Problem with floating atoms

Dear LAMMPS users,

I’m doing a semi infinite ( p p s ) simulation tensile loading ( stretching in z direction). At the point that fracture happens, an atom separate from the system and make the simulation cell in non-periodic direction, i.e. z, growing drastically which is artificial effect to the best of my knowledge. I noticed that in some research, people just delete does " dangling " atoms. However, I wonder what makes the system to do that, and is there any other way to eliminate it ? It is obvious that by having ( p p p ) this phenomenon doesn’t happen.