Creating binary oxide phase diagrams is a time consuming process both experimentally and computationally, and exploring all of chemical space is a daunting task but may be necessary if we are to discover novel functional materials. This repo contains code to predict oxide stability with errors on par with deep learning methods.
When we mix two elements into an oxide, the two elements can have different oxidation states.
This model attempts to describe how this stabilises binary oxides.
If creating an oxide from elements A–B in stoichiometry (A+B):O=2, for instance, we need to make quadratic equations which fit through the most stable points on the A–O and B–O phase diagrams, where the (0, 0) value is AO2 and BO2, and we need quadratic equations for reduction and oxidation. This general representation can be applied to any element, as we enforce that the average oxidation state of non-oxygen elements in ABO4 be +4, and take a weighted sum of the oxidation and reduction equations for any combination of elements in formation_energy_mixer.ipynb
NOTE the code is very version critical, use python version 3.8.12, other versions can have difficulty using the necessary version of pymatgen. Using conda;
conda create -n oxide_mixer python=3.8.12
Then activate the environment
conda activate oxide_mixer
git clone https://github.com/michaelcraiger/fefos
Then in the environment:
conda install matplotlib==3.2.1 numpy==1.22.3 scipy==1.7.3 pymatgen==2022.4.19 notebook ipykernel
Finally to be able to use the environment in ipykernel
python -m ipykernel install --user --name=oxide_mixer
The directory data_gather contains the code to collect the necessary data from Materials Project in mp_data_save.ipynb. This is saved as pickle files since there contains pymatgen structure objects and they are under unary_oxide_data.p for unary oxide information, ele2gs.p which is the single element energy data, required to calculate formation energies, and the binary oxide data under binary_oxide_data.p. These files are needed to fit the coefficients for the parabolas of each of the 3 considered binary oxide types i.e. MO, M2O3 and MO2-type oxides under the make_quadratics directory. Within data_gather another file which saves all the MO/M2O3/MO2-type binary oxides with the appropriate pairs of elements, and formation energy infomation required by FEFOS, is seen in binary_pairing_data.ipynb and saved as binary_pairing_data.p which is required to plot Figures 2 and 4.
In make_quadratics there is code to save the oxidation and reduction parabola coefficients for each considered element, oxide pair under the supplementary_data/{MO,M2O3,MO2}_parabs.csv files containing these coefficients. Note that these coefficients will depend on the saved values from MP that make up unary_oxide_data.p, so if data in MP changes, these coefficients will change. However, the data contained in this repository will be fixed for the sake of the reproducibility of the results presented in the notebooks in this repo.
roost_test_data contains the csv files required to plot the learning curves for Figure 3 in the manuscript, found in the Goodall, Lee Nature Communications paper under 'Data availability'.
The code to plot Figure 1, 2 and 4 is in the notebook formation_energy_mixer.ipynb as this includes the code which carries out the FEFOS procedure once the parabola coefficients and binary oxide data from Materials Project are saved.