Slab stagnation sustained by wet harzburgite and dry basaltic slab assemblages

Abstract

Slab stagnation in the mantle transition zone is critical for the mantle dynamics and evolution of Earth’s interior but remains poorly understood. Here we present seismic images that show two key features near Eurasia’s northeastern margin: 1) a 700-km-deep, 400-km-wide discontinuity within the slab stagnation zone, matching the akimotoite-bridgmanite transition in an anhydrous basaltic assemblage at adiabatic temperature of ~ 1000 K, and 2) the 660-km discontinuity depressed below the Clapeyron slope of the post-spinel phase transition by ~ 15 and ~ 10 km, likely induced by water, respectively in front of the Pacific plate and at the center of the slab stagnation zone. Our mineral physics modeling indicates that slab stagnation is initiated and sustained predominantly by buoyancy from the excess depressions in the pyrolite and harzburgite mantle and aided by buoyancy of the dry basaltic assemblage in the uppermost lower mantle.

Seismic imaging reveals why an oceanic slab stalls in the mantle transition zone for tens millions of years: water deepens harzburgite transitions, creating the main buoyant force, while dry basalt in the uppermost lower mantle provides secondary support.

Introduction

Subducting plates stagnant in the mantle transition zone (MTZ) have been widely observed by seismic tomography studies^1,2. This phenomenon has long been attributed to the negative Clapeyron slopes of the endothermic post-spinel phase transition, which leads to a depression of the 660-km discontinuity and local buoyancy within cold slabs³. Yet, there are also observations of subducted slabs descending straight into the lower mantle without stagnation^1,2. Adding to this paradox are recent high-resolution in-situ X-ray diffraction studies using multi-anvil apparatus, which show that the Clapeyron slopes of the post-spinel reaction are too gentle ( − 1.3 to −0.5 Mpa/K, in the temperature range of ambient mantle and slabs) to generate sufficient depression at the 660-km discontinuity and buoyancy required for slab stagnation⁴. Because of this, slab stagnation has also been attributed to either akimotoite-to-bridgmanite⁵ or garnet-to-bridgmanite+corundum phase transition⁶, which possess steeper negative Clapeyron slopes. Currently, it is debated whether slab stagnation is indeed induced by the post-garnet phase transitions^5,6, or by factors that affect the post-spinel phase transition, such as water, which may lead to downward displacement of the post-spinel phase-change boundary⁷. Other factors that may contribute to slab stagnation include upper-to-lower-mantle viscosity jumps in the ambient mantle⁸ and significant trench retreat at surface^9,10. However, the former are presumably global features and do not explain why some slabs are stagnant while others penetrate to the lower mantle. The latter does not impose direct constraints on the forces that drive slab flattening in the MTZ, nor does it explain why slabs remain in the MTZ for tens of millions of years.

The basaltic crust of a subducted oceanic slab possesses negative buoyancy in the MTZ. It becomes buoyant from the 660-km discontinuity to depths of ~720 km¹¹, or ~800 km¹², and may be gravitationally trapped at the top of the lower mantle to form a garnetite layer, which also possibly impedes penetration of the subducted slab to the lower mantle and hinders whole-mantle convection at some locations¹³. Such a scenario depends largely on whether the basaltic crust can be efficiently separated from the cold and stiff subducted slab^14,15 to form a garnetite layer.

The northeast margin of the Eurasian continent offers an excellent setting for seismic investigation of the stagnation of subducted slabs in the MTZ. Various seismic studies suggest that the modern Pacific plate has subducted from the Japan trench down to, and commenced stagnation, near the base of the MTZ^1,2,16,17. To the southwest of the Pacific subduction zone, tomographic studies also showed conspicuous high-wave-speed anomalies near the base of the MTZ, extending thousands of kilometers from the vicinity of North China to the western Yangtze^1,2 (Fig. 1). Due to their fragmented nature and lack of connection with current subduction of the Philippine Sea plate, these high-wave-speed anomalies beneath Yangtze are considered the remnants of stagnant slabs formed during the late Mesozoic subduction, between South and North China or the paleo-Pacific (Izanagi) plate¹⁸. The northeast-to-southwest variation of the slabs, therefore, contains information about the evolution of stagnant slabs over tens of millions of years. This region also has a dense collection of broadband seismic stations available from the China Digital Seismic Network and various temporary arrays (Fig. 1). The available data, illustrated by the distribution of P-to-S-wave piercing points at 410 and 660 km depths (Supplementary Fig. 1) are sufficiently dense to generate high-resolution P-wave receiver-function (RF) cross-sections across the stagnant slabs in the region.

