Research Area
First-principles methods deliver accuracy but scale poorly; empirical tight-binding scales well but sacrifices transferability. DeePTB closes this gap by learning the tight-binding Hamiltonian directly from ab initio data with deep neural networks, preserving symmetry and locality by construction. The result is electronic structure prediction at ab initio accuracy for systems far beyond the reach of conventional DFT, including finite-temperature and large-scale disordered structures.
Publications in this area
IEEE MoSICom 2025, BITS Pilani Dubai Campus
Zou J., Zhouyin Z., Gu Q., Pandey S.K.
Nature Communications, 15 (2024)
Gu Q., Zhouyin Z., Pandey S.K., Zhang P., Zhang L., E W.
ICLR 2025 (Spotlight)
Zhouyin Z., Gan Z., Pandey S.K., Zhang L., Gu Q.
Other research areas
Exploring spin-orbit assisted Mott insulators and 3d/4d/5d transition metal compounds as QSL candidates using ab initio and renormalization group methods.
Investigating cuprate analogues like AgF₂ using random phase approximation and fluctuation exchange to identify high-Tc unconventional superconductors.
Using dynamical mean-field theory (DMFT) to study how dynamic electronic correlation enhances orbital magnetization in layered materials.
Studying displacive topological phase transitions and hidden-order phases in Weyl semimetals via first-principles and effective theory.
Combining deep learning tight-binding Hamiltonians with non-equilibrium Green’s functions to simulate quantum transport in 2D materials at scale.