Reversal of carbonate-silicate cation exchange in cold slabs in Earth’s lower mantle

The stable forms of carbon in Earth’s deep interior control storage and fluxes of carbon through the planet over geologic time, impacting the surface climate as well as carrying records of geologic processes in the form of diamond inclusions. However, current estimates of the distribution of carbon in Earth’s mantle are uncertain, due in part to limited understanding of the fate of carbonates through subduction, the main mechanism that transports carbon from Earth’s surface to its interior. Oxidized carbon carried by subduction has been found to reside in MgCO3 throughout much of the mantle. Experiments in this study demonstrate that at deep mantle conditions MgCO3 reacts with silicates to form CaCO3. In combination with previous work indicating that CaCO3 is more stable than MgCO3 under reducing conditions of Earth’s lowermost mantle, these observations allow us to predict that the signature of surface carbon reaching Earth’s lowermost mantle may include CaCO3.

C arbon is not the only key to life and Earth's habitability but also traces and modifies geological processes of subduction, partial melting, degassing, and metasomatism, providing valuable insights into Earth's evolution 1 . Over the history of the planet, carbon transport between surface and deep reservoirs has impacted the atmospheric, oceanic, and crustal CO 2 budgets in tandem with the composition and redox state of the Earth's mantle 2,3 . Carbon is transported from Earth's surface to its interior mainly as carbonate minerals in subduction zones and is returned in carbon-bearing gas/fluid through volcanic degassing 2,3 . These processes leave signatures in the mantle, including depletion of incompatible elements 4,5 , diamond formation (and inclusions) 6,7 , and isotopic abundances 8,9 . Carbon flux via subduction to the deep mantle remains uncertain, with estimated magnitudes ranging from 0.0001 to 52 megatons/ year 3,10 . The wide range of these estimates is due in part to limited understanding of the physical and chemical responses of carbonates to mantle pressures, temperatures, and compositional environments.
The dominant carbonates carried into the mantle by subducting slabs, dolomite CaMg(CO 3 ) 2 , magnesite MgCO 3, and calcite CaCO 3 11 , undergo changes in crystal structure or state and chemical reactions at depth. Carbonates are likely to be retained as solid minerals in subducting ocean crust until/unless the solidus of carbonated peridotite 12,13 or eclogite 14,15 intersects with mantle geotherms, initiating melting. These slab-derived carbonatite melts will segregate to the overlying mantle due to low viscosity and density 16 , or be reduced to diamonds at depths greater than~250 km via redox freezing 7,15,17 . However, carbonates are present in the mantle transition zone and possibly lower-mantle depths in some regions, based on direct evidence provided by carbonate minerals found in deep-sourced diamond inclusions 18,19 . Additional evidence from thermodynamic modeling of devolatilization of carbonate-bearing subducting slab 20,21 and melting experiments on carbonates in the MgCO 3 -CaCO 3 system up to 80 GPa 22 supports the preservation of solid carbonates along lowtemperature geotherms in subducting slabs in the lower mantle. However, the temperature is not the only control on the fate of subducted carbonates: carbonates may also interact chemically with the major phases of the ambient mantle or basalt-rich subducted crust. In these compositions in the lower mantle, the silicates potentially reacting with carbonates are bridgmanite (bdg), post-perovskite (pPv), and Ca-perovskite (Ca-Pv).
The presence of the end-member carbonates, MgCO 3 and CaCO 3 (note that (Mg,Ca)(CO 3 ) 2 dolomite breaks down to these end-members above 5 GPa and 1200 K 23 ), together with lowermantle silicates depends on the thermodynamics and kinetics of the carbonate-silicate exchange reaction: Previous experiments 24,25 indicate that CaCO 3 reacts with silicates to form MgCO 3 via the forward reaction up to 80 GPa and 2300 K, i.e., at least to the mid-lower mantle. Theoretical studies further predict that MgCO 3 + CaSiO 3 are enthalpically favored over CaCO 3 + MgSiO 3 throughout the lower mantle pressure and temperature regime [26][27][28][29][30] . However, although many studies have addressed the stability of individual carbonates up to higher pressures [31][32][33] , no experiments examined the carbonate-silicate cation exchange reaction up to core-mantle boundary conditions.
