Silicon carbonate, Si[CO₃]₂, is a potential carbon host in the deep Earth

Abstract

The transport of carbon into the deep Earth is governed by the stability and properties of carbon-bearing phases. However, despite extensive research efforts, it is still an open question whether there are high-pressure minerals that can incorporate both silicon and carbon simultaneously. Multiple theoretical studies suggest that Si─C─O compounds could be stabilized at high pressures, but so far, no reliable experimental evidence for their presence has been presented. Here, we demonstrate that at 40(2) gigapascals and ≈ 1800(200) kelvin, CO₂ reacts with silicic acid or cristobalite and forms the anhydrous silicon carbonate Si[CO₃]₂. The structure consists of [SiO₆] octahedra coordinated by six [CO₃]²⁻ groups. Similar groups occur in many ambient and high-pressure carbonates, suggesting that mixed carbonate-silicate phases may be stable at midmantle pressures. Silicon carbonate decomposes upon decompression at pressures of <6 gigapascals but could act as a potential host for carbon in Earth’s lower mantle.

Synthesizing a carbonate of silicon at the p,T conditions of Earth’s lower mantle.

INTRODUCTION

Carbonates, the main reservoir for carbon in Earth’s crust, are transported into the deep Earth by subduction of oceanic lithosphere (1–3). The total amount of carbon transported into Earth’s mantle is roughly 40 to 60 Mt per year (4, 5). Hence, a prerequisite for understanding the global carbon cycle is to determine structures, stabilities, and properties of high-pressure carbon-bearing phases. The chemically simple anhydrous carbonates Ca[CO₃] and Ca,Mg[CO₃]₂ account for >90% of the carbonates in Earth’s crust (1). Consequently, their p,T phase diagrams have been studied extensively in the last decades and there is a consensus of phase stabilities in these end-member systems (6, 7). Multiple studies suggest that carbonates break down during subduction (8–10), and hence, it is of interest to understand the stabilities and reactions of carbonates and minerals typical for a pyrolitic mantle at high CO₂ fugacities. At high CO₂ fugacities and moderate pressures, anhydrous inorganic pyrocarbonates may form (11–13), and it has long been debated if reactions of carbonates with a pyrolitic mantle would lead to the formation of a silicon carbonate.

However, up to now, no unambiguous evidence, i.e., a reliable structural model derived from a diffraction study, has been presented for a silicon carbonate yet. A previous study investigated the reaction between carbon dioxide and silica at 18 to 26 GPa and 296 to 980 K (14). On the basis of Raman and infrared spectroscopy and synchrotron powder x-ray diffraction, it was claimed that one or more silicon carbonates of unknown composition ${\text{Si}}_x{\mathrm{C}}_y{\mathrm{O}}_{2x+2y}$ had been obtained. However, that study provided neither lattice parameters, nor a structural model, nor unambiguous spectroscopic evidence, such as a well-defined Raman band due to an expected [CO₃]²⁻ stretching vibration. A further experimental study of the system SiO₂-CO₂ (15) was later retracted (16). To facilitate a formation reaction, CO₂-filled SiO₂ zeolite has been studied at high pressures ($\approx 24 \text{GPa}$) and temperatures (up to 825 K), but no chemical reaction was observed (17).

In addition to the experimental studies, several crystal structure prediction approaches have been used to find stable phases in the Si─C─O system (18–22). Morales-García et al. (18) concluded that up to 30 GPa, all investigated compounds with composition ${\text{SiC}}_x{\mathrm{O}}_y$ were thermodynamically unstable, but that the least unstable compound was a monoclinic polymorph of ${\text{SiC}}_2{\mathrm{O}}_6$. Zhou et al. (19) predicted the formation of a compound which crystallizes in a trigonal space group with ${\text{SiC}}_2{\mathrm{O}}_6$ composition and conjectured that it may be synthesized through a reaction of SiO₂ and CO₂ at 20 GPa at low temperature. Marqués et al. (20) predicted a monoclinic structure with ${\text{SiC}}_2{\mathrm{O}}_6$ stoichiometry in a pressure range of 7.2 to 41 GPa. Qu et al. (21) deduced from ab initio molecular dynamics (MD) simulations that CO₂ would react with SiO₂ and would form amorphous compounds at 43 GPa and 800 K. Yong et al. (22) proposed from ab initio MD simulations that a CO₂-SiO₂ solid solution would be formed and be stable above 10 GPa. Clearly, these modeling results are mutually inconsistent and none of them are supported by experiments, e.g., by comparison of a computed vibrational spectrum to an experimentally measured one, let alone by a comparison of a structure derived from diffraction data to a structural model from modeling.

