H 2 mobility and redox control in open vs. closed hydrothermal oceanic systems – evidence from serpentinization experiments

. New hydrothermal experiments in rapid-quench pressure vessels have been performed to investigate the effect of redox state on the serpentinization reaction. The experimental hydrogen fugacity ( f H 2 ) was controlled by monitoring the mobility of H 2 in the reacting system (internal vs. external f H 2 control). This was achieved by using either Au (H 2 impermeable) or AgPd (H 2 permeable) capsules and Ar pressurizing gas to control f H 2 . The experiments were performed with either San Carlos olivine powders or Åheim dunite chips. Water / rock mass ratios of 1–2, a total pressure of 50 MPa, and temperatures of 300 and 350 °C were investigated. Experimental durations of 30, 45, or ≈ 80 d were imposed. Serpentine production is observed in almost all experiments but is favored at 300 °C under external f H 2 control. The serpentine–magnetite assemblage is observed in Au capsules (high f H 2 ) at 300 °C, while the serpentine–hematite(–magnetite) is observed in AgPd capsules (low f H 2 ). At 350 °C, less H 2 is produced than at 300 °C and the serpentine–hematite(–magnetite) assemblage is present in both Au and AgPd capsules. Brucite is absent and this is interpreted to reﬂect both the initially oxidizing conditions and relatively low serpentine production in our experiments. Differences in product phase assemblages found in this study imply that natural serpentinization reaction mechanisms vary with redox conditions, and consequences for H 2 production ﬂuxes and rates can be expected. The high-f H 2 (reduced) internally controlled experiments simulate low-permeability “closed” oceanic hydrothermal systems. The low-f H 2 (oxidized) externally controlled experiments are analogous to “open” oceanic hydrothermal systems where serpentinization is driven by tectonically aided inﬁltration of an external ﬂuid.

The impact of the fluid / rock ratio on serpentinization has been explored in several thermodynamic studies (Klein et al., 2013;Malvoisin, 2015;Ely et al., 2023).Comparatively few efforts have been put into experimentally coupling fluid-rock reaction and fluid circulation (Godard et al., 2013;Farough et al., 2016;Tutolo et al., 2018;Peuble et al., 2019;Osselin et al., 2022).In this paper, we present a set of hydrothermal experiments that enable the effect of H 2 mobility on the serpentinization reaction to be discussed.Conceptually, the mobility of H 2 is used as a proxy of fluid circulation, open vs. closed, in oceanic hydrothermal systems.H 2 is the closest to a perfectly mobile component (i.e., components whose chemical potential in the system is imposed by an external reservoir; Weill and Fyfe, 1964;Korzhinskii, 1965), and f H 2 is the key redox variable in hydrous systems (e.g., Eugster, 1957;Scaillet et al., 1992).In this paper, experiments have been designed in an original way enabling the mobility of H 2 to be controlled and varied.Below, the experimental and analytical methodologies are presented and the experimental result detailed.We show that significant differences appear in experimental phase assemblages and proportions, and in extents of reaction depending on H 2 mobility, and discuss the implication for serpentinization processes at mid-ocean ridges (MORs).

Background
In the following, redox states will be expressed preferentially in terms of f H 2 and to a lesser extent in terms of f O 2 .The f H 2 is a directly measurable variable (Shaw, 1963), and redox states can be conveniently varied experimentally by changing the f H 2 (Chou, 1987;Chou and Cygan, 1990;Scaillet et al., 1992;Schmidt et al., 1995).In hydrous systems, the two redox variables, f H 2 and f O 2 , are related through the following equilibrium: whose equilibrium constant (K (R1) ) is where f H 2 O is the fugacity of water.From Eq. ( 1), f H 2 can be expressed as which illustrates that, for a hydrothermal fluid at constant and H 2 O) are related to their respective fugacities by the following general relation: where φ i is the fugacity coefficient of component i and X i the mole fraction of i in the fluid mixture.For most common geological fluids, X O 2 is negligible, whereas X H 2 is readily measurable, reaching values in the same range as X H 2 O under very reducing conditions (Eugster and Skippen, 1967;Frost et al., 2013).For example, at 300 °C and 50 MPa, MPa for an oxidizing f O 2 corresponding to the hematite-magnetite (HM) buffer, while, for a reducing f O 2 corresponding to the fayalite-magnetite-quartz (FMQ) buffer, f H 2 = 1.7 × 10 −1 MPa (Table 1).Below, two cases will be considered depending on how the f H 2 variable is controlled.In the first case, the f H 2 variable is controlled internally; i.e., H 2 is produced by the advancement of the serpentinization reaction and allowed to accumulate in the reacting system (internal f H 2 control; Fig. 1a).In the second case, f H 2 is imposed externally from outside the reacting system (external f H 2 control; Fig. 1b).

Starting materials
Two distinct starting materials were used in the experiments: San Carlos olivine and Åheim dunite.The San Carlos olivines (Fo 90.5 ; see Table A1 in Appendix A for detailed composition) were separated by hand from slightly disaggregated San Carlos xenoliths.In order to eliminate all contaminants (spinels, pyroxenes), only grains with a sufficient size (larger than ≈ 500 µm to allow a precise examination with binoculars) were kept.After several ultrasonic cleanings followed by abundant rinsing with deionized water, all crystals suspected of surface or inclusion contamination were removed.The Åheim dunite contains, in addition to olivine (Fo 94.0 ; Table A1), about 10 vol % orthopyroxene (En 93.9 ; Table A1), 0.5 vol %-1 vol % chromite, and 1 vol %-5 vol % clinochlore.Hydrous silicates such as serpentine, talc, phlogopite, and tremolite have been reported to be present in trace amounts (Berckhemer et al., 1982;Jackson et al., 1992), although none were found from SEM examination.In the same way and in contrast with the observations of (Jackson et al., 1992), no magnetite was observed in our Åheim dunite sample.This is consistent with the magnetic data (low magnetite mass fraction (MMF) of 2.5 × 10 −3 wt %; see Sect.3.4).

