Asymmetrical retreat of a subducting slab requires the least work to deform the slab, simply because it does not involve extensive internal slab deformation. This would be the case in other types of rollback, such as radial rollback. To allow for asymmetrical slab retreat, the only permanent large-scale deformation the subducting plate has to undergo is the formation of a lithospheric-scale vertical tear on one side of the slab segment (Figure 2). Examples of such tears can be observed in several arcs such as the Tonga Arc and the Ryukyu Arc. The formation of a tear could be facilitated by the presence of a pre-existing weak zone in the subducting lithosphere such as a transform fault, striking at high angle to the arc. The presence of a vertical tear at one cusp of the arc would further facilitate asymmetrical rollback, since it would allow lateral asthenosphere flow around the slab in addition to flow underneath the slab tip (Figure 2c). A continuity of the slab across the other cusp would retard rollback of the slab in the proximity of this cusp, since the asthenosphere would only be allowed to flow underneath the slab tip. This process would then amplify the entire process of asymmetrical rollback and back-arc opening. Asymmetrical rollback could also be facilitated by a density polarity in the subducting lithosphere, parallel to the trench axis. This density polarity could result from a variation in age of the subducting slab parallel to the trench or the presence of buoyant features on the subducting plate. With such a slab configuration, the denser part would have a tendency to roll back faster compared to the lighter part, thus promoting asymmetrical rollback.

In natural arc systems, slab retreat does not result in the separation between the overriding and subducting plates along the trench. Instead, the overriding plate fails at some distance from the trench. In ocean-ocean subduction settings this often occurs along the volcanic arc, resulting in the formation of a remnant and active volcanic arc, separated by a back-arc basin [Karig 1970, 1971a,b; Kobayashi and Nakada 1979]. Examples include the Tonga-Kermadec Ridge, which was split from the Lau-Colville Ridge during formation of the Lau-Havre Basin. Another example is the Izu-Bonin Arc in the Western Pacific, which was split from the Palau-Kyushu Ridge during formation of the Parece-Vela Basin. More recently, the active Mariana Ridge was split from the inactive West Mariana Ridge during formation of the Mariana Trough. In ocean-continent subduction settings, back-arc deformation is more diffuse. However, in a more advanced stage of deformation, retreat of the hinge-line can result in the segmentation of the overriding plate, where a segment is being separated from the continent. One example is the Kuril Arc, which separated from the Okhotsk microplate during opening of the Kuril Basin [Maeda 1990]. Another example is the Ryukyu Arc, which is presently being separated from East China, accompanied by extension in the Okinawa Trough [Uyeda 1977]. Other examples include the Japan Arc [Uyeda and Miashiro 1974] and the South Shetland Arc [Lawver 1995].

A number of arguments can be given for the reason why the overriding plate separates at the arc and not at the trench. First of all, the strength of the overriding plate is at its smallest at the arc due to thermal weakening and its relatively thick crust (Figure 6). Also, the arc has the highest potential energy, further promoting separation along the arc. This excess potential energy results in a net push from the arc towards the overriding plate, which will therefore not allow separation between the overriding and subducting plate during rollback. Injection of asthenosphere material along the subduction boundary does not occur because of this push from the arc region to the subducting plate, the lithostatic pressure exerted by the overlying mass on the boundary and the net downward drag of material along the subduction boundary due to subduction of the slab.

In most experiments, no separation between the overriding plate and the retreating door occurred as well, even though the overriding plate was homogeneous throughout. This could be explained by the fact that in most experiments the buoyancy force was greater than the brittle strength of the overriding lithosphere, resulting in spreading of the overriding lithosphere over the glucose syrup in the space it was allowed to by the retreating door (Figure 7). In an advanced stage of deformation, a ridge segment (i.e. arc) developed in between the retreating door and the principal zone of extension (PZE = a zone which developed in an advanced stage of deformation and subsequently absorbed most or all of the extension during continuing deformation). The location of the PZE and the ridge (arc) could be related to boudinage of the lithosphere due to retreat of the door, where the silicone layer did not favour necking at the contact zone with the door, because this would result in high vertical shear stresses along the contact zone. Thus, it can be concluded that from pure mechanical reasons and even in a homogeneous overriding plate, rollback promotes extension to be localised at some distance from the subduction interface.

An interesting question that arises from experiment 5 (and also experiment 5 and 9) is why the overriding plate did not immediately separate from the retreating boundary, but only did so after some considerable deformation of the overriding plate (Figure 4a,b). One would not expect any deformation in the overriding lithosphere due to spreading, since the buoyancy force (between a column of glass beads lithosphere and a column of glucose syrup asthenosphere) in this experiment is a factor of ~ 8-9 smaller (see Table 2) than the integrated strength of the overriding lithosphere. Evidently, the retreating boundary can transmit deviatoric tensional stresses to the overriding lithosphere, without there being any significant adhesion between the glass beads and the retreating door. An explanation for this behaviour, based on the results of experiment 5, is sketched in Figure 8. From an initial stage (Figure 8a), retreat of the door over a short distance dx leads to the formation of a lithospheric scale gap (Figure 8b), which will not immediately be filled with glucose syrup. This will result in a large potential energy difference between the overriding lithosphere and the gap, leading to the failure and collapse of the overriding lithosphere into the gap (Figure 8c). Extension continues until at some stage the overriding lithosphere separates from the door (Figure 8d).