“…Several large deep‐focus earthquakes in 1994 and 1996 have come as a test of models for the occurrence of deep earthquakes. These models include transformational faulting of olivine to spinel [ Green and Burnley , 1989; Kirby et al , 1992; Green and Houston , 1995; Green and Zhou , 1996]; clinoenstatite to ilmenite [ Akaogi et al , 1987; Hogrefe et al , 1994; Kirby et al , 1996]; serpentinite dehydration [ Meade and Jeanloz , 1991]; plastic instability and shear induced melting [ Griggs and Baker , 1969; McKenzie and Brune , 1972; Ogawa , 1987; Hobbs and Ord , 1988; Lomnitz‐Adler , 1990; Spray , 1993; Kikuchi and Kanamori , 1994; Kanamori et al , 1998; Bina , 1998a]; reactivation of preexisting fault planes [ Silver et al , 1995]; grain‐size reduction causing rheology changes [ Riedel and Karato , 1996, 1997; Karato et al , 2001]; and slab stresses due to an increase in mantle viscosity at 660 km [ Vassiliou et al , 1984; Goto et al , 1987; Vassiliou and Hager , 1988]. Observations show that the largest recent deep event, the 1994 Bolivia earthquake, occurred at the bottom of the subduction zone, that several large earthquakes (1994 Fiji and 1996 Flores Sea, Indonesia [e.g., Tibi et al , 1999]) may have ruptured outside the inner core of the slab as defined by previously located seismicity, and that deep seismicity occurs in areas which may be too warm for olivine to exist metastability.…”