ML phonon calculation scripts

This repository is a collection of scripts for running phonon calculations using machine-learned interatomic potentials (MLIP). It includes scripts for the entire pipeline, from converting CIF files to something more useable, to the phonon calculations themselves.

All the scripts are based on top of the janus-core package.

The pipeline consists of the scripts in the following order:

1. scrape_structures.py for scraping the CSD database to obtain CIF files of the relevant structures.
2. convert_to_xyz.py for converting CIF files obtained from crystal structure databases such as CDS/ICSD/etc. into the much more computer-friendly (and phonopy compatible) VASP POSCAR format.
3. reduce_cell.py for using phonopy to reduce each system into its primitive cell
4. optimise.py for optimising the structure using a given MLIP
5. run_phonon.py for running the phonon calculations on the optimised structure
6. analyse_phonons.py for checking whether phonon calculations concluded successfully
7. plot_abins.py for computing INS spectra from the phonon calculations and plotting the comparison to experimental spectra
8. build_result_db.py for gathering the results from one or multiple runs into csv files. However, note that this script has to be run twice:
   • first normally, which gathers data from MLIP runs
   • second with --update, which adds extra information about systems from the CSD database

Additionally, tools for analysing the results obtained from build_result_db.py are provided in data_analysis.py. Furthermore, example data analyses on our data can be found in the jupyter notebooks:

• data_analysis_full_dataset.ipynb
• data_analysis_reduced_dataset.ipynb