I am simulating a 2830 atom system with timestep = 0.5 fs using AIREBO potential. NPT thermostat for 1,000,000 steps for 300K and 0 bars works. When I next switch to NVE, the system temperature begins to drift upwards.
How can I prevent this drift?
Does the length of thermostatting in NPT affect the drift in NVE?
Should I use a smaller time step or larger atom number?
Is there a general rule of thumb for how long to run a system in thermostat before NVE?
daniel,
I am simulating a 2830 atom system with timestep = 0.5 fs using AIREBO potential. NPT thermostat for 1,000,000 steps for 300K and 0 bars works. When I next switch to NVE, the system temperature begins to drift upwards.
How can I prevent this drift?
to some degree, a drift is unavoidable. however, there are several factors
that can increase it. a too small cutoff or a too large time step for example.
however, too small a time step or large a cutoff will produce more numerical
noise, which will - again - increase the drift. most text books on MD discuss
how this happens in detail.
Does the length of thermostatting in NPT affect the drift in NVE?
no.
Should I use a smaller time step or larger atom number?
this needs to be determined empirically. the optimal time step depends
on the maximal speed of your particles, which in turn depends on mass
and temperature.
Is there a general rule of thumb for how long to run a system in thermostat before NVE?
until it is well equilibrated. 
that is usually the case when the box dimensions and total energy
don't drift anymore.
cheers,
axel.