Efficient Simulation of the 2D Hubbard Model via Hilbert Space-Filling Curve Mapping
Ashkan Abedi, Vittorio Giovannetti, Dario De Santis · 2025
Read Efficient Simulation of the 2D Hubbard Model via Hilbert Space-Filling Curve Mapping by Ashkan Abedi, Vittorio Giovannetti, Dario De Santis (2025) 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-2512-02666.txt.
Introduction
Ashkan Abedi, Vittorio Giovannetti, and Dario De Santis simulate a classic model of electrons on a square lattice. The two dimensional Hubbard model captures how electrons hop and repel, a core puzzle in condensed matter. Tensor network methods must first flatten the square into a chain.
The team maps the lattice with a Hilbert curve, a space filling path that keeps nearby sites close. That compact mapping yields a smaller matrix product state than the usual snake path. Cheaper simulations let physicists probe larger systems in the search for new phases of matter. Flat maps that keep near sites near are a cheap win, and cheap wins stack up in long runs fast.
Catalog record
| CREATOR | Ashkan Abedi, Vittorio Giovannetti, Dario De Santis |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-02666
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Efficient Simulation of the 2D Hubbard Model via Hilbert Space-Filling Curve Mapping Preprint: APS/123-QED Ashkan Abedi, Vittorio Giovannetti, Dario De Santis Abstract We investigate tensor network simulations of the two-dimensional Hubbard model by mapping the lattice onto a one-dimensional chain using space-filling curves. In particular, we focus on the Hilbert curve, whose locality-preserving structure minimizes the range of effective interactions in the mapped model. This enables a more compact matrix product state (MPS) representation compared to conventional snake mapping. Through systematic benchmarks, we show that the Hilbert curve consistently yields lower ground-state energies at fixed bond dimension, with the advantage increasing for larger system sizes and in physically relevant interaction regimes. Our implementation reaches clusters up to 32 × 32 32\times 32 sites with open and periodic boundary conditions, delivering reliable ground-state energies and correlation functions in agreement with established results, but at significantly reduced computational cost. These findings establish space-filling curve mappings, particularly the Hilbert curve, as a powerful tool for extending tensor-network studies of strongly correlated two-dimensional quantum systems beyond the limits accessible with standard approaches. I Introduction The Hubbard model stands as one of the most fundamental paradigms in condensed matter physics, providing an essential framework for understanding strongly correlated electronic systems. Originally proposed by Hubbard in 1963 17 , this deceptively simple model captures the essential physics of electrons moving on a lattice with local Coulomb repulsion, yet exhibits an extraordinarily rich phase diagram encompassing metal-insulator transitions, magnetism, and potentially high-temperature superconductivity 1 ; 32 . Despite decades of intensive research, the two-dimensional (2D) Hubbard model continues to resist a complete analytical solution, making numerical approaches indispensable for understanding its physics. The computational challenge of the 2D Hubbard model derives from the exponential growth of the Hilbert space with system size, making exact diagonalization feasible only for small clusters. Various numerical methods have been developed to tackle this challenge, each with inherent strengths and limitations. Quantum Monte Carlo (QMC) methods 25 ; 39 provide numerically exact results but suffer from the notorious sign problem away from half-filling and at low temperatures, a phenomenon in which the Monte Carlo weights become negative or complex, leading to exponential growth in statistical errors and rendering calculations intractable. Dynamical Mean Field Theory (DMFT) and its cluster extensions 12 ; 4 capture local correlation effects accurately, but may miss important spatial fluctuations and long-range order. Variational methods including neural network quantum states 26 ; 5 have shown promise but require careful optimization and may suffer from bias in the variational ansatz. Tensor network methods, particularly the Density Matrix Renormalization Group (DMRG) 38 ; 34 , have revolutionized the study of one-dimensional quantum systems by exploiting the area law of entanglement entropy 15 ; 10 . This motivates our approach of employing efficient one-dimensional Matrix Product State (MPS) algorithms through a suitable mapping. In particular, by projecting the 2D system onto a 1D chain via space-filling curves, the nearest-neighbor interactions are converted into structured long-range couplings that can still be effectively