Enhancing the ergodicity of Worldvolume HMC via embedding Generalized-thimble HMC (2508.02659v1)

Abstract: The Worldvolume Hybrid Monte Carlo (WV-HMC) method [arXiv:2012.08468] is an efficient and versatile algorithm that simultaneously mitigates both the sign problem and the ergodicity problem -- the latter being intrinsic to algorithms based on Lefschetz thimbles. We consider a situation in which the maximum flow time can be set to a small value, as occurs when WV-HMC is applied to the doped Hubbard model using a nonphysical redundant parameter. An optimal choice of this parameter significantly reduces the sign problem on the original integration surface and allows the maximum flow time to remain small, a feature that facilitates increasing the system size while keeping the computation time modest. However, as the worldvolume becomes a thin layer, it becomes increasingly difficult to explore it efficiently, leading to potential ergodicity issues. To overcome this limitation, we propose embedding the Generalized-thimble HMC (GT-HMC) into the WV-HMC framework. GT-HMC performs HMC updates on a deformed surface at a fixed flow time. Although it suffers from ergodicity issues due to infinitely high potential barriers at the zeros of the Boltzmann weight, it enables more efficient exploration within the allowed region. Furthermore, its molecular dynamics step size can typically be taken to be larger than in WV-HMC. GT-HMC is thus better suited for sampling regions where ergodicity issues are not serious. We provide a proof that GT-HMC can be embedded within the WV-HMC algorithm, and verify that the two methods -- the pure WV-HMC and the combined version -- yield consistent results within statistical errors for the two-dimensional doped Hubbard model on a $6 \times 6$ spatial lattice at $T/\kappa = 1/6.4\simeq 0.156$ and $U/\kappa = 8.0$ with Trotter number $N_t = 20$ ($\kappa$: hopping parameter).