Temperature-induced amorphization in CaCO3 at high pressure and implications for recycled CaCO3 in subduction zones

Calcium carbonate (CaCO3) significantly affects the properties of upper mantle and plays a key role in deep carbon recycling. However, its phase relations above 3 GPa and 1000 K are controversial. Here we report a reversible temperature-induced aragonite-amorphization transition in CaCO3 at 3.9–7.5 GPa and temperature above 1000 K. Amorphous CaCO3 shares a similar structure as liquid CaCO3 but with much larger C-O and Ca-Ca bond lengths, indicating a lower density and a mechanism of lattice collapse for the temperature-induced amorphous phase. The less dense amorphous phase compared with the liquid provides an explanation for the observed CaCO3 melting curve overturn at about 6 GPa. Amorphous CaCO3 is stable at subduction zone conditions and could aid the recycling of carbon to the surface.

L arge amounts of carbon could be introduced into the deep mantle in the form of carbonates (e.g. CaCO 3 , MgCO 3 , and CaMg(CO 3 ) 2 ) by subduction of oceanic crust, constituting an important part of the global carbon cycling. The subducted carbonates pose enormous impacts on the Earth's mantle. Carbonates could drastically reduce melting temperature of peridotite and eclogite [1][2][3] and are important metasomatic agents with a remarkable wetting capability to impregnate silicate minerals [4][5][6] . Carbonate melt is an ionic liquid with ultra-low viscosity 7 and is considered responsible for the conductivity anomalies in the oceanic mantle 3,[8][9][10] . In particular, CaCO 3 acts as a carrier to transport carbon into deep Earth from surface and transfer it back via volcano eruption 11,12 . CaCO 3 , together with CaMg (CO 3 ) 2 , is also a possible calcium source for the formation of perovskite-structured CaSiO 3 observed in deep diamond inclusions, providing insights into the recycling of oceanic crust in the deep mantle 13 . Although CaCO 3 was reported to react with enstatite at upper mantle conditions to form dolomite 14 , diamond inclusions and exhumed ancient subduction-zone rocks evidently show that CaCO 3 can survive to depths of at least the topmost lower mantle 12,15,16 .
Extensive studies on phase transitions of CaCO 3 at high pressure and temperature have been conducted [17][18][19][20][21][22] . Rhombohedral calcite (calcite-I) is stable at ambient conditions and transforms to a monoclinic (P2 1 /c) structure at about 1.5 GPa upon room temperature compression. It further transforms to triclinic calcite-III at about 2 GPa 23 . Upon heating, calcite-III transforms to aragonite 20,24 . However, the phase diagram of CaCO 3 above 3 GPa and 1000 K is still controversial. A disordered calcite crystal phase (resembling calcite-IV or -V) was reported according to energy-dispersive X-ray diffraction (EDXRD) patterns 20 , while Litasov et al. 25 suggested that it might be a new phase. Both studies did not provide any detailed structure information of the high-PT phase. Coincidently, the P-T conditions of the unsolved phase are in accord with the dissolution conditions of CaCO 3 from subducted slabs 11 . However, the large-scale dissolution of CaCO 3 during subduction remains unclear 11,12 . Therefore, exploring properties of the unsolved phase is essential to unveil this mechanism.
This study aims to clarify the nature of the transition from aragonite to the disordered crystal phase or new phase mentioned in Suito et al. 20 and Litasov et al. 25 , respectively, and provide its structure information by in situ measurements up to 7.5 GPa and 1723 K using Paris-Edinburgh press techniques coupled with EDXRD. We report a reversible temperatureinduced amorphization in aragonite. The structure measurements of amorphous CaCO 3 indicate that the amorphous phase is one of the lightest materials in subducting slabs. Correlations between properties of the amorphous phase and the mechanism of largescale CaCO 3 dissolution in subduction zone are subsequently presented.

Results
Phase transitions in CaCO 3 . Upon compression of CaCO 3 in a standard assembly cell for large-volume Paris-Edinburgh press ( Fig. 1) at room temperature, we observed the transition from the rhombohedral calcite (calcite-I) to the monoclinic phase (calcite-II) between 1.4 and 1.9 GPa. The calcite-III phase appeared at 2.5 GPa (Fig. 2). The observed transitions are consistent with prior determinations of the phase boundaries 17,18,20 . Upon heating at 4.8 GPa, we confirmed the transition to aragonite at 473 K, whose XRD pattern can be indexed with an orthorhombic structure (Fig. 3). The transition temperature is in a good agreement with previous observations 20, 24 .
