Semi-empirical Hamiltonian with Machine Learning for accurate and efficient band gap prediction

Given a material, we have developed a method which can reliably and efficiently determine its band gap. This is at a computational cost which is less than that of Density functional theory (DFT) and more accurate. DFT typically underestimates the band gap of materials and improving the DFT predictions requires the use of hybrid DFT and/or GW methods which are compute-intensive. We present an approach which is able to predict reasonably accurate band gap for materials at a cost which is less than that of typical DFT. This technique determines the band gap of materials using a semi-empirical Hamiltonian and then corrects the results via a machine learning (ML) model developed by our group. The efficiency and accuracy of this new approach will be presented for rapid characterization of new materials.