Experimental charges and capsules
Experimental parameters are summarized in Table 2.The San Carlos olivines were used in powder form, obtained by grinding the olivine crystals in an agate mortar.The resulting powders were sieved to different granulometric fractions: 0.5-1 mm, 45-50 µm, 25-32 µm, and < 10 µm.Above 32 µm, granulometric separations were performed with stainlesssteel sieves.Below 32 µm, granulometric separations were performed with nylon sieves and a vibrating table.In both cases, powders were abundantly rinsed with deionized water in order to remove the smallest particle sizes.The Åheim dunite was used as experimental starting materials in the form of millimeter-sized chips.Experimental charges, consisting of either the olivine or the dunite (R) plus demineralized water (W ), were loaded in noble metal capsules (4.2 or 5 mm diameter, 30-35 mm length, wall thickness 0.2 mm) which were welded shut.Capsules were weighted at each step during preparation and cooled during welding to ensure that no water loss occurred.Masses of reactants were adjusted to yield W/R ratios of either 1 or 2 (Table 2).Capsules were checked for leaks (by placing them in an oven and weighting) and then inserted in the pressure vessel.
Control of f H 2 was achieved by using two different capsule materials, Au and Ag 25 Pd 75 or Ag 23 Pd 77 (wt %; AgPd in the following text).Au capsules are almost impermeable to H 2 at the temperatures investigated in this study (Gunter et al., 1987).Therefore, in experiments performed with Au capsules, the f H 2 is intrinsically controlled by and depends https://doi.org/10.5194/ejm-36-555-2024(1964).b From Burnham et al. (1969).c Calculated with SUPCRT92 from Johnson et al. (1992).d From Myers and Eugster (1983) in Chou (1987).e From Haas and Robie (1973) in Chou (1987).on the advancement of the serpentinization reaction (internal f H 2 control).H 2 accumulates inside the capsule, and this leads to high-f H 2 and thus low-f O 2 conditions (Fig. 1a).Our experimental setup does not allow a direct measurement of f H 2 inside Au capsules.However, the f H 2 can be estimated from the H 2 molality in hydrothermal fluids (m H 2 ) determined in other closed-system olivine serpentinization experiments (Berndt et al., 1996;Allen and Seyfried, 2003;Mc-Collom et al., 2016;Fauguerolles, 2016).The f H 2 and the m H 2 are related by where K (R2) , computed in the following using SUPCRT92 (Johnson et al., 1992), is the equilibrium constant of and γ H 2 is the molar activity coefficient of H 2 (aq).The latter can be assimilated to the molar activity coefficient of CO 2 (aq) (Garrels and Thompson, 1962), itself calculated from the work of Drummond (1981).Experimental m H 2 data used to constrain f H 2 in our closed-system experiments are detailed in Table 3.When interpolated at 720 h, the typical duration of our experiments, f H 2 at 300 °C varies between 4.78 and 0.18 MPa, which corresponds to HM = −8.77and HM = −5.89,respectively.Differences in f H 2 between studies reflects the difference in grain size, smaller in the experiments of Berndt et al. (1996) and Fauguerolles (2016), which increases the reaction rate and associated H 2 production and increases f H 2 .Reaction path models (Klein et al., 2013) predict a maximum production of H 2 around 322 °C followed by a drastic decrease at higher temperatures.In contrast, AgPd capsules are quite permeable to H 2 .At 300 °C, H 2 permeability is 9 log units higher in AgPd than in Au and 8 log units higher at 350 °C (Gunter et al., 1987).By using a H 2 permeable capsule, the f H 2 inside the capsule becomes rapidly equal to the f H 2 outside the capsule (i.e., of the pressurizing gas).Volume considerations (the volume of pressurized gas in the vessel is about 10 cm 3 , whereas the volume of the experimental capsule is about 0.1 cm 3 ) dictate that, in this case, the f H 2 within the capsule is imposed by the f H 2 of the pressurizing gas (external f H 2 control; Fig. 1b).The f H 2 imposed by the Ar pressurizing gas has been constrained both directly and indirectly (see Appendix B).At 350 °C, 50 MPa, the f H 2 is 0.9 × 10 −4 MPa, which corresponds to HM = +1.19.At 300 °C, no such measurements are available, but the temperature dependence of the f H 2 in such an oxidizing f O 2 range is small, and a maximum f H 2 of 2.1×10 −4 MPa (corresponding to the value of the HM buffer at 300 °C; Table 1) can be assumed.Therefore, our internally and externally controlled experiments differ by 3.45-4.4log units in f H 2 (5.9-8.8 log units in f O 2 ) at 300 °C and, at 350 °C, by 2.55-3.4log units in f H 2 (5.1-6.8 log units in f O 2 ).

Experimental equipment
In this study, all experiments were performed in René 41 rapid-quench cold-seal pressure vessels (Rudert et al., 1976;Pichavant, 1987;Schmidt et al., 1995) pressurized with Ar.Experimental temperatures, monitored permanently with an external thermocouple, were controlled with a Eurotherm regulator.Temperatures inside the vessel were calibrated under pressure using two thermocouples (precise to ±2 °C according to the manufacturer), spaced 3 cm apart and positioned at the points occupied by the ends of the capsule.In order to minimize the thermal gradient within the capsule, the vessel was positioned in the furnace so that the tempera-ture difference between the calibration thermocouples was less than 1 °C.Pressure was read with a tube manometer to within ±2 MPa.In this study, pressure was kept constant (50 MPa) and two temperatures, 300 and 350 °C, were investigated.Experiments lasted for 30, 45, or 80 d (Table 2).Once completed, the experiment was quenched by removing the pressure vessel from the furnace and allowing it to cool for ≈ 1 h down to room conditions.When cold, the vessel was opened and the capsule extracted.It was cut parallel to its long axis so that the entire sample volume was exposed and recovered.

Analytical methods
Compositions of the starting San Carlos olivine and the starting olivine and orthopyroxene of the Åheim dunite have been determined with a Cameca SX Five electron microprobe (ISTO, Orléans).Analyses were carried out in punctual mode using a 15 kV acceleration voltage, a 6 nA sample current, and a 10 and 5 s acquisition time on peak and background respectively.Solid products recovered after the experiments were dried at 110 °C for 24 h.A fraction was ground and mounted in a capillary for X-ray diffraction (XRD) analysis.An Inel CPS120 diffractometer equipped with a curved detector and a Co anticathode was used (ISTO, Orléans).Magnetic data including hysteresis parameters (saturation magnetization: J s ; saturation remanent magnetization: J rs ; intrinsic coercivity: H c ; remanent coercivity: H cr ) were obtained using a vibrating sample magnetometer at room temperature (MC MicroMag 3900 Series, IPGP Paris).The magnetite mass fraction (MMF) was obtained using the calibration between the amount of magnetite and J s (proportionality factor of 92 A m 2 kg −1 determined from powder mixtures made of synthetic forsterite and magnetite; Malvoisin et al., 2012b).Progress of the serpentinization reaction was calculated from the amount of magnetite produced (MMF) using model 3 of (Malvoisin et al., 2012b).Another fraction of experimental products was selected for detailed textural characterization and phase identification using scanning electron microscopy (SEM) and Raman micro-spectroscopy.Run products were either embedded in epoxy and polished or directly glued on a glass plate.The SEM of the joint BRGM-ISTO facility (Tescan Mira 3 XMU, ISTO Orléans) was used both in secondary (SE) and back-scattered electron (BSE) modes.To reduce charging effects, samples were sputtered with carbon before analysis.Raman characterization of oxides was performed using a Jobin-Yvon Labram HR800 Raman spectrometer (LPG, Nantes) equipped with a CCD detector, a 532 nm solid-state laser, a 1800 grooves mm −1 grating, and a 100× Olympus objective in confocal mode.Acquisition times were 5 × 60 s.To avoid oxide deterioration (de Faria et al., 1997), the laser beam power was optimized on magnetite and hematite powders provided by Sigma Aldrich and comparable in size to those observed in our synthesized products.Thus, a laser power output set at 50 mW (corresponding to 1.5 mW at sample level) was used in conjunction with a filter reducing the laser power by a factor 10. Raman spectra of iron oxides could be acquired only on samples with the largest grain sizes, after pre-localization of iron oxides using SEM images.
Iron oxide characterization (nature and proportion) was carried out by coupling four methods: SEM, XRD, Raman micro-spectroscopy, and magnetic properties.SEM informed on the presence of iron oxides and on textural relations with other phases and provided a relative quantification of their abundances in the samples.Identification of hematite was based on XRD and Raman data.The distinction between magnetite and maghemite (γ Fe 3+ 2 O 3 ), while being ambiguous from XRD (Malvoisin et al., 2012b), was achieved from the Raman data, whereas the magnetic properties, in particular the J s measurements, allowed quantification of amounts of magnetite or maghemite present.
Due to the low serpentine abundances and the similarities between XRD spectra of chrysotile and lizardite, it was not possible to clearly discriminate between the different serpentine polymorphs by this method.In contrast to iron oxides, attempts to characterize serpentine polymorphs by Raman spectroscopy (Lemaire, 2000;Auzende et al., 2004;Schwartz et al., 2013;Rouméjon et al., 2015) were inconclusive, and the identification of these polymorphs in this study is based on textural parameters, notably the characteristic acicular habit (fibrous texture) of the chrysotile.

San Carlos olivine
Fourteen experiments, half in Au and half in AgPd and covering a range of initial grain sizes, durations, and temperatures are reported in Table 2.

