The active subduction of the young Philippine Sea (PHS) plate and the old Pacific plate has resulted in significant seismic heterogeneity and anisotropy in Southwest (SW) Japan. In this work we determined a detailed 3-D P wave anisotropic tomography of the crust and upper mantle beneath SW Japan using ∼540,000 P wave arrival times from 5,249 local earthquakes recorded by 1095 stations. The PHS slab is imaged clearly as a high-velocity (high-V) anomaly which exhibits considerable lateral variations. Significant low-velocity (low-V) anomalies are revealed above and below the PHS slab. The low-V anomalies above the PHS slab may reflect the upwelling flow in the mantle wedge and the PHS slab dehydration, and they form the source zone of the arc volcanoes in SW Japan. The low-V zones under the PHS slab may reflect the upwelling flow in the big mantle wedge above the Pacific slab. The anisotropy in the crust and upper mantle is complex. In Kyushu, the P wave fast velocity direction (FVD) is generally trench-normal in the mantle wedge under the back-arc, which is consistent with the corner flow driven by the PHS slab subduction. The FVD is trench-parallel in the subducting PHS slab under Kyushu. We think that the intraslab seismicity is a potential indicator to the slab anisotropy. That is, the PHS slab with seismicity has kept its original fossil anisotropy formed at the mid-ocean ridge, while the aseismic PHS slab has reproduced the anisotropy according to its current deformation.
 Subduction process is proved to be closely related to seismogenesis, magma generation and metamorphic transformations [e.g., King et al., 2007; Stern, 2002; van Keken and King, 2005]. Therefore, studying the seismic heterogeneity and anisotropy in subduction zones is of great importance in understanding the dynamic process within the Earth. The Japanese islands consist of Northeast (NE) Japan and Southwest (SW) Japan. In NE Japan, the Pacific plate is subducting beneath the Okhotsk plate, intermediate-depth and deep earthquakes form a clear Wadati-Benioff zone in the subducting slab, and active arc volcanoes are distributed along the volcanic front parallel to the strike of the Japan Trench. In SW Japan, the Philippine Sea (PHS) plate is subducting beneath the Eurasian plate from the Sagami and Nankai troughs at a rate of ∼5 cm/yr [Seno et al., 1993; Zang et al., 2002], and the Pacific plate descends beneath the PHS and Okhotsk plates from the Japan and Izu-Bonin trenches at a rate of ∼8 cm/yr [Seno et al., 1996] (Figure 1).
 The geometry of the PHS slab in SW Japan has been investigated by many researchers using the hypocenter distributions, focal mechanisms, receiver-function analyses and tomographic imaging [e.g.,Hirose et al., 2008; Hori, 2006; Ishida, 1992; Nakajima et al., 2009; Ramesh et al., 2005]. Although many studies suggested the existence of an aseismic PHS slab beneath SW Japan, the configuration of the PHS slab including the depth range of the aseismic slab is still in debate. In addition, the relationship between the observed low-frequency (LF) microearthquakes and the subducting PHS slab in SW Japan has been paid much attention [e.g.,Ohta and Ide, 2011; Seno and Yamasaki, 2003; Shelly et al., 2007; Yoshioka et al., 2008; Zhao et al., 2011b]. Seismic tomography, first proposed by Aki and Lee , is one of the most effective methods to explore the heterogeneous structure of the crust and mantle and its relationship to seismic and volcanic activity. There are several recent reviews on the theory and applications of seismic tomography [e.g., Nolet, 2008; Thurber and Ritsema, 2007; Zhao, 2009]. Many researchers have studied the 3-D velocity structure under SW Japan using isotropic tomography, which have greatly improved our understanding of seismotectonics, magmatism and dynamics of this subduction zone [e.g.,Abdelwahed and Zhao, 2007; Honda and Nakanishi, 2003; Nakajima and Hasegawa, 2007; Zhao et al., 1994, 2011b]. Their tomographic results have clearly revealed low-velocity (low-V) anomalies in the crust and upper mantle beneath the active volcanoes, and imaged the subducting PHS plate as a high-velocity (high-V) anomaly. However, the crust and mantle are generally assumed to be isotropic in these previous tomographic studies.
