Composition of the Fluid in Carbonate- and Chlorine-Bearing Pelite near the Second Critical Point: Results of Diamond Trap Experiments

Fluids generated in subduction zones transport slab material to the mantle and therefore play the key role in the generation of mantle magmas (Hermann et al., 2013; Schmidt and Poli, 2014; Keppler, 2017; Manning and Frezzotti, 2020). Being part of subducted slabs, marine/oceanic sediments (pelites) are an important source of fluids in subduction zones, significantly contribute to the transport of water into the mantle, and are the main suppliers of carbon, nitrogen, and large ionic lithophile (LILE) and high-field strength (HFSE) elements (Hermann, 2002; Busigny et al., 2011; Hermann et al., 2013; Schmidt and Poli, 2014; Plank and Manning, 2019). Therefore, the reactions of dehydration and decarbonation in pelites under subduction P–T conditions are of paramount interest. It has been established that phengite is the key water-containing phase in pelitic rocks over a wide range of pressures (3–8 GPa) and at temperatures up to the beginning of melting (Domanik and Holloway, 1996; Schmidt et al., 2004; Schmidt and Poli, 2014). As pressure increases, the essentially hydrous fluid generated during the partial dehydration of pelites becomes capable of dissolving a significant amount of major components. At P > 3–5 GPa, volatile-bearing pelites reach the second critical points, and the composition of the fluid in them becomes the same as the composition of the melt (Shen and Keppler, 1997; Bureau and Keppler, 1999; Schmidt et al., 2004; Kessel et al., 2005; Manning et al., 2010; Dolejš and Manning, 2010). Carbonate phases are believed to be stable in subduction zones. Nevertheless, ca. 30% of subducted carbon returns to the surface as CO₂ (Kelemen and Manning, 2015; Plank and Manning, 2019), and the role of CO₂ in subduction fluids is poorly known.

There are still few publications concerned with studies of the major- and trace element composition of fluids produced during the dehydration of subducted rocks at typical $P–T–fO2$ values (Johnson and Plank, 1999; Kessel et al., 2005; Klimm et al., 2008; Hermann and Rubatto, 2009; Skora and Blundy, 2010). These are mostly experimental studies of fluids formed in carbonate-free metasediments. No systematic data exist on the contents of major elements, LILE, HFSE, and rare-earth elements (REE) in subduction fluids and on the coefficients of distribution of trace elements between fluids and carbonate-bearing pelites at ≥3 GPa. The lack of data on the major- and trace element composition of fluids in carbonate-bearing sedimentary rocks makes it necessary to adapt the method of capture of fluids in diamond traps (Ryabchikov et al., 1989; Stalder et al., 1998; Johnson and Plank, 1999; Kessel et al., 2005; Hermann and Rubatto, 2009) for experiments with a carbonate-bearing material. Such data are needed to elaborate a quantitative model of mass exchange between subduction zones and the mantle.

In this work, we applied the diamond trap method to elaborate an algorithm for reconstructing the composition of the fluid generated in CO₂- and Cl-bearing pelite at 3.0 GPa and 750 and 900 °C. We also obtained the first data on the effect of CO₂ and Cl on the parameters of the second critical points in the pelitic system and on the composition of its stable fluids.

Starting materials. For experiments, we used deep-water marine sediment (pelite) of the Maikop Shale (Taman Peninsula, Russia) (Table 1) (Sokol et al., 2018). Earlier, this pelite was used to study the stability of carbonates under subduction P–T conditions. Pelite consists of (wt.%) muscovite (52), quartz (20), illite (15), albite (5), kaolinite (5), calcite (2), and siderite (1.7). According to data of thermogravimetric analysis, it contains 1.87 wt.% CO₂ and 5.4 wt.% H₂O. The content of Cl^– is 0.1 wt.%. It was determined by the turbidimetric method; the total error of determination (with a confidence level of 0.95) was 15%.

