Faulting triggered by a quasi-diffusionless shear transition of olivine in deep subducted slabs

Abstract

To clarify the process of faulting triggered by the olivine-ringwoodite transition in deep subducted slabs, we conducted deformation experiments on mantle olivine at pressures of 15 to 20 gigapascals and temperatures of 870 to 1320 kelvins. Throughgoing faulting occurred as a result of shear localization to “soft” gouge layers, which mainly consist of kinked olivine grains. Kinking of olivine grains on the [010](100) slip system resulted in a pseudo-martensitic transition of olivine to ringwoodite via poirierite having relationships of [010]_Ol // [010]_Poi and [100]_Ol // [101]_Poi. The release of a high amount of latent heat via the poirierite-ringwoodite transition can induce a significant weakening of the fault gouge, without the aid of grain size–sensitive creep, resulting in the occurrence of faulting. The observed pseudo-martensitic transition of olivine to ringwoodite, which can proceed even at room temperature under stress, explains the cause of high seismicity in bending regions of cold subducted slabs that consist of metastable olivine.

A quasi-diffusionless phase transition of olivine to ringwoodite via poirierite induces deep earthquakes in cold deep slabs.

INTRODUCTION

Many earthquakes occur in subduction zones. At depths greater than 50 km, the occurrence of earthquakes is limited in subducting slabs. Seismicity monotonically decreases with depth due to the positive pressure dependence of frictional strength (i.e., Byerlee’s rule). The cause of intermediate-depth earthquakes (at depths of 50 to 300 km) has been attributed to the dehydration of hydrous minerals (1, 2), the adiabatic instability (3), or the eclogitization reaction (4). In many slabs subducted into the mantle transition zone (MTZ), however, the seismicity increases with depths from ~400 to ~600 km and decreases quite abruptly between 600 and 700 km (5). This anomaly has been attributed to the effect of the pressure-induced phase transitions of (Mg,Fe)₂SiO₄ mantle olivine on deep earthquakes [i.e., transformational faulting (6)]. In the hotter rims of slabs subducted into the MTZ, olivine transforms to wadsleyite (at pressures of ~13 GPa) and ringwoodite (~17 GPa) (7, 8). Seismological observations have pointed out that the occurrence of deep earthquakes requires the existence of a metastable olivine wedge (MOW), which is the inner part of a cold subducted slab filled with metastable olivine, in the subducted slab (9, 10). However, Billen (11) pointed out that deep earthquakes require not only the cold MOW but also strain accumulation (i.e., high strain rate and/or high stress) in the slabs.

Many deformation experiments using Mg₂GeO₄ germanite olivine (12–15) or Fe₂SiO₄ fayalite (16) have been performed to clarify the detailed process of transformational faulting. A combination of multichannel acoustic emission (AE) monitoring and in situ synchrotron x-ray stress/strain measurements (14) enables us to evaluate the brittle behavior of olivine samples at high pressures and high temperatures. It has been realized that formation of nano shear bands filled with nanograins of spinel (15) or the occurrence of shear-enhanced melting (17) causes faulting in germanite olivine. However, the number of previous experimental studies on mantle olivine are very limited because of the difficulty of the experimental techniques for high-pressure generation. Green et al. (18) successfully performed stress-relaxation experiments on mantle olivine at ~15 GPa and 1550 K using a multianvil apparatus and showed that transformational faulting is triggered by the olivine-wadsleyite transformation. Recently, Ohuchi et al. (19) have conducted in situ deformation experiments on mantle olivine under the conditions of deep subducted slabs (11 to 17 GPa and 860 to 1350 K) using a deformation-DIA apparatus combined with a multichannel AE monitoring system installed at the BL04B1 beamline of SPring-8. They showed that the formation of a weak gouge filled with nanocrystalline metastable olivine and wadsleyite causes the faulting in the subducted slabs. However, the role of olivine phase transition in transformational faulting is not yet fully understood.

In this study, we performed triaxial deformation experiments on olivine aggregates at pressures of 15 to 20 GPa and temperatures of 870 to 1320 K (i.e., conditions of the slabs subducted into the middle MTZ; table S1) using a deformation-DIA apparatus at the BL04B1 beamline of SPring-8 (figs. S1 and S2). AE monitoring combined with in situ x-ray analysis (i.e., measurements of pressure, differential stress, and strain) revealed that localization/accumulation of damages and kinking of olivine followed by a quasi-diffusionless pseudo-martensitic transition of olivine to ringwoodite are critical for the occurrence of faulting in the middle and lower MTZ.

RESULTS

Mechanical data and AE records

Time evolutions of differential stress, pressure, strain, temperature, and acoustic activity in the deformation stage of each run are shown in Fig. 1 and figs. S3 and S4. No wadsleyite/ringwoodite peaks were identified in the x-ray diffraction patterns due to sluggish nucleation of high-pressure phases at temperatures lower than 1270 K. Trace amounts of high-pressure phases (<5 vol %) were observed in the recovered samples (table S1). At 1320 K, ringwoodite peaks (220 and 400) were observed during the deformation stage (M4130; fig. S2D). The occurrence of throughgoing faulting was confirmed in three recovered samples deformed at a constant temperature in the range of 970 to 1170 K in path #2 runs (M3564, M3566, and M3845). Faulting was absent in other runs (table S1). Pressure and differential stress gradually increased with strain until the occurrence of yielding at a strain of 0.06 to 0.09 (1120 to 1170 K in runs M3566 and M3564; Fig. 1A and fig. S3A, respectively) or 0.03 (970 K in run M3845; fig. S3B).

Fig. 1. Mechanical (pressure, stress, and strain) and acoustic records in the sample faulted at 1120 K (M3566).

(A) Stress values were obtained from four diffraction peaks of olivine [021 (circle), 101 (diamonds), 130 (squares), and 131 (triangles)]. Pressure and strain are shown by gray open circles. The black and pink arrows indicate the yield and faulting, respectively. Cumulative number of AEs (gray line) and the maximum amplitudes (red lines: AEs from the inside of the sample; green: outside of the sample) are plotted against time. The total number of AE events (N_in: from the inside of the sample; N_out: outside of the sample) is also shown. The red arrow points to the mainshock (~5 V or higher). The pale yellow and pale red areas represent the durations of softening and faulting, respectively. (B) Time-resolved two-dimensional views of AE hypocenters in the deforming sample (blue cylinder) and the pressure medium (black square). The black cross shows the uncertainty range for hypocenter locations. Thick arrows represent the compressional direction.