In this study, we successfully synthesized silicon carbonate, Si[CO₃]₂, in laser-heated diamond-anvil cells (LH-DACs) and characterized its crystal structure using single-crystal x-ray microdiffraction, Raman spectroscopy, and density functional theory (DFT) calculations. We show that silicon carbonate can form at 40 to 45 GPa and $\approx 1800 \mathrm{K}$ through a reaction between silicic acid [SiO_x(OH)_4-2x] and CO₂ or between cristobalite (SiO₂) and CO₂. These p,T conditions correspond to depths of 1000 to 1200 km beneath Earth’s surface (23–25). We chose silicic acid as the starting material in one set of experiments because hydrous silicic fluids are often observed in diamond inclusions (26). The analogous reaction with cristobalite were carried out to test whether Si[CO₃]₂ would also be formed in a “dry” environment. Silicon carbonate remains stable upon decompression to at least 6 GPa at ambient temperature and might be a possible host of carbon in Earth’s lower mantle.

RESULTS

High-pressure experiments attempting the synthesis of silicon carbonate were carried out using LH-DACs between 30 and 45 GPa. Figure 1A shows the sample chamber of the DAC after the loading and compressing the DAC to 30(2) GPa. Silicon carbonate was first synthesized from a mixture of silicic acid and CO₂ that was compressed in a DAC to 40(2) GPa and then laser heated to 1800(200) K for 60 min to promote a reaction toward thermodynamically stable phases (Fig. 1B). Spatially resolved Raman spectroscopy shows that the novel phase forms at the center of the heated area (Fig. 1, C and D), evidenced by the appearance of strong Raman modes at 630 and 1250 cm⁻¹ (Fig. 2C), whereas the initial amorphous silicic acid produces no detectable Raman signal. Unreacted carbon dioxide fully transforms to the CO₂-V polymorph, which is stable at these p,T conditions (27, 28), and is characterized by a strong Raman mode at $\approx 805 {\text{cm}}^{–1}$ (Fig. 2B). Before the laser heating, CO₂-III was present across the sample chamber (Fig. 2A), which remains meta-stable up to very high pressures without heating (27, 29, 30). The experimental spectra of the CO₂ polymorphs III and V are in good agreement with the one derived from our DFT-based calculations (Fig. 2, A and B). Similar experiments at lower pressures suggest that the same silicon carbonate phase forms between 35(2) and 45(2) GPa; at 30(2) GPa, no new Raman modes were observed at $\approx 630 {\text{cm}}^{–1}$ or at $\approx 1250 {\text{cm}}^{–1}$, while at 35(2) GPa, weak peaks appeared. A Raman mode occurring at $\approx 1250 {\text{cm}}^{–1}$ at this pressure could be indicative for the presence of a sp² carbonate. For example, in Mg[CO₃] (50 GPa) or for Be[CO₃] (60 GPa), the [CO₃]²⁻ stretching mode had been observed at $\approx 1200 {\text{cm}}^{–1}$ (31, 32). However, it is well established that even for isostructural carbonates, the position of the [CO₃]²⁻ stretching mode changes substantially with pressure, the size of the cation, and across phase transitions (6, 31–33).

Fig. 1. Synthesis of silicon carbonate at 40(2) GPa.

