The recent increasing number of experimental works leads us to review the elastic properties and mineralogical transformations of mantle minerals. The updated data set is used to compute seismic profiles for two petrological models along three adiabatic temperature profiles. These profiles are chosen to stress out the influence of non-olivine minerals on seismic parameters, and to represent cold and horizontally averaged temperature profiles of the Earth's mantle. In a first part, starting compositions of pyrolite and piclogite and a single layer convection are assumed. The results clearly point out the importance of the non-olivine part of the mineralogy. Two scenarios are found to explain the 660 km depth discontinuity, whatever the starting composition. (1) Ilmenite appears at the expense of garnet at 660 km depth, and then transforms into perovskite and a small amount of garnet at pressures relevant to the lower mantle (case of a 1500 K adiabat with current phase diagrams). The cumulative effects of the breakdown of gamma-spinel and of the reactions involving ilmenite lead to strong seismic discontinuities at 660 km depth followed by relatively small seismic gradients at the top of the lower mantle. (2) Ilmenite is not stable around 660 km depth, and the breakdown of gamma-spinel is the only sharp reaction to occur (case of a 1600 K adiabat). Smaller seismic discontinuities are found at 660 km depth, and higher seismic gradients are obtained at the top of the lower mantle. Taking into account experimental uncertainties, the comparison of our calculations with reference seismic models strongly suggests that ilmenite is present at the upper-lower mantle boundary. Along the cold temperature profile (1000 K adiabat), the reactions involving ilmenite appear at separate depths, leading to a complex upper-lower mantle transition with three separate discontinuities. A case with a stratified convection (thermal boundary layer and different compositions between the upper and the lower mantle) is also studied. The strength of the discontinuity induced by the chemical boundary is in a good agreement with seismic observations when a horizontally averaged temperature profile is used for the mantle. This result implies that it is not possible to discriminate between layered or single cell convection in the Earth's mantle. Along the cold temperature profile, the appearance of ilmenite in the upper mantle leads to a second discontinuity in addition to the one imposed by the chemical boundary. Hence, subduction zones should be characterised by a multiple-step transition from the upper to the lower mantle, whatever the chosen style of convection. The proposed explanation of the 660 km discontinuity is comforted by recent seismic observations: a complex behaviour of the 660 km discontinuity has indeed been found in subducting slabs, and broadband studies of converted waves have suggested a multiple discontinuities pattern to explain this complexity. (C) 1998 Elsevier Science B.V. All rights reserved.
