Graph neural network for Hamiltonian-based material property prediction

Development of next-generation electronic devices calls for the discovery of quantum materials hosting novel electronic, magnetic, and topological properties. Traditional electronic structure methods require expensive computation time and memory consumption, thus a fast and accurate prediction model is desired with increasing importance. Representing the interactions among atomic orbitals in material, a Hamiltonian matrix provides all the essential elements that control the structure–property correlations in inorganic compounds. Learning of Hamiltonian by machine learning therefore offers an approach to accelerate the discovery and design of quantum materials. With this motivation, we present and compare several different graph convolution networks that are able to predict the band gap for inorganic materials. The models are developed to incorporate two different features: the information of each orbital itself and the interaction between each other. The information of each orbital includes the name, relative coordinates with respect to the center of super cell and the atom number. The interaction between orbitals is represented by the Hamiltonian matrix. The results show that our model can get a promising prediction accuracy with cross-validation.