FerroGen

FerroGen consists of three bash shell scripts for sequentially screening diffusion-model–generated inorganic crystal structures to identify promising ferroelectric candidates. This is a multi-stage filtering pipeline that (1) applies symmetry filtering by polar space groups using FINDSYM, (2) computes thermodynamic stability via oxidation state and energy above hull using SevenNet-0, (3) estimates the band gap (a key ferroelectric photovoltaic property) using CGCNN. The workflow is managed by these bash shell scripts, while specific programs (e.g., FINDSYM, SevenNet-0, CGCNN) invoke helper Python scripts where provided.

Prerequisites

The following Python packages are required:

python>=3.10
numpy
pandas
torch    # PyTorch
ase>=3.25.0
pymatgen>=2025.4.24
sevenn>=0.10.4

Publication

Yeo, B. C., Kang, S., & Lee, J.-H. (2025). Diffusion–Model–Driven Discovery of Ferroelectrics for Photocurrent Applications [Working paper]. ChemRxiv. https://chemrxiv.org/engage/chemrxiv/article-details/68d1110df416303770b9e33c
Yeo, B. C., Kang, S., & Lee, J.-H. (2025). Diffusion–Model–Driven Discovery of Ferroelectrics for Photocurrent Applications [Data set]. Zenodo. https://doi.org/10.5281/zenodo.17223175

How to use

1.screening_for_symmetry.sh

• Input:
  • generated_materials_cif/ — folder of structures generated by the diffusion model (Zenodo dataset)
• Process:
  • Uses FINDSYM to analyze the symmetry of the generated structures
• Outputs:
  • 1.cif/ — folder containing symmetry-processed CIF structures
  • 2.cif/ — folder containing only polar space group CIF structures (output from step 1)
  • result_space_group.txt — list of identified space groups
  • result_polar_group.txt — list of structures belonging to polar space groups

2.screening_for_stability.sh

• Input:
  • 2.cif/ — folder containing only polar space group CIF structures (output from step 1)
• Process:
  • Performs oxidation state validation
  • Relaxes structures using SevenNet-0 to compute thermodynamic stability (energy above hull)
• Outputs:
  • result_oxidation_state.txt — file summarizing oxidation state–validated results
  • 4.cif/ — folder containing structures relaxed with SevenNet-0

3.screening_for_band_gap_and_energy_above_hull.sh

• Input:
  • 4.cif/ — folder containing structures relaxed with SevenNet-0 (output from step 2)
• Process:
  • Uses CGCNN to predict band gaps
  • Computes energy above hull values
  • Assigns new sequential numbering (1, 2, 3, …) to the processed structures
• Outputs:
  • 5.cif/ — folder containing candidate CIF structures, renamed sequentially (1, 2, 3, …)
  • result_final_all.txt — file summarizing predicted band gaps (from CGCNN) and energy above hull results