Time evolution of the quantum Ising model in two dimensions using Tree Tensor Networks
Wladislaw Krinitsin, Niklas Tausendpfund, Markus Heyl, Matteo Rizzi, Markus Schmitt · 2025
Read Time evolution of the quantum Ising model in two dimensions using Tree Tensor Networks by Wladislaw Krinitsin, Niklas Tausendpfund, Markus Heyl, Matteo Rizzi, Markus Schmitt (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-2505-07612.txt.
Introduction
Wladislaw Krinitsin and colleagues simulated a two-dimensional quantum magnet out of equilibrium in 2025. Tracking many interacting spins over time in two dimensions defeats most methods, since entanglement grows fast. Tree tensor networks, which organize quantum states on a branching graph, offer a middle path between accuracy and cost, and the team tested how far they reach on the quantum Ising model.
For weak transverse fields the method reproduced known analytic results faithfully on lattices up to a moderate size. In stronger quench settings, rapid entanglement growth set clear limits, which the benchmarks map honestly. Graphics processor acceleration and local operator sums both helped performance. Simulator builders gain a calibrated sense of when tree networks suffice.
Catalog record
| CREATOR | Wladislaw Krinitsin, Niklas Tausendpfund, Markus Heyl, Matteo Rizzi, Markus Schmitt |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-07612
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Time evolution of the quantum Ising model in two dimensions using Tree Tensor Networks Wladislaw Krinitsin, Niklas Tausendpfund, Markus Heyl, Matteo Rizzi, Markus Schmitt Abstract The numerical simulation of two-dimensional quantum many-body systems away from equilibrium constitutes a major challenge for all known computational methods. We investigate the utility of Tree Tensor Network (TTN) states to solve the dynamics of the quantum Ising model in two dimensions. Within the perturbative regime of small transverse fields, TTNs faithfully reproduce analytically known, but non-trivial and physically interesting dynamics of magnetic domains, for lattices up to 16 × 16 16\times 16 sites. Limitations of the method related to the rapid growth of entanglement entropy are explored within more general, paradigmatic quench settings. We provide and discuss comprehensive benchmarks regarding the benefit of GPU acceleration and the impact of using local operator sums on the performance. I Introduction Tensor network algorithms serve as potent and versatile numerical methods in many-body quantum physics, enabling the computation of static and dynamic properties of generic lattice systems Orús 2014 ; Paeckel et al. 2019 ; Silvi et al. 2019 ; Cirac et al. 2021 . These algorithms achieve efficiency by compressing quantum states into a network of low-rank tensors with bonds of a fixed dimension, turning the exponential growth of the Hilbert space with the system size into a polynomial one. The limitations of such a compressed representation are well understood – it is constrained by the states’ entanglement complexity. For one-dimensional, local and gapped systems the entanglement entropy follows the boundary law, enabling ground state search using the highly successful D ensity M atrix R enormalization G roup (DMRG) algorithm White 1992 . Immediate generalizations to higher dimensions suffer from higher algorithmic complexity and typically less favorable entanglement scaling. Nonetheless, recent advances in the area of believe propagation Tindall and Fishman 2023 ; Tindall et al. 2024 and P rojected E ntangled P air S tates (PEPS) Verstraete and Cirac 2004 ; Jordan et al. 2008 ; Lubasch et al. 2014a ; Lubasch et al. 2014b ; Corboz 2016 ; Vanderstraeten et al. 2016 ; Zaletel and Pollmann 2020 ; Puente et al. 2025 provide a promising avenue for efficiently representing ground states in higher dimensional quantum systems Murg et al. 2007 ; Murg et al. 2009 ; Niesen and Corboz 2017 ; Liao et al. 2017 ; Chen et al. 2018 ; Hasik et al. 2022 ; Hasik and Corboz 2024 ; Weerda and Rizzi 2024 . With N eural Q uantum S tates (NQS) Carleo and Troyer 2017 , an alternative family of variational states emerges, whose neural network representation of the wave function offers higher flexibility concerning the lattice geometry of the physical system. While their capability to capture volume law entanglement Deng et al. 2017 ; Levine et al. 2019 ; Denis et al. 2025 and first results for the dynamics of two