[gmx-users] Exact restarts don't seem to be exact

[Mark Abraham]

Hi,

I posted a while back with a lengthy analysis showing that restarting with
gromacs 3.3b on Itanium 2 using .edr and .trr files as input to grompp for
full accuracy wasn't reproducing the pressure at Step 0. Upon comparison
with simulations that "continued on" past the restart, not only was the
pressure not reproduced, but by the first ouput step, all of the energy
components differed between the non-restarted and the restarted runs. This
means that I'm not getting an exact restart.

I've now reproduced this with gromacs 3.2.1 on Pentium 4. If the .edr and
.trr files are binary, why aren't I seeing the step 0 pressure &
subsequent time evolution being reproduced?

Regards,

Mark Abraham