(A) Sample chamber of the DAC with the unheated silicic acid + CO₂ at 30(2) GPa. (B) Silicic acid + CO₂ after laser heating to $\approx 1800\left(200\right) \mathrm{K}$ at 40(2) GPa. (C) Raman map of Si[CO₃]₂-P2₁/n after the synthesis overlaid on a picture of the sample chamber. (D) Raman map of Si[CO₃]₂-P2₁/n ($\approx 630 {\text{cm}}^{–1}$). (E) Raman map of CO₂-V ($\approx 805 {\text{cm}}^{–1}$). (F) XRD map of Si[CO₃]₂-P2₁/n, where the intensity of the reflections at $2\mathrm{\theta }\approx {4.6}^{\circ }$ is plotted.

Fig. 2. Raman spectroscopy at 40(2) GPa.

(A) Raman spectra for CO₂-III before the laser heating. (B) Raman spectra for CO₂-V GPa after the laser-heating. (C) Raman spectra of Si[CO₃]₂-P2₁/n after the synthesis. (D) Enlargement of the frequency range between 100 and 600 cm⁻¹. Experimental Raman spectra are shown in blue, and DFT-based calculations are shown in black. The Raman shifts of the theoretical spectra were scaled by a single factor of 2 to 4% to facilitate a comparison to the experimental data. This scaling effectively accounts for the well-established GGA-PBE underbinding. Peaks of CO₂-V in the Raman spectrum of Si[CO₃]₂-P2₁/n are marked by an asterisk ($\ast$). (E) Eigenvector of the atomic displacements for the characteristic Raman mode of Si[CO₃]₂-P2₁/n at $\approx 630 {\text{cm}}^{–1}$.

X-ray diffraction (XRD) was carried out to determine the chemical composition and crystal structure of the unknown silicon carbonate phase. First, we collected XRD data on a two-dimensional grid across the area where we had located the unknown phase by Raman spectroscopy. The XRD map of the reflections at $2\mathrm{\theta }\approx {4.6}^{\circ }$ (Fig. 1F) are in good agreement with the Raman map (Fig. 1, C and D) of the unknown phase. Hence, we collected diffraction data suitable for single crystal XRD on grid positions where unidentified reflections were present at $2\mathrm{\theta }\approx {4.6}^{\circ }$. We solved the crystal structure of the unknown phase at 40(2) GPa and found that it is a silicon sp² carbonate with Si[CO₃]₂ composition. It crystallizes in the monoclinic space group P2₁/n with Z = 4 (Fig. 3). The lattice parameters at 40(2) GPa are $a=4.141\left(2\right) \mathrm{\AA }$, $b=4.1916\left(9\right) \mathrm{\AA }$, $c=7.2283\left(8\right) \mathrm{\AA }$, and $\mathrm{\beta }=89.88{\left(2\right)}^{\circ }$ [$V=125.47\left(6\right) {\mathrm{\AA }}^3$]. The structural data are listed in tables S1 and S2. The reflection intensity distribution visualized in the XRD map (Fig. 1H) belongs to a combination of the ($0\overline{1}1$), (002), (101), and ($\overline{1}01$) lattice planes of Si[CO₃]₂-P2₁/n.

Fig. 3. Crystal structure of Si[CO₃]₂ obtained from synchrotron single-crystal x-ray diffraction at 40(2) GPa.

The experimental crystal structure of Si[CO₃]₂ (P2₁/n, Z = 4) is shown (A) along the b axis and (B) along the c axis. The positions of the silicon atoms are partially occupied. (C) A [SiO₆] octahedron with selected Si─O bond distances. (D) Geometry of the symmetrically independent [CO₃]²⁻ group with selected C─O bond distances.

In the crystal structure of Si[CO₃]₂-P2₁/n, two Wyckoff positions (2d at $0,\frac{1}{2},\frac{1}{2}$ and 2b at $0,0,\frac{1}{2}$) are occupied by silicon. An unconstrained refinement of the occupancy resulted in a total silicon content of $\approx 1$ per formula unit, with an approximate occupation of 0.75 and 0.25 for site 2b and site 2d, respectively. This is consistent with the charge balance of one Si⁴⁺ cation by two [CO₃]²⁻ groups. A refinement with a full silicon occupancy of the 2b position resulted in a strong increase of R₁ value from 4.1 to 20%. In addition, we solved the crystal structure of another Si[CO₃]₂-P2₁/n crystal in the sample chamber of the DAC at the same pressure. The atomic positions after both structure refinements are identical within the experimental uncertainties. However, the silicon occupancy between the two positions is substantially different, as in the second crystal, and it was 0.95 and 0.05 for the two sites. Again, a refinement with a full silicon occupancy of the 2b position resulted in a substantially increased R₁ value (from 5 to 7%).

