- I-ReaxFF is a differentiable ReaxFF framework based on TensorFlow, with which we can get the first and high order derivatives of energies, and also can optimize ReaxFF parameters with integrated optimizers in TensorFlow.
- ffield,ffield.json: the parameter file for the machine learning potential model.
- train.ipynb: shows a training example of the IRFF-MPNN model with data sets uploaded in the data directory.
- GeomentryOptimization.ipynb: shows how to do geometry optimizations with the IRFF-MPNN model.
- MolecularDynamics.ipynb: shows how to do molecular dynamics with the IRFF-MPNN model.
- StaticCompress.ipynb: shows the computation of the static compression of the solid nitromethane.
(The .py files have the same content as .ipynb files, but supplied as runnable python file).
To run all test files, just using commond like "python test.py".
the following package need to be installed
- TensorFlow, pip install tensorflow --user or conda install tensorflow
- PyTorch, conda install pyTorch
- Numpy,pip install numpy --user
- matplotlib, pip install matplotlib --user
Install this package just run commond in shell "python setup install --user" or "pip install --user irff"
Feng Guo et.al., Intelligent-ReaxFF: Evaluating the reactive force field parameters with machine learning, Computational Materials Science 172 (2020) 109393
Feng Guo et.al., Machine Leaning Potential: A Neural Network Enhanced Version of Reactive Force Field and Message Passing Neural Network, (submitted)