An important goal of computational seismology is to simulate dynamic earthquake rupture and strong ground motion in realistic models that include crustal heterogeneities and complex fault geometries. To accomplish this, we incorporate dynamic rupture modelling capabilities in a spectral element solver on unstructured meshes, the 3-D open source code SPECFEM3D, and employ state-of-the-art software for the generation of unstructured meshes of hexahedral elements. These tools provide high flexibility in representing fault systems with complex geometries, including faults with branches and non-planar faults. The domain size is extended with progressive mesh coarsening to maintain an accurate resolution of the static field. Our implementation of dynamic rupture does not affect the parallel scalability of the code. We verify our implementation by comparing our results to those of two finite element codes on benchmark problems including branched faults. Finally, we present a preliminary dynamic rupture model of the 2011 M w 9.0 Tohoku earthquake including a non-planar plate interface with heterogeneous frictional properties and initial stresses. Our simulation reproduces qualitatively the depth-dependent frequency content of the source and the large slip close to the trench observed for this earthquake.

Near-source ground-motion observations and kinematic source inversions suggest that the rupture process of the 2011 M w 9.0 Tohoku earthquake involved rupture reactivation, that is, repeated rupture nucleation in the same hypocentral area. This unusual phenomenon may have provided a second breath to the rupture that enhanced its final size. Here, we propose that rupture reactivation may have been governed by a slip-weakening friction model with two sequential strength drops, the second one being activated at large slip. Such frictional behavior has been previously observed in laboratory experiments and attributed to pressurization of fault-zone fluids by mineral decomposition reactions activated by shear heating, such as dehydration and decarbonation. Further evidence of this double-slip-weakening friction model is obtained here from the dynamic stress changes in the hypocentral region derived from a finite source inversion model. We incorporate this friction model in a dynamic rupture simulation comprising two main asperities constrained by source inversion models and several deep small asperities constrained by backprojection source imaging studies. Our simulation produces ground-motion patterns along the Japanese coast consistent with observations and rupture patterns consistent with a kinematic source model featuring rupture reactivation. The deep small asperities serve as a bridge to connect the two main asperities, and the rupture reactivation mechanism is needed to reproduce the observed ground-motion pattern. Therefore, we argue that rupture reactivation during the 2011 Tohoku earthquake is consistent with a second strength drop, possibly caused by activation of thermochemical weakening processes at large slip.Online Material: Movie of simulated rupture and wave propagation and ground velocities recorded at the KiK-net and K-NET seismic networks.

Megathrust earthquakes nucleate within the seismogenic zone (Byrne et al., 1988;Hyndman et al., 1997). The updip limit of this region may vary depending on material properties and thermal conditions along the megathrust (Hyndman et al., 1997), but it is commonly defined between 5 and 10 km depth (Scholz, 1998). Yet, megathrust earthquakes may occasionally rupture through the shallow and apparently aseismic region of the fault (<5 km of depth), particularly in areas with sediment-starved trenches and irregular subducting

We perform earthquake cycle simulations with the goal of studying the characteristics of source scaling relations and strong ground motions in multi-segmented fault ruptures. The 1992 M w 7.3 Landers earthquake is chosen as a target earthquake to validate our methodology. The model includes the fault geometry for the three-segmented Landers rupture from the SCEC community fault model, extended at both ends to a total length of 200 km, and limited to a depth to 15 km. We assume the faults are governed by rate-and-state (RS) friction, with a heterogeneous, correlated spatial distribution of characteristic weakening distance Dc. Multiple earthquake cycles on this non-planar fault system are modeled with a quasi-dynamic solver based on the boundary element method, substantially accelerated by implementing a hierarchicalmatrix method. The resulting seismic ruptures are recomputed using a fully-dynamic solver based on the spectral element method, with the same RS friction law. The simulated earthquakes nucleate on different sections of the fault, and include events similar to the M w 7.3 Landers earthquake. We obtain slip velocity functions, rupture times and magnitudes that can be compared to seismological observations. The simulated ground motions are validated by comparison of simulated and recorded response spectra.