In run M3566, the strain rate was almost constant throughout the run (Fig. 1A). The deformation stage was terminated by a discontinuous increase in strain due to the occurrence of faulting at a strain of 0.09. The faulting time has an uncertainty of ±1.25 min (in 41 to 42.5 min after starting deformation; the pink-hatched area in Fig. 1A) due to the limitation of our measurements. The faulting was followed by a blow-out (i.e., a sudden pressure drop due to failure of the second-stage anvils) at 42.5 min. Frequent AE radiation started, and layered clusters of AE hypocenters crossing the sample were continuously observed between 12 and 41 min (Fig. 1B and movie S1). Intense AEs (maximum amplitude of >3 V) radiated intermittently from both the inside and outside of the sample (mainly the heater, MgO sleeve, and alumina pistons; fig. S1A). AEs were absent in the rim of the pressure medium and the gaskets (outside the pressure medium), showing that the pressure generation was stable until the blowout. A huge AE (maximum amplitude of ~5 V or higher) at 27 min was followed by a yield (at 28.5 min). A significant increase in AE rate was observed during the softening (i.e., a continuous decrease in stress associated with an increase in strain rate) (30 to 41 min; the yellow area in Fig. 1A), suggesting localization of damages and the formation of fault planes (Fig. 1B). A huge AE (maximum amplitude of ~5 V or higher) corresponding to the mainshock (at 40.7 min) was followed by the faulting (in 41 to 42.5 min). The faulting is likely to induce a catastrophic adjustment of the pressure medium/gasket, which inevitably leads to a blowout (19). The total number of AEs from the inside of the sample (N_in) reached 129 events.

Figure S5A shows a radiographic image of the sample (in the cell assembly) just after the decompression stage of run M3566. The sample displays barreling of the sides as a result of ductile (and/or semibrittle) deformation. Both ends of the fault (the bright line in the image) are located in the deformed olivine sample and the molybdenum capsule surrounding the sample is not crossed by the fault, showing that the observed faulting was not affected by the events outside of the sample. Namely, the blowout is not the cause of faulting in our cell assembly. Although a blowout occurred during the annealing stage (at 17.4 GPa and 420 K) in run M3847, the development of any faults was not observed in the recovered sample (fig. S5B).

In contrast to run M3566, AEs were hardly detected throughout runs M3564 and M3845 (N_in ≤ 3; fig. S3, A and B), although throughgoing faults were developed in the samples (hereafter, “silent” faulting). The maximum amplitude of detected AEs was less than 1 V. Use of the resonant-type transducers (resonant frequency of ~4 MHz) may be the cause of a very limited number of AE events in runs M3564 and M3845. The transducers are insensitive to the signals having higher/lower frequencies far from the resonant frequency. In particular, the detection of weak AE signals having higher frequencies, which are selectively attenuated in rocks (20), is difficult in our experiments. The grain size of the fault gouge in M3845 is too small (~50 nm) to detect the AEs radiating from the gouge layer because the small source dimension size (i.e., grain size) positively and inversely correlates with the amplitude (21) and corner frequency (20), respectively. Although no discontinuous increases in strain were observed, softening was commonly observed in the late stage of deformation (M3564: at strains >0.08; M3845: ≥0.05). A relatively large fluctuation of pressure (within ±2 GPa of the averaged value) was associated with softening in run M3564.

Developments of large mode-II cracks (length of >200 μm) were observed in two recovered samples at ~18 GPa and 1120 K (M3747 and M3749; table S1). Yielding occurred at a strain of ~0.03 (or less), and then differential stress gradually increased with strain (fig. S3, C and D). Radiation of AEs (N_in = 12 to 22 events) started when strains exceeded 0.03. A few strong AE events (max. amplitude of >3 V) were recorded in run M3747. Like the runs in which faulting occurred (i.e., M3564, M3566, and M3845), softening was observed in the late stage of deformation (M3747: at strains ≥0.1; M3749: ≥0.12). Large fluctuation of stress (within ±1 GPa of the averaged value) was observed throughout the deformation stage of run M3747, probably due to the absence of the annealing stage (path #1) in the run.

Neither any faults nor mode-II cracks were observed in the recovered samples of six runs (M3565, M3748, M3989, M3991, M4127, and M4130; fig. S4). Yielding occurred at strains of 0.02 to 0.03 and 0.04 to 0.07 at lower (<1270 K) and higher (≥1270 K) temperatures, respectively. Softening was hardly observed in the six runs. A sudden stress drop was observed at the end of the deformation stage in run M3991 (fig. S4A). Although Ohuchi et al. (19) reported that softening of olivine was induced by temperature ramping during the deformation stage (i.e., path #3), softening was absent in path #3 runs in the present study (M3565, M3748, and M4127; fig. S4, D to F). This discrepancy might be linked to the difference in the olivine transition process. In Ohuchi et al. (19), the olivine-wadsleyite transition proceeded via grain-boundary nucleation of nanocrystalline wadsleyite (i.e., a diffusional process) at pressures of 14 to 17 GPa. On the other hand, a quasi-diffusionless pseudo-martensitic transition is a major process for the olivine-ringwoodite transition in the present study (see next section). The maximum value of N_in was 11 in the path #2 runs (fig. S4, A to C). The N_in was also limited up to eight events in path #3 runs (fig. S4, D to F), showing that the process of temperature ramping in path #3 runs has no effect on inducing AE radiation. Significant fluctuations of pressure (within ±3 GPa of the averaged value) were associated with AEs that were observed during the late deformation stages in runs M3748 (fig. S4E) and M3989 (fig. S4C). A small amount of wadsleyite and/or ringwoodite (<2 vol % at temperatures of <1170 K; ~3 vol % at 1270 to 1320 K) was observed in the recovered samples (table S1). The observed wadsleyite and/or ringwoodite grains (diameters of <1 μm) were homogeneously dispersed at olivine grain boundaries.

Throughgoing faulting associated with the olivine-ringwoodite transition via a quasi-diffusionless shear mechanism

A thoroughgoing fault (with a displacement of 60 μm) and many mode-II cracks oriented with an angle of 40° to 50° to the compression direction are observed in the recovered sample M3566 (Fig. 2A). Micrometer-scale wadsleyite/ringwoodite was not found in the intact areas far from the fault and mode-II cracks. The formation of damage zones (thickness of <50 μm) is often observed around the fault and mode-II cracks. Gouge layers (thickness of ~10 μm) developed along the fault. Many ultrathin “dark gray” veins (thickness of <200 nm) subparallel to the fault plane are often observed in the gouge (Fig. 2B). A focused-ion beam (FIB) foil was fabricated from the gouge layer (across a dark gray vein with a thickness of ~200 nm; Fig. 2B).

