[gmx-users] RMSD truncation Restart simulation problems

[Henri Mone]

Hi Mark,

Thanks for your feedback. The simulations are running now again.
I got one questions, I think it will be also of interest for the
others on the mailinglist.
I used "grompp" with the "-t state0.cpt"  option, are the starting
velocities taken from the cpt files or are the assigned randomly from
a Boltzman distribution.
Meaning are the new simulations starting fresh or are they
acontinuation from the previous ones?

Thanks,
Henri

On Tue, Mar 8, 2011 at 1:54 PM, Mark Abraham <Mark.Abraham at anu.edu.au> wrote:
> On 8/03/2011 9:41 PM, Henri Mone wrote:
>Ah yes, I remember now. mdrun tries to be smart and
>check that all the files match the state they were in before
>the crash by computing checksums when writing and again
>when reading.
>So the original problem was that the steps recorded in the different
>.cpt are different, unsurprisingly.
>To fix this, get the .mdp used to make the original .tpr, and call
>grompp the same way, plus (e.g.) -t state0.cpt for each of the
>replicas. Now the (matching) state information will be taken by
>grompp from the (matching) checkpoint files, and no checksum
>calculation can fail. Then
>mdrun_mpi -multi 32 -replex 5000 -s new.tpr
>should be fine (and no .cpt should be provided).
>While doing this, I would avoid trying to use mdrun -append,
>by the way. Glue things together later if you need to.
>
>Mark