Subducted seamounts have been linked to interplate earthquakes, but their specific effects on earthquake mechanism remain controversial. A key question is under what conditions a subducted seamount will generate or stop megathrust earthquakes. Here we show results from numerical experiments in the framework of rate- and state-dependent friction law in which a seamount is characterized as a patch of elevated effective normal stress on the thrust interface. We find that whether subducted seamounts generate or impede megathrust earthquakes depends critically on their relative locations to the earthquake nucleation zone defined by depth-variable friction parameters. A seamount may act as a rupture barrier and such barrier effect is most prominent when the seamount sits at an intermediate range of the seamount-to-trench distances (20–100% of the nucleation-zone-to-trench distance). Moreover, we observe that seamount-induced barriers can turn into asperities on which megathrust earthquakes can nucleate at shallow depths and rupture the entire seismogenic zone. These results suggest that a strong barrier patch may not necessarily reduce the maximum size of earthquakes. Instead, the barrier could experience large coseismic slip when it is ruptured.
 Seamounts are ubiquitous topographic features of the seafloor and have profound effects on megathrust earthquakes once they enter subduction zones [Cloos, 1992; Scholz and Small, 1997; Watts et al., 2010; Trehu et al., 2012]. They have been suggested to lead to additionally frictional resistance on a megathrust interface [Scholz and Small, 1997]. Such high-resistance areas may act as asperities where large earthquakes nucleate [Cloos, 1992; Husen et al., 2002; Bilek et al., 2003; Hicks et al., 2012], or as barriers that inhibit coseismic rupture propagation [Kodaira et al., 2000]. If such strong barriers are broken, they may induce large coseismic slip concentrated in the barrier region as suggested for the 2011 off Tohoku, Japan Mw 9.0 earthquake [Simons et al., 2011; Duan, 2012]. On the other hand, subducted seamounts have also been suggested to be associated with lower interplate coupling due to the presence of entrained fluid-rich sediments on the top of seamounts [Mochizuki et al., 2008; Singh et al., 2011], which may stop ruptures but do not generate large earthquakes. Furthermore, it was proposed that a seamount may generate a complex fracture network in its vicinity, which is unfavorable for the generation of large earthquakes [Wang and Bilek, 2011]. Thus the effect of subducted seamounts on megathrust earthquakes could be complex and remains open for debate. In this study we develop a generic subduction fault model to quantitatively evaluate the effect of a seamount on megathrust earthquakes. Assuming that a subducted seamount increases the local normal stress on the fault, we investigate what parameters of a seamount, e.g., its size and relative location, control its effect on megathrust earthquake nucleation and rupture propagation.
2. Method and Model Parameters
 We simulate slip and stress evolution during earthquake cycles on a 2D planar thrust fault, similar to previous numerical studies of earthquake sequences in subduction zones [Stuart, 1988; Kato and Hirasawa, 1997]. The fault is embedded in a homogeneous elastic half space, dips at 12°,and extends 240 km in the down-dip distance. The elastic domains have a shear wave velocity of 3.0 km/s, shear modulus of 30 GPa, and Poisson's ratio of 0.25. A constant long-term plate convergence rate ofVpl= 37 mm/yr is imposed at the down-dip end of the fault. Our numerical experiments are performed in the framework of rate- and state-dependent friction [Dieterich, 1979; Ruina, 1983], with the “ageing” evolution law:
where effective normal stress is the difference between normal stress σ and pore pressure p, a and b are rate and state friction stability parameters, V0 is a reference velocity, dc is the critical slip distance, f0 is the nominal friction at steady state when V = V0, and θ is the state variable. Friction stability parameter a − bdefines the velocity-weakening (VW,a − b< 0, potentially unstable) and velocity-strengthening (VS,a − b> 0, stable) regimes on the fault. Stress transmission between fault segments is computed quasi-dynamically using the solutions of stress induced by shear dislocation in an elastic half space [Okada, 1992] and a radiation damping term that assures the governing equations continue to have well-defined solutions during slip at seismic rates [Rice, 1993]. Although details of rupture propagation such as slip velocity and rupture speed are different for the quasi-dynamic and full-dynamic solutions, qualitative earthquake behaviors such as earthquake nucleation and rupture extent are similar as shown in a direct comparison between these two approaches [Lapusta and Rice, 2003]. Thus, the quasi-dynamic approach is appropriate for our investigation on the nucleation and extent of rupture propagation under the influence of seamounts.
