C.P. Conrad and C. Lithgow-Bertelloni, "How mantle slabs drive plate tectonics," Science, 298, 207-209, 2002. [abstract] [online version] [reprint] [supporting material] [U. Michigan Press Release] [Geotimes Article]
Subduction zones (where an one plate dives beneath another into the mantle) produce "slabs" of subducted lithosphere in the mantle. These slabs are cold and dense, and their descent within the mantle is thought to provide the primary energy source that drives viscous flow in the mantle and, ultimately, the motions of Earth's surface plates. However, exactly how mantle slabs drive plate motions has been the subject of debate.
To constrain how mantle slabs drive plate motions, we have developed models of viscous flow in the mantle driven by the descent of slabs. By balancing the stresses that this flow exerts on the base of the plates, we can obtain a prediction of plate motions that we can compare to observed plate motions. In this way, we can evaluate a given model's ability of predict observed plate motions. We define two mechanisms by which mantle slabs could drive plate motions: "slab suction" and "slab pull".
Slab Suction. In the slab suction mechanism, we assume that slabs are detached from their surface plates and sink within the mantle. This downward motions induces a symmetrical flow pattern that tends to push both the subducting and overriding plates toward the subduction zone at approximately equal rates (Fig. 2). This pattern is evident in the pattern of plate motions that is predicted by this mechanism (Fig. 3). Here overriding plates move in the correct direction, but with speeds comparable to those of the subducting plates (note the primarily yellow and green colors in Fig. 3). By contrast, subducting plates on today's Earth typically move about 4 times faster than non-subducting plates (note the red and pink colors for the Pacific and Indian basins in Fig. 1).
Slab Pull. If, instead, we assume that the slab is mechanically attached to the subducting plate, then the weight of the slab will pull on subducting plate, drawing it toward the subduction zones. In addition, the trenchward motion of the subducting plate will induce a flow in the mantle that will tend to drive the overriding plate away from the subduction zones (Fig. 4). The pattern of plate motions predicted by this mechanism (Fig. 5) shows the correct ratio of subducting to non-subducting plate speeds, but overriding plates tend to move away from subduction zones instead of toward them, as is observed (Fig. 1).
Combined Slab Pull and Slab Suction. By combining "slab pull" from upper mantle slabs and "slab suction" from lower mantle slabs, we are able to obtain a model for how slabs drive plate motions (Fig. 6) that gives a good prediction of plate motions (Fig. 7). In this model, upper mantle slab (above 670 km) are physically attached to their subducting plates. Their weight pulls subducting plates toward subduction zones. Slabs in the lower mantle (below 670 km) are detached from the slabs above them and are instead supported by viscous stresses associated with flow in the surrounding mantle. This flow excites the slab suction plate-driving mechanism, which contributes to plate motions at the surface. There are several possible reasons that slabs below 670 km do not contribute to the pull force. First, slabs may not be strong enough to support the weight of slabs below 670 km. Second, higher viscosity in the lower mantle will provide support for slabs there. We estimate that upper mantle slab pull and lower mantle slab suction account for 60% and 40% of the forces on plates, respectively.