[gmx-users] Gromacs not writing checkpoint files with multidir/replex (v 4.6.4)

[Michael Daily]

Hi all,

I have been attempting to perform successive MD runs using the -multidir
option, in conjunction with replica exchange, using restart from
checkpoint.  The first time the run is performed (see the shell script), a
checkpoint file is created in each of the 6 directories, as expected (see
stderr output in run1.log).  The second time, the runs resume from the
checkpoint files generated by the first cycle, BUT no new checkpoint files
are generated (see run2.log).  Thus, when I attempt a third round, the
simulations restart with the same checkpoints from the first cycle (see
run3.log).  Thus, it is effectively impossible to resume multidir/replex
simulations more than once.  Does anyone know of a fix for this problem, or
will some more elaborate solution be required, such as manually
daisy-chaining a set of short runs in separate directories?

Thanks,
Mike

Additional notes:
1) gmx 4.5 has the same problem
1a) No problem occurs when trying to extend an individual one of these runs
indefinitely without -multidir.
2) the problem occurs whenever using -multidir, not just if replex is used.
3) While I am using short -cpt intervals for testing purposes, the same
problem occurs if the standard interval (15 min) is used.
4) Using -noappend doesn't make a difference

-- 
====================================
Michael D. Daily
Postdoctoral research associate
Pacific Northwest National Lab (PNNL)
509-375-4581