This package, GCN-HEENG (Graph Convolutional Network-driven High-Entropy surface ENerGy prediction), implements a graph neural network for the prediction of adsorption energies of different gas molecules, herein H and CO, onto high-entropy alloys. It constructs graphs based on the chemical environment in vicinity to the adsorbate. The training is performed in a message passing (convolution) scheme.
The framework and results are comprehensively discussed in my paper.
You can install the prerequisite packages in a Conda environment:
conda create --name=gcn_heeng python=3.9.21
conda activate gcn_heeng
conda install numpy=1.24.4
conda install pandas=2.2.3
conda install pytorch=2.5.1
conda install pytorch_geometric=2.6.1
git clone https://github.com/hanao2/GCN-HEENG.git
cd GCN-HEENG
pip3 install . # add user if you don't have user privilage --user
This project is licensed under the GNU General Public License v3.0. See the LICENSE If you use this project in your research, please cite it as follows:
- Hananeh Oliaei, Narayana R. Aluru. "Study of the adsorption sites of high entropy alloys for CO2 reduction using graph convolutional network" APL Machine Learning 2, 026103 (2024).
The HEA_properties.csv file contains the element intrinsic properties that are used as node features. The training dataset including the alloy structures and adsorption energies (for 'CO' and 'H') are utilized from this paper by Pedersen et al., and can be found through this link.