Element selection for crystalline inorganic solid discovery guided by unsupervised machine learning of experimentally explored chemistry

The selection of the elements to combine delimits the possible outcomes of synthetic chemistry. It determines the range of possible compositions – a phase field – their stable structures, and ultimately, properties, that can arise. For example, in the solid state, the elemental components of a phase field will determine the likelihood of finding a new crystalline material. Researchers make these choices based on their understanding of chemical structure and bonding. Extensive data are available on those element combinations that produce synthetically isolable materials, but it is difficult to assimilate the scale of this information to guide selection from the diversity of potential new chemistries.

In this talk, I will show that unsupervised machine learning (e.g. variational autoencoder) can be used to capture the complex patterns of similarity between element combinations that afford reported crystalline inorganic materials. This model guides prioritisation of quaternary phase fields containing two anions for synthetic exploration to identify lithium solid electrolytes in a collaborative workflow that leads to the discovery of Li3.3SnS3.3Cl0.7 among four other unreported materials.

Finally, we will discuss whether this high level of description of materials – elemental combinations – can be used for the early-stage classification of the materials’ properties.