Dear users and developers:
My system is only extended in 1 dimension and there’s longer range pair interaction.
So the ERROR: Cannot use neighbor bins - box size << cutoff (…/neighbor.cpp:1715) appears.
-
I wonder if there’s an easy way to build the bin just in 1D.
Say, in x direction only.
Or if I can skips link-cell method and still set up neighbor list but not update it every steps to boost performance.
I set a huge bin size but I’m still held by the same error.
Any suggestion?
-
About neighbor nsq:
Is the mpi or omp package also parallelize nsq by assigning each processor a fixed subset of atoms?
Thanks,
Dear users and developers:
My system is only extended in 1 dimension and there's longer range pair
interaction.
So the ERROR: Cannot use neighbor bins - box size << cutoff
(../neighbor.cpp:1715) appears.
1.
I wonder if there's an easy way to build the bin just in 1D.
Say, in x direction only.
no, but there is should be no cost to the neighborlist construction, if
you extend the box in the other directions. just make sure that you assign
domains only in that one direction via the processors keyword, to not end
up with processors that have no local atoms.
Or if I can skips link-cell method and still set up neighbor list but not
update it every steps to boost performance.
I set a huge bin size but I'm still held by the same error.
Any suggestion?
difficult to say anything without seeing a simple, specific example that
shows the issue.
2.
About neighbor nsq:
Is the mpi or omp package also parallelize nsq by assigning each processor
a fixed subset of atoms?
all neighbor list construction for *any* style of neighbor list is
parallelized via MPI and always via domain decomposition.
OpenMP support is also available. MPI always parallelizes over domains and
OpenMP over atoms.
this is easily confirmed with some very simple and easy to do test runs.
LAMMPS prints the breakdown of where the time is spent.
axel.