ML_EATOM_REF

ML_EATOM_REF = [real array]
Default: ML_EATOM_REF = 0.0

Description: Reference total energies of isolated atoms used in the machine learning force field method.

This tag is only used if ML_ISCALE_TOTEN = 1.

If ML_EATOM_REF is not provided in the INCAR file then 0.0 is assumed for all species in the system.

By default this tag is not used since all energies are scaled to the average of the training data (ML_ISCALE_TOTEN = 2).

If this tag is used, each reference energy should be obtained from a VASP calculation of an isolated atoms in a sufficiently large simulation box. The reference is then simply taken from the Helmholtz free energy from the OSZICAR file (value following "1 F=" in that file).

The reference energies are simply set in one line as a list for each species, i.e. like the following

ML_EATOM_REF = E_1 E_2 E_3 ...

where E_1, E_2, E_2 etc. are the energies for species 1, 2, 3 etc. (corresponding to the order they occur in the POTCAR file).

The unit of the energies is eV/atom.

Mind: Reference energies are stored in the ML_AB file and are reused whenever the file is read in, i.e, in case of a continued training (ML_MODE = train with ML_AB present), refitting (ML_MODE = refit) or a re-selection run (ML_MODE = select). However, since VASP 6.4.3 the values in the INCAR file take precedence, hence, reference energies from the ML_AB file can be updated by providing new values for the ML_EATOM_REF tag in the INCAR file. In case you are unsure, check the ML_LOGFILE which lists the values actually used.

Related tags and articles