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

ML FF ISAMPLE

From Vaspwiki
Jump to navigationJump to search

ML_FF_ISAMPLE = [integer]
Default: ML_FF_ISAMPLE = 3 

Description: This tag controls the sampling in the machine learning force field method.


The following values can be set for ML_FF_ISAMPLE:

  • ML_FF_ISAMPLE=1: If the estimated error is larger than the pre-determined threshold (ML_FF_CSF for the spilling factor and ML_FF_CTIFOR or it's updated value form ML_FF_LCRITERIA=.TRUE. for the Bayesian error), the data is added to the candidate dataset. If the number of the data in the candidate data set reaches ML_FF_MCONF_NEW the sampling is stopped and a new force field is generated.
  • ML_FF_ISAMPLE=2: If the estimated error is ML_FF_CDOUB times larger than the threshold (ML_FF_CSF for the spilling factor and ML_FF_CTIFOR or it's updated value form ML_FF_LCRITERIA=.TRUE. for the Bayesian error), the sampling is stopped and a new force field is generated based on the new data. Other parts are the same as in the case of ML_FF_ISAMPLE=1.
  • ML_FF_ISAMPLE=3: If the estimated error is ML_FF_CDOUB times larger than the threshold (ML_FF_CSF for the spilling factor and ML_FF_CTIFOR or it's updated value form ML_FF_LCRITERIA=.TRUE. for the Bayesian error) the sampling is stopped and a new force field is generated. Otherwise the program enforces the machine not to sample the ab initio data during ML_FF_NMDINT MD steps from the previous sampling step. After the ML_FF_NMDINT MD steps, the sampling is executed as in the case of ML_FF_ISAMPLE=1. This is the default method if nothing is specified.

Related Tags and Sections

Examples that use this tag


ML_FF_LMLFF,ML_FF_MCONF_NEW,ML_FF_CDOUB,ML_FF_NMDINT