unexpected temperature fluctuations in NVT run

[joachim_bacquet]

Dear all,

I am trying to run MD calculations with an NVT ensemble at 300K. As far as I understand the manual, this implies using the Nose-Hoover thermostat. However, it appears that the temperature fluctuates between zero and almost 10000K during the run (see attached Temperature_by_step image). I do not expect these fluctuations to come from the choice of the SMASS parameter, as there seems to be smaller fluctuations corresponding to it (Zoom_fluctuations). If anyone has experience with such issues, I would be glad to receive pointers on what I did wrong in my INCAR file (included in the run_files.zip).

Thank you and best regards,

Joachim Bacquet

[ferenc_karsai]

Does your system contain Hydrogen? If yes then your POTIM is too large. Please try then with POTIM=0.5. Also maybe you want to switch on calculation of the stress tensor to check the pressure (ISIF=2). Maybe it's extremely high and the structure is too far from stable. In that case it would be maybe good to relax the structure before starting the MD. You can also try the Langevin thermostat for NVT (MDALGO=3 with ISIF=2).
These are very general things to try since you haven't posted your structure. Also look at your structure with a structure viewer like VESTA and see visually what happens (POSCAR and CONTCAR after given number of steps).

[joachim_bacquet]

Thank you for your answer.

Does your system contain Hydrogen? If yes then your POTIM is too large. Please try then with POTIM=0.5.

No my system does not contain any hydrogen, which is why I tried a larger POTIM. As far as I can tell from my tests, the time step does not seem to affect this strange behaviour, no matter how much I increase or decrease the value of POTIM.

Also maybe you want to switch on calculation of the stress tensor to check the pressure (ISIF=2). Maybe it's extremely high and the structure is too far from stable. In that case it would be maybe good to relax the structure before starting the MD.

I actually did make a relaxation run before, along with DOS and bandstructure calculations as I am trying to familiarize myself with VASP calculations at the moment. All my results so far match what can be found in the litterature, hence why I didn't think of adding ISIF=2, but I will try it now, thank you for the suggestion.

You can also try the Langevin thermostat for NVT (MDALGO=3 with ISIF=2).

I could indeed try this as well, thank you. I went with Nose-Hoover because I read a publication where it is mentioned and I wanted to reproduce the results, but changing the thermostat could do the trick.

Also look at your structure with a structure viewer like VESTA and see visually what happens (POSCAR and CONTCAR after given number of steps).

I looked at a few configurations from the XDATCAR file in VESTA and the structure does not change much, even when the temperature reaches its peak, although it might be because I work with pure graphene for now. I will try your suggestions and post again to let you know how it turned out.

[ferenc_karsai]

Ok seems you tried a lot. So after trying the other thermostat it is best to upload your example here (POSCAR, INCAR, KPOINTS, POTCAR, OUTCAR and stdout). Then I will try to run it myself.

[joachim_bacquet]

After trying your suggestions, it appears that I still cannot get the results I want. Therefore, I am attaching to this reply zip files containing all I have for both thermostats. Note that I had to lighten the folder for Nose-Hoover due to the size of some files, so please do not hesitate to ask me for any missing information (the OUTCAR in particular couldn't be fully uploaded).

The INCAR file should be the only difference between both runs, with one yielding the fluctuations previously discussed, and the other giving a constant temperature of 0K before encountering an error.

Thank you for your help with this matter,

Best regards,

Joachim Bacquet

[ferenc_karsai]

Ok, I've ran the calculation. The problem is that at the beginning your temperature is 0 for a very long time. This is because after the positions in the POSCAR file the starting velocities are listed and they are all 0. Since you start from a perfect structure the forces are also more or less zero and no movement happens.

To remedy that delete everything in the POSCAR after the positions. The starting velocities will be drawn randomly from a Boltzmann distribution at the starting temperature (TEBEG=300).