ML_CALGO

ML_CALGO = [integer]
Default: ML_CALGO = 0

Description: Chooses error estimation type for on-the-fly training or reselection of local referenc configurations.

This tag chooes which algorithm is employed for the error estimation in ML_MODE=TRAIN or SELECT. The following two choices are available:

ML_CALGO=0: Bayesian error estimation. Constant or variable threshold. Default.
ML_CALGO=1: Spilling factor. Constant threhold.

In both modes an ab-initio calculation is carried out if the value of the error estimate is above a threshold specified by ML_CTIFOR. In both algorithms the estimators have different units, values and hence defaults for this threshold. In contrast to the Bayesian error estimation which can be run in many different modes for the threhold update (see ML_ICRITERIA), the spilling factor can only be used with a constant threshold (ML_ICRITERIA=0).

Related tags and articles