To understand the role of the partial site occupancies in the experimental structural models, we carried out DFT-based full geometry optimizations. For the DFT calculations, we fully occupied the 2b site and left the other site (2d) empty. This allows to use a cell with the same volume ($\approx 125 {\mathrm{\AA }}^3$) than the experimental structural model. The theoretical Raman spectrum derived from this DFT calculation reproduces the experimental Raman spectrum very well (Fig. 2, C and D). In the second step, we performed a DFT-based full geometry optimization with an inversion of the site occupancies and fully occupied the 2d site. Our analysis showed that an inversion of the site occupancies leads to the same structure, which is related to the first structure by a shift of origin only. Hence, structures with partial site occupancies can be viewed as a combination of two identical structures. The partial site occupancies describe an average occupancy. If a specific site in a specific unit cell is occupied, the likelihood that the close neighbor is occupied is very small. In the next unit cell, the occupancies may be reversed. In the averaged structural model, it then looks like two silicon atoms may be very close to each other, but due to short-range ordering, this is not the case. Raman spectroscopy is a local probe, and as the local environment of the Si atoms is the same for the two sites, Raman spectroscopy is not sensitive to the site occupancies in the present case. We analyzed the eigenvectors of some modes. The dominant Raman mode at $\approx 630 {\text{cm}}^{-1}$ belongs to a complex distortion of the [SiO₆] octahedra (Fig. 2E). In addition, our analysis of the eigenvectors clearly shows that the Raman mode at $\approx 1250 {\text{cm}}^{-1}$ belongs to a [CO₃]²⁻ stretching vibration (fig. S3).

The crystal structure of Si[CO₃]₂-P2₁/n obtained here experimentally is closely related to the results obtained from a crystal structure prediction study by Marqués et al. (20). While the positions of the carbon and oxygen atoms agree with each other, in the crystal structure prediction study, the silicon atom was only located on one position and no partial site occupancies were reported. The one reported position for the silicon atoms agrees with one of the positions in our experimental data, while the second position was unoccupied. Marqués et al. (20) also suggested that the stability field of Si[CO₃]₂-P2₁/n ranges from about 7.2 to 41 GPa, whereas in our experiments, the compound could not be synthesized below roughly 35 GPa. Silicon carbonate remains present upon decompression at ambient temperature to at least 6(1) GPa. Our Raman spectra (fig. S2) indicate a progressive deterioration of crystallinity during decompression, but bands attributed to the high-pressure phase are still clearly observable. Nevertheless the quality of the diffraction data collected at 6 GPa is reasonable and we were able to refine the data (table S1). On further decompression, a refinement of the data was not possible, but in the open DAC at 0 GPa, we could determine lattice parameters of Si[CO₃]₂-P2₁/n. These are in reasonable agreement with those obtained from the DFT calculations. In the 0-GPa dataset, we could identify stishovite as a decomposition product. We find no evidence for carbon-silicon mixing on a single Wyckoff site. In contrast to the findings by Santoro et al. (15), our spectroscopic data clearly show the [CO₃]²⁻ stretching vibration (Fig. 2C), confirming the presence of carbonate groups in the crystal structure. This is independently supported by DFPT calculations, which predict an intense C─O stretching mode (Fig. 2C). We therefore conclude that the absence of this spectroscopic feature is a strong indicator that [CO₃]²⁻ groups are not present.

