Introduction

The Moho discontinuity, defined by a large increase in seismic wave speed and accompanied by increased density, is commonly assumed to represent the petrological crust-mantle boundary. Near-Moho seismicity is rare in most continental lithosphere, but is well-known across the Tibetan plateau, where competing claims as to whether these earthquakes are in the crust or in the mantle stimulated development of the “jelly-sandwich”1,2 and the “crème-brulée”3 models. The former suggests an aseismic lower-crust above a brittle upper-mantle, while the latter argues all earthquakes occur within the crust4. Large uncertainties in earthquake depths inherent in time-to-depth conversions have been overcome by modern methods including S-minus-P delay time comparison5,6, fitting mantle-converted phases7,8, and Sn/Lg amplitude ratio analysis9,10, so that the existence of Tibetan sub-Moho earthquakes now seems incontrovertible.

Both the jelly-sandwich and the crème-brulée models rely on the conventional understanding that earthquakes occur where sufficiently low temperatures and sufficiently high strain rates allow accumulation of sufficiently high stresses to overcome rock strength and cause brittle failure. Earthquakes in the mantle occur in low-strain-rate intra-plate settings only where colder than ~ 600 °C11, but this temperature cutoff increases by ~ 100 °C as strain rates increase by ~ 2 to 2.5 orders of magnitude in tectonically active regions12. Earthquakes occurring above 600 ℃ have been recognized in both oceanic11 and continental upper mantle13,14. Where continental mantle earthquakes appear to have nucleated deeper than the conventional brittle-ductile transition (at low strain rates and high temperatures) appeal has been made to self-localizing thermal runaway mechanisms13,14. However, thermal runaway also requires large differential stresses15 not dissimilar from the stresses required to produce brittle failure. Here we show that brittle behavior can explain the Himalayan sub-Moho earthquakes without appealing to more exotic explanations. We focus on defining the tectonic controls and lithospheric conditions that permit the observed seismicity.

The Himalayan sub-Moho earthquakes are so concentrated in south Tibet and in northwest Tibet, in each area spanning < 300 km along strike of the > 2000 km long orogen, that they cannot be fully explained by a single, orogen-parallel model like slab bending16 or underthrusting cold cratonic Indian mantle17. Hypotheses offered to explain the localized distribution of these sub-Moho earthquakes include lithospheric faults causing seismicity within the ultramafic upper mantle18, and delamination causing seismicity within eclogitized lower crust19,20. Our claim of localized sub-Moho earthquake clusters here assumes stationarity of seismic behavior, even though our observations span a far shorter time than the Himalayan seismic cycle (usually centuries). Nonetheless, the localization of below-Moho clusters is supported by teleseismic re-locations1,2,17, temporary-array recordings5,7,21, and statistical seismic analyses10 across the Himalaya and Tibet. The most parsimonious interpretation is that the Himalayan sub-Moho earthquakes are truly spatially restricted. Localization of seismicity shows that traditional rheologic models (e.g. jelly sandwich and crème brûlée) do not adequately capture rapid lateral variations in lithospheric viscosity.

In south Tibet, the penetrating-fault hypothesis draws support from the shape of the deep seismicity cluster, elongated SE-NW along strike of the Dhubri-Chungtang fault (DCF) from the Shillong plateau to the Himalaya. Many of the sub-Moho focal mechanisms in this area also match DCF’s dextral strike-slip faulting in its Shillong segment18,22 (Fig. 1). Some deep extensional focal mechanisms instead suggest a Moho-penetrating extension of the southern Pumqu-Xainza graben (PXG)19. We argue below that the strain rates on the DCF and PXG are insufficient to embrittle ultramafic upper-mantle at typical Tibetan mantle temperatures (≳600 ℃)23.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

