Re: Random numbers generation

[Stuart Prescott <debian@nanonanonano.net>]

Hi all,

>   /* create a generator chosen by the environment variable GSL_RNG_TYPE */

At the risk of straying further off topic, I can't help reminding
everyone of the importance of selecting a good random number generator
for scientific work (e.g. Monte Carlo simulations). There are far too
many crappy random number generators Out There and many of them are
still in widespread use.

The GSL documentation talks a bit about the different generators
available and I highly recommend some of the references therein (Park &
Miller and L'Ecuyer in particular). 

	http://www.gnu.org/software/gsl/manual/gsl-ref_17.html

Personally, I have made use of the Marsaglia Random Number CD (once the
10MB random bit files are loaded into memory in the disk cache, it's
quite fast) as a good source of reproducible pseudo-random numbers.

	http://stat.fsu.edu/pub/diehard/

or mirrored at (the CD seems to be missing from the original):

	http://www.cs.hku.hk/~diehard/cdrom/

The only disadvantage that I found was a little latency in starting
simulations in a cluster environment as the amount of data to be copied
to the computational node was greater.

cheers
Stuart