithologies that deliver slab fluids into the melt-producing regions of the sub-arc mantle. Antigorite serpentinites represent peak pro-grade conditions, and, with the ability to host up to ~ 13 wt.% structurally bound H2O, likely make the most significant contribution to slab fluids during their decomposition at ~ 1 to 3 GPa and 700–800 °C (Schmidt and Poli 1998; Ulmer and Trommsdorff 1995; Rüpke et al. 2004; Scambelluri et al. 2004; Kendrick et al. 2013). Such rocks are well represented at the surface in exhumed terrains such as the high-pressure serpentinites of the Ligurian Alps (Scambelluri et al. 2004; Cannaò et al. 2016).

Natural antigorite serpentinite from the Western Alps (near the Village of Vara, Voltri Massif) was chosen as starting material for this study (sample VAR10-01; Table 1; Cannaò et al. 2016) due to the sample’s preservation of prograde and peak structures and lack of carbonate veins, and retrograde alteration. Sample VAR10-01 is a deformed mylonitic serpentinite with the main foliation formed by antigorite plus magnetite and minor chlorite and is thought to represent the lithology that delivers the deepest slab fluids into the sub-arc mantle (Cannaò et al. 2016). Figure 2 shows the whole-rock composition of our chosen starting material relative to 186 natural serpentinite whole-rock samples reported in EarthChem/Georoc (; Sarbas and Nohl 2008). The H2O concentration is estimated to be 11.5 wt.% in the bulk rock based on loss on ignition (ibid). Mass balance calculations indicate the starting material is 97 wt.% antigorite and 3 wt.% magnetite, and so the H2O concentration in the antigorite is estimated to be 11.9 wt.%. The Fe3+/ΣFe in VAR 10–01 bulk (0.38) and antigorite (0.28) was determined for this study using Mössbauer spectroscopy. The Mössbauer measurements were performed at Argonne National Laboratory with a WISSEL Spectrometer, operating in constant acceleration mode. The source was 5 mCi 57Co/Pd. The spectrometer was calibrated with standard iron foil. The sulfur concentration (0.1 wt.%) was measured in our starting material powder via infrared absorption at ActLabs.

Table 1 Serpetinite starting material compositions
Fig. 2

Whole-rock composition of starting material VAR 10-01 compared to 186 natural serpentinite whole-rock samples reported in the literature in wt% (; Sarbas and Nohl 2008)

The Fe3+/ΣFe values presented are derived from WinNormos analysis with four different sites, two magnetic and two non-magnetic. The magnetite results are in agreement with literature values in terms of magnetic hyperfine field and isomer shifts. In a study of serpentine polymorph antigorite (Bishop et al. 2008), the results for isomer shift and quadrupole splitting for antigorite are very similar to our finding. However, the relative amounts of each site are sample dependent as well as sample preparation dependent and need not necessarily match. These inconsistencies were well described by Murad (1998).

Sliding binary-alloy redox sensors

To measure the oxygen fugacity of fluids released during the breakdown of the serpentinite starting material, NixPd1−x sliding binary alloy redox sensors were used following the sensor design and calibration of Taylor et al (1992), including the pressure-corrected calibration of Pownceby and O’Neill (1994). Binary alloy and oxide redox sensors have been used since at least the 1960′s (Carapezza 1966), but only more recently have they been well calibrated and tested (Taylor et al. 1992; Pownceby and O’Niell 1994; Righter and Hauri 1998; Stagno and Frost 2010; Matjuschkin et al. 2015). Such sensors are based on well-known oxygen buffer equilibria (e.g., nickel–nickel oxide) and consist of a metal (here, Ni), corresponding metal monoxide (NiO), and a second, diluent metal (Pd), which alloys with Ni resulting in a Ni activity of the system below unity. In the presence of H2O, a relatively small mass of sensor assemblage (e.g., Ni–NiO–Pd) does not buffer experimental fO2 but instead will respond to the ambient fO2 by adjusting the metal alloy composition to be in equilibrium with the pure metal monoxide. In a system more oxidizing than the starting sensor composition, NiO is produced from the reaction of Ni metal with O scavenged from co-existing aqueous fluid, lowering the Ni concentration in the alloy. In a more reducing system, Ni is produced from the breakdown of NiO, increasing the Ni concentration in the alloy. The fO2 of the sensor is related to the fO2 of the sample through diffusive equilibration of H2 and the dissociation constant of H2O. In the presence of some diluent metal and (nearly) pure H2O fluid, the proportion of Ni within the alloy shifts depending on the fH2 of the system, with excess reactive metal being converted to or from its companion monoxide. If the diluent forms a complete solid solution with the reactant metal, as is true for Ni and Pd, the mixture has the capacity as a sliding fH2 sensor in which the mole fractions of both components of the binary alloy respond to and record the ambient oxidation state.

Sensors were prepared by mixing finely ground analytical reagents (99.95–99.999% purity). Prior to serpentine dehydration experiments, we also conducted experiments that employed a sensor alongside a sample composed of a common buffer material (NNO, HM) to confirm that sensors were reacting in an expected way to the fO2 of their environments (see Sects. 2.3 and 4.2).

Experimental design

It was necessary to develop an experimental capsule design that allows for communication between the fluid and redox sensor while isolating the sensor from direct contact with the metal capsule walls or the solid starting material and experimental products. Zirconia inner capsules and lids to house redox sensors were made by slicing 4.5 mm-diameter zirconia ceramic rod into ~ 2 mm-thick pucks and thin lids. Channels to accommodate redox sensors were hand drilled into zirconia pucks using a tungsten carbide microdrill bit. Preliminary experiments utilized alumina capsules rather than zirconia, but electron dispersive spectroscopic analysis of the sensors post-experiment revealed that Ni-spinel crystallized within the sensors, suggesting a reaction between sensor and button material. No reaction was observed with zirconia capsules, and all experiments reported here used zirconia inner capsules. Au outer capsules were chosen to minimize diffusion of hydrogen out of the experiment during the experiment. Large volume 5 mm-diameter capsules allowed for the maximum mass of serpentinite starting material to be loaded relative to the small mass of the sensor, while still fitting into existing experimental assemblies. Au capsules were annealed in a box furnace at ambient fO2 conditions before each run.

Between 30 and 60 mg of VAR 10–01 starting material powder was loaded into Au outer capsules along with NixPd1-x redox sensors encased in milled zirconia ceramic inner capsules (Fig. 3; Table 2). Early experiments employed a ~ 1 mm NiPd metallic pellet surrounded by excess NiO. Later experiments employed “pre-mixed” sensors, where Ni, Pd, and NiO were mixed into a homogeneous powder then loaded into zirconia inner capsules and pressed with a drill blank. A zirconia lid was placed atop the zirconia capsule, and then extra zirconia powder was packed around the inner capsule and lid and pressed flat with a drill blank. VAR 10–01 powder was then loaded into the capsule.

Fig. 3

Schematic of the experimental capsule design. Binary alloy sliding redox sensors were placed inside zirconia sleeves and covered by a zirconia cap. The zirconia sleeve and cap were then placed into an Au capsule, and serpentinite starting material was filled into the remaining void space. Control experiments used Zr (OH)4 as a source of H2 O instead of serpentinite

Table 2 Experimental run conditions and electron microprobe measurements of post-run redox sensors

Serpentinite-free experiments (“control” experiments) were conducted to measure the intrinsic oxygen fugacity of the experimental apparatus and assembly and to ensure that results from serpentinite-bearing experiments reflect the redox potential of antigorite breakdown fluids rather than the oxidizing potential of the experimental setup. Control experiments were constructed in essentially the same manner as serpentinite-bearing experiments. Instead of serpentinite powder, Zr(OH)4 powder, which breaks down to produce ZrO2 + 2H2O around 550 °C, was added to the capsule as a source of H2O, necessary to allow redox sensors to react. A comparison of time series for both experimental series allows for the determination of (and potentially correction for) any rate-limited reaction occurring within sensors.

Once filled, capsules were weighed, triple crimped at their open end, welded shut using a PUK TIG welder with tungsten electrode under Ar gas, and then subsequently re-weighed to check for any mass loss or gain during welding. As a secondary check to ensure capsules were completely sealed, welded capsules were submerged in acetone for ~ 5 min and then reweighed. Any capsules that showed a mass gain after acetone were assumed to have leaked and were discarded. Capsules were enclosed in solid media assemblies sheathed in a graphite furnace enclosed within a BaCO3 sleeve (surrounded by Pb foil). MgO rods were used as spacers above and below the capsule such that the capsule was vertically centered relative to the furnace. Free space on all sides of the capsule was filled with Pyrex glass powder since the low permeability of H though Pyrex has been found to minimize H2O loss from experimental capsules, which can also affect oxygen fugacity (Matjuschkin et al. 2015 and references therein). An alumina disk was placed above the capsule to prevent contact between capsule and thermocouple.

Experiments were carried out at 1 GPa and 800–900 °C (i.e., above the serpentinite-out phase boundary at sub-arc mantle conditions; Table 2; Fig. 4) in a 3/4″ solid-medium Kennedy-style piston-cylinder device in the Experimental Petrology and Igneous processes Center (EPIC) lab at Arizona State University (ASU). Experimental PT conditions were chosen to meet the criteria of: (1) being at depths where subducting lithospheric mantle dehydration occurs (~ 1 to 3 GPa; e.g., Schmidt and Poli 1998); and (2) to be within the range of conditions for which binary alloy redox sensors have been calibrated and tested (up to 1 GPa; e.g., Taylor et al. 1992; Matjuschkin et al. 2015). Slab dehydration thought to supply fluid to regions of sub-arc mantle melting probably occurs predominantly from 2 to 3 GPa in subduction zones, while fluids produced at lower pressure flow into the mantle wedge nose. This highlights the need for calibration of redox sensors at higher pressures. The pressure medium for the piston-cylinder experiments was sintered BaCO3, which was found to have no pressure correction through calibration against the albite = jadeite + quartz and Ca-tschermak pyroxene = anorthite + gehlenite + corundum reactions (Hays 1966; Longhi 2005). Pressures are thought to be accurate to within ± 0.05 GPa. Temperature was monitored using W–Re (Type D) thermocouple, with a temperature correction of 5–40 °C, calculated for each experiment independently depending on the geometry of the capsule and assembly. The degree of correction was determined based on spinel growth temperature calibration experiments after Watson et al. (2002) and were performed over a ca. 10 cm vertical length centered roughly on the hot spot. Temperature was controlled with a Eurotherm 2416 to ± 20 °C. Experimental assemblies were first pressurized to 1.0 GPa at room temperature and held at this pressure while the sample was heated at a rate of 100 °C/min. Experiments were held at the desired P and T between 1–72 h and then quenched by turning off the output power.

