Fault–fracture mesh development produces tectonic tremor in fluid-overpressured serpentinized mantle wedge

Introduction

Episodic tremor and slip (ETS) is a periodic phenomenon of aseismic slow slip and low-frequency seismic activity, such as tectonic tremor and low-frequency earthquakes (LFEs) that are temporally correlated with durations ranging from days to weeks¹. In Cascadia (western North America) and Nankai (Japan), where young and warm oceanic lithosphere is being subducted beneath a continental plate, ETS occurs near the intersection between the tip of the forearc mantle wedge and the subducting oceanic crust at 25–45 km depth, downdip of the megathrust seismogenic zone^2,3,4,5 (Fig. 1). Geophysical observations reveal a landward-dipping layer of low seismic velocity near the ETS zone that is interpreted as overpressurized upper oceanic crust and/or a shear zone along the plate interface^3,6,7 (Fig. 1A), which is capped by a low-permeability barrier⁸.

**Fig. 1: Schematic illustration of fault-valve behavior in a forearc mantle wedge shear zone during ETS.**

It might be expected that the zone of ETS marks a thermally controlled transition in deformation behavior from unstable stick–slip (or brittle failure) to stable sliding (or viscous creep), as inferred for typical crust^9,10. In Cascadia, Mexico, and Nankai, however, a spatial gap exists between the seismogenic and ETS (or LFE) zones^11,12 (Fig. 1A), meaning that in the ETS zone, most strain is accommodated by viscous mechanisms such as dislocation creep or dissolution–precipitation creep inhibiting the nucleation of earthquakes¹³. This apparent contradiction can be reconciled by considering pore fluid pressure, which acts to reduce the effective normal stress and thereby promote brittle failure within a dominantly viscous environment¹³. Recent seismic investigations have shown temporal variations in seismicity, seismic velocity and attenuation, and earthquake focal mechanism during ETS or slow-slip event (SSE) cycles, suggesting fault-valve behavior¹⁴ (Fig. 1C). In this behavior, pore fluid pressure drops markedly during slip events owing to the breaching of low-permeability materials and fluid migration within and across the megathrust, following which pore fluid pressure gradually rises again during the inter-slip period^15,16,17,18. This pattern of processes indicates that the time needed for fracture networks to be sealed by hydrothermal mineral precipitation controls the recurrence interval of ETS and SSEs¹⁹, although it remains unclear which deformation mechanisms and processes operate together with fluid overpressure in the ETS-generating fault zone²⁰.

Continuous fluid supply from the subducting oceanic crust results in extensive serpentinization of the forearc mantle wedge above the deeper (>30–35 km depth) part of the ETS zone²¹ (Fig. 1B). Owing to its possible mechanical weakness^22,23,24,25, antigorite, a high-temperature serpentine mineral, is expected to constitute part of the ETS-generating plate-boundary zone. Several high-pressure (≥1 GPa) deformation experiments on dry intact samples of antigorite have documented stable sliding on a newly forming through-going fault at elevated temperatures (≥400 °C) within the stability field of antigorite^{23,26,27,28,29,30}. This study expands on previous work by exploring the role of pore fluid pressure (or water content) on deformation processes at the pressure–temperature conditions of the ETS zone. We compare our experimental results with the deformation behavior of a mantle-wedge-derived serpentinite shear zone³¹ and propose a geological model for ETS cycles, suggesting that composite extensional and extensional–shear failure events (fault-fracture mesh development³²) within a fluid-overpressured shear zone cause bursts of tectonic tremor and LFEs.

Results and discussion

Mechanical behavior of samples

Axial compression experiments were conducted on antigorite cylinders at a temperature (T) of 500 °C, confining pressures (P_c) of 0.5 and 1.0 GPa, and strain rates (ε̇) of 3.25–4.22 × 10⁻⁶ s⁻¹, under undrained conditions (Supplementary Table 1). We used two types of antigorite samples: intact cores and powders. Distilled water (1.1–12.0 vol%) was added to the cylindrical samples, which were then immediately placed in a mechanically sealed or weld-sealed Ag jacket positioned between the top and bottom alumina pistons (Supplementary Figs. 1–4). Supplementary Fig. 5 and Supplementary Table 1 present the water contents measured before and after each experiment. The experiments can be categorized into “unleaked” or “leaked” groups, in which changes in water content during the experiment are <15 vol% or >50 vol%, respectively.

