AI4Kappa

English | 简体中文

AI4Kappa is a Streamlit application for estimating lattice thermal conductivity (κL) directly from crystal structures in CIF format. The codebase combines CGCNN-based elastic-property inference with two downstream calculation routes:

• KappaP: a Slack-style physical model workflow.
• PINK: a physics-informed workflow that uses predicted elastic quantities and a simplified analytical formula.
• Custom Kappa: a page for user-supplied bulk modulus, shear modulus, and optional Gruneisen parameter.

The public deployment is available at https://kappap-ai.streamlit.app/.

What The Code Actually Does

After a CIF file is uploaded, the application:

1. Converts the uploaded structure to a primitive representation and stores it in root_dir.
2. Builds id_prop.csv for CGCNN inference.
3. Loads the two pretrained models in model one by one to predict:
   • bulk modulus
   • shear modulus
4. Reconstructs crystal descriptors from the primitive structure:
   • number of atoms
   • density
   • volume
   • total atomic mass
5. Computes acoustic quantities from the predicted or user-provided elastic parameters:
   • longitudinal and transverse sound velocity
   • average sound velocity
   • Poisson ratio
   • acoustic Debye temperature
   • Gruneisen parameter
6. Evaluates κL with either the KappaP route or the PINK route at 300 K.

This behavior is implemented across app.py, Pages/KappaP.py, Pages/PINK.py, Pages/CustomKappa.py, streamlit_scripts/file_op.py, streamlit_scripts/calculate_K.py, and predict.py.

Included Pages

• Home: overview, equations, certificate preview, and contact information.
• KappaP: predicts elastic properties from CIF and evaluates the Slack-style expression.
• PINK: predicts elastic properties from CIF and evaluates the physics-informed expression.
• Custom Kappa: lets the user enter elastic parameters manually for up to 5 structures.

Repository Layout

AI4Kappa/
|-- app.py
|-- multipage.py
|-- predict.py
|-- Pages/
|-- cgcnn/
|-- model/
|-- root_dir/
|-- streamlit_scripts/
|-- style/
|-- JMI_Supporting_Information/
`-- KappaP_Supporting_Information/

Important assets:

• model/: pretrained checkpoints for bulk and shear modulus prediction.
• root_dir/atom_init.json: atom embeddings required by CGCNN.
• JMI_Supporting_Information/ and KappaP_Supporting_Information/: tutorial videos and screening result tables distributed with the project.

Installation

The current project is configured around Python 3.9.

git clone --depth=1 https://github.com/Jack-Liu0227/AI4Kappa.git
cd AI4Kappa
python -m venv .venv
.venv\Scripts\activate
python -m pip install --upgrade pip
python -m pip install -r requirements.txt

If you prefer Conda:

conda create -n ai4kappa python=3.9
conda activate ai4kappa
python -m pip install -r requirements.txt

Run Locally

streamlit run app.py

Then open the local Streamlit URL shown in the terminal and upload one or more CIF files from the sidebar.

Inputs And Outputs

Input:

• .cif crystal structure files

Displayed outputs depend on the page, but the tables are built from the following code-level quantities:

• Number of Atoms
• Density (g cm-3)
• Volume (A^3)
• the total atomic mass (amu)
• Bulk modulus (GPa)
• Shear modulus (GPa)
• Sound velocity of the transverse wave (m s-1)
• Sound velocity of the longitude wave (m s-1)
• Speed of sound (m s-1)
• Poisson ratio
• Gruneisen parameter
• Acoustic Debye Temperature (K)
• Kappa_Slack (W m-1 K-1) or Kappa_cal (W m-1 K-1)

Method Notes

• The CGCNN predictor is used here for inference only; training code is not exposed through the Streamlit UI.
• The app currently evaluates thermal conductivity at 300 K in streamlit_scripts/calculate_K.py.
• Custom Kappa accepts at most 5 CIF files in one run.
• Uploaded files are rewritten to primitive structures before most downstream calculations.

Citation

If you use AI4Kappa or the PINK workflow in academic work, cite the paper below in addition to any method-specific references relevant to your study.

@Article{jmi.2024.86,
  AUTHOR = {Yujie Liu and Xiaoying Wang and Yuzhou Hao and Xuejie Li and Jun Sun and Turab Lookman and Xiangdong Ding and Zhibin Gao},
  TITLE = {PINK: physical-informed machine learning for lattice thermal conductivity},
  JOURNAL = {Journal of Materials Informatics},
  VOLUME = {5},
  YEAR = {2025},
  NUMBER = {1},
  ARTICLE-NUMBER = {12},
  URL = {https://www.oaepublish.com/articles/jmi.2024.86},
  ISSN = {2770-372X},
  DOI = {10.20517/jmi.2024.86}
}

Contact

• Prof. Zhibin Gao: zhibin.gao@xjtu.edu.cn
• Yujie Liu: liu_yujie@stu.xjtu.edu.cn