ABSTRACT: Modeling pattern data series with cellular automata fails for a wide range of deterministic nonlinear spatial processes. If the latter have finite spatially-local memory, reconstructed cellular automata with infinite radius may be required. In some cases, even this is not adequate: an irreducible stochasticity remains on the shortest time scales. The underlying problem is illustrated and quantitatively analyzed using an alternative model class called cellular transducers.
University of Illinois, Beckman Institute, Center for Complex Systems Research, Technical Report UIUC-BI-CCSR-92-08.