Serpentinites and related rocks form by hydration of ultrabasic and basic rocks at conditions ranging from those of the surface of the Earth (Etiope et al., 2011), ocean floor (Fruh-Green et al., 2004; Kelley et al., 2001) or Mars (Ehlmann et al., 2009) to those of depths of 150–200 km in subduction zones (Ulmer and Trommsdorff, 1995). Serpentinization is associated with redox reactions affecting iron in hydrous silicates and carbon in fluids, resulting in hydrogen and hydrocarbon production (Andreani et al., 2013) and providing potential niches for early life (Pons et al., 2011; Schulte et al., 2006). Deeper in subduction zones, redox reactions can affect the nature of carbon-bearing phases in association with serpentinization (Vitale Brovarone et al., 2017), providing deep fluid sources of reduced carbon to feed a deep biosphere (Vitale Brovarone et al., 2020). In carbonaceous chondrites, knowledge of the redox state of iron in hydrous silicates is essential for understanding hydrothermal reactions and their relationship to the evolution of organics and carbonaceous matter (Beck et al., 2012; Garenne et al., 2019).

Measurement of the ferric-to-total-iron ratio is essential in quantifying redox processes and relies either on bulk rock composition measurement (Padrón-Navarta et al., 2011; Evans, 2012) or on spectroscopic determination in mineral fractions and modal composition of rocks (Debret et al., 2014, 2015). Conventional Mössbauer spectroscopy has been widely applied for decades but has a resolution no better than a few hundred micrometers, which restricts imaging and mapping (McCammon et al., 1991; Sobolev et al., 1999), a possibility offered by X-ray absorption near-edge spectroscopy (XANES) to within a few micrometers or less (Wilke et al., 2001). Thus, Mössbauer spectroscopy can be applied reliably to rocks with homogeneous mineralogical composition (Fig. 1a) and to separated mineral fractions, while XANES allows the study of complex mineralogy (Fig. 1b), a common case in natural rocks.

Applications of XANES spectroscopy to ferro-magnesian phyllosilicates include chlorites (Muñoz et al., 2006), talc, micas and serpentine minerals (Muñoz et al., 2013). However, antigorites display XANES-determined ferric iron contents (Debret et al., 2014, 2015; Muñoz et al., 2013) that are much higher than those obtained from Mössbauer spectroscopy on samples from similar geological settings (Evans et al., 2012). Wet-chemistry determinations of ferric-to-total-iron ratios in hydrated ultramafic rocks are also questionable as Mössbauer and chemical measurements yield discrepant results (Rozenson et al., 1979).

Here we investigate several antigorite and chlorite samples from high-pressure ophiolites using Mössbauer spectroscopy and estimate redox changes associated with reactions involving serpentine transformation and destabilization. We discuss the relative merits of Mössbauer and XANES for quantifying ferric-to-total-iron ratios in ferro-magnesian phyllosilicates from terrestrial rocks and meteorites.

The samples come from ophiolites and tectonic mélange units of Cenozoic to Paleozoic age. Samples from the Alps are similar in nature to those previously studied by XANES (Debret et al., 2014) as they come from similar high-pressure oceanic units that suffered subduction-type blueschist–eclogite facies metamorphic conditions of about 1.5–2.5 GPa and 450–600 ∘C during the Eocene (Scambelluri et al., 1997; Schwartz et al., 2013). Samples from Baja California Sur belong to the basal serpentinite unit of the Sierra de San Andrés ophiolite, Vizcaíno Peninsula, Mexico (Sedlock, 2003). This mélange comprises tectonic blocks that were metamorphosed under conditions ranging from pumpellyite to blueschist facies at 200–500 ∘C and 0.3–0.8 GPa ca. 180 Myr ago (Moore, 1986). The Ōeyama sample comes from a mélange complex that was metamorphosed under pumpellyite to high epidote–blueschist facies up to 450–550 ∘C and 1.5 GPa ca. 320 Myr ago (Tsujimori and Itaya, 1999).

Minerals were extracted from five serpentinites and two chloritites. Rocks that present a nearly monomineralic composition or large crystals of lizardite in bastite texture were carefully chosen in order to obtain single mineral separates in the required quantity (about 200 mg) for Mössbauer spectroscopy. Sample mineralogy (Table 1) was checked by optical microscopy and Raman spectroscopy (Reynard et al., 2015; Schwartz et al., 2013) on thin or thick sections and on separated and extracted powders.

Scanning electron microscopy and energy-dispersive X-ray (SEM-EDX) analyses were performed only on thin and thick sections. Sample sections were coated with a 20 nm thick carbon layer for analysis (Table 1) with EDX analysis using an AZtec Oxford Instruments system (DDI detector X-Max) installed on a Zeiss Supra VP55 scanning electron microscope operated at 15 kV and in high vacuum, except for sample ZS24, which was analyzed by electron microprobe (Masci et al., 2019). Chlorite and serpentine composition is homogeneous on the micrometer scale, except in BCS16B, where slight zoning causes a higher variability in the Mg and Si contents than in other samples.

