The crystal structures of Fe-bearing MgCO3
 sp
 2- and sp
 3-carbonates at 98 GPa from single-crystal X-ray diffraction using synchrotron radiation

We report the phase transition of Mg0.85Fe0.15CO3 sp 2- to Mg2.53Fe0.47C3O9 sp 3-carbonate (MgCO3–II phase) at 98 GPa and describe their crystal structures by means of single-crystal X-ray diffraction experiments in laser-heated diamond anvil cells.

The crystal structure of MgCO 3 -II has long been discussed in the literature where DFT-based model calculations predict a pressure-induced transition of the carbon atom from the sp 2 to the sp 3 type of bonding. We have now determined the crystal structure of iron-bearing MgCO 3 -II based on singlecrystal X-ray diffraction measurements using synchrotron radiation. We laserheated a synthetic (Mg 0.85 Fe 0.15 )CO 3 single crystal at 2500 K and 98 GPa and observed the formation of a monoclinic phase with composition (Mg 2.53 Fe 0.47 )C 3 O 9 in the space group C2/m that contains tetrahedrally coordinated carbon, where CO 4 4À tetrahedra are linked by corner-sharing oxygen atoms to form three-membered C 3 O 9 6À ring anions. The crystal structure of (Mg 0.85 Fe 0.15 )CO 3 (magnesium iron carbonate) at 98 GPa and 300 K is reported here as well. In comparison with previous structure-prediction calculations and powder X-ray diffraction data, our structural data provide reliable information from experiments regarding atomic positions, bond lengths, and bond angles.

Chemical context
Carbonates and their high-pressure behaviour have attracted significant interest because of their potential role as carbonbearing phases in the deep Earth. Recent discoveries of novel compounds that contain tetrahedral CO 4 4À units (e.g., Merlini et al., 2015;Cerantola et al., 2017) increase the relevance of such studies, as the new high-pressure phases may be stable at conditions prevalent in the deep part of Earth's lower mantle. In addition, theoretical modelling predictions imply potential structural analogues of CO 4 4À -bearing carbonates and silicates, and thus carbonates with tetrahedrally coordinated carbon may be important to understanding the complex geochemistry of Earth's mantle.
Carbonates with tetrahedrally coordinated carbon are not well characterized, despite their potential significance, as structural studies have to be carried out under high-pressure conditions and are therefore challenging. A reliable structural characterization is, however, a prerequisite for determining phase stabilities and to understand, for example, why the p,Tphase diagram of MgCO 3 is relatively simple compared to the dense phase diagram of CaCO 3 (see summary in Bayarjargal et al., 2018).
It is generally accepted that magnesite (MgCO 3 ) transforms to MgCO 3 -II at 80-115 GPa (Isshiki et al., 2004;Boulard et al., 2011Boulard et al., ,2015Maeda et al., 2017). Models based on density functional theory (DFT) (Oganov et al., 2008) and interpretation of X-ray diffraction data and IR spectra imply that MgCO 3 -II contains carbon in a tetrahedral coordination (Boulard et al. 2011(Boulard et al. , 2015. While structure-prediction techniques are undoubtedly useful for preliminary surveys of phase stabilities, they provide a range of possible new phases, derived under constraints such as unit-cell contents. Powder diffraction data obtained at pressures around 100 GPa generally do not yield accurate structure determinations, and typically do not allow unambiguous assignment of the space group or site occupancies. In contrast, single-crystal X-ray diffraction is a powerful and unique tool that can provide accurate structure refinements under these conditions (Boffa Ballaran et al., 2013). Well-established statistical parameters allow an assessment of the reliability of the structural model. Other carbonate structures with tetrahedral CO 4 4À units at extreme conditions have previously been reported using this method, such as the novel phases Fe 4 C 3 O 12 in space group R3c, (Mg,Fe) 4 C 4 O 13 in C2/c (Merlini et al., 2015; and Ca(Fe,Mg) 2 C 3 O 9 in Pnma . These results lead to two conclusions. Firstly, the stability fields of carbonates strongly depend on their composition. Secondly, CO 4 4À units have the ability to form polymeric networks, and thus are potential analogues to silicates.

