The variable structure scenario is of wide applicability. One might argue that many non-linear chaotic dynamical systems could be cast in the form of a VSS. While it is possible to construct a regularly behaved VSS, this VSS scenario does give us a way of approaching the engineering and construction of arrangements for producing irregular and bursty behaviour. An important and increasingly studied VSS is the Cellular Neural Network [13] described by Chua and Yang in 1988. This has important applications to signal processing and artificial vision, among others. Chua has suggested that CNNs are a test bench for the evolution of complexity, which he suggests emerges "on the edge of chaos" when the CNN is set up correctly. The non-linear elements are linked together by coupling and switching elements under software control. Various training algorithms are suggested for particular tasks. VLSI CNN chips are available and would be good candidates for studying the evolutionary behaviour of VSSs.
As we have seen, iterating electronic circuits, non-linear dynamical systems, and recursive network algorithms provide three different kinds of implementation of a VSS. In the case of the two-centre system with its integrators, the delay and memory through the integrator circuits provides the equivalent of the data propagation through the sample-and-hold circuits of the iteration system. In the case of the network example, the previous structure of the cells is stored in computer memory and dictates what happens on the subsequent token pass. Thus, to these authors at least, all these examples are transparently equivalent.
There is a good discussion of the stability of
chaotic processes, and of the behaviour
due to interior crises of the attractor
in dynamical systems, in a paper by Kautz [14].
He says ``when the unstable periodic orbit
defines an interior basin, increasing 
beyond
simply allows the attractor to expand
into other parts of the basin of attraction."
At a critical value of the
bifurcation parameter , the chaotic attractor collides with
an unstable periodic orbit. Now it is a simple
matter to arrange, electronically, for the parameter
to depend on a measure of the average size of the
chaotic trajectory which is visiting
a part of the attractor, and thus to engineer evolution.
There is also a sense in which simple electronic circuits, and their simulations, which rely on perfect diodes for their non-linear properties, may be modelled by VSS. A good example may be found in the two coupled nonlinear LC circuits of Wada [15] et al, in which they observe what they term blowout bifurcation and bursty behaviour similar to our observations, in their Figure 9.
Other examples may come to mind. For example, it is possible that the unpredictable and irregular behaviour of certain classes of computer software may be interpreted as being due to the variable structure in the processes running on deterministic hardware; such processes may interact and retain memory of their most recent invocation in a complex and unpredictable way, and be modellable in terms of a VSS in the software domain. The structure of the software (the bit-image in machine memory at startup) depends on the dynamical history such that even if the user tries to run the software on two successive occasions in exactly the same way, the bit image has changed, and on subsequent runs the behaviour is different.