Finely ground pelite powder was placed in gold or platinum capsules with a wall thickness of 0.2 mm and a diameter of 2 and 10 mm, respectively. The sample weight varied from 9.5 to 504 mg. To study the composition of subduction fluids, we used the diamond trap method adapted for this task. The method was proposed by Ryabchikov et al. (1989) and successfully used elsewhere (Stalder et al., 1998; Johnson and Plank, 1999; Kessel et al., 2004, 2005; Hermann and Spandler, 2008; Dvir and Kessel, 2017). It implies using a layer (ca. 30% of the sample weight) of synthetic diamond powder (ASM grade) with particles 14–20 μm in size into the capsule for trapping the fluid of the samples at high pressures and temperatures. Our approach consists in a significant increase in the sample weight to 400–500 mg and in the diamond trap weight to ca. 250–300 mg. The large volume of the fluid material in the trap permitted ICP AES and ICP MS analysis of the fluid composition in the solution.

The capsules with samples and traps were hermetically arc-welded. The capsules with traps (10 mm in diameter) were placed upside down in the high-pressure cells so that the trap was in the upper part (Fig. 1). Additional gold capsules (2 mm in diameter) with pelite samples were used to study the phase relations in them. The experiments were carried out for 40 h, based on the data of Johnson and Plank (1999). The diamond trap experiments at 2 GPa and 600 °C, which lasted one to eight days, showed that the Ba/Cs and Th/Rb ratios in the traps reach equilibrium values in ca. 40 h. A short duration of the experiments was necessary to prevent the reduction of CO₂ in the fluid as a result of the inflow of hydrogen into the capsules from outside. Earlier we established that the fluid can be significantly reduced even in gold capsules if the experiments last longer than 100 h (Sokol et al., 2004; Kupriyanov et al., 2023).

After the experiments, the capsules with sample and trap were weighed and then pierced. After the removal of liquid and gases from the capsules, they were dried at 100 ºC for 24 h and weighed again. The difference in weight, $G(H2O+CO2)$⁠, marked the amount of liquid and gases (mainly H₂O and CO₂) released from the fluid during its quenching. As the capsules were opened, the trap was separated from the sample.

During the reconstruction of the fluid composition in diamond traps, the most important parameter for mass balance calculations is the degree of the trap filling with fluid quenching products. During the development of the used technique, we tested two options: using the BSE images of diamond trap after the experiments and weighing the diamond trap before and after the dissolution of the fluid contained in it. The second option was chosen as the best. The weight of the fluid material in the diamond trap (G_FM) was determined by the formula: G_FM = G_L – (G_M + G_LOI), where G_L is the weight of the trap sample before its acid digestion, G_M is the weight of pure diamond powder from the trap sample, and G_LOI is the weight of volatile components adsorbed on the trap material after the opening of the capsule (Supplementary Materials, Table S1). Parameter G_LOI was determined by weighing the trap sample before and after annealing at 550 ºC for 4 h; G_M was determined by weighing the diamond powder after acid digestion, washing with distilled water, and drying at 350 ºC for 1 h. The preliminary experiments showed that the trap annealing at 550 ºC and the subsequent multi-acid digestion do not lead to loss in the weight of microdiamonds within the measurement error (0.1 mg).

Experimental technique. Experiments at 3.0 GPa were carried out on a multianvil high-pressure split-sphere apparatus (BARS) (Palyanov et al., 2017). The experimental capsules were placed in high-pressure cells of tetragonal-prism shape, 21.1 × 21.1 × 25.4 mm in size, with a 18.5 mm high graphite heater. The cell pressure was calibrated against phase transitions in Bi at 2.55 GPa and in PbSe at 4.0 and 6.8 GPa and at room temperature. At 900 ºC, the calibration was performed against the quartz–coesite phase transition (Bohlen and Boettcher, 1982). The temperature was measured with a PtRh₆/PtRh₃₀ thermocouple in each experiment. The thermocouple was calibrated at 6.3 GPa against the melting points of Al and Ag (Sokol et al., 2015). Pressure and temperature were measured with an error of ±0.1 GPa and ±20 ºC (Sokol et al., 2015; Palyanov et al., 2017). The samples were quenched at a rate of 150 ºC/s.