Fig. 4

Phase diagram illustrating key reactions, including the serpentinite-out phase boundary (thick solid line; Ulmer and Trommsdorff 1995) and the chlorite-out phase boundary and H2O-saturated mantle solidus (Till et al. 2012). Modeled slab surface P-T paths from Syracuse et al. (2010) are shown in colored dashed lines, with cold subduction zones in cool colors and hot subduction zones in warm colors. Colored dots illustrate the P-T conditions used in our fluid-rock modeling, with the overall modeling space colored in light gray

Analytical methods

Analysis of solid experimental products

Quenched experiments were weighed, punctured, placed in an oven for up to 1 h, and then weighed again to obtain a first-order measure of the mass of volatiles produced during the experiment. A subset of experiments was set aside for fluid composition measurement by ICPMS. All successful experiments showed a weight loss after puncture corresponding to 12–14 wt.% of the mass of starting material powder added to the capsule, in agreement with theoretical estimations for antigorite breakdown and our starting material’s measured loss on ignition (LOI) of 11.5 wt.% (Table 1). Capsules were then peeled open, and redox sensors and post-experiment sample powders were collected. Redox sensors were mounted in epoxy, polished, and carbon coated for chemical analysis via energy-dispersive (EDS) and wavelength-dispersive (WDS) spectrometry on a JEOL JXA-8530F microprobe in the LeRoy Center for Solid State Analyses at ASU. Probe current was 25 nA, voltage was 25 keV, and the beam was set to a focused beam size of 1 µm. EDS element mapping was used to observe heterogeneities within the redox sensors and target sites for analysis with WDS. Several (30–40) individual probe analyses were performed on each sample.

The phase assemblages of post-experiment sample powders and of VAR 10-01 starting material were analyzed using X-ray diffraction (XRD). Some experimental products were also mounted on carbon tape for visual inspection under a scanning electron microscope, and the compositions of mineral phases were measured using EDS on the microprobe.

Fluid analyses by ICP-MS and ion chromatography

For some serpentinite-bearing experiments, experimentally produced antigorite breakdown fluids were collected and their compositions were measured by inductively coupled plasma mass spectrometry (ICP-MS) and ion chromatography (IC). Post-experiment capsules were carefully cleaned and weighed. Experimental fluids were then extracted by first submerging post-experiment capsules in 5 or 10 ml of ultrapure (18.2 MΩ) distilled H2O. The exact mass of H2O was carefully measured to calculate a dilution factor for each individual sample. Once submerged, capsules were punctured with cleaned sharpened stainless-steel tweezers such that sample fluid would be released and mixed into the ultrapure H2O. The capsule was then removed and set aside for other analyses. To mitigate the precipitation of Fe and other insoluble species, ICP-MS samples were filtered directly into a solution containing 100 µl of HNO3 and 25 µl of HCl. Blanks were created by submerging pre-punctured capsules (“capsule blanks”) and/or tweezers (“blanks”) in ultrapure H2O and then acidifying. Samples were not acidified for IC.

Major and trace element concentrations were determined using Q-ICP-MS (Thermo Fisher Scientific ICAP-Q) in the W. M. Keck Foundation Laboratory for Environmental Biogeochemistry at ASU. Samples were analyzed using a multielement internal standard (Sc, Ge, Y, In, Bi) and in-house multielement calibration standards prepared via gravimetric dilution of commercial single-element ICP-MS standards. Analytical precision was typically better than 5% based on repeat analysis of an in-run check standard.

Concentrations of major anions (F, Cl, SO42−, NO3) and major cations (Li+, Na+, K+, Ca2+, Mg2+, NH4+) were determined on separate Dionex DX-600 IC systems using suppressed conductivity detection and operated by Chromeleon software (version 6.8). The anion system employs a potassium hydroxide eluent generator, a carbonate removal device, and AS-18/AG-18 columns. The cation system is equipped with CS-16 and CG-16 columns and cations are eluted isocratically with 19 mM methanesulfonic acid (MSA) at 0.5 ml/min over 58 min. Both systems are plumbed with an external source of deionized water for suppressor regeneration to improve the signal-to-noise ratio of the analyses. Quantification is achieved externally via calibration curves constructed from a series of dilutions of mixed-ion standards (Environmental Express, Charleston, SC, USA). Quantification accuracy is verified daily by analysis of an independent mixed ion standard (Thermo Scientific, Waltham, MA, USA). Uncertainties in reported ion concentrations are estimated to be ± 5%.

Concentration values from ICP-MS and IC measurements were first corrected by subtracting values obtained from blanks. The absolute mass of analyte was determined by multiplying the measured concentration (in mg/kg) with the total mass of H2O in the measured sample. The total H2O mass includes the mass of ultrapure H2O in which the capsule was submerged and the H2O generated experimentally from antigorite breakdown. The former was measured on a balance. Experimentally generated H2O mass was calculated by mass balance, where the mass of the capsule after being punctured and subsequently dried was subtracted from the mass of the capsule measured before puncture. Analyte mass (in mg) was then divided by the mass of experimentally generated H2O (in kg) to yield the concentration of analyte in the experimental fluid (in ppm).

Thermodynamic mass transfer modeling

Fluid–rock mass transfer modeling was performed with the Fortran codes EQ3NR (Wolery 1983a, b) and EQ6 (Wolery 1984) using a custom database of equilibrium constants involving minerals and aqueous ions, metal complexes, and organics calculated using the deep earth water (DEW) model (Sverjensky et al. 2014 and references therein). Recently, a new equation of state for estimating the dielectric constant of water and new equation of state parameters for major solute species has enabled EQ3NR and EQ6 calculations up to 6 GPa and 1000 °C (Facq et al. 2014; Debret and Sverjensky 2017; Huang and Sverjensky 2019; Sverjensky et al. 2014). The model has been shown to successfully predict experimental data from a wide range of systems at high and low pressures (Sverjensky et al. 2014; Sverjensky 2019).

The DEW model employs an extension of the Helgeson–Kirkham–Flowers aqueous species equation of state and a comprehensive series of correlations to estimate the equation of state and standard partial molal properties when experimental data are lacking. The model EQ3NR follows a conceptual scenario in which initial fluid chemistry matching that of our experiments is equilibrated at QFM + 2 and the desired pressure and temperature (held constant for each model run; Fig. 4). This fluid is then input into EQ6, which performs irreversible mass transfer calculations by “titrating” ultramafic mantle rock (as minerals) into the fluid. The moles of minerals titrated at each step as well as the modal abundance of minerals in the titrated lithology is set by the user. The solution is equilibrated, and the properties of the fluid and solids in the system are reported at each step. This method most closely resembles a natural system dominated by channelized fluid flow in the mantle followed by equilibration with an ultramafic rock. The lithology of the solids is monitored throughout the run.

The model fluid fO2 was set based on the measured redox of our experimental fluids. A wide range of fluid compositions was explored, with maximum and minimum solute concentrations chosen based on experimental fluid compositions, mass balance constraints, and literature data on relevant natural and experimental fluids (Table 3 and Sect. 4.4). Fe3+/ΣFe in the bulk mantle mineral assemblage is calculated at each model step and is used as a proxy for redox of the mantle solids. Our modeling assesses the ability of an oxidized fluid to increase the Fe3+/ΣFe in the mantle mineral assemblage by 0.05–0.15, corresponding to the degree to which primitive arc magmas (~ 0.2 to 0.3; Kelley and Cottrell) are oxidized relative to primitive MORB (~ 0.15; ibid). The relationship between Fe speciation in bulk mantle assemblages and Fe speciation in derivative partial melts is poorly understood (Davis and Cottrell 2018), and so here we focus on the relative degree of oxidation of our modeled mantle rather than absolute values. The modeled ultramafic rock starting composition is a depleted MORB-source mantle lherzolite after Workman and Hart (2005). Fluid and ultramafic rock compositions are given in Table 4.

Table 3 Compositions of experimental fluids by IC and ICP-MS (this study) plus concentrations of other relevant experimental equilibrium fluids and natural slab fluids
Table 4 Compositions of fluid and mantle lithologies used in fluid–rock modeling

A limitation of modern thermodynamic modeling of this sort is the lack of solid solution thermodynamic data necessary to describe the incorporation of Fe3+ into many nominally ferrous silicate minerals such as pyroxenes and olivine, as well as thermodynamic data to describe spinel solid solutions, rather than just end-member phases. The only phases containing Fe3+ in the EQ6 model are magnetite, hematite, garnet, and goethite. Of those, the only magnetite is stable in our model runs, and so magnetite is used as a proxy to track the redox evolution of the solids. In an irreversible mass transfer calculation, the initial bulk chemistry of the rock is what dictates reaction behavior, and so the initial phase assemblage does not affect the results. However, once magnetite becomes saturated as a mineral product during the reaction path, the model then departs from the reality of a natural system in which Fe3+ can be accommodated in any other mineral. In a natural peridotite, changes in the redox state of the system will result in an increase or decrease in the magnetite component of Mg–Fe–Al–Cr spinel. In our model, the abundance of magnetite serves as a proxy for the spinel magnetite component. Silicate minerals including clinopyroxene and orthopyroxene may host significant ferric iron (Canil and O’Neill 1996), and this is not represented in our model. For these reasons, we track the bulk Fe3+/ΣFe of the solid assemblage produced during model runs to track redox. Modal abundances of olivine and pyroxenes are used to ensure that the final modeled “mantle rock” matches that of a reasonable natural lherzolite.

Model pressures and temperatures are limited to those for which thermodynamic data is available and calibrated and were chosen to reflect sub-arc conditions in cold subduction zones (4 GPa, 120 km depth), hot subduction zones (2 GPa, 60 km depth), and in the forearc (1 GPa, 30 km depth). This constrains the maximum and minimum pressure range for these reactions taking place in the sub-arc mantle, such that natural slab fluid reactions will occur within this pressure–temperature space (Fig. 4). Three temperatures were modeled for each pressure, from the temperature just above the serpentinite-out phase transition up to 900 °C. Mantle melting is not considered.


