# Non-destructive analysis of a mixed H2O–CO2 fluid in an experimental noble-metal capsule by means of freezing and high-energy synchrotron X-ray diffraction

### Synthesis of fluids and sample preparation

Mixed H2O–CO2 fluids were generated by oxidation of graphite powder (obtained from spectroscopically pure rods gently hand-ground in boron carbide mortar) in ultrapure MilliQ water. Experiments were buffered using the double-capsule technique9 to prevent the direct contact of the sample with the buffering assemblies. An inner H2-permeable Au60Pd40 capsule (outer diameter OD = 2.3 mm; inner diameter ID = 2.0 mm; length ≈ 8 mm) and an outer Au capsule (OD = 3 mm; ID = 2.8 mm) were used. The outer capsule contained the redox buffer fayalite–magnetite–quartz (FMQ) soaked in water, constraining fH2. Quartz (SiO2) was obtained from powdered natural crystals; fayalite (Fe2SiO4) and magnetite (Fe3O4) have been synthesized at 1100 °C in a gas-mixing furnace in a reducing atmosphere (CO2:CO = 10:1), starting from stoichiometrically weighted reagent-grade Fe2O3 (Sigma-Aldrich) and amorphous SiO2 obtained from hydrolyzed tetraethyl orthosilicate (Sigma-Aldrich). At equilibrium conditions, as long as all the buffer phases are present – ​​which has been confirmed by scanning electron microscopy in almost identical runs performed at the same P–T conditions3.6 – the chemical potential of hydrogen is expected to be homogeneous in the inner and in the outer capsules. In the outer capsule, the hydrogen fugacity (fH2) is constrained through the reactions:

$${text{3Fe}}_{{2}} {text{SiO}}_{{4}} left( {{text{fayalite}}} right) + {text{2H} }_{{2}} {text{O}} = {text{2Fe}}_{{3}} {text{O}}_{{4}} left( {{text{magnetite }}} right) + {text{3SiO}}_{{2}} left( {{text{quartz}}} right) + {text{2H}}_{{2}} left( {{1};{text{GPa}},{8}00;^circ {text{C}}} right)$$

(1)

In the inner capsule, the equilibration of the COH fluid is accomplished by the fH2-dependent reaction3:

$${text{C}}left( {{text{graphite}}} right) + {text{2H}}_{{2}} {text{O}} = {text{ CO}}_{{2}} + {text{2H}}_{{2}}$$

(2)

As a consequence, the initial CO2-free aqueous fluid adjusts its CO2/H2O fraction until the equilibrium with the fH2 imposed by the buffer is reached. In a similar fashion, the oxygen fugacity (fO2) is constrained directly in the outer capsule by the FMQ buffer and indirectly in the inner capsule because of the water dissociation reaction:

$${text{2H2O}} = {text{2H2}} + {text{O2}}$$

(3)

In the inner capsule, however, the fO2 will be slightly lower compared to that imposed by the FMQ buffer in the outer capsule since the fluid is not pure H2O but a H2O–CO2 mixture, with a consequent declined fugacity for H2O (and O2)11.

Experiments were performed at 1 GPa at 800 °C using an end-loaded piston-cylinder apparatus. Capsules were embedded in MgO rods (Norton Ceramics) and inserted in graphite furnaces surrounded by NaCl and borosilicate glass (Pyrex®). At the top of the assembly, a pyrophyllite–steel plug was placed to ensure the electrical contact. Temperatures were measured with K-type thermocouples that have an estimated uncertainty of ± 5 °C. An alumina disk was placed at the top of the capsule to avoid direct contact with the thermocouple. Pressure calibration of the apparatus is based on the quartz to coesite transition12 that guarantees an uncertainty of ± 0.01 GPa. Samples were first brought to the running pressure, then heated to 800 °C, with a ramp of 100 °C/min. Run duration was 92 h, in order to approach equilibrium between graphite and CO2-bearing aqueous fluid at 800°C3. Eventually, the experiments were quenched by turning off the power supply, resulting in a cooling rate of > 40 °C/sec.

### X-ray diffraction of the frozen fluid

The inner Au–Pd capsule (Fig. 1a,b) containing graphite and H2O–CO2 fluid (Fig. 1c) was exposed by peeling off the outer gold capsule and removing the redox buffer (FMQ), then mounted on a goniometer head on the sample stage at the beamline ID15A of the ESRF (Fig. 1b). A Debye transmission geometry was used, and the signal was collected with a Dectris Pilatus3 X 2 M detector with a Cd–Te sensor. The beam had an energy of 77 keV (λ = 0.1610 Å) and a spot size of 0.05 × 0.05 mm on the sample. Such an elevated beam energy coupled with a high energy threshold of the detector minimized the absorption effects from the capsule (Au X-ray absorption edge is located at ~ 80.7 keV) and allowed a significant reduction of the background (mainly arising from Au60Pd40 fluorescence), respectively. The X-ray diffraction effects were collected every 2 min while cooling the sample with a cryostat from room temperature down to − 90 °C and from − 90 to − 180 °C with rates of − 3 °C/min and − 6 °C /min, respectively (Fig. 2a). The beam was focused orthogonally to the axis of the rotating capsule and multiple diffraction pattern collections were made at different z-axis positions. For each temperature, multiple scans with z offsets of 0.05 mm in 1 mm z interval were collected. Phase abundances were quantitatively estimated by performing Rietveld refinement and the proportions obtained from multiple acquisitions were used to calculate average values ​​and the related uncertainties. Rietveld analysis was performed with GSAS + EXPGUI software13,14. Crystal structure of graphite, solid CO2, ice and clathrate were taken from the ICSD database. Background, phase scale factors, lattice parameters and two peak profile function parameters (the constant Gaussian term and a term for Lorentian of a pseudo-Voigt function), were refined.

X-ray diffraction microtomography15 was used to acquire a cross section of the middle part of the capsule at T = − 180 °C (Fig. 3), well below the freezing points of both H2O and CO2. An RGB image showing the spatial distribution of the phases in the capsule was obtained by assigning to the color channels the characteristic peaks of solid CO2, clathrate and graphite (red, green and blue, respectively, in Fig. 3; cf. also Supplementary Fig. 1). Background removal was applied to eliminate noise. Pixels of the RGB image were also classified using a K-means clustering method (Wolfram Mathematica).