Introduction

Cenozoic volcanoes are widely distributed in northeast China1, with the active Changbaishan volcanic field (CVF) being the largest among them. The CVF is well known for its Millennium Eruption at Changbai/Paektu Mountain (AD 946-947)2,3, one of the biggest in recorded history4.

The CVF is an intraplate volcanic field located ~1500 km away from the Japan Trench and 1200 km behind the volcanic arc (Fig. 1). The subducted Pacific Plate extends beneath this region in the mantle transition zone (MTZ) out to about 120°E5. Basaltic melts originating from the CVF have ocean island basalt (OIB) affinity with an enriched (EM1) component3,6. Analysis of major and trace elements suggests that these melts are derived from small-fraction partial melting of a garnet-bearing mantle source7,8. The lithosphere beneath the CVF, as well as other Neogene volcanic fields in eastern China, has a thickness of less than 100 km9, thin enough to allow a small fraction of melting of enriched upwelling mantle without requiring particularly high source temperatures. Source mantle potential temperatures are estimated to be 1300–1400 °C10, which is similar to or slightly higher than average mantle values. Water contents of primitive magmas are variously argued to be <1 wt%7 or 1.2–1.8 wt%10, significantly lower than arc magmas. This implies a mantle source region with H2O concentrations ranging between 350 and 550 ppm10, which could correspond to a variety of mantle sources without requiring significant H2O sourced from a subducting slab11.

Fig. 1: Tectonic setting and seismic activity in northeast China and surrounding regions.
Fig. 1: Tectonic setting and seismic activity in northeast China and surrounding regions.The alternative text for this image may have been generated using AI.
Full size image

a Map view of the study region (red rectangle, inset) showing earthquakes (colored dots: depth scale), active volcanoes (red triangles), and tectonic features. Slab geometry (black dashed contours: Slab2.0 model34), major block boundaries (black lines), and cross-section profiles (A–F: blue lines; 1–6: dashed light blue lines, Supplementary Fig. 15) are annotated. The Pacific Plate’s motion relative to Eurasia is indicated (blue arrow). Abbreviations: NC (North China Block), SC (South China Block), PSP (Philippine Sea Plate). b–g Vertical cross-sections of seismicity along profiles A–F in (a), with earthquake depths color-coded as in (a). Source data are provided as a Source Data file.

The formation mechanism of the CVF is currently a subject of intense debate. One widely discussed mechanism, dubbed the “big mantle wedge” (BMW) model12, proposes that mantle melting is triggered by deep dehydration of the subducting slab13 and/or upwelling of hydrous material within the mantle wedge14. However, this mechanism fails to explain why late Cenozoic volcanism (<1.5 Ma) is not widely distributed along the strike of the subduction zone, but rather concentrated in a specific region1. An alternative formation mechanism suggests that sub-slab hot materials upwell through a gap in the slab to supply magma for the Changbaishan volcano15. Recent seismic tomography studies provide for the existence of such a slab gap16,17. Nevertheless, it remains unclear how this slab gap in the MTZ is formed15,17.

In addition to the extensive intraplate volcanism and magmatism, another prominent geological feature in NE China is the occurrence of deep-focus earthquakes (300–650 km)18. Specifically, there exists an elongated cluster of deep-focus earthquakes in the MTZ ~300 km eastward from the CVF at depths ranging from 500 to 600 km (Fig. 1). The isolated nature of this cluster remains enigmatic; the potential mechanisms of deep-focus earthquake generation, including dehydration embrittlement19, transformational faulting of olivine20 and adiabatic shear instability18, are all dominantly temperature-controlled, and the thermal and chemical structure of the slab should be almost constant along-strike beneath East Asia21, so why should deep-focus earthquakes be localized here?

The spatial correlation between the CVF and deep-focus earthquakes raises the possibility that the two could be related. Zhao & Tian13 proposed that the frequent occurrence of deep-focus earthquakes near the CVF could be attributed to a locally more hydrated slab, where dehydration leads to brittle failure in the slab19 and enhanced volatile-flux melting beneath the CVF. However, based on the subduction zone thermal model in this region21, dehydration reactions of various hydrous minerals in the slab are most likely intersected at depths shallower than ~300 km22. In addition, this hypothesis contradicts the inferred presence of a metastable olivine wedge (MOW) in the core of the slab23,24, which indicates a minimal amount of water within it25. The connection between the CVF and deep-focus earthquakes, therefore, remains tenuous.