Textures
Textures of experimental products are illustrated in Fig. 2. For the coarser initial grain sizes (0.5-1 mm), serpentinization products are concentrated within the starting olivine crystals, in large veins (Fig. 2a and b) or gulfs (Fig. 2c) rather than at its free surface.Vein widths are approximately 5 µm, and two different vein types can be distinguished from SEM observations on orthogonal sections.The first type is mainly composed of fibrous material, interpreted to be chrysotile (see Sect. 3.4).Iron oxides occur as small crystals irregularly dispersed along the vein.The vein is not completely filled by product phases, and there is a void between olivine and the material crystallized in the vein (Fig. 2a).The second type, only observed in Au capsules, is texturally similar to the first, being also symmetrical with respect to the center of the vein.It comprises chrysotile and iron oxides, but the fibrous phase displays a contrast difference: lighter near the olivine and darker in the center of the vein.As for the first type, a void is observed between the light band and olivine (Fig. 2b).Whatever the type of vein, iron oxides are almost always sandwiched between olivine and serpentine (Fig. 2a  and b).This textural relationship between minerals is also observed in dissolution gulfs present at the surface of the crystals (Fig. 2c).An extensive search for brucite in the veins with SEM and Raman proved negative.
For smaller initial grain sizes, fibrous chrysotile develops on the surface of the residual olivine crystals where it forms a 5 µm layer (Fig. 2d and e).Olivine can be completely rimmed by serpentine (Fig. 2f).Other reaction products include iron oxides, present locally in very different proportions and sizes (Fig. 2d and g).Brucite is absent.

Mineral assemblages
Olivine, detected by XRD in all charges, persists in all experimental products (Table 2).Similarly, serpentine and iron oxides are present in all experimental charges on the basis of SEM observations (Fig. 2), although they are detected by XRD in only a few experiments (Table 2).Serpentine is detected by XRD in all charges with the smallest initial grain sizes (10 µm) and in only one with an intermediate initial grain size (ol300Pd45; Fig. 3, Table 2).It coexists on XRD patterns with iron oxides in most charges (ol300Pd45, ol300Au10, ol300Pd10 at 300 °C; ol350Pd10 at 350 °C) and is more rarely found in isolation (ol350Au10 at 350 °C; Fig. 3, Table 2).Examination of XRD patterns suggest that serpentine is produced in generally small amounts, except in ol300Pd10 (Fig. 3).Despite extensive tracking with SEM using EDS (energy-dispersive X-ray spectroscopy) and WDS (wavelength-dispersive spectrometry) element maps, brucite was not found and must be considered as absent from San Carlos charges.
Experimental iron oxide products include magnetite and hematite.Magnetite was positively identified by Raman in the four experiments with coarse initial grain sizes, performed both in Au (ol300AuG, ol350AuG) and AgPd (ol300PdG, ol350PdG; Fig. 4, Table 2).Maghemite (spectrum characterized by a doublet at 665 and 621 cm −1 , Legodi and de Waal, 2007, Fig. 4) was not found despite extensive search.In experiments performed with small initial grain sizes in Au at 300 °C (ol300Au10) and in AgPd at 350 °C (ol350Pd10), the iron oxide peaks at 35.13°2θ and 50.52°2θ in XRD spectra are attributed to magnetite (Fig. 3, Table 2) because maghemite is absent in the corresponding experiments performed with large initial grain sizes (respectively ol300AuG and ol350PdG, Fig. 4).Following the same reasoning, the increase of J s (Table 4) relative to this of the starting olivine (baselines in Fig. 5) is attributed to the presence of magnetite and allows the calculation of magnetite mass fraction (MMF) in products (Table 2).Magnetic measurements thus highlight the formation of magnetite in two supplementary samples: ol300Au45 and ol300Pd10 (Fig. 5a, Table 2).They are also in good agreement with the XRD https://doi.org/10.5194/ejm-36-555-2024measurements since the two samples with the highest MMF are those for which magnetite is detected by XDR (Fig. 3, Table 2).Magnetite is thus confirmed in eight charges, half in Au half in AgPd, while it forms preferentially at 300 rather than 350 °C (five vs. three occurrences, respectively).
Hematite was identified by Raman in three of the four coarse-grained samples analyzed (ol300PdG, ol350AuG, ol350PdG; Fig. 4, Table 2) and by XRD in three charges: two with the smallest initial grain sizes (ol300Pd10, ol350Pd10) and, only one, same as for the serpentine, with intermediate initial grain sizes (ol300Pd45; Fig. 3, Table 2).Hematite is preferentially associated with the AgPd experiments.It is absent in Au at 300 °C (ol300AuG) but present in the corresponding charge at 350 °C (ol350AuG; Fig. 4, Table 2).A total of six charges, half at 300 °C and half at 350 °C, contain hematite.

Magnetic properties
The hysteresis data for experimental products (Table 4) are plotted in a J rs /J s vs. H cr /H c diagram (Day diagram;Fig. 6, Day et al., 1977) where J rs /J s is the ratio of saturation rema-nent magnetization to saturation magnetization and H cr /H c the ratio remanent coercivity to intrinsic coercivity.The Day diagram is a useful way to discriminate between various magnetic mineralogies and grain sizes using magnetization and coercivity parameters (Dunlop, 2002;Malvoisin et al., 2012b).For magnetite, the single-domain (SD, J rs /J s > 0.5 and H cr /H c ≈ 1, Fig. 6) corresponds to particle sizes smaller than ≈ 0.05-0.08µm, while the multidomain (MD; J rs /J s < 0.05 and H cr /H c > 4, Fig. 6) is reached for magnetite particles larger than 3-15 µm (Oufi et al., 2002).Between these two grain sizes, the magnetic properties evolve in the pseudosingle-domain (PSD) transition zone.The magnetic properties of our experimental samples (only those with MMF above the baselines; see Fig. 5) are compared with those from other experimental studies and natural serpentinites from the literature.All our experimental magnetites plot in the PSD field (Fig. 6).Magnetites produced in Au capsules at 300 °C are in good agreement, in terms of J rs /J s , with other experimental magnetites at 300 °C (Malvoisin et al., 2012b), although the latter extend to higher H cr /H c values (Fig. 6).Magnetites from AgPd have systematically higher J rs /J s than magnetites from Au capsules (Fig. 6).Taken together,    the experimental magnetites follow a trend broadly similar to that of magnetites from natural systems (Oufi et al., 2002), although, for a given H cr /H c , the experimental magnetites plot at higher J rs /J s than the natural (Fig. 6).