Fig. 1. Plate tectonic setting and seismic stations used in the construction of the seismic cross-sections.

a Blue triangles denote the seismic stations processed in this study, and the white squares denote the seismic stations from the NECsaids and NECESS arrays¹⁶ (relief is from ETOPO1, https://www.ngdc.noaa.gov/mgg/global/relief/ETOPO1/data). White lines show the locations of the seismic cross-sections presented in Fig. 2. Solid color circles show the depth distribution of the seismicity with magnitudes ≥ 4.5 since January 2001 from the USGS catalog, depicting the slab geometry of the modern Pacific and Philippine Sea plates in the region. The purple‒white quadrant circles show the focal mechanisms⁵⁴ of earthquakes with depths ≥ 500 km since January 1976. Red and yellow dashed lines show modern and ancient plate boundaries. White arrows denote the modern plate motions in the International Terrestrial Reference Frame (ITRF2020)⁵⁵. b The tomographic image of the stagnant slabs (featured by bluish velocity anomalies) in the mantle transition zone(MTZ) in the full-waveform S-wave tomographic model FWEA23¹⁷. NC North China, YZ Yangtze, SC South China.

Here, we present the high-resolution seismic imaging and mineral physics modeling of the stagnant slabs in the region. We observe significant excess depressions (~10–15 km) of the post-spinel phase boundary, likely induced by water, and identify a deep discontinuity (~695 km) consistent with the akimotoite-bridgmanite transition in a dry basaltic assemblage. Our results suggest that slab stagnation is primarily initiated and sustained by buoyancy generated from these water-induced excess depressions, with an additional contribution from dry basaltic assemblages localized at the center of the stagnation zone.

Results

The 410- and 660-km discontinuities, or the top and bottom of the MTZ, which are generally linked to the olivine-to-wadsleyite and ringwoodite-to-bridgmanite (post-spinel) phase changes in the pyrolite mantle¹⁹, are clearly imaged in our cross-sections (Fig. 2). The positive and negatives interfaces that are interpreted as the top and bottom of the mantle lithosphere of the Pacific plate in a previous study¹⁶ are also observed in our cross-sections (indicated by ‘X1’ and ‘X2’, respectively, in Fig. 2).

Fig. 2. Receiver function cross-sections of the mantle transition zone(MTZ) structure across the stagnant slabs in the study region.

a–c Receiver function cross-sections along three profiles (indicated by A, B, and C in Fig. 1). Positive and negative converted phases are plotted in purple and cyan, representing interfaces with increasing and decreasing impedance with depth, respectively. The background shows the S-wave tomographic images of model FWEA23¹⁷. Focal mechanisms of earthquakes within a distance of 1° are projected onto the cross-sections. All the coordinates are referred to 117.5°E, 32°N and sea level. Dashed black lines at 410 and 660 km depths are for reference. The crossing points of the cross-sections are indicated by black arrows. For legibility, vertical scales are exaggerated by a factor of 2.85. Nomogram in the upright panel shows the appearance of true dips of 30° and 60° after the same vertical exaggeration. ak akimotoite, br bridgmanite, ri ringwoodite.

The 410-km discontinuity is relatively simple, with an average depth of ~409 km in the region, close to the global average²⁰ (Fig. 2). In contrast, the 660-km discontinuity is more complicated, with significant deepening near the center of stagnant slabs (for example, at locations indicated by ‘5’ and ‘6’ in Fig. 2) and where the subducting Pacific slab appears to be flattened (indicated by ‘4’ in Fig. 2). Outside of the areas with stagnate slabs as defined by high-velocity anomalies in the MTZ, the average depth of the 660-km discontinuity is ~659 km. Generally, a subduction process may lead to shallowing and deepening of the top and bottom boundaries of the MTZ, because the phase changes of the 410- and 660-km discontinuities have positive and negative Clapeyron slopes²¹, respectively. In addition, the topography of the 660-km discontinuity should be much more subdued than that of the 410-km discontinuity for the same thermal anomaly, because the former has a gentler Clapeyron slope⁵. Our observations of a nearly flat 410-km discontinuity at normal depth and obvious deepening of the 660-km discontinuity (Fig. 2) suggest that the thermal (and chemical) perturbations caused by stagnant slabs are significant only in the lower half of the MTZ in the region.