Interactions between the subducting slab and its surroundings are likely to influence the distribution of deep-focus earthquakes in the MTZ. Fukao and Obayashi26 conducted an analysis on the correlation between deep-focus earthquakes and slab morphology, suggesting that slabs stagnating within the MTZ are associated with deep-focus earthquakes limited to depths shallower than ∼620 km, while slabs penetrating into the more viscous lower mantle have events deeper than this threshold. In the Tonga subduction zone, intense seismicity has been attributed to the collision between a relic slab and the subducted Pacific slab27,28,29. More generally, high densities of deep-focus earthquakes are commonly observed in slab regions undergoing folding or more complex macro-scale deformation (e.g., ref. 30), and could be related to localized regions of high deviatoric stress31, otherwise sudden release of strain in the form of earthquakes would be prohibited32.

To better understand the origins of the slab gap in the MTZ and deep-focus earthquakes beneath NE China, as well as their potential links, we have constructed a high-resolution seismic velocity model for the subducting slab within and surrounding the MTZ to better characterize its morphology.

Results

In this study, we employed the teleseismic double-difference tomography method33 to determine the mantle structure beneath northeast China (see Methods). Our final tomographic model clearly shows westward-dipping high Vp anomalies that correlate well with the relocated earthquake hypocenters and Slab2.0 model34 (Supplementary Fig. 2), which can be interpreted as the subducting slab. Figure 2 shows cross sections of the Vp model along profiles that are nearly perpendicular to the trench. To the north (Profile A) and to the south (Profile F) of the cluster of deep-focus seismicity, the deep slab is flat and stagnant in the MTZ. Through the cluster of deep-focus seismicity (Profiles C and D), the high-velocity anomaly continues through the “660 km” discontinuity and down to ~800 km. At the margins of the cluster (Profiles B-C and D-E), there is a transition from slab penetration through “660 km” to stagnancy within the MTZ. Deep earthquakes within the MTZ primarily concentrate along Profile D. Synthetic test with a characteristic velocity model shows these features can be well-resolved (Supplementary Fig. 8).

Fig. 2: P-wave velocity structure (Vp) and seismicity beneath northeast China and surrounding regions.
Fig. 2: P-wave velocity structure (Vp) and seismicity beneath northeast China and surrounding regions.The alternative text for this image may have been generated using AI.
Full size image

a–f Vertical cross-sections of P-wave velocity perturbations (dVp; relative to the initial model) along profiles A–F (Fig. 1a). For each profile, the upper panel displays surface topography, while the lower panel shows velocity anomalies (warmer colors: slower velocities; cooler colors: faster velocities). Relocated earthquakes (gray dots, projected within 0.5° of each profile) and focal mechanisms (Global CMT catalog; red: in-plane thrust faulting; blue: in-plain strike-slip) are overlaid. Dashed black lines mark the 410 and 660 km mantle discontinuities. Source data are provided as a Source Data file.

The local penetration of the slab into the lower mantle can also be illustrated by a horizontal slice through the velocity image at a depth of 760 km, 100 km beneath the nominal “660 km” discontinuity (Fig. 3). This slice clearly depicts a high-velocity zone with the Changbaishan volcano situated around its southwestern corner and the cluster of deep seismicity (hypocenters at 500–600 km depth) aligned along its east edge (Fig. 3). Figure 4 presents a 3D iso-surface view of the imaged high-velocity anomaly. It is evident that this anomaly, interpreted as the subducting slab, extends through the “660 km” discontinuity to nearly 800 km at latitudes ranging from ~42oN to ~46oN, while remaining stagnant on both sides.

Fig. 3: P-wave velocity anomalies and inferred slab structure at 760 km depth.
Fig. 3: P-wave velocity anomalies and inferred slab structure at 760 km depth.The alternative text for this image may have been generated using AI.
Full size image

The P-wave velocity model reveals a high-velocity anomaly (blue rectangle: inferred slab penetration into the lower mantle) beneath northeast China. Black dashed lines mark major tectonic block boundaries. Active volcanoes (red triangles) and deep-focus earthquakes (white dots, depths >500 km) are overlaid. Receiver function-derived 660 km discontinuity depths from prior studies are shown (black: Tang et al.15; gray: Liu et al.38; white: Zhang et al.39; red: Yang et al.61). The 660 km boundary, a key mantle transition zone interface, is locally deflected, consistent with slab penetration dynamics. Source data are provided as a Source Data file.