Influence of capsule material, temperature, and grain size
All 300 °C charges contain serpentine and iron oxides irrespective of the capsule material.However, charge ol300Pd10 has the strongest serpentine XRD signal in our sample set (Fig. 3).In comparison, serpentine is only present in small amounts in the 350 °C charges, whether performed in Au or AgPd (ol350Au10, ol350Pd10, Fig. 3).Therefore, the data suggest a decrease in serpentine production upon increasing temperature from 300 to 350 °C and a higher serpentine production in AgPd than in Au at 300 °C.This latter statement is supported by the XRD detection of serpentine for intermediate grain sizes only in AgPd (ol300Pd45; Fig. 3), although in smaller amounts than for ol300Pd10.We conclude that serpentine preferentially forms at 300 °C in AgPd capsules and in experiments performed with small initial grain sizes.
A first-order change in the nature of the iron oxide phase is observed between the Au and the AgPd experiments at 300 °C.The former contains magnetite and the latter hematite, although a small amount of magnetite is present together with hematite in ol300PdG and ol300Pd10 (Figs. 4, 5, Table 2).This subdivision between Au and AgPd experiments is less strongly marked at 350 °C since hematite coexists with magnetite in ol350AuG, ol350PdG, and ol350Pd10, but magnetite is present in low amounts in ol350AuG (Fig. 4, Table 2).SEM observations show that iron oxides are generally more abundant and grow bigger at 300 °C (up to 5 µm) than at 350 °C (≤ 1 µm; Fig. 2a-d and g), consistent with the evolution of magnetic properties with increasing temperature.At 350 °C, magnetite has a lower H cr /H c and higher J rs /J s than at 300 °C, which indicates a trend toward the SD field and thus smaller grain sizes (Fig. 6).The magnetite production increases with decreasing grain size at 300 °C in Au (Fig. 5a, Table 2).The same type of evolution is observed in AgPd, although the low MMFs (Fig. 5a) indicate that magnetite production is much more limited than in Au.In comparison, at 350 °C, no such increase of the MMF is observed in Au (Fig. 5b and Table 2) and iron oxides occur in amounts insufficient to be detected by XRD (Table 2).In ol350AuG, the Raman data show that hematite is the predominant iron oxide and that magnetite is present only in small amounts (Fig. 4).MMFs are also low in AgPd at 350 °C (Fig. 5b), indicating a very limited magnetite production at the exception of the charge performed with the smallest grain size (ol350Pd10;  2).Relative XRD peak intensities in ol300Pd45 and ol300Pd10 (Fig. 3) imply that hematite production increases with decreasing initial grain size.At 350 °C (ol350Pd45), hematite is not detected by XRD at the difference of at 300 °C (ol300Pd45) for the same initial grain size (Fig. 3, Table 2), indicating a decrease in hematite production from 300 to 350 °C.We conclude that hematite preferentially forms at 300 °C in AgPd capsules and in experiments performed with small initial grain sizes.
Charge ol350Pd10 (Table 2) combines the presence of iron oxides in amounts sufficient to be detected by XRD (Fig. 3), a very high J s (the highest of the entire dataset, Table 4), and a relatively small serpentine production (Fig. 3).Hematite is detected by XRD (Fig. 3), and the other iron oxide in the pattern is interpreted to be magnetite since magnetite (but not maghemite) has been identified in the Raman spectra of the corresponding coarse-grained charge (ol350PdG; Fig. 4).Thus, the strong J s implies a very high MMF, in fact the highest MMF in our sample set (Fig. 5b, Table 2).This indicates an exceptionally high magnetite production in this 350 °C AgPd charge performed with the smallest initial grain size.

Åheim dunite
Two experiments are available, each for internal and external control of f H 2 (i.e., Au and AgPd capsules respectively) at 350 °C for similar durations (≈ 1050 h) and on a relatively coarse-grained starting material (≈ 2000 µm; Table 2).Both olivine and orthopyroxene persist in experimental products, although orthopyroxene appears more reacted.The XRD data indicate that the relative proportions of orthopyroxene decrease more markedly than olivine (Fig. 7), and dissolution textures are more clearly marked on orthopyroxene than on olivine (Fig. 2h).Serpentine appears only as fibrous chrysotile (Fig. 2h), while brucite was not found.In the Au capsules, no iron oxide could be detected either by XRD or SEM, consistent with the MMF being only slightly higher than in the starting material (2.5 × 10 −3 wt %; see Sect.3.1 and Table 2).In contrast, AgPd capsule products contain an iron oxide phase detected by SEM (Fig. 2h).This iron oxide is probably responsible for the weak peaks observed on the XRD pattern (Fig. 7).No reliable Raman analysis could be obtained for this oxide, tentatively interpreted to be maghemite according to XRD pattern, knowing that this method does not allow magnetite to be clearly excluded (see Sect. 3.4).It occurs either in isolation or associated with Fe-Ni sulfides and in fractures or at grain boundaries.The measured MMF is substantially higher than in the Au capsule charge (Table 2), being also 1 order of magnitude higher than the starting material.Using a proportionality factor for maghemite of 78 A m 2 kg −1 (Cullity, 1972), the increase  2).Mineral abbreviations are the same as in Table 2. Peaks which are not highlighted correspond to olivine.All peaks higher than 15 000 counts are clipped.Relative peak intensities demonstrate that orthopyroxene reacts preferentially than olivine.Insets are ×6 magnifications of diffractograms centered on the 50.76°2θ and 67.67°2θ peaks which highlight the presence of iron oxide in du350PdG consistent with the presence of maghemite (see text).
in J s is compatible with a nominal weight of maghemite of 0.036 wt % (0.030 wt % if the iron oxide phase is magnetite).The possible occurrence of maghemite in the Åheim dunite experiment carried out in AgPd capsules is consistent with the preferential formation of hematite in the San Carlos olivine experiments carried out in AgPd.However, the cause of the polymorph change (αFe 3+ 2 O 3 for hematite vs. γ Fe 3+ 2 O 3 for maghemite) remains to be clarified.