In this work, to assess the stability of MgCO 3 and CaCO 3 coexisting with lower-mantle silicates, we conduct a series of experiments on the carbonate-silicate reaction along the lowermantle geotherm. Thin disks of carbonates and silicates were loaded together in laser-heated diamond-anvil cells (LHDAC, Supplementary Table 1, see "Methods" for details). Laser heating at 1600-2800 K and 33-137 GPa was applied for 10-400 min. Run products were examined by in situ synchrotron X-ray diffraction (XRD) and ex situ energy-dispersive X-ray spectroscopy (EDX) analysis with a scanning transmission electron microscope (STEM, see "Methods" for details).

Results
Calcium carbonate reaction to form magnesium carbonate. Experiments assessed thermodynamic stability by using as reactants either (Mg,Ca)CO 3 + (Mg,Fe)SiO 3 (reactants for the forward reaction, hereafter referred to CaC-to-MgC) and (Mg,Fe) CO 3 + CaSiO 3 (reactants for the reverse reaction, hereafter referred to MgC-to-CaC). For reaction CaC-to-MgC, the criterion for determining whether the reaction takes place is the presence of newly synthesized CaSiO 3 -perovskite in the run product. For reaction MgC-to-CaC, newly synthesized MgSiO 3 and CaCO 3 indicate the reaction is favorable. The silicate reaction products are easier to observe through diffraction than carbonates due to higher diffraction intensity.
Experiments with CaC-to-MgC reactants indicate the forward reaction takes place in runs conducted below 83 GPa (runs #1-4), as determined via both EDX and XRD. For example, ex situ electron microscopic analysis of the sample recovered from 33 GPa and 1650 K (run #1) (Fig. 1a-c) reveals a~1-μm-thick layer of CaSiO 3 between the silicate layer and the carbonate layer, coexisting with SiO 2 , FeO, MgSiO 3 , and MgCO 3 . These observations are consistent with in situ XRD patterns of run products after heating (Supplementary Figs. 3b and 5a), which exhibit several new sharp peaks compared to the pattern before heating ( Supplementary  Fig. 3a). The diffraction pattern of run products is consistent with the presence of Ca-Pv, magnesite, bdg, wüstite, stishovite, and monoclinic dolomite III (previously observed at pressure above 36 GPa 34 ). Ca-Pv can be observed in the run products of CaC-to-MgC up to 83 GPa (Supplementary Figs. 3c and 4a, b), in agreement with previous experimental observations 24,25 .
At higher pressures from 91 to 137 GPa, however, we observe no evidence of carbonate-silicate exchange reaction in experiments with CaC-to-MgC reactants. Ca-Pv is not identified in the run products (runs #5-7) through either in situ (Supplementary Figs. 3d, e, 4c, d, and 5b) or ex situ analysis. New, sharp peaks from bdg and pPv can be observed in situ in XRD patterns ( Supplementary Fig. 3d, e), indicating the sample was sufficiently heated to transform starting materials to high-pressure silicate structures, but no carbonate-silicate exchange reaction occurs. Two hypotheses can explain these observations: (1) in contrast to theoretical predictions that the reversal of the carbonate-exchange reaction takes place at higher pressures and lower temperatures [26][27][28][29][30] , CaCO 3 + MgSiO 3 become more favorable than MgCO 3 + CaSiO 3 from 91 to 137 GPa and 2100 to 2800 K; (2) the reaction CaC-to-MgC is hindered by reaction kinetics, and metastable starting materials are observed.
