Insufficient FFT grids for ML_FF production run

[chamara_somarathna]

Dear VASP team,

Could you please clarify a warning related to the ML force field use?
After completing the training and refit procedures, I obtained the ML FF for the fast production run. However, upon starting the production run, I immediately received the following warning:

"Your FFT grids (NGX,NGY,NGZ) are not sufficient for an accurate calculation. Thus, the results might be wrong. Good settings for NGX, NGY and NGZ are 4, 4 and 6, respectively. Mind: This setting results in a small but reasonable wrap-around error. It is also necessary to adjust these values to the FFT routines you use"

My OUTCAR prints the NGX, NGY, and NGZ all to be equal to 2. So I explicitly increased the NGX, NGY, and NGZ values in the INCAR, but they were ignored, and the warning persisted.
For mode=train, I used the default "train" parameters and ENCUT = 500, PREC = Accurate, having NGX=90, NGY=96 NGZ=128 in OUTCAR and the corresponding edges of ~11.72, 13.13, and 17.59 A; 432 atoms.

Should I specify any particular settings during the training or refitting stages to obtain a better FFT grid for the production run? Or should I modify the INCAR specifically for running the ML-FF?

I would greatly appreciate any guidance on resolving this issue.

Many thanks!

Best regards,
Chamara

[marie-therese.huebsch]

Hello,

Thank you for pointing it out! This is an oversight on our side. The warning should only be triggered if ab initio calculations are performed.

In fact, NGX, NGY, and NGZ are set to small values for ML_FF-only calculations to save memory, so setting them in the INCAR file will not affect the internal parameters. I added this to the known issues so that no one worries their calculation may be inaccurate.

Best regards,
Marie-Therese

[chamara_somarathna]

Hi Marie-Therese again,

Thanks for confirming this.

Best,
Chamara