Description of experimental products

The average grain size of solid products from serpentinite-bearing experiments was quite small, with crystals typically ~ 1 µm and rare crystals up to 10 µm (Fig. 5a), making quantitative measurement of the phase chemistry difficult. XRD and EDS analyses indicate the presence of olivine (Fo89-93), orthopyroxene (Mg#90), magnetite, and minor unreacted antigorite. XRD analysis of starting material VAR 10-01 indicates that the material is mostly antigorite with minor magnetite. A comparison between the starting material and experimental products shows a significant decrease in the abundance of antigorite with olivine and orthopyroxene making up the majority of the experimental products, suggesting that most of the serpentinite was expended to produce nominally anhydrous olivine and orthopyroxene. Although our study focused primarily on the experimentally produced fluids, these observations of solid residues agree with other experimental work constraining pressure and temperature conditions and reactive mechanisms for antigorite breakdown (Ulmer and Trommsdorff 1995; Bromiley and Pawley 2003; Padrón-Navarta et al. 2010; Merkulova et al. 2016).

Fig. 5

SEM images of experimental products. (a) Unpolished grain mount of starting material VAR 10-01 after the experiment, illustrating the small grain size. (b) and (c) Post-experiment “pre-mixed” sensors with metallic domains of NiPd (bright) surrounded by an NiO matrix (dark). Orange shaded region in (c) indicates S-bearing domains. (d)–(f) EDS Ni concentration maps superimposed over SEM images of post-experiment redox sensor pellets. All three images at the same scale (scale bar in (d). Green color indicates Ni concentration, with brighter green corresponding to higher Ni concentrations (not quantitative). Progression from shortest (6h) to longest (48h) experiments pictured illustrates growth of reaction rim in redox sensor pellets over time

Description of post-experiment redox sensors

Representative post-experiment redox sensors are shown in Fig. 5. Experiments that employed a ~ 1 mm metal slug surrounded by NiO showed a strong reaction in the rims with cores remaining unreacted. The width of Ni-rich rims increases with experiment duration, indicating the progression of a reaction front with time (Fig. 5d–f). For these sensors, the rims were taken to be in equilibrium with the sample fluid and so only rim measurements were used for calculation of fO2. Experiments that utilized “pre-mixed” sensors produced ~ 10 to 100 µm metallic domains within a NiO matrix (Fig. 5b, c). Metallic domains all displayed systematic, distinct compositional changes from the sensor starting composition and so were interpreted to be equilibrated with the sample. Some metallic zones contained small (few µm), distinct S-rich (< 10 wt.%) regions, typically on rims of metallic blebs, in both serpentinite-bearing and control experiments (Fig. 5c). No sulfide or sulfate minerals were observed (only immiscible sulfur-bearing liquid). Because sulfur was observed in control experiments, which contain no serpentinite starting material, this is assumed to be a minor contaminant either from the zirconia or the sample assembly (e.g., sulfur-bearing lubricants used in experimental assembly). Even if the small sulfur-bearing zones have modest sulfur concentrations, all sulfur is confined to rare (~ 0.1% of the sample) small metallic zones, and so the total amount of sulfur in the bulk sample (zirconia + sensor + capsule + aqueous fluid ± serpentinite) should be exceedingly small (on the order of 100 ppm, equivalent to 1 µmol inside the capsule) and, therefore, will not have a significant effect on redox. This calculation is based on visual observations of our samples, which are assumed to be representative of the whole. Heterogeneities in the distribution and volume of immiscible sulfide liquids could introduce unquantifiable error. However, given the high chalcophile nature of NiPd metal, these zones should represent the vast majority of S in the system. Because S concentrations in these zones are relatively low, and because the formation of sulfide minerals was not observed, it follows that bulk S contents are likely not high enough to affect the redox as recorded by binary redox sensors. Only sulfur-free regions were used to calculate fO2. The presence of sulfur does not interfere with redox sensors since the sulfur-bearing and sulfur-free metallic phases are immiscible at experimental pressure–temperature, and so the activity of Ni in sulfur-free sensor metal may still adjust to satisfy equilibrium between NiPd alloy, NiO, and the fO2 of the system.

Sensors with a Ni-rich (reduced) starting composition far from the expected final composition systematically showed little movement from the starting composition for both serpentinite-bearing and control experiments, regardless of experiment duration. Ni-rich sensors (e.g. XNi = 0.9) placed in a relatively oxidizing environment must scavenge O from the aqueous fluid to decrease the Ni content of the metal (following the reaction Ni + ½O2 ⇌ NiO). Taylor et al. (1992) who established widely used calibration expressions for these sensors found that reversal experiments requiring H2O to propagate the sensor reaction, particularly at lower temperatures, sometimes did not converge on their equilibrium values. Anhydrous experiments by Taylor et al. (1992) fully equilibrated with the shortest experiment times they investigated (12 h). We observe similar behavior where experiments requiring H2O to adjust their sensor composition do not arrive at equilibrium values (compositions unchanged within error) in any of our experiments, which have a maximum experiment time of 72 h. For this reason, and for consistency, we only report results for sensors with Ni-poor (oxidized) starting compositions (Ni0.1Pd0.9, pre-mixed Ni0.1Pd0.9 + NiO, and pre-mixed Ni0.5Pd0.5 + NiO). Agreement between sensors with starting Ni mole fractions of 0.1 and 0.5 support attainment of equilibrium. Only sensors where equilibrium is apparent are reported.

fO2 of experimental fluids from redox sensors

Serpentinite-bearing and control experiments display distinct trends in experimental fO2 (Fig. 6; Table 2). Error bars represent one standard deviation (calculated with the “n = 1” method). Likewise, error for each data series (gray and green bars in Fig. 6), represent one standard deviation about the mean for each series. Control experiments consistently record an ambient fO2 of ~ QFM + 1 (avg. all experiments = QFM + 1.2 ± 0.2) with no change in fO2 with experiment duration, from 1 h up to 24 h. This is within the range of intrinsic fO2 values estimated for similar piston cylinder apparatuses (Matjuschkin et al. 2015). Serpentinite-bearing experiments record an average fO2 of ~ QFM + 2 (avg. all experiments = QFM + 2.1 ± 0.3). Both serpentinite-bearing and control experiments show an increase in measured fO2 in experiments lasting 72 h. We interpret this to be the result of hydrogen diffusion out of the charge through the Au capsule wall (thus leaving excess oxygen inside the capsule). Matjuschkin et al. (2015) showed that hydrogen loss from their charges resulted in an increase of recorded fO2 by ~ 0.25 log units in experiments longer than 24 h. We consider experiments < 24 h representative of the fO2 of the experimentally produced fluid. Two serpentinite-bearing experiments (VAR-07 and VAR-10) show elevated fO2 values within the error of 72 h runs, indicating they may have lost hydrogen or did not fully equilibrate during the run. These experiments both employed pressed metal pellet sensors. The other two 24 h experiments, VAR-NBS0.1-1 and VAR-NBS0.1-2 have fO2 values that agree with shorter duration experiments, and both employed pre-mixed sensors, which were found to more consistently equilibrate.

Fig. 6

Time-series plot showing fO2 values calculated from the compositions of redox sensors in serpentinite-bearing (green dots) and control experiments (black dots) relative to the QFM buffer. Error bars are one standard deviation about the mean of individual probe analyses for each redox sensor (calculated with the “n=1” method). Horizontal bars indicate one standard deviation about the mean for the fO2 of all serpentinite-bearing (green bar, average=2.1 ±0.3) and control experiments (gray bar, average=1.2 ±0.2). The theoretical fO2 of the Ni 0.5Pd0.5 sensors before the experiment (i.e., at duration=0) is ~QFM+4. One control experiment with a run duration of 15 h (NV0.5-1) used a sensor of composition Ni Pd corresponding to an fO of QFM+1.4. The time series indicates that redox sensors adjusted their compositions very quickly, reacting completely within the first hour of the experiment

Fluid compositions

The compositions of fluids extracted from experimental capsules (Table 3; Fig. 7) provide the minimum concentrations of elements in the fluid at high pressure–temperature conditions, since saturated components are likely to exsolve from solutions during quenching to ambient conditions (e.g., Spandler et al. 2007). Slab fluids are expected to be relatively dilute (Manning 2004), with total dissolved solids (TDS) only two to three times greater than seawater (i.e., a few wt.%). Thus, major element concentrations (e.g., Na2O, CaO, Al2O3, etc.) are not expected to be at saturation levels in our experimental proxy for slab fluids, and so the degree to which these fluids are affected by quench may be minor. Despite the potential loss of solute during quenching, absolute concentrations in our most solute-rich experimental fluids agree quite well with those reported in natural antigorite fluid inclusions, perhaps reflecting the naturally dilute nature of subduction zone fluids (Fig. 7b).

Fig. 7

Spider diagram with compositions of experimental fluids extracted at ambient P-T in this study (magenta stars) compared to experimental mantle-water equilibrium fluids (orange squares; Ayers et al. 1997; Brenan et al. 1995; Schneider and Eggler 1986), natural décollement and pore fluids (blue triangles; Fryer et al. 1999; Manning 2004), and high-pressure fluid inclusions in natural antigorites (green dots; Scambelluri et al. 2004). Black dashes and enclosed gray shaded region indicate range of fluid compositions investigated with thermodynamic modeling. Concentrations in (a) normalized to antigorite VAR 10-01

Maximum and minimum fluid mobilities (as concentration in the fluid divided by concentration in the antigorite) for our experimental fluids are plotted in Fig. 7a alongside those calculated for experimental fluids in equilibrium with relevant ultramafic lithologies (Ayers et al. 1997; Brenan et al. 1995; Schneider and Eggler 1986); direct samples of décollement fluids and pore fluids (Fryer et al. 1999; Manning 2004); and high-pressure fluid inclusions in natural antigorites (Scambelluri et al. 2004). The general pattern of major element concentrations in experimental fluids indicates that Na and Cl are the most fluid-mobile, followed by K, Ca, and Mg, consistent with mobilities reported in the relevant literature (also see Dvir et al. 2011; Tatsumi et al. 1986).

Modeling redox potential of slab fluids in the sub-arc mantle

Results from the experiments were used to parameterize fluid-rock mass transfer models to assess the ability of slab fluid to affect the oxidation state of mantle wedge lithologies. Modeling was designed to answer two key questions: first, can slab fluids matching those produced in experiments oxidize a mantle mineral assemblage to the same degree that primitive arc magmas are oxidized relative to primitive MORB (i.e., an increase in Fe3+/ΣFe of 0.5–1.5). And second, if so, how efficient is this process?

