Pressure-induced solid-solid phase changes are responsible for most of the increase of density and seismic velocity with depth in the upper mantle. Lateral variations in temperature cases a similar effect; abrupt changes of density and seismic velocity due to phase changes are superposed on smaller changes associated with thermal expansion. Temperature-induced isobaric phase changes are as important in explaining various recent geophysical data as are the more familar pressure-induced phase changes.

In cold slabs the equilibrium mineral assemblage contains high-density, high-velocity phases which are not stable in hotter mantle. In particular, the ilmenite form of MgSiO3 and the &ggr;-spinel form of Mg2SiO4 have broad stability fields in cold mantle which increase the density and velocity of deep slabs to values in excess of those which have been used in geoid and seismic travel time modeling. Recent arguments for slab penetration into the lower mantle and whole mantle convection are based on thermal models of the slab which ignore the large density and seismic velocity anomalies associated with temperature-induced phase changes. When these effects are taken into account, the geoid and seismic anomalies associated with subducted slabs are consistent with slab confinement to the upper mantle and layered models of mantle convection. The seismicity cutoff and evidence for slab thicknening at 670 km also favor this style of convection. Mantle seismic velocities between 200 and 400 km depth in tectonic and young oceanic regions are lower than in shield regions, and this is due to the presence of a melt phase and lower-velocity, high-temperature phase assemblages. Deep, >200 km, long-lived continental roots, differing in chemistry from ''normal'' mantle, are not required when isobaric phase changes are taken into account. High-velocity subshield mantle is closer to normal subsolidus mantle than is suboceanic mantle which is affected by the presence of high-temperature phase assemblages. the whole mantle convection, thick continental root, and deep slab penetration hypotheses are not supported by seismic and geoid data when isobaric phase changes are included in the analysis. Phase changes are more effective in changing density and seismic velocity than are lateral variations in temperature and composition.

The lack of correlation of the geoid with ridges, shields, heat flow, and upper mantle velocity variations suggests a low geoid sensitivity to the upper mantle, consistent with layered convection. ¿ American Geophysical Union 1987