Requests for technical support from the VASP group should be posted in the VASP-forum.

Difference between revisions of "ML FF IERR"

From Vaspwiki
Jump to navigationJump to search
Line 17: Line 17:
[[Category:INCAR]][[Category:Machine Learning]][[Category:Machine Learned Force Fields]]
[[Category:INCAR]][[Category:Machine Learning]][[Category:Machine Learned Force Fields]][[Category: Alpha]]

Latest revision as of 15:16, 29 February 2020

ML_FF_IERR = [integer]
Default: ML_FF_IERR = 3 

Description: This tag selects the error estimation method used in the machine learning force field method.

The error estimation in the machine learning force field method is used when on-the-fly learning is applied to decide whether the next structure needs to be calculated from first principles (and potentially adding it to the reference structure candidates) or using the already existing machine learned force field. The following cases are possible for this tag:

  • ML_FF_IERR=1: Only the spilling factor is used in the error estimation.
  • ML_FF_IERR=2: Only Bayesian error estimation is used.
  • ML_FF_IERR=3: Use both the spilling factor and Bayesian error estimation.

In case three, which is the default, the machine learned force field is only considered as accurate enough if both the Bayesian error (see tag ML_FF_CTIFOR) and the spilling factor (see tag ML_FF_CSF) are below a given threshold.

Related Tags and Sections

Examples that use this tag