Corresponding author: J. Nakajima, Research Center for Prediction of Earthquakes and Volcanic Eruptions, Graduate School of Science, Tohoku University, Sendai 980-8578, Japan. (firstname.lastname@example.org)
 Using quantitative analysis of waveforms, we investigate a seismic cluster that occurred at a depth of 67 km in the Philippine Sea slab 8 months after the 11 March 2011 megathrust Tohoku-oki earthquake (Mw 9.0). The sequence started with an M 4.1 normal-fault event on 14 November in the deepest part of the cluster, and subsequent earthquakes migrated upward by 6 km along a narrow conduit-like zone. The earthquakes have stress drops of 0.5–40 MPa, and groups of earthquakes with coherent waveforms are observed. We explain these observations in terms of fluid-related embrittlement and the migration of overpressured fluids. The low-permeability plate interface of the Pacific plate may have been broken by the coseismic or postseismic slips of the Tohoku-oki earthquake, with the fluids subsequently being liberated from the underlying crust of the Pacific slab and migrating into the Philippine Sea slab due to the pore pressure gradient. The tensional stresses generated by the coseismic slip promoted the efficient upward migration of fluids and also acted to enhance the deviatoric stress around the source area. The heightened pore pressures and the resulting reduced effective normal stress lowered the strength of the faults sufficiently to bring the system into the brittle regime under the enhanced deviatoric stress. The lag of 8 months may represent the time needed for the unsealing of the plate interface, the upward migration of fluids, and the increase in pore pressure to become sufficient to overcome the lithostatic pressure.
 Since the discovery of the double Wadati-Benioff zone in subducting oceanic lithosphere [e.g., Hasegawa et al., 1978], much attention has been paid to the genesis of intermediate-depth earthquakes. To date, two important hypotheses, dehydration or fluid-related embrittlement [e.g., Raleigh and Paterson, 1965; Kirby et al., 1996; Hacker et al., 2003] and self-localizing shear instability [e.g., Ogawa, 1987; Kelemen and Hirth, 2007; John et al., 2009], have been proposed to explain seismic ruptures at intermediate depths where high pressures render ordinary dry frictional failure unlikely [see Frohlich, 2006 for details].
 Evidence for dehydration embrittlement was derived originally from laboratory experiments and later from seismological observations. Raleigh and Paterson  first showed that dehydration during deformation led to shear fracturing accompanied by a sudden stress drop, and subsequent high-pressure experiments have shown dehydration embrittlement to be a plausible mechanism for the genesis of intermediate-depth earthquakes [e.g., Dobson et al., 2002; Jung et al., 2004]. Recent high-resolution seismic imaging has revealed a spatial correlation between low-velocity zones where fluids may be present and intermediate-depth earthquakes [e.g., Preston et al., 2003; Tsuji et al., 2008; Nakajima et al., 2009a; Shiina et al., 2013]. Studies of focal mechanism solutions have suggested that intermediate-depth earthquakes occur in association with the reactivation of hydrated faults that formed by bending-related tensional faulting at the outer trench slope [e.g., Jiao et al., 2000; Marot et al., 2012; Nakajima et al., 2011a, 2013]. Furthermore, patterns of intermediate-depth seismicity are closely correlated with the predicted depth range for the dehydration locations of hydrous minerals [e.g., Hacker et al., 2003; Omori et al., 2004; Kita et al., 2006; Abers et al., 2013]. These observations suggest that fluids associated with dehydration reactions likely facilitate the generation of intermediate-depth earthquakes through the promotion of fluid-related embrittlement. However, Chermak and Hirth  recently demonstrated that stick-slip instabilities do not develop during the dehydration of antigorite, raising a debate as to whether dehydration does in fact facilitate brittle failure.