ICP AES analysis of the fluid material. The samples of the diamond traps with fluid material after the experiments have specific features: (1) It is impossible to prepare polished thin sections from them for local analysis; (2) the trap is not uniformly filled with the fluid material; (3) the trap with the fluid material weighs no more than 25–50 mg. The contents of major and some trace elements in the traps were determined by ICP AES, which permits a simultaneous analysis of a large number of elements with a wide range of contents and low detection limits (n ∙ 10^–4–n ∙ 10^–5 %). The samples were prepared by the method of alkaline (KOH) fusion, which is the most appropriate for the simultaneous determination of major and trace elements, including B and Li. For effective fusion, the weight ratio of the sample to the flux must be at least 1:4, which yields a high concentration of K in the analytical solution and can lead to the self-absorption of the analytical lines of the elements to be determined (Sedykh et al., 2019). Therefore, we needed to optimize the weight ratio of the sample and flux, having at our disposal 5 to 50 mg of the diamond trap material. Since the starting samples might contain sulfides, we carried out test analyses with the addition of sodium peroxide to the flux in order to determine Fe, S, Zn, etc.

To test and confirm the correctness of the developed technique, we used standard samples of rocks of different compositions: UB-N (serpentinite), Dr-N (diorite), SGD-1a (gabbro), and SGKhM-4 (aluminosilicate deposits). The compositions of all standard samples are given at [http://georem.mpch-mainz.gwdg.de/sample_query.asp]. The element concentrations were calculated by external calibration (I) against the multi-element standard solutions MES-1, MES-2, and MRS-3 produced by Skat R&D enterprise, Novosibirsk. Reference solutions for constructing calibration plots were prepared by diluting the corresponding MES solution with deionized water and nitric acid and adding Sc as an internal standard to all solutions to get a final concentration of 2 mg/L. Calibration of the device (II) was made using solutions prepared from the standard samples UB-N, Dr-N, SGD-1a, and SGKhM-4 (Pupyshev and Danilova, 2007). The analysis was performed on an iCAP-PRO (Duo) (ThermoFisherScientific, USA) ICP atomic-emission spectrometer. The device was operated using the Qtegra software.

Sample preparation. Weighed samples (5 to 50 mg) were fused in 45 ml glassy-carbon crucibles at the flux/sample weight ratio of ca. 1:6. The air-dried sample was placed at the bottom of the crucible, and 1–2 ml of a 0.1 g/ml KOH solution was added. The crucible was placed in a cold muffle furnace and dried at 100–150 °C for 1 h. Then, the temperature in the muffle was raised to 550 °C; after a 5 min exposure, the crucible was removed from the muffle. As the flux cooled, we added ca. 5 ml of water into the crucible and left it for 2–3 h. Then, 2 ml of 15% HCl was added dropwise to the crucible, and the solution was transferred into a 10 ml test tube. To determine the concentrations of major elements, we diluted the solutions with HCl of the same concentration by ten times and added Sc as an internal standard (2 mg/L Sc).

In the experiments with sodium peroxide, 5–25 mg of an air-dried sample was placed in glassy-carbon crucibles, and 100–200 mg of dry KOH and 25–50 mg of Na₂O₂ were added. The following sequence of procedures was the same as for KOH. Possible contamination of the samples during their chemical preparation was controlled by analysis of blank samples.

Choice of the operating conditions and internal standard. To obtain an intense and well-reproduced signal and to keep low noise, we chose the following operating conditions: high-frequency generator power of 1150 W, argon flow rate of 0.65 L/min, auxiliary argon flow rate of 0.5 ml/min, cooling argon flow rate of 12.5 L/min, axial plasma observation, peristaltic-pump speed of 45 rpm, flushing time of 40 s, and signal recording time of 10–20 s. The concentrations of elements were measured at wavelengths without serious spectral overlaps, which ensured the required sensitivity. The introduction of Sc successfully corrects the influence of the matrix in solutions with a high content of KCl (Tiggelman et al., 1990). The analyses were carried out at the lines Sc 255.237, Sc 361.384, and Sc 391.181 nm.

