It looks like restarting a simulation using Nose-Hoover thermostat does not result in exactly same results. Need to investigate deeper.