Fourier-Transformed Crystal Properties (FTCP)

This software package implements Fourier-Transformed Crystal Properties (FTCP) that is an invertible crystallographic representation, and its associative variational autoencoder (VAE) to inversely design new crystals based on material properties, spanning various chemical compositions and crystal structures (achieving general inverse design).

The package provides functions in three major aspects:

• FTCP Representation: Represent crystals using FTCP representation (in data.py)
• VAE Model with Property-Structured Latent Space: Encode represented crystals into a property-structured latent space using VAE, with a target-learning branch to achieve "property-structure" (in model.py)
• Sampling (Inverse Design): Sample the material-semantic latent space according to design target(s) for new crystals (in sampling.py)

The following paper describes the details of the FTCP representation and framework: An invertible crystallographic representation for general inverse design of inorganic crystals with targeted properties at Matter and arXv

Table of Contents

• Fourier-Transformed Crystal Properties (FTCP)
• How to Cite
• Installation
  • Installation in Google Colab
• Usage
• Reproduce Publication Figures
• Authors

How to Cite

Please cite the following work if you want to use FTCP.

@article{REN2021,
  title = {An invertible crystallographic representation for general inverse design of inorganic crystals with targeted properties},
  author = {Ren, Zekun and Tian, Siyu Isaac Parker and Noh, Juhwan and Oviedo, Felipe and Xing, Guangzong and Li, Jiali and Liang, Qiaohao and Zhu, Ruiming and Aberle, Armin G. and Sun, Shijing and Wang, Xiaonan and Liu, Yi and Li, Qianxiao and Jayavelu, Senthilnath and Hippalgaonkar, Kedar and Jung, Yousung and Buonassisi, Tonio},
  journal = {Matter},
  year = {2021},
  issn = {2590-2385},
  doi = {https://doi.org/10.1016/j.matt.2021.11.032},
  url = {https://www.cell.com/matter/fulltext/S2590-2385(21)00625-1},
}

Installation

To install, clone the repository, navigate to the folder, and use pip install -r requirement.txt

Installation in Google Colab

Run the following code at the very beginning to install relevant packages. No need to run !pip install -r requirement.txt as some packages in the requirement.txt are native in Google Colab, and reinstalling may incur version conflicts.

%tensorflow_version 1.x
!pip install pymatgen==2019.12.22
!pip install matminer==0.6.2
!pip install ase

Usage

Run main.py for ground-state properties, and run main_semi.py for thermoelectric power factor design. The two main scripts go through a typical FTCP inverse design pipeline, where (1) crystals are represented using FTCP representation, (2) a property-structured latent space is obtained by training the VAE + target-learning branch model, and (3) new crystals are designed by sampling the latent space with decoding, and postprocessing.

The package contains the following module and scripts:

Module	Description
main.py	Whole inverse design pipeline (Design Case 2 in paper)
main_semi.py	Semi-supervised-learning inverse design pipeline for thermoelectric power factor (excited-state property) designs (Design Case 3 in paper)
data.py	Data-related functions, including data retrieval from Materials Project, and FTCP representation
model.py	Model-related functions, including setting up the VAE + target-learning branch model
sampling.py	Sampling-related functions, including getting the chemical formulas, checking for compositional uniqueness, and outputting CIFs of designed crystals
utils.py	Script for auxiliary functions
requirements.txt	required packages

Reproduce Publication Figures

Run .py files for respective figures in publication figures.

Authors

The code was primarily written by Zekun Ren and Siyu Isaac Parker Tian, under supervision of Prof. Tonio Buonassisi.