 In contrast, self-localizing shear instability has been proposed primarily on the basis of numerical simulations. Shear heating has been shown to promote the onset of highly localized viscous creep in preexisting, fine-grained shear zones, potentially providing a mechanism for intermediate-depth earthquakes [e.g., Kelemen and Hirth, 2007; John et al., 2009]. Because the shear-heating mechanism operates only in a temperature range of 600–800°C, Kelemen and Hirth  concluded that periodic viscous shear-heating instabilities can account for intermediate-depth earthquakes within a narrow temperature interval similar to that associated with the dehydration of hydrous minerals. The development of narrow pseudotachylyte fault veins, which are formed by rapid melting during a seismic rupture, is considered as one of the manifestations of self-localizing shear instability [e.g., John and Schenk, 2006].
 To assess which candidate mechanism plays the more crucial role in generating intermediate-depth earthquakes, we investigate a seismic sequence in the subducting mantle of the Philippine Sea plate beneath Kanto that was activated after the megathrust Tohoku-oki earthquake (Mw 9.0). Because the stress regime beneath the Japanese Islands has been significantly affected by the large coseismic slip (up to 80 m) of the Tohoku-oki earthquake [e.g., Iinuma et al., 2012], the seismic sequences activated after the Tohoku-oki earthquake function as a natural experiment to study the effects of stress change and subsequent physical processes on the genesis of earthquakes.
 Dual subduction of oceanic plates has resulted in the highly complex geological setting beneath Kanto (Figure 1) [e.g., Nakajima et al., 2009b; Uchida et al., 2009], but we hypothesize that earthquakes should be facilitated by a physical mechanism that works everywhere at intermediate depths. Therefore, the generation process of the seismic sequence to be discussed in this study should be applicable to intermediate-depth earthquakes in various tectonic settings, even though the boundary conditions of the tectonic framework could vary region by region.
2 Seismic Sequences Activated After the Tohoku-oki Earthquake
 The Mw 9.0 Tohoku-oki earthquake occurred on 11 March 2011, along the boundary of the Pacific plate east of northeastern Japan, and was accompanied by coseismic slips as large as ~80 m (Figure 1) [Iinuma et al., 2012]. Many aftershocks followed the earthquake [e.g., Hirose et al., 2011; Koper et al., 2011] and intensive seismicity was activated outside the aftershock area (indicated by broken gray lines in Figures 2b and 2d) through dynamic triggering [e.g., Miyazawa, 2011; Yukutake et al., 2011] and/or static triggering [e.g., Okada et al., 2011; Toda et al., 2011]. The activated seismicity included three damaging shallow crustal earthquakes (M 6.7 on 12 March, M 6.4 on 15 March, and M 7.1 on 11 April); the first two occurred ~400 km away from the epicenter of the Mw 9.0 earthquake (Figure 2b). In many cases, seismic sequences were initiated hours to days after the Tohoku-oki earthquake, and the hypocenters migrated with time [e.g., Hirose et al., 2011]. These shallow-crustal sequences are interpreted as having been activated as a result of the stress change caused by the large coseismic slip of the Tohoku-oki earthquake [e.g., Asano et al., 2011; Hasegawa et al., 2011, 2012; Kato et al., 2011; Toda et al., 2011; Yoshida et al., 2012].
 After the Tohoku-oki earthquake, the rates of seismicity increased beneath Kanto (Figures 2d and 3). The most marked activity was in a belt-like area that extends northward from Tokyo Bay (Figure 3b, square brackets). These earthquakes were mostly interplate earthquakes occurring along the upper surfaces of the Philippine Sea and Pacific slabs [Ishibe et al., 2011]. Interestingly, the rates of seismicity increased even in the Philippine Sea slab, and three somewhat isolated seismic clusters (clusters A–C) were activated along a belt-like zone that extends to the south from the southernmost part of the Boso Peninsula (Figure 3b). This area had experienced seismic activity prior to the Tohoku-oki earthquake (Figure 3a), but the rates of seismicity increased markedly after the earthquake. The static stress change induced by the Tohoku-oki earthquake affected the stress regimes over a wide region beneath Kanto (Figure 1), and hence, additional factors must have facilitated the seismic activity in this particular area of the southern part of the Boso Peninsula.