Fig. 2. Microstructures of the sample faulted at 1120 K (M3566).

(A) Backscattered electron (BSE) image of the recovered sample. The throughgoing faults and mode-II cracks are indicated by red dashed lines and blue arrows, respectively. White thick arrows represent the compressional direction. Ol, olivine. (B) Magnified image of the fault (located in area “B” in A). Dark gray veins are distributed along the fault gouge. The FIB foil was prepared from the area of the blue rectangular area. (C and D) transmission electron microscope (TEM) images of poirierite grains taken with two different orientations of the incident beam. Note that the same area was analyzed in (C) and (D). The poirierite grains are next to the host old olivine grain located in the gouge. The analyzed area corresponds to the dark gray vein in the FIB foil (shown in “B”). The SAED patterns of olivine and poirierite along the [001] zone axis of olivine are also shown. Each couple of yellow arrows show the olivine-olivine or poirierite-poirierite subgrain boundary. The couple of light blue arrows points the olivine-poirierite interphase boundary. Whole SAED patterns of poirierite are shown in fig. S6. Poi, poirierite.

The dark gray vein mainly consisted of kinked olivine grains (grain sizes of 2 to 5 μm; Fig. 3A) and angular-shaped ultrafine-grained ringwoodite (diameters of 200 to 500 nm) (Fig. 3, B to D). In the examined area shown in Fig. 3 (B to D), several ringwoodite grains show the same diffraction pattern (i.e., same crystallographic orientation). Numerous dislocations are tangled with each other in the kinked olivine grains, and the kink-band boundaries are subparallel to the [101]_Ol direction (Fig. 3A). However, dislocation density in ringwoodite grains is much lower than that in olivine and straight dislocations (dislocation line // ${\langle 10\overline{1}\rangle }_{\text{Rwd}}$) are dominant (Fig. 3B), suggesting that the ringwoodite grains were not subjected to the high stress that was applied to the kinked olivine grains. Straight subgrain boundaries were developed in the angular ringwoodite grains (Fig. 3D). The ringwoodite grains are slightly enriched in iron (FeO/MgO ~ 0.15 in molar ratio), in comparison with the olivine grains (FeO/MgO ~ 0.13) (Fig. 3E).

Fig. 3. TEM micrographs of the sample faulted at 1120 K (M3566).

(A) Image of a kinked olivine and an SAED pattern. Tangled dislocations form the cell structure. The black dashed arrows show the direction of [101] of the kinked olivine. KB microcrack, kink-band microcrack. (B and C) TEM images of ringwoodite grains and SAED patterns of the ringwoodite. Note that all the ringwoodite grains in this photo show the same diffraction pattern. The white dashed arrow shows the $\left[10\overline{1}\right]$ direction. The black arrows show the dislocations in the ringwoodite grains. Rwd, ringwoodite. (D) Subgrain boundaries developed in the ringwoodite grains (each boundary is indicated by a pair of white arrows). (E) Chemical compositions (molar ratios of iron and magnesium; silicon in wt %) of olivine grains (triangles) and ringwoodite (squares) in the dark gray vein in the M3566 sample.

Many kink bands of olivine and platy poirierite crystals with widths of several tens of nanometers are aligned to the direction of [010]_Ol and [010]_Poi, respectively (Fig. 2, C and D, and fig. S6). The observed selected-area electron diffraction (SAED) patterns of poirierite are consistent with the simulated patterns based on the structure parameters of poirierite reported by Tomioka et al. (22) (fig. S6). The [010]_Ol and [100]_Ol axes of the host olivine grain, which is next to the poirierite crystals, are systematically parallel to the poirierite [010]_Poi and [101]_Poi axes, respectively (Fig. 2C). These observations show that the poirierite crystals are the products of quasi-diffusionless shear deformation (i.e., pseudo-martensitic transition) of olivine along the [010]_Ol direction on the (100)_Ol plane (23). The poirierite crystals can be transformed to ringwoodite via another quasi-diffusionless shear deformation mechanism (22, 23). We use the term quasi-diffusionless for the observed olivine-poirierite transition because the fayalite-ringwoodite pseudo-martensitic transformation is associated with short-range diffusional cation reordering (24), although theoretical studies predicted that the forsterite-ringwoodite pseudo-martensitic transformation is controlled by a diffusionless process (23). The preferential distribution of iron to ringwoodite (Fig. 3E) supports the short-range diffusional cation reordering (24).

Kinking of olivine grains and nucleation of microcracks

Formation of large mode-II cracks at angles of 20° to 80° to the maximum compression direction was observed at a temperature of 1120 K (M3747 and M3749; Fig. 4A and table S1). Orientations of the mode-II cracks are different from those of decompression cracks that are orientated normal to the loading and unloading axes (25). The mode-II cracks are often associated with damage zones, in which numerous microcracks (lengths of <1 μm) perpendicular to the mode-II cracks are developed (Fig. 4B). Transmission electron microscopy (TEM) micrographs of the FIB foils fabricated from a damage zone (across the mode-II crack) in the M3749 sample showed that kink bands (thickness of ~500 nm) subparallel to the [101] or [111] direction of olivine were developed (Fig. 4C). Numerous thin lamellae (thickness of ten nanometers) parallel to the olivine (100) plane also developed in the kinked olivine grains (Fig. 4D). Goetze’s criterion does not account for the developments of large mode-II cracks in our experiments because the stress magnitude (2 to 4 GPa) is much lower than the confining pressure (15 to 20 GPa). The formation of numerous microcracks along the kink-band boundaries (hereafter, kink-band microcracks) suggests the stress concentration resulting from the pileup of dislocations that could trigger brittle failure (26). The large mode-II cracks can be formed by the coalescence of kink-band microcracks in the damage zones once the crack density becomes high enough to coalesce even at high pressures (27).

Fig. 4. Microstructures of the sample deformed at 1120 K (M3749).

(A) BSE image of the recovered sample. The white arrows indicate the mode-II cracks. (B) Magnified view of the mode-II crack with a damage zone (indicated by blue dashed lines) located along area “B” in (A). The numerous kink-band (KB) microcracks are distributed in the damage zone. (C) TEM image of the damage zone consists of olivine kink bands [located in (B)]. The dashed arrows show the directions of [101] or [111] of the kinked olivine crystals determined from the SAED patterns. White thick arrows represent the compressional direction. (D) Magnified view of the kink-band boundary [in area “D” in (C)]. Thin lamellae parallel to the (100) plane are observed in the kinked olivine grain.