Lizardite was extracted from bastite textures of a preserved oceanic serpentinite of Baja California Sur (BCS32), where topotactic replacement of original pyroxene produced lizardite crystals with grain sizes of a few millimeters. SEM showed these grains to contain minor tiny metallic inclusions (<0.1 % in volume) and magnetite (<1 % in volume).

Antigorites were obtained from a foliated serpentinite of Baja California Sur (BCS16A); from an antigorite vein in a serpentinite of the Erro–Tobbio massif, Western Alps (ET5); and from the Monviso massif, Western Alps (Viso4), which comprised two subsamples from the foliated matrix and from a cross-cutting vein. Raman mapping demonstrated a nearly pure antigorite composition and showed structural relationships between veins and foliation (Fig. 1a). In these samples, the main impurity is magnetite (<5 %) and sometimes chlorite (<1 %). Another sample from the same area showed more complex mineralogy partly inherited from oceanic metamorphism and was therefore not considered for bulk Mössbauer analysis (Viso6, Fig. 1b) but could meaningfully be analyzed by high-resolution XANES mapping. Antigorite from a foliated serpentinite of the Ōeyama massif does not contain magnetite but does contain trace amounts of pentlandite (<0.1 %). These homogeneous serpentinite samples contained no olivine or pyroxene.

The two studied chlorites were a Si-rich clinochlore (penninite) extracted from a foliated chloritite (ZS24) from the Zermatt–Saas zone of the Western Alps and a ferrous clinochlore (pycnochlorite) extracted from a foliated chloritite (BCS6F) from Baja California Sur. Magnetite is the main impurity in ZS24 and titanite (∼ 2 %) or clinopyroxene (<1 %) the main one in the BCS6F sample.

The oxidation state of iron was determined using Mössbauer spectroscopy. Extracted samples were gently crushed in alcohol in an agate mortar, and the magnetic fraction was removed by dipping a magnet into the powdered sample. An aliquot of 200 mg powder of each sample was loaded into a plastic holder with 12 mm diameter, giving a Mössbauer thickness between 3 and 8 mg Fe cm-2, except for BCS6F chlorite, which has a thickness of 29 mg Fe cm-2 due to its higher Fe content. Spectra were recorded at room temperature in transmission mode on a constant acceleration Mössbauer spectrometer with a nominal 1.85 GBq 57Co source in a 6 µm Rh matrix where the velocity scale was calibrated relative to α-Fe foil. We collected all spectra at ± 5 mm s-1 and additionally at ± 12 mm s-1 when magnetite peaks were present (ZS24 chlorite). Each spectrum was collected for 2–4 d, except for BCS32 lizardite (5 h) and the Viso4 antigorite vein (12 d).

Mössbauer spectra (Fig. 2) were fit using MossA software (Prescher et al., 2012) in the thin absorber approximation and additionally using the full transmission integral for thick samples (e.g., BCS6F chlorite). We used pseudo-Voigt line shapes to account for next-nearest-neighbor effects and variable area ratios for doublet components to account for preferred orientation. Hyperfine parameters (Table 2) are consistent with literature values for serpentine (Evans et al., 2012). Magnetite was detected in ZS24 by its characteristic sextets (Fig. 2g) that were constrained based on a spectrum collected of the same sample at ± 12 mm s-1. Ferric-to-total-iron ratios were calculated from the relative areas of subspectra.

Values were not corrected for differences in the recoil-free fraction, since such differences have been shown to be minimal when ferrous and ferric iron occupy sites with similar coordination (e.g., Rancourt et al., 1994; Piilonen et al., 2004). Thickness effects can be neglected, since fits using the full transmission integral gave nearly identical results to fits using the thin absorber approximation. Magnetite absorption does not overlap chlorite absorption and does not affect ferric iron determination. The Mössbauer spectrum of titanite overlaps ferric iron absorption in serpentine (Muir et al., 1984), so the ferric-to-total-iron ratio of BCS6F is likely overestimated by 1 %–2 %.

Antigorites have a remarkably constant ferric-to-total-iron ratio in the range 0.17(3)–0.23(4), in spite of the different pressure–temperature metamorphic conditions and ages of formation (Tables 1 and 2). The matrix antigorite Viso4 gives the lowest value, while all other antigorite values are the same within error bars with a mean value of 0.22(3). BCS32 lizardite gives a ferric-to-total-iron ratio of 0.49(5); Fe-rich clinochlore BCS6F and Mg-rich chlorite ZS24 give values of 0.13(3) and 0.26(3), respectively.