 This paper focuses on seismic cluster A of the three clusters, because this cluster is located beneath land and the accuracy in determining the hypocenter locations is therefore better than for the other two clusters. Cluster A is located in the lowermost part of the subducting mantle of the Philippine Sea slab and the source area is concentrated in a small space with dimensions no larger than 10 × 10 × 10 km (Figure 3b). A magnitude versus time (MT) plot shows that the earthquakes in cluster A occurred rather regularly over time before the Tohoku-oki earthquake and that seismicity increased markedly following an M 4.1 earthquake on 14 November 2011 (Figure 4a). The M 4.1 earthquake occurred by normal faulting with the T axis in the NE-SW orientation (Figure 5d). The change in the Coulomb failure function (ΔCFF) [e.g., Stein et al., 1992] due to the Tohoku-oki coseismic slip was positive for the both nodal planes of the M 4.1 earthquake.
3 Data Analysis
 Here we characterize the source properties of the earthquakes in cluster A using high-quality waveform data observed at a nationwide seismograph network in Japan. The analysis includes hypocenter relocations with waveform-derived differential time data, the determination of stress regime around cluster A, estimates of the corner frequencies, source radii, and stress drops of the earthquakes, and the detection of earthquakes with coherent waveforms. These comprehensive approaches provide clues to understand the generation mechanism of the seismic sequence that was activated after the Tohoku-oki earthquake.
3.1 Relative Locations of Hypocenters
 We determined relative hypocenter locations of the earthquakes in cluster A with high accuracy. First, we relocated earthquakes that occurred beneath Kanto from January 2001 to October 2012 with the 1-D seismic velocity model and station corrections of Nakajima et al. [2009b], taking station elevations into account, because it is necessary to consider appropriate station corrections and station elevations when determining the absolute locations of earthquakes beneath Kanto [Nakajima et al., 2009b]. The relocated hypocenters are more accurate in terms of absolute locations than those determined by the Japan Meteorological Agency (JMA), and the focal depth of the M 4.1 event is calculated to be 67 km (Figure 5a), ~6 km above the upper surface of the underlying Pacific plate.
 These relocated hypocenters were then used as the initial hypocenters in the double-difference hypocenter relocation method (hypoDD) [Waldhauser and Ellsworth, 2000]. We applied hypoDD to catalogue-derived differential traveltimes for 358 earthquakes in seismic cluster A. The maximum distance between earthquake pairs was limited to 10 km. The results of the relocations show that the seismic cluster is inclined toward the north (Figure 5b), as is expected from the focal mechanism solution of the M 4.1 event, but the detailed distribution of the seismicity remained still unclear.
 For more precise relative hypocenter relocations, we used hypoDD with waveform-derived differential time data. We selected 69 earthquakes (M ≥ 2) with high signal-to-noise ratios from cluster A for the period from 11 March 2011 to 31 October 2012. Traveltime differences for the 69 earthquakes were calculated at 66 common stations using the cross-spectral method for both P and S waves [Poupinet et al., 1984]. A time window of 3 s was set 0.5 s before the onset of each wave, and delay times were calculated from the phases of cross spectra in a frequency band of 2–12 Hz with a coherency of ≥0.8. This analysis yielded 12,609 and 5814 traveltime differences for P and S waves, respectively.
 Although the initial hypocenters for the 69 earthquakes, which were relocated with the 1-D velocity model and station corrections, show a scattered distribution (Figure 5c), the hypocenters relocated with waveform-derived differential data form a conduit-like cluster dipping toward the NNE, as shown in Figure 5d. The hypocenters lie on a plane with a dip that is consistent with the low-angle nodal plane for the focal mechanism of the M 4.1 event, and the hypocenter of the M 4.1 event is located in the deepest part of the inclined seismic cluster. This seismic activity can be divided into three subclusters on the basis of hypocenter distributions. Subcluster 1 is composed mainly of earthquakes that occurred immediately after (<20 days) the M 4.1 event (Figure 5e). Subcluster 2 is located in the deepest part of cluster A, but it includes earthquakes that occurred both immediately after (<20 days) and long after (>100 days) the M 4.1 event. Earthquakes in subcluster 3 are located in the shallower part of cluster A, and they tended to occur along an inclined zone mainly at 60–180 days after the M 4.1 event (Figure 5f). These results indicate that seismicity migrated toward shallower depths by ~6 km over 6 months, after which seismicity rates decayed with time (Figure 4a).
