[gmx-users] NVT to NPT. Works on a single processor, does not work on 4 ????

[Mark Abraham]

> Hi
>
> I seem to be doing something wrong here, but cannot figure out what.
>
> -  I have run a 200 ps NVT simulation for a lipid bilayer on 4
> processors using the gromacs-3.3.1 build.
>
> - The last frame from the NVT simulation was used in conjunction with
> a newly written .mdp file to restart the simulation using grompp.
> (Before doing this, the trajectory and the final .gro file were
> deshuffled).
>
> - When the NPT .tpr is built for 4 processors (using -shuffle and
> -sort), the NPT simulation immediately explodes perhaps because the
> density is high (1100 kg/m3). However, if the NPT .tpr is built for a
> single cpu, then the simulation does not immediately explode, and has
> run now for at least 3000 steps. (the .edr file from NVT was not
> used). The same mdp file was used to generate both the 4-cpu tpr and
> the 1-cpu tpr.
>
> Why should this happen ??? I am hoping I have made some mistake here.

Numerical simulations do that when you shift to an ensemble where the
present structure is far enough from equilbrium. Try a gentler tau_p
and/or position restraints on your solute.

Mark