 Seismically anisotropic materials have been observed in the Earth's interior. Seismic anisotropy records the present or past tectonic deformation inside the Earth. For this reason, the characterization of seismic anisotropy can give much information on dynamic processes in the Earth's interior [e.g., Becker et al., 2012; Karato et al., 2008; Long and Silver, 2008; Long and Becker, 2010]. Shear wave splitting observations are a straight-forward and powerful method to study the anisotropy of the crust and mantle [e.g.,Maupin and Park, 2007; Savage, 1999; Silver, 1996]. Seismic anisotropy of the crust and upper mantle beneath SW Japan has been revealed by shear wave splitting measurements [e.g., Iidaka, 2003; Salah et al., 2008]. Although shear wave splitting measurements can provide useful information for characterizing the upper mantle deformation, the interpretation of the results in a subduction zone is very difficult because of the existence of complex 3-D mantle flow [Long and Silver, 2008] and different contributions of the minerals and hydrous phases [Mainprice and Ildefonse, 2009]. In addition, the incidence angle of S-wave at a seismic station must be taken into account in the shear wave splitting analysis [Savage, 1999]. These complexities make it difficult to clarify where and how the anisotropy originates with the shear wave splitting measurements alone, especially in subduction zones. In contrast to the shear wave splitting observations, the P wave travel-time data are considered to have high resolution to image both isotropic velocity variation and anisotropy. Hence P wave data have been used extensively to explore the anisotropy of the crust and upper mantle under many regions [e.g.,Bamford, 1977; Eberhart-Phillips and Henderson, 2004; Eberhart-Phillips and Reyners, 2009; Hearn, 1999; Hess, 1964; Hirahara, 1988; Ishise and Oda, 2005; 2008; Wang and Zhao, 2008, 2009, 2010]. Recently, Ishise and Oda  studied P wave anisotropic tomography in SW Japan, and their results are roughly consistent with the shear wave splitting measurements in the crust and reveal anisotropic features of the subducting PHS plate. However, they used a small data set from only 396 seismic events and 55 stations, and did not consider the existing velocity discontinuities (such as the Conrad, Moho and the subducting slab boundary) in their inversion.
 In this work, we have determined a high-resolution P wave anisotropic tomography of the crust and upper mantle under SW Japan using a much better data set containing ∼540,000 high-quality P wave arrival times from 5,249 local earthquakes recorded by 1095 seismic stations (Figure 2). The new result may improve our understanding of the structure and dynamics of the young Nankai subduction zone.
where S is the total slowness, S0 is the azimuthal average slowness (i.e., isotropic component), A1 and B1 are anisotropic parameters, ϕ is the raypath azimuth.
 The inversion for isotropic velocity tomography depends largely on the ray coverage, while the inversion for the anisotropic parameters is related closely to not only the ray density but also the ray azimuthal coverage. Only one grid net was adopted in the previous anisotropic tomography [Wang and Zhao, 2008, 2010]. In this work two 3-D grid nets are set up in the study region. One grid net is used to express the isotropic velocity variations, whereas the other is adopted to express the anisotropic structure. At each point in the modeling space there are three parameters, one of which represents the isotropic velocity perturbation from the starting 1-D P wave velocity model, while the other two parameters represent P wave azimuthal anisotropy. Each of the three parameters at the point is calculated by using a linear interpolation of the values at the eight grid nodes surrounding that point. Thus for local earthquakes within the modeling space, we have the following equation,
where Tmnobs and Tmncal are the observed and calculated travel times from the nth event to the mth station, φn, λn, hn and T0n are the latitude, longitude, focal depth and origin time of the nth event, respectively, Δ denotes the perturbation of a parameter, Vp is the isotropic velocity at the pth grid node, A1q and B1q are the anisotropic parameters at the qth grid node of the second grid net, and Emnrepresents higher-order terms of perturbations and errors in the data. The fast velocity direction (FVD)ψ and the amplitude α of anisotropy can be expressed as:
where V0 denotes the average isotropic velocity, Vf and Vs denote the velocities in the fast and slow directions, respectively.
 As shown in Figure 2, we used ∼540,000 P wave arrival times from 5,249 local earthquakes recorded by 1,088 seismic stations during January 2003 to June 2007. These events were recorded by the seismic stations belonging to the Japanese national universities, Japan Meteorological Agency (JMA), and the High-sensitivity seismic network (Hi-net) that is operated by the National Research Institute for Earth Science and Disaster Prevention in Tsukuba, Japan. The first P wave arrival-time data were collected from the original seismograms, with picking accuracy of 0.05–0.1 s, by the staffs of Research Center for Prediction of Earthquakes and Volcanic Eruptions, Tohoku University. The events used in this study were selected carefully based on the following criteria: (1) all the selected events were recorded by more than 20 seismic stations; (2) the study area is divided into 0.02° × 0.02° × 1 km blocks at depths <30 km and 0.01° × 0.01° × 0.5 km blocks at depths >30 km, and in each block only one event with the largest number of arrival-time data was selected in order to keep the hypocenter distribution as uniform as possible; (3) they have reliable hypocentral locations with uncertainties <5 km.