3.2 Focal Mechanisms and Stress Tensor Inversions
 We determined the focal mechanism solutions for earthquakes with M ≥ 2. The polarities of P waves were manually picked from seismograms for earthquakes that are not reported in the JMA unified catalogue. We then calculated the focal mechanisms of earthquakes with 10 or more polarity data using the method of Hardebeck and Shearer . As a result, we obtained 21 focal mechanisms for earthquakes with 2 ≤ M ≤ 3.5 and with quality ranks of A and B.
 We performed stress tensor inversions [Ito et al., 2009] to calculate the stress regime around cluster A both before and after the Tohoku-oki earthquake. For the stress field prior to the earthquake, we used the earthquake focal mechanisms obtained by Nakajima et al. [2011b], and for the stress field after the earthquake, we used the earthquake focal mechanisms obtained in the present study. Because seismicity around cluster A was not particularly active prior to the Tohoku-oki earthquake (Figure 3a), we used earthquakes from a wider area of the lowermost part of the Philippine Sea slab to estimate the stress regime before the Tohoku-oki earthquake. The numbers of focal mechanisms used for the periods before and after the Tohoku-oki earthquake were 18 and 21, respectively.
 The inversion results indicate that the orientations of σ1 and σ3 before and after the Tohoku-oki earthquake do not show significant differences beyond the estimated errors (Figure 6). The static stress change due to the coseismic slip of the Tohoku-oki earthquake [Iinuma et al., 2012] (marked by squares in Figure 6b) shows orientations of σ1 and σ3 similar to those observed before and after the Tohoku-oki earthquake. This means that the deviatoric stress around cluster A was enhanced by the Tohoku-oki fault slip. In contrast, the stress change due to the postseismic slip of the Tohoku-oki earthquake [Ozawa et al., 2011] (marked by triangles in Figure 6b) is not consistent with the observed stress regime after the Tohoku-oki earthquake, suggesting that small amounts of the static stress change by the postseismic slip may not have had any significant effect on the background stress field in this area.
3.3 Corner Frequency, Source Radius, and Stress Drop
 To estimate the source sizes and stress drops of the earthquakes in cluster A, we determined the corner frequencies of the earthquakes using the spectral ratio method. The spectral ratio for a pair of earthquakes recorded at a common station can eliminate the path-dependent effect and therefore provide reliable source model parameters [Frankel and Wennerberg, 1989]. The amplitude spectrum was calculated at each station for the N-S component of a seismogram, using S wave codas with a time window of 3 s taken at twice the S wave traveltimes, and spectral ratios were then calculated for earthquake pairs at common stations. We limited the earthquake pairs to those with a magnitude difference of at least 0.5 to ensure stable measurements of the corner frequencies. The advantage of using spectral ratios for S wave codas, instead of direct waves, is the production of stable estimates for source model parameters [e.g., Mayeda et al., 2007]. To obtain robust measurements, we stacked the spectral ratios for three windows that overlapped by half their duration (Figure 7a) [Imanishi and Ellsworth, 2006; Uchida et al., 2007]. We used an omega-square source model [Brune, 1970] to model the spectral ratio stacked for the available station pairs (Figure 7b). As a result, we obtained the corner frequencies of 57 earthquakes with 2.0 ≤ M ≤ 4.1.