Calculation of element concentrations. Concentrations of elements were calculated using the formula С_sol = I_sol/I_SS × A, where I are the concentrations of the sample solutions and standard samples (SS) and A is the concentration of element in the SS. Table 2 shows the contents of elements in the SS after fusion, determined from their solutions. They were calculated from the calibration plots constructed by external calibration (I) against the MES solutions (with a confidence level of 0.95). These data showed satisfactory agreement with the certified interval of element contents and no effect of KOH (up to a concentration of 10 mg/ml) on the emission of elements. To assess the matrix effect of Na₂O₂ (Sedykh et al., 2019), we determined the calibration (II) parameters for the solutions of SS fused with a mixture of KOH and Na₂O₂ (Table 3), which were then used to calculate the element contents in the initial pelite (Table 4).

Table 4 presents the contents of trace elements in pelite, calculated from calibration plots (I) and (II), in comparison with the earlier obtained data (Sokol et al., 2018). A significant difference is observed for the content of sulfur (24.4%), which is due to the oxidizing melting of the sample in the presence of sodium peroxide. The contents of boron are also overestimated (8%). The results for the other elements are in agreement with the MES calibration data. The contents of trace elements in the blank sample are determined by the content of impurities in KOH. The addition of Na₂O₂ does not affect significantly the blank sample but increases the content of sulfur in the solution and reduces the signal intensity, especially that of Li. Comparison of the contents determined from the calibration plots constructed over the external aqueous MES solutions and the SS solutions did not reveal a significant discrepancy.

Based on the data obtained, we determined the contents of major components in pelite (Table 5), using calibration plot (I), and compared the results with earlier data (Sokol et al., 2018). The contents of major elements were determined after the fusion of 5–25 mg samples with KOH (the concentration of potassium in the solution was 4–10 mg/ml). For Si and Al, in contrast to trace elements, the calibration plot (I) is linear up to a content of 10 mg (in terms of the initial dry sample per 10 ml of the solution).

Based on the performed studies, we developed a technique for the simultaneous ICP AES determination of major and trace elements (within 2–10 mg/ml) in diamond traps after fusion of the samples with KOH (this technique is inappropriate for determining K and Na). The relative error depends on the element content and is 3–5% for Si and Al and 10–11% for Fe, Ca, and Mg. The contents of trace elements in the blank sample are controlled by the content of impurities in KOH. Comparison of the results for the calibration plots constructed over the external aqueous MES solutions and SS solutions showed no serious difference, which made it possible to use calibration plot (I) based on the MES standard solutions in the following research. A necessary condition for obtaining correct results is equal concentrations of flux (especially with Na₂O₂) in all analytical solutions.

Scanning electron microscopy and probe microanalysis. The textural phase relations and their composition were studied with a Tescan MIRA 3 LMU scanning electron microscope equipped with an INCA EDS 450 microanalysis system with an EDS X-Max-80 Silicon Drift Detector. The operating conditions were as follows: accelerating voltage of 20 kV, beam current of 1 nA, and spectrum integration time of 20 s. The phase compositions were studied by probe microanalysis on a Jeol JXA-8100 analyzer at an accelerating voltage of 20 kV and a current of 40 nA. Analyses of silicate and carbonate phases were performed with a beam 1–2 μm in diameter. The spectrum integration time for each element was 10 s. The following minerals were used as standards: pyrope (O-145) (for analyses for Si, Al and Fe), Cr-garnet (Ud-92), Mn-garnet (Mn-IGEM), diopside (for Mg and Ca), albite (for Na), orthoclase (for K), ilmenite (for Ti,), and spinel (for Ni). The measurement error was within 2 rel.%. Quenched glasses were analyzed on a Jeol JXA-8100 microprobe with a defocused beam 6–10 µm in diameter.

Phase relations in the samples. At 3.0 GPa and 750 and 900 ºC, the pelite samples undergo reactions producing an assemblage of garnet, phengite, clinopyroxene, and coesite (major phases) (Table 6; Figs. 2 and 3) as well as carbonate, zircon, and, in some samples, monazite (accessory phases). The phases are unevenly distributed within the samples. After the experiments, the samples from gold and platinum capsules have an identical phase composition. The only difference is that the samples obtained at 900 ºC have a thin (0.1–0.3 mm) light layer near the walls of the platinum capsules, which is due to the removal of iron. At 3.0 GPa, the solidus of pelite is between 750 and 900 ºC (Schmidt and Poli, 2014; Perchuk et al., 2020). In our samples obtained at 900 ºC, the content of quenched glass reaches ca. 40 vol.% (Table 6; Fig. 3a, b). The glass contains bubbles smaller than 10 µm. The samples with glass contain garnet, kyanite, and rutile.

