A laboratory analogue of a three-layer linear viscous slab-upper mantle-lower mantlesystem is established in a silicone putty, honey and crystallized honey tank experiment.The same setup as in the numerical investigation (part 1) is used. We focus on theinteraction of the slab with the induced passive mantle flow by widely varying the mantlevolume flux boundary conditions. In our numerical experiments the lateral volume fluxwas set to zero. In interpreting the results relative to the real Earth, the base of the box istaken as the bottom of the mantle convection system, while the lateral boundaries may beassociated with the presence of other nearby slabs. Dynamic force equilibrium, assessed onthe basis of an analytical review of forces, is described for four different phases: (1) thesubduction initiation instability, (2) the accelerating dynamic free fall phase of the slab, (3)the dynamic interaction with the 660-km discontinuity, and (4) a final phase of steady statetrench retreat. Phase 3 is an important feature not observed in the numerical experiments.This highly dynamic phase of interrupted trench retreat can therefore be attributed toboundary conditions on mantle volume flux. On the basis of integration constants of forceequilibrium in phases 2 and 4 we identify two different classes of volume flux: one inwhich the lateral boundary can be considered open and the other class where it is‘‘closed.’’ Closed boundary condition cases are obtained if any of the lateral boxboundaries are 600 km away from the slab. Assuming a one-to-one relation betweentrench retreat and back arc spreading, enigmatic observations of episodic opening of backarc basins can be explained by our experimental observations.