Magnesium carbonate reaction to form calcium carbonate. In order to resolve the thermodynamically stable phase assemblage, three separate sets of experiments on the backward reaction (MgC-to-CaC, runs #8-11) were conducted at 35-133 GPa and 1800-2000 K. Elemental mapping of the run products of experiments at 88 GPa (run #10, Fig. 1d-f) and 133 GPa (run #11, Fig. 1g-i) indicates that MgSiO 3 layers formed along with the carbonate-silicate interface, and newly formed CaCO 3 can be observed as well. At 35 GPa, neither EDX nor XRD shows MgSiO 3 formed from MgC-to-CaC reactants (run #8, Supplementary Fig. 9). Observations of the reversal of the reaction confirm that MgCO 3 is unstable and reacts with CaSiO 3 producing CaCO 3 and MgSiO 3 at pressures higher than 88 GPa along a lower-mantle geotherm.
Our results agree with previous experimental constraints ( Fig. 2) below 80 GPa showing: dolomite is unstable relative to CaCO 3 and MgCO 3 at lower-mantle conditions [23][24][25]35 ; neither CaO nor MgO are observed in run products, indicating no decomposition of CaCO 3 and MgCO 3 into oxides plus CO 2 26,27,30 ; MgCO 3 is more favorable in the lower mantle than CaCO 3 up to~80 GPa due to the CaC-to-MgC reaction 24,25 . Since similar previous studies were limited to pressures below 80 GPa, they did not observe the reversal reaction (MgC-to-CaC). Combining our new results with previous results 24, 25 and theoretical predictions indicating a positive Clapeyron slope for this reaction 28-30 , we suggest a reaction boundary above 80 GPa with a positive slope (black dashed line in Fig. 2). We note that the experimental data allow for significant uncertainty in this boundary, but are inconsistent with theoretical predictions [28][29][30] (yellow region, Fig. 2). This discrepancy may have been produced by theoretical approximations at higher temperatures. If density functional perturbation theory and quasi-harmonic approximation have misestimated the volumes of the carbonate phases expected to be stable at~80 GPa and higher pressures, this could lead to the systematic overestimation of Gibbs free energy of CaCO 3 + MgSiO 3 relative to MgCO 3 + CaSiO 3 at higher temperatures.

Discussion
The pressure/temperature conditions of the reversal reaction as constrained by these experiments are similar to those of 900 nm 900 nm polymorphic phase transitions associated with sp 2 -sp 3 bonding changes in both MgCO 3 and CaCO 3 , which suggests these transitions are related to the stabilization of a CaCO 3 + MgSiO 3 assemblage. The transition from sp 2 -to sp 3 bonds in MgCO 3 has been identified at~80 GPa with the stabilization of the C2/m structure 32,33,36 , and the resulting densification of MgCO 3 supports the forward reaction to MgCO 3 + CaSiO 3 . The transition in CaCO 3 from sp 2 -to sp 3 bonds in the P2 1 /c-h structure was experimentally observed at~105 GPa and 2000 K 37 . Computational studies predicted this boundary at~70 29 and~100 GPa 38 at mantle-relevant temperatures (red-shaded region in Fig. 3). While an earlier study that did not include the sp 3 CaCO 3 -P2 1 /c-h structure predicted a crossover in silicate-carbonate-exchange reaction at 135 GPa and 0 K 26 , a later study that predicted the sp 3 CaCO 3 -P2 1 /c-h structure found a silicate-carbonate reaction reversal at 84 GPa and 0 K 30 . This would correspond to sp 2 -sp 3 crossover and stabilization of CaCO 3 + MgSiO 3 in the mid-lower mantle.
