Hi,
do you want to thermalize the structure to a new temperature or model the heating of the system?
Generally you can use any thermostat and any ensemble that takes T as an external parameter, i.e., NVT and NpT. In practice, people use NpT and slowly turn up the temperature.
The "constant T" one often hears when referring to NpT or NVT is a bit confusing in this context. It rather means that the temperature is set to the value from the input via a thermostat. The input takes a starting temperature
TEBEG and a final temperature
TEEND. Hence, the temperature imposed by the thermostat varies over the simulation time. There is an important caveat: The temperature observes fluctuations because it is regulated by a thermostat and not strictly imposed like the volume in NVT. Thus, you need to average the value over a few time steps. If your system is well-thermalized and
TEBEG=
TEEND, there should be no drift in the temperature over the simulation.
TEBEG not equal to
TEEND allows for temperature ramping up or cooling down over the course of a simulation. For thermalization this can be done quite fast. To simulate actual heating or cooling you need very long simulation times such that the system basically thermalizes at each intermediate temperature step. In other words, if you want to extract physical properties from the system, you need to ensure the MD run corresponds to a proper statistical ensemble.
Finally a caveat regarding NpT for ab-initio MD in VASP. It is best practice to chop the total MD run into smaller MD runs. This is because when the volume/cell shape changes also the PAW basis should be reinitialized otherwise the system experiences unphysical stress.
Does this answer your question?
Best regards,
Marie-Therese