This work takes a new approach to a long-standing question in geophysics: Why do intermediate- and deep-focus earthquakes occur? Since deep earthquakes only occur in subducting slabs, suggested explanations for their origin, such as transformational faulting and dehydration embrittlement, are related to processes expected to occur in the slab. In the dehydration embrittlement hypothesis, seawater permeates into outer-rise faults that form in the oceanic crust prior to subduction. As the slab subducts, the weak zone is preserved and then reactivated by dehydration reactions. Thus, the dehydration-embrittlement mechanism, as well as any mechanism that requires the reactivation of surface faults, offers a testable hypothesis: are the fault-plane orientations of outer-rise and deeper events the same after the angle of subduction has been accounted for? Alternatively, are the fault-plane orientations of deep events more consistent with the slab stress field? Previous evidence supporting the pre-existing fault hypothesis has generally come from either studies that identify the fault planes of individual events, often as a by-product of a more comprehensive analysis of the rupture process, or analyses of the statistics of the nodal-plane orientations of numerous earthquakes within a given subduction zone, but without resolving the fault plane ambiguity. This study utilizes a novel method that combines the advantages of both of these approaches , identifying the actual fault planes and analyzing large numbers of events , to test the hypothesis that deep earthquakes occur on faults that initially formed in the oceanic plate prior to subduction. The directivity of hundreds of outer-rise, intermediate- and deep-focus earthquakes with MW=5.7 since 1994 will be systematically analyzed to distinguish their fault planes. Since waveforms are narrowed in the direction of rupture propagation and broadened in the opposite direction, the differences in rupture duration observed over the focal sphere can be compared with the patterns predicted for unilateral and/or bilateral ruptures and, thereby, the rupture direction and fault plane can be identified. Analyzing broadband seismograms, this work employs a recently developed procedure based on the differential duration between pairs of seismograms. This procedure, similar to measuring differential travel times, represents an optimal method for addressing this problem, since it permits the relatively-fast determination of the desired property (the fault plane) for hundreds of earthquakes without a priori assumptions about rupture complexity or the need to measure source duration. Preliminary results from the Tonga subduction zone show that we can identify the fault plane for ~1/3 of large, deep earthquakes, including events as small as MW 5.7, and that the resulting fault plane orientations show systematic patterns that will allow the testing of the fault-reactivation hypothesis. More generally, the expected identification of 150-200 fault planes will help constrain the physical basis for deep-earthquake rupture propagation: Is rupture propagation an isobaric process? How does temperature influence rupture characteristics? Is there a depth-dependent variation in the physical mechanism of deep earthquakes?
