Seed numbers bug in MC

Evgenii feti... at umn.edu
Thu Oct 22 21:01:57 CEST 2015


Hi developers,

I just wanted to point out a bug in MC subroutines regarding streams of 
random numbers.
One would think that if different &GLOBAL&SEED values are specified then it 
should lead to independent MC runs (different sequences of moves etc.). But 
apparently, if one is to look  at line 153 in *mc_runs.F* with the call to 
initialize random numbers, then it does not depend on &GLOBAL&SEED 
resulting in all runs having the same streams.

What would be the best way to quickly fix it? One of the suggestions was to 
use last_rng_stream=force_env_1%globenv%gaussian_rng_stream as an optional 
argument to distinguish between the seeds. But there should probably be a 
better way which is consistent with the rest of the code. Anyway, I believe 
this should be fixed in the upcoming release version.

Thanks!

Evgenii
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20151022/cd665a6f/attachment.html>


More information about the CP2K-user mailing list