NVT vs NVE temperature fluctuations

Hi all,

I guess my questions are more related to MD fundamentals but since I’m using LAMMPS for the simulations and since looking at the books hasn’t (yet) provided an answer I thought to ask here. I use NVT and NPT with the following commands to equilibrate small systems (made of organic molecules in liquid phase with 5000 to 30000 atoms):
fix 1 all nvt temp 300 300 0.06
fix 1 all npt iso 0.0 0.0 0.6 temp 300 300 0.06

After that, I use NVE to check energy conservation. When I plot temperature against time during the whole simulation I observe larger fluctuations in NVT and NPT compared to the subsequent NVE (about 30% larger). This behavior persists with different values of the thermostat’s damping constant. My first reflex was to expect the contrary: a better controlled temperature in NVT then in NVE. So, is this a typical (and physical) behavior? And if so, could someone please explain why NVE does a better job when it comes to temperature fluctuations then the thermostat does?

Thank you!


Open your textbook and read the section on fluctuations. Fluctuations of intensive quantities should be of order 1/sqrt(N). There is good theory describing the effect of different ensembles that you should consult. Generally, fluctuations are a "good" thing, as they mimic the behavior of a real system of the same size.

I don’t know if “better” is the right word here… NVT and NPT are designed to give you realistic temperature fluctuations, “realistic” meaning consistent with the canonical or isothermal-isobaric ensemble, respectively. Assuming your system is properly equilibrated and ergodic etc., then your result would indicate that in the microcanonical ensemble, the temperature fluctuations are smaller than those in the canonical one. I don’t really have intuition as to whether or not this is to be expected or not.