 For expressing the isotropic velocity variations, we set a net of grid nodes with an interval of 0.3° in the longitude direction and 0.25° in the latitude direction, and at depths of 0, 10, 25, 40, 60, 80, 100, 120, 150, 180, 220, 260, 300, and 400 km. For expressing the anisotropic parameters, we set the second net of grid nodes with an interval of 0.4° in the horizontal direction and at depths of 0, 8, 25, 50, 80, 120, 170, 250, and 300 km. Following the previous studies [Zhao et al., 1992, 1994], a slightly modified version of the J-B velocity model [Jeffreys and Bullen, 1940] was used, and the Conrad and Moho discontinuities and the upper boundary of the subducting Pacific slab were considered in the starting velocity model (Figure 3). The geometries of the three discontinuities determined by the previous studies were adopted and they were fixed in the inversion process. In the starting velocity model, the subducting Pacific slab was assumed to be 85 km thick and have a P wave velocity 4% higher than the surrounding mantle, following the previous studies [Abdelwahed and Zhao, 2007; Zhao et al., 1994]. An efficient 3-D ray-tracing scheme was applied to compute the raypaths and travel times [Zhao et al., 1992]. The earthquakes were relocated during the inversion. The large but sparse system of observation equations was solved using the LSQR algorithm [Paige and Saunders, 1982]. The nonlinear tomography problem was solved by iteratively conducting linear inversions. The final results were obtained after four iterations. The damping parameter was 10, 25, 50, and 100 during the four iterations, respectively. An optimal smoothing parameter of 0.008 was adopted after conducting many inversions. To obtain a more stable and reliable inversion result, in this study we first inverted for the isotropic velocity variations, and then used the obtained 3-D velocity model to invert for the anisotropic parameters, in contrast to the joint inversion conducted in our previous studies [Wang and Zhao, 2008, 2010].
 We first evaluated the resolution of the tomographic images using the checkerboard resolution test (CRT) [Zhao et al., 1992] (Figures 4 and 5) before describing the main features of the obtained results. CRT is a convenient and well-used tool to make resolution analysis in tomographic studies [Lévěque et al., 1993; Thurber and Ritsema, 2007; Zhao et al., 1992]. The anisotropic parameters A1 and B1 of same value (±0.02) and velocity anomalies of ±4% (with +4% velocity anomalies already added to the Pacific slab in the initial model) were alternately assigned to the nodes of each grid in the CRT. Following equation (2), the anisotropic components are shown in bars, and their azimuth and length denote the FVD and the amplitude of anisotropy, respectively (Figure 5). In the CRT, the azimuths at two adjacent grid nodes are normal to each other (22.5° and 112.5°), and the amplitude of anisotropy is 2.8% (Figure 5). The theoretical arrival times were calculated for the synthetic model by adding random errors in a normal distribution with a standard deviation of 0.1 s. Then we inverted the synthetic data with the same algorithm as for the real data using an initial 1-D P wave velocity model withA1 = 0.0 and B1 = 0.0. In the output images, ±4% velocity anomalies were subtracted from the inverted velocity perturbations for the Pacific slab to make the CRT results clear. The resolution is thought to be good for the areas where the checkerboard image is recovered well.
 Following Wang and Zhao , we used a parameter MRAGA (maximum ray-azimuth gap angle) to account for the azimuthal coverage of rays at a grid node in the anisotropic inversion. The parameter MRAGA is deduced from the azimuths of the rays passing through each grid node (for details, seeWang and Zhao ). In this study, the anisotropic parameters are inverted only for the grid nodes with at least 20 passing rays and with MRAGA <30 degrees, which are determined by considering the ray coverage in our data set and after many CRTs. Figures 4 and 5 show the CRT results for the isotropic velocity perturbations and FVDs, respectively. The resolution is generally good in most parts of the study area, but the CRT results reveal poor azimuthal coverage of rays in the Chugoku and Shikoku districts at depths >80 km. The anisotropic pattern is poorly recovered at depths >140 km.
 A few previous studies suggested that the unresolved structural heterogeneity could result in the inverted seismic anisotropy [e.g., Backus, 1962; Capdeville et al., 2010; Fichtner et al., 2010]. We conducted two more synthetic tests to investigate the influence of the trade-off between heterogeneity and anisotropy on our inverted results. In the first test, similar to the CRT as mentioned above, velocity perturbations of ±4% were alternatively assigned to the grid nodes but the anisotropy is not considered in the input model.Figures 6 and 7 show the inverted images. In the second test, the input model contains increased alternative velocity perturbations of ±6% without the anisotropy, and the inverted results are shown in Figures S1 and S2 in the auxiliary material. Although no anisotropy is included in the input model, anisotropy shows up in the output images (Figures 7 and S2). However, the inverted anisotropy is quite weak, generally <1.0%, suggesting that our data set (Figure 2) and inversion method can generally distinguish the azimuthal anisotropy from the heterogeneity.
4. Results and Discussion
4.1. Model Comparisons
Figures 8–13 show the tomographic images obtained with the above mentioned starting velocity model (Figure 3, hereafter we call it Model 1) which is composed of the JB velocity model, a high-V anomaly for the subducting Pacific slab, the Conrad and Moho depth variations, and smoothing regularization. We also conducted several more inversions to investigate the influences of the starting velocity model.
Figures S3–S8 show the tomographic images obtained with a starting model similar to Model 1 but the iasp91 Earth model [Kennett and Engdahl, 1991] is adopted. Comparing these images with those in Figures 8–13, we can see that the patterns of velocity perturbations and FVDs are quite similar except for some changes in their amplitudes. The obvious difference is the inverted absolute P wave velocity in the subducting PHS slab.
