Yeah, sorry - I suspected there might be something I was missing. According 
to the wikipedia page about the MT 
<http://en.wikipedia.org/wiki/Mersenne_twister> one of its disadvantages is 
just that:

It can take a long time to turn a non-random initial state—particularly an 
initial state with many zeros—into output that passes randomness tests 
<http://en.wikipedia.org/wiki/Randomness_tests>. A consequence of this is 
that *two instances of the generator, started with initial states that are 
almost the same, will output nearly the same sequence for many iterations 
before eventually diverging.*

So you should probably not do this without significant warm-up, or in some 
other way make sure that the seeds are different enough. If you don’t need 
randoms very often, you’re probably better off with James’ approach.

// T

On Wednesday, July 16, 2014 4:54:07 PM UTC+2, Gray Calhoun wrote:

On Wednesday, July 16, 2014 2:09:41 AM UTC-5, Tomas Lycken wrote:
> > Another way would be to just initialize a `MersenneTwister` stream
> > on each process, and seed them differently. Since the streams are
> > process local, there shouldn't be any communication overhead, and
> > streams with different seeds should be independent.
>
> My understanding is that streams with different seeds are not
> guaranteed to be independent, but there are some RNGs for which they
> are.
>
> --Gray
>
> ​

Reply via email to