Control of redox conditions in the experiments
Results from this study directly depend on the reliability of our methods to control the experimental redox conditions.First, it is necessary that the H 2 generated by the serpentinization reaction remains confined within the Au capsule to sustain a reduced environment throughout the entire duration of the experiment.This necessitates that hydrogen loss (by permeation through the capsule, Harvie et al., 1980) is of negligible importance, despite our relatively long experimental durations.Second, H 2 must escape rapidly out of the AgPd capsule so that oxidized redox conditions, controlled by the Ar pressurizing medium, are imposed to the charge.
Below, we summarize the evidence confirming that both conditions were fulfilled in our experiments.
In experiments performed with Au capsules, the negligible importance of hydrogen loss is demonstrated by the following arguments: 1.Among noble metals used as capsules, hydrogen permeabilities are the lowest for Au (Chou, 1986;Gunter et al., 1987).
2. Au has been classically considered to be almost impermeable to hydrogen at low to medium temperatures, being used as the external container in double capsule assemblies used to control experimental redox states (Eugster, 1957).The permeation of hydrogen in Au is usually considered negligible below 400-450 °C (see Fig. 3 in Chou, 1986).
3. Hydrogen permeation rates have been calculated with equation 10 in Harvie et al. (1980)  For experiments performed in AgPd capsules, the following arguments suggest that oxidized redox conditions were rapidly imposed: 1.A large contrast in hydrogen permeability constants exists between Au on the one hand and AgPd alloys on the other hand.Hydrogen permeabilities in Ag 25 Pd 75 alloys are 8-9 orders of magnitude higher than in Au, being also higher than in pure Pd (Chou, 1986;Gunter et al., 1987).
2. Magnetite is oxidized to hematite in the A1 test experiment performed at 350 °C (Table B1 in Appendix B).In the A4 hydrogen sensor experiment, the determined f H 2 (0.9 × 10 −4 MPa) is lower than that of the HM buffer (Table B1).Therefore, redox conditions more oxidizing than the HM buffer are recorded in the AgPd experiments at 350 °C and also presumably at 300 °C since f H 2 varies little with temperature under strongly oxidizing conditions.
3. Hydrogen permeation rates for Ag 25 Pd 75 capsules have been calculated under conditions same as above using permeability coefficients from Gunter et al. (1987)
The general mechanism can be conveniently described by the following global reaction: Our experimental results under conditions of internal f H 2 control (Au capsule experiments) at 300 °C are broadly consistent with Reaction (R3).Olivine is consumed, and both serpentine and magnetite are produced (ol300AuG, ol300Au45, ol300Au10; Table 2).However, brucite is not present in our charges (whatever the granulometry of the starting material), in contrast with what is expected from Reaction (R3).In comparison, we note that Malvoisin et al. (2012a) reported brucite in their experiments performed under the same conditions (Au capsules and temperatures) and with the same starting material (San Carlos olivine).Several explanations have been proposed for the absence of brucite in experimental and natural serpentinites.For example, a mechanism based on precipitation of carbonates could account for the absence of brucite (Kelemen and Matter, 2008), but it is irrelevant in our case since our experimental products contain no carbonate phase.High aqueous silica activities (a SiO 2 (aq) ) imposed by dissolution of local orthopyroxene or inherited by the fluids from previous encountered lithologies may also explain the absence of brucite (Frost and Beard, 2007;Frost et al., 2013;Tutolo et al., 2018).This mechanism could account for the lack of brucite in the Åheim dunite but not in the San Carlos experiments.However, contamination of silica from the agate mortar during preparation of the starting olivine powders is a possibility.
According to Reaction (R3), serpentinization of Fo 91 (Mg# = 0.91) olivine yields serpentine with Mg# > 91 coexisting with brucite with Mg# < 91 (Bach et al., 2006;Malvoisin et al., 2012a;Klein et al., 2013).The distribution of Mg and Fe between serpentine and brucite determines the amount of "free" Fe which is mobilized as magnetite.For example, for the serpentine-magnetite assemblage, the Mg# of serpentine increases with increasing f O 2 (decreasing f H 2 ) and decreasing a SiO 2 (aq) (Frost et al., 2013)  where the incorporation of Fe 3+ in serpentine (Seyfried et al., 2007;Marcaillou et al., 2011;Andreani et al., 2013) is not considered.The transition from a serpentine-magnetite to a serpentine-brucite assemblage marks a decrease in f O 2 (or increase in f H 2 ) at constant a SiO 2 (aq) (Frost et al., 2013)  Thus, a serpentine-magnetite assemblage is indicative of conditions more oxidizing than a serpentine-brucite assemblage.In our internally controlled experiments, H 2 accumulation in the Au capsule is accompanied by an increase in f H 2 (and a decrease in f O 2 ).However, because serpentine production is small in our experiments (see Sect. 5.2.2), f H 2 did not increase (and f O 2 did not decrease) sufficiently for the stabilization of a serpentine-brucite assemblage, and the system remained within the serpentine-magnetite domain in terms of f O 2 and a SiO 2 (aq) .A similar type of interpretation was proposed by Okamoto et al. (2011), who observed the appearance of brucite only after ≈ 500 h in hydration experiments of olivine plus water at 250 °C vapor-saturated.The late formation of brucite was attributed to a progressive evolution of the fluid from early oxidizing (i.e., sulfate reduction consumes H 2 ) to more reducing (serpentinization leads to H 2 accumulation) conditions during the experiments (Okamoto et al., 2011).For lower f H 2 (and higher f O 2 ) as in the externally controlled experiments in AgPd, the appearance of brucite would be even more unlikely, although we recognize that the a SiO 2 (aq) also comes into play (see Godard et al., 2013).The presence of brucite in the experiments of Malvoisin et al. (2012b, a) could reflect conditions more reducing (higher f H 2 ) than in our internally controlled experiments, which could have been promoted by globally higher serpentine production and lower W/R ratios (up to 5 times lower) than in this study.Indeed, a lower W/R ratio implies, for identical amounts of Fe 2+ oxidized into Fe 3+ and consequently for identical amounts of produced H 2 , a higher concentration of H 2 and so a higher f H 2 (Klein et al., 2013).At 350 °C, the absence of brucite is less surprising since it is actually not predicted to form at the expense of Fo 90 , only small amounts of serpentine and trivial amounts of magnetite being expected (Klein et al., 2009(Klein et al., , 2013)).
Reaction (R4) can also be applied to explain the banded texture of reaction products in veins within the San Carlos olivines (Fig. 2b).This texture is interpreted as a chemical zonation between an early relatively Mg-rich (dark part) and a later Fe-rich (light part) serpentine.According to Reaction (R4), the transition from the dark to the light part would be explained by growth of serpentine under more reducing (higher f H 2 ) and/or higher a SiO 2 (aq) .
For conditions of external f H 2 control (AgPd capsules) at 300 °C, serpentine production is higher than in the Au, and the iron oxide phases correspond to hematite (ol300PdG, ol300Pd45, ol300Pd10), although magnetite can be still present (ol300PdG and ol300Pd10; Fig. 5a, Table 2).Thus, Reaction (R3) can be rewritten as (Mg 0.91 Fe 0.09 ) 2 SiO 4 Olivine(Fo 91 ) Reaction (R6) can be viewed as the combination of Reaction (R3) with the following reaction: which is the HM oxygen buffer reaction.It shows that, with hematite instead of magnetite as a product phase, the proportion of Fe 2+ oxidized to Fe 3+ is higher than in Reaction (R3).This implies in principle that more H 2 should be produced from Reaction (R6) than from Reaction (R3) (e.g., Frost, 1985).However, this conclusion needs to be confirmed in practice with the analysis of product phases in the two model Reactions (R3) and (R6).Andreani et al. (2013) have shown that, during the early stages of serpentinization, Fe 3+bearing serpentine is produced and this generates more H 2 than expected from the amount of crystallized iron oxides (magnetite in their case).

Reaction progress
As is the case for other metasomatic processes, serpentinization is an irreversible transformation (Fonteilles, 1978).Previous experimental work has shown that serpentinization is a relatively slow reaction (Malvoisin et al., 2012a, b).Serpentine progressively forms when increasing the experimental duration, but bulk equilibrium at the scale of the whole charge (which would correspond to a 100 % reaction progress with all olivine present transformed to serpentine; see Sect.5.2.1) is usually not reached.The advancement of the serpentinization reaction has been classically evaluated from the mass fraction of magnetite in reaction products (Oufi et al., 2002;Malvoisin et al., 2012a, b).This approach remains valid as long as magnetite is produced as the only iron oxide in the reaction.This situation occurs in our internally controlled experiments (Au capsules) at 300 °C.Using model 3 of Malvoisin et al. (2012b), which relates the amount of magnetite formed and the reaction progress (S) through the amount of water incorporated in hydrous phases (serpentine, brucite), advancements of 0.9 %, 1.3 %, and 5.4 % are obtained for ol300AuG, ol300Au45, and ol300Au10, respectively (Table 2).Thus, the serpentinization reaction is relatively little advanced in our 300 °C experiments, in agreement with the low amounts of serpentine in XRD spectra and the lack of brucite in experimental products.The S values demonstrate a progressive increase of reaction kinetics with decreasing initial grain size at 300 °C, although durations are not exactly the same between the three experiments (2010 h for ol300AuG vs. 720 h for ol300Au45 and ol300Au10; Table 2).At 350 °C in Au charges, the advancement of reaction cannot be estimated because the MMFs are not systematically higher than the baseline (see Fig. 5).Furthermore, the presence of hematite (at least in ol350AuG) excludes the use of this method.If comparison is made for similar durations and experimental parameters, reaction advancements in our 300 °C Au experiments are significantly smaller than those observed by Malvoisin et al. (2012a).For example, for a 720 h run with an initial grain size of 38-50 µm at 300 °C, an advancement of about 6 % was measured (Run #10 in Malvoisin et al., 2012a), vs. 1.3 % in this study for a similar granulometry (ol300Au45, Table 2).This stresses differences in reaction kinetics between studies, although there is also a substantial intrinsic variability between experiments.
As an illustration, for different runs of 720 h duration with a 5-15 µm initial grain size at 300 °C, reaction advancements cover a large range from 13 to 70 % (Runs #16 and #22, respectively, in Malvoisin et al., 2012a).In ol300Au10 (our experiment the closest in duration and initial grain size for comparison), the advancement is 5.4 % (Table 2).This suggests the existence of rate-limiting mechanisms acting on the kinetics of the serpentinization reaction in addition to the initial granulometry.One of these mechanisms could be armoring due to precipitation of secondary phases, which limits access of the fluid to the olivine surface (Oelkers et al., 2018).This mechanism would be particularly critical if, as suggested by Malvoisin et al. (2012a), reaction kinetics are correlated to surfaces of olivine grains and the dissolution process of olivine acts as the limiting factor.Another indication for the slow serpentinization kinetics in our experiments is provided by the absence of brucite since it is symptomatic of low-f H 2 conditions consistent with a small serpentine production (see above and Okamoto et al., 2011).
In our externally controlled experiments (AgPd capsules), the continuous removal of H 2 from the charge toward the pressurizing medium could act to promote the advancement of the serpentinization reaction (Le Chatelier principle).Since hematite sequesters Fe, the proportion of magnetite is necessarily impacted, and models based on the production of magnetite to quantify the advancement of the serpentinization reaction are no longer applicable.However, the XRD and MMF data show that the highest production of serpentine and magnetite are reached in the externally controlled experiments (ol300Pd10 at 300 °C and ol350Pd10 at 350 °C, respectively).An enhanced advancement of the serpentinization reaction in externally controlled experiments is also consistent with ol300Pd45, the only charge with initial grain size > 10 µm having serpentine and hematite detected by XRD.The increase in serpentine proportion from ol300Pd45 to ol300Pd10 (Fig. 3), i.e., upon decreasing the initial grain size, is not accompanied by an increased magnetite production since MMF values remain practically unhttps://doi.org/10.5194/ejm-36-555-2024changed (Fig. 5a, Table 2).Therefore, in the externally controlled experiments, the advancement of the serpentinization reaction and the production of magnetite are clearly decoupled.This has important consequences for the estimation of serpentinization rates in natural oceanic systems (Cannat et al., 2010).

