Re: Random Seed question

[glen e. p. ropella]

At 05:23 PM 1/31/00 -0500, you wrote:
> The only reasons I can think of why you might *need* to use a more
> sophisticated method to choose seeds are if you're doing cryptography, and
> (perhaps) if it's required to prove the validity of a statistical test
> for your data.

There are at least two other reasons for using a sophisticated method
for choosing seeds (where the second is probably not a problem with
generators with large cycles):

1. If you use coupled random generators (e.g. when doing a HWIL
or a distributed simulation).  In this case, if you always choose
the seed for generatorB to be +1 of the seed for generatorA (or
any fixed algorithm, actually), then you run the risk of tying
the execution of systemA to systemB through more than just the
inherent synchronization of the two systems.

2. If the "state-space" of your generator is larger than the space
of seeds, yielding a 1-many mapping between seeds and initial
states.  Granted, starting the sim with seedA will result in
different phenomenology from one started with seedB; but,
unless the runup process is a good enough "hash", so to
speak, where the seeds cover the state-space in a homogenous
way, artifacts could still arise from preferring seeds in one
region of the seed-space over another (like always using a
number within some sub-range).  Hence, the shape of the runup
process (the method of choosing the seed) could easily affect
the PRN.

glen
--
glen e. p. ropella =><= Feeding the hamster wheel.  Hail Eris!
Home: http://www.swarm.com/gepr                (505) 424-0448
Work: http://www.swarm.com                      (505) 995-0818

                 ==================================
  Swarm-Support is for discussion of the technical details of the day
  to day usage of Swarm.  For list administration needs (esp.
  [un]subscribing), please send a message to <address@hidden>
  with "help" in the body of the message.