Garnet in the samples is present as euhedral, skeletal, or atoll crystals (Fig. 3a, b) measuring 5–30 µm; some crystals reach 60–70 μm in size. The content of garnet in the samples varies from 20 to 33 vol.%. Subhedral phengite grains in the groundmass are 10 to 20 μm in size (Fig. 3a). The modal content of phengite reaches ca. 30 vol.% at 750 ºС, decreasing to zero in the sample with glass at 900 ºС (Fig. 3b). Coesite occurs as anhedral (or, rarely, subhedral) grains up to 100 μm in size (Fig. 3a). Its content varies from ca. 20 vol.% to 0 (in the sample with glass). Clinopyroxene in the obtained samples is present mainly as subhedral grains up to 20 μm in size. Its content in the low-temperature samples reaches ~10 vol.%. The high-temperature samples contain single or no clinopyroxene grains. The contents of subhedral kyanite and rutile and anhedral pyrrhotite are few percent. Their grains are mostly no larger than 30 μm.

The samples with quenched glass are almost free of clinopyroxene, phengite, and coesite (Table 6). This phase relation indicates that the pelite melted following the peritectic reaction Phe + Coe + Cpx = Grt + L (Schmidt et al., 2004; Hermann and Spandler, 2008; Schmidt and Poli, 2014; Perchuk et al., 2020).

Composition of solids. The samples obtained at 750 and 900 ºС contain garnet of almost identical composition (Fig. 4a). The content of the grossular end-member in the garnet varies from 18 to 24%, and the content of pyrope, from 17 to 25%. The content of almandine in the garnet varies from 52 to 57%. Garnet enriched in pyrope (up to 44%) was found only in experiment 2168_2_1. In some samples, garnet grains are zoned (the contents of MgO and TiO₂ increase from core to rim, whereas the contents of FeO and MnO decrease). Note that the garnet obtained at 2.9 GPa and 900 ºС in the system GLOSS–H₂O modeling a global subduction sediment has the following composition: grossular (25%), pyrope (21–27%), and almandine (48–54%) (Perchuk et al., 2020), which is nearly identical to the composition of the garnet produced in the pelitic system. The garnet produced in our experiments contained some amounts of (wt.%): Na₂O = 0.1–0.3, TiO₂ = 0.5–0.7, and P₂O₅ = 0.3–0.5, which are characteristic of garnet in high P–T pelitic systems (Hermann and Spandler, 2008).

Clinopyroxene in the obtained samples corresponds to omphacite. Its grains are homogeneous in composition. With increasing experimental temperature, the portion of hypersthene decreases, whereas the portion of jadeite slightly increases: at 750 ºС, it varies from 63 to 70%, and at 900 ºС, from 65 to 75% (Fig. 4b). The content of K in the produced clinopyroxene does not exceed 0.01 apfu.

Phengite is the main K-containing solid phase in the produced assemblages. At 3.0 GPa and 750 and 900 ºС, it has a high content of the celadonite end-member (Fig. 5). At 750 ºС, the content of Si + Mg in phengite per 11 oxygen atoms varies from 3.6 to 4.0 apfu. At 900 ºС, its average content is more than 3.9 apfu., reaching 4.1 apfu. in some grains (Fig. 5). As the temperature increases, the content of Al regularly decreases from 2.1 to 1.8 apfu. At 2.9 GPa and 750–850 ºС, phengite in the system GLOSS–H₂O contains significantly less celadonite end-member: The content of Si + Mg varies from 3.5 to 3.6 apfu., and the content of Al, from 2.2 to 2.5 apfu. (Perchuk et al., 2020).

Carbonate is a Fe–Mg solid solution with the FeO/MgO (wt.) ≈ 1, CaO = 1–2 wt.%, and MnO < 1 wt.%. This composition of carbonate is due to the low content of CaO in the initial pelite (Table 1). In the system GLOSS–H₂O with similar P–T parameters, carbonate has the composition of dolomite and is poor in FeO (Perchuk et al., 2020).