(a) Tectonic elements of south-central Tibet and Himalaya with SKS splitting distributions24 (short solid and dashed bars). Shear-wave splitting measurements 50 km (solid circle) to 230 km (dashed circle) from the proposed eclogite drip center have their splitting directions colored by angular discrepancy from the radial direction, red (0–30º), orange (30º–60º) and purple (60º–90º). More-distant splits are plotted as blue dashed bars. Dashed green curve is southern limit of the solid green lines marking receiver-function profiles with observed eclogite doublets6,19,25,26,27. Black stars and dots are Sn/Lg-studied earthquakes10 with well- (stars) or poorly- (black dots) constrained source depths. Blue dashed line encompasses all earthquakes plotted on cross-sections. MFT: Main Frontal Thrust. STD: South Tibetan Detachment. YZS: Yarlung-Zangbo Suture. PXG: Pumco-Xainza graben28. DCF: Dhubri-Chungthang fault22. Fine grey lines: other faults. Green overlay: DCF, including possible extension beneath the STD. Red overlay: proposed eclogite drip. (a1) Histogram of splitting directions by their deviation from radial-to-the-drip-center. (a2) Location map. White lines: plate boundaries. Red frame: region in (a) and (b). (b) Moho depth beneath south-central Tibet and Himalaya29,30, focal mechanisms21,22,31,32, and S-P confirmed sub-Moho earthquakes (squares6 and diamonds5) are colored by their depths. Dots and stars as in (a), but colored by Sn/Lg ratios10. Magenta dashed lines: profiles \(AA{\prime}\) (N33ºE, ~ orthogonal to DCF) and \(BB{\prime}\) (N123ºE, ~ parallel DCF). (c,d) Cross-sectional views along \(AA{\prime}\) and \(B{B}{\prime};\) seismicity colored by Sn/Lg ratios10; legends as in (a) and (b) (note, overlapping red eclogite drip and green DCF appears brown). Magenta dashed lines: receiver-function Moho29. Purple dashed lines: local deeper receiver-function converters 6. MHT: Main Himalayan Thrust33.

An alternate to the penetrating-fault hypothesis introduces the foundering of lower crust, through drip or delamination, into the upper mantle and restricts the sub-Moho earthquakes (and brittle deformation) to crustal materials10. Delamination sensu stricto is broadly planar in 3D, characterized by the peeling or rolling back of a layer. Dripping (often labelled as delamination, sensu lato) is typically shown as cylindrical but can also be planar (in this terminology, both a salt wall and a salt stock are positively buoyant drips34), characterized by inward flows surrounding the drip source35. The Vrancea deep seismic zone, for example, has been interpreted as a lower-crustal drip36. Both dripping and delamination can be triggered by over- or underlying structure, like the mantle-lithospheric delamination or dripping often suggested to follow break-off of the Indian slab beneath Tibet37,38. The basic condition for crustal foundering is negative buoyancy, simply achieved by the lower-crustal eclogitization (densification) widely inferred beneath southern Tibet from the presence of a receiver-function doublet5,19,25,39. These eclogite layers have been suggested to have a critical connection with sub-Moho earthquakes20,40,41. Mafic lower crust that is transformed to eclogite becomes denser and stronger than ultramafic upper-mantle so may initiate foundering42 while also remaining brittle to higher temperatures than either parent mafic lower crust or ultramafic upper mantle43.

Distinction between dripping and delamination of the eclogitized lower crust could come from precise earthquake locations, from shear-wave splitting, or from high-resolution tomography. Even the best existing tomography is equivocal as both modes have been inferred38,44,45. However, our sub-Moho seismicity is tightly clustered as if defining a cylindrical drip rather than an extended planar feature (whether a drip or delamination sensu stricto). Further, we observe radial SKS splitting directions surrounding the cluster center24, consistent with inward mantle flow characterizing a cylindrical lower-crustal drip (Fig. 1a). Finally, a deeper and larger foundering of mantle lithosphere (whether dripping or delamination; both have been inferred from tomographic images of Tibet38,44,45) beneath the eclogite layer would create a mantle flow-field that expedites eclogite foundering by dripping rather than delamination (Sect. “Temporally constrained viscosity of a Rayleigh–Taylor instability”). Dripping is therefore our preferred mechanism for eclogite foundering. Local thinning of the receiver-function lower-crustal doublet19 and some localized receiver-function deep converters6 are possible seismological evidence for this mechanism. Localized surface rebound, a consequence of completed foundering in the absence of compensating lower-crustal flow46,47, is not observed in south Tibet, but the continuity of the deep seismicity from near-Moho to ≳100 km suggests that any eclogite drip has not yet detached from the lower crust.

