Creating (Training) a Machine Learning Potential

The mlmd package contains the mlmd_ script, this script works as an int erface to simplify the usage of the mlmd package. To create a machine learning potential (mlp) you will need:

1. A directory containing the dataset with the output of DFT calculations of either Abinit, Fireball or Vasp.

   • In the case of using Abinit, you will need the .out files.
   • In the case of using Fireball you will need the .xyz and .log files.
   • In the case of using Vasp, you will need the OUTCAR and POSCAR files (For more details see the examples in the tutorial).

2. A training file ( Training File)

To create (train) the potential using the mlmd_ script just type on the terminal:

mlmd_ -mode 'create_potential' -training_file 'path_to_the_training_file’

For example, if the training file with name ‘training.in’ is in the same directory in which mlmd_ is invoked then the creating command is :

mlmd_ -mode 'create_potential' -training_file 'training.in'