Perhaps this can be modeled mathematically (non-rigorously)? For some problem space C with dimensionality d, a mechanical or biological system can be described by the tuple s = (x0, x1, ..., xd) which describes a starting, stable configuration of the system, with some room for variance s + y = (x0 + y0, x1 + y1, ... xd + yd). The stable conditions for the system might be described as extrema on a hypersurface or hypervolume of C. Then for some chaotic function f, f(s) -> s', where s' is another point on on hypersurface describing the system, if f is chosen properly, it will result in the system evolving to another saddle point on the hypersurface describing that biological or mechanical system.
The question then is it possible to model the hypersurface with some anayltical equation, and what's the iterative, Chaotic function that will optimize f(s) finding another local saddle point on the hypersurface.