ML CTIFOR: Difference between revisions

From VASP Wiki
No edit summary
(Reference to ML_SCLC_CTIFOR added)
 
(30 intermediate revisions by 3 users not shown)
Line 1: Line 1:
{{TAGDEF|ML_FF_CTIFOR|[real]|<math> 10^{-16}</math>}}
{{DISPLAYTITLE:ML_CTIFOR}}
{{TAGDEF|ML_CTIFOR|[real]|<math> 0.002</math>}}


Description: This flag sets the threshold for the Bayesian error estimation on the force in the machine learning force field method.
Description: This flag sets the threshold for the Bayesian error estimation on the force within the machine learning force field method.
----
----
The use of this tag in combination with the learning algorithms is described here: [[Machine learning force field calculations: Basics#Sampling of training data and local reference configurations|here]]. Generally, first principles calculations are only performed if the Bayesian error estimate of one force exceeds the threshold.


During learning, if a newly considered structure yields  a Bayesian  error in the force larger than {{TAG|ML_FF_CTIFOR}}, a first principles calculation is performed, and the corresponding structure is added to the date set of structures that are used when the force field is updated. This flag is only used if {{TAG|ML_FF_IERR}}=2 or 3 (default is 3). The unit of {{TAG|ML_FF_CTIFOR}} is eV/Angstrom. If the tag {{TAG|ML_FF_LCRITERIA}}=''.TRUE.'' is set, {{TAG|ML_FF_CTIFOR}} is automatically updated using past Bayesian error estimates for the forces.
The initial threshold is set to the value provided by the tag {{TAG|ML_CTIFOR}} (the unit is eV/Angstrom). This threshold can be updated dynamically during ML. The details of the update are controlled by {{TAG|ML_ICRITERIA}}. Typically, after extensive training, attainable values for ML_CTIFOR are 0.02 around 300-500 K, and 0.06 around 1000-2000 K, so temperature but also system dependent. The initial default 0.002 is only sensible, if {{TAG|ML_CTIFOR}} is automatically updated ({{TAG|ML_ICRITERIA}} = 1 or 2). If {{TAG|ML_ICRITERIA}} = 0 is used, it is necessary to use significantly larger values around 0.02-0.06 for {{TAG|ML_CTIFOR}}.


For more details see {{TAG| ML_FF_ISAMPLE }}.  
The related tag {{TAG|ML_SCLC_CTIFOR}} determines how many local reference configurations are chosen from each first principles calculations.


== Related Tags and Sections ==
== Related tags and articles ==
{{TAG| ML_FF_ISAMPLE }}, {{TAG|ML_FF_LMLFF}}, {{TAG|ML_FF_IERR}}, {{TAG|ML_FF_CSF}}, {{TAG | ML_FF_LCRITERIA}}, {{TAG | ML_FF_CDOUB}}
{{TAG|ML_LMLFF}}, {{TAG|ML_ICRITERIA}}, {{TAG|ML_SCLC_CTIFOR}} , {{TAG|ML_MHIS}}, {{TAG|ML_CSIG}}, {{TAG|ML_CSLOPE}}, {{TAG|ML_CDOUB}}, {{TAG|ML_CX}}, {{TAG|ML_NMDINT}}, {{TAG|ML_MCONF_NEW}}


{{sc|ML_FF_C[[Category:VASP6]]
{{sc|ML_CTIFOR|Examples|Examples that use this tag}}
TIFOR|Examples|Examples that use this tag}}
----
----


[[Category:INCAR]][[Category:Machine Learning]][[Category:Machine Learned Force Fields]][[Category: Alpha]]
[[Category:INCAR tag]][[Category:Machine-learned force fields]]

Latest revision as of 06:46, 13 October 2023

ML_CTIFOR = [real]
Default: ML_CTIFOR =  

Description: This flag sets the threshold for the Bayesian error estimation on the force within the machine learning force field method.


The use of this tag in combination with the learning algorithms is described here: here. Generally, first principles calculations are only performed if the Bayesian error estimate of one force exceeds the threshold.

The initial threshold is set to the value provided by the tag ML_CTIFOR (the unit is eV/Angstrom). This threshold can be updated dynamically during ML. The details of the update are controlled by ML_ICRITERIA. Typically, after extensive training, attainable values for ML_CTIFOR are 0.02 around 300-500 K, and 0.06 around 1000-2000 K, so temperature but also system dependent. The initial default 0.002 is only sensible, if ML_CTIFOR is automatically updated (ML_ICRITERIA = 1 or 2). If ML_ICRITERIA = 0 is used, it is necessary to use significantly larger values around 0.02-0.06 for ML_CTIFOR.

The related tag ML_SCLC_CTIFOR determines how many local reference configurations are chosen from each first principles calculations.

Related tags and articles

ML_LMLFF, ML_ICRITERIA, ML_SCLC_CTIFOR , ML_MHIS, ML_CSIG, ML_CSLOPE, ML_CDOUB, ML_CX, ML_NMDINT, ML_MCONF_NEW

Examples that use this tag