Structure to Property: Machine Learning Methods for Predicting Electronic Properties of Crystals

We present a general-purpose machine learning model for predicting properties of crystals. Specifically, energy of formation, Fermi level energy, band gap, partial charges, and bulk modulus as well as spectral properties, including electronic and phonon densities of states are targeted. Thus, our model can be used to screen materials for specific properties. The model is based on atomic representations which enables it to effectively capture complex information about each atom and its surrounding environment in a crystal. The accuracy achieved for band gap values exceeds results previously published. By design, our model is not restricted to electronic properties discussed here but can be extended to fit diverse chemical databases.