Formalizing models

[Roger M. Burkhart]

To continue the discussion on formalization to model complex systems,
I'd like to separate the more modest goal of formalizing the
*specification* of a complex system model from the analysis or
interpretation of any behavior produced as it runs.  A goal of Swarm
is to make the specification of a complex system model entirely explicit
and precise, so that it can be published, repeated in experiments, etc.
The formalization of a model needs to include not only its schema
(generic types and constraints) but also all the instance-level detail
that is produced during the history of a model run.  It does not
require any formalization of abstractions put forward as candidates to
explain or compress the system behavior.  Those abstractions are needed
too, but it's a completely different set of concerns.

It's important to have a good formalization of model content so that
our artificial models can really become the shared cultural artifacts
we seek them to be.  Right now in Swarm all the model specification and
history is buried in the O-O program structures we build in the Swarm
library frameworks.  I'm still eager to see these mapped to more
language-neutral structures that still capture the structure and
dynamics of a working simulation model.

There are strong reasons to develop better content formalizations anyway,
independent of any concerns for theory building.  Existing model
definition techniques are still very weak in expressing important aspects
of realistic complex processes, such as concurrent interaction, emergent
levels, and dynamically determined behavior.  The structures of typical
O-O analysis methods (including the new Unified Modeling Language that
has just been standardized by the O-O community) tend to define only
the most basic elements of a statically engineered information system.
Nevertheless, specifying discrete dynamic models to run in a computer
should require only computer science fundamentals, integrated across a
fairly broad space.  It's my hope that the practical requirements of
building Swarm models can generate spinoff benefits in more expressive
model definition techniques, and that these will also prove useful to
those who build systems to do specific work.

Roger Burkhart


                  ==================================
   Swarm-Modelling is for discussion of Simulation and Modelling techniques
   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
   please send a message to <address@hidden> with "help" in the
   body of the message.
                  ==================================