This repository contains a collection of notebooks and data that investigate correlations between vibrational properties and diffusive properties in solids. The outputs obtained by myself can be found in the data repository. More background can be obtained from the last Results Chapter of my thesis.

Phonon data is obtained from the phonon calculation database created by Atsushi Togo. Please refer to (https://github.com/WMD-group/phononDB) to query the phonon data.

The notebook 01_get_data.ipynb contains functions required to organise the phonon density of states (DOS) data from PhononDB into a single folder, as well as functions to calculate phonon descriptors obtained from the DOS data: phonon band-centre, relative spread, DOS first peak, and DOS spectrum featurisation.

After running 01_get_data.ipynb a dataframe data.csv, and feature vectors for the DOS (viball_feature.npy, vibli_feature.npy, vibtot_feature.npy) are obtained. data.csv includes composition data as well as descriptor data for each Li-containing material found in the PhononDB database.

Then, in the 02_labelling_conductivies notebook, conductivity labels are added to the dataframe using digitized_data_for_SSEs.csv, a database obtained from (https://github.com/FALL-ML/materials-discovery).

The labelled data is first visualised in 3aa_visualisation.ipynb and fast-ion conducting outliers are identified. Unsupervised clustering is also attempted in 3ab_clustering.ipynb. Different candidates are obtained from the visualisation and the clustering. The candidates are then fully investigated in M3GNet simulations. The results are investigated in 3ac_candidates_investigation.ipynb.

M3GNet simulations were ran for all the materials in the PhononDB database by Kasper Tolborg (data_kasper_full.csv). Direct correlations between phonon descriptors and diffusivity are investigated in 3b_firect_correlations.ipynb.

An attempt to reduce the dimensionality of the DOS feature vectors using a (Variational) AutoEncoder can be found in aa_AE.ipynb and bb_VAE.ipynb.

This work builds upon the work of Amelia Hu realised as part of a UROP summer internship. Her original work can be found at (https://github.com/AmeliaHu0920/urop-project). The internship was co-supervised by Anthony Onwuli and myself.