I carried out an ice slab simulation with tip4p/pppm and non-periodic boundary conditions (in my case, in z-direction). The pppm algorithm divides the forces in two parts: short- and long-range ones. The long-range would be calculated using FFT.

Because of my non-pbc, the pppm algorithm cannot consider effects of an eventual image due to pbc. For example, the effects could be discarding energy contributions from supercell electric configurations (dipoles, quadrupoles, etc). It is a consequence for any kind of calculation considering non-periodic condition, but not a difficulty for using PPPM since FFT could be calculated for non-periodic functions.

Is there any point about PPPM and non-periodica condition that I cannot see at the moment?

please note that the non-periodic PPPM algorithm works the following way:

you enlarge the box in z direction by the given factor (typically 3.0, as required by the poisson solver)

you make a fully periodic calculation

you compute the residual dipole for the slab

you apply a poisson solver to compute the interactions between the slabs from that residual dipole (it will only converge, if the distance between the slabs is sufficiently large) and subtract that from the periodic calculation. any charged particle that moves into the extra vacuum will thus cause inaccuracies.

this will not fully decouple the interactions. higher order terms are still included but should be rather small.

the only option in LAMMPS to completely decouple periodic images is through MSM, which - in the current implementation in LAMMPS - is either slow or not very accurate. improvements to the algorithms have not been implemented yet.