Si[CO₃]₂-P2₁/n is an sp² carbonate with nearly trigonal-planar [CO₃]²⁻ groups as central building blocks (Fig. 3D). One symmetrically independent [CO₃]²⁻ group is present in Si[CO₃]₂-P2₁/n. The C─O bond distances within the [CO₃]²⁻ group range from 1.240(4) to 1.269(3) Å and scatter around the values derived from our DFT calculations (1.261 to 1.265 Å). In addition, the range of experimental values for the O─C─O angles [119.0(1)° to 120.6(2)°] is slightly larger than the range of the values obtained from DFT (119.5° to 120.3°). The Mulliken population analysis of the C─O bonds in the [CO₃]²⁻ groups yielded nearly identical values (0.89 and 0.90 e⁻/ų). The geometry of the [CO₃]²⁻ groups is in agreement with the geometry observed in other sp² carbonates hosting small cations such as Be[CO₃] or Mg[CO₃] at elevated pressures (32, 34, 35). In the crystal structure of Si[CO₃]₂-P2₁/n, the [CO₃]²⁻ groups are arranged in layers parallel to the b axis.

The crystal structure of Si[CO₃]₂-P2₁/n is also characterized by [SiO₆] octahedra (Fig. 3C). The Si─O bond distances range from 1.720(2) to 1.738(2) Å at 40(2) GPa. The experimental Si─O bond distances are in agreement with our DFT-based calculations (1.71 to 1.73 Å). The DFT calculations reveal a predominantly covalent character of the Si─O bond with bond populations of 0.39 and 0.40 e⁻/ų. The Si─O bond distances are in very good agreement with the ones observed in stishovite (SiO₂-P4₂/mnm) at the same pressure ($\approx 1.71 \mathrm{\AA }$) (36). Structural units of the type [M(CO₃)₆]^n–, in which an octahedrally coordinated cation is surrounded by six parallel carbonate groups, occur in common carbonates such as calcite (Ca[CO₃]), magnesite (Mg[CO₃]), and siderite (Fe[CO₃]), as well as in less common high-pressure Be-carbonate (Be[CO₃]) and Al-carbonate (Al₂[CO₃]₃) (32, 37, 38). These units accommodate a remarkably wide range of cation sizes within the [MO₆] octahedron, from about 1.00 Å for Ca²⁺ down to about 0.54 Å for Al³⁺, 0.45 Å for Be²⁺, and 0.40 Å for Si⁴⁺ (32, 37–39).

Figure 4 shows a comparison of the [MO₆] octahedra in Si[CO₃]₂ with the calcite-type carbonates Ca[CO₃] and Be[CO₃]. The [MO₆] octahedra generally assemble into close-packed layers through shared carbonate groups: In calcite-type structures, the layers are flat and stacked in an ABCABC pattern (fig. S6), while in Al₂[CO₃]₃, the layers are buckled and follow an ABAB sequence. In silicon carbonate, the layers are flat but stack in a single-layer AAA sequence (fig. S6), and partial Si occupancies help maintain charge balance and prevent formation of short Si-Si distances between [SiO₆] octahedra from adjacent layers. The [SiO₆] octahedra of the partially occupied silicon atoms share faces along the b axis (fig. S6). Along the a and c axes, the [SiO₆] octahedra are connected by sharing a [CO₃]²⁻ group. These structural similarities among carbonates of the most commonly encountered metal cations in the mantle (Mg, Ca, Fe, Si, and Al) suggest the possibility of forming solid solutions between end-members of high-pressure carbonate-silicate phases. Experimental and theoretical studies show that silicate melts can dissolve substantial amounts of CO₂, with solubility increasing with pressure (40). Our observations indicate that [Si(CO₃)₆]⁸⁻ units can readily form under lower mantle conditions, suggesting that they may act as rigid structural fragments in models of carbonate-silicate melts in Earth’s deep interior.

Fig. 4. [MO₆] octahedra coordinated by six parallel [CO₃]²⁻ groups in different carbonates.

(A) In Si[CO₃]₂-P2₁/n at 40(2) GPa. (B) In Be[CO₃]-$R\overline{3}c$ at 60(2) GPa (32). (C) In Ca[CO₃]-$R\overline{3}c$ at ambient conditions (37).

