[CP2K-user] [CP2K:13023] Wanting to get synchronized trajectories

Noam Bernstein noam.b... at gmail.com
Thu Mar 26 13:44:02 UTC 2020

As Patrick said, _any_ difference in calculation will result in growing variation in the trajectory, since it is chaotic. However, computer operation is deterministic (if using deterministic algorithms), and so even if the solution of the equations is approximate, there’s no inherent reason that it can’t be deterministic. if you start from the same numbers and do the same operations on them, you’ll end up with the same results. 

At one point CP2K was using deterministic algorithms for ab initio MD, except for parallel reduction operations. The BLACS_REPEATABLE flag causes a blacs_set call to request topologies that lead to repeatable parallel operations. That was enough back then. Not sure if it still is now.


More information about the CP2K-user mailing list