Basaltic assemblage concentrated at the center of the stagnation zone

Double seismic phases are observed at the upper- and lower-mantle boundary near the center of the stagnation zone (indicated by ‘1’ and ‘2’ in Fig. 2). Phase ‘1’ appears at a depth of ~670 km. Phase ‘2’ is observed at ~695 km depth (Fig. 2a), beyond the depth range allowed for the post-spinel phase transition⁵, but consistent with the occurrence of the transition from akimotoite to bridgmanite in a relatively cold region^5,22. Both phases can be simultaneously and consistently observed on the non-stacked receiver functions derived from earthquakes at various epicentral distances and back-azimuths (Fig. 3 and Supplementary Fig. 2). They have moveout curves similar to that of P660S, but different from those of major known body-wave-generated noise, involving phases PcP, PP, pP and sP. These observations indicate that the double phases are unlikely to be an artifact of stacking derived from the effects of unaccounted seismic anisotropy, inhomogeneity along the ray paths, or the contamination of these other-body-wave-generated noises. Other phases at similar depths of ~700 km (indicated by ‘7’ and ‘8’ in Fig.2) are either too weak or near the margin of our data coverage to be confidently interpreted.

Fig. 3. Receiver functions (RFs) for the double phases ‘1’ and ‘2’ near the base of the mantle transition zone (MTZ).

a The RFs plotted in a time-distance format. Gray curves show the theoretical arrival times of known phases (PcP, PP, pP and sP) relative to P for earthquakes of different focal depths (indicated by numbers in brackets) and epicentral distances as predicted by the model IASP91⁵⁶. Phases ‘1’ and ‘2’, indicated by brown and cyan arrows, have moveout curves that are similar to those of P410s and P660s (indicated by green and brown curves) but different from those of the above-mentioned body waves. b The RFs stacked by back-azimuth, showing that the double phases can be consistently observed at different back-azimuths. c Distribution of the back-azimuths and the epicenters (indicated by small red circles) of the RFs observed for the double phases.

Based on the depths observed for phase ‘2’ and the Clapeyron slope curve for the akimotoite-to-bridgmanite phase transition⁵, we obtained a mantle temperature of ~1150 K (or a mantle adiabatic temperature of ~1000 K), near the site of the double phases (Fig. 4a and Supplementary Fig. 3a). Our mineral physics modeling with the Perple X code²³ and the elastic parameter database of ref. ²⁴ show that this estimated temperature is consistent with the appearance of the akimotoite to bridgmanite phase transition at a depth ~25 km below the 660-km discontinuity (Supplementary Fig. 4).

Fig. 4. Predicted and observed depths and receiver function (RF) amplitude ratios of the seismic phases.

a Phase transitions near 660 km depth predicted by recent in situ X-ray diffraction studies^5,6. Color-filled circles show the depths observed for phases ‘1’, ‘2’, ‘5’ and ‘4’, plotted on the 1000 K and 1560 K mantle adiabats, respectively. b Theoretical variations of RF-amplitude ratio of phase ‘2’ to phase ‘1’ with basalt fraction (f) along the 1000 K mantle adiabat. Yellow-filled circles are the ratios observed along profile A in Fig. 2. ak akimotoite, br bridgmanite, ri ringwoodite, fp ferropericlase, gt garnet, cr corundum.

