Geochemical models have frequently divided the mantle into depleted upper a
nd undepleted lower mantle reservoirs, usually taken as indication for a la
yered style of convection. This is difficult to reconcile with seismologica
l and geodynamical evidence for substantial mass flux between lower and upp
er mantle. Various models have been proposed to jointly interpret the evide
nce, including that of G.E Davies [J. Geophys. Res. 89 (1984) 6017-6040] in
which the author suggested that lumps of primitive material may exist in t
he lower mantle, representing reservoirs for undepleted basalts. Mixing cal
culations have suggested, however, that such blobs could not survive 4 Ga o
f convection. Calculations by M. Manga [Geophys. Res. Lett. 23 (1996) 403-4
06] on the other hand showed that high-viscosity blobs could persist in con
vective cells for geologically long times without being substantially defor
med and mixed with the surrounding flow. We investigate a blob model of con
vection based on these ideas and consider dynamical, thermal, geochemical a
nd rheological consequences. The radiogenic heat production in the primitiv
e blobs would lead to higher temperatures. However, these would be modest (
Delta T < 300 K) for sufficiently small blobs (radius < 800 km). The result
ing thermal buoyancy can be offset by a small intrinsic density excess (< 1
%) so that blob material is hidden from the ridges but sampled by rising pl
umes. To satisfy geochemical constraints, blobs would have to fill 30% to 6
5% of the mantle (less if they are taken to be enriched rather than primiti
ve). Thermal considerations require that they be surrounded by depleted mat
erial of lower viscosity that would convectively transport heat to the surf
ace. The thermal decrease in blob viscosity would be about one order of mag
nitude but constrained to the interior; the stiffer 'shell' can then be exp
ected to control the dynamical mixing behavior. On average, the viscosity o
f the lower mantle would be increased by the presence of the blobs; if they
were 100 times more viscous than the surrounding mantle the net effect wou
ld be to increase the effective viscosity approximately 5-fold. The origin
of the proposed blobs is an unresolved problem. We suggest that perovskite/
magnesiowustite ratio variations could be the reason, which would yield an
intrinsic density contrast as well. Blob geometries are at the current reso
lution limit of global tomographic models, and the trade-off between temper
ature and compositional effect on seismic wave speeds tends to blur the sig
nal. However, joint P- and S-wave inversions and scattering studies may ult
imately approach the necessary precision to detect blobs. Under the simplif
ying assumptions employed in this paper, we find that the viscous blob mode
l is internally self-consistent and feasible. The model may explain the out
standing problem of incongruous geochemical and geophysical data. (C) 1999
Elsevier Science B.V. All rights reserved.