Figure 2A, B and Supplementary Fig. 6 show differential stress (Δσ) versus axial strain (ε) curves for antigorite samples from “unleaked” and “leaked” experiments. For all core samples, the differential stress increased linearly prior to yielding at ε = 1%–3% (Fig. 1A and Supplementary Fig. 6). After yielding, most of the core samples reached initial peak stress and then underwent transient strain weakening, followed by quasi-steady-state behavior or strain hardening. In contrast, for all powdered samples, the differential stress increased nonlinearly until the initial peak stress at ε = ca. 10%–20%, followed by strain weakening (Fig. 2B). Despite the different stress–strain behaviors, core and powdered samples of the “unleaked” experiments have similar strengths at a given range of water contents measured before the start of the experiment (H₂O_initial), with the strengths decreasing with increasing H₂O_initial. The differences in mechanical behavior between the core and powdered samples may be attributed to differences in the initial porosities of the cylindrical samples (see below and Supplementary Note 1). For the more porous powdered samples, compaction may have continued during the deformation stage.

**Fig. 2: Stress–strain curves of antigorite samples and the relationship between sample strength, pore fluid pressure ratio, and final water content.**

In Fig. 2C, the final differential stress (Δσ_final) for all core and powdered samples is plotted as a function of water content measured after the end of the experiment (H₂O_final). The final differential stress ranges from 135 to 1193 MPa and tends to decrease exponentially with increasing H₂O_final. In contrast, the Δσ_final values for core samples from “leaked” experiments are highly scattered at a given H₂O_initial (Supplementary Fig. 7A), indicating that not all but most of the experiments with higher Δσ_final values underwent a larger amount of water leakage.

Macro- to micro-scale deformation processes

Figure 3 presents backscattered electron (BSE) images of entire cross-sections and rose diagrams showing fracture orientations, obtained for three core samples (H₂O_initial = 0, 1.1, and 3.0 vol%) from “unleaked” experiments. The samples deformed at H₂O_initial = 0 and 1.1 vol% contain several through-going faults that are oriented at 30°–40° to the direction of maximum compression (i.e., σ₁), consistent with the results of previous high-pressure (≥1 GPa) experiments on the axial deformation of antigorite core samples^26,29. For the sample deformed at H₂O_initial = 3.3 vol%, a network of conjugate faults oriented 20°–40° to the σ₁ direction or σ₁-subparallel fractures developed, particularly in the upper part of the sample, resulting in the formation of rhomboidal or hexagonal blocks (up to ca. 10 mm length). The long axis of the blocks is oriented subparallel to the σ₁ direction. The network of σ₁-parallel (extensional) and conjugate (extensional–shear) fractures, referred to as a fault–fracture mesh¹⁴, was only observed for unleaked core samples with H₂O_final ≥ 3.2 vol% and two leaked samples (Runs GPA0033 and 0046) with H₂O_initial = 6.0 vol%, which is interpreted to represent failure events at low differential stress and elevated pore fluid pressure (see below).

**Fig. 3: Microstructures of entire core samples.**

Detailed SEM observations (Fig. 4 and Supplementary Fig. 8A–C) show that the through-going faults observed in the core samples with H₂O_initial = 0 and 1.1 vol% consist of subparallel and closely spaced fault surfaces, along which microcracking occurs, and in some places, comminuted fine-grained zones with a thickness of 10–125 μm are formed (Fig. 4A, B and Supplementary 8A–C). In addition, microcracked grains along portions of the through-going faults contain olivine, interpreted to result from dehydration of antigorite, forming narrow (<5 μm) anastomosing networks (Supplementary Figs. 8A and 9). The formation of dehydration products might have been related to mechanochemical processes by which grain-size reduction by (001)-parallel microcracking and comminution in antigorite increases surface energy and promotes the dehydration reaction^29,33. For the core sample deformed at H₂O_initial = 3.0 vol%, fracture zones developed in regions between rhomboidal or hexagonal blocks (Fig. 4C). The fracture zones typically have thicknesses of 50–750 μm and are characterized by finer-scale aligned blocks or more finely cracked or comminuted clasts (Fig. 4D, E). In some larger-scale blocks, there are incipient conjugate faults that would have caused fragmentation of the blocks if further strain had accumulated (Fig. 4C). Fragments and pores are observed between the finer-scale blocks (Fig. 4F). See Supplementary Note 1 and Supplementary Figs. 8D–F and 10 for detailed descriptions of the macro- and micro-structures of the unleaked powdered samples.

