A parameter-free approach for estimating materials novelty along chemical and structural axes using mutual information-informed density functions. This method computes material density in a way that balances the local and global datastructure to provide a simple means of assessing novelty. The manuscript (NoveltyMatRevC.pdf
) is still a work and progress and some changes are expected as it moves towards publication.
If you have feedback or ideas for improvement, I would love to hear it!
Two Jupyter notebooks demonstrate the application of this method:
-
control_novelty.ipynb
: Demonstrates the method using a control dataset of materials with varying degrees of similarity. Shows how the method distinguishes between different types of novelty. -
Li_novelty.ipynb
: Applies the method to analyze lithium-containing compounds from the GNOME dataset relative to known materials in the Materials Project database.
NOTE that the MP materials are not provided with this repository, but can be downloaded through the get_mp_data.py
file.
The core functionality is provided through three main modules:
distance.py
: Distance calculations using ElMD and LoStOP metricsmi_density.py
: Mutual information analysis and density calculationsutils.py
: Data processing and visualization utilities
@article{falkowski2024mutual,
title={Mutual Information Informed Novelty Estimation of Materials Along Chemical and Structural Axes},
author={Falkowski, Andrew and Sparks, Taylor},
year={2024}
}