[gmx-users] precision of coordinates
mark.j.abraham at gmail.com
Wed Jan 25 22:29:03 CET 2017
The trr by definition gives you the full single precision result of what
was computed. But you'd surely compute different approximations to the
exact forces if you changed the atom order, or rotated the box, or changed
the number of domains (try the latter with mdrun -rerun -ntmpi x). The
practical consequence is that the number of significant force digits
depends quite a bit on the largest component that contributes to each.
Various implementation details that are algebraically equivalent also
affect this, in practice.
But it's worse than that. The updates take the form x' = x + dx, with dx
sampled from something vaguely normal. But the size of the smallest dx that
can affect whether x' != x depends on the size of x. So if the simulation
cell size ranges over two orders of magnitude, then the updates near the
origin probably retain more significance than those far away. Someone is
supposed to be studying that, but no progress has happened yet.
So, surely only some of the position digits are significant. How badly that
is dominated by model effects such as the choice of vdw cutoff or lack of
polarization surely depends on the system, too :-)
On Wed, 25 Jan 2017 21:21 Irem Altan <irem.altan at duke.edu> wrote:
> I was wondering, what is the reasonable precision to expect from heavy
> atom coordinates in a trajectory? I’m currently printing higher precision
> data from the .trr file using the -ndec flag, but I would like to know if
> this is meaningful.
> Gromacs Users mailing list
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
More information about the gromacs.org_gmx-users