Covariance / predict-gibbs-energies

predict-gibbs-energies

Repository associated with Physical descriptor for the Gibbs energy of inorganic crystalline solids and temperature-dependent materials chemistry article used to predict the temperature-dependent Gibbs energies of inorganic crystalline solids.

This repository is a rewritten copy of original repo.

Data files

table-S1.csv

Table of 440 compounds used for training and testing the SISSO-learned descriptor.

masses.json

Dictionary of atomic masses (amu) in format {element : mass}.

gels.json

Dictionary of experimental Gibbs energies (chemical potentials) for the elements in format {temperature (K) : {element : G (eV/atom)}}.

POSCAR.mp-1143_Al2O3

Structure file for Al2O3 from Materials Project to demonstrate use of descriptor.

CLI

CLI is provided for the Gibbs energy approximation. An example usage of it for Al2O3:

./cli.py --formula Al2O3 --H=-3.442 --structure=data/POSCAR.mp-1143_Al2O3

Implementing descriptor

predictor.py

Contains class for implementing SISSO-learned descriptor; see comments within and arXiv link for detailsю

formula.py

Contains chemical formula parser; see comments within for usage examples.

Changelog

• Formula standardizer extracted into separate class, now it is much more time-efficient (from 9x to 28x on different benchmarks) and able to process more complicated formulas;

• Tests provided to cover formula standartization;

• Predictor refactored and type hints provided in accordance with MyPy guidelines;

• Integrational tests for matching results with previous implementation provided;

• CLI interface for predictor provided;

• CI/CD pipelines added for code quality control and testing.

TBD

• Move floating-point calculations to high-precision class, such as Decimal;

• Provide GUI-backed application for the predictor.