To interpret the generation of back arc basins as an effect of the slab pull mechanism (Fig. 2b) it is necessary to explain how the roll back of the slab may induce a tensional deviatoric stress in the overriding plate able to cause extensional deformation. A number of authors (e.g., Forsyth and Uyeda, 1975; Chapple and Tullis, 1977) invoked the existence of a trench suction force, induced by slab pull, which would cause the trenchward motion of the arc and its consequent divergence from the overriding plate. However, the results of numerical (Pacanovsky et al., 1999) and analogue (Shemenda,1993) modelling indicate that the tensional stress induced by this force in the overriding plate may range between 5 and 40 MPa, i.e. values significantly lower than the average tensional strength of the intact lithosphere, which is estimated in the range 200-260 MPa for oceanic plates (Scholz and Campos, 1995; Mueller et al., 1996) and 100-200 MPa for continents (Dunbar and Sawyer, 1989; Fadaie and Ranalli, 1990; Okaya et al., 1996). To find a tentative solution of this problem, Shemenda (1993) suggested that back arc basins could correspond to zones previously weakened by uprising of magma from mantle sources. However, this hypothesis contrasts with the fact that in most subduction zones the magmatic arc and the back arc basin are clearly separated (e.g. Sibuet et al., 1987) and that some subduction zones, as the Sunda and Chilean ones, are characterized by well developed volcanic activity, but do not present any evidence of back arc opening.

The feasibility of the slab pull mechanism is also conditioned by the length of the slab. For instance, numerical experiments (e.g. Hassani et al., 1997) have shown that the slab pull force is unable to produce extension in the overriding plate, even weakened, until the slab has reached a length of roughly 300 Km. Thus, the possibility that roll back and back arc opening are driven by a dense slab in the initial stage of its development can reasonably be excluded.

Another factor which controls the occurrence of slab roll back and of the consequent back arc extension is the mechanical coupling between the subducting and overriding plates. Numerical and analogue experiments (e.g. Shemenda, 1993, Hassani et al., 1997; Chemenda et al, 2000) point out that tensional stresses disappear from the overriding plate when the friction coefficient at the subduction fault is larger than 0.2. Considering that both observational evidence and theoretical considerations suggest that realistic values of this coefficient may range in the interval 0.4 - 0.6, due to the nearly hydrostatic fluid pressure regime in the crust (Townend and Zoback, 2000), it seems that the conditions for the occurrence of extension in the back arc zone are not generally fulfilled.

On the other hand, if extension cannot occur in the overriding plate, the roll back of the slab is inhibited as well, since the separation of the subducting lithosphere from the overriding plate is not allowed by the lithostatic pressure (e.g. Shemenda, 1993). Thus, the steepening of the slab would be the only feasible effect of the slab pull force, as the Shemenda's experiments clearly show.

Other arguments useful to evaluate the possibility that slab pull results in a simple slab steepening, instead of trench retreat and slab roll back, could be derived by comparing the energy expenses implied by these two kinds of deformation. The first pattern (Fig. 2b) would involve a number of tectonic processes, as trenchward displacement of the arc, thrusting activity at the accretionary belt, lithospheric thinning in the back arc zone, flexure of the subducting lithosphere, friction along the subduction fault, penetration of the slab into the asthenosphere and lateral displacement of asthenospheric material on both sides of the roll backing slab. Slab's steepening, instead, provides a fixed position of the trench with respect to the overriding plate and does not require any deformation of the overriding lithosphere. With respect to the first solution, this pattern involves no energy dissipation for back arc opening and for trenchward arc displacement and presumably involves less energy expense for the lateral displacement of asthenosphere at both sides of the slab. All the other sources of energy expense listed above seem to be more or less comparable in the two cases. Thus, the minimum work principle would suggest that a simple slab steepening would be the most probable effect of the slab's negative buoyancy, unless very peculiar structural-rheological conditions considerably reduce the energy budget of the tectonic processes implied by the slab pull model.

The evidence that most slabs are not vertical and that no simple relation exists between slab dip and age of subducted lithosphere (see e.g., Taylor and Karner, 1983) could imply that other forces act on the slab in addition to its negative buoyancy. These forces could be, e.g., lithosphere resistance to bending (Conrad and Hager, 1999), suction between the overriding and underthrusting lithosphere along their frictional interface (Jischke, 1975), uplift of the subducted slab due to the flow induced in the asthenosphere (Tovish et al., 1978; Turcotte and Schubert, 1982) and viscous resistance to slab migration through the mantle (Davies, 1980; Scholz and Campos, 1995). Which the resultant effect of all these (and possibly others) forces could be, is not clear yet.

Most arcs associated with back arc extension are characterized by a significant curvature (e.g. Uyeda and Kanamori, 1979; Uyeda, 1986). In the Mediterranean area, for instance, all major examples of these systems (Balearic, Tyrrhenian, Aegean and Pannonian) involved a migrating arc which started from a more or less straight initial configuration to reach a final shape where the trends of its various segments show remarkable differences, even larger than 90° (see, e.g., Dercourt et al., 1986; Royden, 1993b; Mantovani et al., 2000a). A similar behavior has been observed in several circum-Pacific T-A-BA systems (e.g. Uyeda, 1986; Lee and Lawver, 1995; Hall, 2001). In our opinion, to explain such phenomenon as an effect of slab-pull forces one must face a number of problems. For instance, one should understand why slab sinking rates are higher in the central sectors of the arc with respect to the lateral ones. This behavior could be due, e.g., to different densities in the various parts of the subducting lithosphere. However, it seems rather unlikely that such peculiar density distribution was systematically present in all T-A-BA systems of the world. Some authors argued that arc buckling might be due to the earth's sphericity (e.g. Bayly, 1982; Yamaoka, 1988). However, this hypothesis cannot explain why such effect occurred in some subduction zones and not in others (e.g. Uyeda, 1986) and, also, it can hardly account for the extremely variable curvature of arcs at consuming boundaries.

Discussions as to the plausibility of the slab pull model are also reported in other works (e.g., Taylor and Karner, 1983; Uyeda, 1986; Mantovani et al., 1997, 2000a; Flower et al., 2001).