TorchMD Autodiff Example

Introduction

This example demonstrates the benefits of automatic differentiation within a molecular dynamics engine. We first simulate a waterbox and frequently save the coordinates and velocities. Then we try to infer the force field parameters from this trajectory.

Requirements

• torchmd
• tqdm
• torch (tested with version 1.4.0)
• numpy
• matplotlib

Scripts and Files

• waterbox.py -- Definition of the molecular system based on files in the torchmd test suite.
• simulate.ipynb -- Jupyter notebook for simulating the water box.
• learn.ipynb -- Jupyter notebook for learning the parameters from the trajectory.
• integrator.py -- A version of the torchmd integrator without inplace operations
• plot.ipynb -- plotting script
• writeup.ipyng -- for the paper