It has often been proposed that the 660 km discontinuity may represent a change in chemical composition, and although its interpretation as a phase boundary is now established, recent work [ Jeanloz, 1991; Ita and Stixrude, 1992; Stixrude et al., 1992], has indicated that a change in bulk chemistry may be necessary in addition to a phase transition, in order to simultaneously satisfy the observed density and seismic velocity jumps. An intrinsic chemical change may also arise from consideration of thermodynamic equilibria [ Bina and Kumazawa, 1993].
Early modeling [ Christensen and Yuen, 1984] assumed a chemical boundary that was initially coincident with the phase transition, and showed how an intrinsic chemical density difference added linearly to the phase change buoyancy effect in determining the transition between layered and whole mantle convection. A general assumption in the mantle modeling community has been that the Earth would inevitably become increasingly mixed over geological time, and research has tended to focus on how slowly this mixing could take place, either from an initially layered state [ Gurnis and Davies, 1986a; Todesco and Spera, 1992] or with continuously introduced subducted oceanic crust [e.g., Gurnis and Davies, 1986b; Kellogg and Turcotte, 1990]. However, Weinstein [1992] showed that the opposite may occur, namely that chemical layering could be dynamically induced through a 'filter effect' of the endothermic phase transition, in which heterogeneities of different chemical compositions pass through the phase transition with different levels of difficulty, resulting in an intrinsic chemical difference between the upper and lower mantles where none was present initially. This type of dynamic ``unmixing'' process may provide the key to reconciling geochemical and geodynamic data and results, and is an important topic for future research.
Another component in reconciling geochemical evidences is that during periods of ultrafast subduction, which are likely to accompany avalanche events, recycling of volatiles into the mantle is much more efficient, producing localized mantle chemical anomalies [ Staudigel and King, 1992].