The shape of the Philippine Sea Plate subducting beneath western Japan is a crucial factor in understanding earthquakes and volcanic activity. We propose that the subducted plate was split along an extinct ridge due to an abrupt change of subduction direction, followed by elastic deformation of the plate and an accumulation of stress near the ridge. This shape is consistent with receiver function images, the distribution of deep tremor sources, the seismicity and focal mechanisms of intraplate earthquakes, and the distribution of anomalous ratios of helium isotopes in ground water. The movement history of the plate can explain active tectonics in western Japan in the last 2–4 Ma. The location history of the volcanic front and the tear control where fluids ascend in the crust, and these fluids are responsible for the generation of large earthquakes and volcanoes.
 Western Japan is densely populated, and has suffered disasters from offshore megathrust earthquakes like the 1944 Tonankai and 1946 Nankai earthquakes, and inland earthquakes like the 1995 Kobe earthquake. Tectonics in western Japan is complex and not so well understood, mainly due to insufficient knowledge about the Philippine Sea (PHS) Plate subducting beneath western Japan (Figure 1). The possibility that the plate lies beneath the Chugoku region was proposed in 1980 [Nakanishi, 1980], but it was only recently confirmed by receiver function (RF) analyses [Shiomi et al., 2004, 2008; Ueno et al., 2008], and seismic tomography [Nakajima and Hasegawa, 2007]. Although the shape of the plate was modeled in those recent studies, significant mismatches exist among the models. In particular, the shape of the plate between the Kinki and Shikoku-Chugoku regions is not well constrained.
 While the top of the PHS Plate was defined in previous models as a continuous surface, here we propose a model which includes a tear between the Kinki and Shikoku-Chugoku regions. Worldwide, many tears in subducting plates have been reported [e.g., Yamaoka et al., 1986; Wortel and Spakman, 2000; Obayashi et al., 2009], and a tear in the PHS Plate has been suggested in its shallower seismic part [Ishikawa, 2001; Ueno et al., 2008], and considered as a possible mechanism to explain anomalous ratios of helium isotopes in ground water [Sano and Nakajima, 2008; Sano et al., 2009]. However, the exact shape of the plate and the possible mechanisms for producing a tear in the plate were unknown.
2. Historical Plate Motion and the Tear
 The PHS Plate subducting under western Japan was formed at a NNW-striking back-arc spreading center active from 30-15 Ma [Okino et al., 1994, 1999]. The extinct ridge is still visible in bathymetry as the Kinan Seamount Chain (KSC) (Figure 1). At 15 Ma, the plate was subducting beneath the Japanese Islands in a NNW direction, almost parallel to the ridge [Seno and Maruyama, 1984; Takahashi, 2006], and this direction differs from the current plate motion direction of N55°W [Miyazaki and Heki, 2001]. These two directions are visible as lineaments in the distribution of deep tremor sources in the Shikoku region [Ide, 2010]. The time of the change in direction has not been determined precisely, but it was probably between 2 and 4 Ma, given that all present tectonic movements started around that time [e.g., Kimura et al., 2005; Sato et al., 2009; Ikeda et al., 2009].
 Therefore, when the plate movement changed direction, it had subducted for more than 10 million years towards the NNW. If the velocity of subduction were similar to the present-day velocity of 6–7 cm/yr [Miyazaki and Heki, 2001], the plate must extend at least 500 km from the trench, and it must bend downwards and have been accompanied by a volcanic front. Since the exact shape of the plate is unknown, we assume a very simple shape (Figure 2a), together with a volcanic front in the Chugoku region, where there was a lot of volcanic activity between 12 and 4 Ma [Kimura et al., 2005].
 After the change in movement direction, the plate was folded from the WNW so that it dips towards the new direction, similar to its current disposition beneath the Kyushu region. Since the plate behaves elastically, the deformation brought up parts of the subducted plate beneath the Chugoku region to compensate for downward movement in the Kyushu region. This folding movement would have caused an accumulation of strain in the PHS Plate, and the plate would have split along its weakest part, the extinct ridge.
3. Current Shape of the Philippine Sea Plate
 The current shape of the PHS Plate can be deduced from the hypocenters of intraplate earthquakes (Figure 2b). These earthquakes occur near the oceanic Moho, and deeper than the plate interface [Baba et al., 2002; Shelly et al., 2006]. Based on the separation between intraplate seismicity and the interface identified by low frequency earthquakes in western Shikoku, we assume that the interface is 7 km above the zone of intraplate seismicity. We extrapolate the intraplate seismicity downwards to satisfy the biharmonic equation that minimizes free energy for a linear elastic thin plate [Smith and Wessel, 1990]. Figure 2c shows the extrapolated surface, which has a valley structure at the east of the subducted extinct ridge. The surface shape is almost symmetrical across this valley, but seismicity exists only in the eastern side. There is large strain near the subducted extinct ridge, a region lacking in seismic activity. If, on the other hand, we assume a discontinuity exists along the subducted extinct ridge, and extrapolate surfaces in the eastern and western plates independently (Figure 2d), the strain becomes small, consistent with the lack of seismicity. It should be noted that the offset at the tear is up to 50 km in Figure 2d, which is smaller than the typical thickness of oceanic lithosphere. Therefore, the tear may not penetrate through the whole subducted lithosphere, but it decouples stress in two regions.
