Using tip4p force field potential in non-orthorhombic cells is very
important for me and my colleagues.
I would like to know is there any chance that the problem of tip4p to be
solved?
yes, we *are* looking into it and a first set of related changes will be
in the next LAMMPS patch.
however, as already noted, there are multiple problems that need to be
addressed and some are quite complex to track down and correct.
furthermore, different LAMMPS developers have different are expertise, and
in this case multiple people have to look into different aspects. yet,
because of the complexity of the issue, it is best to solve each part of
the problem individually and completely before moving on. add to that, that
this is not the only known issue and that other ongoing development work
needs to be done.
in short: please have a little patience.
please also consider, that you can also use an orthogonal supercell as a
workaround.
axel.
hiya!
the next step is solving this problem is now available as a pull request: https://github.com/lammps/lammps/pull/544
however, there is still a final challenge: the resulting code creates memory corruption for MPI parallel runs when the box is tilted (serial and OpenMP with a single MPI rank works). i currently have no idea, where this is coming from, possibly “change_box”?
axel.
hiya!
the next step is solving this problem is now available as a pull request:
https://github.com/lammps/lammps/pull/544
however, there is still a final challenge: the resulting code creates
memory corruption for MPI parallel runs when the box is tilted (serial and
OpenMP with a single MPI rank works). i currently have no idea, where this
is coming from, possibly "change_box"?
final update,
i figured out the reason for the memory corruption: one needs to
(re-)define the kspace style after changing the box to triclinic, as memory
allocation dimensions/offsets change. the updated pull request is now
cleared for inclusion into the next LAMMPS patch release.
axel.