[gmx-users] protein jumps 10nm south in the ~20000th frame

[Paul Barrett]

Id like to report odd behaviour in mdrun that I *suspect* is a bug.

Im simulating a globular protein with a resident ATP and Mg2+. The
system is water solvated with periodic boundary conditions.

All goes well for the first couple of hundred thousand steps (2 fs
steps) and then the problem arises.

When the .xtc is viewed in VMD you see the whole protein jump (intact)
about 10 nm "south" leaving the ATP and MG where they were. Other than
that all is fine. This configuration stays for a few frames, then
reverts to the correct config, only to lapse again a few frames later.

This continues over a couple of hundred frames then seems to fix
itself.

I note that the amount it jumped was similar to the size of my unit cell
for periodic boundary conditions. Maybe VMD is interpreting the two
chains as being in neighbouring cells when the molecule has drifted out
of the unit cell. ??? mdrun performs a MOD function happily bringing the
calculations back in frame, but VMD looks at the bare un-MODed
coordinates???? Just a guess.

I plan to just delete the funnny frames. Makes me a bit nervous
though...

any comments?

Paul