Serpentine’s wet breakdown path and enhanced water flux in cold subduction zones Authors: Heehyeon Sim, Yoonah Bang, Huijeong Hwang, Nico Giordano, Dongzhou Zhang, Hyun Hwi Lee, Taeyeol Jeon, Robert Farla, Hanns-Peter Liermann, Yongjae Lee Published: 2026 ABSTRACT AbstractSerpentine stability dictates deep water cycling in subduction zones. We investigated pure serpentinites (≈ 13 wt.% H2O) using in-situ synchrotron X-ray diffraction and ultrasonic velocity measurements under two cold-slab P–T paths, varying excess water (0.9–31.5 wt.%). Here we report that under a cold core (or ultracold Moho) geotherm with >6.9 wt.% excess water, serpentinites undergo deep hydration to form the 3.65-Å phase and phase E above 9.3 GPa and 420 °C (ca. 280 km depth), increasing total water to ≈ 19.3 wt.%. With <6.9 wt.% excess water, serpentine transforms to phase E without water loss. Conversely, under a cold Moho geotherm, serpentinites dehydrate into phase E and clinoenstatite above 7.4 GPa and 585 °C (ca. 220 km depth), releasing ≈ 5.3 wt.% water. These divergent reactions highlight an underestimated water transport capacity of subducting slabs, generating lateral contrasts in hydration, density, and seismic velocity.Subject terms: Geophysics, Mineralogy, Geochemistry Introduction A large amount of water has been transported into the Earth’s interior since the initiation of modern-style subduction of oceanic plates. The amount of subducted water has recently been estimated to be ~3.0 × 109 Tg Myr−1 ( ~ 2.3 × 1011 Tg) in the form of free water (fluid or ice) or structurally bonded water (within hydrous minerals)1–3. It is the latter form of water that is carried deep into the Earth’s interior by subducting slabs, composed of sedimentary layers (~1 km), oceanic crustal rocks (~6 km), and mantle rocks (~50–100 km)4. Some (or all) of the water contained in these minerals have generally been known to be released with increasing depths, i.e., dehydrated with increasing pressure and temperature (P-T)5,6. Along warm subduction zones, as estimated for the Cascadia and Mexico subduction systems with thermal gradients above ~15 °C/km7, dehydration processes are completed at relatively shallow depths by ≈ 115 km3. In contrast, along cold and ultracold subduction zones, such as NE Japan (e.g., Honshu) and Tonga subduction systems, where thermal gradients are below 5–10 °C/km, dehydration processes are inhibited to depths of up to ≈ 240 km3,8,9. Major hydrous minerals in a subducting slab include kaolinite (≈ 14 wt.% H2O) and phengite (≈ 4 wt.% H2O) in the sedimentary layer, talc (≈ 5 wt.% H2O), lawsonite (≈ 11 wt.% H2O), zoisite (≈ 2 wt.% H2O), Mg-chloritoid (≈ 8 wt.% H2O), and amphiboles (≈ 2 wt.% H2O) in the oceanic crust, and chlorite (≈ 13 wt.% H2O), phlogopite (≈ 4 wt.% H2O), and serpentines (≈ 13 wt.% H2O) in the mantle layer10,11. The stability and thermochemical characteristics of each hydrous mineral differ and are highly sensitive to the P-T conditions. Thus, the stability fields of these minerals have been extensively investigated to understand subduction-related geochemical processes8,10,12. For example, under the warm subduction geotherm of Cascadia, dehydration of lawsonite (CaAl2Si2O7(OH)2·H2O) has been reported to occur at ≈ 60 km depth to form zoisite (4 lawsonite → 2 zoisite (Ca2Al3Si3O12(OH)) + 1 kyanite (Al2SiO5) + 1 coesite (SiO2) + 7 H2O) that releases 87.5% of the structurally bound H2O by mass13. Contrastingly, in the cold slab of Tonga, lawsonite remains stable down to ≈ 300 km depth (≈ 10 GPa), where it transfers H2O to Dense Hydrous Magnesium Silicates (DHMS)14–16. The fluids released by various hydrous minerals at different depths could be trapped as an interstitial component to be transported downward by the descending slab17 or migrate upward via percolation into the overlying layers and up to the mantle wedge of the overriding plate18. It is thus conceivable to infer that subducting minerals can react with fluids originating from the dehydration of the other subducting minerals19. Recently, hydration reactions, rather than dehydration reactions, have been reported to occur in various mineral assemblages when additional fluids are present under cold subduction conditions12,20–22. Subducting kaolinite (Al2Si2O5(OH)4), one of the representative oceanic clay sediments in the ASH (Al2O3–SiO2–H2O) system, increases its H2O transport capacity via the so-called super-hydration, i.e., a counter-intuitive mineral reaction in which a hydrated mineral uptakes more water upon P-T increase to form a further hydrated mineral; kaolinite becomes super-hydrated (Al2Si2O5(OH)4·3H2O) at a depth of about 75 km along a fluid-rich cold subduction interface, which subsequently breaks down near 200 km depth into other known minerals in the ASH system12. Furthermore, talc (Mg3Si4O10(OH)2), an important mineral in subducting lithospheres (crust and mantle), has been known to transform into the 10-Å phase (Mg3Si4O10(OH)2·nH2O) through interlayer expansion by water intercalation at ~4–7 GPa and ~450–650 °C10,23. When exposed to alkaline salty water, talc has recently been reported to transform into the super-hydrated 15-Å phase along the cold subduction geotherm at a depth of ≈ 90–95 km and then converts to the 10-Å phase through dehydration at ≈ 125 km20. The 10-Å phase then decomposes into a 3.65-Å phase (MgSi(OH)6) + stishovite (SiO2) assemblage at about 10–12 GPa and 450–550 °C24. Such (super-)hydration reactions might well be common processes in subducting materials where water can percolate through and/or accumulate. On the other hand, talc-related high-pressure minerals such as talc-like and the 10-Å phases have frequently been reported to occur during the dehydration processes of serpentines and other minerals of the MSH (MgO–SiO2–H2O) system25–28. However, their occurrence strongly varies with experimental factors such as P-T conditions and the composition of the starting materials. Serpentine minerals are widespread but heterogeneously distributed in the subducting mantle, as they are produced by hydrous alteration of ultramafic and mafic rocks through fault-controlled fluid infiltration, from mid-ocean ridges29 to subduction zones30,31. Serpentines are also thought to be the source materials for the stable DHMSs, which can transfer water through the mantle transition zone (MTZ) into the lower mantle via penetrating (cold) slabs32–34. There are various types of serpentines, including three main varieties (i.e., lizardite, antigorite, and chrysotile) as well as other, less common forms such as conical, polygonal, and polyhedral serpentines35–37. They are all 1:1 (Mg-octahedral:Si-tetrahedral) layered magnesium silicates with the ideal chemical composition of Mg3Si2O5(OH)4. Slight chemical variations among serpentines and interlayer stress compensation resulting from dimensional misfit between octahedral and tetrahedral sheets give rise to the structural variations of serpentines38. Lizardite is the most abundant serpentine mineral at ambient and sub-surface conditions (<300 °C), while antigorite is favored in high-grade metamorphic conditions with a modulated layer structure39. Chrysotile, characterized by cylindrical fibrous structure, is less stable than antigorite and lizaridte40. Serpentine minerals are stable up to the temperature range of 400–600 °C, with variations depending on the type of serpentines, above which they break down depending on the pressure conditions. Under the high-T (> 600 °C) and moderate-P (< 5 GPa) conditions, serpentine minerals decompose into forsterite (Mg2SiO4) ± enstatite (MgSiO3) ± talc or 10-Å phase + H2O, i.e., a dehydration reaction releasing H2O41,42. Under moderate-T (400–600 °C) and high-P (5–8 GPa) conditions, i.e., along a cold subduction geotherm, serpentine minerals decompose into phase A (Mg7Si2O8(OH)6) + enstatite ± 10-Å phase + H2O, thus retaining ≈ 3.9 wt.