Fault gouge formed at the onset of silent faulting

The formation of faults was also observed in two “silent” runs (M3564 and M3845). The faults are oriented with an angle of 40° to 50° to the compression direction and the displacements range from 40 to 50 μm (Figs. 5 and 6). Formation of large mode-II cracks was not observed (M3564; Fig. 5A) or limited to one (M3845; Fig. 6A) in the “silent” runs, although many large mode-II cracks were developed in the runs in which more than 12 AE events were recorded [M3566 (Fig. 1A), M3747 (fig. S3C), and M3749 (fig. S3D)]. Nucleation of a small amount of wadsleyite and ringwoodite (<2 vol %) with grain sizes less than 1 μm was observed on olivine grain boundaries in the off-fault intact areas of the M3564 sample deformed at 1170 K (Fig. 5). The maximum thickness of the damage zones (corresponding to the gouge layers) is ~50 μm, and the zones consist of pulverized grains of olivine with diameters of 5 μm or less in the M3564 sample. A small amount of wadsleyite and ringwoodite (<2 vol %), with grain sizes less than 1 μm, were also observed in the damage zones, similar to the off-fault intact area.

Fig. 5. Microstructures of the sample faulted at 1170 K (M3564).

(A) BSE image of the recovered sample. The throughgoing fault is highlighted by the red dashed line. (B) Magnified image of the intact area [located in area “B” in (A)]. Light gray grains are wadsleyite or ringwoodite (Wad/Rwd). (C) Fault and the associated damage zone (surrounded by white dotted lines). (D) Pulverized olivine grains in the damage zone [located in area “D” in (C)].

Fig. 6. Microstructures of the sample faulted at 970 K (M3845).

(A) BSE image of the recovered sample. The throughgoing fault is highlighted by the red dashed line. (B to D) Magnified images of the fault. FIB foils were prepared from the blue rectangular areas in (C) and (D). (E) TEM image of the thick gouge layer in (C) and a typical SAED pattern of the nanograins filling the gouge layer. The spots corresponding to a d-spacing of ~8 nm (indicated by the white arrows in the SAED pattern) suggest the existence of wadsleyite and/or ahrensite nanograins. (F) TEM image of the thin gouge layer (in D) developed along a kinked olivine grain. The SAED pattern was taken from the kinked olivine grain. The white arrows show the kink-band boundaries. The black dashed arrow shows the [100] direction. (G) Chemical compositions (iron and silicon in wt %) of olivine polymorphs (triangles), silicon-rich patches (squares), and the iron-rich phase (circles) in the thick gouge layer (in C). The blue dashed lines are the mixing lines between olivine polymorphs and stishovite (Sti).

Neither micrometer-scale wadsleyite nor ringwoodite was found in the M3845 sample deformed at 970 K (Fig. 6). The maximum thickness of the gouge layer is about 1 μm in the M3845 sample. Developments of damage zones were absent around the fault in the M3845 sample, which may partly contribute to the occurrence of silent faulting in run M3845. Two FIB foils were fabricated from the gouge layer developed in the central part of the M3845 sample (Fig. 6, C and D). TEM micrographs of the gouge layer show that the gouge is filled with nanograins of round-shaped equigranular olivine grains (diameters of 20 to 50 nm) and some platinum blobs (diameters of less than 200 nm) (Fig. 6E). The existence of platinum blobs strongly suggests the melting of the platinum strain marker at temperatures higher than ~2500 K (28), which is more than 1500 K higher than the background temperature (970 K).

The averaged concentrations of iron and silicon in kinked olivine grains are FeO = 10.2 wt % and SiO₂ = 41.0 wt %, respectively (Fig. 6G). However, many olivine nanograins are depleted in iron (FeO ≤ 9 wt %). A few particles are enriched in iron (FeO = 18 to 21 wt % and SiO₂ ~ 44 wt %), showing that the iron loss from olivine nanograins is compensated for by the formation of the iron-rich phase. Silicon concentrations in the analyzed nanograins continuously vary from SiO₂ = 41 to 72 wt %, resulting from the contamination of stishovite (SiO₂ = 100 wt %) particles in the gouge layer. The combination of an iron-rich and silicon-rich phases was also reported in fault gouges developed in olivine at pressures of ~15.5 GPa with a background temperature of 1160 K (19). We consider that the observed iron-rich and silicon-rich phases associated with the depletion of iron in olivine nanograins in the gouge layer (Fig. 6G) imply the occurrence of incongruent melting of olivine polymorphs at temperatures of >2400 K at 17 GPa (19, 29). Unfortunately, particle sizes of the iron-rich/silicon-rich phases were too small (<20 nm) to constrain their crystallographic structures. The possibility that the iron-rich/silicon-rich phases are a metastable phase or a quench product cannot be excluded.

The SAED spotty ring patterns obtained from the nanograined areas contain spots with a d-spacing of ~0.8 nm (Fig. 6E), implying the existence of wadsleyite (30) and/or γ-Fe₂SiO₄ ahrensite (31) in the gouge layer. Note that a d-spacing of ~0.8 nm cannot be explained by other phases (i.e., olivine, ringwoodite, fayalite, stishovite, and iron oxides). The thinner gouge layer (located on the middle-upper area of the sample) consists of olivine nanograins with diameters of 5 to 10 nm (Fig. 6F). Many kink bands (width ~50 nm) parallel to the olivine [100] direction developed in old olivine grains around the gouge layer (Fig. 6F).

DISCUSSION

Pseudo-martensitic transition of olivine to ringwoodite via poirierite

It has been considered that formation of weak gouge layers filled with nanograins of high-pressure phases is essential for the occurrence of transformational faulting because grain size–sensitive (GSS) creep of nanograins of olivine polymorphs induces shear localization (6, 14, 15, 19). Our results on run M3845 (Fig. 6) are consistent with the model of transformational faulting aided by adiabatic shear instability (19), although AEs were hardly detected in this run [note that many AEs were associated with the faulting in previous studies (14, 15, 19)]. On the other hand, the gouge layer consists of angular-shaped ultrafine-grained olivine and ringwoodite (diameters down to 200 nm) in run M3566 (Fig. 3), suggesting the absence of adiabatic shear instability throughout the faulting. Hereafter, we discuss the process of faulting in run M3566 from the viewpoint of the pseudo-martensitic transition of olivine to ringwoodite.

