Simple example of creating, training, and using a Machine Learning Potential for Molecular Dynamics

This repo contains an example of creating, training, and using a machine learning potential. For simplicity a 2D Lennard Jones system will be used as the dataset.

The main notebook is simple_mlp.ipynb

The explains the steps and runs all of them.

The scripts it uses are in three separate python files:

• generate.py - creates the dataset
• train_mlp.py - defines the MLP and trains it
• md_with_mlp.py - uses the MLP to run MD

A further notebook: MD_with_autograd/md_with_autograd.ipynb gives an example of using autograd to compute the force from a molecular potential energy function.