Theoretically, an increase in the basalt fraction leads to a decrease in the RF amplitudes from the 660-km discontinuity²⁵ and an increase in the RF amplitudes from the akimotoite-to-bridgmanite phase-change boundary (Supplementary Figs.  5 and 6). Indeed, phase ‘1’ has lower amplitude than the 660-km discontinuity in the rest of the study region and phase ‘2’ (Fig. 2). Following ref. ²⁵, we used the RF amplitude ratios of phase ‘2’ to phase ‘1’, which eliminate the effects of ray parameters (Supplementary Fig. 6), to estimate basalt fractions (f), and obtained f = ~0.6 to ~0.9, which suggests a high concentration of basaltic contents in the site of the double phases (Fig. 4b). Such a high concentration is likely formed from, 1) the results of basaltic crust detached from¹⁶, or peeling off²⁶, the top of the plate, or 2) the results of subducted basaltic ridge trapped at the top of the lower mantle²⁷. Although geodynamic modeling has not demonstrated an easy separation of the basaltic crust at MTZ depths^14,15, the first mechanism cannot be ruled out. The stagnant cold slabs may become substantially weaker due to grain-size reduction associated with the phase transformations in the MTZ²⁸, and cold, wet slabs are most susceptible to transition-induced weakening, with a transient weakening of up to three orders of magnitude during the olivine to ringwoodite phase transitions²⁹. Such an interpretation is also consistent with the observation of an aseismic feature in the MTZ near the center of the stagnation zone (Figs. 1, 2). This place is also near the proposed location of the extinct Izanagi-Pacific spreading ridge that merged with the trench of the long-lasting subduction zone along the eastern margin of Asia from ~60 to 50 Ma³⁰. Therefore, the second interpretation cannot be ruled out either. Nevertheless, the lack of observation of the akimotoite-to-bridgmanite phase boundary elsewhere suggests that the basaltic contents are relatively low under most of the 660-km discontinuity (f < 0.4 and Supplementary Fig. 6).

Excess depressions imaged in the post-spinel transitions

Some of the observed depths of the 660-km discontinuity near the center of the stagnation zone (for example, ~678 km, at the place indicated by ‘5’ or ‘6’ in Fig. 2) exceed the maximum depth limit (~668 km) of the Clapeyron slope curve for that phase change by ~10 km (Fig. 4a and Supplementary Fig. 3a) and cannot be accounted for by under-correction of velocity inhomogeneity (under-correction of upper-mantle low-velocity anomalies may contribute to deepening of both the 410- and the 660-km discontinuities, while under-correction of high-velocity anomalies in the MTZ may contribute to an apparent shallowing rather than deepening of the 660-km discontinuity; see Methods). Similarly, at a depth of ~683 km, phase ‘4’ appears deeper than the limit expected for the normal post-spinel phase change by ~15 km (Fig. 4a). Similar depressions (~20–30 km) were observed previously in a narrow region along the strike in the front of the Pacific subducting plate hundred(s) kilometers north of our profiles B and C^31,32. One interpretation is to associate this phase with the akimotoite-to-bridgmanite phase change at a mantle temperature of ~50 K higher than that of phase ‘2’ (Supplementary Fig. 3a), and attribute the disappearance of the 660-km discontinuity to an extreme concentration of basaltic component (f ≈ 1.0; cf. Supplementary Fig. 6a), or a coincidence of the negative phase ‘X2’, at the location ‘3’ (Fig. 2a). The second interpretation is to associate this phase with the post-spinel phase change and attribute the observed deepening of the boundary to a delay in the phase transition due to kinetics of the mantle under the inflection point of the subducting Pacific slab. This requires downwelling of the mantle beneath the slab deflection point and a dry mantle condition that maintains slow reaction rates, both of which are unlikely in this region. The third and our preferred interpretation is to associate this phase with the influence of minor elements (e.g., Fe and Al)^21,33 or volatiles (e.g., water)^7,34. In contrast to Fe, which may shift the post-spinel transition boundary to lower pressure²¹, an increase in Al content may shift the post-spinel transition to higher pressure³³. However, the depression effects of Al may be largely counteracted by Fe³⁴, due to the strong coupling between Fe and Al in perovskite³⁵, resulting in no depression of the post-spinel discontinuity as observed at location ‘1’, where the basaltic slab assemblage (enriched with Fe and Al) dominates. Therefore, the minor elements Fe and Al may not be the primary factors for the excess depression. Association of the depression with water is consistent with recent experimental studies that suggest hydrous conditions are necessary for the observation of phase ‘X2'¹⁶. Previous seismic waveform modeling³⁶ and electromagnetic studies³⁷ suggest that the western Pacific subduction zone is particularly wet, with water contents as high as ~0.5–1.0 wt% in the MTZ³⁷. Petrological experiments⁷ indicate that ~2.0 wt% water in ringwoodite may shift the phase change boundary downwards by ~15 km at 1473 K, and ~30 km at a cold mantle temperature of ~1000 K. Therefore, the excess depression of ~15 km at the location ‘4’ can be explained by the existence of ~2.0 wt% water near the bottom of the MTZ. Similarly, the extra deepening of ~10 km at the location ‘5’ or ‘6’ can be explained by the existence of ~0.7 wt% water, given the nearly linear relationship between excess depression of the post-spinel phase transition and water content³⁸.