Madon and Poirier (23) predicted that phase transitions of olivine to ringwoodite proceed via an intermediate structure (poirierite) when shear deformation of a single-crystal olivine proceeds (// [010]_Ol direction on the (100)_Ol plane). Such a diffusionless transition is expected to be insensitive to temperature (32), and thus formation of high-pressure phases can effectively proceed even at low temperatures. Recently, Okuchi et al. (33) conducted a series of shock compression experiments on single crystals of mantle olivine at room temperature and showed the occurrence of the olivine-ringwoodite transition via a shear mechanism within a few tens of nanoseconds.

In general, formation of kink bands is the result of shear localization in a single crystal [local shear stress // kink-band boundary; (34)]. Shi et al. (35) reported formation of many kink bands due to the activation of the [001](010) slip system in Mn₂GeO₄ olivine at 4 to 5 GPa and 673 to 1323 K. We showed that kinking (i.e., shear deformation) of olivine single crystals via the [010](100) slip system results in formation of poirierite grains with colinear relationships of [010]_Ol // [010]_Poi and [100]_Ol // [101]_Poi (Figs. 2C and 7). Burnley (36) also reported formation of spinel lamellae parallel to the olivine (100) plane in deformed Mg₂GeO₄ germanite olivine grains. Successive kinking of poirierite single crystals via the [101](010) slip system of poirierite induces another pseudo-martensitic transition from poirierite to ringwoodite (23) (Fig. 7). Origin of the observed ringwoodite grains having the same crystallographic orientation (Fig. 3, B to D) is explained by the shear transition of an old olivine single crystal to the ringwoodite grains via the poirierite structure. Recently, a phase transformation of mantle olivine to ringwoodite at 15 to 28 GPa and room temperature has been reported as a result of severe shear deformations (37). Volume change via the olivine-ringwoodite transition [−8.5%; (38)] would also contribute to nucleation/growth of kink-band microcracks. Kinking of olivine crystals is inevitably associated with nucleation/growth of kink-band microcracks because a propagating kink band and a mode-II shear crack have a similar strain field in front of the kink/crack tip (39). Formation of kink bands accompanying the propagation of kink-band microcracks has been commonly reported in hexagonal alloys (40, 41) and orthorhombic metals/alloys (42) deformed at ambient pressure and room temperature.

Fig. 7. Process of the pseudo-martensitic transition of olivine to ringwoodite via poirierite under the conditions of the deep subducted slab.

Localization of shear strain (black arrows) to kink bands induces the pseudo-martensitic transition of olivine. Ol, olivine; Poi, poirierite; Rwd, ringwoodite.

The formation of olivine kink bands by the activation of some slip systems [other than [010](100)] was also observed in our experiments, although pseudo-martensitic transitions of olivine did not proceed in the case that the direction of olivine [010] is oriented with a high angle to the kink-band boundary (Figs. 3A and 4C). These observations also support the validity of the model by Madon and Poirier (23).

Superheating by the transition of poirierite to ringwoodite

It has been reported that poirierite maintains the highest enthalpy among olivine polymorphs at a wide range of pressures [at least up to 20 GPa; (22)]. A net temperature increment across the phase transition in simple isobaric heating ΔT is roughly described as (43)

$$∆T \sim \frac{-∆H}{C_p}$$ (1)

where ΔH is the relative enthalpy, and C_p is the isobaric heat capacity. Tomioka et al. (22) reported that the value of ΔH for the olivine-poirierite (ΔH_ol-poi) is positive, and thus superheating is not expected in the olivine-poirierite transition. The ΔT for the direct olivine-ringwoodite transition (ΔH_ol-rwd < 0) is about ~200 K at 20 GPa (38, 44). A much higher temperature increment is expected in the poirierite-ringwoodite transition (Fig. 7) because not only ΔH_ol-rwd but also ΔH_ol-poi contribute to the increment of ΔT [= −(ΔH_ol-rwd − ΔH_ol-poi)/C_p] in the transition. Considering that the absolute value of ΔH_ol-poi is estimated to be higher than that of ΔH_ol-rwd at 20 GPa (22), weakening of the damage zone would proceed drastically if the pseudo-martensitic transition from poirierite to ringwoodite proceeds in the gouge layer.

Shear localization induced by the olivine-ringwoodite transition

Uniaxial/triaxial strain accommodated by the slip of a fault (ε_slip) is expressed as

$${\mathrm{\epsilon }}_{\text{slip}}=\text{ln}\left(\frac{l}{l-D·\text{cosθ}}\right)$$ (2)

where D is the fault displacement, θ is the dip angle of the fault, and l is the sample length. The ε_slip is estimated to be ~0.03 for run M3566. The discontinuous increase in strain (Δε = 0.05) at the time of faulting (41 to 42.5 min; Fig. 1A) can be explained by the unstable slips of the main fault and another minor fault (Fig. 2A) under the assumption that the sample strain was constant during the blowout in run M3566. On the other hand, the estimated values of ε_slip (~0.03) for runs M3564 (Fig. 5) and M3845 (Fig. 6) are attributed to the “silent” stable slips of faults that are presumed to proceed continuously for 13 to 27 min during softening (fig. S3, A and B). Note that the values of t_slip (= 13 to 27 min) for runs M3564 and M3845 might be overestimated because we ignored the contribution of unstable slips to the values of ε_slip.

Dynamic shear strain rates of the fault can be roughly estimated by the equation ${\dot{\mathrm{\gamma }}}_d$ = D/(H∙t_slip), where H is the thicknesses of the gouge layer, and t_slip is the duration of the fault slip. Substituting the parameters from Figs. 1 and 2 (D ~ 50 μm, H ~ 10 μm, and t_slip < 100 s) into the equation, the ${\dot{\mathrm{\gamma }}}_d$ is estimated to be >5 × 10⁻² s⁻¹ for the fault slip in run M3566 (Fig. 8A). Similarly, the values of ${\dot{\mathrm{\gamma }}}_d$ are roughly estimated to be ~2 × 10⁻³ and ~3 × 10⁻² s⁻¹ for runs M3564 and M3845, respectively (Fig. 8, A and B). The relatively high strain rate in run M3845 results from the thickness of the gouge layer in run M3845 (H ~ 0.05 to 1 μm).

Fig. 8. Conditions for localized heating in metastable olivine at pressures of 15 to 17 GPa.

(A) Background temperature of 1120 K (M3566: red) or 1170 K (M3564: blue) and a grain size of 200 nm. (B) Background temperature of 970 K and a grain size of 50 nm (M3845: blue). The thick solid curve (pink) shows the critical strain rates for adiabatic instability. The thin curves represent the creep strengths of ultrafine-grained olivine controlled by diffusion creep (green), superplasitcity (purple), or the Peierls creep (black: at background temperature; gray: superheated) without (ΔT = 0 K: solid) and with (ΔT = 200 K: dot-dashed) the effect of superheating. Our experimental results are shown by the blue squares (A: M3564; B: M3845) and the red triangle (A: M3566). Crosses show the estimated strain rates at the time of throughgoing faulting (see text).