Figures S9–S14show the tomographic images obtained with a starting model similar to Model 1 but the Pacific slab was not included. The subducting Pacific slab is not imaged as a high-V anomaly (seeFigure S9and section A-A′ inFigure S10). Comparing these images with those in Figures 8–13, we can see that the large difference is in the velocity image along the northern cross-sections inFigure S10, which may be induced by the different raypaths within the Pacific slab where the slab is shallow.
Figures S15–S20 show the tomographic images obtained with a starting model similar to Model 1 but with flat Conrad and Moho at 12 and 26 km depths. Comparing these images with those in Figures 8–13, we can see some differences in the absolute P wave velocity of the crust and the uppermost mantle (see Figures S17 and S19).
Figures S21–S26 show the tomographic images obtained with a starting model similar to Model 1 but without the smoothing regularization in the inversion. Comparing these images with those in Figures 8–13, we can see some minor differences between them. It is clear that the inversion without smoothing regularization resulted in a rough model of velocity tomography (Figures 8 and S21).
 The final RMS travel-time residuals for the five inversions are 0.28 s, 0.34 s, 0.29 s, 0.31 s, and 0.29 s, respectively. Although there are some differences among these results, all the inversions show the following features: (1) the PHS slab is imaged clearly as a high-V anomaly and it exhibits different shapes beneath SW Japan; (2) a prominent low-V anomaly is visible under the PHS slab beneath the Chugoku and Shikoku districts; (3) significant low-V anomalies are visible in the crust and uppermost mantle beneath the active arc volcanoes and in the fore-arc mantle wedge beneath Kyushu, (4) the FVDs are generally trench-parallel in the subducting PHS slab beneath Kyushu. We focus on the tomographic images obtained with the starting velocity model (Model 1) and briefly address others in the supplementary materials.
 Subduction of the PHS plate at the Sagami trough and the eastern part of the Nankai trough has caused arc-arc collision of the Izu-Bonin-Mariana Arc with the Honshu Arc in central Japan, and the Izu peninsula is in collision at present [Soh et al., 1998, 1991]. The tectonic complexity makes it difficult to clarify the PHS slab in central Japan [Hori, 2006; Nakajima and Hasegawa, 2007; Nakajima et al., 2009; Sato et al., 2005; Yamamoto et al., 2009]. Beneath the Kanto-Chubu district, the PHS slab is imaged clearly as a high-V anomaly subducting toward the northwest of the Izu peninsula to depths of ∼140 km in contact with the Pacific slab (see section AA′ inFigure 9), which is consistent with the recent tomographic result [Nakajima et al., 2009].
 Beneath the eastern end of the Nankai trough, a high-V zone is imaged down to depths of ∼200 km with the seismicity down to depths ∼50 km (see sections BB′ and CC′ inFigure 9). It is visible that the PHS slab seems to be separated into two high-V zones here, which is in consistent with the result ofNakajima and Hasegawa  and the supposed model with the splitting of the PHS slab into two parts by Ishida . However, there is no slab gap in the recent tomographic image [Nakajima et al., 2009]. The difference may be induced by the different Pacific slab models used in these inversions. In our tomography, we used the Pacific slab geometry from Zhao et al.  and Nakajima and Hasegawa , but the modified upper boundary of the Pacific slab in Nakajima et al. is generally shallower beneath the Chubu district. Beneath the active volcanoes especially along section BB′, a prominent low-V anomaly is clearly visible in the crust and upper mantle wedge above the subducting PHS slab, which reflects the arc magmatism caused by the PHS slab dehydration and corner flow in the mantle wedge.
 Beneath the Kinki-Chugoku district, the high-V zones are discontinuous and the subducting PHS slab seems complex with the seismicity down to a depth of ∼60 km (see sections DD′ and EE′ inFigure 9). Especially along the section EE′, the PHS slab is imaged as a low-V anomaly with intraslab seismicity occurred in the low-V zone, which has been known to be a unique feature beneath the Kii Peninsula, in agreement with the previous studies [Honda and Nakanishi, 2003; Nakajima and Hasegawa, 2007; Salah and Zhao, 2003; Seno et al., 2001]. Many studies using the helium isotopic data have revealed anomalously higher 3He/4He ratios in Kii Peninsula than those in other fore-arc regions of the Japan Islands [Matsumoto et al., 2003; Sano and Wakita, 1985; Umeda et al., 2006; Wakita and Sano, 1987]. There are two interpretations of the anomalous features in the Kii Peninsula. One is that the PHS slab beneath SW Japan is younger and hotter than the Pacific slab, hence the slab is serpentinized, and aqueous fluids released by dehydration of the serpentinized slab could acquire MORB (mid-ocean ridge basalts) source helium from the mantle wedge to the Earth's surface through the existing normal faults under NE-SW extension [Matsumoto et al., 2003; Umeda et al., 2006]. The second interpretation is that the PHS slab might be more hydrated than the surrounding regions because of the interaction with a large-scale upwelling flow below the PHS slab around Kii Peninsula as suggested byNakajima and Hasegawa . Although our tomographic image shows a similar large low-V anomaly below the PHS slab, the influences of the upwelling flow need more detailed investigations through isotope tracer studies and high-resolution seismic tomography in Kii Peninsula.