Structural commentary
Under ambient conditions (Mg 0.85 Fe 0.15 )CO 3 crystallizes in the calcite-type structure in space group R3c. Iron and magnesium share the same crystallographic site (Wyckoff position 6b; site symmetry 3.) and are coordinated by six oxygen atoms, while the CO 3 2À units form planar equilateral triangles with pointgroup symmetry 32 (e.g. Lavina et al., 2010). After compression to 98 (2) GPa at ambient temperature, X-ray diffraction data of (Mg 0.85 Fe 0.15 )CO 3 can still be indexed in the R3c space group (Fig. 1, Table 1). However, the unit-cell volume is decreased by nearly 32% compared to ambient conditions. This result challenges a recent suggestion based on DFT-based calculations that predicted a structural transformation of MgCO 3 to a triclinic phase at 85-101 GPa and 300 K (Pickard & Needs, 2015). At 98 GPa, the C-O bond length [1.195 (8) Å ] has decreased only by $7% compared to the structure at ambient conditions, thus reflecting the highly incompressible nature of the CO 3 2À units. On the other hand, the (Mg/Fe)-O bonds [1.855 (5) Å at 98 GPa] display a much more compressible behavior ($12% bond-length and $32% octahedra-volume shrinkage compared to ambient conditions). On a last note, it is well known that rhombohedral carbonates can be described as a distortion of the NaCl (B1) structure. Previously, the t parameter, t ¼ 4a= p 2c, where a and c are the lattice parameters) has been used to evaluate the degree of distortion (Gao et al., 2014). We observed that at 98 GPa and 300 K, t '1 for (Mg 0.85 Fe 0.15 )CO 3 , which means that at the conditions of our experiment the (Mg/Fe) cations and the CO 3 2À anions are arranged in the manner of a nearly ideal NaCl (B1) structure.
After annealing at 2500 K and 98 GPa, we observed a phase transition to a polymorph in which carbon is tetrahedrally coordinated by oxygen. The newly formed phase with chemical formula (Mg 2.53 Fe 0.47 )C 3 O 9 (as determined from structural refinements, see below) has monoclinic symmetry, and the diffraction pattern indicates space group C2/m (Fig. 2, Table 1). We identify this phase as the MgCO 3 -II structure that was previously predicted (Oganov et al., 2008;Boulard et al., 2015). In contrast to previous studies, we provide an accurate structure solution and refinement based on single crystal X-ray diffraction data. The structure consists of threemembered C 3 O 9 6À rings formed by corner-sharing CO 4 tetrahedra ( Crystal structure of (Mg 0.85 Fe 0.15 )CO 3 at 98 GPa and prior to laserheating shown in a projection along the c axis. The building blocks of the unit cell appear on the right. Here, iron occupies the same sites as the magnesium atoms.
(1) The M1 site is located on a twofold rotation axis (Wyckoff position 4g) and is occupied by Mg and Fe in a 0.917 (17):0.083 (17) ratio. This site is surrounded by eight oxygen atoms forming a distorted square antiprism (dark blue); (2) The M3 site is situated on a mirror plane (4 i) in a 0.61 (2):0.39 (2) Mg:Fe ratio and a coordination number of 10 (blue; can be described as half cuboctahedra merged through hexagonal-based faces with hexagonal pyramids); (3) M2 is likewise situated on a mirror plane (4 i) and is fully occupied by Mg in [MgO 6 ] octahedra (magenta). The maximum and minimum bond lengths of each cation site from its neighbouring oxygen atoms are shown in Table 2. At 98 GPa the C-O bond lengths of the two different CO 4 4À carbonate groups [C1 is located on a general site (8 j) and C2 on a mirror plane (4 i) vary from 1.287 (18)-1.409 (13) Å and the C-O-C inter-tetrahedral angle is $112 .
From all proposed structural models for MgCO 3 -II over the last two decades, only one appears to successfully match the structure model that we report here. On the basis of powder X-ray diffraction (PXRD) experiments and variable-cell simulations, Oganov et al. (2008) suggested several energetically favourable structural models for MgCO 3 -II, one of which is in space group C2/m. While our structural solution and refinement from the experimental data is clearly similar to the theoretical predictions by Oganov et al. (2008), the different composition of the materials and the small differences in the structural parameters required us to check additionally whether theoretical calculations with our model as the starting one would lead to the same result as that reported by Oganov et al. (2008). We performed such a test and confirm that our results and those of Oganov et al. (2008) are the same within the accuracy of the methods. More concretely, we performed DFT-based model calculations using the plane wave/pseudopotential CASTEP package (Clark et al., 2005). Pseudopotentials were generated 'on the fly' using the parameters provided with the CASTEP distribution. These pseu-  (1974)  dopotentials have been tested extensively for accuracy and transferability (Lejaeghere et al., 2016). The pseudopotentials were employed in conjunction with plane waves up to a kinetic energy cutoff of 1020 eV. The calculations were carried out with the PBE exchange-correlation function (Perdew et al., 1996). For simplicity, we assumed that all three M1, M2 and M3 positions are fully occupied by Mg 2+ . The calculations revealed that the energies of our structural model and that of Oganov et al. (2008) are indeed, identical. The DFT calculations gave C-O distances in good agreement with experimental data. Each carbon atom is coordinated by two oxygen atoms that are each shared with another tetrahedrally coordinated carbon, and two that are not shared. The C-O distances for the latter are significantly shorter [1.29 Å < d(C-O) < 1.32 Å ] than the former [1.33 Å < (C-O) < 1.41 Å ]. A Mulliken bond-population analysis shows that for the long C-O bonds there is a significant bond population of $0.5 e À Å À3 . This is less than the value for the short bonds, where the bond population is $0.9 e À Å À3 , but this still is a predominantly covalent bond, and justifies the description as a tetrahedrally coordinated carbon atom. The formation of (C 3 O 9 ) 6À carbonate rings was previously observed in Ca(Fe,Mg)C 3 O 9 (dolomite-IV) after laser heating of Ca(Fe,Mg)CO 3 at 115 GPa . However, dolomite-IV is topologically different from the MgCO 3 -II structure that we report here. Unlike (Mg 2.53 Fe 0.47 )C 3 O 9 , Ca(Fe,Mg)C 3 O 9 crystallizes in the orthorhombic system (space group Pnma), thus highlighting the significance of the metal cations that are present in the carbonate. Upon decompression at ambient temperature, (Mg 2.53 Fe 0.47 )C 3 O 9 reflections become broad and weak, and almost disappear at $74 GPa (Fig. 3a-c). This may be an indication of either amorphization or sluggish back-transformation to a carbonate with trigonal symmetry. Anticipating that further heating would aid recrystallization, we laserheated the sample at 74 GPa and 2000 (150) K for a few seconds. Wide images collected on the temperature-quenched sample indicated the formation of the calcite structure-type carbonate (Fig. 3d).

