ML WTOTEN: Difference between revisions

From VASP Wiki
No edit summary
No edit summary
 
(6 intermediate revisions by 2 users not shown)
Line 1: Line 1:
{{DISPLAYTITLE:ML_WTOTEN}}
{{TAGDEF|ML_WTOTEN|[real]}}
{{TAGDEF|ML_WTOTEN|[real]}}
{{DEF|ML_WTOTEN|0.005|if {{TAG|ML_IWEIGHT}}{{=}}1|1.0|otherwise}}
{{DEF|ML_WTOTEN|0.005|if {{TAG|ML_IWEIGHT}}{{=}}1|1.0|otherwise}}


Description: This tag sets the weight for the scaling of the total energy in the training data within the machine learning force field method.
Description: Sets a scaling of the fitted potential energy.  
----
----


{{TAG|ML_IWEIGHT}}, {{TAG|ML_WTOTEN}}, {{TAG|ML_WTIFOR}}, {{TAG|ML_WTSIF}} form a group of tags which set the normalization and weighting of ab initio training data, i.e.  energies, forces and stresses of the training structures. The main control tag is {{TAG|ML_IWEIGHT}}, please also have a look at its detailed description. If {{TAG|ML_IWEIGHT}}=1 the weight has a unit of eV/atom and is used to divide the data by it. For {{TAG|ML_IWEIGHT}}=2 and 3 the weights are unitless and the data is multiplied by the weight.
For {{TAG|ML_IWEIGHT}}=2 and 3 (default), the potential energy in the training data set is multiplied by {{TAG|ML_WTOTEN}} (unitless). We recommend increasing {{TAG|ML_WTOTEN}} if you plan to apply the force field in a simulation where the accuracy of the total energy is most important. This puts a focus on the energy error and is desirable, for instance, for the computation of defect-formation energies.


== Related Tags and Sections ==
For {{TAG|ML_IWEIGHT}}=1, {{TAG|ML_WTOTEN}} has the unit of eV/atom, and the potential energy in the training data set is divided by it.
{{TAG|ML_LMLFF}}, {{TAG|ML_IWEIGHT}}, {{TAG|ML_WTIFOR}}, {{TAG|ML_WTSIF}}, {{TAG|ML_IALGO_LINREG}}
== Related tags and articles ==
{{TAG|ML_IWEIGHT}}, {{TAG|ML_WTIFOR}}, {{TAG|ML_WTSIF}}, {{TAG|ML_IALGO_LINREG}}, {{TAG|ML_LMLFF}}


{{sc|ML_WTOTEN|Examples|Examples that use this tag}}
{{sc|ML_WTOTEN|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 08:17, 3 October 2023

ML_WTOTEN = [real] 

Default: ML_WTOTEN = 0.005 if ML_IWEIGHT=1
= 1.0 otherwise

Description: Sets a scaling of the fitted potential energy.


For ML_IWEIGHT=2 and 3 (default), the potential energy in the training data set is multiplied by ML_WTOTEN (unitless). We recommend increasing ML_WTOTEN if you plan to apply the force field in a simulation where the accuracy of the total energy is most important. This puts a focus on the energy error and is desirable, for instance, for the computation of defect-formation energies.

For ML_IWEIGHT=1, ML_WTOTEN has the unit of eV/atom, and the potential energy in the training data set is divided by it.

Related tags and articles

ML_IWEIGHT, ML_WTIFOR, ML_WTSIF, ML_IALGO_LINREG, ML_LMLFF

Examples that use this tag