 Beneath the eastern and central Shikoku-Chugoku districts, the mantle wedge is not well developed, and the upper boundary of the subducting PHS slab seems to be located directly beneath the inland crust (see sections FF′ and GG′ inFigure 11), which is consistent with the previous results of teleseismic receiver-function analysis [Yamauchi et al., 2003] and local isotropic tomography [Nakajima and Hasegawa, 2007]. Along the section FF′, our tomographic image does not show a continuous aseismic PHS slab extending down to 200 km depth under the Japan Sea as revealed by the teleseismic tomography [Abdelwahed and Zhao, 2007]. Beneath the western Shikoku-Chugoku district, the PHS slab is imaged clearly as a high-V anomaly subducting gently to a depth of ∼80 km with the intraslab seismicity down to a depth of ∼60 km (see section HH′ inFigure 11).
 A large-scale low-V anomaly of amplitude ∼5% is visible below the PHS slab under the Chugoku and Shikoku districts, with lateral and depth extents of ∼300 km and ∼200 km in this study, respectively. However, the size of the inverted low-V zone is constrained by the raypaths, e.g., the western edge of the low-V zone is not clear because there is no resolution beneath the Japan Sea in this study. The large-scale low-V anomaly seems to be connected with the Quaternary volcanoes in Chugoku, which is in agreement with the suggestion that the low-V zone below the PHS slab acts as the primary source of magmas for Quaternary volcanoes in the Chugoku district [Nakajima and Hasegawa, 2007; Sano et al., 2006]. If the low-V anomaly reflects hot upwelling mantle materials from the lower mantle below the Pacific slab at depth >400 km, the3He/4He ratio at the volcanic front in Chugoku should be very high, which is inconsistent with the observed normal 3He/4He ratios (the highest value is 5.88 Ratm) [Sano et al., 2006]. The teleseismic tomography [Abdelwahed and Zhao, 2007] has revealed a low-V anomaly below the PHS slab and above the Pacific slab under SW Japan. Therefore, a plausible interpretation is that the large-scale low-V anomaly was induced by the deep dehydration process of the subducted Pacific slab in the mantle transition zone (at depths between 410 km and 660 km) [Ohtani et al., 2004] and the convective flow in the big mantle wedge [Abdelwahed and Zhao, 2007; Zhao and Ohtani, 2009; Zhao et al., 2011a].
Kelbert et al. derived a global-scale 3D model of electrical conductivity variations in the Earth's mantle by inverting long-period geomagnetic response functions and they found that the electrical conductivity is high in the cold and seismically fast areas where slabs have subducted into or through the mantle transition zone, which supports that cold subducting slabs have carried some water into the transition zone. However, because of the enormous H2O storage capacities for wadsleyite and ringwoodite [Kohlstedt et al., 1996; Smyth et al., 1997], the water released from the cold subducting slabs should be stored in the transition zone. As a result, the deep dehydration of the subducted Pacific slab may not be a direct contributor to cause the large-scale low-V anomaly because of the difficulty for water escaping from the mantle transition zone.Huang et al.  presented electrical conductivity measurements of wadsleyite and ringwoodite for a range of H2O content and suggested ∼0.1–0.2 wt. % H2O in the transition zone beneath the north Pacific, which is much larger than that in the upper mantle (100–500 ppm). Several studies have suggested that the large contrast in H2O storage capacity between the mantle transition zone and overlying upper mantle could result in negatively buoyant partial melt of material advecting across the 410-km discontinuity, and the H2O released from the partial melt zone may be delivered to the upper mantle [e.g., Kawamoto et al., 1996; Litasov and Ohtani, 2002; Revenaugh and Sipkin, 1994; Young et al., 1993]. Thus, the water expelled from the partial melt zone atop of the 410-km discontinuity and convection in the big mantle wedge may be the contributors to the large-scale deep low-V anomaly revealed by this study. When the Pacific slab was subducting into the transition zone in the past, a vast amount of ascending water-rich wadsleyite (forced up by the downward flux of subducting slab) might rise out of the mantle transition zone into the upper mantle and undergo dehydration-induced partial melting that filtered out incompatible elements, and the filtered, dry and depleted solid phase may continue to rise to become the source material for MORB [Bercovici and Karato, 2003]. Because the Pacific slab has subducted into the transition zone beneath eastern North China Craton [Zhao, 2009; Huang and Zhao, 2006], this hypothesis may explain the intraplate magmatism and lithospheric thinning in eastern North China Craton during the Mesozoic-Cenozoic. However, it should be confirmed or denied by more pieces of evidence from geology, geophysics and geochemistry.
