Simulating dynamics of the two-dimensional transverse-field Ising model: a comparative study of large-scale classical numerics

Joseph Vovrosh, Sergi Julià-Farré, Wladislaw Krinitsin, Michael Kaicher, Fergus Hayes, Emmanuel Gottlob, Augustine Kshetrimayum, Kemal Bidzhiev, Simon B. Jäger, Markus Schmitt, Joseph Tindall, Constantin Dalyac, Tiago Mendes-Santos, Alexandre Dauphin · 2025

Read Simulating dynamics of the two-dimensional transverse-field Ising model: a comparative study of large-scale classical numerics by Joseph Vovrosh, Sergi Julià-Farré, Wladislaw Krinitsin, Michael Kaicher, Fergus Hayes, Emmanuel Gottlob, Augustine Kshetrimayum, Kemal Bidzhiev, Simon B. Jäger, Markus Schmitt, Joseph Tindall, Constantin Dalyac, Tiago Mendes-Santos, Alexandre Dauphin (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-2511-19340.txt.

Introduction

Joseph Vovrosh and thirteen co-authors benchmark classical methods for simulating a two-dimensional quantum magnet, the transverse-field Ising model. Simulating many-qubit dynamics classically is a key yardstick for judging quantum processors. The team gathers a broad toolbox of leading numerical approaches.

Three tensor network methods, plus a neural quantum state Monte Carlo method, are run on two standard protocols: annealing through a critical point and dynamics after a sudden quench. The extensive results show where each method's predictions agree and where they strain. The benchmark links classical simulability to physical regimes and sets reference data for future experiments.

Catalog record

CREATORJoseph Vovrosh, Sergi Julià-Farré, Wladislaw Krinitsin, Michael Kaicher, Fergus Hayes, Emmanuel Gottlob, Augustine Kshetrimayum, Kemal Bidzhiev, Simon B. Jäger, Markus Schmitt, Joseph Tindall, Constantin Dalyac, Tiago Mendes-Santos, Alexandre Dauphin
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE108 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-19340

