My Community

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

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.