In this study, we aim to discriminate between the penetrating-fault and eclogite-foundering hypotheses through the analysis of multiple data constraints and re-evaluation of Tibetan rheology. We first estimate strain rates and temperatures within the sub-Moho seismic zone, then build lithospheric rheology profiles to identify lithologies that can remain brittle (seismogenic) to ≳100 km. Our geodynamic simulation of Rayleigh–Taylor instability48 constrains the viscosity range that permits dripping to ≳100 km in the available ~ 10 Myr timescale constrained by Himalayan geological history. We show that eclogitized lower crust meets temperature and viscosity criteria to remain seismogenic to ≳100 km, but to meet temporal constraints, it likely requires water input that could be facilitated by crustal-scale faulting transecting the eclogite layer and Moho.

Physical conditions in sub-Moho seismogenic zone

Sub-Moho seismicity provides a window into present-day temperatures and viscosities beneath the Indo-Eurasian continental plate convergence. For simplicity, we assume that the stress and strain condition remains approximately fixed with respect to the topographic crest of the Himalaya, and affects different rocks at different times.

The earthquakes span ~ 50–100 km across the DCF (N53ºW, Fig. 1c), ~ 200 km along strike (Fig. 1d), and from ~ 50–110 km vertically (Fig. 1c,d). They lie ~ 100–200 km north of the Himalayan thrust front, where temperatures range from ~ 400–800 °C23 (Fig. 2a). Lithologies of the deep earthquake zone likely include pyroxene-plagioclase rocks (diabase/gabbro or mafic granulite), garnet-pyroxene rocks (eclogite, transformed from mafic lower crust by high-pressure phase change), and olivine-dominated rocks (ultramafic upper mantle)4,39.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

(a) Default model set-up. (b) Default model after 20-Myr. Orange-to-red curves: evolution of Moho (below eclogite); light- to dark-blue curves: evolution of lithosphere-asthenosphere-boundary (LAB). Velocity field is a snapshot at 10 Myr. (c–e) Regime diagrams of drip length \({d}_{e}\) (colored circles) as a function of \({\eta }_{m}\) and \({\eta }_{e}\) for (c) \({\rho }_{e}\) = 3400 kg/m3 after 20 Myr; (d) as (c) but after 5 Myr; and (e) \({\rho }_{e}\) = 3300 kg/m3 after 20 Myr. Green curve marks \({d}_{e}\) = 40 km (drip reaches 110 km depth). Blue circles: dripping too slow to reach 110-km depth; red/orange circles: drip, and consequent seismicity, reach 110 km depth. Forbidden zone: eclogite weaker than ultramafic mantle.

The sub-Moho hypocenters cluster beneath the intersection of the dextral DCF and extensional PXG (Fig. 1). The geodetic slip rate of DCF is \(\dot{s}\) = 0.4–0.8 mm/yr49; the geochronological slip rate across PXG is \(\dot{s}\) ~1 mm/yr50. Under the penetrating-fault hypothesis, shearing of the DCF or opening of the PXG span the sub-Moho seismicity orthogonal to these faults for a distance \({L}_{h}\) ~100 km, so the characteristic seismogenic strain-rate is \(\dot{\epsilon }\) = \(\dot{s}/{L}_{h}\) ≈ 0.3 × 10–15 s–1. Strain rates will be faster if shearing is accomplished in a high-strain zone narrower than \({L}_{h}\), so we take \(\dot{\epsilon }\) ~ 1 × 10–15 s–1 as an order-of-magnitude estimation.

