Tutorials

Creating (Training) A Machine Learning Potential With Abinit Data

Go to the testing_data directory inside the mlmd directory containing the setup.py file. Inside testing_data there are three directories (mlmd_data_test_abinit , mlmd_data_test_fireball, mlmd_data_test_vasp), go inside the mlmd_data_test_abinit directory, inside this directory, there are three files (input.md, input.stru, training.in), and one directory (dft_data ). To train the machine learning potential, mlmd_ needs (see :ref: chap-mlp) the input.stru file containing the information for the machine learning potential (see :ref: chap-training_file), and the dft_data directory containing the results of DFT calculations performed with abinit.

To train a machine learning potential, execute the command:

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

The calculation may take some time depending on the processing power of the system carrying out the calculation, at the end of the calculation a directory named potential_test, this directory contains the machine learning potential.

Creating (Training) A Machine Learning Potential With Fireball Data

Go to the testing_data directory inside the mlmd directory containing the setup.py file. Inside testing_data there are three directories (mlmd_data_test_abinit , mlmd_data_test_fireball, mlmd_data_test_vasp), go inside the mlmd_data_test_fireball directory, inside this directory, there are three files (input.md, input.stru, training.in), and one directory (dft_data ). To train the machine learning potential, mlmd_ needs (see :ref: chap-mlp) the input.stru file containing the information for the machine learning potential (see :ref: chap-training_file), and the dft_data directory containing the results of DFT calculations performed with fireball.

To train a machine learning potential, execute the command:

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

The calculation may take some time depending on the processing power of the system carrying out the calculation, at the end of the calculation a directory named potential_test, this directory contains the machine learning potential.

Creating (Training) A Machine Learning Potential With Vasp Data

Go to the testing_data directory inside the mlmd directory containing the setup.py file. Inside testing_data there are three directories (mlmd_data_test_abinit , mlmd_data_test_fireball, mlmd_data_test_vasp), go inside the mlmd_data_test_vasp directory, inside this directory, there are three files (input.md, input.stru, training.in), and one directory (dft_data ). To train the machine learning potential, mlmd_ needs (see :ref: chap-mlp) the input.stru file containing the information for the machine learning potential (see :ref: chap-training_file), and the dft_data directory containing the results of DFT calculations performed with vasp.

To train a machine learning potential, execute the command:

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

The calculation may take some time depending on the processing power of the system carrying out the calculation, at the end of the calculation a directory named potential_test, this directory contains the machine learning potential.

Performing Molecuar Dyanmics With A Machine Leaning Potential

Go to the testing_data directory inside the mlmd directory containing the setup.py file. Inside testing_data there are three directories (mlmd_data_test_abinit , mlmd_data_test_fireball, mlmd_data_test_vasp), select whichever of those three as long as you already made a machine learning potential in choosen one.

To perform Molecular Dynamics with mlmd_ you need two files the structure file (:ref: chap-stru_file) and the md file (:ref: chap-md_file), in addition you need a machine learning potential (:ref: chap-mlp).

To make mlmd_ performs Molecular Dynamics execute the command

mlmd_ -mode 'perform_md' -potential_dir 'potential_test' -structure_file 'input.stru' -md_file 'input.md'

The result of the Molecular Dynamics a xyz file with the trayectory of the structure undergoing the Molecular Dynamics.

Calculate The Structural Information Fiter Features Only

To calculate the The Structural Information Fiter Features (SIFF) features only, you can use the Jupyter Notebooks stored in mlmd/testing_data/jupyter_notebooks, there are four notebooks (one per every DFT code supported, and one extra where the inputs are only xyz files).

In case you want to calculate the SIFF features for a structure or many structures without processing them through a DFT code, you can do it by putting the structures in a .xyz file (the energy and temperature can be 0.00, it does not matter for the feature calculation), once the structures are in a .xyz file format you can use the feature_calulation_from_xyz.ipynb notebook to see how to obtain the SIFF features.