Can MLFF way be effective in studying magnetic materials?

[wdovew]

Dear, VASP Expert,

Thank you very much for your reading and answers. The problems are follows:

By utilizing the better output files（ML_FF and ML_AB by ML_MODE =test） and later calculating the energy in different magnetic configurations( ML_MODE =run). I'm confused about such use, I wonder if vasp can use the machine learning potential directly for total energy calculations for different magnetic structures after obtaining ML_FF. Is the energy difference between the different magnetic structures within the error allowance of the direct self-consistent calculations? Have you tested this for some magnetic materials?

Looking forward to your reply.

Best wishes.

User, wdovew

[ferenc_karsai]

At the moment there is no quantity such as the magnetic moment implemented in the MLFFs which could distinguish different magnetizations. So if magnetization plays a strong role for structure distinction the force-fields cannot predict the right structure solely on the lattice and atom positions itself.
If magnetization plays a minor role for structure distinction one can get away with the currently implemented force field.