SlaKoNet-VQD: A universal Slater-Koster tight-binding Hamiltonian for variational quantum band-structure calculations on near-term hardware

Abstract

Abstract Variational quantum algorithms such as the variational quantum eigensolver (VQE) and variational quantum deflation (VQD) have emerged as leading candidates for computing electronic structure on near-term quantum hardware. For periodic solids, however, their reach has been limited not by the quantum algorithm itself but by the cost of constructing a faithful second-quantized Hamiltonian for each material, typically via density-functional theory (DFT) followed by Wannierization or by hand-fitted empirical tight-binding parameters. SlaKoNet, a recently developed method, significantly reduces the computational cost of electronic Hamiltonian construction with a small accuracy tradeoff. It integrates modern deep-learning methods and the Slater-Koster tight-binding formalism to fit thousands of onsite and distance-dependent hopping and overlap tight-binding parameters across 65 elements of the periodic table. After fitting, these parameters can be used to deterministically construct an electronic Hamiltonian for any crystal made up of any of the 65 elements in the training set, and the crystal's band structure can subsequently be computed. In this work, we combine a SlaKoNet model trained on the JARVIS-TBmBJ dataset with a Qiskit-based variational quantum deflation algorithm, replacing the traditional and costly Hamiltonian construction step with a universal neural Hamiltonian generator. Within the 65-element set, the resulting SlaKoNet-VQD workflow is structure-agnostic, fully differentiable, fast, and suited to high-throughput screening of band structures. We benchmark the workflow on silicon, recovering the full eight-band band structure along the standard high-symmetry k-path with a mean absolute deviation of 1.78 meV from exact diagonalization on a 3-qubit statevector simulator. We further compute the full band structure for five additional conventional superconductors, Al, Ta, Nb, V, and ZrN, achieving a mean absolute deviation within approximately 1.8 meV, and demonstrate full execution of SlaKoNet-VQD on IBM Quantum hardware for a representative k-point ground-state calculation on aluminum with a mean absolute error of roughly 0.37 eV. Beyond the single-particle level, we promote the same SlaKoNet Hamiltonian to a correlated Hubbard model and solve it with dynamical mean-field theory, recovering the weakening of correlations across the group-5 metals and strong quasiparticle renormalization in the cuprate La₂CuO₄, and we identify the resulting impurity problem as the natural target for a quantum solver. The pipeline opens a path to high-throughput benchmarking of variational quantum algorithms across the periodic table and to gradient-based co-optimization of ansatz and Hamiltonian for materials discovery. An interactive web app is available at https://atomgpt.org/quantum.