On Fri, Aug 24, 2018 at 5:47 AM, Helena Damtoft Tjørnelund <
s153842.student.dtu.dk> wrote:
> Hello Vinicius, Jason, Bill,
>
> Thank you for your comments.
>
> I did try all you of your suggestions.
>
>
> I tried running a few steps of production run using pmemd.mpi (around 2000
> frames)and then switching to pmemd.cuda. It still crashes at exactly 500th
> step.
> Just running my system using pmemd.MPI takes about 3000hours for 100ns.
>
Is the simulation stable with pmemd.MPI? Look closely at the energy terms
and make sure they're sensible. It's possible there's a pmemd.cuda bug in
there somewhere for CpHMD explicit solvent simulations that's causing your
simulations to blow up.
It may be that a protonation step that is accepted in GB causes large
forces when solvent relaxation begins that's unstable with pmemd.cuda's
precision model.
A few ideas to try and narrow down the problem:
1. Try pmemd.MPI and see if the simulation stays stable for an extended
period of time.
2. Try pmemd.cuda_DPDP and see if increased precision leads to a stable
simulation.
Also, as Vinicius stated, you should use 0.1 for the saltcon parameter.
This won't affect simulation stability because it has a small impact on
just the protonation state change event. However, the reference energies
were computed using 0.1 for this parameter, so you should use the same
value as well.
HTH,
Jason
--
Jason M. Swails
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Aug 24 2018 - 07:30:01 PDT
