Investigating the Fermi-Hubbard model by the Tensor-Backflow method (2507.01856v1)

Abstract: Recently, a variational wave-function based on the tensor representation of backflow corrections has achieved state-of-the-art energy precision in solving Fermi-Hubbard-type models. However, the Fermi-Hubbard model is very challenging to solve, and the validity of a method relies on investigating the ground state's physical property. For simplicity, we name the tensor representation of backflow corrections as the Tensor-Backflow in this work. We apply the Tensor-Backflow method to investigate the Fermi-Hubbard model on two-dimensional lattices as large as 256 sites, under various interaction strengths $U$, electron fillings $n$ and boundary conditions. Energy precision can be further improved by considering more backflow terms, such as considering backflow terms from next-nearest-neighbours or from all sites. Energy extrapolations on 64-site lattices give competitive results to the gradient optimized fPEPS with the bond dimension as large as $D$=20. For cases of $n$=0.875 and $U$=8 on the $16\times 16$ lattice under open boundary condition, by considering nearest-neighbour backflow terms, obtained energy is only $4.5\times 10^{-4}$ higher than the state-of-the-art method fPEPS with the bond dimension $D$=20. For periodic boundary condition, the variational wave-function is not enforced on any prior symmetry, meanwhile linear stripe order is successfully obtained. Under the same filling and $U$=10,12, energies obtained from initializations with the obtained wave-function for $U$=8 are lower than that from direct optimizations, meanwhile energies are competitive to or even better than state-of-the-art tensor network methods. For cases of $n$=0.8 and 0.9375, results consistent with the phase diagram from AFQMC are obtained by direct optimizations.