In our thermodynamic fluid–rock modeling, a relatively oxidized aqueous fluid with a solute composition and fO2 measured from the experiments is reacted step-wise with a relatively reduced ultramafic rock, and the system experiences a wide range of fluid/rock ratios in which the system comes to equilibrium at each step (Fig. 8). At the beginning of the reaction, the system is fluid-dominated, such that the redox state of the system will be dictated by the fluid. As more rock is added to the system, the fluid/rock ratio decreases and the rock will exert more control over the redox state of the system. This continues until we reach the rock-dominated regime, in which the system’s redox state resembles that of the starting rock. At moderate fluid/rock ratios, the redox potential of the fluid is balanced by that of the rock, and an equilibrated assemblage is generated where the rock has been oxidized and the fluid reduced. Here, the fluid/rock ratio required to elevate mantle Fe3+/ΣFe to values matching those in primitive arc magmas is taken as a measure of the efficiency of the redox reaction. Lower fluid/rock indicates a more efficient reaction (i.e., less fluid is required to oxidize the same mass of rock).

Fig. 8

Results of thermodynamic mass transfer modeling of the interaction between our experimental slab fluids and a MORB-source mantle lherzolite that illustrates the evolution of Fe /ΣFe in mantle rock as a function of fluid/rock ratio in the system. Modeling simulates the stepwise titration of mantle rock with a starting Fe /ΣFe of 0.15 (MORB-source value) into a fluid whose composition and fO2 matches that of our experimentally produced fluids. The Fe /ΣFe is monitored throughout the reaction. Nine discrete model runs are shown here: three temperatures spanning from the serpentinite-out line to 900°C at 1 GPa (a), 4 GPa (b), and 2 GPa (c; see Fig. 4). The horizontal yellow bar indicates the range of Fe /ΣFe values measured in primitive arc magmas (0.18–0.32; Kelley and Cottrell, 2009). Blue vertical bars indicate fluid/rock ratios that produce a metasomatized MORB-source lherzolite with Fe /ΣFe within the range of arc magma values. The lower panel in (c) illustrates the evolution of the activity of H in the fluid for 2 GPa model runs

To test the effect of fluid chemistry on redox potential, we modeled a wide range of solute compositions constrained by our experimental fluid analyses, mass balance calculations using our starting material and fluid compositions (i.e., assuming a given element is completely partitioned into the fluid), and the relevant data from the literature summarized above. The range of fluid compositions is given in Table 4 and indicated by the gray shaded region in Fig. 7. Model results discussed below are for runs with a median solute load.

Over the entire parameter space investigated, fluid/rock ratios ca. 1–100 are required to elevate the Fe3+/ΣFe from MORB values to the range of arc values (indicated by the yellow shaded region in Fig. 8). At 850–900 °C, where hydrous mantle melting begins, the required fluid/rock ratios range from < 1 to ~ 10. In other words, < 1–10 kg of slab fluid is required to oxidize 1 kg of ultramafic mantle rock to the oxidation states observed in arc magmas.

Over the range of fluid compositions tested, solute load has only a small effect on the redox potential of the fluids. This effect is most pronounced at the highest model pressures, where solute solubilities increase significantly (Fig. 9). In 4 GPa runs, fluids with minimum and median solute loads show essentially no change in redox potential. Fluids with maximum solute loads, however, have a significantly higher redox potential, where the fluid/rock ratio corresponding to arc-like Fe3+/ΣFe in the solids decreases by an order of magnitude for all temperatures. This is facilitated by an increased total Fe concentration, which allows for more electron exchange between Fe2+ and Fe3+ in the fluid.

Fig. 9

Effect of solute load on model results. (a) Effect of solute load on redox in 4 GPa model runs at all three investigated temperatures. Minimum and median solute loads produce almost identical results, while maximum solute loads are more effective oxidizers at lower fluid/rock ratios. (b) Effect of solute load on mineralogy during 4 GPa 800 °C model runs. All runs converge at the same redox state and mineralogy at very low fluid/rock ratios (<0.1)

A change in solute load also affects the modal mineralogy of precipitated phases throughout the reaction, although all runs converge on the same mineralogy (olivine, orthopyroxene, clinopyroxene, magnetite, clinochlore) at very low fluid/rock ratios (< 1; Fig. 9b). Note that clinochlore replaces spinel as the Al-bearing phase after reaction with the fluid. As with redox potential, mineralogy is essentially unchanged between the minimum and median solute runs. At maximum solute loads, there are more distinct changes. Brucite, Mg(OH)2, and clinochlore, (Mg,Fe2+)5Al(Si3Al)O10(OH)8, are stabilized early in the reaction due to abundant Mg, Fe, and Al in the fluid. Brucite completely disappears around fluid/rock = 30, at the expense of olivine, which does not precipitate until later in the reaction compared to minimum and median solute models. With a median solute load, olivine is stabilized over brucite, which does not crystallize. Clinochlore abundance is initially much higher (up to ~ 10 wt.% of the mode) in maximum solute models but is decreased to the abundances in lower solute models (~ 2 wt.%) by fluid/rock ca. 10. Major modal components olivine, orthopyroxene, and clinopyroxene all precipitate slightly later during the reaction in maximum solute runs. Magnetite, which is our proxy for Fe3+ in the solids, has a lower modal abundance early in the reaction in maximum solute models due to the enhanced precipitation of brucite, but this does not change the concentration of Fe2+ or Fe3+ in the solid and thus redox is unaffected.

In addition to EQ6 forward modeling, phase diagram sections (i.e., pseudosections) from 1–4 GPa and 500–1000 °C were computed using Perple_X, a Gibbs free energy minimization software (Connolly 2009). Perple_X (version 8.8.9) was used together with the 2002 revision of the Holland and Powell mineral database (Holland and Powell 1998; Ghiorso et al. 2002). Phase diagram modeling was done as a comparison point to EQ6 model results and was specifically chosen since Perple_X considers multicomponent solid solution models for spinel and oxides that are not considered by EQ6 (Supplementary Table A1). Because EQ6 calculations are equilibrated at each step along the reaction path, equilibrium assemblages predicted by EQ6 at discrete steps were compared to equilibrium assemblages predicted by Perple_X. A direct comparison was achieved by taking bulk compositions of the solids from EQ6 as inputs for Perple_X, which was then run at water-saturated conditions. For all compositions considered, fO2 was fixed to − 8.55, equivalent to QFM + 2 at 4 GPa and 800 ∘C. EQ6 and Perple_X results were compared at 4 GPa and 800 °C.

The resulting stable mineral assemblages from Perple_X calculations are remarkably similar to the results from EQ6 (Supplementary Table A2). Even when considering a more complete set of solid solution models, Perple_X predicted assemblages are dominated by end-member minerals for the bulk compositions targeted here. In Perple_X runs, the oxide solid solution Eskol(C) = corundum-hematite (Chaterjee et al. 1982) is present from ~ 3 to 11 mol% (2–7 vol%), increasing with reaction progress, and in all cases is dominated by the hematite end member (~ 90%). For the bulk composition achieved early in the EQ6 reaction, Perple_X predicts the pure end-member phase magnesioferrite at ~ 4 mol% in addition to Eskol(C) at ~ 5 mol%, however, magnesioferrite is not stable in other assemblages. Small percentages of magnetite (4 mol%) and clinochlore (2 mol%) appear in the EQ6 calculations but are not present in Perple_X calculations. Spinel solid solution (Sp(HP)) is not present in any of the Perple_X calculated assemblages. We find that, overall, the predicted phase proportions from both models are quite similar. The differences in mineral proportions or composition are so small as to be inconsequential, indicating that the inclusion of spinel and oxide solid solutions does not significantly affect our results or interpretations.


Identifying redox reactions in our model

To identify whether an aqueous fluid at QFM + 2 has a positive redox potential at sub-arc mantle pressure–temperature conditions as suggested by the modeling results, we must demonstrate a reaction that transfers electrons between an oxidant within the fluid and a reduced multivalent element in the mantle mineral assemblage (e.g., Fe2+). Our modeled system contains no S or C, leaving only Fe and H as potential redox couples, via the reduction half-reactions:

$$2{\text{H}}^{ + } + 2{\text{e}}^{ - } \leftrightarrow {\text{H}}_{2} \left( {{\text{aq}}} \right),$$
$${\text{Fe}}^{3 + } \left( {{\text{aq}}} \right) + {\text{e}}^{ - } \leftrightarrow {\text{Fe}}^{2 + } \left( {{\text{aq}}} \right).$$

Using the valence of Fe in the solids as a measure of the redox state of the mantle, the full redox reactions are:

$$2{\text{Fe}}^{2 + } \left( {\text{s}} \right) + 2{\text{H}}^{ + } \left( {{\text{aq}}} \right) \leftrightarrow 2{\text{Fe}}^{3 + } \left( {\text{s}} \right) + {\text{H}}_{2} \left( {{\text{aq}}} \right),$$
$${\text{Fe}}^{2 + } \left( {\text{s}} \right) + {\text{Fe}}^{3 + } \left( {{\text{aq}}} \right) \leftrightarrow {\text{Fe}}^{3 + } \left( {\text{s}} \right) + {\text{Fe}}^{2 + } \left( {{\text{aq}}} \right) .$$

These reactions can also be written in terms of oxide components as:

$$2{\text{FeO}}\left( {\text{s}} \right) + {\text{H}}_{2} {\text{O}} \leftrightarrow {\text{ Fe}}_{2} {\text{O}}_{3} \left( {\text{s}} \right) + {\text{H}}_{2} \left( {{\text{aq}}} \right),$$
$$2{\text{FeO}}\left( {\text{s}} \right) + {\text{Fe}}_{2} {\text{O}}_{3} \left( {{\text{aq}}} \right) \leftrightarrow {\text{Fe}}_{2} {\text{O}}_{3} \left( {\text{s}} \right) + {\text{FeO}}\left( {{\text{aq}}} \right) .$$

None of these species need to be physically transferred to satisfy redox equilibrium; they must only facilitate electron exchange.