Role of temperature
Experiments performed on San Carlos olivines have demonstrated a decrease in serpentine production upon increasing temperature from 300 to 350 °C (see Sect. 4.1.4).In the same way, there is a general decrease in the MMF and magnetite production from 300 to 350 °C with the exception of one charge (ol350Pd10).Overall, these observations indicate that the serpentinization Reactions (R3) and (R6) are fully operative at 300 °C and much less developed at 350 °C, although serpentine is still present as indicated by results both on the Åheim dunite and the San Carlos olivines (Table 2).This has the consequence that less H 2 is generated at 350 than at 300 °C, so f H 2 reaches lower values in Au capsule experiments at 350 than at 300 °C, which explains the presence of hematite in ol350AuG (Table 2).These results are consistent with previous experimental and theoretical studies where 350 °C was found to be approximately the upper temperature limit for serpentinization (McCollom and Bach, 2009;Klein et al., 2013).Thermodynamic modeling of serpentinization reactions has shown that, for an oxidizing fluid and a high W/R ratio, the serpentine stability field can slightly exceed 350 °C; under these conditions, the magnetite proportions increase and hematite appears as a reaction product (Malvoisin, 2015;Ely et al., 2023).Therefore, our 350 °C experimental results are qualitatively consistent with these calculations, with the exception of one AgPd charge (ol350Pd10).This 350 °C charge is characterized by a very important iron oxide production (see Sect. 4.1.4).It has a MMF much higher than in the corresponding charge at 300 °C (ol300Pd10) and which also markedly exceeds that in the corresponding Au charge at 350 °C (ol350Au10; Fig. 5; Table 2).The increase in MMF with decreasing grain size is also very abrupt in the 350 °C AgPd experiments (Fig. 5b), thus stressing the anomalous character of the charge.Since iron oxide production in ol350Pd10 is decoupled from serpentine formation (see Sect. 4.1.4),one possibility to be considered to explain the abnormally high magnetite production and presence of hematite is direct oxidation of olivine (e.g., Khisina et al., 1998;Knafelc et al., 2019).However, such mechanisms have been shown to work under anhydrous conditions and at temperatures generally higher than in this study.If operative at 350 °C under hydrothermal conditions, we note that olivine oxidation should be also observed in the 350 °C Au charge (ol350Au10), which is not the case (Fig. 5b; Table 2).Therefore, a satisfactory explanation of the ol350Au10 charge must await further studies.

Iron oxides as redox indicators in serpentinization environments
H 2 production in serpentinization environments is coupled with the oxidation of Fe 2+ into Fe 3+ incorporated in serpentine and iron oxides (see Sect. 1).Among those, magnetite (Fe 2+ Fe 3+ 2 O 4 ) is the most common iron oxide in serpentinites.It is also the main iron oxide produced in experimental and thermodynamic studies of serpentinization (see Sect. 1 for references).Magnetite is emblematic of the high f H 2 (and low f O 2 ) characteristic of serpentinization environments.This is illustrated by our internally controlled (Au capsule) experiments at 300 °C where H 2 accumulates inside the capsule, leading to high-f H 2 and thus low-f O 2 conditions (Fig. 1a; Table 3).Magnetite forms as the main iron oxide phase (Fig. 5a; Table 2) together with serpentine according to Reaction (R3) (see Sect. 5.2.1).However, magnetite is stable in a wide range of f O 2 , being limited by the hematite-magnetite (HM) and iron-magnetite (IM) reactions at high and low f O 2 respectively.Note that wüstite (Fe 2+ x O) formation is not expected for the relatively low temperatures at which the serpentinization process operates (i.e., < 560-570 °C; Haas and Robie, 1973;Myers and Eugster, 1983).Therefore, the presence of magnetite only provides broad constraints on redox conditions in serpentinites.The occurrence of awaruite (Fe 0 3 Ni) in natural (Chamberlain et al., 1965;Frost, 1985;Klein and Bach, 2009) and possibly in experimental (McCollom et al., 2020a, b) serpentinization environments testify to the highly reductive conditions that can be reached ( IM ≈ +1 according to Frost et al., 2013).Furthermore, awaruite precipitation indicates that H 2 -consuming reactions (which reduce part of the Fe 3+ involved in magnetite formation) are also taking place.This illustrates that the f H 2 of serpentinites is determined by the combination of several reactions, some acting in opposed directions.
Our externally (AgPd capsules) controlled experiments carried out at 300 °C demonstrate that hematite (αFe 3+ 2 O 3 ) appears as a reaction product of a serpentine-producing reaction (Reaction R6; see Sect.5.2.1).The H 2 generated is continuously extracted and transferred through the permeable capsule to the Ar pressurizing medium (Fig. 1a).Hematite appears in isolation, i.e., without being associated with magnetite as demonstrated by the near baseline MMF in ol300Pd45 (Fig. 5a).This, as well as the f O 2 conditions above the HM buffer in the externally controlled experiments, suggests that hematite is directly formed instead of representing a product from the transformation of an earlier magnetite.Hematite formation has been expected theoretically (e.g., Frost, 1985) and hematite identified as a product phase in thermodynamic modeling of serpentinization reactions (Malvoisin, 2015;Ely et al., 2023).Hematite has also been reported in several experimental studies.It occurs in open-system reactive percolation experiments of CO 2enriched fluids in olivine aggregates and serpentinite (Peuble et al., 2019;Osselin et al., 2022).Closed-system carbona-tion of partially serpentinized and weathered peridotites have also yielded hematite-bearing product assemblages (Hövelmann et al., 2011).Lafay et al. (2018) observed the appearance of hematite in closed-system carbonation experiments of olivine single crystals.In contrast, parallel hydration experiments yielded only magnetite without hematite (Lafay et al., 2018).Peuble et al. (2019) assigned the oxidation of Fe 2+ to the reduction of CO 2 rather than of H 2 O as classically proposed in serpentinizing environments.For CO 2 -free conditions, hematite appears only in the reactive percolation study of Godard et al. (2013) performed with an oxidizing ingress fluid equilibrated with atmospheric O 2 .It is notably absent from the high-pressure closed-system peridotite hydration experiments of Nakatani and Nakamura (2016), despite one buffered at HM.In this particular experiment, the H 2 generated in the reaction chamber was extracted (as in the present study) and consumed to oxidize hematite of the external buffer, and only magnetite formed with serpentine.We conclude that hematite formation in previous studies is a consequence of the addition of an oxidizing chemical component to the system (either CO 2 or O 2 ).In contrast, in the present study, a causal relation is demonstrated between the presence of hematite and the redox parameters.The appearance of hematite (and the transition from magnetite to hematite) is the result of imposing specific redox conditions (low f H 2 and high f O 2 ) during the serpentinization process.
In natural contexts, iron oxide phases other than magnetite, such as maghemite, have been reported (Prévot et al., 1981;Krammer, 1990;Nazarova et al., 2000).Hematite is also present although rarely found in the serpentinized oceanic crust (Smith and Banerjee, 1985;Bach et al., 2004;Huang et al., 2017a).When present, natural hematite has been interpreted as related to a seawater-rock interaction process that postdates the serpentinization event (Bach et al., 2004).Following the final and most advanced stages of serpentinization, fluid circulations lead to the dissolution of magnetite and its replacement by hematite as the iron oxide phase.This mechanism occurs under oxidizing conditions as indicated by the abundance of hematite in Hole 1268A and iowaite in Hole 1272A (Bach et al., 2004).

