Hamiltonian Monte Carlo enhanced by Exact Diagonalization
Finn L. Temmen, Martina Gisti, David J. Luitz, Thomas Luu, Johann Ostmeyer · 2026
Read Hamiltonian Monte Carlo enhanced by Exact Diagonalization by Finn L. Temmen, Martina Gisti, David J. Luitz, Thomas Luu, Johann Ostmeyer (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-17788.txt.
Introduction
Temmen and co-authors merge exact diagonalization with Hamiltonian Monte Carlo for fermionic systems. Exact methods scale badly, while Monte Carlo fights sign problems and slow mixing. Their hybrid, H2MC, targets 2D arrays of coupled quantum wires. It scales far past exact diagonalization alone, they report. It also softens the sign problem and cuts autocorrelation times versus plain Monte Carlo. Wider parameter ranges open up for Hubbard chain arrays.
Catalog record
| CREATOR | Finn L. Temmen, Martina Gisti, David J. Luitz, Thomas Luu, Johann Ostmeyer |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 74367 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.17788
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Hamiltonian Monte Carlo enhanced by Exact Diagonalization Finn L. Temmen, Martina Gisti, David J. Luitz, Thomas Luu, Johann Ostmeyer Abstract Strongly correlated fermionic systems are of great interest in condensed matter physics and numerical methods are indispensable tools for their study. However, existing approaches such as exact diagonalization (ED) and stochastic quantum Monte Carlo methods each suffer from fundamental limitations: ED is hindered by exponential scaling in system size, while Monte Carlo methods are plagued by sign problems and long autocorrelation times. These limitations restrict the accessible parameter space and developing algorithms that efficiently alleviate them remains a central challenge in computational physics. In this work, we propose a hybrid algorithm that combines ED and Hamiltonian Monte Carlo (HMC) to simulate 2D arrays of coupled quantum wires, modeled as interacting fermionic Hubbard chains. We demonstrate how our hybrid implementation of HMC, which we dub H 2 MC, outperforms either method alone across several key simulation facets. When compared to pure ED, H 2 MC has a much more favorable computational scaling, which allows us to push simulations to much larger 2D arrays. H 2 MC also greatly alleviates the sign problem and reduces autocorrelation times when compared to pure HMC formulations utilizing either real or imaginary auxiliary fields. Our formalism demonstrates how complementary strengths of seemingly disparate methods can be leveraged to enable feasible simulations in an extended parameter space. I Introduction Higher dimensional quantum many-body systems of strongly correlated fermions give rise to a wide range of collective phenomena that remain difficult to access from first principles, as analytical approaches become intractable in regimes where interactions dominate. Consequently, numerical simulations have become an indispensable tool for extracting quantitative insight into such systems, and the development of efficient and reliable algorithms remains a central challenge in computational physics. Broadly speaking, common numerical approaches can be divided into methods that evaluate many-body quantities directly within the Hilbert space, and stochastic methods that reformulate models in terms of a path integral and evaluate it stochastically using Monte Carlo techniques. Direct methods operate on an explicit representation of the many-body Hilbert space and therefore provide numerically exact results. In particular, exact diagonalization (ED) allows for the computation of observables in the thermodynamic equilibrium up to machine precision. However, the dimension of the Hilbert space grows exponentially with system size, fundamentally limiting the accessible lattice sizes. This scaling barrier, often referred to as the curse of dimensionality , restricts direct approaches to comparatively small systems despite their high accuracy. Tensor network methods (TN) 34 partially alleviate this constraint by exploiting the entanglement structure of quantum states. In some cases (especially in 1D), the ground state of a gapped system can be represented with polynomial rather than exponential effort. Unfortunately, excited state properties like the aforementioned gap are typically much harder to extract from TN simulations 44 . Moreover, while one-dimensional matrix product states (MPS) 41 have been very successful, simulations of 2D systems are significantly more challenging. Projected Entangled Pair State (PEPS) 55 ; 56 ; 11 are the most natural generalisation of MPS