Here, we compare the background strain rates and the estimated fault slip rates with the steady-state creep strength of olivine (in simple-shear geometry). At low temperatures and high differential stresses corresponding to the present experimental conditions, deformation of coarse-grained olivine is expressed by the following exponential flow law for triaxial deformation (45)

$$\dot{\mathrm{\epsilon }}=A_t·\text{exp}[-\frac{E^{\ast }+P{V}^{\ast }}{\mathit{RT}}{(1-\frac{\mathrm{\sigma }}{{\mathrm{\sigma }}_p})}^2]$$ (3)

where $\dot{\mathrm{\epsilon }}$ is the strain rate, A_t is the pre-exponential constant, E* is the activation energy, P is the pressure, V* is the activation volume, R is the gas constant, T is the temperature, σ is the steady-state creep strength, and σ_p is the Peierls stress. The σ_p at high pressures was calculated from the Peierls stress at ambient pressure (σ_p⁰ = 9.1 GPa) and shear modulus (46). We used the reported flow-law parameters for dry olivine (46, 47) (table S2). We assume the relationship for shear strain rate $\dot{\mathrm{\gamma }}=2\dot{\mathrm{\epsilon }}$ and shear stress τ = σ/2 for simple-shear geometry in Fig. 8 (A and B).

Weakening of the gouge layer is expected due to grain size reduction via pulverization and dynamic recrystallization. It has been shown that the relationship between the dynamically recrystallized grain size (G_r in mm) and the differential stress (σ in MPa) for deformed olivine rocks is empirically given by the following equation (48)

$$G_r=15\times {\mathrm{\sigma }}^{-1.33}$$ (4)

Substituting a representative value for the differential stress in our experiments (σ = 3000 MPa) into Eq. 4, G_r is obtained to be ~350 nm. This value is comparable to the diameters of cell structures formed by tangled dislocations (200 to 500 nm; Fig. 3A) and grains in the gouge layer developed in the M3566 sample (typically ~200 nm; Fig. 3C). However, dynamic recrystallization of olivine cannot account for the origin of the ultrafine-grained gouge layer in the M3845 sample (~50 nm; Fig. 6E).

Steady-state creep strength of nanocrystalline aggregates is described as the power-law equation given as

$$\dot{\mathrm{\gamma }}=A_s\frac{{\mathrm{\tau }}^n}{G^p}\text{exp}(-\frac{E^{\ast }+P{V}^{\ast }}{\mathit{RT}})$$ (5)

where A_s is a pre-exponential constant for shear deformation [= 3^(n+1)/2A_t; (49)], n is the stress exponent, G is the grain size, and p is the grain size exponent. In the case of diffusion creep (i.e., n = 1), Eq. 5 is rewritten as follows

$$\dot{\mathrm{\gamma }}=13.3\times 3^{(n+1)/2}\frac{\mathrm{\tau }}{G^2}[D_{0,l}\text{exp}(-\frac{H_l^{\ast }}{\mathit{RT}})+\frac{\mathrm{\pi \delta }}{G}{D}_{0,\mathit{gb}}\text{exp}(-\frac{H_{\mathit{gb}}^{\ast }}{\mathit{RT}})]\frac{\mathrm{\Omega }}{\mathit{RT}}$$ (6)

where D₀ is the pre-exponential factor (l is the lattice diffusion; gb is the grain boundary diffusion), H^* is the activation enthalpy for diffusion, δ is the width of the grain boundary, and Ω is the molar volume. We consider two GSS creep mechanisms for olivine [superplasticity (50) and diffusion creep (51, 52)] (table S2), which are expected to be effective in the deep subducted slabs (19). In the calculations, we consider a gouge layer (or damage zones) with a grain size of 200 nm (Figs. 3C and 5D) or 50 nm (Fig. 6E) with/without the effect of superheating via the direct olivine-ringwoodite transition [ΔT = 200 K; (38, 44)] at a background temperature (T_bkg) in the range of 970 to 1170 K. We do not consider the case of superheating via the poirierite-ringwoodite transition because the value of C_p for poirierite is unknown.

Figure 8 (A and B) shows that the Peierls creep is effective (i.e., GSS creep is ineffective) at low background temperatures. The background strain rates of olivine are consistent with the values calculated from the Peierls creep (within the error bars) in the runs M3564, M3566, and M3845. The estimated fault slip rates for run M3566 (>5 × 10⁻² s⁻¹ in the case of T_bkg = 1120 K; Fig. 8A) and M3845 (3 × 10⁻² s⁻¹ in the case of T_bkg = 970 K; Fig. 8B) can be explained by the Peierls creep if we assume a superheated condition of ΔT = 200 K. A relatively low fault slip rate for run M3564 (2 × 10⁻³ s⁻¹ in the case of T_bkg = 1170 K; Fig. 8A) is marginally consistent with the Peierls creep at 1320 K (i.e., a superheated condition with ΔT = 150 K). On the other hand, both the GSS creep mechanisms cannot account for the fault slip rates in the three runs even if we assume the superheated condition. The Peierls creep, which has a high activation energy and a high activation volume (table S2), can accelerate deformation effectively during superheating. In summary, transition in the dominant creep mechanism from the Peierls creep to GSS creep is not a necessary condition for the weakening of the fault gouge.

Ohuchi et al. (19) discussed that the initiation of adiabatic instability plays an important role in transformational faulting. In Fig. 8 (A and B), the critical shear strain rate, $\dot{{\mathrm{\gamma }}_c}$, for the initiation of adiabatic instability was calculated from a theoretical equation (49)

$${\dot{\mathrm{\gamma }}}_c \sim \frac{h\mathrm{\rho }{c}_p\mathrm{\kappa }R{T}^2}{\mathrm{\tau }{L}^2(E^{\ast }+P{V}^{\ast })}$$ (7)

where h is the work hardening coefficient (= 1), ρ is the density (= 3.6 g cm⁻³ at 13 GPa) (53), c_p is the specific heat (= 817 J kg⁻¹ K⁻¹) (54), κ is the thermal diffusivity (= 0.7 mm² s⁻¹) (55), τ is the shear stress (= σ/2), L is the sample size (= 1.2 mm in length for the present study), E^* is the activation energy for deformation (= 502 kJ mol⁻¹ for the Peierls creep) (47), and V^* is the activation volume for deformation (= 30 cm³ mol⁻¹ for the Peierls creep) (46). The fault slip rate for run M3845 is above threshold values (Fig. 8B), supporting the occurrence of flash heating from the background temperature of 970 to ~2500 K or above at the onset of faulting associated with the melting of the platinum strain marker (Fig. 6E). Although the fault slip rate for run M3566 is also above the threshold value for the initiation of adiabatic instability (Fig. 8A), the actual peak temperature during the flash heating is unknown in run M3566 due to the absence of platinum markers in the gouge layer (Figs. 2 and 3).

