ML_LFAST

From VASP Wiki
Revision as of 14:26, 31 March 2023 by Karsai (talk | contribs)

ML_LFAST = [logical]
Default: ML_LFAST = .FALSE. 

Description: This tag switches on the descriptors for refitting in the fast execution mode within machine learning force fields.

Mind: This tag is only available as of VASP.6.4.0

This tag switches on the descriptors for the fast execution mode. To use the fast execution mode the force field first has to be refit with the fast descriptors. As usual, the resulting ML_FFN has to be copied to ML_FF and the fast code will be automatically run in production mode runs by setting (ML_MODE=RUN or ML_ISTART=2).

To retrain the force field set the following:

or

The speedup of the fast method compared to the regular method is increasing with increasing number of local reference configurations.

It should be noted that in the fast version no Bayesian error estimation is available.

ML_LFAST sets the defaults ML_IALGO_LINREG=4, which is fitting with singular value decomposition and regularization (ML_SIGW0 and ML_SIGV0).

ML_LFAST cannot be used with ML_IALGO_LINREG<3.

Since the calculation time of the fast version is of the same order of magnitude as the timing for the output of the molecular-dynamics results, we advise decreasing the output frequency for molecular dynamics. This is controlled by the tag ML_OUTBLOCK. By default, it writes out all molecular-dynamics results at every molecular-dynamics step. Additionally, the calculation and output of the pair-correlation function is suppressed by ML_OUTPUT_MODE=0 as default.

Related tags and articles

ML_LMLFF, ML_ISTART, ML_IERR, ML_OUTBLOCK, ML_OUTPUT_MODE