 Beneath Kyushu, the PHS slab is imaged clearly as a high-V anomaly subducting steeply to a depth of ∼200 km with intermediate-depth seismicity (see sections II′, JJ′ and KK′ inFigure 11), which is in agreement with the recent tomographic results [Nakajima and Hasegawa, 2007; Zhao et al., 2011b]. Two prominent low-V zones are visible above the subducting PHS slab. One exists in the crust and upper-mantle wedge beneath the active volcanoes and back-arc areas, which originates from ∼150-km depth and reflects the source zones of arc and back-arc magmatism caused by the PHS slab dehydration and corner flow [Cagnioncle et al., 2007; Stern, 2002; Zhao et al., 2011a]. Another low-V zone exists in the crust and uppermost mantle beneath the fore-arc region, which originates from ∼50-km depth and reflects the slab dehydration and fore-arc mantle serpentinization [e.g.,Hilairet and Reynard, 2009; Hyndman and Peacock, 2003; van Keken, 2003; Xia et al., 2008].
 The inverted absolute P wave velocity of the subducting PHS slab seems larger than 8.0 km/s at depths <100 km, which is in accordance with the results from the reflection/refraction surveys [Iidaka et al., 2003] and tomography [Nakajima and Hasegawa, 2007], and increasing to approximately 8.4 km/s with depth (see Figures 10 and 12). The absolute P wave velocities of the large-scale low-V anomaly mentioned above, below the PHS slab under the Chugoku and Shikoku districts, are between 7.6 km/s and 8.0 km/s (see sections FF′, GG′ and HH′ inFigure 12).
 Low-frequency (LF) microearthquakes, which lack sharp onsets and last from a few minutes to a few days with a predominant frequency of 1 to 10 Hz, have been observed around the Moho discontinuity in SW Japan [e.g.,Zhao et al., 2011b]. As shown in Figures 9–12, some of the LF events occurred in the lower crust and uppermost mantle beneath the active volcanoes in SW Japan, and they are good indicators of the magmatic and volcanic activity [Hasegawa and Yamamoto, 1994]. Many LF events also take place at 30–40 km depths along the upper boundary of the subducting PHS slab under SW Japan, but no LF events occur in Kyushu where the PHS slab dips steeply. Our results also show that the LF events in the fore-arc regions generally occurred in the low-V zones, suggesting that fluids released from the PHS slab may affect the generation of the LF events (see a recent review byObara ). Different from other fore-arc regions in western Japan, two groups of LF events occurred beneath Kii Peninsula (see section EE′ inFigure 9), which may be associated with the slab segmentation beneath Kii, as suggested by a recent receiver-function study [Shiomi and Park, 2008].
4.3. Crustal Anisotropy
Figure 13 shows the obtained P wave anisotropic tomography at six depth slices beneath SW Japan. The result of the upper crust displays a complex image of anisotropy (Figure 13a). Microcracks, cracks or fractures located in the vicinity of active faults, and lattice preferred orientation (LPO) of mineral crystals are suggested to be the main contributors to the crustal anisotropy [Godfrey et al., 2000; Kaneshima, 1990; Kern, 1993; Kern and Tubia, 1993]. The anisotropic pattern in the upper crust is roughly consistent with the result of Ishise and Oda  in the Chugoku and Shikoku districts. Ishise and Oda suggested that their obtained P wave FVDs of the upper crust are E-W in the Shikoku district, in close agreement with the fast polarization directions estimated from the shear wave splitting observation and nearly parallel to the maximum principal stress direction and strike directions of zonal geological structure. Their P wave FVDs of the upper crust are not consistent with the maximum principal stress of NW-SE direction in the Chugoku district. In fact, their obtained P wave FVDs are actually NE-SW rather than E-W in the Shikoku district, in accordance with our present result. The NE-SW P wave FVDs are consistent with the strike of the faults there (for more information on the faults in Japan, see the Webhttp://riodb02.ibase.aist.go.jp/activefault/cgi-bin/search.cgi?%20search_no=j024&version_no=1&search_mode=0). However, our obtained P wave FVDs in the upper crust beneath SW Japan (Figure 13a) are very complex and nearly random. Note that the S-wave fast polarization direction may differ from the P wave FVD when the waves propagate through a cracked solid in the upper crust [Crampin, 1978; Zheng, 2000]. Many studies explicitly suggested that the faster splitting shear waves generally parallel with the strike of the cracks and the horizontal direction of maximum compression stress in the upper crust [Crampin et al., 1989; Savage, 1999; Zheng, 2000], but there is no defined orientation of the P wave FVD for crack-induced anisotropy in the upper crust. In addition, the lateral resolution of 0.4° of our P wave anisotropy tomography may be too coarse to explain the small-scale anisotropic structure induced by microcracks and cracks in the upper crust.