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Simulating dynamics of the two-dimensional transverse-field Ising model: a comparative study of large-scale classical numerics Joseph Vovrosh, Sergi Julià-Farré, Wladislaw Krinitsin, Michael Kaicher, Fergus Hayes, Emmanuel Gottlob, Augustine Kshetrimayum, Kemal Bidzhiev, Simon B. Jäger, Markus Schmitt, Joseph Tindall, Constantin Dalyac, Tiago Mendes-Santos, Alexandre Dauphin Abstract The quantum dynamics of many-qubit systems is an outstanding problem that has recently driven significant advances in both numerical methods and programmable quantum processing units. In this work, we employ a comprehensive toolbox of state-of-the-art numerical approaches to classically simulate the dynamics of the two-dimensional transverse field Ising model. Our methods include three different tensor network techniques—matrix product states, tree-tensor networks, and two-dimensional tensor-networks under the belief propagation approximation—as well as time-dependent variational Monte Carlo with Neural Quantum States. We focus on two paradigmatic dynamical protocols: (i) quantum annealing through a critical point and (ii) post-quench dynamics. Our extensive results show the quantitative predictions of various state-of-the-art numerical methods providing a benchmark for future numerical investigations and experimental studies with the aim to push the limitations on classical and QPUs. In particular, our work connects classical simulability to different regimes associated with quantum dynamics in Rydberg arrays - namely, quasi-adiabatic dynamics, the Kibble-Zurek mechanism, and quantum quenches. I Introduction Figure 1: We consider a square layout of atoms and study the Transverse Field Ising model with two types of protocols: annealing and quench. We then use a toolbox of state-of-the-art classical solvers to replicate two types of observables: magnetization and two-point correlations. The last decade has witnessed the rapid development of quantum processing units (QPUs), supported both by academic research and growing industrial involvement. These versatile platforms allow for precise control and manipulation of many-body quantum systems, and therefore offering a new tool for the study of these systems in complex regimes. In particular, a key promise of these devices is their potential to simulate many-body quantum systems in regimes which are inaccessible on classical computers and therefore being used to discover or probe emergent, non-equilibrium phenomena. Indeed, in the absence of noise, it has been shown that QPUs can efficiently simulate the dynamics of strongly correlated quantum systems 22 ; 50 . However, current devices are noisy and, in parallel, unbiased classical numerical methods have been actively developed, giving rise to a stimulating competition pushing the boundaries of classical numerical methods and quantum algorithms for simulations. QPUs in digital mode have allowed for the large-scale quantum simulation of the kicked Ising model 41 and for the prethermalization study of the digitized Ising model 29 . QPUs in analog mode, meanwhile, have opened the door to the large-scale study of non-adiabatic effects near spin-glass transitions 41 ; 42 , of thermalization and criticality in the one-dimensional XY model 3 , and of Haar random distributions in one-dimensional Ising models 17 ; 88 . While QPU platforms have shown remarkable advances, the limits of classical simulability remain poorly understood, particularly in connection with the diverse dynamical phenomena that can be naturally accessed on quantum devices. In recent years, many works based on classical numerical methods have softened, or even overturned, claims of quantum advantage 97 ; 54 ; 99 . Addressing this gap is therefore essential for clarifying the boundary between classical and quantum computational capabilities in the study of many-body dynamics. Numerous numerical methods have been proposed to address many-body systems. Whereas state-vector simulations are limited by the exponential scaling of their memory requirements with system size, more efficient approaches to simulating quantum dynamics - based on compact variational representations of quantum states, such as tensor networks (TN) and Neural Quantum States (NQS) - have been developed. Tensor network ansätze include, among others, one-dimensional Matrix Product States (MPS) 92 , Tree Tensor Networks (TTN) 95 ; 60 ; 83 , and higher-dimensional tensor networks (2D-TN and 3D-TN) 104 ; 35 ; 99 . These architectures can be used to study quantum dynamics by means of tailored algorithms such as the time-dependent variational principle (TDVP) 28 and the time-evolving block decimation (TEBD) 106 ; 105 ; 109 . In the case of non-tree architectures, the choice of tensor network contraction algorithm becomes important, with choices ranging from highly efficient but approximate BP-based methods 34 ; 2 ; 98 ; 6 ; 1 ; 47 ; 64 to more accurate but more expensive approaches such as boundary MPS 51 ; 104 ; 26 and the corner transfer matrix renormalization group method 61 ; 62 . NQSs are represented by various types of artificial neural networks (ANNs), including restricted Boltzmann machines and convolutional neural networks. When combined with time-evolution algorithms - such as time-dependent Variational Monte Carlo (tVMC) for NQS - both approaches have successfully simulated the dynamics of relatively large quantum systems, going beyond brute-force state-vector simulations. Finally, recent works tackled successfully quantum dynamics with Pauli propagation methods 69 ; 70 ; 11 ; 12 ; 7 ; 71 . However, it is evident that these numerical approaches are not able to accurately capture real-time dynamics across all regimes, facing limitations in many physically relevant cases. For tensor-network methods, this limitation arises from the rapid growth of entanglement, which cannot be faithfully represented with restricted bond dimensions. For NQS-based approaches, understanding their limitations remains an open question. Numerical evidence suggests that these challenges are not necessarily related to entanglement, but rather to instabilities or sampling complexity in the tVMC algorithm. The lack of numerical techniques that are valid across all parameter regimes makes a comprehensive and extensive comparison and benchmark of existing numerical techniques a crucial requirement. In this work, we focus on many-body dynamics compatible with Rydberg atom QPUs operating in the analog mode. These versatile platforms have several strengths: arbitrary two-dimensional array geometries, continuous-time evolution, dynamical control of system parameters, and long coherence times 13 ; 74 ; 32 . They therefore allow for the exploration of a wide range of dynamical phenomena. Notable examples considered here include annealing protocols for preparing magnetically ordered states 80 ; 20 ; 86 , investigations of non-adiabatic effects associated with universal features 38 and coarsening dynamics 52 , post-quench protocols for studying quantum thermalization 88 ; 17 ; 89 ; 18 , and protocols to study phenomena related to high-energy physics 107 ; 53 ; 25 . To assess the capabilities of classical numerics and find challenging regimes compatible with Rydberg atom QPUs in a two-dimensional square array, we first consider a cross-benchmark of a toolbox of numerical approaches to simulate both annealing and post-quench dynamics; see Fig. 1 . Specifically, we employ three tensor-network methods - MPS, TTN, and 2DTN - as well as tVMC with NQSs. In an attempt to compare these methods equally, we restrict ourselves to equal hardware, namely, access to A100 Nvidia GPUs. This analysis is complemented by more conventional investigations of the errors associated with each individual method, followed by a proposed framework to verify expected symmetry constraints of observables during the dynamics. We further verify that mean-field and semiclassical methods cannot capture the behavior described by the previous methods—an indication of strong correlation of the studied post-quench dynamics. The main contribution of this work is to delineate the capabilities of state-of-the-art numerical methods in describing different dynamical phenomena that are realizable

Read the complete text