We note that no significant excess depression of the 660-km discontinuity at the location ‘1’ (Fig. 4a) suggests that the water content is low near the center of the basaltic mantle assemblage. This may be due to a relatively low water solubility of majorite garnet in the MTZ³⁹ and the transition-zone water filter⁴⁰. The basaltic mantle assemblage in the uppermost lower mantle is depleted of water due to the transition zone filter effect and buoyant (see Fig. 5b). A possible consequence of such local buoyancy is the stagnation or rise of the basaltic mantle assemblage, resulting in a dry post-spinel phase boundary.

Fig. 5. Theoretical densities (ρ) and density contrasts (Δρ) for the stagnant slabs (1000 K) and surrounding mantle (1560 K).

a Theoretical variations of densities with depth of the oceanic basalt crust (f = 1.0; in green) and harzburgite mantle lithosphere (f = 0.0; in cyan) of the slabs and the pyrolitic surrounding mantle (f = 0.2; in red), respectively along the 1000 K and the 1560 K mantle adiabats (cf. Supplementary Fig. 4). b Theoretical density contrasts of the basalt crust and harzburgite mantle lithosphere of the slabs (1000 K) with respect to the pyrolitic surrounding mantle (1560 K). c,d Theoretical density contrasts of the harzburgite slab (1000 K, assuming a thickness of 100 km) with respect to the pyrolitic surrounding mantle (1560 K), without and with an excess depression of 10 km in the phase transition, respectively. R stands for the ratio of the buoyancy to the negative buoyancy. e Theoretical density contrasts derived from the depression of the post-spinel phase-transition in the pyrolitic surrounding mantle (1560 K). The effects from adiabatic thermal expansion due to the downward displacement are ignored in (c–e). Brown and cyan shaded areas are proportional to the upward and downward buoyancies. ol olivine, wa wadleyite, ak akimotoite, br bridgmanite, ri ringwoodite, fp ferropericlase, gt garnet.