dimensional quantum systems Schmitt and Heyl 2020 ; Gutiérrez and Mendl 2022 ; Schmitt et al. 2022 ; Sinibaldi et al. 2023 ; Mendes-Santos et al. 2023 ; Mendes-Santos et al. 2024 ; de Walle et al. 2024 ; Sinibaldi et al. 2024 outline a promising avenue, their limitations and shortcomings are not yet fully understood. Therefore, tackling dynamics in two-dimensional systems remains a major challenge for all computational methods. On the tensor network side, a less explored architecture are so called T ree T ensor N etworks (TTNs) – their hierarchical structure can be exploited to respect the local nature of interactions present in typical Hamiltonians, while still remaining loop-free and generalizable to lattices of any dimension. So far they have been successfully applied to the ground state search Tagliacozzo et al. 2009 ; Murg et al. 2010 ; Ferris 2013 ; Schuhmacher et al. 2025 and recently also used to tackle questions related to the dynamics of two-dimensional quantum systems Schröder et al. 2019 ; Kloss et al. 2020 ; Pavešić et al. 2024 ; Jaschke et al. 2024 . In the following we expand upon these results, focusing on the example of the quantum Ising model, which serves as a paradigmatic model in the study of quantum phase transitions and other collective phenomena in and out of equilibrium. In recent years, it has become highly relevant as the effective model describing the main interactions in Rydberg atom arrays Greiner et al. 2002 ; Kinoshita et al. 2006 ; Hild et al. 2014 ; Georgescu et al. 2014 ; Martinez et al. 2016 ; Choi et al. 2016 ; Gross and Bloch 2017 ; Jurcevic et al. 2017 ; Bernien et al. 2017 ; Zhang et al. 2017 ; Gärttner et al. 2017 ; Choi et al. 2017 ; Levine et al. 2018 ; Barredo et al. 2018 . With local addressibility becoming an increasingly common capability of such quantum simulators, specific inhomogeneous product states can be prepared in order to observe their relaxation dynamics Manovitz et al. 2024 . This opens the possibility to investigate interesting theoretical predictions. In particular, we will address various domain wall initial conditions, which exhibit different inhibiting mechanisms, that result in unusually slow thermalization. On the one hand, approximate dynamical constraints lead to Hilbert space fragmentation in the perturbative limit of small transverse fields, leading to effecive Wannier-Stark localization of interfaces and extremely long-lived pre-thermal states Balducci et al. 2022a ; Balducci et al. 2022b . Away from the perturbative regime, the stability of domain wall initial conditions persists as long as the system remains within a smooth interface regime below a roughening cross-over Krinitsin et al. 2024 . The associated slow growth of entanglement renders tensor network techniques a suited tool to investigate the characteristics of such dynamics. In this work we investigate the capabilities of TTNs in regimes of constrained dynamics and paradigmatic quench settings. We provide a comprehensive study of the numerical method, including detailed performance benchmarks. We start by recapitulating the general concepts behind TTNs in Section II.1 and the T ime- D ependent V ariational P rinciple (TDVP) used to perform the time evolution in Section II.2 . Benchmarks and discussions of strategies enhancing the performance of the algorithm used in this work, like the usage of local sums instead of M atrix P roduct O perators and the performance gain on GPU s, are provided in Section II.3 . We introduce the transverse field Ising model and analytical predictions related to its perturbative limit in Section III.1 , after which we turn to a numerical investigation of said predictions for various classes of initial states in Section III . Away from the perturbative limit, we provide an analysis for quenches of different strengths, pointing out limitations of the method. A summary of the main results and outlook in Section IV conclude this work. II Numerical and theoretical Details II.1 Tree Tensor Networks Tree tensor networks provide a highly versatile, loop-free network architecture, which makes them easily applicable to arbitrary lattice geometry while keeping the computational complexity tractable. By alternating the direction of inter-layer connections, i.e. for a two dimensional lattice alternating between connections oriented along the x- and y-direction, see Fig. 1 a), the tree structure is able to provide a coverage of higher-dimensional lattices,