660-km discontinuity

The seismic discontinuity at a depth of approximately 660 km (the 660) as the boundary between upper and lower mantle plays an important role in the dynamic state of the Earth's interior. The 660 might be a barrier to whole mantle convection and the horizontal boundary between different convection cells in a layered convection model (i.e. upper and lower mantle).
Both, phase changes and chemical composition are able to explain the increase in density and seismic velocity of about 6% - 11% (Christensen, 1995).
A chemical boundary at a depth of 660 km implies that two chemically separate layers exist in the mantle without mixture of material of the two reservoirs in the upper and lower mantle. Two kinds of chemical changes have been suggested: a change in iron content (Anderson, 1967; Sawamoto, 1977; Stixrude et al., 1992) and a change in silica content (Anderson, 1970; Stixrude et al., 1992).
The phase change model suggests the transition from $\gamma$-Spinel ($\gamma$-sp) $\rightarrow$ Perovskite (pv) + Magnesiowüstite (mw) in the olivine component and the transformation of Garnet (gt) $\rightarrow$ Perovskite (pv). Both occur within the appropriate temperature and pressure interval. The pressure and temperature conditions of these two transitions are, however, slightly different. Consequently, different depths of these phase changes should exist. This will result in a complex structure of the discontinuity, which would cover a large depth interval. The $\gamma$-Spinel transition occurs over a narrow depth interval (< 4 km), but the change within the Garnet phase might have a broadening effect as it occurs over a wider depth interval (Ito and Takahashi, 1989).
However, this mineralogical observation is in disagreement with seismological observations, which show reflections of high frequency PKPPKP (P'P') waves at the 660 (Benz and Vidale, 1993), suggesting a transition width $\leq$ 4 km. The seismic observations would exclude the gt $\rightarrow$ pv transition. However, the P'$^{660}$P' observations are scattered, possibly implying that the properties of the discontinuity, like reflectivity and thickness, vary strongly. Also, long-period P-SV conversion studies suggest that the 660 is a velocity and density gradient of 20 - 30 km thickness (Petersen et al., 1993), in contradiction to the P'P' studies.
In the long-period stacks of Flanagan and Shearer (1999), the P$^{660}$P underside reflection is absent, though the S$^{660}$S reflection is strong. This leads to Earth models with a smaller P-velocity jump in comparison to standard Earth models like IASP91 due to a reduced change in bulk modulus without a change in the density or S-velocity structure (Estabrook and Kind, 1996). A grid search with a variation of the three parameters P-velocity jump, S-velocity change and density contrast by Shearer and Flanagan (1999) to fit their PP and SS long-period stacking data, results in a model with a strongly decreased density jump compared to PREM. Their best fitting model for the 660 has $\Delta\alpha$ = 2.0%, $\Delta\beta$ = 4.8%, and $\Delta\rho$ = 5.2% (PREM: $\Delta\alpha$ = 4.6%, $\Delta\beta$ = 6.5%, $\Delta\rho$ = 9.3%). As result of a small density contrast across the 660 ($\sim$5%, Shearer and Flanagan (1999)) layered convection would be less likely than whole mantle convection.
The Clapeyron slope of the $\gamma$-Spinel $\rightarrow$ pv + mw transition is negative, indicating an endothermic phase change. In agreement with seismic observations this transformation predicts a depression of the 660 in regions of subduction, where the cold slabs interact with the hotter surrounding mantle (Revenaugh and Jordan, 1991a). Seismological studies show a variation of apparent depth of the 660 of 30 - 40 km (Revenaugh and Jordan, 1991a; Shearer, 1990).
Theoretical studies of the seismic velocity change due to phase transitions probably indicate that hybrid models of a discontinuous change of chemical composition and a phase change are most likely able to explain the reflections found by seismological studies (Lees et al., 1983).