% water to be transported into the deep mantle32,42,43. To better understand the stability of serpentinites in cold subducting slabs, we perform in-situ high-pressure, high-temperature (HP-HT) X-ray diffraction experiments on serpentinites (lizardite and antigorite aggregates) in excess-water conditions in the range of ≈ 0.9–31.5 wt.%, using a resistive-heated diamond-anvil cell (RH-DAC) up to 785 °C and 13 GPa, simulating endmember geotherms within cold and ultracold subducting slabs, i.e., one path along the cold core region (the coldest part of a subducting slab) or ultracold Moho region (as in the case of Tonga), and another path along the cold slab Moho (as defined on the basis of the global average 7-km-thick crust). We also perform, separately, in-situ ultrasonic wave velocity measurements using a multi-anvil press (MAP) to observe the changes in wave velocity prior to and following the mineral breakdown reactions. We report here contrasting transformations in serpentinites depending on the thermal path as well as the excess-water conditions and discuss their impacts in terms of enhanced water transfer/storage and lateral contrasts in hydration level, density, and seismic velocity of a subducting slab. Results Hydration and water-retention reactions in serpentinites along the ultracold Moho and cold core geotherms In-situ HP-HT X-ray powder diffraction (XRD) experiments on serpentinites (lizardite powder as a starting material) in ≈ 31.5 wt.% water were performed up to 13.1 ± 0.3 GPa and 520 ± 35 °C to follow the cold core geotherms of the Honshu thermal model, representative of cold subduction zones (Fig. 1, Supplementary Fig. 1a, and Supplementary Table 1). These geotherms also correspond to the Moho geotherms of the Tonga thermal model, a typical example of ultracold subduction zones. Under such water-excess and cold conditions, the stability of lizardite is extended to a depth of ≈ 280 km, i.e., up to 9.3 ± 0.3 GPa and 420 ± 35 °C, below which it breaks down into the 3.65-Å phase (MgSi(OH)6) and phase E assemblage, as below: lizardite/antigorite water 3.65-Å phase phase E. Experimental pressure-temperature (P-T) conditions of serpentinite with water along cold subduction geotherms. a Geotherm-dependent stability of serpentines along the Honshu thermal model using an RH-DAC (open symbols) and a MAP (open symbols with a central dot) setup. Blue symbols represent experiments conducted along cold-core (or ultracold Moho) conditions, whereas purple symbols represent experiments conducted along cold Moho conditions. Error bars indicate pressure and temperature uncertainties. b Previously established stability fields of serpentinite performed without excess water in the same PT-range as in (a). Colored solid lines define the stability field of various serpentinites26,28,32,41–43,48,49,52 including those identified in this study, while thin dashed lines indicate the stability of related minerals24,92,93. The bold solid lines denote the P-T paths of subducting slab surfaces, slab Moho, and cold core, respectively. Slab Moho is about 7 km below the slab surface, while the cold core delineates the lowest geotherm. These geotherm models have been extrapolated from previously published datasets and are based on the models of ref. 94 for the slab surface and Moho, and ref. 89 for the cold core. MAP multi-anvil press, RH-DAC resistive-heated diamond-anvil cell, MSH MgO–SiO2–H2O, MASH MgO–Al2O3–SiO2–H2O, FMSH FeO–MgO–SiO2–H2O, MFASH MgO–FeO–Al2O3–SiO2–H2O. Phase abbreviations: lz lizardite, atg antigorite, E phase E, Al-E aluminous phase E, 3.65 Å 3.65-Å phase, D phase D, A phase A, tlc talc, 10 Å 10-Å phase, en enstatite, cen/oen clino/orthoenstatite, bal balangeroite, fo forsterite, st stishovite, grt garnet, sur Mg-sursassite, brc brucite, W water. This breakdown reaction differs in terms of the change in water content. Upon the formation of the two DHMS phases, there is a net 6.3 wt.% increase (13.0 → 19.3 wt.%) in the crystal water content, which can be described as a form of hydration reaction occurring in cold core (and ultracold Moho) environments. Since phase E has a non-stoichiometric chemical formula44,45, its chemical formula was modeled by the Rietveld method using the measured XRD data (Supplementary Fig. 2a and Supplementary Table 5) as (Mg,Fe)2.3Si1.3O6H2.2. The phase E compositions observed in this study can be described using a modified charge-balance expression based on previously reported compositions and characteristics of phase E44–47 as M3-0.5x-ySixO6H6-3x+2y (M = Mg, Fe²⁺), where x represents the Si content and y represents the additional M-site vacancy. In this formulation, an increase in the M-site vacancy decreases the divalent-cation content while requiring additional H for charge compensation. In the case, antigorite was used as a starting sample, the breakdown into the 3.65-Å phase and phase E assemblage is confirmed at slightly deeper depth of ≈ 290 km, i.e., near 9.6 ± 0.7 GPa and 555 ± 90 °C (Fig. 1, supplementary Fig. 3, and Supplementary Table 3). Although lizardite and antigorite show some differences in stability28,48,49, both exhibit the same hydration reaction under the cold core (and ultracold Moho) conditions with excess-water. In our experiments, we note that the temperature stability of the 3.65-Å phase persists to 330 km, i.e., 590 ± 90 °C at 10.9 ± 0.7 GPa. Such DHMS phases are known to be stable enough to transport water to greater depths; phase E contains ≈ 11.3–18.7 wt.% H2O and is known to be stable up to 14–18 GPa44,45,50 where it breaks down to another DHMS, phase D or superhydrous phase B with H2O content of 14–18 or 5.8 wt.%, respectively32,51. Previous studies have also shown that the 3.65-Å phase with ca. 34.9 wt.% water can remain stable up to 640 °C at 14 GPa24, corresponding to the P-T range of a cold subducting slab penetrating the MTZ. To further evaluate the effect of water content on the observed hydration reaction, two additional sets of in-situ HP-HT experiments were conducted. The first set used silicone oil as pressure-transmitting medium (PTM) to simulate fluid-limited conditions (i.e., 0.9 wt.% of surface-absorbed H2O), while the second set utilized a higher water content of 6.9 wt.%. Both experiments followed the same geotherms as in the run with 31.5 wt.