 We estimated the source radius from the estimated corner frequency using the circular crack model [Sato and Hirasawa, 1973], assuming an S wave velocity of 4.5 km/s and a value of 1.9 for the constant C. The stress drops were calculated using the formula of Eshelby . For the calculations, the seismic moment M0 was estimated from the Japan Meteorological Agency (JMA) magnitude (MJMA), with logM0 = 1.5MJMA + 9.1. The estimated corner frequencies for the 57 earthquakes appear to be inversely proportional to the cube root of the seismic moment (Figure 7c). Stress drops are estimated to be 0.5–6 MPa for earthquakes with M ≤ 3.3, and only the three largest earthquakes have stress drops of >10 MPa. Even though the estimates of stress drop depend largely on the estimates of the seismic moment, the calculated stress drops are compatible with those estimated for intermediate-depth earthquakes, 1–10 MPa for Mw 3.7–7.4 earthquakes in Vrancea, Romania [Gusev et al., 2003], an average value of 34 MPa for Mw 4.1–6.7 earthquakes in Mexico [Garcia et al., 2004], and 1–33 MPa for M 2.0–3.2 earthquakes in Japan [Nakajima et al., 2013]. The source radius of the M 4.1 event is estimated to be 420 m, and the source radii for smaller earthquakes range from 140 to 380 m.
3.4 Detection of Earthquakes With Similar Waveforms
 Visual inspection of seismograms revealed that the seismic cluster contains earthquakes with similar waveforms. For quantitative evaluations, we calculated waveform similarity for the 69 earthquakes used in the hypocenter relocations. A time window of 30 s was set for the vertical component, beginning 3 s before the onset of the P wave, and coherency was calculated for each pair of earthquakes in a frequency band of 2–12 Hz for common stations. When the calculated coherencies were larger than 0.9 at three or more stations, we regarded the earthquake pair as part of the same group. In this manner, we categorized 17 earthquakes into six groups with highly coherent waveforms (Figure 8). MT plots of these earthquakes show that with the exception of Group 6, the earthquakes in the same group occurred successively over a short period of time (Figure 9).
 The earthquakes in each group are concentrated in a volume with dimensions smaller than hundreds of meters (Figure 10). For example, the earthquakes in Group 1 lie on a plane that is consistent with one of the nodal planes of the focal mechanism solutions, and they appear to have ruptured complementary portions of the fault plane. Although the numbers of earthquakes in the other groups are too small to discuss the sequences of seismic ruptures in detail, the earthquakes in each group tend to lie on a fault plane that is compatible with one of the nodal planes of the focal mechanism solutions. The earthquakes with coherent waveforms are concentrated around the hypocenter of the M 4.1 event, but their focal depths are slightly shallower than the M 4.1 event (Figure 11).
3.5 Summary of Key Observations
 In summary, the important seismic features of seismic cluster A are as follows:
 The seismicity was activated 8 months after the Tohoku-oki earthquake, following an M 4.1 normal-fault earthquake on 14 November 2011. The M 4.1 event occurred in the deepest part of the cluster at a depth of 67 km and was located 6 km above the upper surface of the Pacific plate. Seismicity following the M 4.1 event migrated upward by 6 km over 6 months, and seismicity rates decayed with time.
 The static stress change calculated from the coseismic slip of the Tohoku-oki earthquake is consistent with the orientations of σ1 and σ3 observed around the seismic cluster both before and after the Tohoku-oki earthquake.
 Static stress drops and source radii for the earthquakes are estimated to be 0.5–40 MPa and 140–420 m, respectively.
 The seismic cluster involves subsets of earthquakes with similar waveforms that occurred over a short time interval, and the earthquakes appear to have ruptured complementary portions of the fault plane.
 Seismic activity in cluster A had occurred regularly prior to the Tohoku-oki earthquake, but seismicity rates increased drastically after that event (Figures 3 and 4a). The activated seismicity is therefore likely to have been affected by physical processes associated with the Tohoku-oki earthquake. Our attention now turns to considering the processes that could have generated seismic activity in the lowermost part of the Philippine Sea plate and the subsequent upward migration of hypocenters.
4.1 Model for the Generation of Intermediate-Depth Earthquakes
 Prior studies have shown that aftershock activity at intermediate depths often spreads out around the fault plane that was ruptured by the main shock, and the extent of aftershocks is generally comparable to the size of the main shock fault plane. Such aftershock activity has been widely observed, for example, for the 1993 Kushiro-oki earthquake (Mw 7.8) [Ide and Takeo, 1996], the 2005 Tarapaca earthquake (Mw 7.8) [Peyrat et al., 2006], the 2005 off-Miyagi earthquake (M 7.0) [Okada and Hasegawa, 2003], and the 2011 off-Miyagi earthquake (M 7.1) [Nakajima et al., 2011a]. These aftershocks were likely facilitated near the fault plane of the main shock by stress disturbance due to coseismic slip.