Recently the crystal structure of U[CO₃]₂-C2, a chemically simple carbonate also hosting a tetravalent cation, has been described at 20(2) GPa (41). Similar to Si[CO₃]₂-P2₁/n, the crystal structure of U[CO₃]₂-C2 is characterized by trigonal-planar [CO₃]²⁻ groups. However, in U[CO₃]₂-C2, the [CO₃]²⁻ groups are not oriented in layers. Instead, in a first approximation, they are randomly distributed around the uranium atoms. In addition, the uranium coordination by oxygen atoms is noticeably higher (10 and 12) and nearly spherical in contrast to the [SiO₆] octahedra in Si[CO₃]₂-P2₁/n. This is probably due to the much larger ionic radius of tetravalent uranium cations (0.89 Å in sixfold coordination) in contrast to the small silicon cations (0.40 Å) (39).

In a second set of experiments, synthetic cristobalite, obtained by heating SiO₂ glass, was used as a starting material. After laser heating the cristobalite and CO₂ mixture at conditions similar to those used in the first experiment, we could identify Si[CO₃]₂ by its distinct Raman spectroscopic signature (fig. S7). Moreover, the anhydrous composition is charge balanced. Also, our DFT calculations confirm the anhydrous structural model. Furthermore, no Raman signal at high wave numbers (3000 to 3500 cm⁻¹) has been observed after heating the cristobalite + CO₂ mixture, which could be indicative for the presence of O─H bonds.

We used our DFT calculations to obtain the p,V relation for Si[CO₃]₂-P2₁/n between 0 and 50 GPa. The calculated p,V data were used to derive the bulk modulus (K₀) and its pressure derivative (K_p). From an EoS fit to the p,V data (fig. S4), we obtained a bulk modulus of K₀ = 53.1(4) GPa with K_p = 6.40(5). The ambient pressure bulk modulus is in good agreement with the one derived from our stress-strain calculations [52(1) GPa]. Without using a correction scheme for van der Waals (vdW) interactions during the calculations, the bulk modulus derived from the EoS fit is noticeably lower [K₀ = 37(1) GPa with K_p = 7.3(1)]. The elastic stiffness coefficients are listed in table S3. The bulk modulus is noticeably lower than for other sp² carbonates hosting sixfold coordinated cations, such as Ca[CO₃] [67(2) GPa], Mg[CO₃] [107(1) GPa], or Be[CO₃] [153.0(3) GPa] (32, 42). In contrast, the pressure derivative of the bulk modulus is noticeably higher for Si[CO₃]₂-P2₁/n. The compression mechanism for Si[CO₃]₂-P2₁/n is strongly anisotropic. Si[CO₃]₂-P2₁/n is highly compressible along the b axis and incompressible along the a and c axes (fig. S5). This compression behavior can be explained by the layered crystal structure of Si[CO₃]₂-P2₁/n. The [CO₃]²⁻ groups within the a, c planes are incompressible, but during compression, the space between the [CO₃]²⁻ group layers is strongly reduced. Experimentally, the monoclinic angle is always found to be very close to 90° at all pressures. Because of the limited access to the reciprocal space in the DAC, the error on the lattice parameter may be too large to distinguish between orthorhombic and monoclinic symmetry. However, the crystal structure could not be solved in a higher space group symmetry. This is consistent with our DFT-based calculations, which show that the relaxed structures always maintain the monoclinic P2₁/n space group symmetry over the whole investigated pressure range while the monoclinic angle remains between 90.19° (50 GPa) and 90.35° (0 GPa).

DISCUSSION

We have synthesized the anhydrous silicon sp² carbonate Si[CO₃]₂-P2₁/n through the reaction of silicic acid and CO₂ and by the reaction of cristobalite with CO₂ under lower-mantle pressure-temperature conditions [40(2) GPa and 1800(200) K]. At $\approx 40$ GPa, the density of Si[CO₃]₂-P2₁/n is approximately 3.9 g cm⁻³, slightly lower than that of typical uppermost lower-mantle assemblages ($\approx 4.4 \text{to} 4.6 \mathrm{g} {\text{cm}}^{-3}$) (23). Thus, once formed, this phase would have neutral to slightly negative buoyancy and could represent a localized, but viable, host for carbon in the deep mantle. At 40 GPa, the seismic velocities computed from the elastic stiffness tensor are v_p = 11.3 km s⁻¹ and v_s = 6.3 km s⁻¹, which is very close to the corresponding values of the preliminary reference earth model (PREM) (23), so that the presence of Si[CO₃]₂ would not have an identifiable seismic signature. To understand whether Si[CO₃]₂-P2₁/n may occur at the p,T,x conditions of Earth’s lower mantle, it is now of relevance to benchmark its stability in the presence of the main mantle minerals.