% water. In the fluid-limited experiment using silicone oil, serpentine (starting sample of lizardite) transforms directly to phase E near 10.4 ± 0.3 GPa and 530 ± 35 °C, i.e., ≈ 310 km depth conditions (Fig. 1, Supplementary Fig. 4a, and Supplementary Table 1). In the experiment using ≈ 6.9 wt.% water, transformation of serpentine to phase E was also observed, accompanied by a trace amount of the 3.65-Å phase at slightly lower P-T conditions of 10.0 ± 0.3 GPa and 510 ± 35 °C, i.e., ≈ 300 km depth, than in the run with 30.6 wt.% water. Given that no other phases were observed in the XRD patterns (Fig. 1, Supplementary Fig. 4b, and Supplementary Table 1), the observed transformations can be described as a water-retention reaction by assuming the non-stoichiometric composition of phase E as below: lizardite/antigorite phase E. The above composition for phase E has been confirmed by semi-quantitative SEM-EDS analysis performed on the recovered sample from the ≈ 6.9 wt.% water run, which indicated the chemical formula of phase E to be Mg1.9Fe0.2Si1.3O6H2.6, assuming the H content for charge compensation, to validate the water-retention reaction described by Eq. (2) (Supplementary Fig. 5). By the depth near the 410 km discontinuity at 13.3 ± 0.6 GPa and 705 ± 90 °C, the formation of phase E + phase D assemblage is observed (Fig. 1), which has been reported as the second breakdown products from serpentine, e.g., serpentine → phase A → phase E + phase D, in previous studies32. Therefore, the observed hydration reaction described by Eq. (1) of serpentine appears to be specific for cold core (and ultracold Moho) conditions when certain excess-water conditions are assumed, i.e., when surrounded by at least > 6.9 wt.% water. Dehydration reaction of serpentines along the cold Moho geotherm In-situ HP-HT XRD experiments on serpentinites (lizardite powder as a starting material) in ≈ 8.4 wt.% H2O were performed up to 13.0 GPa and 785 ± 90 °C to follow the cold Moho geotherms of the Honshu models (Fig. 1, Supplementary Fig. 1a, and Supplementary Table 2). Serpentine remains stable up to 7.4 ± 0.7 GPa and 585 ± 90 °C, corresponding to ≈ 220 km depth, where it starts to form phase E and clinoenstatite plus a minor amount of balangeroite (Mg21Si8O27(OH)20)52. Upon a marginal increase in P-T to 8.1 ± 0.7 GPa and 680 ± 90 °C, i.e., ≈ 240 km depth, diffraction peaks corresponding to balangeroite disappear, and the phase E and clinoenstatite assemblage remains. Considering the transient nature of balangeroite, the breakdown of serpentine along excess-water, cold Moho conditions can be summarized as below: lizardite/antigorite phase E clinoenstatite water. The chemical composition of phase E and its H2O content were modeled by the Rietveld method to confirm the reaction described by Eq. (3) (Supplementary Fig. 2b and Supplementary Table 6). The observed breakdown can thus be categorized as a dehydration reaction releasing ≈ 5.3 wt.% H2O (i.e., ≈ 13.0 → 7.7 wt.%) by the depth down to ≈ 240 km. This contrasts with the previous finding that serpentinites dehydrate through crystallization of phase A in the depth range of 180–240 km along slightly warmer geotherms in water-unsaturated conditions32,41,43. Since the 10-Å phase did not appear in our experiments conducted in excess-water conditions along cold subduction geotherms, in contrast to some previous results53,54, we performed additional in-situ HP-HT XRD experiment on the same sample with a 9.4 wt.% H₂O to investigate the relationship between the formation of the 10-Å phase and the P–T–H₂O conditions during serpentinite reactions (Fig. 1, Supplementary Fig. 4c, and Supplementary Table 2). In this experiment, the conditions were extended to slightly higher temperatures (>550 °C) and pressures (>5 GPa). The 10-Å phase appeared at 5.4 ± 0.3 GPa and 540 ± 35 °C and coexisted with lizardite up to 6.0 ± 0.7 GPa and 705 ± 70 °C (Supplementary Fig. 4d). These results indicate that the stability field of the 10-Å phase would be shifted to higher temperatures under excess-water conditions. Ultrasound velocity of subducting serpentinites along the cold core and Moho geotherms To establish a correlation between the observed, geotherm-dependent breakdown reactions of serpentines and changes in the resulting seismic velocity, in-situ HP-HT ultrasonic velocity measurements were conducted on a mixture of serpentinite (starting sample of lizardite powder) and water using the MAP at the P61B beamline at PETRA-III (see “Methods”, Fig. 2a, b). To monitor the reaction products, diffraction patterns were obtained using energy-dispersive X-ray diffraction (ED-XRD) (Fig. 1, Supplementary Fig. 1b, and Supplementary Table 4), while changes in sample length through the reactions were measured using X-ray radiography. Under the cold core conditions at 525 °C and 10.0 ± 0.4 GPa, the ED-XRD on the mixture of serpentine and 20.9 wt.% water confirmed the hydration reaction described by Eq. (1) to form the 3.65-Å phase and phase E assemblage, as observed in the diamond-anvil cell (DAC) experiments. Upon this reaction, the sample length decreased by approximately 10%, and the measured P-wave velocity increased by approximately 17%. Under the cold Moho conditions at 573 °C and 7.7 ± 0.2 GPa, the serpentinites (lizardite powder as a starting material) plus 25.9 wt.% water also confirmed the dehydration reaction described by Eq. (3) to form the assemblage of phase E and clinoenstatite (plus the transient formation of balangeroite), as observed in the corresponding DAC experiments. Concomitant to this reaction, the sample length decreased by approximately 15%, and the measured P-wave velocity increased by approximately 20%. Such changes in the seismic wave velocity and volume advocate discrete yet contrasting reactions in serpentinites in the cold core (and ultracold Moho) and cold Moho regions. Changes in the seismic velocity and density of serpentine along cold Moho and cold core (or ultracold Moho) geotherms. a P-wave velocities were derived from the MAP experiments (symbols) and compared to the calculated estimations. The P-wave velocity of the solid assemblage (gray solid lines) was calculated using the VRH average, and this value was used to estimate the water-solid mixture, whose upper and lower limits (red solid and dot lines) were constrained by the Hashin–Shtrikman (HS) bounds. Error bars indicate velocity and pressure uncertainties. b X-ray radiographic images were used to derive changes in the sample length and to correlate with the respective reactions, together with changes in density and sample length in the MAP experiments. Abrupt changes in the sample length and density are observed concomitant to the hydration (cold core) and dehydration (cold Moho). Error bars indicate uncertainties in sample length, velocity and pressure. MAP multi-anvil press, Vp P-velocity. Phase abbreviations: lz lizardite, E phase E, 3.65 Å 3.65-Å phase, cen clinoenstatite, bal balangeroite, W water. The stability limits and reaction products of lizardite and antigorite observed in our experimental conditions are summarized and compared with previous experimental results in Table 1. We observed that both the stability and the reaction pathways of serpentinite differ from those reported earlier given the applied excess-water conditions26,28,32,41–43,48,49,52 (Fig. 1). The enhanced stability of serpentine under cold-core conditions is consistently observed across all the experiments conducted with excess-water (i.e., 0.9, 6.9, and 31.5 wt.