Various models for the generation of tsunami earthquakes have been proposed, including shallow earthquake slip through low strength materials. Because these physical fault conditions would likely affect other earthquakes in the same rupture zone, source properties of other events may provide a guide to locations of tsunami earthquakes. The 25 October 2010 Mw = 7.8 Mentawai tsunami earthquake and surrounding events provide a test of this hypothesis. We determine slip patterns for the mainshock and relocate aftershocks, with the majority occurring in the near trench region. The two largest magnitude aftershocks occurred within the downdip end of the mainshock rupture area and have long moment-normalized rupture duration, likely related to fault zone conditions. Several older relocated earthquakes at the northern edge of the 2010 rupture area also have long duration character, suggesting both spatial and temporal consistency in the conditions needed to produce slow seismic processes along this margin.
 Given that models link shallow slip in weak near-trench materials to tsunami earthquake occurrence, an important question is whether these fault conditions also impact the rupture of other events in the same area. Our efforts here address this question by first defining the rupture area and source characteristics of the 2010 event as well as relocating aftershocks of the event. We then compare the rupture extent to source parameters computed for other regional earthquakes in order to assess the possibility of consistent slow behavior along specific patches of the plate interface.
 Our source analysis of the 2010 mainshock incorporates both broadband P and SH waves recorded at teleseismic distances (30–90°) and a large suite of long period vertical component seismograms for the Rayleigh wave (R1) signal used in the relative source time function deconvolution. We also determine source parameters for 74 moderate magnitude events from 1990–2009 (Mw > 5.5, thrust mechanism based on Global Centroid Moment Tensor (GCMT, www.globalcmt.org) solutions) using teleseismic P and SH waves.
3.1. Source Parameters: Mainshock
 Estimates of the rupture extent are initially developed using a theoretical Green's function deconvolution [e.g., Lay et al., 2009] with 59 R1 surface wave records that are azimuthally well distributed around the source. Point source synthetic seismograms are created using normal mode summation for each distance and azimuth range represented by the data collected using PREM [Dziewonski and Anderson, 1981] for the velocity model. These synthetic seismograms are deconvolved from the observed seismograms to remove theoretical path effects, leaving a relative source time function (STF) to describe the source effects at each station [e.g., Lay et al., 2009] (Figure 2a). We pick the onset and end of each STFs main source pulses (dt) and using the definition for the directivity parameter [Γ, Γ = cos(station_azimuth − rupture_azimuth)/phase_velocity], solve for the best fit rupture duration, azimuth, and duration length (Figure 2a).
 We also use broadband body waves (P and SH) to invert for fault finiteness, obtaining estimates of the overall rupture duration and slip distribution on the fault plane (M. Kikuchi and H. Kanamori, Note on teleseismic body-wave inversion program, 2006, http://www.eri.u-tokyo.ac.jp/ETAL/KIKUCHI). We use a grid of 25 km along strike by 40 km in dip direction to define moment subevents on a fault plane defined by the GCMT geometry. Subevent moment pulses, defined with 5 overlapping triangles of 10 s duration, are removed from the P and SH waveforms at various timing and fault grid positions, allowing for synthetic seismograms to be constructed for each of the observations. Moment rate function and moment distribution at each grid point are computed, and we choose the solution that minimizes misfit between the observed and synthetic seismograms (Figure 2b). Slip is computed from the moment distribution, using a value of 20 GPa for fault rigidity, appropriate for the shallow region of this fault zone based on global and regional estimates from earthquake parameters [Bilek and Lay, 1999; Bilek, 2007]. Peak slip is 4.7 m in the area near the epicenter, similar to the 4.5 m determined by Lay et al. , but less than the 9.6 m estimated by Newman et al. . Moment rate functions, duration, and spatial distribution of slip are similar between each study.
3.2. Source Parameters: Moderate Magnitude Events
 For moderate magnitude earthquakes that occurred prior to the 2010 mainshock as well as 3 aftershocks, we also determine source time functions and depth estimates for each event by deconvolving a point source Green's function from the P and SH waves, solving for the source time function at a range of depths between 3–65 km [e.g., Ruff and Miller, 1994; Bilek and Lay, 1999]. Synthetic seismograms generated for each depth are compared with the observed seismograms, with the optimal time function and depth producing the lowest misfit. In all cases these depths are compared to the relocated depths (section 3.3). Duration is measured from the final event time function as the time from onset to the first zero crossing, capturing the majority of the moment release of the event. These durations are then scaled by the cube root of seismic moment (Mo) to remove the effect of increasing duration with increasing seismic moment, and normalized by the duration of an Mw = 6.0 event [e.g., Bilek and Lay, 1999].
3.3. Earthquake Relocations
 Teleseismically recorded earthquakes along the Sunda system have been relocated using the Engdahl, van der Hilst, and Buland (EHB) method [Engdahl et al., 1998]. This method uses reported International Seismic Centre (ISC) and National Earthquake Information Center (NEIC) reported phases and increases location accuracy using iterative relocation with dynamic phase identification, variable phase weighting, ellipticity and station patch corrections, and the ak135 velocity model. Only a small percentage of depth phases (pP, pwP, sP) are reported to ISC/NEIC, but the inclusion of these phases can significantly improve depth determination.
