The 2011 Tohoku-Oki earthquake (Mw9.0) ruptured a large portion of the plate boundary where the coupling was considered weak and represented by sparsely distributed small asperities. A typical asperity break produced an event of around Mw7.5 accompanied by a subsequent expansion of aftershock activity beyond the main rupture zone. Unlike previous large earthquakes, the Mw9.0 earthquake sequence does not show much spatial expansion of aftershock activity, but the areas of slip deficit in the main rupture zone are progressively filled in by aftershocks. We show a clear image of segmentation of physical properties in the subducted slab that has been revealed by joint tomography of P and S wave arrivals. The mapped variations separate the zones of dominant high-frequency radiation down-dip and dominant low-frequency radiation up-dip on the mainshock fault. These changes indicate significant variations in the effective plate coupling that interlocked the source areas of the past distinct earthquakes.
 Over the last few decades an asperity model [Ruff and Kanamori, 1980, 1983; Lay and Kanamori, 1981; Lay et al., 1982] has been developed to characterize the ruptures of large shallow subduction zone earthquakes in context of the strength of plate coupling. The 2011 Tohoku-Oki earthquake (moment magnitude, Mw 9.0) ruptured a large portion of the boundary between the subducting Pacific and the overriding Okhotsk plates where the coupling was considered weak and represented by sparsely distributed small asperities [e.g.,Tajima and Kanamori, 1985a, 1985b] so that such a great earthquake had not been anticipated. In the region off the coast of Miyagi-prefecture in northeast (NE) Japan an earthquake of a magnitude (M) around Mw7.5 was expected with a recurrence interval of 30–40 years based on the past activity over the last 150 years.
 Two days before the great 2011 March 11 earthquake, an Mw7.3 event occurred on the same subduction zone, somewhat further offshore than the anticipated event. A spokesman on a television news program warned that the aftershock activity could last with events of M > 6 for a week or so, presumably based on the assumption that it should decrease with time as represented by the Modified Omori law [Utsu, 1961]. Such a formulation does not account for the spatial extent of the affected area. The March 9 event turned to be a mere foreshock to the Mw9.0 March 11 earthquake that released about 180 times as much energy as an Mw7.5 event, and produced several dozens of aftershocks of M > 6. In reality we only learn whether we have a foreshock, mainshock or aftershock, after the sequence, and did not know in advance the nature of the driving force that controls the propagation of seismic rupture that ended up as an Mw9 event.
 This 2011 megathrust earthquake is one of the greatest ever instrumentally recorded, comparable to the 1960 Chilean (Mw9.5), the 1964 Alaskan (Mw9.2) and the 2004 Sumatra-Andaman (Mw9.0-9.3) events. The previous major earthquake catastrophe in Japan was the 1995 Kobe disaster (Mw6.8). Since then, Japan has installed one of the world's densest real-time seismometer monitoring systems, and the most extensive earthquake early-warning system. The excellent distribution of seismic stations and plentiful earthquake sources allow the production of high resolution seismic tomography models and models for seismic rupture. The advances in the available information have been very significant. Nonetheless, the potential for an impending Mw9 earthquake was not broadly recognized, although analysis of earlier events suggested that only a fraction of the accumulated tectonic stress was being relieved, and that there was indeed a possibility of an event that combined near-trench and near coast slips that in the past few centuries had only occurred in distinct earthquakes [Igarashi et al., 2003; Kanamori et al., 2006].
2. Plate Coupling and Aftershock Expansion Pattern
 In the aftermath of a seismic event there is an expansion of aftershock activity reflecting stress adjustment outside the ruptured areas. A simple tool has been developed to characterize the spatial expansion patterns of aftershocks with time, based upon the amount of seismic energy released from aftershocks per unit area and unit time [Tajima and Kanamori, 1985a]. A linear expansion ratio ηl for the characteristic size of the aftershock zone and an areal expansion ratio ηS are calculated as a function of time. The ratios are calculated from the size at a certain time after the mainshock taft(day) divided by that at 1-day, since the 1-day aftershock area is considered to represent the rupture of an asperity. The results indicate that typical expansion patterns of aftershock areas for major subduction events can be classified into three groups [Tajima and Kanamori, 1985b]: (1) very little expansion associated with a megathrust earthquake, e.g., the 1964 Alaskan earthquake (Mw9.2); (2) little expansion along the strike and some expansion widthwise, e.g., large events in the Kurile subduction zone; (3) large areal and linear expansions, e.g., 1968 Tokachi-oki (Mw8.2), 1978 Miyagi-oki (Mw7.5) in the NE Japanese subduction zone.