% H2O). This observation suggests that, consistently with previous studies, serpentine stability is primarily governed by the presence of H2O26. More importantly, Eqs. (1) and (2) indicate that the amount of excess H2O controls the formation of the 3.65-Å phase, which is an indicator for water-excess conditions. Table 1Stabilities of serpentines with excess water (this study), compared to the previous results performed without excess waterExcess water (wt.%)Sample typeMax. PressureMax. TemperatureMax. DepthaDHMS productsReferenceUltracold Moho/cold core conditions31.5lz9.3 GPa420 °C279 kmE + 3.65 ÅThis study20.9lz9.4 GPa476 °C282 kmE + 3.65 Å16.4atg9.6 GPa555 °C288 kmE + 3.65 Å6.9lz9.3 GPa490 °C279 kmE0.9 (silicone oil)lz9.7 GPa500 °C291 kmE≈0atg8 GPa400 °C240 kmA + en (± 10 Å)Irifune et al. (1998)32Cold Moho conditions7.4lz7.4 GPa585 °C222 kmE + en ± balThis study25.9lz7.8 GPa546 °C234 kmE + en ± bal≈0atg (MSH)6.6 GPa520 °C198 kmA + enKomabayashi et al. (2005)43≈(3.5)batg (MSH)6.9 GPa540 °C207 kmA + en ± 10 ÅUlmer & Trommsdorff (1995)41DHMS dense hydrous magnesium silicates, MSH MgO–SiO2–H2O. lz lizardite, atg antigorite, E phase E, 3.65 Å 3.65-Å phase, A phase A, en enstatite, 10 Å 10-Å phase, bal balangeroite.aDepth was converted to pressure using a lithostatic gradient of 1 GPa ≈ 30 km, a commonly adopted approximation in subduction zone studies52,88.bThe value in parentheses indicates the amount of water predicted from the dehydration of brucite. Stabilities of serpentines with excess water (this study), compared to the previous results performed without excess water DHMS dense hydrous magnesium silicates, MSH MgO–SiO2–H2O. lz lizardite, atg antigorite, E phase E, 3.65 Å 3.65-Å phase, A phase A, en enstatite, 10 Å 10-Å phase, bal balangeroite. aDepth was converted to pressure using a lithostatic gradient of 1 GPa ≈ 30 km, a commonly adopted approximation in subduction zone studies52,88. bThe value in parentheses indicates the amount of water predicted from the dehydration of brucite. In all reactions described by Eqs. (1)–(3), we observe that lizardite transforms directly into an assemblage containing phase E, bypassing phase A. To confirm this represents (near-)equilibrium reaction, we conducted additional long-duration annealing experiments (6–12 h; see Supplementary Fig. 6), considering the potentially sluggish kinetics of phase A formation. However, no evidence for the formation of phase A was found, and the proportion of the reaction products remain unchanged over time. These observations are difficult to reconcile with simple kinetic limitations or disequilibrium effects and instead suggest the presence of previously unrecognized P–T phase boundaries or a compositional shift in phase stability fields. The excess-water conditions and bulk composition used in this study may have contributed to these differences. In fact, the stability fields of DHMS are known to vary as a function of water content55; therefore, under the given P–T conditions, the presence of excess water may have further enhanced the stability of hydrous phases such as the highly hydrated 3.65-Å phase and phase E, which possesses a relatively flexible hydrogen-bonding network45 compared to phase A. Consistent with this flexibility, phase E in Eqs. (1)–(3) exhibits a range of estimated H contents of 2.2–2.6 per O6, depending on reaction stoichiometry, coexisting assemblages, and water contents. The derived compositions in the modified charge-balance expression correspond to x = 1.20–1.33 and y = 0.050–0.335. In addition to such water effects, the presence of Fe and Al has also been reported to play a key role in stabilizing phase E at relatively low pressures52. Discussion Lateral contrasts in water content and density in a subducting slab It is important to correlate the observed geotherm-dependent paragenetic evolutions to corresponding changes in the net water content and crystalline density within a subducting slab. Under the cold core (or ultracold Moho) geotherms, serpentine breaks down near 280 km depth to form the 3.65-Å phase + phase E assemblage when enriched in water, as observed in the DAC and MAP runs with ≈ 31.5 and ≈ 20.9 wt.% H2O, respectively. Alternatively, serpentine transforms directly into phase E near ≈ 295 km depth when water availability is limited, as observed in the water-limited conditions by less than ≈ 6.9 wt.% H2O in the DAC runs. The reaction observed under water-rich conditions (by more than ≈ 6.9 wt.%) is defined as a hydration reaction, as in Eq. (1), as the net water content increases from ≈ 13.0 to 19.3 wt.%. This is due to the formation of the 3.65-Å phase, one of the most hydrated DHMS with ≈ 34.9 wt.% H2O. The transformation observed under the water-limited conditions can then be defined as a water-retention reaction, as in Eq. (2), as the net water content is maintained at ca. 13.0 wt.%. In the experiment under water-rich, cold Moho conditions, we observed the formation of the phase E + clinoenstatite (+ balangeroite) assemblage as breakdown products of serpentine near 220 km depth (Fig. 1). However, the depth range where the three-phase assemblage exists is narrow (≈ 20 km), while the diffraction peaks identified as balangeroite are weak and broad, which led us to conclude that the formation of balangeroite is transient, possibly related to its narrow stability field52. At depths > 240 km, the two-phase assemblage of phase E and clinoenstatite persists down to ≈ 390 km depth. If we consider the net water content of the phase E + clinoenstatite assemblage, where phase E contains ≈ 12 wt.% H2O, the breakdown of serpentine into the phase E + clinoenstatite assemblage releases ≈ 5.3 wt.% H2O. This is about 60% of the previously reported water release of ≈ 9.0 wt.%, based on the breakdown of serpentine into the phase A + clinoenstatite assemblage42. On one hand, under water-rich, cold Moho conditions, more water, at least 7.7 wt.%, can thus be retained as the breakdown products of serpentine to be transported deeper into the Earth. Through the dehydration reaction under the water-rich, cold Moho conditions, the net crystalline density increases by ≈ 16.0% from 2.84 ± 0.01 g/cm3 (serpentine) to 3.29 ± 0.01 g/cm3 (phase E + clinoenstatite) near 240 km depth (Fig. 2b and Supplementary Fig. 7a). On the other hand, through the hydration or water-retention reaction under the water-rich or water-limited cold core (or ultracold Moho) conditions, the increase in the net crystalline density is modulated to only ca. 8.0 or 13.1%, from 2.89 ± 0.01 g/cm3 (serpentine) to 3.11 ± 0.01 g/cm3 (3.65-Å phase + phase E) or 3.27 ± 0.01 g/cm3 (phase E), respectively (Fig. 2b and Supplementary Fig. 7b). By the depth of ≈ 300 km, where the contrasting reactions (i.e., hydration/water-retention and dehydration) are completed along the cold subduction zone, the assemblage from serpentinite breakdown in the cold Moho would become denser and drier by up to ~7.1% and ~11.6 wt.