Yes, it does…
But it gives “-nan” as the total energy, van der waals (pair) energy and pressure. It seems like some atoms overlap. But I have checked and made sure it does not happen so.
for more than one processor, sometimes it shows, “angle atoms … missing on proc … on step 0” and sometimes “bond atoms … missing on proc … in step 0”
Yours faithfully
Yes, it does…
But it gives “-nan” as the total energy, van der waals (pair) energy and pressure. It seems like some atoms overlap. But I have checked and made sure it does not happen so.
for more than one processor, sometimes it shows, “angle atoms … missing on proc … on step 0” and sometimes “bond atoms … missing on proc … in step 0”
Ok. If it runs in serial, but crashes in parallel, then it is likely your domains change too much during neighbor list updates, e.g. when the box changes rapidly, for example through bad choice of box dimensions for shrink-wrap boundaries.
Also, are you using the current version of LAMMPS? If not, always verify, if the same thing happens with the current version and thus make sure you don’t make people chase an issue that has already been addressed.
Axel
If it gives NaNs in serial, it’s not likely to magically
work in parallel. Are you using the special_bonds command
to exclude pairwise interactions for atoms in the same
bond or angle?
Steve
If it gives NaNs in serial, it's not likely to magically
work in parallel. Are you using the special_bonds command
to exclude pairwise interactions for atoms in the same
bond or angle?
i downloaded and ran the posted input example, and it works in serial
with the latest version of lammps. the included output indicates that
this input is run with a >1.5 year old version of LAMMPS. so the NaNs
are likely a bug that we fixed already.
the missing bonds are indeed the result of incorrect use of
shrink-wrap boundaries. the initial box is too large and thus shrinks
massively at the first reneighbor bonds are lost because of that.
PEBCAC.
axel.