CGWGAN

Crystal Generative Framework based on Wyckoff Generative Adversarial Network

We present the Crystal Generative Framework based on the Wyckoff Generative Adversarial Network (CGWGAN). CGWGAN utilizes a strategy that focuses on generating crystal templates while effectively masking the occupancy information of elements at specific sites within the crystal structure.

Resources

• Crystal templates: Available on Hugging Face.
• Novel crystal data: Available on Figshare.
• CGWGAN generator: Located in the 'model' folder.
• Atom infill and high-throughput filter: Found in the 'opt_db' folder.

Prerequisites

• Ensure that the following packages are installed: phonopy, pymatgen, ase, and a surrogate model such as m3gnet.

Example Setup

• This example uses m3gnet as the surrogate model.
• Provide the path to the database that stores structures with substituted elements.
• Specify this in the ./opt_db/run_all.py file:

file_path = "path_2_db"
db_path = f"{file_path}/data.db"
cif_processor = CIFProcessor(file_path)
structure_processor = StructureProcessor(file_path, db_path)
cif_processor.process_files()
cif_processor.clean()
structure_processor.process_structures()

Contact Information

• Mr. SU Tianhao
  Email: thsu0407@gmail.com

• Mr. Cao Bin
  Email: bcao686@connect.hkust-gz.edu.cn

Acknowledgement

If you utilize the data or code from this repository, please reference our paper.