Upon further heating at 4.8 GPa, we observed the aragonite phase stable up to 1073 K (Fig. 3). During the heating cycle, the recorded diffraction peaks showed some variation in peak intensity due to aragonite grain growth. At 1273 K, the majority of sharp peaks disappeared, and two broad bands emerged at dspacing values of~1.0 and~1.3 Å. At higher temperature, only three broad peaks were observed up to 1673 K (Fig. 3). The broad peaks over this temperature range cannot be caused by melting because the temperatures are too low to initiate melting. We also confirmed melting that occurred at temperatures >1873 K by monitoring the movement of Pt spheres in the molten sample. Below the melting temperature, no movement of Pt spheres was observed (Fig. 4). It had been reported that the incorporation of water could significantly reduce the melting temperature of CaCO 3 to less than 1127 K 26,27 . However, water (or water source, e.g., Ca(OH) 2 ) should be sealed in Pt (or Au) capsule in largevolume press experiments in order to maintain the equilibrium  28 and Pt sphere is expected to drop. On the contrary, we observed the movement of the Pt sphere only when temperature reaches melting >1873 K. The likely explanation for the observed diffraction features is due to a transition to an amorphous phase. Amorphous CaCO 3 could be further confirmed by multiangle EDXRD at 2θ = 3°, 4°, 5°, 7.4°, 9°, 12°, 16°, and 22° (Fig. 5a). No crystal peak of the sample was observed at 5.2 GPa and 1773 K in all the EDXRDs. On the other hand, the intensity of the broad peaks is large enough to exclude the possibility of background. More importantly, the broad peaks shift to lower energy when the diffraction angle increases. All the features conclusively verify that CaCO 3 becomes an amorphous phase, ruling out the causes by melting, significant grain growth, or single crystal. We subsequently obtained EDXRD patterns during the cooling cycle. The three broad peaks persist to temperatures as low as 1173 K. When the temperature decreased to 1073 K, the sample immediately transformed back to aragonite characterized by its sharp diffraction peaks (Fig. 3). We also observed similar change of the patterns during the heating and cooling cycles in different experiments. These results indicate that the transition between aragonite and the amorphous phase is reversible.
A disordered phase of CaCO 3 at 6.16 GPa and 1473 K was reported by Suito et al. 20 . The disordered phase resembles calcite-IV or calcite-V with space groups of R 3c and R 3m, respectively 29 , based on 3 diffraction peaks (102), (104), and (110). The diffraction patterns we collected at temperatures >1473 K and 5 GPa are consistent with features of an amorphous phase, and they are distinctively different from those in Suito et al. 20 . We also observed some sharp peaks along with the emerging of broad peaks in the temperature range of 1073-1273 K. During ascending temperature, some sharp peaks occasionally persist to temperatures as high as 1673 K and randomly disappeared at a temperature between 1273 and 1673 K in different heating cycles (Fig. 2). Therefore, we cannot rule out the existence of a disordered phase before final amorphization. We note that a weak broad peak at~56 keV (d =~2.1 Å) could be observed at 1473 K in Fig. 4 of Suito et al. 20 . The weak feature could be due to relatively short collection time. The d-value of the weak peak is consistent with the third broad peak in Fig. 3, indicating that amorphous CaCO 3 was also presented in the study of Suito et al. 20 . The other two broad peaks shown in Fig. 3 could not be identified in Suito et al. 20 , which might be covered by strong graphite peaks at 38 keV. 5  We conducted a total of seven independent experiments and repeated the experiments to cross the crystal-amorphous phase boundary and characterized this boundary by the diffraction data of the high-PT phase. As discussed above, there is some uncertainty regarding when the diffraction peaks of the crystalline phase completely disappear during the heating cycle. However, during descending of the temperature, we consistently observed amorphous CaCO 3 without crystalline diffraction peaks until it transformed to aragonite at a definitive temperature. We determine the phase boundary in Fig. 2 via EDXRD data at descending temperature. The boundary determined in this study is about 200 ± 100 K lower than that reported by Suito et al. 20 and Litasov et al. 25 . Our boundary may be considered as a low bound for the transition.
