This synthesis links many seismic and tectonic processes at subduction zones, including great subduction earthquakes, to the sinking of subducted plate. Earthquake data and tectonic modeling for subduction zones indicate that the slab pull force is much larger than the ridge pusJa force. Interactions between the forces that drive and resist plate motions cause spatially and temporally localized stresses that lead to characteristic earthquake activity, providing details on how subduction occurs. Compression is localized across a locked interface thrust zone, because both the ridge push and the slab pull forces are resisted there. The slab pull force increases with increasing plate age; thus because the slab pull force tends to bend subducted plate downward and decrease the force acting normal to the interface thrust zone, the characteristic maximum earthquake at a given interface thrust zone is inversely related to the age of the subducted plate. The 1960 Chile earthquake (M w 9.5), the largest earthquake to occur in historic times, began its rupture at an interface bounding oceanic plate < 30 m.y. old. However, this rupture initiation was associated with the locally oldest subducting lithosphere (weakest coupling); the rupture propagated southward along an interface bounding progressively younger oceanic lithosphere, terminating near the subducting Chile Rise. Prior to a great subduction earthquake, the sinking subducted slab will cause increased tension at depths

A relocated earthquake sequence in the western Aleutian arc lasting from February 1965 through December 1968 is shown to map an episode of subduction of the Pacific plate. The February 4, 1965, Rat Island main shock (Ms = 8.1) was a complex rupture involving greatest underthrusting at lock zones beneath the Aleutian arc transverse canyons. The primary tectonic consequences of the main shock rupture were (1) the downslab diffusion of a compressional pulse, (2) the oceanward diffusion of an extensional pulse in the oceanic lithosphere, and (3) the slow rebound of the continental plate. The velocities of these strain pulses (actual plate motions) are determined by the viscosity of the constraining mantle material; observed strain velocities indicate a Newtonian mantle viscosity of 6×1019 P. The extensional pulse that propagated oceanward is shown to determine the locations and normal‐faulting mechanisms of the trench earthquakes of this sequence. It is inferred that the entire subducting oceanic lithosphere contains a sequence of near‐vertical faults that strike parallel to the trench axis, reflecting former trench‐related normal‐faulting earthquakes. This condition is supported by considerable data and suggests a slabbing, escalatorlike descent mechanism for the upper 100 km of the subducting Pacific plate. A working hypothesis is presented that relates the primary features of the typical volcanic arc and interarc basin to episodic downslab compressional pulses. This hypothesis involves a sequence of high‐loading‐rate mantle compression, viscoelastic rebound, pressure reduction, increased partial melt concentrations, and counterflow in a thin tabular zone above the subducting plate.

The great 1977 Sumba earthquake occurred at the eastern Sunda trench, just west of the collision of Australian continental lithosphere with the island arc. The length of the aftershock zone of this normalfaulting earthquake is about 200 km. Aftershocks are concentrated 65-115 km east of the main shock epicenter, with very few aftershocks in a 50-km-long segment that spans the main shock epicenter. Relocated hypocenters and focal mechanism data are consistent with normal faulting throughout the upper 28 km of the oceanic lithosphere. There is no evidence for thrust faulting of the deeper aftershocks. These data imply that the neutral bending surface must be at least 35-40 km deep. A second aftershock zone, about 180 km northwest of the main shock, became active immediately following the main shock, but events were concentrated during days 50-52. This zone is a 70-km-long lineation that trends toward the main shock epicenter and reflects right-lateral, strike-slip faulting within the subducted oceanic plate. Seismicity exists to a depth of about 650 km in the very old plate beneath the Sunda-Banda arc, and that plate's negative buoyancy causes very large slab pull forces. Great interface thrust earthquakes are absent at the Sumba region, and slab pull forces are inferred to have partially decoupled the subducted plate from the overriding plate. This decoupling permits slab pull stresses to be guided updip to the region of the Sumba main shock. Such shallow-acting slab pull provides a bending moment at the trench and explains the deformation and timing observed for the entire Sumba earthquake series. In this model, slab pull forces stretch the subducted plate until the increasing stresses at the shallow subduction zone lead to a subduction zone earthquake. Postseismically, the released oceanic plate undergoes a pulse of downdip strain, returning the plate to a less extended state. The moment of this downdip plate motion could exceed the seismic moment of the main shock.