Deep seismicity in bending regions of subducted slabs induced by the pseudo-martensitic transition of olivine

We conclude that the observed faulting was caused by the accumulation of damage (i.e., kinking of olivine and propagation of kink-band microcracks) to a localized area followed by the pseudo-martensitic transitions of olivine to ringwoodite via poirierite. A recent experimental study on Mn₂GeO₄ olivine (35) showed that kink-band boundaries are the preferred sites for nucleation of the high-pressure phase. As pointed out by Shi et al. (35), formation of kink-band boundaries is mandatory for the initiation of shear localization. However, nucleation of high-pressure phases is controlled by diffusion processes and thus less active at low temperatures in deep subducted slabs. On the other hand, the quasi-diffusionless pseudo-martensitic transition of olivine to ringwoodite (i.e., very low activation energies for nucleation of ringwoodite) is effective even at low temperatures (22, 33), corresponding to the central coldest part of the Mariana slab [depths of 630 km and 820 K; (56)].

It has been speculated that deep earthquakes nucleate within the rim of MOW (9, 57). This speculation would be justified if the formation of nanograins of wadsleyite/ringwoodite in the gouge layers is controlled by grain-boundary nucleation, which is accelerated at high temperatures (19). The hypocenters of deep earthquakes are distributed within the rim of MOW [corresponding to the warmer regions >973 K; (9)] and the cold core of MOW is aseismic in the subduction zones of Tonga (9), Izu-Bonin (57, 58), and southwest Japan (59). In contrast to the cases in those subduction zones, many deep earthquakes occur in the cold core of MOW in the Mariana subduction zone (56, 60). Considering that the strain rate (proportional to the probability of shear localization associated with the kinking of metastable olivine) is estimated to be elevated in the Mariana slab at depths of ~600 km (11), the pseudo-martensitic transition of olivine to ringwoodite can account for the cause of deep earthquakes in the deep Mariana slab.

The pseudo-martensitic transition of olivine to ringwoodite observed in this study is not the only transition possible; a transition to wadsleyite via poirierite is also possible (22, 23). The pseudo-martensitic transitions (i.e., shear transitions) of olivine are consistent with the fact that deep earthquakes require sufficient strain accumulation in bending regions of subducted slabs (11). Nucleation of deep earthquakes needs to be considered from the viewpoints of both thermal (i.e., nucleation of wadsleyite/ringwoodite) and mechanical (i.e., pseudo-martensitic transition) processes in the MOW.

MATERIALS AND METHODS

Starting materials

Olivine aggregates used in deformation experiments were prepared from gem-quality crystals of San Carlos olivine that has no inclusions. These crystals were finely ground for 2 hours in an agate mortar. The fine-grained olivine was placed into a nickel capsule (inner diameter: 8 mm; length: 11 mm) and sintered at 4 GPa and 1220 K for 1.5 hours by using a Kawai-type multianvil apparatus at Geodynamics Research Center (GRC), Ehime University. The entire cell assembly for the synthesis of olivine aggregates was dried at 383 K in a vacuum oven before the sintering experiment to remove moisture from the cell assembly. The average grain size of olivine in the recovered samples is typically 10 to 20 μm. The recovered sample was cut then core drilled with a diameter of 1 mm and a length of 0.6 mm.

In situ deformation experiments

The design of the cell assembly for deformation experiments is based on Ohuchi et al. (19), although cobalt-doped magnesia was replaced by chromium-doped magnesia for the material of pressure medium to improve the efficiency of pressure generation (fig. S1A). A cubic pressure medium (5 mm edge length) made from semisintered chromium-doped magnesia was surrounded by six second-stage anvils with a 3-mm truncation edge length. Five second-stage anvils were made from tungsten carbide. An x-ray transparent second-stage anvil made from sintered diamond anvil was placed on the downstream side. Two of the four slide guide blocks placed on the downstream side have a conical x-ray path with a maximum 2θ angle of ~10°. Two core-drilled sintered olivine samples were sandwiched between two hard-alumina pistons (diameter: 1.0 mm; length: 0.3 mm) in a molybdenum capsule (outer diameter: 1.1 mm; wall thickness: 0.05 mm; length: 1.8 mm). Surfaces of the hard-alumina pistons were coated with platinum to prevent chemical reaction between pistons and samples. Each piece of olivine sample and hard-alumina piston were separated by a platinum foil strain marker (diameter: 0.8 mm, thickness: 10 μm). The capsule was surrounded by a magnesia sleeve, which was put into the inner bore of a boron nitride composite heater (TiB₂ + BN + AlN). The heater was inserted into a LaCrO₃ thermal insulator. The x-ray path (with a diameter of 1.4 mm) in a cell assembly was replaced by a mixture of amorphous boron and epoxy resin (at a ratio of 4:1 by weight). Thermocouples were not used in the experiments because it could damage the top of the sintered diamond anvil at high temperatures. The temperature was estimated from the relationship between the temperature and the power calibrated by Ohuchi et al. (19).

On the basis of the protocols in Ohuchi et al. (19), we adopted three pressure-temperature-time (P-T-t) paths for our experiments (fig. S1B). In the path #1 run (M3747), the main ram load was increased to a desired value (1.1 MN: corresponding to P ~ 18 GPa) at room temperature (cold-compression stage). After the cold-compression stage, the temperature was increased to 1120 K, and then the sample was deformed by advancing both the upper and lower first-stage anvils with a constant stroke rate (4 μm min⁻¹) (deformation stage). In the path #2 and #3 runs, the main ram load was increased to 0.55 MN at room temperature and the temperature was raised to 1470 K at a rate of ~50 K min⁻¹. The temperature was kept at 1470 K (corresponding to the olivine stability field at P = 9 to 10 GPa), and then the sample was annealed for 20 min to reduce the densities of cracks and dislocations formed by cold compression (annealing stage). After the annealing stage, temperature was decreased to 870 K or room temperature and then the main ram load was increased to a desired value (0.9 to 1.35 MN: P = 15 to 20 GPa). The maximum generated pressure was limited to ~16 GPa in earlier runs (M3564, M3565, and M3566), due to low-pressure generation efficiency (because the main ram load was increased at 870 K). To improve the pressure generation efficiency, the compression stage was performed at room temperature in path #2 (M3749, M3845, M3989, M3991, and M4130) and path #3 runs (M3748 and 4127). The temperature was again increased to a desired value (870 to 1320 K) at a rate of ~50 K min⁻¹ just before the deformation stage. In the deformation stage, the sample was deformed by advancing both the upper and lower first-stage anvils with a constant stroke rate (1 to 4 μm min⁻¹). The main ram load was gradually increased with a rate of +0.1 MN hour⁻¹ to avoid the depressurization during the deformation stage. The samples were deformed at a constant temperature in the path #2 runs. The procedure before the deformation stage for the path #3 run is the same as that for the path #2 runs, although the temperature increased at a rate of ~30 K min⁻¹ during the middle deformation stage. The ratios of the temperature ramping rate and the strain rate in the path #3 runs (T/ε = 1.0 to 1.4 × 10⁴ K) are comparable to those in natural subduction zones (2).