Slab stagnation explained by excess depressions of the transition

To understand the dynamics of slab stagnation in the region, we calculated the density contrasts of a cold slab with respect to the pyrolitic surrounding mantle (Fig. 5 and Supplementary Fig. 7). Our results show that buoyancy due to density contrasts derived from thermal and petrological difference (including the normal anhydrous post-spinel and post-garnet phase transitions) is relatively small, and present only between ~660 and ~720 km depths (indicated by brown area in Fig. 5b), no matter whether basalt fraction f is 0 or 1 (that is, no matter whether the akimotoite-to-bridgmanite or garnet-out phase boundaries is visible or not). We also modeled the contribution to the buoyancy from the (excess) depression of the post-spinel phase-transition boundary beneath the center of the stagnation zone (Fig. 5d) and at the front of the subducting Pacific plate (Fig. 5e). Our results show that the buoyancy increases drastically with the excess depression (cf. Supplementary Fig. 7c–e), and the ~10 km excess depression near the center of the stagnation zone can generate a significant increment of buoyancy (see brown areas in Fig. 5c, d). Without the contribution from this excess depression, the thermo-chemical buoyancy of the slab with respect to the pyrolitic surrounding mantle (indicated by brown area in Fig. 5c), is much lower than the negative buoyancy of the slabs in the MTZ (indicated by cyan area in Fig. 5c), with a positive-to-negative buoyancy ratio (R) of only ~27.5%, assuming the slabs above the 660-km discontinuity have an average thickness of ~100 km as estimated from the distance between phases ‘X1’ and ‘X2’. The excess depression of 10 km increases the ratio R to ~97.5%. Therefore, together the buoyancy from the excess depression of the 660-km discontinuity and the normal thermo-chemical buoyancy of the slabs are approximately in balance of the downward force of the slabs in the MTZ, maintaining slab stagnation (Fig. 5d). In the front end of the subduction plate, the buoyancy derived from the total depression of ~25 km (relative to the normal mantle outside of the slab stagnation and subduction zone, including the excess depression of ~15 km) of the post-spinel phase boundary (proportional to the brown shaded area in Fig. 5e) is, however, much smaller compared to the overriding burden of the subducting Pacific plate that possesses large positive density contrasts above the 660-km discontinuity (denoted by the cyan shaded area in Fig. 5b, assuming the plate is composed of 10% basaltic crust and 90% harzburgite mantle lithosphere). In light of the observed bending of the plate, the gentle dip (~25°), and rheological weakness of the cold slabs due to small spinel grain-size²⁸, we infer that the upward phase buoyancy, together with the mantle viscous resistance⁸, provides the bending moment to deflect the subducting plate to commence stagnation in the MTZ (Supplementary Fig. 8).

Discussion

Our findings suggest that the excess depression of the post-spinel phase-change boundary, most likely induced by water, plays a dominant role in slab flattening and stagnation in the MTZ. In comparison, the normal thermally induced post-spinel and akimotoite-to-bridgmanite⁵ phase transitions combined contribute no more than half of the buoyancy required for the stagnation (cf. Fig. 5c, d). Our results suggest only a localized high (f = 0.6–0.9) concentration of basaltic content in the center of the stagnation zone, within a 400-wide zone near the upper-lower mantle boundary. Our observations do not support the presence of a widely distributed garnetite layer that plays a major role in slab stagnation¹³ in this region.

Our observations provide an explanation for the long-term (likely >tens Ma) presence of the slab stagnation zone in the region. With a higher water solubility than the upper and lower mantle^41,42, the MTZ is capable of preserving water⁴⁰ and thus the excess depressions of the 660-km discontinuity, for an exceptionally long (>1Gyr) period⁴³. Although tectonic reconstruction and numerical modeling show that the history of the Izu-Bonin trench is dominated by fast, episodic retreat from 40 to 6 Ma⁴⁴, which may have resulted in the gentle subduction that facilitated the stagnation¹⁰ of the Pacific plate in the MTZ in the earlier Cenozoic, trench retreat, by itself, does not explain the forces required for the tens-of-millions-of-years stagnation of the Mesozoic slabs thousands of kilometers away to the southwest, nor the forces at the inflection point of the Pacific slab where it is deflected horizontally.

Methods

Seismic data and construction of receiver-function cross-sections

We collected seismic data from broadband seismic stations in China and adjacent regions, including permanent stations from the China Earthquake Administration⁴⁵, temporary stations from ChinArray⁴⁶, NECsaids (the Northeast China Seismic Array to Investigate Deep Subduction), NECESSArray (the Northeast China Extended Seismic Array)¹⁶ and NETS (NorthEastern Tibet Seismic) experiments⁴⁷. We processed waveform data from teleseismic events with magnitudes larger than 5.0 and epicentral distances ranging from 30 to 90°. We used the time-domain iterative deconvolution technique⁴⁸ in the calculation of the RFs, in which the Gaussian filter factor was set to 2 to suppress high-frequency ( ≥ 1 Hz) noise. We visually examined all waveform data, corrected for sensor misorientation and discarded data with abnormal waveforms and low coherence. We finally obtained a total of 65,655 high-quality RFs from 1339 seismic stations.

