Deep Neural Networks for Accurate Predictions of Crystal Stability - Unofficial Jupyter Notebook Implementation
This repository contains an unofficial Jupyter Notebook implementation of the research presented in "Deep neural networks for accurate predictions of crystal stability".
Ye, W., Chen, C., Wang, Z., Chu, I. H., & Ong, S. P. (2018). Deep neural networks for accurate predictions of crystal stability. Nature communications, 9(1), 3800.
Predicting the stability of crystals stands at the heart of materials science. Despite the capabilities of approaches like density functional theory (DFT) calculations, they remain resource-intensive and face challenges in scaling with system size. In this notebook, we replicate the promising results achieved using deep neural networks that primarily utilize two descriptors: the Pauling electronegativity and ionic radii. The DFT formation energies of C3A2D3O12 garnets and ABO3 perovskites are predicted with astonishing accuracy. These models hint at a significant stride in machine learning, by showcasing a glimpse into efficient exploration across expansive chemical terrains, possibly unlocking the discovery of groundbreaking materials.
- Python (version, e.g., 3.8+)
- Jupyter Notebook
- numpy, pandas, tensorflow, matplotlib
-
Clone the repository:
git clone https://github.com/shadzzz90/Deep-Neural-Network-for-Predcition-of-Crystal-Stability
-
Navigate to the cloned directory and install the required libraries:
pip install -r requirements.txt
-
Launch Jupyter Notebook:
jupyter notebook
-
Open the provided notebook and run the cells to see the implementation in action.
- Follow the steps within the notebook. Each cell is documented to guide you through the implementation process and to help you understand the methodology.
MIT
This Jupyter Notebook serves as an unofficial implementation inspired by the paper "Deep neural networks for accurate predictions of crystal stability". All credit for the original research, methodologies, and data goes to the authors and associated entities of the paper.