to two dimensions and fermionic versions are readily available 12 . Unlike MPS, however, the exact contraction of 2D PEPS, necessary to compute the expectation values, cannot be performed in polynomial time 45 and all practical algorithms must resort to approximations; periodic boundary conditions are then virtually impossible to implement. Isometric tensor network states and their fermionic counterparts have been proposed precisely to circumvent the contraction difficulties, offering a numerically stable subclass of PEPSs that is canonical by construction in 2D 61 ; 13 and 3D 52 ; however, the isometric constraint imposes fundamental limitations on the states expressivity. Alternative 2D ansätze such as tree tensor networks (TTNs) 48 ; 31 and their augmented variants (aTTNs) 42 , which incorporate a layer of disentanglers to better capture the entanglement area law in two dimensions, offer a computationally cheaper alternative to PEPS, albeit these represent a smaller class of quantum states than PEPS. Thus, the simulation of large fermionic 2D systems with a small gap remains very challenging if not infeasible for TNs. Stochastic approaches overcome the scaling limitation by reformulating the evaluation of thermal traces in terms of path integrals, which are then evaluated via stochastic sampling using Monte Carlo integration 38 . Specifically, auxiliary fields are introduced by means of a Hubbard-Stratonovich transformation to decouple many-body interactions, and subsequently sampled using a Markov chain Monte Carlo algorithm. Among these algorithms, the Hamiltonian Monte Carlo (HMC) method 16 (initially hybrid Monte Carlo) enables efficient global updates through molecular dynamics evolution and has long been established as a standard tool in high energy physics and lattice field theory, while increasingly gaining traction in condensed matter applications 6 ; 7 ; 54 ; 47 ; 8 ; 35 ; 36 ; 43 ; 57 . Nevertheless, practical simulations face two central challenges. First, complex (or non-positive) statistical weights introduce noise into the estimation of expectation values, giving rise to the notorious sign problem 53 ; 27 , which causes statistical errors to grow. We remark that formally TNs are not free of a sign problem either 10 , but its severity is typically negligible in practice. Second, potential barriers and multimodal distributions hinder the efficient exploration of the configuration space, leading to large autocorrelation times and in-practice ergodicity violations that can systematically bias results 60 . A natural strategy to mitigate these difficulties is to hybridize direct and stochastic methods, exploiting the complementary strengths of each method. With this approach, one can access system sizes well beyond the reach of full ED and avoid the sampling pathologies that afflict purely stochastic formulations, rendering a larger portion of the parameter space accessible. In this proof of principle work, we employ ED to develop a hybrid HMC (H 2 MC) framework for a two-dimensional system of interacting Hubbard chains. The considered model consists of one-dimensional Hubbard chains coupled via density-density interactions, forming a genuinely two-dimensional geometry in which inter-chain correlations are driven purely by interactions. Two dimensional systems are of particular numerical interest, as they are particularly challenging for all known methods but seem to lie in reach for improvements compared to three dimensional systems. We demonstrate that the hybrid scheme correctly reproduces exact results and yields a significant reduction in both the sign problem and autocorrelation times compared to the fully stochastic treatment, enabling access to regimes that are simultaneously beyond the reach of full ED and problematic for conventional auxiliary field quantum Monte Carlo (AFQMC) approaches. We further incorporate stochastic trace estimators to improve computational scaling of the algorithm, and probe the two-dimensional character of the model on large systems approaching the computational frontier of the method. Related hybrid schemes combining MPS with AFQMC methods have been explored previously 5 , though not in conjunction with the HMC algorithm and its associated advantages. We focus here on the same model as pursued in Ref. 5 for comparability. More broadly, hybridization strategies have recently been pursued for sign problem reduction via MPS trial wavefunctions in AFQMC for fermionic systems 21 . Hybrid quantum-classical approaches utilizing Lanczos and MPS based techniques have similarly been explored for solving electron-phonon systems 30 . The remainder of