Our methods and tools for modelling, optimisation, and control depend
heavily on exploiting problems structure. Understanding the relationship and constraints
underlying the problem structure enables predicting system behaviour as well
as potentially controlling behaviour. Decomposing problem structure, associating first principles
with the elements resulting from this decomposition, then recomposing these
principles into an overall mathematical or computational model are typical
steps of systems modelling. Frequently, this does not work. Understanding
why and developing real insights into complex problems involves what
Sage and Rouse (1999) describe as the modelling challenge.
Meeting the modelling challenge is complicated by the fact
that not all critical phenomena can be fully understood, or
even anticipated, based on analysis of the decomposed elements of
the overall system. Complexity not only arises from there being
many elements of the system, but also from the possibility
of collective behaviours that even the participants in the system
could not have been anticipated (Casti, 1997).