The solubility of Fe in our modeled fluids is exceedingly small, and the a priori speciation of Fe in the fluid assumed at the start of model runs does not have an effect on Fe3+/ΣFe evolution during modeling. This suggests that the reduction of H+ to H2 must be the predominant reaction facilitating electron exchange in our model. This is consistent with the model results, which show an increase in the activity (and concentration) of dissolved H2 in the fluid as the reaction progresses (Fig. 8c). Here, H+ dissolved in the fluid is being reduced to H2 (gaining an electron). This electron is scavenged from Fe2+ in the rock, which is oxidized to Fe3+ in the rock. In these reactions, electrons will flow from the material with higher electron activity to the material with lower electron activity, toward redox equilibrium. Reactions (1a), (2a), and (3a) are thus favored to progress to the right given fluids with a sufficient dissolved H+ concentration (i.e., dissolved H+ ions in excess of the hydrogen in H2O molecules in fluids with sufficiently low pH) since electron activity, pe is anticorrelated with pH. The reaction will stop once pe is balanced in both rock and fluid.

Hydrogen, and aqueous fluid in general, is a well-known oxidizer in shallow systems such as ore bodies (Pirajno 2009), nuclear waste management (Perez et al. 2005), oxidative weathering (White and Yee 1985), and seafloor hydrothermal systems (Bilenker et al. 2016). However, aqueous fluid has been thought to not be an efficient oxidizer in the mantle in part because of the inefficiency of H2O to transfer O2 molecules via dissociation (H2 + ½O2 = H2O; Frost and Ballhaus 1998). But, the invoked redox reactions need not involve oxygen at all, since oxidation and reduction are defined as the loss or gain of electrons, respectively. Further, the generation of H2 in reaction (1a) does not require that H2 gas physically leave the system. Because aqueous fluid is a powerful solvent, the produced H2 remains dissolved within aqueous fluid, which contains dissolved H2, H+, and O2 ions in excess of the hydrogen and oxygen atoms that make up H2O molecules. The ability of aqueous fluid to contain appreciable dissolved H2 has long been exploited by the experimental petrology community, which has developed several devices to measure H2 dissolved in aqueous fluid in hydrothermal experiments (see, e.g., Chou and Eugster 1976; Ding and Seyfried 1995; Gunter et al. 1979). Using our model results, the following worked example quantitatively demonstrates how electron transfer via the reduction of H+ to H2 in the fluid can efficiently oxidize the mantle.

Assume 1 kg of a bulk mantle lherzolite with 10 wt.% FeOtotal. This equates to 7.77 wt.% total Fe ions of all valence states. A MORB mantle with Fe3+/ΣFe = 0.15 contains ~ 1.17 wt.% Fe3+ ions. To increase the Fe3+/ΣFe to a value of 0.2, as found in primitive arc basalts (i.e., to 1.55 wt.% Fe3+), 3.89 g or 0.07 mol of Fe2+ ions must be oxidized to Fe3+ ions in 1 kg of rock. The half reaction in Eq. (1b) dictates that every mole of Fe2+ oxidized to Fe3+ in mantle rock liberates 1 mol of electrons. Thus, the oxidation of 1 kg of our mantle lherzolite requires the transfer of 0.07 mol of electrons. Now consider 1 kg of aqueous fluid matching the composition and oxygen fugacity of our experiments. During model runs at 2 GPa and 800 °C, the mass of H2 dissolved in 1 kg of fluid increases by 0.0086 g or 0.0043 mol H2. Equation (1a) dictates that for every mole of H2 generated from the reduction of H+ in the fluid, 2 mol of electrons much be transferred into the fluid. Therefore, each kilogram of fluid reacted with lherzolite can transfer 0.0086 mol of electrons, and so ~ 8 kg of fluid is required to oxidize 1 kg of lherzolite. This is demonstrated for a range of oxidized mantle Fe3+/ΣFe values in Fig. 8. This specific example corresponds to the 800 °C curve in Fig. 8c when Fe3+/ΣFe = 0.2.

Our model results quantitatively demonstrate that aqueous fluids at subduction zone conditions can efficiently oxidize mantle rock via electron transfer, without the need to transfer O2 molecules directly, consistent with the thermodynamic treatment of works such as Frost and Ballhaus (1998). The electron transfer mechanism discussed here can progress via Fe3+ substitution in relevant mantle minerals without the need of a companion oxygen to maintain charge balance. The manner in which charge balance is maintained will depend on the substitution mechanism in the mineral of interest. For example, it has been shown that charge balance for Fe3+ substitution in olivine is accommodated by an M site vacancy (Nakamura and Schmalzried 1983), which can be described with the reaction \(3{\text{Fe}}^{2 + } \leftrightarrow 2{\text{Fe}}^{3 + } + { }\)\({\text{Fe}}^{2 + } \leftrightarrow 2{\text{Fe}}^{3 + } + { }\)\({\text{Fe}}^{2 + } \leftrightarrow 2{\text{Fe}}^{3 + } + { }\)\({\text{Fe}}^{2 + } \leftrightarrow 2{\text{Fe}}^{3 + } + { }\)\({\text{Fe}}^{2 + } \leftrightarrow 2{\text{Fe}}^{3 + } + { }\)☐ (vacancy). This well-known mechanism for charge balance has been used to explain the formation of Fe3+-bearing olivine under oxidized mantle conditions without the diffusion of oxygen (Ejima et al. 2018; Cline et al. 2018).

Because of the inability of aqueous fluid to transfer O2, the subduction zone community has investigated other oxidants, with a particular focus on S due to its ability to exist in a wide range of valence states, from S2− to S6+, which gives one mole of S the ability to transfer up to eight electrons. The presence of S in subducting slabs would serve to make redox reactions more efficient than those shown here, as postulated by modeling (Debret and Sverjensky 2017) and experimental studies (Merkulova et al. 2017).

With the application of new thermodynamic data describing complexing in aqueous fluids at high P and T, we have demonstrated that aqueous fluids can be a powerful oxidant in the mantle in subduction zones. Even in the absence of S or other redox-sensitive elements, the reduction of H + in the fluid can increase the oxygen fugacity of mantle rock by two orders of magnitude at reasonable fluid/rock ratios. In contrast with previous work that suggests that aqueous fluids are poor oxidizers, this mechanism does not call upon the direct transfer of O2 or Fe3+, or S or C electron transfer between fluid and rock to facilitate an increase in redox state in the mantle. Moreover, it does not require the diffusion of hydrogen out of the system, as called for by Frost and Ballhaus (1998). Instead, the H2 remains dissolved in the aqueous fluid, in equilibrium with oxidized mantle rocks.

Transport of redox potential via mass transfer

It is generally agreed that fluids sourced from the dehydration of hydrous slab lithologies (primarily serpentinized lithosphere) oxidatively metasomatize adjacent mantle material at least locally, that is, near the slab-wedge interface (e.g., Tumiati et al. 2015, Malaspina et al. 2017). There, where fluid/rock is high, it is expected that a metasomatic front will develop from the slab to overlying mantle wedge, possibly through dissolution-reprecipitation of oxides and silicates along veins and fractures. Redox potential and trace elements may then be transported into regions of sub-arc mantle melting either by continued transit of fluids or by advective transport of diapirs (Marschall and Schumacher 2012; Tumiati et al. 2013). Observations of fluid inclusions (Scambelluri et al. 1997; Song et al. 2009; Kawamoto et al. 2013) and multiphase solid inclusions (Malaspina et al. 2006, 2010; Scambelluri et al. 2008; Vrijmoed et al. 2008) in supra-subduction peridotites provide direct evidence for the direct transport of fluids deep into the mantle wedge. More recent evidence suggests that slab fluids can maintain their oxidizing capacity during kilometer-scale transit to regions of arc magma genesis and that this signature can be effectively transferred to primitive arc magmas during mantle melting (Bénard et al. 2018). If fluid flow from the slab to the region of mantle melting is primarily via channelized flow, then the fluid/rock ratio will remain elevated during transport, such that it will preserve its oxidizing capacity. Alternatively, if the fluid flow is dominantly via porous flow, low temperatures and continued fluid flow over time may facilitate oxidation of the mantle melt source regions. Modeling results in Fig. 8 illustrate how a decrease in temperature of 200 °C lowers the efficiency of redox transfer by up to an order of magnitude (corresponding fluid/rock ratios < 10 at 900 °C and < 100 at 700 °C). Slab surface temperatures at the point of slab release are thought to be ~ 700 °C compared to 850–1200 °C in regions of mantle melting (Grove et al. 2012).

Sulfur-enhanced oxidation of the sub-arc mantle

Due to the challenging and unconventional nature of both the experiments and the modeling performed in this study, we purposely chose a relatively simple system—that is, “pristine” antigorite free of appreciable sulfur or carbon species that participate in redox reactions. Sulfur in particular is of great interest, due to its presence in many serpentinites (ranging from nearly absent to several wt.%; e.g., Klein and Bach 2009) and its tremendous redox potential. With valence states ranging from S2− to S6+, one mol of S has the ability to move up to eight mol of electrons. Serpentinites formed in the deep lithosphere at mid-ocean ridges may contain substantial amounts of reduced sulfides (pyrite and pyrrhotite; Alt et al. 2013), while those formed at high water–rock ratios at the seafloor will instead host sulfur as oxidized sulfate (barite, anhydrite, or S6+-bearing serpentine minerals; Debret et al. 2017). Thus, S in serpentinite can act as a strong reducer or oxidizer.

It is unclear how much of the sulfur hosted in serpentinites is transferred to the sub-arc mantle and how much is retained within the subducting slab to great depths (> 5 GPa). Many studies suggest that only a small percentage of sulfur will be released from serpentinites during dehydration in subduction zones, with estimates ranging from < 1% (Evans and Powell 2015) to 15–30% (Evans 2012; Jego and Dasgupta 2013). Even in small abundances, however, sulfur is still a powerful redox agent. Thermodynamic modeling by Debret and Sverjensky (2017) showed that the addition of increasing amounts of pyrrhotite (S as S2−) to a dehydrating serpentinite assemblage drastically changed the evolution of fO2 of the fluid released during the reaction. Interestingly, even with S present as reduced pyrrhotite, modeling precited that oxidized sulfate species always made up a significant portion of S species in the generated fluid. The addition of sulfate to fluids delivered into the mantle wedge in subduction zones can, therefore, enhance the oxidative potential of serpentinite breakdown fluids by providing stronger pathways for electron exchange compared to those explored in this study. Altogether, this suggests that subducting slabs should have variable oxidation potentials, depending on a multitude of factors including the abundance and speciation of sulfur in hydrated subducting lithologies. In this study, we have investigated the baseline oxidative potential of the S-free antigorite system and how oxidation potential varies with slab dip angle and PT pathway. Further experiments to empirically determine the redox potentials of sulfide- and sulfate-bearing serpentinite dehydration fluids are a logical next step in applying these findings to natural subduction zones around the world and could be carried out using the experimental, analytical, and modeling procedures presented in this work.

