Polymerized 4-Fold Coordinated Carbonate Melts in the Deep Mantle

Our understanding of the deep carbon cycle has witnessed amazing advances in the last decade, including the discovery of tetrahedrally coordinated high pressure (P) carbonate phases. However, little is known on the properties of molten carbonates at depth, while their properties at lower mantle conditions are unknown. Here, we report the structure and density of FeCO3 melts and glasses from 44 GPa to 110 GPa by means of in situ x-ray synchrotron diffraction, and ex situ Raman and x-ray Raman spectroscopies. Carbon is fully transformed to 4-fold coordination, a bond change recoverable at ambient P. While low P melts react with silica, resulting in the formation of silico-carbonate glasses, high P melts are not contaminated but still quench as glasses. Carbonate melts are therefore polymerized, highly viscous and poorly reacting with silicates in the lower mantle, in stark opposition with their low P properties.


INTRODUCTION
Although the lower mantle is mostly a reducing environment with the presence of reduced Fe (Frost et al., off-axis heating system to avoid using carbon mirrors that would add to the x-ray background signal and 63 compromise processing of the scattered signal, T could not be measured by pyrometric techniques. FeCO 3 64 melting curve has only been measured up to 20 GPa (Kang et al., 2015), where it reaches 1865 K. The 65 stishovite to CaCl 2 SiO 2 transition has been investigated up to 90 GPa (Fischer et al., 2018), this constrains 66 T to a maximum of 2300 K at 79 GPa and 2500 K at 83 GPa as CaCl 2 is the observed SiO 2 structure for the three highest P runs, while stishovite is observed below. We therefore consider that x-ray diffraction 68 patterns were collected on molten FeCO 3 within the 2000 K-2500 K interval except for the highest P 69 point that is only constrained to below 3500 K from extrapolation of the stishovite-CaCl 2 Clapeyron slope 70 (Fischer et al., 2018). P is measured at room T using fluorescence of a ruby sphere added in the sample 71 chamber (Mao et al., 1986) and SiO 2 equations of state (Andrault et al., 1998;Nishihara et al., 2005) 72 for quenched samples, and using only SiO 2 equations of state for molten samples with error bars on P 73 including the effect of a 2000 K-2500 K T -range, and up to 3500 K for the 110 GPa data point.
74 75 X-ray diffraction methods 76 We collected in situ high P -T x-ray diffraction data in laser-heated diamond anvil cells at the extreme 77 conditions beamline P02.2 at the PETRAIII synchrotron. We used symmetric diamond-anvil cells equipped 78 with 70 • opening Boehler-Almax seats in order to access a wider q-range up to 10Å −1 , and reduce the 79 diamond Compton contribution as Boehler-Almax anvils are only 1.5 mm thick. The x-ray monochromatic 80 beam (42.7 keV) was focussed down to a size of 4 × 6 µm 2 , allowing high spatial resolution in direct 81 space. To limit iron migration away from the laser heating spot due to Soret effect, the laser shutters were 82 opened only once the targeted power was reached, and held open for 10 s during which 10 x-ray diffraction 83 patterns of 1 s acquisition time were recorded on a Perkin-Elmer 2-D detector. 2-D patterns were integrated 84 using the Fit2D software (Hammersley et al., 1996). In order to isolate the scattered intensity from the 85 molten FeCO 3 only, each sample was removed from the gasket, and the gasket put back in place to collect 86 x-ray data on the empty cell. Obtained patterns were then scaled vertically to match the baseline of x-ray 87 patterns collected on the starting crystalline sample under P (Sanloup and de Grouchy, 2018). This last 88 step ensures that any P effect on the background is corrected for. Amongst eight successful runs (Table 1) 89 for which full melting was observed, intensity from molten FeCO 3 could only be processed for the highest 90 P run for which the sample vs SiO 2 platelets thickness ratio was slightly higher, the scattered intensity 91 being too weak for the lower P points. All glass patterns could be processed. The x-ray diffracted intensity 92 data are converted into the structure factor, S(q) (Fig.?? and Fig.4), using the Ashcroft-Langreth formalism.

