On October 6, 2000, an Mw = 6.7 crustal earthquake occurred in western Tottori prefecture, southwest Japan. Beneath the focal region of the earthquake, deep low-frequency (DLF) earthquakes were observed at depths of around 30 km. Five DLF earthquakes were detected within 3 years before the mainshock and more than 60 DLF earthquakes were observed during the 13 months after the mainshock. We investigated the focal mechanism of the DLF earthquake that occurred 9 hours before the mainshock, using amplitude ratios of the S-waves to the P-waves and polarization patterns of the S-waves. Both analyses indicated that a single-force source mechanism is more preferable than a double-couple source mechanism, which suggests the transport of fluid such as water or magma. This event is probably another example of DLF earthquakes that occur beneath active fault zones.
 On October 6, 2000, the 2000 Western Tottori Earthquake (Mw 6.7) shook western Tottori prefecture, Chugoku district, southwest Japan. Beneath the focal region of the quake, deep low-frequency (DLF) earthquakes were observed before the mainshock. During the period from June 1999 until the occurrence of the mainshock, five (5) DLF earthquakes were observed. DLF activity was also observed after the mainshock and during the 13 months from October 2000 to October 2001, with more than 60 DLF earthquakes detected. Figure 1a shows the distribution of DLF earthquakes from the earthquake catalogue of JMA (Japan Meteorological Agency) together with the ordinary earthquakes which are mainly the aftershocks of the 2000 Western Tottori earthquake.
 Around the world, DLF earthquakes have been usually observed near active volcanoes or the volcanic front and attributed to magma activity [e.g. Hasegawa and Yamamoto, 1994; Ukawa and Obara, 1993]. However, there are no active volcanoes in the focal region of the Western Tottori Earthquake. This is probably a new example of DLF activity that is associated with active faults.
 According to the JMA earthquake catalogue from October 1997, four (4) DLF earthquakes were observed before the mainshock, located about 8 km west of the epicenter of the mainshock at about 30 km depth. Three events occurred in June and July 1999 and one in June 2000. Another DLF earthquake, which took place 9 hours prior to the mainshock, was detected by inspecting continuous records of the station TRT (Figure 1a). The hypocenter is in the nearby region where the other four DLF earthquakes were located. Location error in hypocenter calculation is less than 500 m both in horizontal and vertical directions.
 The DLF activity increased after the mainshock and more than 60 DLF events were observed during the 13 months after the mainshock (October 2000–October 2001). Magnitudes of these DLF events range from 1.4 to 2.2. Figure 1b shows the space-time plot of DLF events together with ordinary earthquakes from October 1997 to October 2001.
Figure 2a shows the vertical component seismograms of 5 DLF events that occurred before the mainshock, observed at station TRT. Figure 2b is the three-component seismogram of the DLF earthquake on October 6, 2000, that occurred 9 hours before the mainshock. Observed features in the waveforms of these DLF events are as follows; (1) Predominant frequency is 2 Hz–4 Hz, however they are not monochromatic events, (2) Both P- and S- waves are observed however S-waves have larger amplitudes, and (3) The onset of the P-waves have a high-frequency component. Duration of these events are one minute or longer, and in some cases they exhibit durations of several minutes. A DLF event on June 2000 had a duration of more than 5 minutes and continuous occurrence of DLF events were observed in a tremor-like waveform.
 Low-frequency features of the source mechanism is supported from the following observation; (1) High-frequency components in the P-wave onset portion are observed, and (2) Seismic rays of ordinary earthquakes which are several tens kilometers away that samples the focal region of the DLF earthquakes remains high-frequency components. Thus it is hard to assume a strong attenuation in the focal area of the DLF events, and we conclude the low-frequency features are originated from the source.
3. Focal Mechanisms of DLF Earthquakes
 Since magnitudes of these DLF earthquakes are small, it is difficult to adopt waveform inversion techniques, such as moment tensor inversions, to investigate the source mechanisms of the DLF earthquakes. Therefore, we estimated the source mechanism of a DLF earthquake in the same manner that Ukawa and Ohtake  analyzed a DLF event beneath Izu Ooshima volcano. We analyzed the focal mechanism of a DLF earthquake that occurred 9 hours before the mainshock on October 6, 2000. This event was recorded at more than 10 seismic stations, where short-period (1 sec) seismometers are installed (Figure 3).
 First we compared the amplitude ratios of the S-waves to the P-waves with those of three theoretical source models. In order to reduce the effect of site amplification, we used seismograms recorded at Hi-net stations which are installed at the bottom of 100–200 m boreholes to estimate amplitude ratios. Figure 4a shows the amplitude ratios of the maximum S-wave amplitude, As, to the maximum P-wave amplitude, Ap, as a function of azimuth. We obtained Ap from the P-wave amplitude of the vertical component of the seismogram and As from the amplitudes of the SH and SV waves, that is . Amplitudes of the incident SH and SV waves were estimated from the transverse and radial components of the horizontal seismograms taking into account the effect of the Earth's surface.
 The amplitude ratios of the S-waves to P-waves averaged on the focal sphere for three theoretical source models are shown in Figure 4a. The average amplitude ratios for tensile-crack, single-force, and double-couple models are 2.1, 4.2, and 6.4, respectively [Aki and Richards, 1980; Ukawa and Ohtake, 1987].