A compilation of determinations of Fe3+ / Fetotal ratios in serpentines and Mg-rich chlorites with XFe<0.2 by Mössbauer spectroscopy (Aja and Dyar, 2002; Bertoldi et al., 2001; Billault et al., 2002; De Grave et al., 1987; Evans et al., 2012; Fuchs et al., 1998; Goodman and Bain, 1979; Gregori and Mercader, 1994; Lougear et al., 2000; Malmström et al., 1996; Mellini et al., 2002; Mitra and Bidyananda, 2001; O'Hanley and Dyar, 1993, 1998; Peretti et al., 1992; Rozenson et al., 1979; Smyth et al., 1997; Votyakov et al., 2005; Zazzi et al., 2006), the present data and XANES (Andreani et al., 2013; Debret et al., 2014, 2015; Masci et al., 2019; Muñoz et al., 2013; Rigault, 2010; Trincal et al., 2015; Vidal et al., 2006) reveals important trends (Fig. 3).

Box plots and histograms emphasize the large overlap of Fe3+ / Fetotal ratios from XANES and Mössbauer data in lizardite and a small overlap of XANES and Mössbauer data in antigorite. For chlorites, overlap between XANES and Mössbauer data is larger than in antigorites. Mössbauer analyses show higher ferric iron content (Fe3+ / Fetotal>0.3) where secondary continental alteration by meteoritic waters was inferred for lizardites (Votyakov et al., 2005) and is supposed for antigorite and chlorite populations, as discussed in Sect. 4.1. Mean values, standard errors, standard deviations and median values are reported in Table 3 for all populations.

Typically, uncertainties in the means (2 SE, Table 3) are lower than 10 % and often lower than 5 %. Standard deviations pertinent to individual analyses are lower than 10 % for Mössbauer analyses of antigorites and chlorites and 20 % for XANES and Mössbauer data of lizardites.

Ferric iron determination influences phase equilibria modeling (Evans et al., 2012) because it is used to define the ferrous and ferric end-members of the lizardite, antigorite and chlorite solid solutions. Mg-rich terrestrial basic to ultrabasic compositions are not amenable to the use of XANES to determine accurate ferric iron contents in spite of the method's many advantages such as speed and spatial resolution, likely because of photo-oxidation. Nevertheless, XANES remains a reliable method for TOT phyllosilicates (talc and micas) and in terrestrial or extraterrestrial rocks with much higher iron contents (Beck et al., 2012; Garenne et al., 2019). Its application to Mg-rich serpentines and chlorite requires reevaluation and the use of well-characterized standards of the same mineralogical and compositional nature as the studied samples, instead of calibrations based on anhydrous compounds that are not significantly affected by photo-oxidation.

Mössbauer data are preferred over XANES data for estimating oxidation state budgets in hydrated mafic and ultramafic rocks of subduction zones (Evans, 2012; Mayhew and Ellison, 2020). Using the average ferric-to-total-iron ratios from Mössbauer data (Fig. 3) and modal mineralogies of high-pressure ultramafic rocks (Debret et al., 2014, 2015), the bulk ferric-to-total-iron ratios of the hydrated ultrabasic rocks change from 0.65(15) to 0.25(15) at the lizardite-to-antigorite transformation in Alpine ophiolites and from 0.4(1) to 0.15(5) at the antigorite–serpentinite-to-chlorite–harzburgite transition in Cerro del Almirez, at nearly constant bulk Fe content levels. This is to be compared to 0.70(15) to 0.50(15) and 0.55(5) to 0.15(5), respectively, using XANES data (Debret et al., 2014, 2015). The most important redox change is associated with the lizardite–antigorite transition occurring at 300–400 ∘C (Schwartz et al., 2013; Evans, 2004) rather than with the dehydration of antigorite occurring at 500–650 ∘C (Ulmer and Trommsdorff, 1995; Hilairet et al., 2006). With a large decrease in ferric iron and relatively small fluid release, the lizardite–antigorite transition is expected to release oxidized fluids in the cold mantle wedge. The major dehydration of antigorite is associated with a smaller decrease in ferric iron in the solid residue and should result in the release of large quantities of mildly oxidized fluids. The present estimates of Fe3+ / Fetotal ratios will affect the oxidation state of dehydration fluids in subduction zones (Debret and Sverjensky, 2017; Piccoli et al., 2019) and budgets of associated reactions affecting redox-sensitive elements such as carbon (Vitale Brovarone et al., 2017; Galvez et al., 2013). Continental alteration likely increases the Fe3+ / Fetotal ratio of hydrated ultramafic rocks, in agreement with conclusions reached from statistical analysis of available data (Mayhew and Ellison, 2020).