 The plate coupling in NE Japan is characterized by sparsely distributed small asperities, each of which could produce an event of Mw∼7.5 to lower 8 (Figure 1a). A typical asperity break shows a complex source rupture process [e.g., Seno et al., 1980], but the driving force of rupture propagation may not be large enough to break through a broad region, as was in the case in the 2011 Mw9 earthquake. The weaker zones surrounding the asperities are likely to be broken by smaller events, aftershocks or repeating small events [Matsuzawa et al., 2002; Igarashi et al., 2003]. The aftershock activity expands significantly into the surrounding areas (asperity group-3). This pattern is in contrast to a fault zone that is characterized by a uniform, large asperity, the rupture of which could produce a megathrust event, e.g., the 1964 Alaska earthquake with a characteristic source-timeτof ∼180 s (group-1). The aftershock area does not show much expansion at taft∼large (≥100 days) (Figure 1b) and the inter-plate coupling is considered strong.
 The rupture pattern of the 2011 Tohoku-Oki earthquake sequence that started with the Mw7.3 foreshock is compared with the patterns postulated previously for this region. The 2-day aftershock area of the March 9 foreshock mainly expanded trenchward until the March 11th event took place at the western edge of the zone (Figure 1c). It seems there is a distinct zone with a strong coupling strength associated with a nucleation of the Mw9 megathrust event. After the March 11 earthquake ruptured the broad region in τ∼ 150 s, however, the aftershock area did not show much expansion over time as compared with the 1-day area in spite of the numerous aftershocks (Figure 1d). Unlike the previous large earthquakes in this subduction zone, this expansion pattern is similar to that of asperity group-1 (Figure 1b). Here we consider the aftershock area to be linked to the main rupture zone, and the large events (M > 6) in Niigata or Shizuoka, which were apparently induced after March 11 as a separate feature.
where the seismic wavespeeds are given by β2= G/ρ, ϕ2 = KS /ρ , in terms of the shear modulus G, bulk-modulusKS, and density ρ. To first order, the measure R is simply the difference (δβ /β − δϕ /ϕ) and so enhances the subtle differences in the behavior of the shear and bulk moduli, since the density ρ is common to both wavespeeds.
 Tomographic images are taken on a plane approximately coincident with the fault surface. Rshows a distinct reduction to zero and slightly negative values (light orange) just up-dip of the mainshock hypocenter (Figure 2b). The down-dip edge of this zone corresponds to the separation between the regions of dominant high-frequency radiation and dominant low-frequency radiation. The March 9 foreshock lies just outside the anomalous zone, as does the 1978 Mw7.5 event. In the lower two panels the images forδϕ/ϕ and δβ/β are shown separately. There is a significant gradient in δϕ/ϕ, a slight negative anomaly near the trench, which then grades to stronger positive anomalies at depth (Figure 2c). In contrast δβ/β is generally strongly positive with just a slight reduction at depths near 25 km, the hypocentral depth of the March 11 event (Figure 2d). The zone of reduced R is largely associated with this reduction in δβ/β with the effects enhanced by the increase in δϕ/ϕ, and appears to have a strong influence on plate coupling over the whole of the rupture area. The edges of the anomalous region we have delineated act as the initiation points for rupture process of the March 2011 sequence as well as for 1978 (Mw7.5) and 1981 (Mw7.0) events (Figure 2a), and these locations will be where the strongest contrasts exist in physical properties. Regional P wave tomography [Zhao et al., 2011] indicates additional significant wavespeed variations above the slab interface that could affect the frictional properties at the plate boundary.