Monazite has the following composition (wt.%): Ce₂O₃ (26–30), La₂O₃ (11–14), Nd₂O₃ (7.2–11.1), Pr₂O₃ (2.2–3.3), and Sm₂O₃ (0–1.3). Kyanite contains 0 to 0.8 wt.% FeO. Rutile contains 1.8–2.3 wt.% Al₂O₃ and 0.8–1.2 wt.% FeO. Zircon bears impurities of HfO₂ (1.5–2.2 wt.%) and Nb₂O₅ (up to 1.5 wt.%).

Composition of the fluid. At the end of the experiments, quenching of the samples led to the separation of the fluid into gas, liquid (in room conditions), and solid; the latter was then preserved in the interstices of the diamond trap. Supplementary Materials include a video (Video S1) taken immediately after the piercing of Pt capsules 10 mm in diameter (after experiment 2103_2_2) containing pelite and a trap. One can see the intense release of gas (mainly CO₂) and water with dissolved salts from the hole in the capsule wall. This process lasted at least 30 s. The amount of the released gas and liquid components (hereafter, H₂O + CO₂) was determined by weighing the capsules before and after opening (with additional drying at 100 ºC for 24 h). The liquid contained a significant amount of dissolved salts, mostly potassium and sodium chlorides. After the experiments at 750 ºC, the average content of released H₂O + CO₂ was 4.9 wt.%, and at 900 ºC, 1.2 wt.% (per weight of the pelite sample). The formation of 4.9 wt.% H₂O + CO₂ corresponds to the retention of 30 wt.% phengite and 2 wt.% carbonate in the sample, which is in agreement with the real phase relations in the samples. The underestimated content of H₂O + CO₂ in the samples after the experiments at 900 ºC is due to the formation of glass with bubbles filled with a fluid. Since it is impossible to determine the content of H₂O + CO₂ in the fluid in the samples obtained at 900 ºC, the composition of the fluid/melt was not reconstructed. The degree of filling of the diamond trap with the fluid material was determined following the procedure discussed in METHODS. As shown by measurements after the experiments at 3.0 GPa and 750 ºC, the content of the “dry” fluid material varied from 23 to 25 wt.% in three traps (G_FM) and was 12 wt.% in only one trap (Suppl. Mat., Table S1). The data on the degree of filling of the diamond trap allowed us to calculate the coefficient of conversion of the contents of major and trace elements in diamond traps obtained from ICP AES data (Table 7) to their absolute contents in the fluid (Table 8).

Position of the second critical point for CO₂- and Cl-bearing pelite. Above the second critical point (2CP), the aqueous fluid produced during the partial dehydration of pelites can dissolve a significant amount of silicate components (Shen and Keppler, 1997; Bureau and Keppler, 1999; Schmidt et al., 2004; Kessel et al., 2005; Dolejš and Manning, 2010; Manning et al., 2010). At the 2CP parameters, the composition of the fluid becomes indistinguishable from the composition of the melt (Fig. 2). In our case, dehydration and decarbonatization of pelite led to the release of water and carbon dioxide. According to mass balance calculations, the H₂O/(H₂O + CO₂) weight ratio in the fluid at 3.0 GPa and 750 ºС was no lower than 0.7, and the content of chloride ions reached 1 wt.%. In order to establish whether the system was above 2CP, it is necessary to determine the quantitative ratio of H₂O + CO₂ in the produced fluid to the total content of the dissolved substance. The content of H₂O + CO₂ in the samples was determined by weighing the capsules before and after piercing, with their additional drying. The weight of the solid material from the fluid was determined by weighing the diamond trap before and after the dissolution of the fluid material. We obtained all the necessary data to evaluate the above ratio in the experimental products at 3.0 GPa and 750 ºС (Table 6; Suppl. Mat., Table S1). However, it is unclear how to interpret the data on the loss on ignition in the diamond trap (G_LOI). To reconstruct the fluid composition, we subtracted G_LOI from the weight of the fluid material (Suppl. Mat., Table S1), assuming that volatiles might have been adsorbed from the air. These might be carbonates and hydroxides produced in the diamond trap during the fluid quenching. In this case, they were components of the fluid material. At present, we cannot unambiguously reconstruct the origin of the volatiles considered as G_LOI. This uncertainty inputs the main error in the determination of the ratio of H₂O + CO₂ to major components dissolved in the fluid. However, with and without G_LOI, the content of H₂O + CO₂ in the fluids of three samples varied within the narrow range 26–34 wt.%; in one sample it reached 42–54 wt.%.