 The rate and state friction parameter a − balong the subduction fault is converted from temperature-dependent wet granite friction data [Blanpied et al., 1998] using a Cascadia thermal model [Peacock et al., 2002], following Liu and Rice . Table 1 lists their values at depth. Normal stress on the fault is assumed as σ = ρcgz, where ρc is density of crust, g is gravitational acceleration, and z is depth. Pore pressure pin subduction zones has been proposed as over-hydrostatic, even near-lithostatic at the up-dip and down-dip ends of the seismogenic zone [Saffer and Tobin, 2011]. We here use a constant effective normal stress at depth for simplicity (e.g., = 50 MPa) [Rice, 1992]. In addition, the effective normal stress is taken to be time-constant, thus does not incorporate any potential pore pressure changes induced by dilatancy or thermal pressurization during fault slip [Liu and Rubin, 2010; Segall et al., 2010; Noda and Lapusta, 2010].
 In our model a seamount is approximated by a strong patch with an elevated effective normal stress Δ from the background (Figure 1). The amplitude of Δ is proportional to the height of the seamount, or more precisely, the seamount-induced flexural displacement of the upper plate [Scholz and Small, 1997]. Assuming the seamount will flex the upper plate by Δh, Δ is approximately , where D is flexural rigidity of the upper plate, w is basal width of the seamount, and C is a constant in the range of 1/3 to 1 associated with pore pressure variation [Scholz and Small, 1997]. The range of Δh depends on the sediment thickness in the subduction channel and properties of the overriding plate [Watts et al., 2010]. By assuming a flexural displacement up to half of the seamount height for sediment-starved arcs with little accretionary prism,Scholz and Small  estimated Δ = 100 MPa for a 4-km-high and 60-km-wide seamount. If the upper plate is composed of accretionary sediments, Δ could be smaller [Honkura et al., 1999]. However, shearing off seamounts during subduction will smear out the excess mass and redistribute the elevated normal stress over a broader area, resulting in Δ of orders of magnitude smaller, e.g., 6.5 MPa [Scholz and Small, 1997]. Therefore we examine the sensitivity of our results to the seamount-induced Δ in the range of 0 to 200 MPa.
Linear interpolation was applied between pivotal points.
 Generally seamounts have sizes ranging from a few to tens of kilometers in width. Subducted seamounts have been discovered at depths from a few km to 40 km [Watts et al., 2010; Singh et al., 2011]. Therefore, we model a seamount with different sizes from a very shallow depth to the down-dip VS zone. In order to compare and apply our results to a wide range of subduction zone models and rock rheology, we adopt dimensionless quantities for the down-dip distanced0 and basal width w of a seamount, , where Lis the distance between the trench and the up-dip edge of the earthquake nucleation zone in the reference model (L ≈ 50 kmFigure 2a). Consequently, we define the dimensionless down-dip distance on the fault, , where xis the down-dip distance (0–240 km). We also use to represent the non-dimensional amplitude of the increase in effective normal stress induced by the seamount.
 We investigate the role of seamount on earthquakes by changing d* from 0 to 1.8, w* from 0 to 1.0, and σ* in the range of 0 to 4. Simulation results of a reference case without seamount show four megathrust earthquakes for the simulated 1,000 years (Figure 2a). Ruptures nucleate at approximately x* = 1 and break the entire VW zone. The fault at greater depth slips stably under VS condition.
 In the experiments with a seamount, we find that its down-dip locationd* is a critical parameter for the seamount to act as a barrier. Honkura et al. suggested that a seamount at shallow depth might prevent ruptures from breaking to the trench, thus reducing the tsunamigenic potential. In our model with a small up-dip VS zone, however, a shallow seamount sitting up-dip or near the shallow VS-to-VW transition zone (d* < 0.2) cannot stop ruptures (Figure 2b). A seamount can significantly slow down or even stop ruptures if it is subducted to an intermediate distance above the nucleation zone (0.2 < d* < 1.0) (Figures 2c and 4). For a seamount subducted further down-dip into the nucleation zone, ruptures propagate through the VW area without significantly slowing down, resulting in higher coseismic slips (Figure 2d). If a seamount has subducted beyond the VW zone, then it has nearly no effect on earthquake ruptures (Figure S1 in the auxiliary material).
 The amplitude of σ* also plays an important role on the barrier effect (Figure 3). At a given seamount location, a greater value of σ* makes a seamount more likely to be able to stop ruptures unless d* is too small. As stated above, a shallow subducted seamount cannot impede ruptures even with an extreme value of σ* = 4. In the intermediate range of the seamount-to-trench distance, however, a much smallerσ* would make the seamount a barrier. The minimum σ* that is required for a seamount to stop at least one rupture in a sequence of modeled earthquakes also depends on the seamount width (Figure 3b). In our simulation, it is easier for a larger seamount to impede a rupture. Our results show that a seamount is able to stop ruptures in the normalized down-dip distanced* ∼ 0.5–1 with σ* = 0.1 (Figure 3c). Taking the upper plate properties assumed by Scholz and Small , Δ = 5 MPa (σ* = 0.1) translates to tens of meters of flexural displacement of the upper plate for a large seamount, e.g., w = 40 km.