 The seismicity that followed the M 4.1 earthquake of the present study migrated upward by ~6 km and extended to a much larger area than the fault size of the M 4.1 earthquake (a radius of 420 m) (Figure 11). Aftershock activity at intermediate depths lasts for months to years in some cases, but the migration of aftershocks beyond the main shock fault plane, as observed in this study, has been seldom reported. One previous example is an aftershock sequence that followed an Mw 5.7 earthquake, which occurred at a depth of 113 km in the subducting Nazca plate, where the aftershocks extended along a plane with dimensions of 40 km × 10 km over 24 days following the main shock [Marot et al., 2012]. The extent of the aftershock activity was larger than the rupture area expected for the main shock.
 Self-localizing shear instability can potentially cause repeated earthquake generation via viscous shear heating [Kelemen and Hirth, 2007], but it is unclear whether this mechanism results in the temporal and spatial migration of hypocenters beyond the source area of the main shock. Furthermore, the values of stress drop (500–750 MPa) expected from self-localizing thermal runaway [e.g., Kelemen and Hirth, 2007; John et al., 2009] are much larger than those observed at intermediate depths (< ~40 MPa) [e.g., Gusev et al., 2003; Garcia et al., 2004; Nakajima et al., 2013]. This is true even if about half the stress drop occurs during shear-heating events, and hence, the stress drop estimated from seismic data samples only part of the total relaxation [Kelemen and Hirth, 2007]. Periodic viscous shear-heating instabilities operate in a temperature range of 600–800°C [Kelemen and Hirth, 2007], but this range is higher than predicted for the Philippine Sea slab at the depth of the seismicity observed in this study [e.g., Iwamori, 2000; Seno, 2007]. Therefore, self-localizing shear instability cannot explain many of the seismological observations at intermediate depths, suggesting that it does not play a dominant role in facilitating such earthquakes.
 Temporal-spatial migrations of seismicity can instead be explained by the migration of fluids [e.g., Chen et al., 2012]. With regard to the activated seismicity analyzed in this study, a low-velocity and high-Vp/Vs layer exists immediately below seismic clusters A–C (Figure 3b). The layer is interpreted to be hydrated, overpressured crust of the Pacific slab [e.g., Matsubara et al., 2005; Nakajima et al., 2009b]. If fluids are supplied from the overpressured crust underlying the area of activated seismicity, as discussed further below, they will facilitate seismicity as a result of enhanced pore fluid pressures and reduced effective normal stress, thereby lowering the strength of the rocks sufficiently to bring the system into the brittle regime [e.g., Kirby et al., 1996; Sibson, 2002]. The lower strength of the rocks is also consistent with the observed stress drop of as low as 0.5–40 MPa, which is much smaller than the lithostatic pressures at intermediate depths. These observations suggest that fluid-related embrittlement is a likely candidate for the generation of intermediate-depth earthquakes.
4.2 Effective Upward Migration of Fluids After the Tohoku-Oki Earthquake
 The Tohoku-oki earthquake may have affected the physical states of the crust and mantle beneath Kanto through the coseismic and postseismic slips along the upper interface of the Pacific plate and the resultant static stress change. Even though a coseismic slip of up to 10 m did not propagate toward the south beyond the northeastern limit of the Philippine Sea plate [e.g., Iinuma et al., 2012; Ozawa et al., 2011], small amounts of the coseismic slip may have propagated toward Kanto. Moreover, the upper interface of the Pacific plate beneath Kanto certainly underwent the postseismic slip [e.g., Ozawa et al., 2011; Suito et al., 2012]. The amount of postseismic slip was as much as 0.4 m in the first 14 days and reached 0.8 m over 180 days [Suito et al., 2012].