Implication for serpentinization processes at MORs
Experiments of this paper demonstrate that the H 2 mobility has an influence on the serpentinization reaction.The internally controlled f H 2 experiments lead to the confinement of H 2 , whereas, in the externally controlled f H 2 experiments, H 2 is extracted from the reacting serpentinization system.Contrasted redox (reducing vs. oxidizing) conditions and product phase assemblages (magnetite vs. hematite) characterize the two regimes.The experimental methodology implemented for H 2 in this paper is closely analogous to the "internally controlled" vs. "externally controlled" cases classically considered in the modeling of metasomatic processes such as serpentinization (Evans et al., 2013).Below, we use the mobility of H 2 as a proxy of fluid circulation, open vs. closed, in oceanic hydrothermal systems and examine the implications of fluid circulations for redox control of serpentinization processes at MORs.
In our internally controlled f H 2 experiments, the f H 2 is imposed by the fluid-rock interaction process (serpentinization).This situation is analogous to the one prevailing in rocks having low permeabilities and whose behavior approaches that of a "closed" system.The permeability data available for ultramafic rocks (Villeneuve et al., 2014;Farough et al., 2016;Tutolo et al., 2018) show that such closed-system situations generally apply to non-fractured lithologies.It is also worth emphasizing that, because of the volume expansion that accompanies serpentinization (Kelemen and Hirth, 2012;Evans et al., 2020;Klein and Le Roux, 2020;Malvoisin et al., 2020), porosity will tend to decrease upon hydration of ultramafic rocks, thus lowering their permeability (Tutolo et al., 2018).Ocean floor serpentinites from the MARK area record serpentinization stages formed in a closed, diffusive system, and their high serpentinization rates imply large volume expansions of the rocks (Andreani et al., 2007).Under these conditions, the fluid composition, including the H 2 concentration and the f H 2 , will be determined by rock-dominated fluid-rock interactions (e.g., Frost et al., 2013).It is important to emphasize that this is the case of most experimental serpentinization studies, which as a rule are performed with W/R ratios < 3 (Seyfried et al., 2007;Malvoisin et al., 2012a, b;Malvoisin and Brunet, 2014;Klein et al., 2015;McCollom et al., 2016McCollom et al., , 2020a, b;, b;Fauguerolles, 2016; Table 2).These studies yield high concentrations of dissolved H 2 (m H 2 ) in the fluid, high f H 2 , and low f O 2 (Table 3).Small W/R ratios are also considered some thermodynamic simulations of serpentinization (e.g., McCollom and Bach, 2009;Klein et al., 2013).Natural serpentinites have mineral assemblages that closely correspond to the experimental and thermodynamic simulations.They comprise serpentine, brucite, and magnetite as dominant phases (Mével, 2003;Bach et al., 2006;Klein et al., 2014).Properties of product phases such as magnetite (magnetic properties) are similar between natural and experimental serpentinites (Fig. 5).The common occurrence of awaruite (FeNi 3 ) as an accessory phase (Frost, 1985;Klein and Bach, 2009;Foustoukos et al., 2015) is also consistent with a "closed" system in which conditions are kept reducing because a low permeability.
Although the preceding situation is very close to the one simulated in our serpentinization experiments performed under internal f H 2 control, one important difference between natural and experimental systems is to be noted.Experiments all start from oxidized charges and low-f H 2 conditions, so the H 2 concentration progressively increases with time and advancement of the reaction.In contrast, in natural systems, rock-buffering through serpentine-brucite-magnetite fluidmineral equilibria is assumed during the initial stages of serpentinization (Beard et al., 2009;Frost et al., 2013).This imhttps://doi.org/10.5194/ejm-36-555-2024plies initially high f H 2 and highly reducing conditions, although the influence of initial fluids of variable f O 2 has been considered in some simulations (Malvoisin, 2015).This difference in initial f O 2 between experimental and natural systems is considered significant.As noted earlier, it is probably responsible for the lack of brucite in our experimental product assemblages.
Our externally controlled f H 2 experiments correspond to a situation where the f H 2 is imposed from outside the reacting system.This situation is analogous to a serpentinization environment where the f H 2 is controlled by an externally derived fluid.We assume that serpentinization is a hydration process driven by seawater and, consequently, the external fluid should have a low f H 2 .It is now accepted that serpentinization takes place in an active tectonic environment where fracturing of rock and fluid infiltration are intimately associated (Andreani et al., 2007;Iyer et al., 2008;Plümper et al., 2012;Rouméjon and Cannat, 2014).For example, it has been proposed that mesh texture initiation results from tectonically controlled penetration of seawater-dominated fluids within peridotites and that the last stage of serpentinization takes place in the brittle-fracturing regime where advective fluid transfer dominates (Andreani et al., 2007;Rouméjon and Cannat, 2014).During these two stages, the fluid circulation regime is controlled by active fracturing and thus approaches an "open" system regime.In this case, certain components of the fluid phase can be externally controlled.H 2 almost certainly belongs to this group given its very high mobility, but the other fluid components (e.g., a SiO 2 (aq) ) might behave differently.Under this situation, simulated in our externally controlled experiments performed with AgPd capsules, the serpentinization reaction is modified (Reaction (R6); see Sect.5.2.1).The appearance of hematite as a product of the reaction implies the possibility of low f H 2 (high f O 2 ) during serpentinization and confirms previous expectations (e.g., Frost, 1985).It is worth noting that, in thermodynamic simulations (Malvoisin, 2015;Ely et al., 2023), hematite appears only for elevated W/R ratios (> 20 and > 6.5, respectively), i.e., for conditions analogous to our "open" system situation.
The near-absence of hematite in serpentinized peridotite (Smith and Banerjee, 1985;Bach et al., 2004;Huang et al., 2017a) is probably to be related to the very low f H 2 conditions necessary for hematite to be stable in natural hydrothermal oceanic systems (f H 2 < 2.1 × 10 −4 MPa at 300 °C; Table 1).Such low-f H 2 conditions appear to be reached in specific contexts, so a serpentine-hematite-brucite assemblage can be expected to be produced in place of serpentinemagnetite-brucite usually found in serpentinization environments.However, an appropriately low-f H 2 range might be rarely attained, and hematite-bearing serpentinites might be difficult to preserve since hematite would presumably react and transform if subjected to later hydrothermal stages involving higher f H 2 fluids.Yet, the presence of hematite, although rare, attests to the possibility of low-f H 2 conditions either locally or temporarily and, by inference, of varying f H 2 during hydration of the oceanic crust.This stresses the need to consider the redox parameters as key process variables in serpentinization environments.

