[lammps-users] Lost atoms with fixed boundary condition

Steve, LAMMPS users,
I am trying to run a NVT simulation with the script below with fixed boundary condition in the z-direction (the domain involves a silicon and aluminum layer adjacent to each other). No matter how small I make the timestep, I still seem to end up with some lost atoms. If instead of the fixed boundary condition in the z-direction, I put a periodic condition, the simulation works fine even with a timestep of 0.00055 (i.e. a factor of 100 higher than what I have below).
I am wondering what else I can do to avoid losing atoms from the domain (I assume it is undesirable to lose atoms from the domain).

Steve, LAMMPS users,
I am trying to run a NVT simulation with the script below with fixed
boundary condition in the z-direction (the domain involves a silicon
and aluminum layer adjacent to each other). No matter how small I make
the timestep, I still seem to end up with some lost atoms. If instead
of the fixed boundary condition in the z-direction, I put a periodic
condition, the simulation works fine even with a timestep of 0.00055
(i.e. a factor of 100 higher than what I have below).
I am wondering what else I can do to avoid losing atoms from the
domain (I assume it is undesirable to lose atoms from the domain).

how close is your z-boundary to the atoms?
have you considered making your simulation cell
a bit larger in z-direction to give your atoms
some room to vibrate, or use a shrink-wrapped
boundary in z direction?

cheers,
   axel.

Could it be the LJ potentials that are blowing the atoms apart? Their
epsilon values seem a little high.

Zhun-Yong