Structure information of the amorphous phase and liquid CaCO 3 . Broad diffraction peaks, observed at temperatures between 1473 and 1673 K (Fig. 3), are consistent with diffraction signal from an amorphous phase. In order to confirm the amorphization of CaCO 3 at high temperature and to understand the differences between amorphous and liquid CaCO 3 , we conducted multiangle EDXRD measurements at diffraction angles of 3-22°up to 1923 K at 3.9-7.5 GPa. Figure 5 presents the typical multiangle EDXRDs of amorphous and liquid CaCO 3 . One or two broad peaks were observed in the multiangle EDXRD at each 2θ (Fig. 5). As the 2θ increases, the broad peaks shifted to lower energy and new broad peak may appear at higher energy. From the multiangle EDXRD, we can get the structure factor [S(Q)] and real-space pair distribution function [G(r)] 28 . Figure 6a shows Fourier transformation of S(Q) yielded the real-space pair distribution function, G(r) (Fig. 6b) 31 . The G(r) patterns of liquid CaCO 3 are similar to those reported in previous studies 7, 32 . The amorphous CaCO 3 has a similar G(r) pattern to that of the liquid CaCO 3 . However, differences can be observed in the bond lengths of C-O, Ca-O, and Ca-Ca, represented by the first, second, and third peaks, respectively ( Table 1). The bond lengths of amorphous and liquid CaCO 3 were calculated from peaks in G(r) patterns using Gaussian fitting and those of aragonite were estimated according to its thermal equation of state and crystal structure 25,33 . The C-O bond length of the amorphous phase is almost same as aragonite, while it is much larger than that of the liquid phase. The Ca-O bond length of the amorphous phase is identical to those of liquid CaCO 3 and aragonite. It is difficult to get the Ca-Ca bond length because of the asymmetry of the Ca-Ca peak. However, we can qualitatively compare the differences by the topmost position (TP) of asymmetric Ca-Ca peaks. The Ca-Ca TPs of liquid and aragonite are~6.3% and 7.9% smaller than that of the amorphous phase, respectively. These data indicate that the amorphous phase at 1573 and 1773 K has basically the same structure as liquid but with different C-O and Ca-Ca bond lengths.
Although it is difficult to obtain an accurate density from the diffraction data of liquid or amorphous phase alone, the measurements provide qualitative comparison of densities for different phases because the density is proportional to the FSDP position 34 . The center of the FSDP of amorphous CaCO 3 is~2.00 Å −1 , while that of the liquid phase is~2.15 Å −1 . The former is about 7% smaller, which is significantly larger than the uncertainty (<1%) 35 in determining the center of the FSDP. Our diffraction data indicate that the density of the amorphous CaCO 3 is smaller than that of the liquid phase based on the extracted structure factor of the amorphous and liquid phases. This can be further confirmed by the comparisons of the bond lengths in Table 1. Zhang and Liu 36 theoretically calculated the densities of aragonite and liquid CaCO 3 and found that the latter is~13% smaller. In Table 1 Mechanism. Pressure-induced amorphization is fundamentally interesting in physics, chemistry, material, geoscience, and industrial applications 38,39 . On the other hand, temperature is mainly treated as a crystallization stabilizer. Only hexagonal ice, SiO 2 -stishovite, and zeolites were reported to transform to an amorphous phase upon heating [40][41][42] . The temperature-induced amorphization (TIA) in CaCO 3 is more intriguing because (1) amorphous CaCO 3 is a stable phase and does not recrystallize before melting; and (2) the aragonite-amorphous phase transition is reversible. We observed direct melting from the amorphous phase without recrystallization in CaCO 3 , which behaves totally different from the crystal-amorphous phase-recrystallization in the meta-stable stishovite at ambient pressure. We also noticed that the amorphous CaCO 3 with a relatively larger volume (the sample size is) could be stabilized at high pressure and high temperature compared with the metastable small amorphous CaCO 3 particle (~1 μm) intergrown with aragonite at ambient conditions 43,44 , indicating that amorphous CaCO 3 would exist in volumetrically large size in subduction zone. The reported mechanisms for TIA include chemical disorder, thermodynamic melting followed by immediate vitrification, and mechanical collapse 42,45 . Chemical disorder is considered in zeolite where Al substitutes Si in TO 4 framework 45 , whereas it can be excluded in the case of amorphous CaCO 3 because there is no another element to substitute carbon in the framework. Thermodynamic melting followed by immediate vitrification was favored in amorphization of hexagonal ice 42,45 . Taking into account the melting temperature determined by Suito et al. 20 and Li et al. 37 , the observed amorphization of CaCO 3 occurs at much too low temperature (~800 K lower than the melting temperature) to be explained by vitrification. Consequently, mechanical collapse is the most plausible mechanism for the TIA in CaCO 3 . The crystal structure might be destroyed by the high mobility of the Ca 2+ cation and the CO 3 2− anion in the amorphous phase. This is consistent with the moderate decrease of the measured resistance across the aragonite to amorphous phase transition below the melting temperature 37 . The increase of the Ca-Ca bond length of the amorphous phase compared to that of aragonite also supports the increased mobility of Ca 2+ cation that could eventually lead to the destruction of the Ca 2+ lattices to form the amorphous phase.