 One consequence of the large coseismic (>80 m) slip of the Tohoku-oki earthquake is that it caused a static stress change over a wide area beneath the Japanese Islands [e.g., Yoshida et al., 2012]. The large coseismic slip generated subhorizontal tension oriented NE-SW around the area of activated seismicity [e.g., Ozawa et al., 2011; Iinuma et al., 2012]. We propose that these tensional stresses promoted the upward migration of the fluids in the Philippine Sea plate, once the fluids had been supplied from the underlying overpressured crust of the Pacific plate.
 The overpressured crust of subducting lithosphere, which is a phenomenon prevalent in many subduction zones [e.g., Shelly et al., 2006; Audet et al., 2009; Kato et al., 2010], suggests that the low-permeability plate interface is able to capture fluids efficiently in the subducting crust [e.g., Peacock et al., 2011]. A possible and important consequence of large slips along the plate interface would be the large-scale fracturing of the seal of the plate interface. The breaking of the permeability seal after large megathrust earthquakes has been suggested for the Mw 8.0 Antofagasta earthquake in Chile [Husen and Kissling, 2001]. In the present study, the unsealing of the permeability barrier by the coseismic or postseismic or both slips could have promoted the efficient release of fluids from the underlying Pacific plate. As a result, fluid flow would have been preferentially enhanced due to the combined effect of the breaking of the permeability barrier and the tensional stresses generated by the static stress change of the coseismic slip (Figure 12). It is noted that the presence of seismicity in the mantle of the Philippine Sea plate before the Tohoku-oki earthquake suggests that small amounts of fluid had been intermittently escaping from the overpressured crust of the Pacific plate.
 The fluids liberated from the underlying Pacific plate may have migrated upward into the mantle of the Philippine Sea plate due to the pore pressure gradient along preexisting and highly fractured faults. Although fluid flow into the mantle potentially results in the serpentinization of peridotite under low-temperature conditions beneath Kanto [e.g., Iwamori, 2000; Seno, 2007], it is likely that serpentinization was already completely developed along localized preexisting fractures through which fluids have repeatedly migrated over a geological timescale. Therefore, abundant fluids resulting from the increased rates of fluid flow efficiently raised pore pressures to near-lithostatic pressure and lowered the strength of the preexisting faults, resulting in the occurrence of earthquakes in response to the static stress change brought by the Tohoku-oki earthquake. The decay of seismicity rates with time (Figure 4a) probably resulted from a decrease in the rate of fluid supply from the Pacific plate.
 The conduit-like distribution of aftershocks (Figure 5d) can be explained by the localized migration of fluids along narrow preexisting fractures. We interpret that the groups of earthquakes with similar waveforms observed in cluster A are manifestations of complementary seismic ruptures on adjacent patches of the same fault plane, as observed for intermediate-depth earthquakes in the Pacific slab beneath northeastern Japan at a depth of 155 km [Nakajima et al., 2013]. Complementary ruptures can be facilitated by the spatial-temporal migration of overpressured fluids and the resultant reactivation of variously oriented preexisting faults around the narrow conduit-like zone.
 The 8 month lag may represent the time needed for the breaking of the permeability seal, the subsequent upward migration of fluids, and the increase in pore pressure required to overcome the lithostatic pressure. The timing of the unsealing of the plate interface has not been constrained, but we can estimate a lower bound of the permeability in the source area, provided that the coseismic or early-stage postseismic slips broke the permeability seal. Assuming dynamic viscosity of 10−4 Pa s and storage coefficient of 10−10 m−1 as in Husen and Kissling , we obtain the permeability of ~10−19 m2 for the activation of the seismicity at 6 km above the Pacific plate with the 8 month delay as well as for the upward migration of the seismicity by 6 km over 6 months. However, given storage coefficient of 10−6 m−1 [e.g., Ingebritsen and Minning, 2010], the permeability will become 4 orders of magnitude high, providing a possible range of the permeability of 10−15–10−19 m2. This range spans estimates for the permeability of 10−18–10−19 in the continental lower crust [e.g., Ingebritsen and Minning, 2010] and of 10−16–10−17 m2 in the megathrust zone in Chile [Husen and Kissling, 2001]; however, further quantification is difficult because of large uncertainty in the storage coefficient and the timing of fluid migration.
