(High) Electric field, temperature control, thermostat

Dear All,

I am running some simulations that require applying a relatively high electric field pulse (~10ps) to a solid material.

The magnitude of the electric filed is normally around 1MV/cm to 4MV/cm (0.01eV/A - 0.04eV/A).

I use “fix NPT” to control the temperature. Using the “metal” unit, I set the Tdamp to be 1.0 (that is 1.0ps) and use a time step of 1fs (sometimes 0.5fs).

I think my Tdamp is reasonable.

I notice that if the electric field is high, for example, 2.0 MV/cm, the temperature will shoot up for about ~50K.

In practice, we can tune Tdamp (e.g., Tdamp=0.05) so as to maintain the temperature during the application of electric field. But, I am not sure whether that is a physically correct way to do:

By applying the electric field, the system is gaining energy. Assuming that the rate of energy exchange between the system and the environment (thermostat) is constant, the temperature of the system should rise up.

The situation now is Tdamp will influence the kinetics of my system. (Tdamp=1.0 will give very different result when Tdamp = 0.05).

I am eager to know your suggestions or comments.

Sincerely,

Shi

If you are adding a big burst of energy
to the system, then trying to use a thermostat
to prevent the temperature from rising is an
unphysical thing to do. That’s also not

what thermostats are designed for. Why
shouldn’t the temperature rise?

Steve

You should look in the literature first to see how people deal with this kind of problems. High and localized energy bursts are very common in radiation damage studies. There, the tendency is to avoid thermostats and run NVE simulations with large system sizes that allow the damping of the large temperature spike initially created without modifying the EOM. Of course, you have the extra complication that your energy burst is periodic in time. Again, look in the lit because the solution might be tailored to the property/behaviour you intend to calculate/simulate. Also remember that large perturbation might require modifications of the simulation timestep.

Carlos