CGCNN Workflow
CGCNN_VFE is a repository conisting of training and testing data for both Bader charges and vacancy formation energies for each atom within a Ni-Cu-Au-Pd HEA alloys phase space. This includes data from 18 binary constituent alloys, 18 ternary constituent alloys, and 5 quaternary alloys. Updated crystal graph convolutional neural networks (CGCNN) are used to predict Bader charges and vacancy formation energies. Examples for both are found in the '/bc_files/' and '/vfe_files/' folders, respectively. To create new data, use the 'create_model_inputs.ipynb' Notebook in the '/pre_and_post_processing/' folder by loading in a relaxed structure and its Bader charge and vacancy formation energy data.
CHGNet Workflow
CHGNet is also fine-tuned and utilized to relax HEA structures with good accuracy. An example of how to fine-tune CHGNet is found in the '/chgnet_files/' folder. Once structures are relaxed (either via DFT or CHGNet), they can be passed through the saved Bader charge and vacancy formation energy models to obtain both accurately for any given structure.
To install the modified CGCNN model and setup a virtual environment with the appropriate requirements, it is recommended to first set up a virtual environment using Conda:
conda create -n cgcnn_vfe python=3.12
conda activate cgcnn_vfe
Clone the main brach of the repository:
git clone https://github.com/Materials-Computation-Data-Science-MCDC/CGCNN_VFE.git
Or clone a specific branch:
git clone -b <branch_name> https://github.com/Materials-Computation-Data-Science-MCDC/CGCNN_VFE.git
Then move to cgcnnVFE directory and install in editable (-e) mode.
cd cgcnnVFE
pip install -r requirements.txt
pip install -e .
If you use these models or the workflow, please cite the following publication:
Files and Jupyter Notebooks for recreating the figures in the publication are found in the '/pre_and_post_processing/' folder.