Skip to content

Interface for on-the-fly training of Machine Learning Interaction Potentials

Notifications You must be signed in to change notification settings

LAM-GROUP/LASP2Interface

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

LASP2 Interface

This repository provides an interface for 'on the flight' training of a high dimensional neural network. This software is an interface between three programs (n2p2, LAMMPS, and VASP). The LAMMPS software is used for the dynamics of the simulation, using a potential created with the n2p2 software for high dimensional neural network potentials. VASP is the reference method with which the potentials are trained.

Basic usage

LASP2 requires as input configuration files for n2p2, LAMMPS and VASP, as well as a own LASP2 configuration file. The n2p2 input that is required consists of:

  • The n2p2Input directory containing subdirectories for each of the different seeds used. The subdirectories must contain the input.nn and input.data files used for training.
  • The PotentialsComplete directory contanining subdirectories for each of the different seeds used. The subdirectories must contain the files necessary to load an hdnnp potential in LAMMPS.
  • A completeinput.data file with the complete database used for training.

For LAMMPS, the required input consists of:

  • An input file named input.lmp containing the commands to run the desired simulation.
  • A second input file named restart.lmp with instructions that need to be defined again after restarting the simulation. The simulation will stop when training is needed, and it will be restarted after the neural network potential is trained again.
  • The structure used as starting point for the simulation.

In order to use VASP, the interface requires the following input files:

  • A directory named vaspInput containing the INCAR, KPOINTS, and POTCAR files, as well as a copy of the VASP binary.

Output

When the simulation is finished, the interface will produce a plot of the dispersion between the potentials with different random seeds as a function of the steps of the simulation. The output of the VASP calculations will be stored in a directory named DFT, numbered by the number of computations made throughout the computation. Similarly, the output of n2p2 training can be found in the directory named Training, containing subdirectories for the short and long trainings numbered also by the number of trainings performed throughout the simulation. The last potentials used in the simulation can be found in a subdirectory named Potentials under the Training directory.