Dear Nigel,
There is one domain, that I am aware of, where large-scale models will
definitely yield different results: ecology, or population dynamics, in
general. That is because of genetic drift (and to some extent, density
dependent effects). Populations tend to oscillate a lot and in small
systems, oscillations lead to extinction much easier. In one branch of my
recent work I am, with co-authors, blending ideas from ecological models
with ideas from innovation and economics research (e.g., the dynamics of
niche markets). Experiences there lead me to think that the effect of small
population size may be an important factor in these areas as well.
Another domain where system size may matter is information diffusion
(a.k.a., percolation, contagion, etc.) models, or more generally, dynamics
on (social) networks. That is because some interaction topologies have the
small world property, while others don't. (In some topologies the average
peer-to-peer distance scales with log N, where N is the number of agents,
while in others it may scale with sqrt(N), or with N.) For small systems,
the difference between a small world and a "large" one may be negligible,
while in larger systems the "opening gap" between these may lead to
important differences in the aggregate system behavior.
My 2 cents. I hope this helps!
Best regards,
-- Laszlo
--
dr. Gulyás László | Laszlo Gulyas, PhD
kutatási igazgató | dir. of research
AITIA International Zrt. | AITIA International Inc.
----- Original Message -----
From: "Nigel Gilbert" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Sunday, August 10, 2008 11:17 PM
Subject: [SIMSOC] Whendo big models give different results from small ones?
In what circumstances might it be necessary to implement agent-based models
of thousands or millions of agents, rather than the more typical tens or
hundreds? Do you know of any examples where scaling up the model (in terms
of the number of agents) gives qualitatively different results? Do you know
of any literature that addresses this issue?
Išd be very interested to hear of any suggestions.
Thank you,
Nigel
_______________________________________________________________________
Professor Nigel Gilbert, Editor, Journal of Artificial Societies and
Social Simulation, <http://www.soc.surrey.ac.uk/JASSS/>
Centre for Research on Social Simulation (CRESS)
Department of Sociology, University of Surrey, Guildford, UK.
Tel:+44 1483 689173 [log in to unmask]
<http://cress.soc.surrey.ac.uk/>
|