However, in the area of sub-Moho seismogenesis, receiver-function images5,19 support a \(H\) \(\lesssim\) 20 km thick eclogite layer above a ~ 70 km Moho. A Rayleigh–Taylor instability or eclogite drip may have been triggered by the last major slab breakoff/rollback event in the southern Lhasa terrane at ~ 20 Ma that is indicated by the consequent ~ 18–8 Ma adakitic magmatism37 characteristic of melting of delaminated garnet-bearing lower-crust47. Several recent seismic tomographies44,45,51 recognize high-velocity anomalies within the asthenosphere beneath the underthrust Indian mantle lithosphere, and interpret them as evidence of this ~ 20 Ma slab break-off.

If eclogite drips to ~ 110 km, it crosses the depth range of the seismogenic zone, \({L}_{z}\) ~60 km (measured from Fig. 1). The timing of any putative eclogite drip is controlled by the convergence rate between India and the south Tibetan detachment (STD, directly above our proposed eclogite drip) across the Main Himalayan Thrust (MHT), ~ 15 mm/yr52. The low-angle geometry of the MHT implies that Indian lower crust does not reach depths (pressures) appropriate for eclogitization until the MHT reaches ≳20 km depth as it passes beneath the Lesser Himalayan thrust ramp53 ~ 50 km north of the Main Frontal Thrust (MFT, thrust front). The sub-Moho earthquakes occur 100–200 km north of the Himalayan thrust front, in rocks that cannot have begun to eclogitize until ~ 5–10 Ma b.p. If the available timescale for an eclogite drip is \(T\) = 5–10 Myr, the strain rate \(\dot{\epsilon }\) ~ \({L}_{z}/({L}_{H}T)\)  ~ 10–14 s–1 in the putative drip.

In summary, we seek a geologic scenario in which brittle earthquakes occur at ~ 400–800 °C at 50–100 km depth at averaged strain rates of 10–15–10–14 s–1 within ~ 5–10 Myr of entry into the subduction zone, in a region where slab breakoff or lithosphere removal may have occurred at 20 Ma.

Temporally constrained viscosity of a Rayleigh–Taylor instability

We next investigate the range of viscosities that permit a drip of the size of the sub-Moho seismogenic zone to form in the available 5–10 Myr, using pure viscous modeling of Rayleigh–Taylor instabilities (Fig. 2). Note that our Rayleigh–Taylor models have simplified geometries and (constant) viscosity assigned to each geological layer. Our model is intended to show the relation between eclogite viscosity and the dripping timescale, and not to reproduce the seismogenic process. We run simulations with different lithospheric-mantle viscosity \({\eta }_{m}\), eclogite viscosity \({\eta }_{e}\), and eclogite density \({\rho }_{e}\). We require eclogite to form a drip length \({d}_{e}\) > 40 km from a static environment at 20 Ma. In our model, dripping is triggered by pre-existing topography of the Moho and LAB (Fig. 2a). The drip of lithospheric mantle is mimicking the slab breakoff 20 Myr ago.

The eclogite drip length \({d}_{e}\) is a strong function of viscosity. For eclogite density \({\rho }_{e}\) = 3.40 × 103 kg/m3, corresponding to ~ 40–60% lower-crustal eclogitization54, in general \({d}_{e}\) ≥ 40 km for \({\eta }_{e}\) \(\lesssim\) 5 × 1021 Pa⋅s (Fig. 2c). Modeling with our default parameters (Supplementary Material S1), we show that drips initiate simultaneously in the upper-mantle lithosphere and eclogite layers, but the upper mantle drips much faster because it has lower viscosity and is settling into the asthenosphere that has even lower viscosity (Fig. 2b). After 10 Myr the eclogite drip has barely initiated, but lithospheric mantle has already reached the bottom of our model (400 km), inducing inward lithospheric and asthenospheric mantle flow at the tail of the drip. From 10–20 Myr mantle flow continues as the trailing stem of the drip gradually closes, while the eclogite drip grows faster, reaching > 100 km depth by the end of the simulation (Supplementary Material S2). Therefore, although our modeling timescale is 20 Myr, the effective timescale for the eclogite drip is \({T}_{e}\) ≲10 Myr. Our 5-Myr simulations (Fig. 2d) also follow a linear scaling, and require \({\eta }_{e}\) ≲ 1 × 1021 Pa⋅s.

