Re: [Swarm-Modelling] general question on complexity

[Marcus G. Daniels]

Jacob Lee wrote:
> Is the complexity of a system strongly correlated to the computational 
> complexity of a model of that system?
Even if a model is functionally accurate in the sense of being validated 
(e.g. retrodictive out of sample) and predictive, there's the matter of 
whether the algorithms that implement it are close enough to what 
happens in the real world.   Compare a software CPU simulator that 
describes all of the internal details of its silicon sibling.  The 
software can predict what the hardware will do and exactly how fast it 
can do it, but the software may be orders of magnitude slower to give 
that answer.  Further, the real time scaling of the two won't 
necessarily the same if, say, the simulator processes execution units of 
the CPU one at a time and then consolidates the results (doing 
bookkeeping  arithmetic to work out the simulated real time that would 
have passed).  That kind of labor serialization would  lack realistic 
physical concurrency. 

It seems to me accurate ABMs that don't involve `representative agents', 
or aggregates e.g. models of low-level physical processes, models that 
are sufficiently detailed and concurrent (executed that way on as may 
CPUs as needed), and models that are executed dense in time would be 
strongly correlated.   By "dense in time" I mean in Swarm terminology 
schedules with nonsequential events on them. In Swarm if there are 
events at t = 0, t = 2, and t = 1000, the computational cost is 3, not 1000.

Overall, I'd say model and system computational time (as opposed to 
Kolmogorov description) complexity could be strongly correlated if the 
ABM was developed with that in mind and there was detailed information 
available on the mechanics of the process itself.