**Fig. 4: Backscattered electron images of unleaked core samples deformed at water contents (H₂O_initial) of 0–3.0 vol%.**

Comparison with a shallow mantle-wedge-derived shear zone

Geological observations of shallow (ca. 30 km) mantle-wedge-derived bodies³⁴ enclosed in pelitic schists of the Sanbagawa belt in the Tomisato region, central Shikoku, Japan, demonstrate that the bodies have undergone complete antigoritization under conditions of the lower amphibolite facies (T ≈ 500 °C) and exhibit fault–fracture meshes throughout the bodies³¹ (Fig. 5A–C) that are similar to those produced in our experiments. The key difference is that interpenetrating antigorite blades were newly precipitated as a matrix to fill pore spaces between the fractured blocks in nature (Fig. 5D, E), while the dissolution of antigorite at the edges of blocks did not occur during the period of heating (ca. 1 day) in our experiments. Furthermore, the matrix regions lack finely cracked or comminuted clasts as observed in the fracture zones developed between the larger-scale blocks (Fig. 5D, E). This implies that in nature, most of the antigorite clasts were entirely dissolved into the surrounding fluids.

**Fig. 5: Geologic evidence for the development of fault–fracture mesh preserved in a shallow mantle-wedge-derived serpentinite body.**

Hirauchi et al.³¹ proposed that in nature, a localized viscous shear zone formed along longer fractures that were favorably oriented relative to σ₁ (Fig. 5D, F). Within this shear zone, newly precipitated antigorite aggregates show undulatory extinction and subgrain boundaries, along with a clear crystal preferred orientation characterized by [010] axes oriented subparallel to the shear direction and [001] axes oriented subnormal to the shear zone (Fig. 5G and Supplementary Fig. 11). These microstructural features suggest that dynamic recrystallization via subgrain rotation occurred within the newly precipitated antigorite aggregates³¹. However, the observed crystal preferred orientation (Fig. 5G) does not solely indicate the operation of dislocation creep, as it can also result from grain rotation and anisotropic grain growth through dissolution–precipitation creep²⁴. We, therefore, propose that in regions of finely fragmented clasts that existed immediately after the failure events, grain size sensitive dissolution–precipitation creep is likely to dominate over dislocation creep, whereas relatively larger antigorite crystals, which have undergone substantial grain growth, will deform predominantly through grain size insensitive dislocation creep²⁴. The longer fractures that represent the coalescence of shorter extensional–shear (mode I–II) fractures (Fig. 5A, B, D, F) appear to be the same as those observed in the core samples that deformed with H₂O_final = 3.3 vol% (Figs. 3C and 4C).

Fault–fracture mesh development at high pore fluid pressures

On the basis of a Mohr circle analysis with two Coulomb failure envelopes of antigorite serpentinite, we estimate pore fluid pressures (P_f) and then pore fluid pressure ratios (λ = P_f/P_c) at the end of the experiments (Fig. 2D and Supplementary Figs. 7B and 12–15; see “Methods” for details). Although it is possible that the high P_f values (λ > 0.8) estimated using a Coulomb failure envelope constructed from friction experiments³³ were underestimated (since P_f values never become negative), the values of λ tend to increase exponentially with increasing H₂O_final (Fig. 2D; up to 0.93 at H₂O_final = 11.7 vol%). Our experiments also show that the failure patterns of antigorite serpentinite vary with P_f and λ. For core samples with low water contents (H₂O_initial ≤ 1.1 vol%), subparallel through-going shear (mode II) failure surfaces form at an angle of ca. 30° to the σ₁ direction (Fig. 3A, B), suggesting that most of the strain is accommodated by stable frictional sliding on the through-going faults. When H₂O_final exceeds initial porosities of 0.3–1.1 vol%, which yield excess pore fluid pressures with λ of >0.7, a network of conjugate extensional (mode I) and extensional–shear (mode I–II) failures develops (Fig. 3C). This development of fault–fracture meshes is considered to occur when P_f > P_c¹⁴. The existence of extensional fractures oriented subparallel to the σ₁ direction in the sample deformed at λ = 0.7 (Fig. 3C) implies that P_f was spatially heterogeneous within the sample column (Fig. 3C). Indeed, fault–fracture meshes are more predominant in the upper (i.e., remaining) part of the sample column. The formation of connected pore space by fracturing would lead to dilatancy-induced strain-hardening behavior (Fig. 2A).