The lithospheric mantle viscosity \({\eta }_{m}\) has little influence on the eclogite drip rate if \({\eta }_{m}\lesssim\) 0.2 \({\eta }_{e}\). The mantle-vs.-eclogite density difference also has relatively little influence: setting \({\rho }_{e}\) = 3.30 g/cm3, only 50 kg/m3 above the lithospheric-mantle density and corresponding to just 10–30% eclogitization54, barely changes the maximum \({\eta }_{e}\) from ~ 5.1021 Pa⋅s (Fig. 2c) to ~ 4.1021 Pa⋅s (Fig. 2e). This small decrease is far less than predicted by the linear relation \(\eta\) ~ \(\Delta \rho g\lambda T\) for Rayleigh–Taylor instabilities55. Our supplementary simulations (S3) simulating eclogite dripping in a static lithospheric mantle show that eclogite drips much faster in the presence of lithospheric delamination, due to the mantle flow field created by earlier lithospheric delamination. If viscous drag is the major driving force for an eclogite drip, no large density difference (or equivalently, no high degree of eclogitization) is required, and the eclogite drip initiates at the very beginning of eclogitization. Viscous drag from large-scale mantle flow has recently been inferred to trigger lithospheric dripping at the base of the North American craton56, and beneath southern Tibet57. In our models, because of the viscous drag, whether the deeper lithospheric foundering is by delamination or by dripping has no significant effect on the eclogite foundering rate (Figs. 2, S3), and the eclogite always founders by dripping (Supplementary S8).

Although we have modelled the conventional “lithosphere-denser-than-asthenosphere” condition, it has recently been shown that even with the “asthenosphere-denser-than-lithosphere” condition, the intra-lithosphere downward density decrease due to thermal expansion can drive intra-lithospheric overturn confined to the upper 200 km of mantle58. This overturn instability proceeds on the same ~ 20-Ma timescale of our models and similarly exerts shear stresses on the Moho that entrain lower crust58.

In summary, for an eclogite drip to reach and create seismicity at 110 km depth since 5–10 Ma, \({\eta }_{e}\) ≤ 1–5.1021 Pa⋅s, and likely \({\eta }_{m}\) ≤ 1021 Pa⋅s. These maximum viscosities are higher than most previous estimates from Tibet and India, of lower-crustal viscosity 1018,19,20,21 Pa⋅s59,60,61,62 and lithospheric-mantle viscosity 1018–21 Pa⋅s63,64. Delamination modeling studies also use similar or smaller viscosities: eclogite 1019–21 Pa⋅s65, and dry/wet olivine 1016–21 Pa⋅s66,67.

Brittle and ductile strengths of south Tibetan lithosphere

Earthquakes occur in materials that are brittle at the imposed strain rates, pressures, and temperatures because their ductile failure modes cannot dissipate the build-up of stress. The differential stress required for brittle failure increases with confining pressure (depth), whereas the stress needed to reach a specific ductile flow rate decreases with temperature (depth). The intersection of the curves representing these two strength laws (Fig. 3) is the brittle-ductile-transition depth, hence maximum depth of seismicity, which we estimate using strain rates for both tectonic hypotheses, laboratory-measured constitutive equations for different rock types (supplementary material S4) and temperature models23 (Supplementary Material S5).