 The observed amplitude ratios are almost greater than 2.1, that indicates the tensile-crack model does not coincide with the observation. Therefore, next we considered the single-force and the double-couple type source model using the spatial distribution of the S-wave polarization on the focal sphere. We calculated particle motions of the initial portion of the onset of the S-waves. The polarization angles defined by tan−1 (SH/SV), were measured from the direction of the particle motion.
 To avoid the effect of the critical reflection at the surface, seismic velocity profiles of the boreholes were used to estimate the incident angles of the ray for the Hi-net stations. Other than Hi-net stations, a P-wave velocity of 4.0 km/s and an S-wave velocity of 2.0 km/s are assumed. With this criteria, 12 stations were used for analysis. Amplitudes of the SH and SV waves were estimated in the same manner which was used for the amplitude ratio estimation.
 Then we fit the observed polarization data to single-force and double-couple source models. The best fit models were obtained by minimizing the sum of squared residuals (SSR), that is SSR = ΣRi2, where Ri is a residual of the polarization angle at the i-th station. The best fit direction was searched over the focal sphere for a grid points on a 1° mesh.
Figure 4b shows the best fit solutions for each theoretical model. Since the number of degrees of freedom of the two theoretical models are different, the AIC (Akaike's Information Criteria) was used to select the final model. In this case, AIC parameter is expressed as AIC = N log(2π) + N log(SSR/N) + N + 2k, where N and k are number of data and free parameters, respectively. If the difference in AIC between two models are greater than 2, the difference is regarded as significant [Ukawa and Ohtake, 1987]. AIC values for the best fit single-force and double-couple models are 17.7 and 22.5, respectively, therefore the single-force model is a preferable fit to data.
 In our analysis, we selected the best source model out of three simple theoretical models. Here we consider the case of combined source model. A combined source model was proposed by Hill  to explain the swarm activity around volcanoes. It is a combination of share fracture and tensile-cracks. Shimizu et al.  analyzed the volcanic earthquakes in the Miyake-jima volcano assuming a tensile-shear crack model like Hill . Their result shows that the tensile-shear crack model has rather good excitation of P-waves by the effect of tensile-crack. As we showed before, P-wave amplitude of the observed DLF is very small, and thus we suppose it is difficult to assume tensile-shear crack model in our case.
 In previous studies, DLF earthquakes are mostly observed near active volcanoes or the volcanic front and attributed to magmatic activity [e.g. Hasegawa and Yamamoto, 1994; Ukawa and Obara, 1993]. In the focal region of the 2000 Western Tottori earthquake, there are no active volcanoes but a Quaternary volcano, Mt. Daisen is located about 30 km away.
 Seismic tomography analysis [Zhao et al., 2000] indicates the existence of a low-velocity body around 40 km depth in the Chugoku district. The hypocenters of the DLF earthquakes are located near the edge of the low-velocity body. Low-velocity bodies in the lower crust and upper mantle identified from the tomographic analysis are suggested to be due to the effects of slab dehydration and indicate existence of fluids [Zhao et al., 2000].
 The DLF earthquake analyzed in this study is thought to have a single-force type source mechanism, which may be attributed to the transport of fluids such as magma or water [e.g. Ukawa and Ohtake, 1987].
 The depth of the Moho discontinuity in the Tottori region obtained from the explosion seismology is about 38 km [e.g. Hashizume et al., 1966]. Thus the DLF earthquakes are located around or just above the Moho discontinuity. We suggest that these DLF events are evidence of fluid activity in the the western Tottori region at depths around the Moho discontinuity.
 Recent studies [e.g. Hasegawa et al., 2001] indicates the existence of fluids beneath active faults that cause the decrease of shear strength of the fault. DLF earthquakes observed in the western Tottori region are probably another piece of evidence that indicates the existence of fluid beneath an active fault zone.
 Deep low-frequency (DLF) earthquakes were observed beneath the focal region of the 2000 Western Tottori earthquake, southwest Japan. Five DLF earthquakes were detected within three years before the mainshock and more than 60 DLF earthquakes were observed within one year after the mainshock. They are located at depths around or just above the Moho discontinuity. The focal mechanism of a DLF earthquake that occurred 9 hours before the mainshock indicates a single-force source. Seismic tomography analysis shows the existence of a low-velocity body in the lower-crust to upper-mantle portion near the hypocentral region of the DLF earthquakes. Therefore, we conclude that these DLF earthquakes are probably evidence for fluid activity beneath the focal region of the 2000 Western Tottori earthquake. It is important to understand the nature of DLF earthquakes beneath active faults, in relation to the behavior of fluids in the focal region that might trigger the rupture of the fault.
 We are grateful to the National Research Institute for Earth Science and Disaster Prevention (NIED) for providing us with the Hi-net waveform data. We also thank the Japan Meteorological Agency for allowing us to use their waveform data and the Preliminary Determined Earthquake catalogue. This catalogue was compiled from analyzing waveform data of the seismic networks operated by the national universities (Kyoto University and other universities), JMA, and NIED. The General Mapping Tool [Wessel and Smith, 1998] was used for drawing figures.