Geological model of ETS cycles

Peak metamorphic P–T conditions of the Sanbagawa pelitic schists (P = 1 GPa, T = 480–540 °C³⁵) in the Tomisato region correspond to the depth range of ETS in warm subduction zones such as Cascadia and Nankai³⁶. In the depth range, large amounts of fluids expelled via dehydration reactions in the more deeply subducting oceanic crust at the amphibolite-to-eclogite transition enter the base of the overlying mantle wedge³⁷. Based on field and experimental observations, a schematic model illustrating fracturing, viscous flow, and mineral dissolution–precipitation in mantle-wedge serpentinite over ETS cycles was constructed (Fig. 6). Continuous fluid supply to the serpentinite-rich plate-boundary fault zone maintains a high pore fluid pressure (λ_average ≥ 0.7) resulting in low effective normal stress. In this environment, a network of extensional (mode I) and extensional–shear (mode I–II) fractures (i.e., fault–fracture meshes) develops throughout the serpentinite, which may represent bursts of tectonic tremor and LFEs (stage 2 in Fig. 6). The formation of pore spaces (i.e., dilation) accompanying the fracturing and inferred rapid fluid flow along the plate boundary may result in a sudden decrease in P_f and the initiation of antigorite precipitation. We note that the absence of stress oscillations, such as those shown in Fig. 1C, in the stress–strain behavior of unleaked core samples with λ > 0.7 (Fig. 2A) can be attributed to the undrained conditions of our experiments (see Methods for details). Under these conditions, even if fault–fracture meshes (i.e., fluid pathways) developed, the water contained within the Ag jacket could not escape. As a result, pore fluid pressure (and thus effective confining pressure) remained nearly constant, preventing stress fluctuations. These experimental limitations need to be improved in future studies.

**Fig. 6: Conceptual model of ETS cycles.**

Recent deformation experiments on core antigorite samples using a gas-medium apparatus demonstrated that conjugate extensional–shear fractures formed under P–T conditions beyond the stability field of antigorite³⁸. This suggests that in addition to fluid supply from slab dehydration, the dehydration of serpentinite itself could elevate P_f and lead to the formation of fault–fracture meshes similar to those observed in our experiments. Such dehydration may occur when mantle-wedge serpentinite is dragged down by the traction of the slab³⁹.

Tectonic tremor and LFEs will re-occur when P_f increases (as a result of the closure of pore spaces) until a failure condition is attained again (stage 3 in Fig. 6), suggesting that the dissolution–precipitation rates of antigorite control the recurrence interval (months to years) of ETS, although these rates have not yet been thermodynamically determined. The failure condition might be also modulated by variations in stress and strain rate within a fluid-overpressurized, block-in-matrix serpentinite shear zone (Fig. 1B), where the stronger blocks make load-bearing force chains, leading to local stress buildup^40,41. Numerical modeling by Beall et al.⁴⁰ demonstrated that the fracturing of the blocks might redistribute stress into the weaker matrix, switching the shear zone to predominately viscous creep (related to short-term SSEs) with an associated increase in strain rate. We suggest that in the Tomisato serpentinite body, at the relatively low P_f conditions following dilational fracturing, newly precipitated antigorite grains on fracture surfaces that are favorably oriented relative to σ₁ deform predominantly by dislocation and dissolution–precipitation creep²⁴ (Figs. 5 and 6). Following the findings from Beall et al.⁴⁰, the viscous creep localized within the newly-precipitated antigorite matrix that immediately occurred after the fracturing represents high-strain-rate deformation under excess-shear-stress conditions³¹ (stage 3 in Fig. 6) and may correspond to short-term SSEs.