Whether a crossover in the carbonate-silicate exchange reaction takes place in the deep Earth depends on whether carbonates are preserved in Earth's lower mantle to at least 1800-km depth. Previous studies have identified barriers to carbon subduction and stability in the lower mantle, particularly melting 15,39 and reduction [40][41][42] . If carried in cold subducting slabs, MgCO 3 and CaCO 3 may avoid melting as their melting temperatures 22 are higher than some predicted cold slab geotherms 36 . Any solid carbonate in the mantle will be in contact and may equilibrate with silicates in all mantle environments and with free silica in basalt-rich compositions. MgCO 3 and CaCO 3 have been observed in experiments [40][41][42] to undergo decarbonation reactions with free silica over a pressure range of~40 to 60 GPa. However, the Clapeyron slope of CaCO 3 + SiO 2 → CaSiO 3 + CO 2 is positive and takes place at pressure/temperature conditions warmer than the coolest slab geotherms 40 . Observations that MgCO 3 is less thermally stable than CaCO 3 support the survival of CaCO 3 rather than MgCO 3 along a cold subducted slab geotherm to the lowermost mantle 36,41 (Fig. 3). In this study, we report a reversal in the Mg-Ca silicate-carbonate cation exchange reaction at~90 GPa, making MgCO 3 + CaSiO 3 favorable in the upper part of the lower mantle, while CaCO 3 + MgSiO 3 is preferred in the lower part of the lower mantle (Fig. 3). However, the question of whether any carbonate persists to these depths in the coldest subducting slabs remains unresolved. If MgCO 3 remains present in cold slabs, and the reaction CaC-to-MgC proceeds throughout most of the mantle eliminating CaCO 3 , the reversal MgC-to-CaC reaction may transform MgCO 3 back to CaCO 3 in the lowermost mantle (Fig. 4). CaCO 3 could thus be found in the lowermost mantle coexisting with silicates and reduced iron.
The reduced nature of the Earth's mantle, with oxygen fugacity inferred to be near the iron-wüstite buffer in the transition zone and greater depths 43 , stabilizes diamond or Fe carbide as longterm hosts of carbon, owing to their chemical refractoriness and dynamic immobility 44 . Similarly, our experimental observations support CaCO 3 as a refractory, stable host for oxidized carbon in the middle to the lowermost mantle, in particular, the highpressure polymorph of CaCO 3 (CaCO 3 -P2 1 /c-h) with tetrahedral bonds 37 . Experimental observations also suggest that CaCO 3 is more resistant to redox breakdown reaction with iron under reduced conditions than MgCO 3 35 . In addition, due to the cation    Thomson et al. 22 . Four typical mantle geotherms are modified from Maeda et al. 36 . The red-shaded region indicates the transition boundary of CaCO 3 from sp 2 to sp 3 structure predicted by density functional theory computations 29,38 . exchange between carbonate and silicate, the relative stability of MgCO 3 or CaCO 3 will change in the lowermost mantle, and depending on conditions one of these phases may buffer the redox state of the mantle through an influx of oxidized carbon in the form of solid carbonate 45 .
The Mg-Ca silicate-carbonate-exchange reactions along with subduction pressure-temperature (P-T) conditions may impact observable signatures of Mg and Ca isotopes in mantle silicates under certain special conditions, or in carbonate inclusions in diamonds. Subducting carbonates carry low-δ 44/40 Ca and lowδ 26 Mg signatures relative to the heavier mantle ratios, but although previous studies have observed heterogeneity in the Ca and Mg isotope signatures in basalts and mantle peridotites, these studies determined that lighter ratios cannot be simply interpreted as evidence of recycled marine carbonates 46,47 . The Mg-Ca silicate-carbonate-exchange reactions along with subduction P-T conditions may contribute to these variable Mg and Ca isotopic compositions. The reaction CaC-to-MgC in the transition zone and upper part of the lower mantle would transfer light Ca isotopes from subducted CaCO 3 to CaSiO 3 (Ca-Pv) (Supplementary Note 1 and Supplementary Fig. 12). Isotopically light Ca-Pv can then be trapped in diamond inclusions and return to the surface 48 , while the Ca isotopic signature of upwelling rocks would remain variable, as it undergoes continuous fractionation within peridotitic mantle lithologies 46,[49][50][51] . The modification of carbonate-silicate phase equilibria observed in this study provides a new process that could alter Mg and Ca isotopic composition in such lithologies (Supplementary Note 1 and Supplementary Fig. 12). While the isotope signature of MgSiO 3 produced by reaction MgC-to-CaC would not be observable due to the small masses involved relative to the vast lower-mantle reservoir of MgSiO 3 , any CaCO 3 produced in the deep lower mantle by this reaction would carry a heavier deep mantle δ 44/ 40 Ca signature that would distinguish it from surface-derived carbonate. If preserved in diamond inclusions and returned to the surface, heavy CaCO 3 could be used to trace the presence of oxidized carbon in the lowermost mantle. The potential of CaCO 3 to be a signature of an ultradeep carbon cycle reaching the core-mantle-boundary region may help to reveal other mysteries of the deep mantle, such as heat budget related to radioactive elements stored in Ca-bearing silicates 52 , and compositions of heterogeneities that may record Earth's early history 48,53 .