P-wave receiver function (RF) technique is an effective tool for imaging slab interfaces⁴⁹ and mantle discontinuities down to the MTZ depths¹⁶. We implemented a procedure similar to the common-conversion-point method but without the plane-wave assumption; RFs were migrated from time to space in a 3-D volume via dynamic ray tracing⁵⁰ from the sources to the seismic stations according to the Earth reference model AK135 (cf. Supplementary Fig. 1c) and then corrected for the Earth’s crust, ellipticity and station elevation. To improve depth accuracy, we used seismic tomographic models to correct for 3-D velocity inhomogeneity along the ray paths.

Let subscripts i and j index the ray segments along the raypaths of the P-to-S converted phase and the direct P phase, respectively. For each segment, let Δt denote the reference-model travel time through that segment, and let δm denote the relative velocity perturbation in the tomographic model (δVp/Vp for P wave, δVs/Vs for converted S wave). The delay time of the S-wave converted at a mantle interface can be theoretically obtained by:

$${\mathit{t}}_{\mathit{ps}}={\sum }_{\mathit{i}}^{\mathit{ray}\mathit{Ps}}\frac{{\mathrm{\Delta }\mathit{t}}_{\mathit{i}}}{\mathbf{1}+{\mathrm{\delta }\mathit{m}}_{\mathit{i}}}-{\sum }_{\mathit{j}}^{\mathit{ray}\mathit{P}}\frac{{\mathrm{\Delta }\mathit{t}}_{\mathit{j}}}{\mathbf{1}+{\mathrm{\delta }\mathit{m}}_{\mathit{j}}}=\left[{\sum }_{\mathit{i}}^{\mathit{ray}\mathit{Ps}}{\mathrm{\Delta }\mathit{t}}_{\mathit{i}}-{\sum }_{\mathit{j}}^{\mathit{ray}\mathit{P}}{\mathrm{\Delta }\mathit{t}}_{\mathit{j}}\right]+\left[{\sum }_{\mathit{i}}^{\mathit{ray}\mathit{Ps}}\frac{{-\mathrm{\delta }\mathit{m}}_{\mathit{i}}}{\mathbf{1}+{\mathrm{\delta }\mathit{m}}_{\mathit{i}}}{\mathrm{\Delta }\mathit{t}}_{\mathit{i}}-{\sum }_{\mathit{j}}^{\mathit{ray}\mathit{P}}\frac{{-\mathrm{\delta }\mathit{m}}_{\mathit{j}}}{\mathbf{1}+{\mathrm{\delta }\mathit{m}}_{\mathit{j}}}{\mathrm{\Delta }\mathit{t}}_{\mathit{j}}\right]$$ 1

The items in the first and second brackets correspond to 1) the time-to-space migration along the ray-path for a spherically symmetric Earth, and 2) the travel-time correction accounting for 3-D velocity inhomogeneity in this study.

The final seismic cross-sections are obtained by projecting and stacking all the corrected RFs within 1° of the cross-sections onto the vertical plane. Our final cross-sections have a bin size of 25 × 0.5 km along the cross-section and the depth directions, respectively.

Validation of the excess depressions by different tomographic models

We used three different tomographic models, TX2019slab⁵¹, UU-P07⁵² and FWEA23¹⁷, to correct for the 3-D velocity inhomogeneity and validate the observation of the excess depressions. TX2019slab is a global travel-time tomographic model that incorporates a 3-D thermal model of subducting slabs in the inversion. Since the crustal anomalies were removed as station statics and not included in TX2019slab, we performed a crustal correction to the RFs when using this model. FWEA23 is a regional full-wave tomographic model for East Asia. It is worth noting that the delay-time anomalies derived from excess depression of the discontinuity may be partly explained by the velocity perturbations above the discontinuity in the inversion of full-waveforms that contain converted waves (such as P410S and P660S), especially in places that lack crossing rays, such as the east margins of our profiles. In addition, FWEA23 has better resolution for the S-wave than for the P-wave. All these may lead to an over-correction and a more flattened discontinuity observed in the RF profiles, though whether padded with a global model or not does not have an observable influence on the results. In contrast, UU-P07 is a global high-resolution P-wave tomographic model. Previous tomographic models^17,51 show that the S-wave velocity anomalies are similar in pattern to, but about 1.5 ± 0.5 times of, the P-wave anomalies (with correlation coefficients of about 92% and 87% between the P- and S-wave anomalies in FWEA23 and TX2019slab, respectively). Therefore, it is possible to use UU-P07 to correct for the 3-D velocity inhomogeneity by assuming a scale factor between the P- and S-wave velocity anomalies. Indeed, assuming a larger scale factor may lead to an overcorrection, and a discontinuity to be imaged shallower than its true position, due to the fact that the tomographic model is dominated by large-scale low-velocity anomalies in the upper mantle and some relatively small high-velocity anomalies in the MTZ^17,51(Supplementary Fig. 3a).

