Machine learning method for tight-binding Hamiltonian parameterization from ab-initio band structure

Abstract The tight-binding (TB) method is an ideal candidate for determining electronic and transport properties a large-scale system. It describes the system as real-space Hamiltonian matrices expressed on manageable number of parameters, leading to substantially lower computational costs than ab-initio methods. Since whole defined by parameterization scheme, choice TB parameters decides reliability calculations. typical empirical uses directly from existing parameter sets, which hardly reproduces desired structures quantitatively without specific optimizations. thus not suitable quantitative studies like property derives results through transformation basis functions, achieves much higher numerical accuracy. However, it assumes prior knowledge may encompass truncation error. Here, machine learning proposed, within neural network (NN) introduced with its neurons acting matrix elements. This can construct model that given energy bands predefined accuracy, provides fast convenient way construction gives insights into applications in physical problems.