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

(a) Himalayan geotherms from Western Dharwar (red, preferred) and Bundelkhand (blue)23 at the proposed eclogite-drip center. Dashed lines: temperature uncertainties (Supplementary Material S5). (b,c): Lithosphere yield-strength envelopes for Western Dharwar geotherm for different stress systems and representative lithologies. Transparent swaths: ranges due to temperature uncertainty. Brittle-failure curves (Byerlee’s Law)68 plotted in black with frictional-coefficient angle φ = 30º for compressional, strike-slip, and extensional stress systems (principal faulting mechanism under each hypothesis emphasized with thicker solid lines) and in red for strike-slip failure with φ = 2º. Yellow dashed lines mark the range of crossing points between strength and brittle-failure curves, i.e. likely deepest seismicity under different hypotheses. (b) \(\dot{\varepsilon }\) = 10–15 s-1, consistent with DCF penetrating into olivine-rich upper mantle (solid green lines), or into an eclogite-drip (translucent eclogite and clinopyroxene strength curves). (c) \(\dot{\varepsilon }\) = 10–14 s-1, eclogite-drip hypothesis; dashed magenta line represents the strength curve if eclogite adiabatically drips 40 km deeper.

The penetrating-fault hypothesis requires the dominant upper-mantle mineral, olivine, to remain brittle beneath the ~ 70 km Moho and down to 110 km depth to satisfy the sub-Moho seismicity observations. For strain rate ~ 10–15 s-1, in the modeled temperature range the olivine brittle realm is almost entirely above the Moho, whether olivine is wet, dry or undergoing Peierls creep, for all faulting mechanisms for a 30º frictional angle (Fig. 3b, with the deepest transition point at 65 km). Even at 10–14 s-1 strain rate (Fig. 3c) the deepest brittle-ductile transition for strike-slip faulting is barely below the Moho (71 km). Only if the frictional angle is as low as ~ 2º as sometimes arbitrarily modelled65 to represent the effect of decreased friction at high pressure and temperature69 or viscous strain softening at high strain rates70, does the brittle-ductile transition occur well below the Moho, but still only at ~ 80 km, far shallower than the deepest seismicity (Fig. 3b). Hence an olivine-rich mantle below the ~ 70-km Tibetan Moho is likely to be aseismic and cannot explain the ≳100-km depth seismicity.

The eclogite-drip hypothesis, in contrast, requires eclogite to remain brittle at typical upper-mantle depths. For strain rate ~ 10–14 s-1, the brittle-ductile transition of eclogite is 5–10 km below the 70-km Moho for the three kinds of faulting (Fig. 3c). Note that horizontal compression is equivalent to vertical extension which represents the dripping condition. An eclogitic region lying on the modelled geotherm therefore may also not predict seismicity to the observed ~ 110 km depth. However, as discussed above, if there is an eclogite drip beneath the Himalaya, it has dripped ~ 40 km downward in only ~ 5–10 Myr, much faster than its thermal-diffusion timescale (~ 50 Ma, supplementary material S6). The core of the drip is effectively isothermal for ~ 50 Myr, allowing it to remain brittle to 110-km depth (adiabatically shifted dashed curve, Fig. 3c). In addition, constitutive equations for clinopyroxenite, likely the controlling component of eclogite, show even deeper brittle-ductile transition depths, also permitting seismicity below 100 km. Although the creep curves have large uncertainty due to compositional variability (Supplementary Material S5&7), clearly an eclogite drip is more likely to remain seismogenic to > 100 km than olivine-dominated upper mantle.

Discussion: eclogite drip or Moho-penetrating fault?

The drip hypothesis clearly best fits the brittle-failure criterion. Though further studies might be useful to fully explore the conditions under which gravitationally unstable eclogite forms a drip as modelled here, rather than a delaminating layer, no matter what geometry we use for the initial state of our Rayleigh–Taylor models the eclogite tends to drip at similar rates. Any eclogite drip can achieve 40-km deepening in 5–10 Myr when enabled by viscous drag from mantle flow caused by Miocene mantle-lithospheric delamination, while remaining highly viscous (brittle). The viscous drag is a pre-requisite for rapid drip formation, but recent data assimilation studies show the detached Indian slab creates large drag forces with maximum shear stress inside the lithospheric root57. Our drip model can also explain why, despite the eclogite doublet being observed25,71 for 1400-km along-strike, sub-Moho Himalayan earthquakes are restricted to the part of the orogen where seismic tomography has found deep high-velocity anomalies45,72 likely representing the lithospheric delamination at ~ 20 Myr37,73.