Global fluid flux in subduction zones

To further test the viability of the proposed mantle wedge oxidation mechanism, we use the modeled fluid/rock ratios to calculate the annual slab fluid flux (mass per time) required to oxidize a sufficiently large mass of the mantle to produce the global annual volcanic flux (erupted products plus unerupted parental material) at arcs. This represents the minimum fluid flux required to explain oxidized signatures in arc magmas, assuming that the source of oxidation is entirely inherited from the mantle source. The flux of mantle material required to account for arc volcanic output can be calculated as:

$$Q_{{\text{M}}} = \frac{{Q_{{\text{V}}} }}{{\left( {1 - F} \right) \times P}} ,$$

where QM is the annual mantle mass required, QV is the annual magmatic mass flux supplied to the crust, F is the fractionation factor required to derive the average magma composition from a primary parent melt, and P is the partial melt fraction to generate a primary melt from the mantle source.

The annual fluid flux required to oxidize the mantle can then be calculated as:

$$Q_{{{\text{Fl}}}} = Q_{{\text{M}}} \times R ,$$

where QFl is the annual fluid mass flux and R is the fluid/rock ratio taken from our modeling results.

The average global arc volcanic magmatic output rate is estimated at 10–5–10–2 km3/year (White et al. 2006). Assuming an average magma density of 2.8 g/cm3, this equates to 2.8 × 107–2.8 × 1010 kg/year of erupted material. Only a portion of magma generated at arcs will be erupted, with the rest stalling in the crust as magmatic intrusions. On average, we assume that erupted arc lavas represent a 20% residual liquid from a primary mantle melt (F = 0.8). This is consistent with chemical mass balance modeling of arc lavas (e.g., Grove et al. 2002, 2003) and average intrusive:extrusive mass ratios at arcs (~ 5:1; White et al. 2006). We also assume that primary arc magmas are 10% partial mantle melts (P = 0.1; Till et al. 2012). These values give an estimated required mantle mass flux of approximately 1.4 × 109–1.4 × 1012 kg/year, which would require the infiltration of 1.4 × 109–1.4 × 1014 kg/year of slab fluid beneath the arc front, assuming fluid/rock ratios ranging from 1 to 100. This fluid flux overlaps with estimated global H2O return fluxes from slab to mantle beneath the arc front (15–100 km depth) in subduction zones of 2.5 × 1011–7.5 × 1011 kg/year, equating to approximately one trillionth of the Earth’s modern ocean mass circulating through subduction zones each year or 0.01% of the ocean mass per billion years (Schmidt and Poli 2003; van Keken et al. 2011; Parai and Mukhopadhyay 2012). Thus, given estimated fluid fluxes in subduction zones, oxidizing (QFM + 2) slab fluids alone have the capacity and thermodynamic impetus to oxidize the sub-arc mantle wedge in the region of melt generation to the values observed in primitive arc magmas.


The results presented here demonstrate that aqueous fluids produced by the dehydration of serpentinite in the subducted slab alone have the capacity to oxidize the sub-arc mantle and produce the oxidized signatures measured in arc lavas. Our experiments and thermodynamic modeling demonstrate that H2O is an effective mantle oxidant in the absence of other significant redox sensitive solutes. Oxidation of Fe in the mantle is balanced by the reduction of H+ to H2 in the fluid. Consequently, moderately oxidizing fluids produced during serpentinite breakdown (QFM + 2) can shift mantle redox from MORB-source mantle values (Fe3+/ΣFe = 0.15) to arc magma values (Fe3+/ΣFe = 0.2–0.3) at fluid/rock ratios thought to be present in the sub-arc mantle. Mass balance calculations suggest that current estimates for the global flux of slab fluids are overall of sufficient magnitude to produce the volume of arc magmas that carry oxidized signatures. The oxidizing effect of serpentinite-dehydration fluids on sub-arc mantle will vary in hotter or colder subduction zones globally (with oxidation potential increasing with temperature) and with the amount of fluid flux from the slab, producing the natural range of arc magma fO2. In subduction zones where slabs carry down abundant S (which has a very large oxidation potential, from S2− to S6+) or C, the oxidation potential of derivative fluids will be enhanced and may explain elevated redox states in cold subduction zones or those with lower fluid flux.


  1. Alt JC, Garrido CJ, Shanks WC, Turchyn A, Padrón-Navarta JA, Sánchez-Vizcaíno VL, Gómez-Pugnaire MT, Marchesi C (2012) Recycling of water, carbon, and sulfur during subduction of serpentinites: a stable isotope study of Cerro del Almirez, Spain, earth and planet. Sci Lett 327–328:50–60

    Google Scholar 

  2. Alt JC, Schwarzenbach EM, Früh-Green GL, Shanks WC, Bernasconi SM, Garrido CJ, Crispini L, Gaggero L, Padrón-Navarta JA, Marchesi C (2013) The role of serpentinites in cycling carbon and sulfur: seafloor serpentinization and subduction metamorphism. Lithos 178:40–54

    Google Scholar 

  3. Ayers JC, Dittmer SD, Layne GD (1997) Partitioning of elements between peridotite and H2O and 2.0–3.0 GPa and 900–1100 °C, and application to models of subduction zone processes. Earth Planet Sci Lett 150:381–398

    Google Scholar 

  4. Bali E, Audétat A, Keppler H (2010) The mobility of U and Th in subduction zone fluids: an indicator of oxygen fugacity and fluid salinity. Contrib Mineral Petrol 161:597–613

    Google Scholar 

  5. Bénard A, Klimm K, Woodland AB, Arculus RJ, Wilke M, Botcharnikov RE, Shimizu N, Nebel O, Rivard C, Ionov DA (2018) Oxidising agents in sub-arc mantle melts link slab devolatilization and arc magmas. Nat Commun 9:3500

    Google Scholar 

  6. Bilenker LD, Romano GY, McKibben MA (2016) Kinetics of sulfide mineral oxidation in seawater: implications for acid generation during in situ mining of seafloor hydrothermal vent deposits. Appl Geochem 75:20–31

    Google Scholar 

  7. Bishop JL, Dyar MD, Skulte EC, Drief A (2008) Physical alteration of antigorite: a Mössbauer spectroscopy, reflectance spectroscopy and TEM study with applications to Mars. Clay Miner 38:55–67

    Google Scholar 

  8. Brandon AD, Draper DS (1996) Constraints on the origin of the oxidation state of mantle overlying subduction zones: an example from Simcoe, Washington, USA. Geochim Cosmochim Acta 60(10):1739–1749

    Google Scholar 

  9. Brenan JM, Shaw HF, Ryerson FJ, Phinney DL (1995) Mineral-aqueous fluid partitioning of trace elements at 900 °C and 2.0 GPa: constraints on the trace element chemistry of mantle and deep crustal fluids. Geochim Cosmochim Acta 59:3331–3350

    Google Scholar 

  10. Bromiley GD, Pawley AR (2003) The stability of antigorite in the systems MgO–SiO2–H2O (MSH) and MgO–Al2O3–SiO2–H2O (MASH): the effects of Al3+ substitution on high-pressure stability. Am Mineral 88:99–108

    Google Scholar 

  11. Brounce MN, Kelley KA, Cottrell E (2014) Variations in Fe3+/ΣFe of Mariana arc basalts and mantle wedge fO2. J Petrol 55(12):2513–2536

    Google Scholar 

  12. Canil D, O’Neill H, St C (1996) Distribution of ferric iron in some upper-mantle assemblages. J Petrol 37:609–635

    Google Scholar 

  13. Cannaò E, Scambelluri M, Agostini S, Tonarini S, Godard M (2016) Linking serpentinite geochemistry with tectonic evolution at the subduction plate-interface: the Voltri Massif case study (Ligurian Western Alps, Italy). Geochim Cosmochim Acta 190:115–133

    Google Scholar 

  14. Carapezza M (1996) A method for continuously variable control of the fugacity of gas in hydrothermal synthesis at high pressure. Geochem Int 3:819–823

    Google Scholar 

  15. Chatterjee ND, Leistner H, Terhart L, Abraham K, Klaska R (1982) Thermodynamic mixing properties of corundum-eskolaite, α-(Al, Cr + 3)2O3, crystalline solutions at high temperatures and pressures. Am Mineral 7–8:725–735

    Google Scholar 

  16. Chou I-M, Eugster HP (1976) A sensor for hydrogen fugacities at elevated P and T and applications. Eos Trans Am Geophys Union 57:340 (abstract)

    Google Scholar 

  17. Connolly, JAD (2009) The geodynamic equation of state: what and how. G3, v. 10.

  18. Cline II CJ, Faul UH, David EC, Berry AJ, and Jackson I (2018) Redox-influenced seismic properties of upper-mantle olivine. Nature 555:355–358

    Google Scholar 

  19. Davis FA, Cottrell E (2018) Experimental investigation of basalt and peridotite oxybarometers: Implications for spinel thermodynamic models and Fe3+ compatibility during generation of upper mantle melts. Am Mineral 103:1056–1067

    Google Scholar 

  20. Debret B, Andreani M, Muñoz M, Bolfan-Casanova N, Carlut J, Nicollet C, Schwartz S, Trcera N (2014) Evolution of Fe redox state in serpentine during subduction. Earth Planet Sci Lett 300:206–218

    Google Scholar 

  21. Debret B, Bolfan-Casanova N, Padrón-Navarta JA, Mertin-Hernandez F, Andreani M, Garrido CJ, Sánchez-Vizcaíno VL, Gómez-Pugnaire MT, Muñoz M, Trcera N (2015) Redox state of iron during high-pressure serpentinite dehydration. Contrib Mineral Petrol 169:36

    Google Scholar 

  22. Debret B, Sverjensky DA (2017) Highly oxidizing fluids generated during serpentinite breakdown in subduction zones. Sci Rep 7:10351

    Google Scholar 

  23. Debret B, Andreani M, Delacour A, Rouméjon S, Trcera N, Williams H, EIMF (2017) Assessing sulfur redox state and distribution in abyssal serpentinites using XANES spectroscopy. Earth Plant Sci Lett 466:1–11

    Google Scholar 

  24. Ding K, Seyfried WE Jr (1995) In-situ measurement of dissolved H2 in aqueous fluid at elevated temperatures and pressures. Geochim Cosmochim Acta 59:4796–4773

