It is sometimes stated that in order to be amenable to a simple statistical
description. a system needs to have many coupled degrees of freedom. In th
is view, the statistical limit is to be understood from the dynamics. Here
we discuss a complementary point of view where the role of 'size' is to ens
ure that the probabilities of simple events, which are technically marginal
probabilities, settle down to a canonical distribution, conditioned by add
itive constants of the motion. The requirement that the conditioning is on
additive variables plays an essential technical role and it is not dear tha
t this condition can be relaxed. In the view discussed here, the effective
'size' is determined by the variance of the physical variables of interest.
Implications for Monte Carlo sampling are also discussed, with examples. (
C) 2001 Elsevier Science B.V. All rights reserved.