Our drip hypothesis has two distinctions from previous interpretations. Our drip is centered ~ 150-km north of the Himalayan front, but gravity profiles have previously been used to suggest that Indian lower-crustal density only reaches upper-mantle density, enabling drip initiation, ~ 200 km north of the front35,67, consistent with the slow kinetics of anhydrous eclogitization70. However, the mismatch between our proposed early eclogitization and the later eclogitization inferred from gravity anomalies39 is well within the resolution limit of density models. In supplementary material S9 we show the trade-off between deepening the eclogite layer by 5 km (the potential uncertainty in Moho depth) and densifying the eclogite layer by ~ 200 kg/m3, so confirming that our drip location is consistent with gravity data. In order to expedite eclogitization within only a few million years, grain-boundary water is certainly required, though perhaps only a few ppm74 or a few hundred ppm42. Sufficient water may be pre-existing in the Indian lower-crustal granulites75, but this assumption does not explain the spatial localization of the below-Moho earthquakes. Alternatively, seismic pumping – our preferred hypothesis – can introduce water through an active fault reaching the eclogitizing layer38,69. Seismic pumping is highly potent during eclogitization because densification leads to cracking, permeability increase, and so sucks more water into the rock, causing volumetrically pervasive transformation41.

A second distinction is that most of the deep earthquakes beneath south Tibet have strike-slip focal mechanisms15,16,47 (supplementary S12, S13), in contrast to the vertical extension nominally created by dripping or slab necking as observed beneath the Vrancea22,23 and Hindu Kush22,23. The seismogenic middle crust below the MHT implies the DCF or other structures remain active beneath the High Himalaya76, so may pump surface fluid down to the lower crust as well as create dextral shear within the proposed eclogite drip. Even though an isolated eclogite drip is an insufficient explanation of our observations, we suggest that the sub-Moho seismogenesis, and supra-Moho seismogenesis in the same region, results from crustal-scale faulting enabling lower-crustal eclogitization that triggers dripping. In this hypothesis, pre-existing whole-crustal faults (DCF and possibly PXG) guide fluid into the lower crust, triggering fast eclogitization42 (Fig. 4a). As the eclogitized lower crust becomes denser than the upper mantle, the mantle flow field created by lithospheric delamination ~ 20 Myr ago triggers an eclogite drip to \(\gtrsim\) 110 km depth, bringing brittle materials down to today’s sub-Moho seismogenic zone. Byerlee’s Law incorporating Andersonian mechanics68 implies that the required differential stress is the greatest for compressional events, the stress system we would most expect in the Himalaya, compared to the other faulting types. However, the deeply-penetrating faults that we infer allow the sub-Moho eclogite to break via weaker strike-slip or extensional faulting mechanisms, enabling brittle faulting to greater depth (Fig. 3b) and explaining the observed strike-slip and normal-fault mechanisms. The receiver-function Moho offsets further north along the PXG may mark an earlier phase of lower-crustal dripping triggered by the graben19 (Fig. 4b).

Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

(a) Cartoon coupling mechanism combining faulting, fluid intrusion (blue water drops), lower-crustal densification, and dripping. Stars mark the rough locations of observed seismicity. (b) North–south profile through seismicity10 and intersecting west–east receiver-function profile and interpretation19. Stars are previously-studied earthquakes10 colored by their Sn/Lg values (red earthquakes with high Sn/Lg ratios are likely sub-Moho; blue ones with lower Sn/Lg ratios are likely above-Moho).