Methods
Starting materials. To investigate phase equilibria in the carbonate-silicate system in Earth's lower mantle and control for effects of reaction kinetics, both CaC-to-MgC and MgC-to-CaC experiments were carried out in symmetric diamond-anvil cells (DAC) with flat-top double-sided laser heating 54 . For CaC-to-MgC, natural dolomite with homogeneous composition of (Mg 0.38 Ca 0.59 Fe 0.03 )CO 3 was used as a carbonate reactant, the composition, and structure of which has been characterized by X-ray fluorescence spectroscopy and X-ray diffraction, respectively 35 . Febearing enstatite synthesized at École Polytechnique Fédérale de Lausanne with a composition of (Mg 0.5 Fe 0.5 )SiO 3 was used as a silicate reactant 55 . For MgC-to-CaC, natural ferromagnesite (sample from Princeton University) was used as a carbonate reactant, with composition determined to be (Mg 0.87 Fe 0.13 )CO 3 by wavelength dispersive X-ray spectroscopy in a Cameca SX100 Electron Probe Microanalyzer at the University of Michigan. Pure calcium silicate (CaSiO 3 , Alfa Aesar) was used as a silicate reactant. The chief advantages to the abovementioned starting compositions are that recognition of a carbonate-silicate exchange reaction only requires identification of the presence of newly synthesized silicates in quenched run products, i.e., Ca-perovskite (Ca-Pv) in CaC-to-MgC and bridgmanite (bdg) in MgCto-CaC; and Fe-bearing enstatite and ferromagnesite can serve as laser absorber during the forward CaC-to-MgC and reversal MgC-to-CaC experiments, respectively.
LHDAC experiments. The dolomite, enstatite, and calcium silicate samples were separately ground under acetone in an agate mortar for~2 h each to achieve homogenous, finely powdered samples with grain size typically less than~2 µm. A single ferromagnesite crystal was double-side polished to~10-micron thickness. All starting materials were dried in an oven at 120°C overnight before loading, and the powder samples were subsequently pressed in a DAC to form thin foils approxi-mately~8-10-µm thick. The enstatite foils and ferromagnesite crystals were sandwiched between iron-free dolomite and calcium silicate, respectively, serving as thermal insulators in symmetric DACs for CaC-to-MgC and MgC-to-CaC (Supplementary Figs. 1 and 2). No other pressure standard or medium was loaded to prevent reactions with other components and contamination of the chemical system. The sample sandwiches were loaded in sample chambers with diameters approximately halves of the anvil culet sizes drilled into Re gaskets pre-indented to a thickness of ∼30 μm, by using the laser drilling system at HPCAT (Sector 16) of the Advanced Photon Source (APS), Argonne National Laboratory (ANL) 56 . Diamond anvils with flat culets of 300 μm were used for experiments under 60 GPa, beveled culets of 150/300 μm for experiments under 100 GPa, and beveled culets of 75/300 μm for experiments up to 140 GPa.