Conclusions
Our experiments highlight the key role played by the redox conditions and mechanisms of redox control during serpentinization.We have demonstrated experimentally that serpentinization can take place in a large range of f H 2 (and in a correspondingly large range of f O 2 ).However, although the general mechanism remains the same, notable differences were found between the internally and externally controlled experiments.At 300 °C, in Au capsules (internal f H 2 control), H 2 accumulation leads to a relatively high-f H 2 (low-f O 2 ) environment characterized by the serpentine-magnetite assemblage.In comparison, in AgPd capsules (external f H 2 control), continuous removal of the produced H 2 yields a lowf H 2 (high-f O 2 ) environment characterized by the serpentinehematite(-magnetite) assemblage.At 350 °C, the lesser efficiency of the serpentinization reaction limits H 2 production, and both the internally and externally controlled experiments are characterized by the serpentine-hematite(-magnetite) assemblage.The maximum serpentine production is observed at 300 °C under external f H 2 control, which is in the optimum temperature range for serpentinization, the reaction being enhanced because of continuous removal of H 2 .Differences in product phase assemblages found in this study imply that natural serpentinization reaction mechanisms vary with redox conditions, and consequences for H 2 production fluxes and rates can be expected.
The externally controlled experiments at 300 °C demonstrate that hematite can appear as a product of the serpentinization reaction.In contrast with previous studies where the presence of hematite was the consequence of the addition of an oxidizing chemical component, the present study establishes a causal relation between hematite and the imposed redox variables.The lack of brucite probably reflects conditions insufficiently reducing (f H 2 not high enough), and brucite appearance would be expected in longer internally controlled experiments (to promote reaction advancement) and for lower W/R ratios.Brucite would be even more unlikely under low-f H 2 environments as in the externally controlled experiments.We emphasize the significant difference between, on the one hand, most serpentinization experiments which start from oxidized, low-f H 2 conditions and, on the other hand, natural systems where serpentinization is assumed to start at equilibrium with serpentine-brucitemagnetite and so under high f H 2 .
Our experiments under external f H 2 control have no counterparts among natural serpentinites mainly because these rocks typically lack hematite.However, the behavior of H 2 and its differential mobility in our two types of experiments has applications for natural oceanic hydrothermal systems.

Figure 2 .
Figure 2. SEM images of reacted San Carlos olivines (a-g) and Åheim dunite (h).For each view, the scale bar is 10 µm.Mineral abbreviations are the same as in Table 2. (a-d, f, h) Polished sections and (e, g) unpolished preparations.(a) Vein developed in San Carlos olivine crystal; the middle of the vein is occupied by fibrous chrysotile; iron oxides are concentrated in the void left between olivine and the chrysotile band (BSE, ol350PdG).(b) Texture similar to (a) except that the band is filled with chrysotile showing different contrast (light: lB and dark: dB; BSE, ol350AuG).(c) Dissolution gulf filled with large iron oxides and chrysotile (BSE, ol300PdG).(d) ≈ 5 µm thick chrysotile layer surrounding olivine crystal and containing < 1 µm iron oxides in very limited amounts (BSE, ol350Au45).(e) Fibrous texture of chrysotile (SE, ol300Au45).(f) Chrysotile layer rimming olivine crystal (ol350Pd45).(g) Chrysotile and numerous iron oxides with sizes up to 5 µm developed on the surface of olivine crystals (BSE, ol300Au45).(h) BSE image of dunite reacted in AgPd capsule (du350PdG).Orthopyroxene is more markedly dissolved than olivine.The iron oxide is most probably maghemite (see text).The fibrous material developed at the grain interface is chrysotile.

Figure 3 .
Figure 3. XRD diffractograms of experimental charges produced from San Carlos olivine emphasizing phases other than primary olivine (Table2).Mineral abbreviations are the same as in Table2.Peaks which are not highlighted correspond to olivine.All peaks higher than 15 000 counts are clipped.Insets are ×6 magnifications of diffractograms.San Carlos olivine is still present at the end of all experiments.Relative olivine and serpentine peak intensities demonstrate limited formation of serpentine except in ol300Pd10.ol300Pd45 is the only charge with initial grain size > 10 µm showing detectable serpentine and hematite peaks (inset centered on the 38.71°2θ peak).Relative peak intensities suggest a more advanced reaction in ol300Pd10 (< 10 µm initial grain size) than in ol300Pd45 (45-50 µm).Hematite is identified in experiments performed in AgPd capsules (ol300Pd10, ol350Pd10, and ol300Pd45).Peaks observed in ol350Pd10 and ol300Au10 at 35.13°2θ and 50.52°2θ (insets) are attributed to magnetite (see text).

Figure 4 .
Figure 4. Representative Raman spectra of iron oxides.Raman spectra have been normalized to 1 based on the maximum intensity recorded for each spectrum.Mineral abbreviations are the same as in Table2.Hm Std. and Mgt Std. 1 spectra have been acquired on hematite and magnetite standard powders, respectively.Mgt Std. 2 spectrum corresponds to the R061111 magnetite sample from the RRUFF database (rruff.info,2023), while maghemite standard (Mgh Std.) comes fromLegodi and de Waal (2007).Maghemite is characterized by a doublet at 665 and 621 cm −1 , while magnetite is characterized by a single peak in the 660-681 cm −1 range (the area colored in blue delimits the magnetite peak position in Mgt Std.1 and Mgt Std. 2).None of the samples studied contains maghemite, while magnetite is present in ol300AuG and ol350PdG and also in lesser amounts in ol300PdG and ol350AuG (red arrows).Except in ol300AuG, hematite is clearly identified in the three other samples.

Figure 5 .
Figure 5. Amounts of magnetite in reaction products (magnetite mass fraction, MMF, Table 2) plotted against the initial grain size for the San Carlos olivine experiments.Experiments are distinguished by capsule material, either Au or AgPd.The method for constructing the starting olivine baseline is detailed in Appendix C. Hm and Mgt correspond to hematite and magnetite, respectively.(a) 300 °C charges.Note the systematic increase of MMF with decreasing grain size observed in experiments with Au capsules.Experiments with AgPd capsules show a modest increase of MMF < 10 µm.(b) 350 °C charges.No significant increase of MMF is observed except for the experiment carried out with the smallest initial grain size in AgPd capsule (ol350Pd10).

Figure 6 .
Figure 6.J rs /J s as a function of H cr /H c plot(Day diagram, Day  et al., 1977)  for selected experimental products.Only charges with magnetite mass fractions (MMFs) higher than the baseline (see Fig.5) and thus containing a substantial amount of magnetite are plotted.SD: single domain, PSD: pseudo single domain, and MD: multi domain.The yellow field corresponds to the experiments ofMalvoisin et al. (2012b) on San Carlos olivine plus pure water at 300 °C.The blue field corresponds to natural serpentinite samples (DSDP Sites 395, 556, 558, and 560 and ODP Sites 670, 895, and 920, as compiled byOufi et al., 2002).

Figure 7 .
Figure 7. XRD diffractograms of experimental charges produced from Åheim dunite (Table2).Mineral abbreviations are the same as in Table2.Peaks which are not highlighted correspond to olivine.All peaks higher than 15 000 counts are clipped.Relative peak intensities demonstrate that orthopyroxene reacts preferentially than olivine.Insets are ×6 magnifications of diffractograms centered on the 50.76°2θ and 67.67°2θ peaks which highlight the presence of iron oxide in du350PdG consistent with the presence of maghemite (see text).

Table 1 .
Calculated f H 2 and f O 2 for HM and FMQ buffers at 300 and 350 °C and 50 MPa.Calculations are made using a Lewis and Randall mixing rule for the H 2 O-H 2 fluid.
a From Shaw and Wones

Table 2 .
Experimental parameters and summary of results obtained on solid products.
a J s and J rs are weighted by the analyzed sample mass.b na: not analyzed.

Table 2 )
, discussed separately below.We conclude that magnetite preferentially forms at 300 °C in Au https://doi.org/10.5194/ejm-36-555-2024Eur.J. Mineral., 36, 555-579, 2024 . Hydrogen transfer from inside to outside the capsule is extremely fast.For external and internal f H 2 of 2.1×10 −4 and 4.78 MPa respectively, all the H 2 present inside the capsule is lost in less than 1 min.Durations of hydrogen transfer vary with the internal f H 2 but remain extremely short.Therefore, Ag 25 Pd 75 capsules cannot keep a high internal f H 2 for a long time if placed in a low-f H 2 external environment (Ar pressure medium), and oxidized conditions are almost instantaneously imposed to the charge.
Mg 0.91 Fe 0.09 ) 2 SiO 4 position of reaction products and of the partitioning of Fe 2+ and Fe 3+ between product phases.