[CP2K-user] [CP2K:13022] Wanting to get synchronized trajectories
patri... at gmail.com
Thu Mar 26 12:56:11 UTC 2020
Even if you use only methods (such as thermostats, barostats, etc.) that do
not rely on random numbers, inherent numerical errors will accumulate over
time and affect the simulated trajectory over a sufficiently long time
period. If it is absolutely necessary to sample the exact same trajectory
twice, there are several things you can try to do. You can tighten the
convergence criteria for the SCF loop. You can reduce the time step of the
simulation. You can reduce EPS_DEFAULT or the specific EPS_ variables
governing the various aspects of numerical accuracy. However, as we're
dealing with a chaotic system, even negligible errors and inconsistencies
accumulate and propagate to dramatically change the resulting trajectory.
You can make your calculations ten times more accurate and burn X times
more computational resources, but this will probably buy you just a few
more consistent steps in the parallel trajectories before you see them
diverge. This is not a bug of CP2K, but a feature of the underlying
physical and computational problem.
In any case, if you make sure you are properly converging your wave
function at each simulation step, and that your time step is short enough
to sample all vibrational modes in the system, then even microscopically
diverging trajectories should contain the same macroscopic information. For
example, the vibrational spectrum inferred from your dynamics, the radial
distribution functions, the average potential energy, the computational
cell volume, etc. should be (more or less) identical for both parallel runs.
On Thu, 26 Mar 2020 at 13:19, Abdullah Bin Faheem <
abdullahb... at gmail.com> wrote:
> Hello everyone,
> I had a question regarding AIMD. Is it possible to get the same xyz
> coordinates from two parallel runs with the same input files and starting
> What I mean is that can Run 1 and Run 2 produce the same xyz coordinates?
> As when I do that using the same input files for both, the difference in
> xyz coordinates increases as time passes, which is something I do not want
> to happen.
> I have attached the sample input files.
> Thank you for your time.
> Abdullah Bin Faheem
> You received this message because you are subscribed to the Google Groups
> "cp2k" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to cp... at googlegroups.com.
> To view this discussion on the web visit
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the CP2K-user