%, respectively, than that in the cold core. Contrasting changes in the P-wave velocity for the two geotherms In order to better understand the seismic velocity model of a cold subducting slab, we have estimated the correlation between the geotherm-dependent stabilities of serpentine and the resulting seismic velocity changes. We estimated the P-wave velocity of the observed phase assemblages in both water-limited and water-rich cases, the latter with the amount of water as in the MAP experimental conditions, using the Voigt-Reuss-Hill (VRH) average and the Hashin–Shtrikman (HS) bounds (Fig. 2a and Supplementary Table 7)56,57. While the VRH average was applied for the P-wave velocity of the solid-phase assemblage, the HS method was added to incorporate the effects of H₂O to constraint the upper and lower bounds for the P-wave velocity of the solid-phase + water assemblage. The dehydration of serpentinite observed along the water-rich, cold Moho path would account for the 10.3-% increase of P-wave velocity, whereas the hydration reaction observed along the water-rich, cold core (or ultracold Moho) path would account for the 7.5-% increase. When comparing the calculated and observed velocities, we find that the observed P-wave velocity under the cold-core condition lies between the upper and lower HS estimates, whereas that under the cold-Moho condition is closer to the upper bound. Overall, the calculated and observed velocities show good agreement within experimental uncertainty, confirming the role of excess water and the contrasting changes after the reactions. While further studies are required to account for the influence of serpentine and its reaction assemblages on seismic velocities in actual subduction zone topography, the observed changes in the P-wave velocity conform to the respective hydration and dehydration reactions, accompanied by a contrast in net crystalline density between the cold core and Moho regions. Implications of the hydration reaction for deep water transfer Serpentine, as the most abundant hydrous mineral in subduction systems and most stable at depths <200 km, plays a major role in water transportation and its deep cycling into the mantle30,58. Based on our experimental results, we have re-evaluated the water transport ability of contemporary cold subducting slabs (Table 2, Fig. 3a, and supplementary Tables 8–10). Among the 56 main subduction zones, 16 can be classified as cold (or ultracold) based on their thermal characteristics8. Table 2Reevaluation of the H2O flux in the 16 global cold subduction zonesParametersValueReference1. Thickness of the hydrated slab mantle2. H2O content in the slab mantle25 kmAverage 2 wt.%Cai et al. (2018)1Hacker (2008)63. Initial H2O flux of the slab mantle in the global cold subduction zones ➢ serpentine ➢ chlorite~1.4 × 1015 g/yr~1.2 × 1015 g/yr~2.0 × 1014 g/yrCalculated based on Schmidt and poli (1998)59, Cai et al. (2018)1, van Keken et al. (2011)3, and references therein(supplementary Table 8)4. H2O fluxes by serpentine and its breakdown products0–210 km subducting serpentine~1.2 × 1015 g/yr* The geotherms of the Central Honshu were applied to represent coldsubduction system89.* Dehydration of a slab mantle down to a thickness of 4 km, considering P-T conditions with subduction depth.* Since more than 0.9 wt.% water is available for retention/hydration reaction, we assume that the cold core undergoes only preservation orhydration reactions to allow a clear comparison. 210–260 kmSlab Moho (7–11 km thickness) phase A (ref. 42) or phase E (Eq. (3)at >220 km) assemblagesCold core (11–25 km thickness) serpentine~1.0 × 1015 g/yr~1.0–2.1 × 1014 g/yr(Average ~1.6 × 1014 g/yr)~8.4 × 1014 g/yr >300 kmSlab MohoCold core (Eqs. (1) or (2) at >260 km) Case 1. water-retention Case 2. hydration + water-retentiona~1.0–1.1 × 1015 g/yr1.0–2.1 × 1014 g/yr(Average ~1.6 × 1014 g/yr)~8.4 × 1014 g/yr~9.0 × 1014 g/yr5. H2O flux by chlorite0–210 km subducting chlorite210–270 km Mg-sursassite assemblage270–330 km 23 Å-phase assemblage~2.0 × 1014 g/yr~2.5 × 1013 g/yr~0.3 × 1013 g/yrCalculated based on Van Keken et al. (2011)3 and Cai & Inoue (2019)61 (supplementary Table 8)6. The total amount of H2O available for hydration reactionDehydration-induced free H2O (+) ➢Mg-sursassite dehydration ➢ Serpentine dehydrationLoss (−) ➢ nominally anhydrous minerals (~1000 ppm) ➢loss due to upward movementb~0.6 × 1014 g/yr~2.2 × 1014 g/yr~2.2 × 1013 g/yr1.5–2.6 × 1014 g/yr(Average ~2.0 × 1014 g/yr)~1.6 × 1014 g/yr~5.9 × 1013 g/yr0.7–1.3 × 1014 g/yr(Average ~1.0 × 1014 g/yr)Calculated based on Van Keken et al. (2011)3Cai & Inoue (2019)61aSince it is difficult to quantitatively constrain the spatial accumulation of fluids in the cold core of the mantle layer, we did not estimate the boundary condition, i.e., the minimum water content for the hydration reaction, between Case 1 and Case 2.bConsidering the typical dip angle (45°)90,91 and depth extent of bending faults, free fluids formed at depths shallower than 240 km are inferred to migrate up-dip due to pressure gradients rather than being consumed in hydration reactions along the fault network.*A thermal parameter of >40 was assumed to define the 16 cold subduction zones amongst the total 56 global subduction zones8. Reevaluation of the H2O flux in the 16 global cold subduction zones 1. Thickness of the hydrated slab mantle 2. H2O content in the slab mantle 25 km Average 2 wt.% Cai et al. (2018)1 Hacker (2008)6 3. Initial H2O flux of the slab mantle in the global cold subduction zones ➢ serpentine ➢ chlorite ~1.4 × 1015 g/yr ~1.2 × 1015 g/yr ~2.0 × 1014 g/yr 4. H2O fluxes by serpentine and its breakdown products 0–210 km subducting serpentine * The geotherms of the Central Honshu were applied to represent coldsubduction system89. * Dehydration of a slab mantle down to a thickness of 4 km, considering P-T conditions with subduction depth. * Since more than 0.9 wt.% water is available for retention/hydration reaction, we assume that the cold core undergoes only preservation orhydration reactions to allow a clear comparison. 