Fig. 4: Slab geometry, deep seismicity, and Late Cenozoic volcanism in northeast China.
Fig. 4: Slab geometry, deep seismicity, and Late Cenozoic volcanism in northeast China.The alternative text for this image may have been generated using AI.
Full size image

Map view of the high-velocity slab interface (1% P-wave velocity anomaly) derived from the tomographic model. Relocated deep-focus earthquakes (>500 km depth) are denoted by red dots. Focal mechanisms (Global CMT catalog, indexed as in Supplementary Fig. 16) for three representative events are overlaid. Late Cenozoic volcanoes (red triangles) and regional topography highlight the spatial relationship between slab structure, deep seismicity, and surface volcanism. Source data are provided as a Source Data file.

Discussion

The Vp model obtained by teleseismic double-difference tomography33 in this study clearly reveals a north-south progression from stagnation to penetration and then back to stagnation for the subducting Pacific slab in northeast China (Figs. 2, 4). This pattern can also be seen, albeit less clearly, in the velocity images of ref. 35 and ref. 16 (Supplementary Fig. 12), although it was not mentioned or discussed in those papers. One interesting feature of our tomographic images along Profiles C and D is that the high-velocity anomaly bifurcates at around 600 km depth, with a 300–500 km long high velocity “limb” lying flat along the “660 km” discontinuity, and a longer “limb” penetrating into the lower mantle. This feature appears to be well-resolved (see Supplementary Information). Given the high viscosity of subducting slabs, we feel that this is unlikely to be “splitting” of the slab, and may instead represent one contiguous slab and a stranded slab fragment. This feature could represent the footprint of the slab morphology change from slab stagnation within to penetration through the MTZ. This high-velocity bifurcation differs from the features we observed in the Izu-Bonin subduction zone, where some high-velocity anomalies are observed in the uppermost lower mantle. Those anomalies probably represent the slow sinking of “stagnant” slab materials through the 660 km interface, which is not a perfect barrier to convection36.

It is well-known that the penetration of the slab into the lower mantle can cause depression of the “660 km” discontinuity. Several receiver function imaging studies15,37,38,39 have revealed a prominent depression of the “660 km” discontinuity in this region, which is consistent with the penetration of a cold subducting slab marked by the high-velocity belt imaged at 760 km depth at ~129oE longitude in Fig. 3. Taken together, these results give us confidence in the fine-scale features in our double-difference tomographic images.

Our study demonstrates that the previously reported slab gap by ref. 15 is the result of a narrow tongue-shaped portion of the subducting slab penetrating into the lower mantle, flanked to the north and south by stagnant sections of the slab (Figs. 4, 5). Importantly, our image of the penetrating slab tongue allows us to rule out other possibilities for a slab gap, including viscous thinning and thermal erosion, better constraining deformation mechanisms in the deep mantle.

Fig. 5: Isometric sketch of the coldest part of the subducting slab beneath NE China, based on our tomographic results and the Slab2.0 model34.
Fig. 5: Isometric sketch of the coldest part of the subducting slab beneath NE China, based on our tomographic results and the Slab2.0 model34.The alternative text for this image may have been generated using AI.
Full size image

The view is toward the SE. Black lines represent depth contours spaced 100 km apart, and blue lines are the intersection of the slab with the profiles given in Fig. 1a. The location of the deep-focus earthquake cluster mentioned in the text is marked in blue. The volcanoes of the Changbaishan volcanic field, marked by red triangles, are projected onto the 0 km depth surface. The 670 km depth surface, approximately marking the top of the lower mantle, is shown in purple. Red arrows mark potential paths of upwelling from the lower mantle through the slab gap created by the penetrating slab segment.

