Page 1 of 1

ML_WTSIF in MLFF

Posted: Thu Oct 17, 2024 5:04 am
by reza_namakian1

Dear VASP Team,

I have a question regarding the use of ML_WTSIF in MLFF.

The documentation says:

"If you train a system containing a surface or isolated molecules, do not train the stresses. For that, set ML_WTSIF to a very small value in the INCAR file (e.g., ML_WTSIF=1E-10)."

I want to train the MLFF for a bulk phase and then further train it for the same material system but in a slab configuration with free surfaces. In this scenario, should I set ML_WTSIF to 1e-10 for both cases during the training, or only for the system with free surfaces?

Thank you in advance for your help!


Re: ML_WTSIF in MLFF

Posted: Thu Oct 17, 2024 7:58 am
by ferenc_karsai

If you train the bulk material first it doesn't matter for the bulk calculation what you set, because in the continuation run you would anyway fit with practically zero weight. If you train on surface first and then bulk, then you should set it in both cases.