NewtonNet

A Newtonian message passing network for deep learning of interatomic potentials and forces

Installation and Dependencies

We recommend using conda environment to install dependencies of this library. Please install (or load) conda and then proceed with the following commands:

conda create --name newtonnet python=3.12
conda activate newtonnet

Now, you can install NewtonNet in the conda environment by cloning this repository:

git clone https://github.com/THGLab/NewtonNet.git

and then runnig the following command inside the NewtonNet repository (where you have access to pyproject.toml):

pip install torch
pip install -e .

Once you finished installations succesfully, you will be able to run NewtonNet modules anywhere on your computer as long as the newtonnet environment is activated. If you have trouble installing torch_geometric, torch_scatter, or torch_cluster, please refer to the PyG documentation page. Optionally, if you want to use Weights & Biases for logging, you can initialize it with

wandb login

Training and Inference

You can find several run files inside the scripts directory that rely on the implemented modules in the NewtonNet library. The run scripts need to be accompanied with a yaml configuration file. You can run an example training script with the following command:

python newtonnet_train.py --config config.yaml

or resume a checkpoint of an interupted training with the following command:

python newtonnet_train.py --resume md17_model/training_1

Optionally for large datasets, you might want to process the data on a CPU node with larger memory using:

python preprocess.py --root md17_data/aspirin/ccsd_train

All models are assumed in ASE units, such as eV and Ang. You can call an ASE calculator from newtonnet.utils.ase_interface. An example MD script can be found in simulate.py.

The documentation of the modules are available at most cases. Please look up local classes or functions and consult with the docstrings in the code.