Tang et al.15 proposed that the slab gap beneath Changbaishan plays a crucial role in its volcanic history, because it allows hot materials to migrate upwards from the lower mantle (Fig. 5). The geodynamic modeling shows that Izanagi plate has been totally fallen into the lower mantle since ~10–20 Ma in east Asia, which can trigger the upwelling of hot lower mantle materials40. This is also supported by the existence of double “660-km” interfaces beneath northeast China41, which could be formed due to combined effects of post-spinel and post-garnet phase transitions induced by upwelling hot lower mantle materials42. Recent seismic anisotropy analysis further supports the idea of mantle flow migrating upwards through the slab by revealing a localized region of vertical flow directly beneath Changbaishan volcano43. By better resolving the location of the slab gap, our findings indicate that other nearby volcanoes, such as Jingpohu and Longgang, are also situated around this same slab gap, suggesting a shared origin (Figs. 3–5).

It is important to recognize that the length and position of the Pacific slab, and volcanic activity in the CVF have changed through time. The Pacific slab probably only reached the MTZ at ~20 Ma40 or ~15 Ma44. The earliest massive basalt eruptions in the CVF can be traced back to ~16 Ma45, probably predating the presence of a stagnant slab44. At ~15 Ma, the Pacific slab started to stagnate in the MTZ, and the sub-horizontal slab continued to advance westward44. We propose that the formation of the slab gap and associated mantle upwelling may be associated with the shield-forming stage of CVF volcanism from ~5 to 1 Ma3,6,45. We note that this interpretation is not mutually exclusive of other mechanisms of mantle upwelling, such as lithospheric dripping beneath the Songliao Basin44.

The reason for localized penetration of the slab into the lower mantle may be related to slab shape. The dip direction of the subducting Pacific Plate undergoes a rotation from north-northwestward to west-northwestward around 45°N, accompanied by a corresponding rotation in the orientation of the trench, resulting in the formation of a subduction cusp (Fig. 1). Geodynamic modeling results indicate slab penetration is favored near to this subduction cusp where the slab dip angle is the smallest and the trench retreat velocity is slowest46. Away from the cusp, the velocity of trench retreat is higher, and the slab is more likely to flatten and stagnate within the MTZ46.

The deep-focus earthquake (500–600 km) cluster is mainly distributed along the hinge and warped zone where the slab penetrates through the 660 km discontinuity (Figs. 3–5). Focal mechanisms from the Global Centroid Moment tensor catalog indicate down-dip compression and a mixture of slab-normal and along-strike extension47 (Supplementary Fig. 16). In the center of the penetrating section of the slab (Fig. 2, Profile D), most of the focal mechanisms suggest along-strike extension and down-dip compression (i.e., strike-slip motion in the plane of the slab). At the edges of the penetrating slab segment (Profiles C and E), slab-normal extension and down-dip compression is seen (i.e., thrust faulting in the plane of the slab), similar to shallower events47,48. Along-strike extension occurs within other deeply subducting slabs that are undergoing bending, such as Kermadec30. One explanation for these observations is that the slab-normal thickening promoted by resistance to flow through the “660 km” discontinuity is moderated by warping or folding of the penetrating slab segment. The dip variation of the penetrating slab tongue from steep to shallow to steep imaged in this study (Fig. 4) bears a remarkable resemblance with the shape exhibited by the Kermadec slab.

Based on the spatial correlation between the deep-focus earthquake cluster and the portion of the slab that penetrates into the lower mantle, we propose that high strain rates resulting from penetration have led to incipient bending, producing a cluster of deep-focus earthquakes. The relative absence of deep-focus seismicity to the north and south, where the slab stagnates in the mantle transition zone, suggests lower stresses and consequent strain rates in those regions. Our result provides direct support for the hypothesis that deep-focus earthquakes are limited by the rate of slab deformation in the MTZ31,49.

In the context of the physical mechanisms underlying deep-focus earthquakes, we speculate that if these earthquakes result from transformational faulting in olivine, the distribution of observed earthquakes might reflect regions where the stress/strain rates are sufficiently high to trigger subsequent thermal runaway50,51. This dual mechanism of transformational faulting and runaway provides a natural explanation for the observation of large earthquakes only in areas exhibiting tomographically imaged bending. The initiation of deep-focus earthquakes within the MTZ could be attributed to the transformational faulting of metastable olivine, while large events likely rupture beyond the wedge through the shear heating instability24,50. In warm slabs where a metastable olivine wedge cannot exist within the MTZ, shear heating instability might be the sole physical mechanism for deep-focus earthquakes52.

