Text S1. DEM Modeling

In our implementation of the discrete-element-method (DEM), we simulate an assemblage of discrete spherical particles that interact with each other according to elastic-frictional (Hertz-Mindlin) contact law. We introduce cohesion by adding a mechanical bond at interparticle contacts (Morgan, 2015). This property can allow us to simulate cohesive materials as may occur within the deeper accretionary prism. The more frontal region of a prism is approximated as non-cohesive, i.e., lacking interparticle bonds. The combination of pre-assigned interparticle contact parameters, in combination with the mechanical properties of the particles themselves, define the overall behavior of the particle assemblage. This study uses a version of DEM implemented in RICEBAL. Details about the method are provided in Morgan and Boettcher (1999), Guo and Morgan (2004; 2006), and Morgan (2015). Continuum approximations of the bulk properties and behavior of the numerical model are derived using the contact force distribution and displacement gradients. By averaging continuum properties over finite volumes, stress and strain fields can be calculated for the domain (Thornton and Barnes, 1986; Morgan and Boettcher, 1999; Morgan and McGovern, 2005a; 2005b; Morgan, 2015).
Figure 1 shows the general model setup. The initial length of the simulated wedge is set to 200km, comparable to the dimensions of the rupture lengths along the Chile Margin (Moreno et al. , 2010; Contreras-Reyes et al. , 2017). To best balance model run time and model resolution, the upper wedge is constructed of approximately 200,000 discrete particles with radii of 100, 120, 160 and 200 m. Particles are randomly generated within a two-dimensional domain 400\(\times\)60 km, allowed to settle under gravity, and then trimmed to the desired wedge geometry.
The mechanical properties of the simulated system are defined by the assigned particle properties and interparticle friction coefficients (Table S1 and Table S2). The gray particles defining the basal sliding surface are fixed in space, their small radii (10 m) ensure a relatively smooth sliding surface, unimpeded by particle roughness.
The contacts between particles, distinguished by labeled surface types (Figure 1), is controlled by the interparticle friction coefficients. The basal sliding surface is defined by Surface Type 4. The upper plate as is partitioned into an inner (velocity-weakening) wedge (Surface Type 1) and outer (velocity-strengthening) wedge (Surface Type 3) domains. The splay fault boundary defined by Surface Type 1 and Surface Type 2. In addition, cohesiveness, as an option, can be implemented by adding bonds to every contact to simulate more realistic cohesive material. The mechanical properties of the domain and the fault interface are controlled by the particle properties and interparticle friction coefficients. To simulate a predefined fault, relatively low or zero interparticle friction coefficients can be assigned to the particles that define the fault interface, with no cohesion at the contacts between those particles.