In the system metapelite–H₂O, the melt below 2CP cannot contain more than 25–30 wt.% H₂O (Fig. 2) (Hermann and Spandler, 2008; Hermann et al., 2013). Thus, some of the reconstructed phases rich in volatiles had maximum allowable contents of H₂O + CO₂, and some had a composition typical of supercritical fluids at parameters above 2CP. There are different viewpoints on the P–T conditions of 2CP in a pelitic system with >5 wt.% H₂O: 3.6 GPa and 750 ºС (Hermann and Spandler, 2008; Hermann et al., 2013) and 5.3 GPa and 850 ºС (Schmidt and Poli, 2014).

Our data show that at 3.0 GPa and 750 ºС, the studied pelitic system with 5.4 wt.% H₂O, 1.9 wt.% CO₂, and 0.1 wt.% Cl was either near or even above 2CP. This conclusion is rather nontrivial. Note that the presence of CO₂ and Cl can cause different effects on the solubility of major and trace components in the fluid (Frezzotti and Ferrando, 2015; Keppler, 2017; Barnes et al., 2018; Manning, 2018; Macris et al., 2020). Carbon dioxide, a nonpolar solvent, poorly dissolves silicates and oxides. Therefore, the higher the concentration of CO₂ in the fluid, the lower the concentrations of silicates and oxides in it. Chlorine reduces the solubility of SiO₂ (Cruz and Manning, 2015), but its complexation with metals increases the solubility of other major components and REE (Macris et al., 2020).

Fluid speciation. There is a common concept that supercritical fluids in a pelitic system are enriched in Si, Al, and K (Schmidt et al., 2004; Hermann et al., 2013; Schmidt and Poli, 2014; Keppler, 2017). The fluid in the studied pelitic system with H₂O, CO₂, and Cl is also enriched in Si and Al and contains other elements, with their contents decreasing in the order: K > Na > Сa » Fe > Mg > Mn > Ti » P (Table 8; Figs. 6 and 7). Earlier, Hermann et al. (2013) reported the extremely low solubility of mafic components in the system metapelite–H₂O; thus, the content of MgO + FeO is <3 wt.%. Obviously, it is the presence of chlorine that ensures the high solubility of Fe, Ca, and Mg in the fluid. Comparison of our data with literature ones (Kessel et al., 2005; Hermann and Spandler, 2008; Hermann et al., 2013) shows that the composition of the reconstructed fluid falls in the field of melts in the (Na + K)/Al–Si/Al (mol.%) diagram (Fig. 6). Note that the reconstructed fluid composition differs significantly from the composition of the melt (quenched glass) generated in the same pelite samples at 3.0 GPa and 900 °C (Tables 8 and 9). The fluid and melt compositions (normalized, without H₂O and CO₂, to 100%) have similar contents of K₂O and Na₂O. The fluid contains less SiO₂ but more Al₂O₃, FeO, CaO, MgO, and MnO. It is remarkable that the composition of the melt generated at 900 °C (quenched glass) is similar to the compositions of the melts obtained by other researchers in pelitic systems under similar P–T conditions. The reason is that all the melts formed by the peritectic reaction Phe + Coe + Cpx = Grt + L (Schmidt et al., 2004; Hermann and Spandler, 2008; Schmidt and Poli, 2014; Perchuk et al., 2020). The compositions of the melts (quenched glasses) obtained in the system GLOSS–H₂O at 3.5 GPa and 750 and 900 °C (Table 9) differ insignificantly, only in Na₂O content (Hermann and Spandler, 2008).