Implications for ETS source characteristics

The development of fault–fracture mesh at near-lithostatic pore fluid pressure in regions downdip of a megathrust seismogenic zone is not limited to the shallow forearc mantle wedge³¹ but has also been recognized in underthrust pelitic sediments^20,42,43,44, where temporal fluctuations in P_f occur in association with cyclic fracturing and sealing (quartz precipitation)⁴⁵. The development of fault–fracture mesh is coeval with viscous, SSE-like deformation (dissolution–precipitation creep) in nearby localized shear zones at strain rates higher than those inferred from plate-motion rates²⁰. Therefore, we suggest that the development of a distributed fault–fracture mesh and contemporaneous localized viscous shearing across a subduction interface represents a universal mechanism for ETS where pore fluid pressure locally approximates or exceeds lithostatic pressure, although the density, geometry, and length of fault–fracture meshes are expected to vary with the nature of preexisting anisotropy (e.g., foliation) and physical properties (e.g., porosity, rigidity, and strength) of fault-rock assemblages and influence the magnitude of LFEs.

Our experimental and field observations support a model of ETS generation as coupled brittle–viscous deformation in a thick (mélange) shear zone, rather than unstable slip on a single planar fault in a rate-and-state friction framework with velocity-neutral or slip-rate-dependent friction^46,47. The infiltration of hydrous fluids along fractures in the serpentinized mantle wedge leads to the precipitation and growth of antigorite and the occurrence of block-in-matrix structures³¹, whereas in the presence of fluid-mobile elements (Ca and Si) in aqueous fluids, metasomatic alteration results in the formation of hydrous minerals (e.g., talc and tremolite) from antigorite in a highly sheared matrix^48,49,50. Therefore, variations in slip rate during SSEs are probably related to the effective thickness (a few to several hundred meters) and rheological properties of the weak matrix.

A fundamental problem in this study is whether the fault–fracture meshes observed in experiments and nature can be scaled up to generate the large magnitudes of LFEs⁵¹ and whether precipitation kinetics are fast enough to control P_f cycling to match ETS cyclicity. Although the dimensions of individual blocks (up to a few meters in length³¹) are smaller than the length scales of LFEs (>10–10²m), upscaling might be possible if fracturing occurs by connecting the individual blocks (Fig. 3C), as shown in two-dimensional numerical models used to examine transient-slip characteristics of block-in-matrix shear zones⁵¹. Furthermore, depending on the scale of individual blocks in a weaker viscous matrix, cascading or simultaneous failure of multiple rupture surfaces within a fluid-overpressurized shear zone might be required to produce events detectable by seismological instruments^51,52. It is widely accepted that LFEs generate double-couple focal mechanisms and represent shear slip on the megathrust⁵³. However, LFEs in several subduction zones deviate from the moment–duration scaling and Gutenberg–Richter relationship characteristic of regular earthquakes^52,54,55. In addition, their seismic radiation patterns are inconsistent with a double-couple, shear rupture mechanism⁵⁶. We propose that these seismological characteristics can be better explained if LFEs are interpreted as composite extensional and extensional–shear failure events (i.e., fault–fracture mesh development), consistent with field and experimental observations from our study.

Methods

Starting material

The starting material was a serpentinite block (sample No. SK-18) obtained from the Nagasaki metamorphic rocks in southwestern Japan. The serpentinite has a porosity of 0.3–1.1 vol% and is composed of 98% antigorite, <2% magnetite, and trace amounts of tremolite. Antigorite displays an interpenetrating texture⁵⁷, consisting of randomly oriented antigorite blades, and trails of magnetite define a foliation (Supplementary Fig. 1). To ensure that the rheological effects of magnetite are consistent across all experiments, the serpentinite block was cored at 45° to the magnetite foliation. Cylindrical core samples with a diameter of 6 mm were cut to a length of approximately 13 mm, and the ends were ground parallel. To maximize the amount of water entering the sample, the sample for Run GPA0015 was cut in half perpendicular to the cylinder axis, whereas for Run GPA0018, a bore with a diameter of 1.6 mm and a length of 8 mm was drilled on the upper surface of the sample cylinder. To prepare powdered samples, we crushed and sieved the serpentinite block to obtain a grain size of <100 μm. An amount of 1 g of the powders was cold-pressed at atmospheric pressure into cylinders with a diameter of 6.22 mm and length of 18 mm. The porosity of the powdered samples is approximately 30%. All samples were dried at 120 °C for one day (minimum) before use in each experiment.

