Results from recent tomographic imaging of the mantle have revealed plume-l
ike structures under some hotspots and renewed the interest in the theoreti
cally possible forms of ascending jets in mantle convection. It is now a cl
assical view that plumes reaching the lithosphere from below can, in princi
ple, develop from boundary layers either at 660 km or at 2900 km depth. If
both types are present in the mantle, the 660 km boundary layer, possibly d
ue to the endothermic spinel-perovskite phase transition, must be partially
penetrable. The present study shows that, with a partially penetrable phas
e boundary at 660 km depth, a further kind of plumes can develop, namely fr
om below the 660 km boundary layer. These 'mid-mantle plumes' have no root
in the deep lower mantle. If, as recent viscosity inversions suggest, a sec
ond low viscosity zone exists under the 660 km discontinuity, then this 'se
cond asthenosphere' represents a well-focused source volume for the mid-man
tle plumes. These upwellings are the counterparts of avalanche-like downwel
lings crossing the phase boundary in an intermittent manner. The condition
for the development of mid-mantle plumes is that the phase boundary acts as
a strong, but not fully impenetrable barrier to vertical flow. In two- and
three-dimensional numerical simulations using a compressible fluid in a Ca
rtesian box, it has been found that the critical parameters of mantle conve
ction (Rayleigh number, phase transition characteristics) closely meet this
condition. Mid-mantle plumes develop with an eruptive vigor, much faster t
han the boundary layer plumes and can produce huge plume heads, exceeding 1
000 km in radius. They can thus explain very extensive, episodic flood basa
lt volcanism on the surface. If mid-mantle plumes really exist, they can co
ntribute to the explanation of the diversity of hotspot basalt isotopic sig
natures since they sample a geochemical reservoir distinct from the classic
al plume sources. (C) 2000 Elsevier Science B.V. AU rights reserved.