The simplest way to conduct a molecular dynamics simulation is to set up some initial conditions and propagate the system in time by solving Newton's equations of motion. When one does not interact with the system at all(no thermostat or barostat) this will be an NVE simulation, since the number of particles, the volume and the total energy is conserved.
My problem is, that I see no reason why(or how) such a simulation without external influences conserves volume. Consider simulating a cluster of water: due to the random motions, every once in a while a molecule on the surface will have sufficient energy to break away from the cluster, and fly away into what is effectively a perfect vacuum. In fact, that is exactly what I would expect from a tiny droplet of water in perfect vacuum: to evaporate and become a gas.
So my question is, how do MD simulations preserve volume?