Using constraints from geologic history and active seismotectonics, together with strain-rate, temperature, and viscosity estimates, our viscous-drip modeling and brittle-ductile-transition analysis suggests the sub-Moho Himalayan earthquakes occur within an eclogite drip that required crustal-scale faulting for its initiation. Although each data constraint has significant uncertainty, collectively they are sufficient to begin to test different models for this unusual, spatially isolated, sub-Moho seismic zone. We infer that these continental mantle earthquakes depend on the rare spatial coincidence of subduction of an active strike-slip fault cutting mafic lower crust in a region of geologically recent slab breakoff or delamination. Spatially localized earthquake clusters imply that, in contrast to traditional, implicitly one-dimensional, rheologic models (e.g., jelly-sandwich and crème-brûlée terminology), real lithospheric rheology may vary laterally over short distances.

Methods

Viscous drip modeling

We model pure viscous 2D Rayleigh–Taylor dripping48 using Underworld277, finite-element geodynamics software which has been widely applied in lithospheric viscous simulations35,78 (Supplementary Material S10).

Our model is 400 km deep × 800 km wide, with four layers: crust, eclogite, ultramafic upper mantle, and ultramafic asthenosphere (Fig. 2a). The initial thickness of each layer is prescribed based on the receiver-function doublet, and Moho and LAB depths beneath south Tibet29,79. Densities are set at 2600 kg/m3 (crust), 3400 or 3300 kg/m3 (eclogite), 3250 kg/m3 (lithospheric mantle), and 3200 kg/m3 (asthenosphere). Viscosities for the crust and asthenospheric mantle are set at 50 and 0.01 × 1022 Pa⋅s respectively. We vary the eclogite and lithospheric-mantle viscosities from 0.2–1.0 × 1022 Pa⋅s and 0.02–0.4 × 1022 Pa⋅s respectively. In order to trigger instabilities, we assign a Gaussian perturbation of 20 km to the base of the eclogite (Moho) and lithospheric-mantle (LAB) using horizontal standard deviations of 100 km (length scale of the deep earthquake cluster) and 200 km (length scale of the tomographic + 2% shear-velocity anomaly45). For detailed parameters see supplementary materials S1.

Modeling commences from a static state (a fixed-India reference frame) and runs for 20 Myr, the suggested timing of Indian lithospheric delamination, with a focus on processes requiring only 5 Myr, our more restrictive constraint from the plate-convergence rate. The depth reached by the bottom of the eclogite drip (i.e. 70 km Moho depth plus its elongation below Moho, \({d}_{e}\)), is taken to represent the extent of the brittle seismogenic zone. We adjust the viscosity of eclogite \({\eta }_{e}\) from 2–10 × 1021 Pa⋅s (0.5–2 × 1021 Pa⋅s for 5-Myr model runs) and the upper mantle \({\eta }_{m}\) from 0.2–4 × 1021 Pa⋅s (0.1–2 × 1021 Pa⋅s for 5-Myr model runs), with the additional constraint \({\eta }_{e}\ge {\eta }_{m}\) because eclogite is stronger than upper mantle (Fig. 2c–e).

Rheology profiles

We use existing south-Tibet lithospheric thermal models23 to establish temperature profiles and uncertainties at the location of the sub-Moho seismicity (Supplementary Material S5). We take creep laws from representative rocks, using felsic granulite and diabase80,81 for the upper and lower crust, wet and dry olivine82,83 for the lithospheric mantle and asthenosphere, and eclogite/clinopyroxene43,84 for an eclogite drip, where present. For detailed rock ductile parameters see supplementary material S4. We also plotted envelopes for Peierls creep that is thought to be the dominant low-temperature failure mode for olivine82.

In Fig. 3 we plot brittle-failure curves using the minimum differential stress for brittle-slip laws68 for extension (across PXG), strike-slip (along prolongation of DCF), and compressional (equivalent to vertical extension caused by dripping) fault mechanisms (Supplementary Material S11). We plot the ductile curves for the preferred ‘Western Dharwar’ temperature model23 for strain rates of 10–15 s–1 (faulting hypothesis) and 1014 s–1 (dripping hypothesis). We show equivalent curves for a cooler model in Supplementary Material S7.