Random number seed for Langevin MD

Valentin evang.v... at gmail.com
Thu Oct 13 17:19:23 UTC 2016


my post concerns the problem that the value of the global seed does not 
affect the initialization of the parallel random number generator that is 
used for the Wiener process in the Langevin ensemble. For cp2k version 4.1*, 
*the first particle-specific seed is set in line 95 of 
*motion/wiener_process.F* by calling *next_rng_seed()* without arguments, 
thus resulting in 12345.0.

I observed (using version 2.4) that setups that differed only in their 
initial velocities showed strong correlations among each other in their 
dynamics after a short time, which I would attribute to the noise term 
being de facto not random.

A similar issue for MC calculations has been posted in this forum before 
(subject "Seed numbers bug in MC") and the problem has been solved in 
version 4.1 by specifying 
*last_rng_stream=force_env_1%globenv%gaussian_rng_stream*. Could you fix 
this problem for the Wiener process as well?

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20161013/5b4aa414/attachment.htm>

More information about the CP2K-user mailing list