Sample assembly

Schematic diagrams of the sample assembly are presented in Supplementary Figs. 2–4. NaCl and a composite of NaCl and polyethylene were used as the confining medium on the outside and inside of the graphite furnace, respectively. Temperature was measured using a Pt–Pt13%Rh thermocouple shielded in mullite and placed next to the center of the sample. An amount of 1.1–12.0 vol% of distilled water was added to the sample using a pipette and/or by immersing the sample under vacuum in distilled water. Supplementary Fig. 4 presents schematic diagrams of the sample assembly to show how distilled water added to the sample was sealed with an Ag jacket. For Runs GPA0016–0018, 0032, 0033, 0035, 0049, and 0051, each core sample was wrapped with two layers of Ag foil (wall thickness of 0.02 mm), then mechanically sealed by placing one to three Ag disks (wall thickness of 0.02 or 0.1 mm) at each end of the sample and folding the ends of the Ag foils over Ag disks, and then slid into an outer Ag tube (wall thickness of 0.2 mm) that slightly overlapped the end pistons (composed of alumina). For Runs GPA0032–0035, each core sample was weld-sealed by placing one Ag cup (wall thickness of 0.1 mm) at each end of the sample. For Runs GPA0045-0047, each core sample was placed inside an inner Ag tube (wall thickness of 0.15 mm) mechanically sealed using three Ag disks (wall thickness of 0.1 mm) at each end of the sample and then slid into an outer Ag tube. For Runs GPA0036-0038, 0044, and 0048, each powdered sample was placed inside a mechanically sealed, tubulated Ag foil with one Ag disc (wall thickness of 0.02 mm) at each end of the sample and then slid into an outer Ag tube. We weighed the sample, end pistons, and Ag jacket just before the start of each experiment.

Deformation experiments

Axial compression experiments were conducted within the stability field of antigorite at a temperature (T) of 500 °C and confining pressures (P_c) of 0.5 and 1.0 GPa, using a modified Griggs-type solid-medium apparatus installed at Kochi Institute for Core Sample Research, Kochi, Japan. The experimental P–T conditions remain below the stability limit of antigorite²⁶. Samples were initially pressurized to 250 MPa at ambient temperature. The pressure and temperature were then raised alternatingly to the desired values over a period of 7 h. The samples were annealed at the desired pressure and temperature for 6–13 h, during which the piston was advanced to come into contact with the sample. All samples except for Run GPA0036 were deformed at constant axial displacement rates of 0.0452–0.0587 μm/s to achieve a strain rate (ε̇) of 4.22 × 10⁻⁶ s⁻¹. The strain rate for Run GPA0036 was 3.25 × 10⁻⁶ s⁻¹. Two hydrostatic experiments (Runs GPA0034 and 0035), each conducted for approximately 20 hours, were performed for comparison. At the end of the experiments, the samples were quenched under load at a rate of 2 °C/s to 300 °C to preserve microstructures. The pressure and temperature were then reduced to ambient conditions over a period of 3 h.

After quenching and depressurization, the piston–sample–jacket assembly was weighed using an electric balance after the Ag foil/tube had been perforated by a micro-drill (0.5 mm in diameter) and then dried for 1 day at 120 °C, after which changes in water content during deformation experiments were calculated. The error in the water content estimation is ±1.5 mg, which corresponds to a porosity variation of ±0.4 vol% for the core sample size used in this study. For the powdered samples, we note that the antigorite powder may have absorbed some moisture when the sample assembly was set up, while this effect likely added less than 1 vol% of water to the samples.

Stress measurements

Axial load and axial displacement were measured using an external load cell and an external displacement transducer, respectively. Force-displacement data were corrected to account for apparatus distortion (100.5 kN mm⁻¹) and dynamic friction caused by the advance of the σ₁ piston. The baseline for the dynamic friction correction was provided by linear extrapolation of the initial load increase prior to the hit point, where the hit point represents the intersection of the dynamic friction baseline and a linear fit to the initial elastic loading curve. The resolution for the stress measurement is ±25 MPa with a precision of 2 MPa. Experimental conditions and results are summarized in Supplementary Table 1.

