I'm working on parallel MCMC algorithms both in R and C, but I'm kindoff worried about the independence between the sequences in my

*p*parallel streams/threads.

I know setting a new random seed for each stream would be required but does that really insure independence? I don't think it provides any reinsurance that my sequences are truly independent.

PRNGs are typically also

*periodic*, which means that the sequence will eventually repeat itself every length T (see e.g. --> http://www.random.org/randomness/)

Another way this could be done is by making sure I choose starting positions in the PRNG sequence that don't overlap for each p streams (basically dividing the sequence T in p parts) but this doesn't really sound feasible.

Anybody worked with this problem before and got tips? I'm sure people with backgrounds in computing, math and stats have run into this problem before (other than in my field, Ecology, that is).

Any suggestions will be greatly appreciated!

TE