 For earthquakes prior to 2010, we have integrated additional depth phases derived using automated methods to better constrain hypocentral depths. This frequency-based technique identifies depth phase onset times on both velocity and displacement waveforms by searching for abrupt changes in the gradient of the power spectral density function (see Pesicek et al.  for further details). For the 2010 mainshock and aftershocks, depth phases were derived using the automated method but also verified by analyst (Figure 3). Analyst verification was done because depth phases generally have low frequency content as they propagate through near surface layers where there is high attenuation and may be obscured in the P-wave coda or noise, making auto-identification difficult. Additionally, not all ISC phase picks for the 2010 sequence have been reported due to the 2-year lag in compiling such a large database. For this study, it is imperative that reported depth phases for potentially shallow and complex earthquakes be correct.
 We find a common depth that satisfies both arrival-based and waveform modeling depth constraints for all earthquakes along this margin (Figure 4) (see auxiliary material). Earthquakes with multiple local minima in the waveform modeling misfit function are flagged, and the arrival residual associated with the corresponding earthquake location re-analyzed. In many cases, the arrival data can help identify the correct local minima to interpret for the waveform model, or vice versa.
 The source parameters determined for the mainshock suggest that the 2010 event was indeed a tsunami earthquake, with rupture duration of ∼130 s. The mainshock ruptured NW from the epicenter, extending for approximately 110–150 km in length (Figure 2). This rupture length is consistent with the extent of the relocated aftershocks updip of the mainshock and to the NW. The northern edge of the rupture area extended into a portion of the megathrust zone where long duration events occurred in 2005 and 2007 with normalized durations of 5.9 s and 6.9 s, respectively (Figure 4). These past events were deeper (30 km and 28 km) than the majority of slip observed in the 2010 mainshock.
 The two largest aftershocks of the 2010 mainshock had high signal to noise ratios and are suitable for the body wave deconvolution analysis. These events had longer than average NSD (5.5 s and 8.7 s) relative to the mean value for the entire margin (4.7 s). These events are deeper than the mainshock and the majority of the rest of the aftershocks, but lie in the depth range with the majority of the other past events in our dataset (Figure 4). The other relocated aftershocks are located in the very shallow portion of the seismogenic zone, shallower than much of the previously observed seismicity in this region.
5. Discussion and Conclusions
 The observed shallow slip and long duration source characteristics of the mainshock suggest that this event was indeed a tsunami earthquake. Our relocated aftershocks provide further evidence of very shallow, near trench rupture. We also show that the rupture extends NW into an area of the fault that exhibited other slow processes, specifically the long normalized rupture durations of previous events.
 In addition, we show that two of the largest aftershocks also had longer than average durations. Only one of the older earthquakes occurred in the aftershock region, and this event had a shorter than average duration (2.6 s). This difference may indicate that the mainshock slip could modify fault zone conditions at the downdip edge to cause these long duration events. However, because we have only one older event in this region, we cannot rule out the possibility that the fault conditions responsible existed previously and were not significantly modified during mainshock slip. Unfortunately, the smaller magnitude aftershocks at the updip edge had poor signal to noise ratio records and we are unable to assess their source durations with the methods used here.
 Even with the limited dataset, our results suggest a spatial link between slow seismic processes over a range of depths along this segment of Sumatra. Short duration events occur along much of the margin, however, only a few isolated regions based on the catalog to date, appear to produce these long duration events. Events that have NSD 1 standard deviation above the mean value locate in 3 regions: near the 1907 tsunami earthquake, in the region just to the north of the 2010 rupture extent, and one event at ∼5.5°S.
 One model to explain this is the presence of distinct, spatially constrained patches on the megathrust that provide the required heterogeneous frictional conditions to produce both larger magnitude tsunami earthquakes and the more frequent smaller magnitude long duration events [Bilek and Lay, 2002]. Additional geophysical characterization of this segment of the Sumatra margin would be necessary to determine if there are other variations in subduction inputs that might influence the frictional conditions.
 Interestingly, Kanamori et al.  characterized the 1907 earthquake as a tsunami earthquake based on historical seismogram analysis. The 1907 earthquake occurred beneath Simeulue Island, a region where we also find a long duration event (1993/09/01). This region is also at the boundary between the southern rupture extent of the 2004 Aceh-Andaman Mw = 9.2 and the 2005 Nias-Simeulue Mw = 8.7 events [e.g., Ammon et al., 2005; Hsu et al., 2006]. Dean et al.  note that this boundary between the 2004 and 2005 ruptures corresponds to changes in the likely decollement and lowermost sediment properties. These changes in sediment properties are one possible link to variations in frictional stability of portions of the fault zone. Briggs et al.  and Hsu et al.  suggest that the fault beneath central Simeulue Island tends to slip aseismically given the small amounts of coseismic slip there in the past great earthquakes. However, it may be that this area is another spatially constrained patch with fault conditions ripe for slow seismic processes, but not for coseismic slip in great earthquakes.
 In addition to the spatial patterns, our results also suggest that these fault zone conditions may persist for years. The events with long duration in the northern section of the 2010 rupture area occurred in 2005 and 2007. The occurrence of the 1907 tsunami earthquake in the region of another long duration event in 1993 suggests that these fault conditions may persist for decades. Additional analysis of events in other tsunami earthquake regions can be used to examine these patches in other areas, but in some cases, the earthquake catalog might not be sufficient to see this effect. The Sumatra margin is an area that has experienced many earthquakes in the last decade, allowing us to define these patches of long duration events as areas that may also produce tsunami earthquakes.
 We gratefully acknowledge NSF funding for this project, NSF-OCE 0840908 (SLB) 0841040 (ERE) and 0841022 (HRD). All waveform data was obtained from the IRIS Data Management Center.
 The Editor thanks Richard Briggs and an anonymous reviewer for their assistance in evaluating this paper.