[gmx-users] random number generation

Erik Lindahl lindahl at stanford.edu
Tue Mar 4 02:59:57 CET 2003


On Monday, Mar 3, 2003, at 17:50 US/Pacific, Jim Dix wrote:

>   To answer my own question: Gromacs uses a uniformly-distributed
> random number generator with a period of 10^8. The random numbers so
> generated are transformed into a Gaussian-distributed random numbers.
> Thus, if I figure correctly, on a 1000 atom BD simulation with a
> random force on each of the three dimensions, there would be at most 3
> x 10^4 steps available before the "random" numbers repeated
> themselves. The lore is that one should not use a random number
> generator for more than 10-20% of its period, so that means that there
> would be on the order of only 5000-10,000 BD steps available for a
> given seed to the random number generator. (?)
>   Whether the random numbers generated by a particular seed are good
> random numbers still needs to be tested.
>

Hi Jim,

Good to know; I'll add one of the "good" random number generators for 
the next release so we don't have to worry about it.

Cheers,

Erik






More information about the gromacs.org_gmx-users mailing list