Focal mechanism solutions for shallow earthquakes throughout the Cascadia plate system indicate that the primary regional stress is northerly compression, even though the Juan de Fuca plate generally is thought to be subducting N50øE. This compressional stress is pervasive throughout the Gorda-Juan de Fuca-Explorer plate system and much of the adjoining section of North American plate. Modeling, using a discrete element code, shows that this stress primarily is due to the Pacific plate being driven into the Gorda block and Juan de Fuca plate (at the Mendocino and Blanco fracture zones). This plate collision causes compression of the offshore plate system northward through the 45øW trending Vancouver Island and leads to transfer of northerly compression into the overriding plate. The northerly compression in the Cascadia plate system has caused large, crustal earthquakes in the regions of Vancouver and northern California and may be capable of causing large earthquakes also in and offshore of Washington and Oregon. Several independent lines of evidence confirm that the Cascadia subduction interface is coupled strongly and that subduction primarily must be accompanied by significant earthquakes, rather than occurring aseismically. For numerous earthquakes within the subducted plate, focal mechanisms indicate extensional stresses that trend downdip. These earthquakes reflect plate extension that is due to sinking of the more deeply subducted plate and to plate motion being resisted at the shallow subduction interface. The slab pull force of the sinking plate also can lead to earthquakes at the shallow subduction interface and the potential exists for M 7.5-8.0 subduction earthquakes to occur at segments of the subduction boundary at Washington and Vancouver and less frequently so at segments of the subduction boundary at Oregon. The in-plate driving forces of the offshore Cascadia plate system do not appear to contribute significantly to regional stresses. Stresses and earthquakes at the Cascadia province primarily result from the action of the Pacific plate on the Gorda-Juan de Fuca-Explorer plate system, a transfer of this action into the overriding North American plate, and the independent sinking and slab pull force of the subducted oceanic lithosphere.

S U M M A R YOuter-rise seismicity and dynamics are examined using inelastic models of lithospheric deformation, which allow a more realistic characterization of stress distributions and failure behaviour. We conclude that thrust-and normal-faulting outer-rise earthquakes represent substantially different states of stress within the oceanic lithosphere. Specifically, the normal-faulting events occur in response to downward plate bending, which establishes the 'standard', bending-dominated state of outer-rise stress, and the thrust-faulting events occur in response to an elevated level of in-plane compression, which develops only in response to exceptional circumstances. This interpretation accounts for the observation that normal-faulting outer-rise earthquakes occur more frequently and are more widely distributed than their thrust-faulting counterparts, an observation for which the simple bending model offers no explanation. In addition, attributing both thrust-and normal-faulting outer-rise earthquakes to plate bending implies that both classes of events should occur within relatively close lateral proximity to one another because both are allegedly a manifestation of the same bendingdominated stress distribution, whereas, in reality, this is not observed. We propose that the tendency for thrust-faulting outer-rise earthquakes to exhibit greater source depths than their normal-faulting counterparts (an observation that is frequently cited in support of the bending interpretation of the former) is merely a consequence of the fact that bending-induced tension is confined to the upper lithosphere. Our model predicts that outer-rise in-plane-force variations may promote thrust-faulting outerrise activity prior to an underthrusting interplate subduction earthquake and normalfaulting outer-rise activity following such an earthquake, but that both forms of outerrise activity are unlikely to be associated with the same subduction earthquake. A corollary implication of our model is that subduction earthquakes are likely to be either preceded by or followed by an absence of large outer-rise earthquakes. Levels of in-plane compression necessary to generate thrust-faulting outer-rise earthquakes are attributed to stress concentrations within the subducting plate that are induced by relatively localized resistance to regionally distributed plate-driving forces. Resistance of this nature may result from either the attempted subduction of relatively buoyant (i.e. isostatically compensated) bathymetric features or the existence of strong interplate asperities.