I would like to run a Lammps simulation using the minimum image convention
but without using any cut off (ie. I would like to include ~N^2
interactions, but only to the closest image). Is there a way to do this in
Lammps?
As far as I know, setting cut off < half box size is a typical way to
guarantee using the minimum image convention in Lammps. Since I would like
to include all interactions, "cut off = sqrt(3) * half box size" (for
cubic box) would be more appropriate. But with this cut off, I expect to
get multiple images onto the neighbor list. I would be even OK with the
extra cost of keeping these multiple images on the neighbor list, as long
as this extra energy term would be ignored.
It sounds like you want to have a periodic system, but
include interactions for each particle in a cube
surrounding it, rather than a sphere. I agree with
Paul there is no way to do this currently. Why
is that a good model for what you want to simulate?
My simulations with cut off (say optimization of lattice parameters) weren’t quite well-behaved, and I was advised that the simulations might be more well-behaved without cut off, just with the minimal image convention.
There has been a good experience with such approach in the past and I thought it could help, since the number of interactions stays constant.
Again, what is the mental picture you have of what
surrounding atoms interact with a particular atom?
There is a difference between all interactoins in
a non-periodic system, or full long-range interactions
in a periodic system, or a "cube" of interactions
in a periodic system. LAMMPS can do the first two,
but not the 3rd. And I've never heard of a model where
the 3rd makes sense.