210–260 km Slab Moho (7–11 km thickness) phase A (ref. 42) or phase E (Eq. (3)at >220 km) assemblages Cold core (11–25 km thickness) serpentine ~1.0 × 1015 g/yr ~1.0–2.1 × 1014 g/yr (Average ~1.6 × 1014 g/yr) ~8.4 × 1014 g/yr >300 km Slab Moho Cold core (Eqs. (1) or (2) at >260 km) Case 1. water-retention Case 2. hydration + water-retentiona ~1.0–1.1 × 1015 g/yr 1.0–2.1 × 1014 g/yr (Average ~1.6 × 1014 g/yr) ~8.4 × 1014 g/yr ~9.0 × 1014 g/yr 5. H2O flux by chlorite 0–210 km subducting chlorite 210–270 km Mg-sursassite assemblage 270–330 km 23 Å-phase assemblage ~2.0 × 1014 g/yr ~2.5 × 1013 g/yr ~0.3 × 1013 g/yr 6. The total amount of H2O available for hydration reaction Dehydration-induced free H2O (+) ➢Mg-sursassite dehydration ➢ Serpentine dehydration Loss (−) ➢ nominally anhydrous minerals (~1000 ppm) ➢loss due to upward movementb ~0.6 × 1014 g/yr ~2.2 × 1014 g/yr ~2.2 × 1013 g/yr 1.5–2.6 × 1014 g/yr (Average ~2.0 × 1014 g/yr) ~1.6 × 1014 g/yr ~5.9 × 1013 g/yr 0.7–1.3 × 1014 g/yr (Average ~1.0 × 1014 g/yr) Calculated based on Van Keken et al. (2011)3 Cai & Inoue (2019)61 aSince it is difficult to quantitatively constrain the spatial accumulation of fluids in the cold core of the mantle layer, we did not estimate the boundary condition, i.e., the minimum water content for the hydration reaction, between Case 1 and Case 2. bConsidering the typical dip angle (45°)90,91 and depth extent of bending faults, free fluids formed at depths shallower than 240 km are inferred to migrate up-dip due to pressure gradients rather than being consumed in hydration reactions along the fault network. *A thermal parameter of >40 was assumed to define the 16 cold subduction zones amongst the total 56 global subduction zones8. Schematic diagram illustrating the H₂O transport capacity through the geotherm-dependent serpentine’s wet breakdown paths. a The geotherm-dependent breakdown of serpentine is illustrated with a schematic subduction diagram. The numbers in circles denote each equation in the manuscript: (1) hydration, (2) water-retention, and (3) dehydration. The isotherm lines are adapted from ref. 89. The left inset highlights the faulting and fault reactivation affecting the subducting slab due to bending, while the right inset highlights reactions along the cold core and Moho regions resulting from fluid infiltration along the reactivated faults. The dehydration curve for serpentine is from this study, whereas that for Mg-sursassite is taken from ref. 61. b Estimations of H2O fluxes by subducting serpentines in the global subduction zones; H2O fluxes in the sediment/crust (yellow) and mantle excluding the 16 cold subduction zones (gray) were estimated from ref. 3 and ref. 1. The contributions by serpentines in the 16 cold subduction zones were estimated in two ways: (dark green, this study) based on the results with excess water and (light green) based on the dehydration reactions performed without excess water as reported by ref. 42. For the degree of serpentinization in the slab mantle, the same criteria were applied, i.e., hydrous alteration by 2 wt.% H2O down to 25 km from Moho in the 16 cold subduction systems. SZs subduction zones. Phase abbreviations: srp serpentine, E phase E, A phase A, bal balangeroite, 3.65 Å 3.65-Å phase, cen clinoenstatite, sur Mg-sursassite. Following ref. 59, we assume that the initial composition of the slab mantle is that of a partially hydrated harzburgite, with serpentine and chlorite as the principal hydrous minerals (Supplementary Table 8). Recent studies indicate that serpentinization can progress to depths of at least 25 km beneath the slab Moho in cold subduction zones1,60. Based on these constraints, we adopt a simple linear (-proportional) hydration model to estimate the degree of fault development and serpentinization, i.e., up to 30% serpentinization at the Moho and progressively reduced values to <1% at 25 km beneath the Moho. Under these conditions, the average water content of the hydrated mantle is ~2 wt.%, and the initial water flux subducting into the mantle by the 16 cold subduction zones is estimated to be ~1.4 × 1015 g yr−1 13. The subducting hydrated mantle may be supplied with free water through fluid redistribution along hydrated fracture zones19. Within this simplified compositional model of the slab mantle, the primary water sources are the dehydration of serpentine and Mg-sursassite, with the latter forming through chlorite breakdown. For both cases, a V-shaped breakdown field is expected in subduction zones as dehydration would occur at progressively shallower depths closer to the slab Moho (Fig. 3a)61. Although the stability of serpentinite is likely to be controlled by its composition and the presence of excess water as observed in this study, the natural compositions of subduction zone serpentinites suggest that dehydration begins at around 210 km under dry condition52,62. Similarly, dehydration of Mg-sursassite is generally thought to occur at depths of 210–270 km61. Traditionally, released fluids in the slab mantle have been considered to migrate upward along the pressure gradient63. However, recent studies suggest that such fluids may also migrate toward the cold core in response to variations in tectonic pressure induced by slab unbending64,65. Furthermore, chemical-potential gradients may influence fluid redistribution and percolation pathways within the slab and along the plate interface in subduction zones66,67. We should also consider that serpentinization of the lithospheric mantle results from localized water-rock interaction driven by fluid percolation along deep bending faults or reactivated abyssal-hill faults cutting through the lithospheric mantle to the brittle-ductile transition68. These faults may be reactivated via mineral dehydration, thereby providing local fluid pathways60,69,70. Integrating previously proposed scenarios for fluid redistribution along pre-existing faults in metasomatized peridotites19,71 with pressure-driven fluid migration in the slab mantle64,65, we suggest that fluids released by serpentine dehydration near the slab Moho and by Mg-sursassite dehydration within the hydrated mantle migrate along these faults. This process supplies water to the surrounding serpentinite, ultimately driving local fluid accumulation. This process can drive local variations in mantle composition19, thereby increasing the stability of fluid-exposed serpentinite and promoting reactions that require excess water. On this basis, we could estimate the depth-dependent water transport capacity by global cold subductions, assuming our observed hydration (3.65-Å phase + phase E), water-retention (phase E), and dehydration (phase E + clinoenstatite) reactions of serpentinites, and compared them to the conventional estimates, which consider only the dehydration reaction of serpentinite into phase A42 (Fig. 3a and Table 2). By a depth of ~210 km, the amount of water supplied by serpentine is approximately ~1.2 × 1015 g yr−1. In the deeper 210–260 km interval, serpentinite breakdown occurs beneath the cold Moho, and depending on local fluid availability and compositional conditions, either phase A or phase E assemblage would form, releasing approximately ~2.6 × 1014 g yr−1 or ~1.5 × 1014 g yr−1 of water, respectively. In addition, breakdown of Mg-sursassite within the mantle can contribute an additional ~2.2 × 1013 g yr−1 of free fluid. Because part of these fluids may ascend before reaching the depths of the DHMS stability field, or get stored in nominally anhydrous minerals (NAMs)62,64, the amount of free fluid available to form the 3.65-Å phase and/or phase E assemblage under cold-core conditions at depths greater than 260 km would likely be limited to less than 27% (Table 2). However, it is worth noting that even a small amount of free water under cold-core conditions (0.9 wt.