X-ray data analysis

X-ray radiographic images were taken with 3 s of exposure time using monochromatic x-rays (energy of 60 keV) at BL04B1, SPring-8. The length of a deforming sample was measured from the distance between the platinum strain markers in the radiographic images (fig. S2A). The natural strain ε of the deforming sample was calculated by ε = −ln(l/l₀), where l is the length of the sample during deformation and l₀ is the initial length of the sample. The uncertainties in strain are within 10% (equivalent to the thickness of the strain marker). The central strain marker placed between the olivine samples was used as an indicator of the occurrence of throughgoing faulting.

Two-dimensional radial diffraction patterns of monochromatic x-rays were taken with 70 to 110 s of exposure time using a WidePix 5×5 imaging detector with CdTe sensors (Advacam Co.) (fig. S2B). The acquired half-cycle radial diffraction patterns were subdivided into 36 sectors having a uniform azimuth angle of 5°. The position of olivine peaks (021, 101, 130, and 131) in subdivided sectors were semiautomatically determined to evaluate the pressure and differential stress (fig. S2, C and D). The pressure was determined from the unit-cell volume using an equation of state of olivine (61). The differential stress was evaluated from the azimuth angle dependency of the d-spacing (62) using olivine single-crystal elastic constants (53, 63). The uncertainty of the stress resulting from the accuracy of the measurements is defined as the 1-sigma in the least-squares fit of the Singh’s theoretical curve of the azimuth angle dependency of the d-spacing (62).

AE monitoring and analysis

The setup of AE measurements is the same as in our previous study (19). AE events were detected by six piezoceramic lead-zirconate-titanate (PZT) transducers glued on the sides of second-stage tungsten carbide anvils. Each transducer with a size of 4 mm in diameter and thickness of ~0.5 mm (resonant frequency of ~4 MHz) was electrically isolated from the second-stage anvil by a doubly mirror-polished alumina plate with a thickness of ~0.5 mm. We did not use any broadband-type transducers for accurate determination of AE hypocenter locations. AE signals detected by each transducer were first amplified by an ultrasmall amplifier (20 dB gain) located at ~150 mm from the transducer. The amplified signals were further amplified by a low-noise preamplifier (30 dB gain). Amplified signals exceeding a threshold value (>100 mV of the maximum amplitude) were recorded by a high-speed multichannel waveform recorder at a sampling rate of 100 MHz (16 bit, 8192 samples). The P-wave arrival time was automatically picked on the basis of the Akaike Information Criterion Approach for the obtained waveforms. The hypocenter location of each AE event was determined by the difference in arrival time between the first and other transducers (64). Following the definition of the body wave earthquake magnitude, the relative magnitude (M_ae) of an AE event is determined according to the logarithm of the maximum amplitude of the AE signal (V_max) detected on a chosen channel, that is, M_ae = log₁₀V_max. See Ohuchi et al. (19) for the detailed procedures for the calculations of hypocenter locations and uncertainties.

Microstructural observations and phase identification

Recovered samples were cut with a low-speed saw (with a thickness of 0.15 mm) and mounted in epoxy resin. The surfaces of the mounted samples were polished using alumina powder followed by a colloidal silica suspension. The mirror-polished surfaces of the samples were examined using a field-emission scanning electron microscope (FE-SEM; JEOL JSM-IT500HR) equipped with an energy-dispersive x-ray (EDX) detector installed at GRC, Ehime University. The FE-SEM was operated at an accelerating voltage of 15 kV and a beam current of 1 nA.

Nanostructures of three samples (M3566, M3749, and M3845) were also examined using a transmission electron microscope (TEM) (JEOL JEM-2010) or a field-emission scanning TEM (STEM; JEOL JEM-2100F) equipped with an EDX detector (at GRC, Ehime University) operated at an accelerating voltage of 200 kV. Thin foils (thickness of ~100 nm) of the desired areas in the samples were prepared using FIB systems (FEI Scios and Scios2) at an accelerating voltage of 30 kV. The ReciPro software (65) was used for the calculations of theoretical SAED patterns of the minerals examined in TEM micrographs. In particular, phase identification of poirierite (grain sizes of <100 nm) was performed using the observed and theoretical SAED patterns of single crystals (Fig. 2, C and D, and fig. S6). The chemical compositions were determined from the STEM/EDX data using the Cliff-Lorimer equation (66). We used representative chemical compositions of olivine grains analyzed using the FE-SEM/EDX as an internal standard for determining Cliff-Lorimer k-factors. The uncertainty of the calculated chemical composition reflects the error in the counts of characteristic x-rays.

Unpolarized Raman spectra of the recovered samples were acquired using a JASCO NRS-5100gr (intrinsic resolution of ~1 cm⁻¹) installed at GRC, Ehime University. A microscope with a ×50 objective lens was used to focus an incident 532-nm Nd:YAG laser (Neodymium-doped Yttrium Aluminum Garnet laser) into a 2-μm spot on the polished surface of the sample. Spectra were acquired over 120 s using a laser power of 5.4 mW. Olivine, wadsleyite, and ringwoodite having grain sizes larger than 2 μm were identified on the basis of the existence of peaks [at 824, 855, 920, and 961 cm⁻¹ for olivine (67), 719 and 916 cm⁻¹ for wadsleyite (68), and 796 and 836 cm⁻¹ for ringwoodite (69)].

Supplementary Materials

The PDF file includes:

Figs. S1 to S6

Tables S1 and S2

Legend for movie S1

Other Supplementary Material for this manuscript includes the following:

Movie S1