 To the south of rupture zone of the 2011 Tohuku event, we note a further anomaly in R, but in this case we have strong reductions in both the bulk-sound speed and shear wavespeed, a very different configuration from further north. In the study ofLoveless and Meade  this southern zone shows a very different style of plate coupling to the locked area to the north. Once again it appears that structural information can provide useful indications of the state of the subduction system.
 A seemingly robust approach [Ide et al., 2011] used the Mw7.3 foreshock as an empirical Green's function exploiting the similar focal mechanisms and location proximity, and determined the mainshock source process with the features that consist of a small initial phase (area S1 near the hypocenter), deep rupture radiating high frequency energy for up to 40 s (S2), extensive shallow rupture at 60 to 70 s with relatively low-frequency energy radiation (S3), and continuing deep rupture lasting more than 100 s (S4). A unified source model [Koketsu et al., 2011] combining results of many different types, delineates the large slip in the up-dip zone that creates the seafloor movement and tsunami. The presence of the significant R anomaly at the trench line may well be linked to the large coseismic slip. However, we have to be cautious in any interpretation of the very shallowest features since they are likely to have a strong dependence on the event locations used in the tomographic inversion and the particular slice through the model. The tomography image [Gorbatov and Kennett, 2003], created prior to 2003, indicates significant clues to the impending March 2011 rupture process.
4. Interlocking of Effective Plate Coupling
 In the light of the results above, we attempt to formulate a conceptual model of rupture propagation at the subducting plate interface in NE Japan. The distribution of plate coupling strength (or asperities) is illustrated in Figure 3a. The asperities which caused the 1978, 1981, and 2011 March 9 earthquakes are illustrated by dark shading. The area in lighter shade extending to the north corresponds to the zone with R < 0, on which the black zone is superposed to show the initiation of the March 11 mainshock. The edges of the distinct rupture zones are expected to have localized stress concentrations that are relieved in the aftershock activity. Expansions of the aftershock areas have occurred where the stress concentrations migrate beyond the rupture zones of the past large earthquakes [Tajima and Kanamori, 1985a]. Nonetheless, the narrow area of the March 11 mainshock initiation remained intact until the tectonic stress level reached the maximum coupling strength there, which is higher than in the surrounding areas. As noted by Lay et al.  the slip displacement for the same moment release is enhanced by the reduced seismic wavespeeds toward the trench. Thus comparable strength asperities can lead to much larger slip in the near trench zone as in zone S2.
 In Figure 3b the progressive tectonic stress level is drawn schematically in relation to the heterogeneous plate coupling strength associated with the major seismic ruptures since the 1978 earthquake. The distribution of the heterogeneous shear strengths represents the effective plate coupling for interlocking the distinct source areas. When the March 11 mainshock rupture started, the tectonic stress levels in the past distinct source areas were close to the maximum shear strength for these zones. This condition promoted interlocking rupture propagation through the broad region. On the other hand the set of ruptures during the mainshock removed the slip deficit on parts of the plate interface. The rupture gaps are progressively filled in by the aftershocks, particularly those of larger size that fit in the larger ‘holes’ in the main rupture slip distribution (Figures 1d and 3a).
 The combined exploitation of joint seismic tomography and seismic rupture patterns as used for the Tohoku-Oki earthquake may be utilized to extract clues for potential for future interlocking megathrust events (see the image of relative variations ofδβ/β and δϕ/ϕ for the broader region around the Japan islands [Kennett et al., 2011]).
 F.T. appreciates the supports and research environment provided at LMU. She has benefitted from instruction from the late Leon Knopoff who introduced her to the concept of a rupture driving force in heterogeneous plate coupling. Collaboration in Munich was made possible by Humboldt Foundation support to B.K. The review comments by an anonymous reviewer helped to improve the clarity of the paper. “GMT” [Wessel and Smith, 1995] was used to produce Figure 1.
 The Editor thanks the anonymous reviewer for their assistance in evaluating this paper.