Thus, the analysis of the data obtained shows the formation of a near-supercritical fluid with 30–50 wt.% H₂O + CO₂, up to 1 wt.% Cl, and 1.5 wt.% P₂O₅ at 3.0 GPa and 750 °C (Fig. 2). In the Si/Al and (Na + K)/Al ratios it is similar to hydrous melts (Fig. 6) but differs from them in lower contents of SiO₂ and higher contents of FeO, CaO, MgO, and MnO. This fluid composition is due to a decrease in the solubility of SiO₂ in it in the presence of CO₂ (nonpolar solvent) and Cl and an increase in the solubility of metals as a result of their complexation with Cl (Frezzotti and Ferrando, 2015; Cruz and Manning, 2015; Keppler, 2017; Barnes et al., 2018; Manning, 2018; Macris et al., 2020). As the temperature grows from 750 to 900 ºС, the fluid gradually turns into a melt, with a drastic increase in the degree of pelite melting to ca. 40 wt.% and a nearly four-fold decrease in the Cl content in it. Thus, the melt becomes identical in composition to melts of pelitic systems (Table 9).

Element fractionation. The obtained data on the fluid composition allow us to evaluate the mobility of major and trace elements during the dehydration and decarbonatization of pelites in subduction zones at depths of ca. 100 km with typical thermal conditions. Fractionation of elements between the silicate matrix and the fluid depends significantly on the stability of mineral-hosts in the system under experimental P–T conditions (Hermann and Rubatto, 2009). Figure 8 shows the pelite-normalized contents of elements in the reconstructed fluid. It is seen that P, Mn, Sr, and B are incompatible with the eclogite-like phase association (excess Mn might be from metal inclusions in the synthetic trap diamond). Our data confirm the high mobility of Sr and B (Leeman, 1996; Marschall et al., 2007; Palmer, 2017) during the separation of fluids from metasediments in subduction zones. The redistribution of phosphorus into the fluid probably reflects the high solubility of monazite and, possibly, an additional source of phosphorus (organics?) in the supercritical fluid (Keppler, 2017). In the presence of phengite and omphacitic pyroxene, K and Na are moderately mobile, whereas Ca, Mg, and Fe, which are concentrated in garnet, pyroxene, and carbonate, show a low mobility (Ca > Mg > Fe). Titanium and sulfur present in stable rutile and pyrrhotite are almost not redistributed into the fluid. As established earlier, Li can be mobile in subduction zones during slab melting (Ryan and Langmuir, 1987; Brenan et al., 1998; Caciagli et al., 2011). In our samples, Li is lowly mobile, most likely because of the incorporation into phengite. In general, the data obtained indicate a high mobility of P, Sr, and B and a low mobility of Li during the fractionation of elements between the solids of the eclogite-like association and the fluid, which is a common phenomenon at low temperatures. Sulfur is almost immobile under redox conditions near the Ni–NiO buffer.

We have developed an algorithm for reconstructing the composition of the fluid formed in the system pelite–volatile near 2CP. Water and carbon dioxide released during the dehydration and decarbonatization of pelite make the basis of the fluid phase. At the same time, the fluid has high contents of dissolved major components and trace elements. The data obtained showed that an almost supercritical fluid formed in carbonate- and chlorine-bearing pelite even at 3.0 GPa and 750 °C. The fluid is Si- and Al-rich and contains 30–50 wt.% H₂O + CO₂ and up to 1 wt.% Cl. The contents of other major elements decrease in the order: K > Na > Сa ≈ Fe > Mg > Mn > Ti ≈ P. Compared with supercritical fluids generated in the systems pelite–H₂O and eclogite–H₂O, the fluid with high CO₂ and Cl contents is richer in Fe, Ca, Mg, and Mn but poorer in Si. Silicate melt generated in this system at 900 ºС has a composition typical of pelitic systems. The experiments have revealed a set of fingerprints of element fractionation between a supercritical fluid and solids forming an eclogite-like association, namely, high mobility of P, Sr, and B and low mobility of Li, which are typical of low-temperature processes. Sulfur under redox conditions near the Ni–NiO buffer is poorly soluble in the fluid.

We are grateful to Yu.M. Borzdov, Yu.N. Palyanov, A.F. Khokhryakov, and reviewers O.G. Safonov and A.V. Girnis for useful critical remarks, which helped to improve the paper.

The analyses were carried out at the Analytical Center for Multi-Elemental and Isotope Research SB RAS. The work was financially supported by grant 22-17-00005 from the Russian Science Foundation.