Graph neural network for Hamiltonian-based material property prediction

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