Fermionic-propagator and alternating-basis quantum Monte Carlo methods for correlated electrons on a lattice

TL;DR

This paper introduces two novel quantum Monte Carlo methods, FPQMC and ABQMC, for simulating correlated electrons on a lattice, demonstrating their effectiveness in equilibrium and out-of-equilibrium scenarios with different sign problem characteristics.

Contribution

The paper develops and benchmarks two discrete-time QMC methods, FPQMC and ABQMC, for the Hubbard model, with unique real/reciprocal space representations and improved computational efficiency.

Findings

FPQMC has excellent average sign in equilibrium.

ABQMC produces good results despite worse average sign.

ABQMC sign problem is time-independent, unlike FPQMC.

Abstract

Ultracold-atom simulations of the Hubbard model provide insights into the character of charge and spin correlations in and out of equilibrium. The corresponding numerical simulations, on the other hand, remain a significant challenge. We build on recent progress in the quantum Monte Carlo (QMC) simulation of electrons in continuous space, and apply similar ideas to the square-lattice Hubbard model. We devise and benchmark two discrete-time QMC methods, namely the fermionic-propagator QMC (FPQMC) and the alternating-basis QMC (ABQMC). In FPQMC, the time evolution is represented by snapshots in real space, whereas the snapshots in ABQMC alternate between real and reciprocal space. The methods may be applied to study equilibrium properties within grand-canonical or canonical ensemble, external field quenches, and even the evolution of pure states. Various real-space/reciprocal-space correlation functions are also within their reach. Both methods deal with matrices of size equal to the number of particles (thus independent of the number of orbitals or time slices), which allows for cheap updates. We benchmark the methods in relevant setups. In equilibrium, the FPQMC method is found to have excellent average sign and, in some cases, yields correct results even with poor imaginary-time discretization. ABQMC has significantly worse average sign, but also produces good results. Out of equilibrium, FPQMC suffers from a strong dynamical sign problem. On the contrary, in ABQMC, the sign problem is not time dependent. Using ABQMC, we compute survival probabilities for several experimentally relevant pure states.