The most fundamental question for evaluating the thermal regime of the lower mantle is the degree to which the upper and lower mantle are chemically homogeneous. If the mantle is compositionally layered, a thermal boundary layer must exist at or near the 660-km discontinuity that will exert a profound influence on temperatures in the lower mantle. Due to the overriding importance of this question, a large number of studies have been devoted to this issue, of which only a small sampling can be incorporated into this review.
On the basis of thermoelasticity data for magnesiowüstite and silicate perovskite, Stixrude et al. [1992] and Hemley et al. [1992] concluded that an upper mantle composition does not completely match seismic data for the lower mantle, and specficially, there could be an enrichment of silicon in the lower mantle (see also Zhao and Anderson [1994]). The tradeoff between temperature and iron content has also been explored by these authors who found that iron enrichment of the lower mantle is required for the higher temperatures that would be expected in the deep interior of a chemically stratified planet. On the other hand, Wang et al. [1994] concluded from an analysis of lower pressure equation of state data on silicate perovskite that a uniform mantle composition is consistent with seismic data. A thermodynamic analysis of data on the thermal expansivity of silicate perovskite has been carried out by Anderson and Masuda [1994]. The degree to which compositional differences can develop as a result of the coupling of a chemical change and a phase transition in a multivariate system has been examined by Bina and Kumazawa [1993].
This question of compositional stratification has also been approached using numerical simulations of mantle flow. Glatzmaier and Schubert [1993] used a three-dimensional, spherical convection model to investigate how the style of mantle convection is influenced by layering. They concluded that whole mantle convection models better reproduce geophysical observations. Morgan and Shearer [1993] combined seismic velocity heterogeneities, discontinuity topography, and radial viscosity models to compute radial mantle flow distributions that showed no reduction across the 660-km discontinuity, consistent with whole mantle convection.
Numerical simulations of convection have raised the possibility of a new solution to the above controversy: namely, that the mantle is partially stratified, with the degree of stratification varying in space and time. These results stem from the development of numerical mantle models that include the effects of phase boundaries [ Machetel and Weber, 1991]. Two-dimensional and axisymmetric calculations showed that the presence of the endothermic 660-km boundary promotes layered or intermittently layered convection [ Zhao et al., 1992; Weinstein, 1992; Peltier and Solheim, 1992; Solheim and Peltier, 1994]. The sensitivity of these results to variation of a range of parameters was investigated by Ita and King [1994]. Three-dimensional calculations also show complex, time-dependent behavior, including avalanching of cold material across the boundary layer [ Honda et al., 1993; Tackley et al., 1994]. When the Clapeyron (P-T) slope of the phase transition is included, downwellings are found to be impeded at the boundary while rising plumes are unaffected [ Liu, 1994].
Recent comparisons between convective flow calculations and seismic
tomography are suggestive of a convective style that is dominantly
whole mantle but allows for localized layering [ Jordan et al.,
1993; Woodward et al., 1994]. Weinstein [1993] has
argued that episodic flow across the 660-km discontinuity could
produce temperatures in the transition region that are 
250 K
cooler than the surrounding mantle due to the temporary stagnation
of cold subducting material in this region. Additional features that
need to be incorporated into numerical models before more definitive
conclusions can be drawn include temperature-dependent viscosity,
rigid plates, and three-dimensional calculations at higher (more
Earth-like) Rayleigh numbers.