 The crustal anisotropy induced by cracks and microcracks may be limited to the upper 10–15 km depth of the crust [Crampin, 1994; Kaneshima et al., 1988] because of the closure of cracks at pressures greater than ∼200–300 MPa (corresponding to ∼10–15 km depth) [Hrouda et al., 1993; Kern, 1990]. The P wave anisotropy in the lower crust is probably caused by the LPO of phylosilicates aligned during ductile flow of the lower crust [McNamara and Owens, 1993]. The obtained P wave FVD is roughly NW-SE at 25 km depth beneath the land except for the volcanic front (VF) area in SW Japan (Figure 13b), which is roughly consistent with the average absolute plate motion (APM) direction (∼N 65°W, denoted by the red arrow in Figure 13b) deduced from the HS3-NUVEL 1a model [Gripp and Gordon, 2002]. In addition, the obtained P wave anisotropy at 25 km depth may be affected by the magmatic underplating and coupled with the anisotropy of the uppermost mantle.
 Although the anisotropic amplitude is affected largely by the values of damping [Wang and Zhao, 2008] and smoothing parameters (see Figure S26), the tomographic images (Figure 13) clearly show that the anisotropic amplitude in the upper crust is not much larger than that in other portions beneath SW Japan. However, Huang et al. [2011a, 2011b]revealed strong anisotropy in the upper crust but very weak anisotropy in the mantle wedge under NE Japan using shear wave splitting measurements. The inconsistency may result from the differences between the P wave anisotropy and S-wave anisotropy in their sensitivity to the saturated cracks and the dependence to frequency. It is generally considered that fluids exist widely in the crust and upper mantle in subduction zones [Hirschmann, 2006; Mainprice and Ildefonse, 2009; Peacock, 1990; Zhao et al., 2011b]. The strength of P wave anisotropy decreases as the cracks become saturated [Crampin, 1978; Nur, 1972], while S-wave anisotropy is insensitive to the saturation of the cracks [Crampin, 1978]. P wave anisotropy is not very sensitive to frequency when propagating through cracks [Crampin, 1978]. In contrast, recent studies have paid close attention to the frequency-dependence of shear wave anisotropy [e.g.,Faccenda et al., 2008; Greve and Savage, 2009; Huang et al., 2011b; Marson-Pidgeon and Savage, 1997]. Therefore, these factors should be taken into account when comparing P wave anisotropy with the S-wave polarization in the crust and upper mantle.
4.4. Anisotropy in the Upper-Mantle Wedge and PHS Slab
 Beneath Kinki, Shikoku and Chugoku, the anisotropic structures are complex in the mantle wedge, which may be influenced by the interactions of the slab-driven corner flow, slab dehydration and the upwelling flow from depth. It is difficult to separate these influences by the present tomography alone. Although the inverted P wave FVDs are complex in the mantle wedge and the PHS slab, an obvious feature is visible that P wave FVDs are trench-normal (NW-SE) in the mantle wedge under the back-arc areas but trench-parallel (NE-SW) in the subducting PHS slab beneath the Kyushu district (seeFigure 13 and sections II′, JJ′, KK′ and LL′ in Figure 11). The trench-normal P wave FVD in the back-arc mantle wedge is consistent with the model in which the olivine axis aligns with the transport direction induced by the slab-driven corner flow, which agrees with the results in other subduction zones [e.g.,Eberhart-Phillips and Reyners, 2009; Wang and Zhao, 2008, 2010; Huang et al., 2011a, 2011b]. Long et al. presented a series of 2-D numerical simulations in a subduction zone, and used the trench-normal flow in the mantle wedge with B-type olivine fabric dominating in the region to explain their observed trench-parallel S-wave polarization beneath the Kyushu district. It is visible that the anisotropic amplitude of ∼5% in the PHS slab is larger than that in the overlying mantle wedge beneath Kyushu, which is consistent with the anisotropy in the Pacific slab [e.g.,Eberhart-Phillips and Reyners, 2009; Wang and Zhao, 2010]. Although the FVD in the fore-arc mantle wedge is not clear for the tomographic inversion with a lateral grid interval of 0.4°, our result suggests that the trench-parallel anisotropy within the subducting PHS slab is the important contributor to the observed ∼1 s S-wave polarization in Kyushu.
 Beneath Chubu, Kinki and Shikoku, the PHS slab shows both trench-parallel and trench-normal P wave FVDs at shallow depths, whereas trench-normal FVDs dominate in the deeper parts of the PHS slab, which is different from the uniform trench-parallel anisotropy in the subducting PHS slab under Kyushu. One plausible interpretation for the trench-parallel anisotropy in the PHS slab is that the deformation in slab would produce B-type anisotropy due to water-rich, low temperature and high stress, which was suggested to account for the trench-parallel anisotropy within the Pacific slab [Eberhart-Phillips and Reyners, 2009; Wang and Zhao, 2010]. However, the B-type olivine would exist in the fore-arc mantle wedge rather than within the subducting oceanic slabs [Karato et al., 2008; Kneller et al., 2005].