treated within 1D tensor network algorithms 6 . The challenge with naive 2D-to-1D mappings is that they can introduce very long-range interactions in the 1D representation. A recent study by Baldelli et al. 2 pointed out that a straightforward snake-like mapping leads to long-range hopping terms that cause an exponential growth in computational cost with lattice width, limiting traditional DMRG to very narrow cylinders - quasi-2D geometries where a 2D lattice of width W and length L\gg W is treated with periodic (cylinder) or open (strip) boundary conditions in the width direction (typically width 6-8). This fundamental limitation motivates the search for more locality-preserving mappings that can maintain the efficiency of 1D tensor network methods while tackling truly 2D systems. Recent advances in tensor network algorithms have further enhanced their applicability to 2D systems. The development of improved optimization schemes 36 , and better handling of symmetries 19 have pushed the boundaries of accessible system sizes and accuracy. Furthermore, the integration of machine learning techniques with tensor networks 23 has opened new avenues for efficient representation and optimization of quantum states. The choice of mapping from 2D to 1D is crucial for the efficiency of tensor network simulations. Cataldi et al. 6 demonstrated that the Hilbert curve 16 mapping significantly outperforms conventional approaches for the 2D quantum Ising model. The Hilbert curve, a continuous fractal space-filling curve, possesses unique locality-preserving properties that minimize the average distance between neighboring sites after mapping, resulting in more efficient tensor network representations. Furthermore, a recent study by Bellwood et al. 3 examined a wide range of 2D-to-1D path mappings and confirmed that fractal space-filling curves like the Hilbert curve generally yield faster convergence and higher accuracy than the standard snake path, with the benefit increasing for larger systems. They propose methods to construct optimal paths by tiling smaller fractal segments, further validating the approach we adopt here. In this work, we extend and optimize the Hilbert curve mapping approach for the fermionic 2D Hubbard model, which presents additional challenges due to the presence of both spin and charge degrees of freedom and the need to properly handle fermionic statistics. To our knowledge, this is the first application of the Hilbert curve mapping to a fermionic system with spin-up and spin-down degrees of freedom, extending beyond the spin models studied in previous works 6 . We perform comprehensive benchmarking against the conventional snake mapping and other recent approaches, demonstrating substantial improvements in both accuracy and computational efficiency. Our key contributions include: (i) systematic analysis of how the Hilbert curve preserves locality in the fermionic Hubbard model, (ii) quantitative comparison of convergence rates between different mappings for systems up to 32\times 32 sites, and (iii) demonstration that the Hilbert mapping enables more accurate simulations in parameter regimes where traditional approaches struggle. The remainder of this paper is organized as follows. In Section II , we describe the space-filling curve mappings and a brief introduction to the MPS formalism and the numerical methods we used in our simulations. Section III presents our numerical results, comparing ground-state energies and convergence properties for different system sizes and parameter regimes, studying both half-filled and doped cases with a particular focus on the intermediate coupling regime. Finally, Section IV summarizes our findings and discusses future directions for applying space-filling curve mappings to other strongly correlated systems. Our results establish the Hilbert curve mapping as a powerful tool for studying 2D strongly correlated systems, offering a favorable balance between accuracy and computational cost. This approach opens new possibilities for investigating the phase diagram of the Hubbard model and related models in parameter regimes that were previously inaccessible to accurate numerical treatment. II Preliminaries II.1 Space-Filling Curves Mapping a two-dimensional lattice onto a one-dimensional chain is a prerequisite for applying highly efficient tensor-network algorithms such as the MPS ansatz. The choice of mapping, however, is not merely a technical detail: it determines how local couplings in the original lattice are translated into interactions in the effective 1D Hamiltonian. Poor mappings can transform short-range 2D couplings into long-range 1D terms, increasing entanglement across MPS bonds and inflating the bond dimension required for accurate simulations. Space-filling curves provide a systematic class of mappings that traverse every