93
The radial distribution function g(r) (Fig.3B), that describes ion-ion contributions in real space, is obtained 94 by Fourier transforming of S(q), where n = ρN A M , N A is the Avogadro number, M the mean atomic molar mass, and ρ the density. there should not be any signal, i.e. below the minimum interatomic distance (r < 0.95Å here). This method 100 requires that the background, essentially the Compton signal from the diamond anvils that dominates the 101 total diffracted intensity, is perfectly subtracted.
As the C-O contribution is distinct on g(r) of quenched glasses up to 83 GPa, we also ran consistency checks by fixing the C-O coordination number to 4 as indicated by x-ray Raman spectra (cf Results section), 104 and simulating the C-O contribution using the obtained density values against a gaussian with the following 105 equation: and

114
Raman and x-ray Raman spectra were collected at ambient conditions on glassy FeCO 3 recovered from 115 x-ray diffraction experiments and from additional laser-heated diamond anvil cell synthesis respectively.

116
X-ray Raman data were collected at an incident energy of 9.7 keV at the C K-edge on beamline ID20 of heating duration in this work); alternatively, Fe stabilizing effect on high P carbonates could be at stake. 139 We observe no disproportionation of Fe as was reported in the crystalline state in some studies (Boulard

143
A striking characteristic of glassy FeCO 3 is its strong first sharp diffraction peak (FSDP) that persists in 144 the structure factor up to the highest P investigated (Fig.3A), indicative of a strong medium-range order.

145
This is in stark contrast to silicate glasses that lose their medium-range order with increased P (Sato and For glasses quenched at 11 GPa and 15 GPa, the x-ray structure factor, S(q), is intermediate between that

156
of pure SiO 2 glass (Sato and Funamori, 2008) and high-P FeCO 3 glasses (Fig.4). SEM image of sample 8 157 (15 GPa, Fig.2) shows heterogeneities in the quenched glass, which indicates that the x-ray structure factor 158 likely averages at least two types of glass structure and therefore data cannot be interpreted quantitatively.

159
The x-ray Raman C K-edge spectrum of quenched FeCO 3 glass shows no presence of sp2 3-fold 160 carbon characterized by an intense π * peak at 290 eV (Fig.5, π * peak). Only the σ * peak of tetrahedrally

DISCUSSION
The 3-fold to 4-fold transition occurs in molten Fe-carbonates at P less or equal to 53 GPa, compared is also opposite to the behaviour of molten basalt that systematically quenches as crystalline phases above Lobanov, S. S., Stevanovic, V., Gavryushkin,P. N.,Litasov,K. D.,Greenberg,E.,Prakapenka,V. B.,297 Oganov, A. R. and Goncharov, Al. F. (2017). Raman spectroscopy and x-ray diffraction of sp (3)  Maeda, F., Ohtani, E., Kamada, S., Sakamaki, T., Hirao, N., and Ohishi, Y. (2017). Diamond formation in FIGURE CAPTIONS Figure 1. Microphotograph of the sample after laser heating at 110 GPa. Single shot laser heating resulted in the formation of a quasi-spherical pure carbonate glass that was removed from the gasket for EPMA and/or SEM analyses.   . X-ray Raman spectra collected at the carbon K-edge on crystalline siderite and high P -quenched FeCO 3 glasses at ambient conditions. The disappearance of the π * feature, which is solely related to the three-fold coordinated carbon, is a spectroscopic evidence of a full four-fold coordination state in the glassy structure of FeCO 3 . 11.6/ -not recovered, reaction confirmed by XRD (Fig.4 Figure 7. Density evolution of glassy, molten and cristalline siderite with pressure. Molten low P siderite (plain curve), high P data on glass (black points) and the highest P melt (red point), crystalline equation of state (dashed curve) includes the transition from high spin siderite I to low spin siderite II at 50 GPa (Liu et al., 2015).