[gmx-users] g_msd bug

[Matthew Zwier]

I just experienced this myself.  The problem appeared to manifest
itself when I was using -mol on a molecule that straddled the box
wall.  Memory usage was extremely high and the resulting MSD plot did
not show any linear behavior.  Imaging the trajectory with -pbc nojump
made g_msd's memory usage drop dramatically and produce a MSD plot
more like what I would expect -- a linear region on the 10s of ns
scale bookended by nonlinear regions.  In this latter case, the
results with and without -mol, though not numerically identical, are
at least approximately equal.  So, it looks like PBC handling may be
at fault for one or both of the problems (memory usage and results
with/without -mol) here.

Regardless of whether this is a bug or a natural consequence of trying
to do a MSD calculation on broken molecules, the documentation for
g_msd implies that g_msd handles periodic boundary conditions
implicitly; perhaps the documentation should be updated to emphasize
the (in hindsight, completely logical) fact that molecules need to be
whole before running g_msd -mol.

MZ