%) is sufficient for these reactions, and the amount of fluid that could be supplied, estimated at ~0.6 × 1014 g yr−1 wt.%, is locally available to create such conditions. However, given that the spatial extent and degree of progress of fluid-induced reactions cannot be directly quantified, our estimate should be regarded as an upper-bound scenario in which hydration or water-retention reactions occur wherever local fluid availability and thermal conditions allow it. Considering these geotherm-dependent lateral variability in reactions schemes by a depth of 300 km, up to 1.0–1.1 × 1015 g yr−1 H2O could be supplied through cold subduction zones globally, which translates to 83–92% efficiency of water transportation by subducting serpentinites. This is ~3.1–3.3 times higher than the conventional estimates assuming only dehydration into phase A. Applying this to the global subduction system, the amount of water to be transported to (and possibly stored in) the MTZ would then become ~1.2–1.3 × 1015 g yr−1, about 2.2–2.3 times of the previous estimates1,3,42 (Fig. 3b). Moreover, in the ultracold subduction zones as exemplified by Tonga, the region where deep hydration reactions occur may expand, potentially leading to a further increase in water transport. Methods To investigate the stability and physical properties of serpentinite under high-pressure and high-temperature conditions, we performed three complementary experiments: (1) in-situ synchrotron angle-dispersive XRD using a DAC, (2) in-situ synchrotron ED-XRD and radiography using a MAP, and (3) acoustic reflection measurements of ultrasonic wave velocities. Details of the sample preparation and experimental procedures are described below. Sample initial characterization We have used powdered samples of lizardite (1T-type, Boryeong area in Chungcheongnam-do, Korea, Mg2.98Fe0.12Al0.01Ca0.01Si1.95O5(OH)3.91) and antigorite (Sta Olalla de Cala in Huelva, Spain, Mg3.04Fe0.05Si1.96O5(OH)3.98). The initial chemical composition was characterized using X-ray fluorescence (Tornado M4) at Yonsei University and thermogravimetric analysis (TGA, SDT 500) at the Yonsei Center for Research Facilities (YCRF) to determine the water content as well as cationic compositions (Supplementary Figs. 8 and 9). The weight loss observed in the TGA at temperatures below 300 °C indicates the presence of surface-adsorbed water in the powdered samples. In fact, previous study on lizardite powder showed its decomposition by releasing structurally bound water above 400 °C72. The amount of surface-adsorbed water is approximately 0.9 wt.% (for lizardite) or 2.6 wt.% (for antigorite). In addition, field emission scanning electron microscopy (JEOL-7800F) at the YCRF and synchrotron XRD at the Extreme Conditions Beamline (ECB) P02.2 of PETRA-III, Germany, were conducted to examine morphological features and to identify crystalline phases, respectively (Supplementary Figs. 8 and 9). At beamline P02.2, the X-ray beam generated by the undulator source was tuned to a wavelength of 0.4834(1) Å (25.650 keV) and focused to 8 (H) × 3 (V) µm2 in size (FWHM) using Compound Refractive Lense (CRL) optics. Diffraction data were collected using a Perkin Elmer XRD 1621 detector, positioned ~401 mm away from the sample, with an exposure time of 10 s. The two-dimensional synchrotron XRD images were processed to one-dimensional patterns using the Dioptas program73, which were then analyzed using the LeBail method within the GSAS suite of programs74,75 and TOPAS76. The refined unit-cell parameters of lizardite-1T at ambient conditions are a = 5.3180(3) Å and c = 7.3106(6) Å in space group P31m, while those of antigorite are refined to a = 43.559(3) Å, b = 9.2337(8) Å, c = 7.2612(2) Å and β = 89.38(1) in space group Pm. Transformation of lizardite to antigorite Antigorite is generally more stable than lizardite in high-temperature environments above 300 °C77. Our experimental conditions following a cold subducting slab would therefore exert the transformation of lizardite to antigorite around 120 km depth conditions. When lizardite was used as a starting sample, we observed the formation of antigorite via in-situ HP-HT XRD and ex-situ Raman measurements (Supplementary Fig. 10). In the XRD measurement, we identified the characteristic diffraction peaks of antigorite in the d-spacing range of 2.4–2.6 Å above 300 °C78. In the Raman spectroscopy measurement, we identified the characteristic shifts in the 3600–3800 cm⁻¹ regions from the quenched lizardite sample after heating to 300 °C at 4 GPa for 6 h77,79. However, we observed the coexistence of lizardite and antigorite within their stability range in our experiment. In-situ high-pressure, high-temperature synchrotron X-ray powder diffraction In-situ synchrotron XRD experiments were performed at the GSECARS 13-BM-C beamline at Advanced Photon Source, USA, the 5 A beamline at Pohang Light Source-II (PLS-II), Korea, and the ECB P02.2 at PETRA-III, Germany. At beamline 13-BM-C, monochromatic X-ray beam with a wavelength of 0.4340(1) Å (28.568 KeV) from the bending magnet source was focused to 12 (V) × 18 (H) µm2 in FWHM using a set of KB-mirrors. A PILATUS 1 M detector was used to measure XRD data with 150 s of exposure time. At beamline 5 A, monochromatic X-ray beam with a wavelength of 0.4340(1) Å (28.568 KeV) from the undulator source was cut down to 90 (V) × 90 (H) µm2 FWHM using a pinhole. A Mar345 image plate detector was used to measure XRD data with 300 s of exposure time. At beamline P02.2, monochromatic X-ray beam with a wavelength of 0.4834(1) Å (25.650 KeV) from the undulator source was focused to 8 (H) × 3 (V) µm2 in FWHM using CRL optics. A Perkin Elmer XRD 1621 detector was used to measure XRD data with 10 s of exposure time. A symmetric-type DAC was used as a pressure vessel in combination with a membrane pressure control system. The culet diameter of the DAC used was 300 or 500 µm. A 99.9% rhenium gasket of 200 µm in thickness was pre-indented to 30–50 µm thickness, where a hole of 150 or 280 μm in diameter was made using laser drilling or electric discharge machine (EDM), respectively. A powdered sample of lizardite was initially mixed with a < 2 wt.