Over the years I have repeatedly encountered the problem that NVT dynamics of simple liquids and solids with a Langevin thermostat do not manage to keep the desired temperature. I have observed this behavior for several atomic and molecular liquids with VASP as well as with ASE. Sometimes this can be fixed by reducing the timestep and/or increasing the coupling constant (within reasonable limits), but in some cases, even this does not help.

My current system is pre-equilibrated liquid pentane at T = 300 K at its experimental volume with all masses set to 10 amu and a semi-empirical GFN0-xTB Hamiltonian. I have tried increasing the coupling constant from 0.02 au (suggested value) to 0.05 au and 0.10 au and decreased the time step from 4 fs to 2 fs and 1 fs, but even after 20 ps of simulation time and a pre-equilibration of 10 ps with an even higher coupling constant the average T of the simulation remains at ~290 K instead of the desired 300 K.

I have made similar observations in countless VASP simulations of atomic liquids and solids (DFT Hamiltonian), so I'm starting to think this is a fundamental problem of the LV thermostat? If I'm not completely mistaken that behavior means that the thermostat can not put energy into the simulation fast enough. But where is the energy going? With such a short time step energy conservation should be really good but obviously it isn't.

What is my misconception, what am I doing wrong, and how can I fix that behavior?

Any help would be greatly appreciated.

Jan