Analytical procedures

The recovered samples were cut in half parallel to the cylinder axis, and one half was impregnated with epoxy resin to prepare polished thin sections. Petrographic analysis was then performed using an optical microscope and a field-emission–scanning electron microscope (FE–SEM).

Minerals were identified using a laser Raman spectrometer (NRS-7100) at Shizuoka University, Shizuoka, Japan. A microscope with a ×100 objective lens was used to focus the incident laser beam (532 nm line, green laser) into a 1 μm spot size. Spectra were acquired for 60 s at a laser power of 11.3 mW. The spectral resolution was 1 cm⁻¹.

An electron backscatter diffraction (EBSD) map was acquired using the FE–SEM (JEOL JSM-IT700HR) equipped with a Symmetry EBSD detector (Oxford Instruments) at Shizuoka University. For the EBSD mapping, the XZ surface of one serpentinite sample was polished using diamond paste (1.00 and 0.25 μm) and colloidal silica. The analytical conditions were an accelerating voltage of 15 kV, a sample tilt of 70°, and a probe current of 6 nA. EBSD patterns were automatically indexed using AZtec software on a rectangular grid (ca. 625 μm × 400 μm) with a step size of 0.85 μm. Post-acquisition treatment with AZtecCrystal software included the removal of isolated single pixels (i.e., wild spikes) and the assignment of the average orientation of ≥6 neighboring pixels to non-indexed pixels. In addition, systematic indexing errors resulting from pseudosymmetry in the antigorite structure, which leads to similar electron diffraction patterns for crystals rotated by 60° and 120° around the [001] axis²², were partly corrected. The EBSD data were then processed using the MTEX toolbox for MATLAB to produce pole figures. Pole figures are presented as lower-hemisphere equal-area projections and are calculated from the orientation distribution function (ODF) using a half-width of 10°. Contours are multiples of uniform distribution (m.u.d.). We measured the strength of crystal preferred orientations (CPOs) using the J-index, which is equal to unity for a random fabric and increases to infinity for a single crystal⁵⁸. The strength of the CPO was also determined using the M-index technique⁵⁹, in which M-index values of 0 and 1 represent a random fabric and a single crystal, respectively. The sharpness of each pole figure is defined by the pfj index, which has a value of 1 for a random distribution.

Estimates of pore fluid pressure

The mechanical behavior of rocks is described by a Mohr–Coulomb criterion of the form

$$tau ={mu }_{{{mathrm{int}}}}({sigma }_{{{{rm{n}}}}}-{alpha P}_{{{{rm{f}}}}})+C$$

(1)

where τ is the shear stress; μ_int is the internal coefficient of friction; σ_n is the normal stress; α is a constant that is related to the ratio of the real area of contact (A_r) to the nominal area of contact (A), with α = 1 − A_r/A⁶⁰; P_f is the pore fluid pressure; and C is the cohesion. To estimate pore fluid pressures (assuming α = 1) at the end of the experiments, we first constructed a Mohr–Coulomb failure envelope for antigorite serpentinite, using values of final differential stress (Δσ_final) for dry core samples at P_c = 0.5 and 1.0 GPa, which yield μ_int = 0.30 and C = 131 MPa (Supplementary Fig. 12A). We assumed that the strength at low (<0.4 GPa) normal stresses follows a failure envelope based on the frictional properties of wet antigorite, with μ_int = 0.63 and C = 0 MPa obtained at P_c^eff (= P_c − P_f) = 70 MPa and T = 500 °C³³. Next, Mohr circles (Δσ_final) for the core and powdered samples obtained from the leaked and unleaked experiments were shifted left until the point of tangency with the composite failure envelope (Supplementary Figs. 13–15). The amount of shift can be regarded as the magnitude of pore fluid pressure (Supplementary Fig. 12B), which is then utilized to calculate the pore fluid pressure ratio λ (=P_f/P_c). It is important to note that as P_f values are estimated from the mechanical data, the estimation of P_f and P_c^eff is not influenced by uncertainties in the water content measurements. Instead, the uncertainty in the P_f estimation arises from the accuracy and precision of the stress measurement.