% of gold powder (pressure marker) and loaded into the sample chamber with H2O as both a PTM and a source of water to simulate fluid-bearing cold subduction environment. The sample-to-water ratio was estimated based on the derived phase fractions of lizardite (or antigorite) and ice VII before its melting via the Rietveld method using the GSAS suite of programs74,75. Alternatively, for comparison, silicone oil was applied as a PTM to model a water-limited cold subduction environment. Simultaneous P-T conditions were achieved by resistive-heating (RH) methods using either Kanthal coil heater (at beamlines 13-BM-C and 5A) or a graphite foil heater (at beamline P02.2)80 that enclosed each diamond anvil. Temperature was monitored using a K-type or R-type thermocouple affixed to the pavilion of the anvil close to the sample chamber, which endowed maximum uncertainties of ± 3 °C. The equation of state of Au was used to calculate the pressure at a given temperature. Temperature uncertainties in our in-situ RH-DAC experiments were evaluated to be ± 35 °C below 530 °C when using the Kanthal coil heater and ± 35 °C below 530 °C and ± 90 °C between 530 and 1130 °C when using the graphite foil heater80. In-situ high-pressure, high-temperature ultrasonic wave velocity measurement In-situ ultrasonic wave velocity measurements, coupled with combined ED-XRD and X-ray radiography, were carried out using the Aster-15 MAP at the P61B beamline at PETRA-III, Germany81. At beamline P61B, the white beam from an array of 10 damping wigglers was reduced to 50 × 300 µm2 in size by adjustable slits before reaching the MAP. ED-XRD data spanning the 30–160 keV range were collected using two Germanium solid-state detectors calibrated at two different diffraction angles (3° and 5°) using an exposure time of 120 s. In order to monitor changes in the sample dimension in conjunction with the ultrasonic wave and ED-XRD measurements, X-ray radiography images were recorded with a beam size of 2000 × 1500 µm2 within 0.1 s using an X-ray microscope equipped with a sCMOS PCO Edge 5.5 camera. At each P-T condition, X-ray radiography and ED-XRD measurements were performed sequentially in about 5-min intervals. ED-XRD data analysis was performed using the PDIndexer and the GSAS II programs82. The MAP at the P61B beamline is a 6-ram Hall-type press that applies isotropic compression to a cell assembly consisting of tungsten carbide (WC) cube-shaped anvils in a 14/8 assembly, resulting in quasi-hydrostatic pressure conditions up to 10 GPa (Supplementary Fig. 11). A cylindrical graphite heater surrounding the inner components of the 14/8 assembly was used to create a high-temperature environment up to 700 °C. Temperature was monitored using a K-type thermocouple inserted above the pressure marker pellet composed of a Mo and CsCl mixture, and the equations of state for Mo and CsCl were employed to determine the pressure at the measured temperature83–85. The powdered sample of lizardite was first processed into a pellet, then placed inside a titanium capsule, followed by the addition of distilled water to simulate a fluid-bearing cold subduction environment as used in the DAC experiments. A set of 2-µm-thick Platinum foils was placed on both ends of the capsule between the titanium and the sample as a marker to measure the sample length. The capsule was then sealed using arc welding by ensuring the containment of the water medium. The ratio of water to the sample was quantitatively determined by measuring the weight change. For effective measurement of ultrasound velocity, a buffer rod (BR) was placed between the anvil and the sample assembly, while a backing plate (BP) was positioned between the pressure marker and the sample. Ultrasonic wave velocity measurements were conducted using the pulse-echo overlap method commissioned at the P61B beamline86. Fixed-frequency sinusoidal pulses generated by an arbitrary waveform generator (AFG3252C) were sent to both the 10° Y-cut LiNbO3 transducer attached to the anvil and the oscilloscope (MSO 6 series) via a high-speed relay developed at the Bayerishes Geoinstitut (BGI), University of Bayreuth. The transducer, when subjected to sinusoidal pulses, generates compressional and shear waves, which propagated within the assembly to be reflected upon encountering boundaries of different materials. The reflected waves delivered back to the transducer are converted into electrical signals by the oscilloscope. The transducer is capable of simultaneously generating a P-wave with a resonance frequency of ~50 MHz and an S-wave with ~30 MHz, but only the P-wave velocity was analyzed due to the solid–liquid nature of our sample. The data collection was automated using a Python script at the P61B beamline, analyzed using a LabView-based program called Echofinder, also developed at the BGI. Based on the time difference between two successive echoes, the two-way travel time (2t) through the sample of length l can be transformed into the P-wave (vp) and S-wave (vs) velocities using the following relations: Data visualization XRD patterns and plots in the main text and Supplementary Information were prepared using OriginPro, Version 2021b (OriginLab Corporation, Northampton, MA, USA). Supplementary information Supplementary Information Transparent Peer Review file Acknowledgements Synchrotron experiments were conducted at beamlines 5A and 3D at PLS-II, GSECARS 13-BM-C at APS, and ECB P02.2 and P61B at PETRA-III. This research used resources of the Advanced Photon Source, a U.S. Department of Energy (DOE) Office of Science User Facility operated for the DOE Office of Science by Argonne National Laboratory under Contract No. DE-AC02-06CH11357. The work performed at GeoSoilEnviroCARS (The University of Chicago, Sector 13), Advanced Photon Source (APS), Argonne National Laboratory, was supported by the National Science Foundation—Earth Sciences (EAR—1634415). We also acknowledge DESY (Hamburg, Germany), a member of the Helmholtz Association HGF, for the provision of experimental facilities and the scientific exchange and support of the Center for Molecular Water Science at DESY. Beamtime was allocated for proposals I-20220264 and I-20231080. The beamline P61B is funded by the ErUM-Pro programme (grant number 05K16WC2) of the German Federal Ministry of Research, Technology and Space. The authors thank Gerald Bauer, Sven Linhardt, and Yusuke Seto for their computer programs and technical supports regarding the ultrasound and ED-XRD experiments and Sung-Joon Chang of Kangwon National University for helpful discussion. Author contributions H.S. conducted the synchrotron experiments and the data analysis with the help of Y.B. and H.H. H.-P.L., R.F., H.H.L., T.J., D.Z., and N.G. helped experiments at synchrotron laboratories. Y.L. supervised the research and collaborated with all authors on the manuscript. Peer review Peer review information Nature Communications thanks Thomas Ferrand, Joerg Hermann and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Funding Y.L. and H.S. disclose support for the research of this work from the Leader Researcher program (grant number NRF-2018R1A3B1052042) of the Korean Ministry of Science and ICT (MSIT). The remaining co-authors declare no relevant funding. Data availability Data generated or analyzed in this study are provided in main text and Supplementary Information. Additional source data not included in the main text or Supplementary Information have been deposited in Figshare and are available at 10.6084/m9.figshare.3226708887. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-75935-4. References Associated Data Supplementary Materials Supplementary Information Transparent Peer Review file Data Availability Statement Data generated or analyzed in this study are provided in main text and Supplementary Information. Additional source data not included in the main text or Supplementary Information have been deposited in Figshare and are available at 10.6084/m9.figshare.3226708887.