Before laser heating, each sample was compressed to the target pressure at 300 K, and after heating each sample was quenched to ambient pressure at 300 K to limit and preserve reactions at target conditions. Pressures were determined from the Raman shift of the singlet peak of the diamond anvil at the culet surface 57 , and post-heating pressures were typically within 3% of the pre-heating pressure. Thermal pressure during heating may be estimated to be~10% GPa higher than the pre-heating pressure at the modest temperatures 58,59 . High-temperature conditions were achieved by using a double-sided ytterbium fiber laser heating system at beamline 13-ID-D (GeoSoilEnviroCars) of APS, ANL 54 , with two 1.064 μm laser beams focused down to a flat-top spot with a diameter of 10-12 μm on both sides of the sample. Temperatures of the heated samples were determined by fitting the measured thermal radiation spectra using the Planck radiation function under the graybody approximation 54 . The temperature reported in Supplementary  Table 1 is the temporal average of multiple temperature measurements over the heating duration. Temperature fluctuations over this timescale were less than the specified uncertainty, which is derived from a standard deviation of temperature measurements from both sides of the laser-heated sample (typically ± 100 K below 2000 K and ±150 K above 2000 K) ( Supplementary Figs. 10 and 11). Experiments were held at temperatures between 1600 and 2800 K for~30 min in CaC-to-MgC experiments and up to 400 min in MgC-to-CaC experiments.
In situ XRD. Phases synthesized at high P/T and achievement of chemical steadystate were determined by in situ angle-dispersive X-ray diffraction (XRD) measurements performed before, during, and after heating at beamline 13-ID-D (GeoSoilEnviroCars) of APS, ANL. The incident X-ray beam was focused down to less than 3 × 4 μm 2 with a monochromatic wavelength λ = 0.3344 Å. Diffracted X-rays were recorded using a MAR 165 detector or Pilatus 1 M CdTe pixel array detector. NIST standard LaB 6 was used to calibrate the detector distance, tilt angle, and rotation angle of the image plane relative to the incident X-ray beam. Exposure times were typically 30 s. The XRD patterns were integrated to produce 2θ plots using the software DIOPTAS 60 . MgSiO + CaCO 3 3 S u b d u c ti n g c a rb o n Fig. 4 Schematic illustration of the fate of carbonates in the oceanic crust (dark blue) subducted to the lower mantle. Through subduction, the carbonates may undergo melting (red arrow), redox freezing with metallic iron (purple arrow), decarbonation reaction with free silica (blue arrow), and exchange reaction with lower-mantle silicates (green arrow). Based on the observation of reversal of the carbonate-silicate cation exchange reaction at conditions relevant to cold subducted slabs at mid-lower-mantle depths, CaCO 3 is the potential stable phase that hosts oxidized carbon in the lowermost mantle.
Ex situ EDX. After complete pressure release, each sample was recovered from the LHDAC, and then sectioned along the compression axis through the laser-heated spot and over the entire thickness of the DAC sample (~5-20 μm), using a focused ion beam (FIB) coupled with a field-emission scanning electron microscope (FE-SEM) at IPGP (Paris, France) or the Michigan Center for Materials Characterization at the University of Michigan (Ann Arbor, USA). A~30-nm-thick Au layer was coated on each sample to reduce charging in the scanning electron microscope, and a 2-μm-thick Pt layer was deposited across the center of each heated spot to protect the sample from damage by the Ga + ion beam. Thin sections of each heated spot were extracted and polished to electron transparency (∼100-nm thickness). Textural and chemical characterization of recovered samples was performed with scanning transmission electron microscopy (STEM) and energy-dispersive Xray spectroscopy (EDX) in a JEOL 2200FS field-emission TEM (Center for Advanced Microscopy, MSU), operated at 200 kV to image the sample in brightfield. EDX maps were scanned over 512 × 384 pixel areas with a pixel dwell time of 50 μs. Typical count rates were~2000 counts per second. Chemical mapping rather than point measurement approach prevents migration of elements due to damage by the electron beam. Uncertainties in compositions were determined from standard deviations of EDX measurements obtained from selected regions within multiple grains.

Data availability
Additional diffraction and spectroscopy data and metadata are available in the Supplementary Information and from the corresponding authors upon request.