    Google Scholar 

  25. Dvir O, Pettke T, Fumagalli P, Kessel R (2011) Fluids in the peridotite-water system up to 6 GPa and 800 °C: new experimental constrains on dehydration reactions. Contrib Mineral Petrol 161:829–844

    Google Scholar 

  26. Ejima T, Osanai Y, Akasaka M, Adachi T, Nakano N, Kon Y, Ohfuji H, and Sereenen J (2018) Oxidation states of Fe in constituent minerals of a spinel lherzolite xenolith from Tariat Depression, Mongolia: The significance of Fe3+ in olivine. Minerals 8:204

    Google Scholar 

  27. Evans KA (2012) The redox budget of subduction zones. Earth Sci Rev 113:11–32

    Google Scholar 

  28. Evans KA, Powell R (2015) The effect of subduction on the sulphur, carbon, and redox budget of lithospheric mantle. J Metamorph Geol 33:649–670

    Google Scholar 

  29. Facq S, Daniel I, Montagnac G, Cardon H, Sverjensky DA (2014) In situ Raman study and thermodynamic model of aqueous carbonate speciation in equilibrium with aragonite under subduction zone conditions. Geochim Cosmochim Acta 132:375–390

    Google Scholar 

  30. Frost BR, Ballhaus C (1998) Comment on “Constraints on the origin of the oxidation state of mantle overlying subduction zones: an example from Simcoe, Washington, USA” by A.D. Brandon and D.S Draper. Geochim Cosmochim Acta 62:329–331

    Google Scholar 

  31. Fryer P, Wheat CG, Mottl MJ (1999) Mariana blueschist mud volcanism: Implications for conditions within the subduction zone. Geology 27:103–106

    Google Scholar 

  32. Ghiorso MS, Hirschmann MM, Reiners PW, Kress VC (2002) The pMELTS: a revision of MELTS for improved calculation of phase relations and major element partitioning related to partial melting of the mantle to 3 GPa. G3 3(5):1–35

    Google Scholar 

  33. Groppo C, Castelli D (2010) Prograde PT evolution of a lawsonite eclogite from the Monviso meta-ophiolite (western alps): dehydration and redox reactions during subduction of oceanic FeTi-oxide Gabbro. J Petrol 51:2489–2514

    Google Scholar 

  34. Grove TL, Parman S, Bowring S, Price R, Baker M (2002) The role of an H2O-rich fluid component in the generation of primitive basaltic andesites and andesites from the Mt. Shasta region. N Calif Contrib Mineral Petrol 142:375–396

    Google Scholar 

  35. Grove TL, Elkins-Tanton LT, Parman SW, Chatterjee N, Müntener O, Gaetani GA (2003) Fractional crystallization and mantle-melting controls on calc-alkaline differentiation trends. Contrib Mineral Petrol 145:515–533

    Google Scholar 

  36. Grove TL, Till CB, Lev E, Chatterjee N, Medard E (2009) Kinematic variables and water transport control the formation and location of arc volcanoes. Nature 459:694–697

    Google Scholar 

  37. Grove TL, Till CB, Krawczynski MJ (2012) The role of H2O in subduction zone magmatism. Annu Rev Earth Planet Sci 40:413–439

    Google Scholar 

  38. Gunter WD, Myers J, Girsperger S (1987) Hydrogen: metal membranes. In: Ulmer GC, Barnes HL (eds) Hydrothermal experimental techniques, 4th edn. Wiley, New York, pp 100–120

    Google Scholar 

  39. Hawkesworth CJ, Turner SP, McDermott F, Peate DW, Van Calsteren P (1997) U-Th isotopes in arc magmas: implications for element transfer from the subducted crust. Science 276(5312):551–555

    Google Scholar 

  40. Hays JF (1966) Lime–alumina–silica: Carnegie Institution of Washington year book, pp 234–239

  41. Holland T, Powell R (1998) An internally consistent thermodynamic data set for phases of petrological interest. J Metamorph Geol 16:309–343

    Google Scholar 

  42. Huang F, Sverjensky DA (2019) Extended deep earth water model for predicting major element metasomatism. Geochim Cosmochim Acta 254:192–230

    Google Scholar 

  43. Jego S, Dasgupta R (2013) Fluid-present melting of sulfide-bearing ocean-crust: experimental constraints on the transport of sulfur from subducting slab to mantle wedge. Geochim Cosmochim Acta 110:106–134

    Google Scholar 

  44. John T, Scambelluri M, Frische M, Barnes JD, Bach W (2011) Dehydration of subducting serpentinite: implications for halogen mobility in subduction zones and the deep halogen cycle. Earth Planet Sci Lett 308:65–76

    Google Scholar 

  45. Iwamori H, Zhao D (2000) Melting and seismic structure beneath the northeast Japan arc. Geophys Res Lett 27:425–428

    Google Scholar 

  46. Kawamoto T, Yoshikawa M, Kumagai Y, Mirabueno MHT, Okuno M (2013) Mantle wedge infiltrated with saline fluids from dehydration and decarbonation of subducting slab. Proc Natl Acad Sci USA 110:9663–9668

    Google Scholar 

  47. Kelley KA, Cottrell E (2009) Water and the oxidation state of subduction zone magmas. Nature 325:605–607

    Google Scholar 

  48. Kelley KA, Cottrell E (2012) The influence of magmatic differentiation on the oxidation state of Fe in a basaltic arc magma. Earth Planet Sci Lett 329–330:109–121

    Google Scholar 

  49. Kendrick MA, Scambelluri M, Honda M, Phillips D (2011) High abundances of noble gas and chlorine delivered to the mantle by serpentinite subduction. Nat Geosci 4(11):807–812.

    Article  Google Scholar 

  50. Kendrick MA, Honda M, Pettke T, Scambelluri M, Phillips D, Giuliani A (2013) Subduction zone fluxes of halogens and noble gases in seafloor and forearc serpentinites. Earth Planet Sci Lett 365:86–96

    Google Scholar 

  51. Kessel R, Schmidt MW, Ulmer P, Pettke T (2005) Trace element signature of subduction-zone fluids, melts and supercritical liquids at 120–180 km depth. Nature 437(29):724–727

    Google Scholar 

  52. King PL, Hervig RL, Holloway JR, Delaney JS, Dyar MD (2000) Partitioning of Fe3+/Fe total between amphibole and basanitic melt as a function of oxygen fugacity. Earth Planet Sci Lett 178:97–112

    Google Scholar 

  53. Klein F, Bach W (2009) Fe–Ni–Co–O–S phase relations in peridotite–seawater interactions. J Petrol 50(1):37–59

    Google Scholar 

  54. Lee C-TA, Leeman WP, Canil D, Li Z-XA (2005) Similar V/Sc systematics in MORB and arc basalts: implications for the oxygen fugacities of their mantle source regions. J Petrol 46(11):2313–2336

    Google Scholar 

  55. Longhi J (2005) Temporal stability and pressure calibration of barium carbonate and talc/pyrex pressure media in a piston-cylinder apparatus. Am Mineral 90:206–218

    Google Scholar 

  56. Malaspina N, Hermann J, Scambelluri M, Compagnoni R (2006) polyphase inclusions in garnet-orthopyroxenite (Dabie Shan, China) as monitors for metasomatism and fluid-related trace element transfer in subduction zone peridotite. Earth Planet Sci Lett 249:173–187

    Google Scholar 

  57. Malaspina N, Scambelluri M, Poli S, Van Roermund HLM, Langenhorst F (2010) The oxidation state of mantle wedge majoritic garnet websterites metasomatized by C-bearing subduction fluids. Earth Planet Sci Lett 298:417–426

    Google Scholar 

  58. Malaspina N, Tumiati S (2012) The role of C–O–H and oxygen fugacity in subduction-zone garnet peridotites. Eur J Mineral 24:607–618

    Google Scholar 

  59. Malaspina N, Langenhorst F, Tumiati S, Campione M, Frezzotti ML, Poli S (2017) The redox budget of crust-derived fluid phases at the slab-mantle interface. Geochim Cosmochim Acta 209:70–84

    Google Scholar 

  60. Manning CE (2004) The chemistry of subduction-zone fluids. Earth Planet Sci Lett 223:1–16

    Google Scholar 

  61. Marschall HR, Schumacher JC (2012) Arc magmas sourced from mélange diapirs in subduction zones. Nat Geosci 5:862–867

    Google Scholar 

  62. Matjuschkin V, Brooker RA, Tattitch B, Blundy JD, Stamper CC (2015) Control and monitoring of oxygen fugacity in piston cylinder experiments. Contrib Mineral Petrol 169(9):1–14

    Google Scholar 

  63. Merkulova M, Muñoz M, Vidal O, Brunet F (2016) Role of iron content on serpentinite dehydration depth in subduction zones: Experiments and thermodynamic modeling. Lithos 264:441–452

    Google Scholar 

  64. Merkulova MV, Muñoz M, Brunet F, Vidal O, Hattori K, Vantelon D, Trcera N, Huthwelker T (2017) Experimental insight into redox transfer by iron- and sulfur-bearing serpentinite dehydration in subduction zones. Earth Planet Sci Lett 479:133–143

    Google Scholar 

  65. Moussallam Y, Oppenheimer C, Scaillet B (2019) On the relationship between oxidation state and temperature of volcanic gas emissions. Earth Planet Sci Lett 520:260–267

    Google Scholar 

  66. Mungall JE (2002) Roasting the mantle: slab melting and the genesis of major Au and Au-rich Cu deposits. Geology 30:915–918

    Google Scholar 

  67. Murad E (1998) Clays and clay minerals: what can Mössbauer spectroscopy do to help understand them? Hyperfine Interact 117:39–70

    Google Scholar 

  68. Nakamura A, Schmalzried H (1983) On the nonstoichiomety and point defects of olivine. Phys Chem Minerals 10:27–37

    Google Scholar 

  69. Padrón-Navarta JA, Hermann J, Garrido CJ, Sánchez-Vizcaíno VL, Gómez-Pugnaire MT (2010) An experimental investigation of antigorite dehydration in natural silica-enriched serpentinite. Contrib Mineral Petrol 159:25–42

    Google Scholar 

  70. Padrón-Navarta JA, Sánchez-Vizcaíno VL, Hermann J, Connolly JAD, Garrido CJ, Gómez-Pugnaire MT, Marchesi C (2013) Tschermak’s substitution in antigorite and consequences for phase relations and water liberation in high-grade serpentinites. Lithos 178:186–196