4.3 Mechanism for the Activation of Seismicity in Confined Areas
 The directions and magnitudes of the static stress change induced by the Tohoku-oki earthquake do not vary substantially in space [e.g., Suito et al., 2012], and the low-permeability barrier is likely to have been broken along a wider area of the plate interface where the coseismic and postseismic slips propagated. Under such a situation, it is unclear why seismicity was activated only along a belt-like area (clusters A–C) that extends to the south of the Boso Peninsula (Figure 3b). Here we propose a model for such a localized activation of seismicity in the Philippine Sea slab in terms of the stress regime prior to the Tohoku-oki earthquake.
Nakajima et al. [2011b] showed that stresses with downdip tension or downdip compression are dominant in the subducting mantle of the Philippine Sea slab beneath Kanto. However, an area with subhorizontal σ1 and σ3 exists in the lowermost part of the mantle beneath the southern part of the Boso Peninsula, where σ1 and σ3 are oriented NW-SE and NE-SW, respectively (Figure 6a). This stress regime is approximately the same as the stress generated by the Tohoku-oki coseismic fault slip (Figure 6b). From this, we infer that the coseismic slip preferentially enhanced the existing deviatoric stress and produced a marked increase in seismicity rates in clusters A–C with the assistance of overpressured fluids. In other areas of the subducting Philippine Sea slab, the background stress regime was substantially different from the stress generated by the coseismic slip of the Tohoku-oki earthquake [Nakajima et al., 2011b]. In such cases, the static stress change by the coseismic slip did not enhance the existing deviatoric stress to cause brittle failure, even with lowering of the fault strength by overpressured fluids supplied from the underlying Pacific plate.
 Seismic clusters A–C form a lineation along the southwestern edge of the contact zone between the Philippine Sea and Pacific slabs (Figure 3b). Because the lowermost part of the mantle of the Philippine Sea slab near the edge of the slab contact zone is close to the hot asthenosphere, the slab is likely to have undergone abrupt changes in thermal regimes and frictional properties, as well as in mechanical interactions with the Pacific slab. This tectonic setting may result in a unique background stress regime only in the lowermost part of the Philippine Sea slab, as observed by Nakajima et al. [2011b]. Quantification of the stress regime in the Philippine Sea slab associated with the slab contact zone is important and will be a subject of future studies.
 Using quantitative waveform analysis, we investigated a seismic cluster in the subducting mantle of the Philippine Sea plate that was activated 8 months after the Tohoku-oki earthquake. The seismicity migrated upward by 6 km over ~6 months, and the rate of seismic activity gradually decreased with time. The temporal-spatial migrations of hypocenters, the values of stress drop of as low as 1–40 MPa, and the successive occurrence of earthquakes with coherent waveforms favor fluid-related embrittlement, rather than self-localizing shear instability, as the dominant process in facilitating intermediate-depth earthquakes. Fluids were probably supplied from the underlying crust of the Pacific plate, and the upward migration of the fluids along preexisting fractures was preferentially promoted by the tensional stresses caused by the coseismic slip generated by the Tohoku-oki earthquake. Comprehensive seismological studies in combination with the consideration of realistic, kinematic, and chemical processes will provide a better understanding of the genesis of intermediate-depth earthquakes in subducting oceanic plates.
 We thank I. Katayama and T. Iinuma for fruitful discussions. Constructive and careful reviews by two anonymous reviewers and Mark Behn, the Associate Editor, improved the manuscript. We used the waveform data from the nationwide seismograph network in Japan and obtained the arrival time data from the unified catalogue of the JMA. All the figures were drafted using GMT [Wessel and Smith, 1998]. This work was supported by the Ministry of Education, Culture, Sports, Science, and Technology of Japan, under its Observation and Research Program for the Prediction of Earthquakes and Volcanic Eruptions; by the Global COE Program, Global Education and Research Center for Earth and Planetary Dynamics, Tohoku University; and by JSPS KAKENHI grant 24740300.