MATERIALS AND METHODS

A detailed description of the experimental and computational methods is available in the Supplementary Materials.

Sample synthesis

The high-pressure experiments were carried out using laser-heated Boehler-Almax–type DACs (43). For the loading of the DAC, we used in the first set of experiments commercial silicic acid [SiO_x(OH)_4–2x] powder (99.9% purity, Sigma-Aldrich, Merck KGaA, Darmstadt, Germany) without further purification. In the second set of experiments, we used phase pure synthetic cristobalite (SiO₂) obtained by heating silica glass to 1823 K in an oxygen-controlled high-temperature oven for 6 hours following an earlier approach (44). CO₂ gas was used as purchased (Nippon gases, purity ≥99.996%) and loaded as dry ice by cryogenic loading into the DACs using a custom-built cryogenic loading system (45). The DAC was cooled down to $\approx 120 \mathrm{K}$ for the cryogenic loading. CO₂-I (dry ice) was directly condensed from a gas jet into the sample chamber. The pressure during compression was determined from the position of the high-frequency edge of the diamond Raman band (46). At the target pressure of the experiment, laser heating was performed from both sides using a custom-built setup equipped with a Coherent Diamond K-250 pulsed CO₂ laser (λ = 10,600 nm) (6).

Sample characterization

Raman spectroscopy was performed in DACs at elevated pressures using an Oxford Instruments WITec alpha 300R Raman imaging microscope. The WITec alpha 300R was equipped with an Olympus SLMPan N 50× objective, and the measurements were performed using a 532-nm laser and the 1800 grooves mm⁻¹ grating of the WITec UHTS 300S (VIS-NIR) spectrograph. Raman maps were measured on a grid with a step size of 0.5 μm. Single-crystal synchrotron XRD was carried out at two different synchrotron beamlines. At 40(2) GPa, the experiments were carried out at the synchrotron PETRA III (DESY) in Hamburg, Germany, at the Extreme Conditions Beamline P02.2 (47). The beam size on the sample was $\approx 2\times 2 {\mathrm{\mu m}}^2$ [full width at half maximum (FWHM)] using a wavelength of 0.2903 Å (43.7 keV), and the diffraction data were collected using a Perkin Elmer XRD1621 detector. At 6(1) GPa and ambient conditions, the experiments were carried out at the European Synchrotron Radiation Facility (ESRF) in Grenoble, France, at the Materials Science Beamline line ID11 (48). The beam size on the sample was $\approx 0.6\times 0.6 {\mathrm{\mu m}}^2$ (FWHM) using a wavelength of 0.2846 Å (43.6 keV), and the diffraction data were collected using an Eiger2 X 4M CdTe detector. We rotated the DAC by ±34° around the vertical axis perpendicular to the beam while collecting frames in 0.25° steps with 2 s acquisition time per frame. The structure solution and refinement on the single-crystal diffraction data were performed using the software package OLEX2 using SHELXT for the crystal structure determination and SHELXL for the refinement (49–51).

DFT-based calculations

First-principles calculations were carried out within the framework of DFT, using the Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional and the plane wave/pseudopotential approach implemented in the CASTEP simulation package (52–54). CASTEP and auxiliary programs, including those for a symmetry analysis, an analysis of the electron density distribution, and for visualizations, were run within the BIOVIA Materials Studio suite of programs (55). We used the correction scheme for vdW interactions developed by Tkatchenko and Scheffler (56). Phonon frequencies were obtained from density functional perturbation theory (DFPT) calculations (57, 58). Raman intensities were computed using DFPT with the “2n + 1” theorem approach (59).

Supplementary Materials

This PDF file includes:

Supplementary Materials and Methods

Supplementary Text

Figs. S1 to S7

Tables S1 to S3

References