You’re right, the problem persists without the gpu package. I tried creating a smaller system with the same triclinic geometry (9000 atoms), but it worked fine. My actual system is about 940k atoms, which may be unwieldy for you to work with, but if you have the time and resources, I’ve included a link to a 16mb compressed tar file with a simplified script, data file, and potential, perhaps you could take a look. I forgot to mention earlier, but I have a shrink-wrapped boundary in one direction. If I make the system fully periodic, the problem goes away. So it’s likely related to the boundary condition… Thanks for your help.
Are you saying run for 1000 steps to produce
a restart file, then that restart file has the problem?
How many procs do I need to run on to reproduce the error?
I don’t really want to run 1M atoms for 1000 steps.
Can I see the problem with less atoms or less steps?
Thanks for the response, you can run any number of steps I believe, I just tested with 10 steps and it still gives the same error. It takes about 10 seconds total to run on 16 procs. You can modify that input file, or you can use this slightly updated one.
input.0 (631 Bytes)
ok - I fixed it. It was a problem peculiar to triclinic
boxes with shrink-wrap boundaries. Will be
in the next patch.
Note, however, that in your system you are defining
tilt factors that include z (e.g. xz and yz) while z
in non-periodic. This means there is no way for
the system to enforce the tilt and the atoms will
not be held in positions that reflect the tilt. That
is b/c it is only the periodic images of atoms in z
that would enforce an xz or yz tilt. So your model
will likely not do what you want.