As shown in this study, the existence of high stress/stain rates due to slab deformation is the necessary condition for the occurrence of observed deep-focus earthquakes in the MTZ, as also supported by observations in Kermadec30, Izu-Bonin53, Tonga29, and Fuji31. For example, the 2015 Bonin M7.9 earthquake occurred near the fold hinge where the slab transitions from near-vertical to overturned, leading to large slab deformation or high local strain rates53. This could partly explain why the source region of deep-focus earthquakes is associated with high anisotropy observed in many subduction zones54, as deformation could locally enhance anisotropic fabrics in the slab55. The region of slab penetration exhibits strong, approximately NS-oriented azimuthal anisotropy at depths of 500 and 800 km56 and no radial anisotropy in the MTZ57, consistent with the orientation of fractures formed by local slab penetration through the 660-km interface. Therefore, we propose that as for shallow earthquakes, deep-focus earthquakes are induced by localized deformation and the resulting high strain rates49, as supported by recent findings that shallow and deep earthquakes have similar seismic moment release process58.

It is noted that deep seismicity clustering is only seen along the warping region of the slab tongue, and not along slab tears created when slab transits from stagnation above to penetration through the 660 km discontinuity and then reverses. This is also the case observed in the Izu-Bonin subduction zone53. This may be because shear heating and conductive heating near the slab tears render temperatures too high for brittle deformation to occur. From vertical cross-sections of seismicity shown in Fig. 1 and Supplementary Fig. 15, the local slab penetration beneath the CVF mainly affects the seismicity deeper than ~400 km based on the fact that the slab seismicity is distributed above ~400 km outside the slab penetration region.

In this study, teleseismic double-difference tomography was employed to delineate fine structures of the Pacific slab in the MTZ beneath NE China. High-resolution seismic images reveal that the slab is mostly stagnant within the MTZ, with a narrow tongue of material penetrating through the 660 km discontinuity. This distinctive change in slab morphology naturally gives rise to a slab gap within the MTZ, facilitating upward migration of sub-slab hot materials, which may subsequently trigger volcanic activity associated with the CVF, as well as potentially other volcanoes in this region (Fig. 5). The penetrating tongue appears to be undergoing incipient bending, evident from a change in slab dip and a cluster of deep-focus earthquakes with in-plane strike-slip focal mechanisms. Therefore, both Changbaishan volcanism and deep-focus earthquakes are facilitated by localized penetration of the slab beneath northeast China (Fig. 5).

Methods

Teleseismic double-difference tomography for the mantle structure in northeast China

The teleseismic double-difference tomography method utilizes both regional and global seismic stations to better resolve seismic velocity structures within and around seismically active source zones33,53. We assembled three sets of P-wave arrival-time data, including the selected ISC-EHB Bulletin dataset comprising ~770,000 P-wave arrival times from 3158 events occurring in proximity to the study region59, the selected China Digital Seismic Network dataset consisting of ~60,000 P-wave arrival times from 889 events, and the NorthEast China Extended SeiSmic Array (NECESSArray) dataset encompassing 11,000 P-wave arrival times from 145 events from September 2009 to August 201115. Overall, we gathered a total of 850,000 P-wave arrival times from 4192 events recorded by ~10,426 stations (Fig. 1 and Supplementary Fig. 1). From these absolute arrival times, a total of 3,500,000 P-wave differential arrival times are constructed with an average inter-event distance of 140 km.

The teleseismic double-difference tomography method employs a nested regional-global gridding scheme for inversion. In the target study area, the grid spacing is finer, with an interval of 1o in longitude, an interval of 1.5o in latitude, and varying intervals in depth. Within the MTZ, the grid spacing is set at 50 km in depth, which is finer than used in previously published velocity models. We utilize the LLNL_G3Dv360 as our initial global velocity model in three dimensions. The grid intervals of the global model are set to be 5o in both longitude and latitude while ranging from 55 to 225 km in depth. The checkerboard model resolution test and the model restoration test demonstrate that the model is well-resolved for depths from 160 km down to 1200 km, particularly within the MTZ (refer to Supplementary Information). Furthermore, we also test tomographic inversions with different initial crustal models and global models, and the final tomographic models are shown to be robust (Supplementary Figs. 9, 10).