    Google Scholar 

  71. Parai R, Mukhopadhyay S (2012) How large is the subducted water flux? New constraints on mantle regassing rates. Earth Planet Sci Lett 317:396–406

    Google Scholar 

  72. Parkinson IJ, Arculus RJ (1999) The redox state of subduction zones: insights from arc-peridotites. Chem Geol 160:409–423

    Google Scholar 

  73. Perez JR, Banwart SA, Piugdomenech I (2005) The kinetics of O2(aq) reduction by structural ferrous iron in naturally occurring ferrous silicate minerals. Appl Geochem 20:2003–2016

    Google Scholar 

  74. Pirajno F (2009) Water and hydrothermal fluids on Earth. Hydrothermal processes and mineral systems. Springer, Dordrecht, pp 1–71

    Google Scholar 

  75. Poli S, Schmidt MW (2002) Petrology of subducted slabs. Annu Rev Earth Planet Sci 30:207–235

    Google Scholar 

  76. Pownceby MI, O’Neill H, St C (1994) Thermodynamic data from redox reactions at high temperatures. III. Activity-composition relations in Ni–Pd alloys from EMF measurements at 850–1250 K, and calibration of the NiO + Ni–Pd assemblage as a redox sensor. Contrib Mineral Petrol 116:327–339

    Google Scholar 

  77. Righter K, Hauri EH (1998) Compatibility of rhenium in garnet during mantle melting and magma genesis. Science 280:1737–1741

    Google Scholar 

  78. Rudnick RL, Gao S (2003) Composition of the continental crust. Treatise on geochemistry, 3rd edn. Elsevier, Pergamon

    Google Scholar 

  79. Rüpke LH, Morgan JP, Hort M, Connolly JAD (2004) Serpentine and the subduction zone water cycle. Earth Planet Sci Lett 224:17–34

    Google Scholar 

  80. Sarbas G, Nohl U (2008) The GEOROC database as part of a growing geoinformatics network. In: Brady SR, Sinha AK, Gundersen LC (eds) Geoinformatics 2008—data to knowledge. Proceedings of the U.S. Geological Survey scientific investigations report 2008–5172, pp 42–43

  81. Scambelluri M, Piccardo GB, Philippot P, Robbiano A, Negretti L (1997) High salinity fluid inclusions formed from recycled seawater in deeply subducted alpine serpentinite. Earth Planet Sci Lett 148:485–499

    Google Scholar 

  82. Scambelluri M, Fiebig J, Malaspina N, Müntener O, Pettke T (2004) Serpentinite subduction: implications for fluid processes and trace-element recycling. Int Geol Rev 46(7):595–613

    Google Scholar 

  83. Scambelluri M, Pettke T, Van Roermund HLM (2008) Majoritic garnets monitor deep subduction fluid flow and mantle dynamics. Geology 36:59

    Google Scholar 

  84. Scambelluri M, Tonarini S (2012) Boron isotope evidence for shallow fluid transfer across subduction zones by serpentinized mantle. Geology 40(10):907–910

    Google Scholar 

  85. Schmidt MW, Poli S (1998) Experimentally based water budgets for dehydrating slabs and consequences for arc magma generation. Earth Planet Sci Lett 163:361–379

    Google Scholar 

  86. Schmidt MW, Poli S (2003) Generation of mobile components during subduction of oceanic crust. In: Treatise on geochemistry, vol 3. The crust, pp 567–591. Elsevier, New York

  87. Schneider ME, Eggler DH (1986) Fluids in equilibrium with peridotite minerals: implications for mantle metasomatism. Geochim Cosmochim Acta 50:711–724

    Google Scholar 

  88. Song S, Su L, Niu Y, Lai Y, Zhang L (2009) CH4 inclusions in orogenic harzburgite: evidence for reduced slab fluids and implication for redox melting in mantle wedge. Geochim Cosmochim Acta 73:1737–1754

    Google Scholar 

  89. Spandler C, Mavrogenes J, Hermann J (2007) Experimental constraints on element mobility from subducted sediments using high-P synthetic fluid/melt inclusions. Chem Geol 239:228–249

    Google Scholar 

  90. Stagno V, Frost DJ (2010) Carbon speciation in the asthenosphere: experimental measurements of the redox conditions at which carbonate-bearing melts coexist with graphite or diamond in peridotite assemblages. Earth Planet Sci Lett 300:72–84

    Google Scholar 

  91. Sverjensky DA, Harrison B, Azzolini D (2014) Water in the deep Earth: the dielectric constant and the solubilities of quartz and corundum to 60 kb and 1,200 °C. Geochim Cosmochim Acta 129:125–145

    Google Scholar 

  92. Sverjensky DA (2019) Thermodynamic modelling of fluids from surficial to mantle conditions. J Geol Soc 176:348–374

    Google Scholar 

  93. Syracuse EM, van Keken PE, Abers GA (2010) The global range of subduction zone thermal models. Physics Earth Planet Int 183:73–90

    Google Scholar 

  94. Tatsumi Y, Hamilton DL, Nesbitt RW (1986) Chemical characteristics of fluid phase released from a subducted lithosphere and origin of arc magmas: evidence from high-pressure experiments and natural rocks. J Volcanol Geotherm Res 29:293–309

    Google Scholar 

  95. Taylor JR, Wall VJ, Pownceby MI (1992) The calibration and application of accurate redox sensors. Am Mineral 77:284–295

    Google Scholar 

  96. Tenthorey E, Hermann J (2004) Composition of fluids during serpentinite breakdown in subduction zones: evidence for limited boron mobility. Geology 32:865–868

    Google Scholar 

  97. Till CB, Grove TL, Withers AC (2012) The beginnings of hydrous mantle wedge melting. Contrib Mineral Petrol 163:669–688

    Google Scholar 

  98. Tumiati S, Fumagalli P, Tiraboschi C, Poli S (2013) An experimental study on COH-bearing peridotite up to 3.2 GPa and implications for crust-mantle recycling. J Petrol 54:453–479

    Google Scholar 

  99. Tumiati S, Godard G, Martin S, Malaspina N, Poli S (2015) Ultra-oxidized rockes in subduction mélangees? Decoupling between oxygen fugacity and oxygen availability in an Mn-rich metasomatic environment. Lithos 226:116–130

    Google Scholar 

  100. Ulmer P, Trommsdorff V (1995) Serpentinite stability to mantle depths and subduction-related magmatism. Science 268:858–861

    Google Scholar 

  101. Ulmer P, Trommsdorff V (1999) Phase relations of hydrous mantle subducting to 300 km. In: Fei Y, Bertka CM, Mysen BO (eds) Mantle petrology: field observations and high pressure experimentation. Special publication no. 6. The Geochemical Society, Houston, pp 259–281

  102. van Keken PE, Hacker BR, Syracuse EM, Abers GA (2011) Subduction factory: 4. Depth-dependent flux of H2O from subducting slabs worldwide. J Geophys Res 116:1–15

    Google Scholar 

  103. Vrijmoed J, Smith D, Van Roermund H (2008) Raman confirmation of microdiamond in the Svartberget Fe–Ti type garnet peridotite, Western Gneiss Region, Western Norway. Terra Nova 20:295–301

    Google Scholar 

  104. Wang J, Xiong X, Takahashi E, Zhang L, Li L, Liu X (2019) Oxidation state of arc mantle revealed by partitioning of V, Sc, and Ti between mantle minerals and basaltic melts. J Geophys Res Solid Earth 124:4617–4638

    Google Scholar 

  105. Waters LE, Lange RA (2016) No effect of H2O degassing on the oxidation state of magmatic liquids. Earth Planet Sci Lett 447:48–59

    Google Scholar 

  106. Watson E, Wark D, Price J, Van Orman J (2002) Mapping the thermal structure of solid-media pressure assemblies. Contrib Mineral Petrol 142:640–652

    Google Scholar 

  107. White AF, Yee A (1985) Aqueous oxidation–reduction kinetics associated with coupled electron–cation transfer from iron-containing silicates at 25 °C. Geochim Cosmochim Acta 49:1263–1275

    Google Scholar 

  108. White SM, Crisp JA, Spera FJ (2006) Long-term volumetric eruption rates and magma budgets. Geochem Geophys Geosyst 7(3):1–20

    Google Scholar 

  109. Wolery TJ (1983) EQ3NR: a computer program for geochemical aqueous speciation-solubility calculations: user’s guide and documentation: UCRL-53414 (Lawrence Livermore Lab, Univ Calif)

  110. Wolery TJ (1984) EQ6—a computer program for reaction-path modelling of aqueous geochemical systems: user’s guide and documentation: UCRL-51 (Lawrence Livermore Nat Lab, Univ Calif)

  111. Wood BJ, Bryndzia LT, Johnson KE (1990) Mantle oxidation state and its relationship to tectonic environment and fluid speciation. Science 248(4953):337–345

    Google Scholar 

  112. Workman RK, Hart SR (2005) Major and trace element composition of the depleted MORB mantle (DMM). Earth Planet Sci Lett 231:53–72

    Google Scholar 

Download references


This project was supported by NSF FESD-1338810 to A. Anbar and NSF EAR-1447342 to C.B. Till. The authors thank Chris Clark and two anonymous reviewers for comments on an early version of the manuscript. The authors also thank M. Scambelluri for supplying natural starting material and for important discussions about the role of serpentinite in subduction zones; E. Alp and the Argonne National Lab for performing Mössbauer analysis; A. Wittmann and the ASU Eyring Center for assistance on the electron microprobe at ASU; S. Romaniello for ICP-MS measurements at the ASU Keck Facility; K. Fecteau for ion chromatography measurements; D. Sverjensky and J. Leong for assistance in EQ3/6 modeling; and E. Shock and A. Regberg for illuminating discussions on redox reactions in aqueous fluids.

Author information



Corresponding author

Correspondence to Kayla Iacovino.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Communicated by Timothy L. Grove.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary file1 (DOCX 13 kb)

Supplementary file2 (DOCX 16 kb)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Iacovino, K., Guild, M.R. & Till, C.B. Aqueous fluids are effective oxidizing agents of the mantle in subduction zones. Contrib Mineral Petrol 175, 36 (2020).

Download citation


  • Redox
  • Oxygen fugacity
  • Subduction zones
  • Mantle
  • Serpentinite
  • Experiments