 The split model (Figure 2d) is consistent with the distribution of seismicity and focal mechanisms. Hypocenter depths differ significantly across the tear (Figure 3). The changes in depth occur within 20 km of the assumed tear. While earthquakes occur down to depths of 70 km on the eastern side, there is no sign of deep subduction on the western side. Many earthquakes have a focal mechanism that shows a vertical nodal plane with a northwestward strike, suggesting subsidence of the northeastern block (Figure 3). The tear corresponds to the western edge of a band-like distribution of non-volcanic tremor sources in the region from Chubu to Kinki [Obara, 2002], which also suggests the discontinuity of the plate at a depth of about 30 km. The difference between the eastern and western parts of the plate seems to continue to the trench axis. In the shallower part of the subducting PHS Plate, the stress field is discontinuous along the eastern side of the KSC [Mochizuki et al., 2010]. This observation supports the idea that the subducting surface of the PHS Plate is mechanically partitioned, at least to a depth of 30 km.
 A tear, such as we describe here, has not been suggested in previous studies of the plate's subsurface structure. However, as a matter of fact we can see evidence of the tear in some RF studies [Shiomi et al., 2004, 2008; Ueno et al., 2008]. Beneath the Chugoku region, for example, previous workers interpreted the surface of the PHS Plate in our split model as the continental Moho. Since the depth of the plate surface is about 40 km on the western side, and very close to the continental Moho, it is difficult to discriminate between the two discontinuities. We then reinterpreted the previous RF image [Shiomi et al., 2008] with additional data, and assigned positive phases (red) close to the surface of the split model (Figure 2d) to the oceanic Moho (Figure 2e). Figure 4 shows that the positive phase is continuously distributed and descends to the north, corresponding to the oceanic Moho within the subducting plate. Dip angles in the two cross-sections are completely different, with a gentle dip beneath the Shikoku region, and steeper angles of subduction under the Kii region. It is difficult to identify the oceanic Moho near the tear, suggesting that the structure along the tear is very complex (Figure 2e). The location of the northern edge of the subducted plate is not clear. The presence of positive phases under the Oki Islands (a question mark in Figure 4) may be an indication that the plate extends beneath those islands.
4. Discussion and Conclusions
 Extrapolating from the depths of intraplate earthquakes and high RF amplitudes, we present our preferred model for the split PHS Plate in Figure 2f. The model provides a new perspective for earthquakes and volcanoes in western Japan. Most Quaternary volcanoes in western Japan [Committee for Catalog of Quaternary Volcanoes in Japan, 1999] are located between two assumed volcanic fronts, shown by the red lines in Figure 2f, except for the Chubu region where the effect of the Pacific plate is not negligible. We propose that underground sources of magma near the surface of the PHS Plate migrated between the two volcanic fronts. Some of the roots of the present volcanoes might be distributed during this migration.
 Near these volcanoes, many deep low-frequency earthquakes (LFEs) occur around the continental Moho [Hasegawa and Yamamoto, 1994; Ohmi and Obara, 2002]. There are also clusters of LFEs in the western Kinki region that are quite distant from the volcanoes. While the LFEs near volcanoes are probably induced by magma-related fluids, the LFEs in western Kinki are probably induced by water ascending along the tear of the plate. The location of the tear is also on the western edge of the region where anomalously high amounts of the helium-3 isotope are observed in ground water, which suggests upwelling of water from the deep mantle [Sano and Nakajima, 2008; Sano et al., 2009]. These previous studies concluded that a continuous slab model similar to Figure 2c is favorable, because they could not explain why the region of high anomaly is located only in the eastern side of the extinct ridge. However, our split model explains it, because upwelling may occur only in the wake of the tear moving to the west.
 We point out that all 16 large crustal earthquakes (MJ > 6, as determined by the Japan Meteorological Agency) in western Japan (Figure 2f), since 1960, occurred within 20 km of Quaternary volcanoes or LFEs, or in other words, at points where fluids are ascending. For older events before 1960, we see a similar tendency, but the correlation with LFEs is weaker. Nevertheless, all the large events between 1925 and 2010 took place close to the two hypothetical volcanic fronts, or in the eastern side of the tear. Earthquakes have generally been considered to occur on active faults. However, active faults do not allow us to predict the location of large earthquakes efficiently because the faults have a very wide distribution. Instead, we suggest that the strong correlation between fluid ascension points and large earthquakes is an indication that it is fluids that control large earthquakes. The spatial pattern of the fluid ascension points is controlled by the deformation history and the present shape of the PHS Plate. This change of perception is important in terms of the future risk assessment of destructive inland earthquakes.
 We thank H. Kawakatsu, M. Kinoshita, and J. Ashi for helpful discussion. We also thank for constructive comments from D. Wilson and an anonymous reviewer. This work was supported by JSPS KAKENHI (20340115), MEXT KAKENHI (21107007). GMT software [Wessel and Smith, 1991] was used to draw figures.