Machine learning lattice constants for spinel compounds

Abstract

Spinels can house a large variety of elements into the crystal structure. As a crystallographic parameter, the lattice constant, a, is highly sought in further investigations into materials properties. Experimental approaches to obtain the lattice constant are resource-intensive, and limit the exploration into non-synthesized spinels. Here, we develop the Gaussian process regression model to shed light on the relationship among ionic radii, electronegativities, and lattice constants. 167 samples with lattice constants between 8.044 Å and 11.660 Å are investigated. The model provides more accurate predictions than previous studies based on linear regressions, and statistical relationships between descriptors and the target.