yes, i am creating the particle in random positions.
Stating the obvious here, but more particles equals higher chances to
get random overlaps.
Take a look at the minimize command, it might help you take out some of the potential energy due to the overlap
Obviously, you want to start off with a smaller timestep. Just keep decreasing the timestep until you no longer get lost atoms. Then, once the system is equilibrated (i.e. overlaps are eliminated, or some more stringent criterion), the usual way to set a good timestep dt is to check for convergence of the velocity autocorrelation function as a function of dt.