- Python 100%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
| data | ||
| model | ||
| results | ||
| scripts | ||
| LICENSE | ||
| README.md | ||
| requirements.txt | ||
Struct2Flat
This repository contains the code, datasets, and candidate results for the structure-informed discovery of flat bands in two-dimensional (2D) materials. Our framework integrates a physically motivated flatness score, combination of band dispersion and density of states, with a multi-modal deep learning model trained on atomic structures and compositional descriptors. Without requiring electronic band structure inputs, the model enables scalable and interpretable prediction of flat-band candidates across large 2D material databases.
Features
Physically motivated flatness score combining band dispersion and DOS sharpness Multi-modal model using graph-based encoders and text-based descriptors Fully structure-only prediction — no DFT band structure required at inference time Applied to predict 10,000+ materials from C2DB Includes DFT validation and topological analysis for selected candidates
Repository Structure
Struct2Flat/
data/ # Processed and labeled datasets
model/ # Deep learning architecture
scripts/ # Flatness score computation, kagome-like and atom projection screening
results/ # Candidate summaries and plots
requirements.txt # Environment of the deep learning model
README.md
Data
This repository includes only the processed datasets with algorithm-labeled flatness scores used for model training and inference.
The original raw data can be accessed from the official databases:
2DMatPedia: http://www.2dmatpedia.org
C2DB: https://cmr.fysik.dtu.dk/c2db/c2db.html
Getting Started
-
Clone the repository
cd Struct2Flat
- Set up the environment
pip install -r requirements.txt
- Train the model
All parameters and file paths can be configured in configs.py.
cd model
python train.py
- Predict with the model
python prediction.py
We provided a trained model in results folder.
- Run inference for c2db
python predict_c2db.py
Results
Over 10,000 2D materials screened
Selected candidates form a new flat-band database
Several top candidates validated with DFT and topological quantum chemistry
Results and figures from the paper are available under /results/.