Steve,
It was resolved. I think the problem was that the number of mesh points along the periodic direction (z) was getting smaller as I increased the other dimensions when I allowed the mesh sizes to be automatically set. I manually changed the mesh sizes and there was no problem. I switched to using a slit geometry instead of a channel for now anyway, since getting the forces accurate (especially in the z direction) was going to be expensive.
There isn’t a version of Ewald/PPPM/PME for 1D is there? I have not found anything on this. I tried to get the multipole method (DPMTA) to work in NAMD, but was unable to figure out how to compile it. NAMD doesn’t really support it anymore, but the code is still there if you can figure out how to cobble all the pieces together (it uses PVM instead of MPI). If a cutoff method isn’t satisfactory for the channel, I may have to try to get a multipole method to work with LAMMPS, or just waste a lot of time simulating empty space.
Brian
I don't know of any 1d PME or PPPM. Paul may know of one,
or if it would be hard to implement, similar to what he did for 2d slab.
Steve
I've never heard of a 1d Ewald, PME, or PPPM.
There is a 2D Ewald for doing slab geometries, but it is expensive compared with the 3D Ewald with slab correction or 3D PPPM with slab correction. So we've implemented the latter in LAMMPS that is a good approximation to 2D Ewald. And yes, you end up simulating "empty space" but that's cheap compared with the real 2D Ewald.
Paul
I've never heard of a 1d Ewald, PME, or PPPM.
actually, there is a 1d derivation for a regular Ewald sum.
a colleague of mine was supposed to implement it for our local
MD code when i was in grad school (i.e. quite a while ago).
however, it turned out that it was next to impossible to make
it work well for the systems that we were working on (water in
cylindrical "pores" with a functionalized wall simulated by
point charges). the error would depend on the distance from
the center of the cylinder and we were particularly interested
in the functionalized walls.
There is a 2D Ewald for doing slab geometries, but it is expensive
compared with the 3D Ewald with slab correction or 3D PPPM with slab
correction. So we've implemented the latter in LAMMPS that is a good
approximation to 2D Ewald. And yes, you end up simulating "empty
space" but that's cheap compared with the real 2D Ewald.
according to the works of my phd advisor, the errors resulting
from a 3d-ewald vs a reald 2d-ewald are most prominent for relatively
small system sizes. the typical calculation that you'd use LAMMPS
on is much larger compared to what we could work on in those days,
so unless one is working on a pathological problem, i'd expect that
what is in LAMMPS should work well.
cheers,
axel.