Synthesis and crystallization
Magnesium carbonate crystals with 15(AE4) mol% Fe were grown following the procedure reported by Chariton et al. (2020). The composition of the starting material was determined by single-crystal X-ray diffraction under ambient conditions as (Mg 0.85 Fe 0.15 )CO 3 . A single crystal of $7 mm size in all dimensions was loaded inside the sample chamber of a BX90-type diamond anvil cell equipped with bevelled Boehler-Almax type diamonds (culet diameter 80 mm). Rhenium and neon were used as the gasket material and pressure-transmitting medium, respectively. The pressure was determined using the equation of state (EoS) of solid Ne (Fei et al., 2007). First, the sample was compressed up to 98 GPa and a single-crystal collection took place at 300 K. Consequently, the same crystal was laser-heated from both sides up to 2500 (150) K for a few seconds and then quenched to room temperature. Finally, we performed a 5Â5 grid of still-image collection with a 2 mm step and 1 s exposure time around the center of the sample. This strategy was used to locate the most heated area of the crystal and the best spot to collect singlecrystal X-ray diffraction patterns during rotation of the cell. Single-crystal data collection was performed as a series of ! scans over the range AE35 with a step of 0.5 .

Refinement
Details of the data collection, structure solution and refinement are summarized in Table 1. In the case of the (Mg 0.85 Fe 0.15 )CO 3 dataset collected at 98 GPa, the limited number of available reflections required us to fix the Fe content according to our ambient condition estimates (see also "Synthesis and Crystallization" section). On the other hand, during the structure refinements of (Mg 2.53 Fe 0.47 )C 3 O 9 all three cation sites (i.e. M1, M2 and M3) were tested for their ability to host Fe by refining the site occupancies. As described above, only the M1 and M3 sites were eventually found to accommodate $16(AE3) mol % Fe in total. Note that the resulting 5.38 Mg:Fe ratio of (Mg 2.53 Fe 0.47 )C 3 O 9 is almost  identical to the starting 5.67 Mg:Fe ratio of (Mg 0.85 Fe 0.15 )CO 3 within the accuracy of our method. Therefore, it is safe to conclude that nearly none or only a negligible amount of Fe was lost during the observed phase transition. The crystal structure of (Mg 2.53 Fe 0.47 )C 3 O 9 solved at 98 GPa was used for the structure refinements of the data of the same phase collected during decompression. Due to the limited angular range caused by the laser-heated DAC, the resolution of the data set was not sufficient to refine the anisotropic displacement parameters. Therefore, all atoms were refined with the isotropic approximation.  (Palatinus & Chapuis, 2007); program(s) used to refine structure: Jana2006 (Petříček et al., 2014); molecular graphics: VESTA (Momma & Izumi, 2011); software used to prepare material for publication: publCIF (Westrip, 2010). Weighting scheme based on measured s.u.'s w = 1/(σ 2 (F) + 0.000144F 2 ) (Δ/σ) max = 0.001 [a] All atomic displacement parameters were refined in the isotropic approximation