Our sensitivity tests using the three models show that our key results for the excess depressions of the 660-km discontinuity remain unchanged when corrected with different models (Supplementary Fig. 3a). For example, phases ‘1’ and ‘2’ appear at similar depths. Though the maximum depression center changes from location ‘5’ to ‘6’, the maximum depression magnitude remains unchanged when we corrected with FWEA23 instead of TX2019slab and UU-P07. A notable difference is observed for phase ‘4’, which appears at much shallower depths (though still much below the depth limit of anhydrous post-spinel Clapeyron curve) if corrected with FWEA23 and possibly a consequence of over-correction. In short, using different tomographic models did not alter our core conclusions. The receiver functions shown in Fig. 2 were corrected using TX2019slab, which is independent of the converted waves.

Mineral physics modeling of post-spinel phase transition

We computed the effects of temperature and composition on mineral phase equilibria, and the depth variation of seismic velocities and densities with the Perple X code²³ and the elastic parameter database of ref. ²⁴. We consider the subducted slab to be either an equilibrium assemblage (EA) or a mechanical mixture (MM) of basalt (i.e., oceanic crust) and harzburgite (i.e., oceanic mantle lithosphere). The mechanical mixture is a good approximation when the re-equilibration rate of basalt and harzburgite in the slab is considered to be slow⁵³, and is more convenient to be used for comparison under the first-order approximation. The end member compositions basalt and harzburgite are from ref. ⁵³ (see Supplementary Table 1).

Based on thermodynamic relations, water incorporation into ringwoodite can expand its stability field to higher pressures relative to anhydrous system⁷. We modeled such delays in the phase transitions by shifting the relevant velocity and density curves downwards by the corresponding depths, irrespective of which minor element indeed caused the additional (or excess) depression of the discontinuity.

Author contributions

D.S. and Y.S. conducted the study, performed the seismological analysis and wrote the paper. D.S., R.Z. and J.S. performed the data processing. D.S. and H.X. performed the mineral physics modeling. C.C., X.X., Y.Z., Y.X., Y.L. and Y.Y. contributed to the data collection and improvement of the manuscript.

Peer review

Peer review information

Nature Communications thanks Zhouchuan Huang, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

D.S. acknowledges funding from the National Natural Science Foundation of China [grant number 42174109] and the National Key Research and Development Program of China [grant number 2016YFC0600201]. Y.S. acknowledges funding from the U.S. Natural Science Foundation [grant number 1927133]. R.Z acknowledges funding from the China Geological Survey [grant number DD20221829]. Y.L. acknowledges funding from the National Natural Science Foundation of China [grant number 42174169].

Data availability

The receiver function data generated in this study have been deposited in Figshare (https://figshare.com/s/9551770029e23ef9257f). The receiver function data from the NECsaids and NECESS arrays are available from World Data System for Geophysics at ftp://159.226.119.161/data/NECsaids/RF2020. Raw seismic waveform data from CDSN and CHINArray are available from the DMCs of China National Seismic Network at Institute of Geophysics, China Earthquake Administration (http://www.seisdmc.ac.cn, http://www.chinarraydmc.cn). Raw waveform data from the stations of the WDSN and the northeastern Tibet seismic experiments are available from the IRIS DMC.

Code availability

The seismic ray tracing program used in the construction of the RF images is available at http://geodynamics.org/cig/software/packages. The thermodynamic code Perple X used for mineral physics modeling is available at https://www.perplex.ethz.ch. The GMT software package used for production of most figures is available at https://www.generic-mapping-tools.org. All the specific data processing codes and scripts are available from the corresponding authors.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-75639-9.