 We find an interesting feature that the trench-parallel anisotropy generally exists in the PHS slab where the intraslab earthquakes occur actively (Figure 14). The magnetic lineation is found to be oriented in the NW–SE direction in the Shikoku basin southeast of the Nankai trough [Weissel et al., 1981], which is perpendicular to the trench-parallel direction. It is well known that the anisotropy in the oceanic crust and mantle are induced by fossil fabric formed at the spreading mid-ocean ridge with the fast axis parallel to the spreading direction [Christensen, 1984; Hess, 1964; Raitt et al., 1969; Shearer and Orcutt, 1986]. Hence we propose the following scenario to explain the anisotropy in the subducting PHS slab beneath SW Japan: (1) the frozen-in anisotropy in the PHS plate exhibits a trench-parallel direction normal to the magnetic lineation when the PHS plate was formed; (2) the older and colder seismic PHS slab subducting steeply beneath Kyushu has kept the trench-parallel frozen-in anisotropy; (3) the younger and warmer seismic PHS slab descending gently beneath Chubu, Kinki and Shikoku still keeps the trench-parallel frozen-in anisotropy, whereas the aseismic PHS slab in deeper areas has reproduced their anisotropy to trench-normal in consistent with the subduction direction at present. Previous studies have suggested that deformed olivine crystals would keep the frozen-in anisotropy and may not easily be reoriented below a critical temperature of 900°C [Estey and Douglas, 1986; Goetze and Kohlstedt, 1973; Silver and Chan, 1988; Vinnik et al., 1992]. We suggest that the intraslab seismicity may be a potential indicator of the slab anisotropy, in addition to the critical temperature. That is to say, the anisotropy in the seismic slab may reflect the past deformation, while the anisotropy in the aseismic slab may reflect the present deformation. This may also explain the trench-parallel anisotropy in the cold subducting Pacific slab found by the previous tomographic studies [e.g.,Ishise and Oda, 2005; Eberhart-Phillips and Reyners, 2009; Wang and Zhao, 2010].
 We determined a detailed 3-D P wave anisotropic tomography beneath SW Japan using a large number of high-quality arrival-time data from local earthquakes. Main results of the present work are summarized as follows.
 1. The subducting PHS slab is generally imaged as a high-V zone except for the Kii Peninsula where the slab exhibits low velocity. The PHS slab shows considerable variations in its dipping angle, subduction depth and intraslab seismicity beneath SW Japan.
 2. A large-scale low-V anomaly is revealed in the upper mantle below the PHS slab under the Chugoku and Shikoku districts. We suppose that the fluids expelled from the partial melt zone atop the 410-km discontinuity and convection in the big mantle wedge above the Pacific slab may have caused the large-scale low-V anomaly.
 3. Two significant low-V anomalies are visible above the PHS slab in Kyushu: one exists in the crust and upper-mantle wedge beneath the active volcanoes and extends down to ∼150 km depth under the back-arc area; the other exists in the uppermost mantle under the fore-arc area.
 4. Many low-frequency events occurred in the lower crust and uppermost mantle beneath the active arc volcanoes. Some LF events took place in the low-V zones in non-volcanic areas beneath SW Japan, which may be induced by the slips between the gently subducting PHS slab and the overlying island crust, and the fluids released from the PHS slab may promote the slips. Two groups of LF events exist in the fore-arc beneath Kii Peninsula, which may reflect the PHS slab segmentation there.
 5. The P wave anisotropy in the upper crust is complex and different from that in the lower crust. We consider that various factors may cause the different anisotropies in the crust: the anisotropy may be mainly caused by cracks in the upper crust while largely influenced by the plastic flow deformation in the lower crust.
 6. The P wave FVD is trench-normal in the mantle wedge under the back-arc area and it becomes trench-parallel in the subducting PHS slab beneath Kyushu. The FVDs show both trench-normal and trench-parallel in the subducting PHS slab beneath Chubu, Kinki and Shikoku. We think that the intraslab seismicity may be a potential indicator to the slab anisotropy. That is, the subducting PHS slab with seismicity may keep its original fossil anisotropy formed at the mid-ocean ridge, while the aseismic slab may reproduce the anisotropy in accordance with its current deformation. This is a new interpretation. The present results may provide new insights into seismic anisotropy and dynamics of subduction zones.
 We used a large amount of high-quality data provided by the data center of the Kiban seismic network and the JMA unified catalog. The arrival-time data were collected from the original seismograms by the staffs of Research Center for Prediction of Earthquakes and Volcanic Eruptions, Tohoku University. Thorsten Becker (the Senior Editor), an anonymous Associate Editor, and two anonymous reviewers provided thoughtful review comments and suggestions that helped to improve our manuscript considerably. This research was supported by grants (40974026, 41074060) from the National Science Foundation of China to J. Wang and grants (Kiban-A 17204037, Kiban-S 11050123) from Japan Society for the Promotion of Science to D. Zhao, as well as by the Global-COE program of Earth and Planetary Sciences of Tohoku University. Figures are made using the GMT software [Wessel and Smith, 1995].