Tutorial on NNP MD simulation for methane

This is a tutorial for the following chapter. Please cite the chapter if you follow this tutorial:

Jinzhe Zeng, Liqun Cao, Tong Zhu (2023). Chapter 12 - Neural network potentials. Pavlo O. Dral (Eds.), Quantum Chemistry in the Age of Machine Learning (pp. 279-294), Elsevier.

Preparation

Before starting this tutorial, you need to download and install the following software:

• DeePMD-kit with LAMMPS support
• ReacNetGenerator

It's also recommended to have a GPU environment.

Download this repository:

git clone https://github.com/tongzhugroup/Chapter13-tutorial
cd Chapter13-tutorial

Training the model

Execute

dp train methane_param.json

After the training is completed, freeze and compress the model:

dp freeze -o graph.pb

You will get the model file called graph.pb. We've provided graph.pb in this repository to continue the next step.

Then compress the model:

dp compress -i graph.pb -o graph_compressed.pb -t methane_param.json

Running the simulation

Execute

lmp -in input.lammps

Analyze the simulation

Execute

reacnetgenerator -i methane.lammpstrj --dump -a C H O --nohmm