Discussion
The observed amorphization in CaCO 3 occurs at a relatively low temperature, comparable to the conditions relevant to subduction zone environments. At depth of 30-50 km, calcite will transform to aragonite and further to amorphous phase around 70-120 km (Fig. 7). The bond length comparison and the negative slope of the melting curve support that amorphous CaCO 3 is less dense than the liquid counterpart. Most recently, Hudspeth et al. 32 obtained the densities of liquid CaCO 3 based on structure measurements, which range from 2.4 to 2.6 g cm −3 in the pressure range of 4.7-8.7 GPa at~2000 K. In contrast, the densities of aragonite are 2.9-3.1 g cm −3 at these P-T conditions according to Litasov et al. 25 . Liquid CaCO 3 is~16% lighter than aragonite. The density reduction of liquid CaCO 3 compared with aragonite is also supported by first principle calculations. Zhang and Liu 36 calculated the densities of aragonite and liquid CaCO 3 and found the latter is~13% smaller. Since the amorphous CaCO 3 is less dense than the liquid CaCO 3 , it would be at least 16% smaller than that of aragonite. The density profile of aragonite 25   The phase boundary of aragonite and amorphous CaCO 3 intersects the melting curve at~13 GPa, which corresponds to a depth of 400 km. In the depth between 70 and 400 km of subducted slabs, the most plausible form of CaCO 3 is amorphous phase according to the phase diagram in Fig. 2. The density difference between the amorphous CaCO 3 and other subduction zone assemblages indicates that the amorphous CaCO 3 would possess an ultra-high buoyancy. On the other hand, amorphous CaCO 3 is a weak solid with shear modulus of 14 GPa at ambient conditions 49 , which is only~17% of olivine 50 and~41% of aragonite 51 . The ultra-high buoyant and soft amorphous CaCO 3 has important implications for carbon recycling in subduction zones. Indeed, the majority of CaCO 3 would likely escape the downgoing slabs and ascend to magmas in the overlying mantle through dissolution 11 , rather than subducting to the deep mantle (Fig. 7). Decarbonization in fluids from the shallow subducting slabs only accounts for a portion of CO 2 released in volcanic gas emissions [52][53][54] . CaCO 3 has been considered to fill the gap between the CO 2 outputs via volcanism and inputs in subducting slabs 12 . However, the mechanism for a large-scale CaCO 3 dissolution from the slab is unknown 11 . We propose a novel regime of carbon input where CaCO 3 readily concentrates through percolation of the amorphous phase and migrates from the subducting slabs to the mantle wedge. The migration of CaCO 3 in the mantle wedge could be complex depending on the thermal structure of the slab and mantle wedge. Because the mantle wedge is hotter than the subducted slabs, the percolated amorphous CaCO 3 from the subducted slabs would be further stabilized and migrate faster when CaCO 3 enters the mantle wedge (Fig. 7). On the other hand, the higher temperature in the mantle wedge would dilate the amorphous CaCO 3 , so as to increase its buoyancy. In the shallow part of the mantle wedge, the temperature decreases as depth becomes smaller. The buoyancy of the amorphous CaCO 3 would be reduced and the migration of the amorphous CaCO 3 would be slower. As the amorphous CaCO 3 migrates to a depth where the temperature is lower than the phase boundary between aragonite and the amorphous phase, it would transform back to aragonite. Assimilation of the amorphous CaCO 3 into arc magmas could serve as an important CO 2 source for volcanic eruptions. Additionally, this mechanism may lead to the formation of calcite-rich carbonatites 12,55 .

Methods
Experimental set-up. The EDXRD measurements on CaCO 3 at high pressure and temperature were conducted using a large-volume Paris-Edinburgh press at 16-BMB synchrotron beamline at the Advanced Photon Source, Argonne National Laboratory. Two cup-shaped tungsten carbide anvils were used to generate high pressures. The starting material, CaCO 3 (purity 99.99%, Alfa Aesar company) dried for 24 h in an oven at 110°C, was loaded into a graphite capsule (Fig. 1). A boron nitride sleeve was used to separate the graphite heater from the capsule. An MgO sleeve outside the graphite heater served as both pressure medium and pressure marker. Pressure was determined from the XRD data of MgO 56 at a fixed angle (2θ) of 15°using a Ge solid-state detector. Temperature was determined according to the calibrated temperature-power relationship 28 . The pressure and temperature uncertainties are 0.3 GPa and 100 K, respectively. Pt spheres of 80-100 μm in diameter were loaded in the sample chamber to monitor the melting of sample.
where Q ¼ 4πE 12:3984 sinðθÞ, I m 2θ; E ð Þis the observed energy EDXRD spectrum at a given 2θ, S ′ 2θ ð Þ is proportional to cos 2 ð2θÞ, I P E ð Þ is the primary white beam profile, A E ð Þ is the X-ray attenuation, C E ð Þ is any other energy-dependent term, I inc 2θ; E ð Þis incoherent scattering from the sample, f 2 2θ; E ð Þcan be calculated according to parameters in International Tables for X-ray Crystallography (Ed. Ibers and Hamilton). The detailed data analysis method can be found in Kono et al. 28 .

Data availability
The data that support the findings of this study are available from the corresponding author upon request.