ML_REG

From VASP Wiki
(Redirected from ML REGCAR)

This file contains the regression results of the force field training compared to the ab-initio training data in the last training step. This file is written for ML_MODE = train, select.

A sample output should look like the following:

****************************************************************************************************
     Results at         52 MD step.
====================================================================================================
     Total energies (eV)
----------------------------------------------------------------------------------------------------
       -16.991157   -17.523764
       -17.130082   -17.637286
       -17.327379   -17.802581
              ...          ...
====================================================================================================
     Forces (eV ang.^-1)
----------------------------------------------------------------------------------------------------
         1.180622     0.723282
         0.136526     0.018732
        -0.033733     0.103882
         0.888107     0.516528
        -0.148552    -0.401311
         0.159014     0.315306
         0.689279     0.218246
         0.718439     0.750706
              ...          ...
====================================================================================================
     Stress (kbar)
----------------------------------------------------------------------------------------------------
       208.393665   178.750015
        15.421420     8.376722
        -5.264501    -3.403226
              ...          ...

The first column contains the ab-initio data. The second column contains the fitted data.

  • Total energies (eV): The number of lines corresponds to the number of training structures (The number of configurations in the corresponding ML_ABN file).
  • Forces (eV ang.^-1): Contains 3 lines (Cartesian directions) for each atom in each training structure. The direction varies first, then the atoms, and last the training structures.
  • Stress (kbar): Contains 6 lines for the tensor components of each training structure.