 In addition to acting as a barrier, a subducted seamount may also change the location where ruptures nucleate on a megathrust. For instance, we consider a seamount of w* = 0.24 subducted to d* = 0.75 (Figure S2). With σ* = 0.1, the seamount inhibits ruptures of Events 3 and 5 (Figure S2a). Consequently, post-seismic and inter-seismic shear stresses up-dip of the barrier are increased (Figure S2b), which led to the nucleation of Event 4 in the shallow part of the VW region. This is consistent with the static stress transfer induced seismicity on continental faults [King et al., 1994]. Rupture of Event 4 appears to be slow and does not break the entire VW region because most tectonic stresses there have already been released by previous events (Figure S2b). However, the seamount does not always stop ruptures and could be swept by a large megathrust earthquake, e.g., Event 6 (Figure S2a).
 Furthermore, the seamount may become an asperity that initiates megathrust earthquakes (Figure 4). Event 4 is stopped by the seamount, but ∼30 minutes later another earthquake (Event 5) initiates on the seamount and ruptures the entire VW zone (Figure 4). Similar features are also found for seamounts of different sizes and locations (Figure S3). This illustrates that megathrust earthquake could initiate on a subducted seamount, as was suggested as a source for the 1990 Mw 7.0 earthquake in the Gulf of Nicoya, Costa Rica [Husen et al., 2002]. If the seamount is located at a shallow depth, as in Figure 4, shallow megathrust earthquake nucleation may also become plausible, contrary to the conventional view that large earthquakes always nucleate near the base of the seismogenic zone [Das and Scholz, 1983].
 In our generic model, we have used wet granite friction data, which are most systematically documented in fault-sliding laboratory experiments and most commonly adopted in numerical studies of earthquake nucleation and earthquake cycles. Application of different rock friction properties, such as gabbro or olivine, in the current model will change the depth distribution of friction parameters and hence the size of the VS and VW zones. For example, application of the hydrothermal gabbro gouge friction data [He et al., 2007] leads to a VW-to-VS transition at a greater depth and a much broader transition zone than that using the wet granite data [Liu and Rice, 2009]. The nucleation depth, amplitude of coseismic slip, and recurrence interval of megathrust earthquakes are consequently different. However, the qualitative behaviors such as a coseismic rupture impeded by a highly compressed patch (seamount) are similar (Figure S4), and the main conclusions of this study remain valid.
 Subducted seamounts may also weaken the coupling between the overriding and downgoing plates due to the presence of entrained fluid-rich sediments [Mochizuki et al., 2008; Singh et al., 2011]. The sediments in the vicinity of a seamount may increase the local thickness of the gouge layer on the megathrust fault. A thicker fault gouge could stabilize slip and thus impede rupture propagation due to an increase in the friction stability parameter a − b and/or an increase in the characteristic slip distance dc [Marone, 1998]. Kaneko et al.  have investigated the relation between interseismic coupling and earthquake rupture patterns through numerical experiments, in which a VS patch separated two VW segments along the strike direction. They found that the VS patch could also be broken by accumulated stress due to ruptures on the neighboring segments. The probability for an earthquake to break through the VS patch was shown to correlate with the average degree of interseismic coupling. Although we have considered a seamount as a strong patch on the fault, our results show that the seamount may not act as a permanent barrier (i.e., one that always stops ruptures) in earthquake cycles.
 Our model has produced rich earthquake phenomena, including rupture barrier, stress transfer, and nucleation of megathrust earthquakes at relatively shallow depth, despite the simple model approach to treat a subducted seamount as a patch under a higher normal stress. The presented work clearly shows that the distance between a subducted seamount and the earthquake nucleation zone, as defined by depth-dependent rate and state friction parameters, is a key factor to determine whether seamounts would stop or generate megathrust earthquakes. When it is located at shallow depth or deeper than the seismogenic zone on a megathrust, the seamount has little effect on megathrust earthquakes. If it sits in the intermediate distance range up-dip of the nucleation zone, however, the seamount can inhibit or nucleate earthquake ruptures depending on the stress conditions. Our results thus provide insights into understanding the role of mechanical heterogeneities in subduction zone seismic cycles.
 This work is supported by the NSF Grant EAR-1015221 and WHOI Deep Ocean Exploration Institute awards 27071150 and 25051162. We benefited from discussion with Pablo Canales, Yoshihiro Kaneko, Min Xu, and the WHOI Marine Tectonics Group. We are grateful to Anne Trehu and one anonymous reviewer for providing insightful comments, which have improved the paper.
 The Editor thanks Ann Trehu and an anonymous reviewer for their assistance in evaluating this paper.