Monte Carlo sampling from a projected entangled-pair state in simulations of quantum annealing in the three dimensional random Ising model

arXiv:2603.16509 · quant-ph, cond-mat.dis-nn, cond-mat.str-el · Submitted 2026-03-17 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: Today's paper: "Monte Carlo sampling from a projected entangled-pair state in simulations of quantum annealing in the three dimensional random Ising model".

Mira: Monte Carlo sampling from a projected entangled-pair state in simulations of quantum annealing in the three dimensional random Ising model addresses computational bottlenecks in simulating quantum phase transitions by proposing…

Kai: First, who's behind it and why it matters.

Paper summary: Kai: So, we've established that the paper, "Monte Carlo sampling from a projected entangled-pair state in simulations of quantum annealing in the three dimensional random Ising model," is about tackling the computational difficulty of simulating quantum phase transitions in three dee tensor network simulations. It sets out to propose an efficient Monte Carlo approach for evaluating expectation values within these complex three dee networks.

Mira: Precisely. The central thesis is that this method shifts the primary computational burden away from intensive expectation value evaluation and places it more squarely on time-evolution simulation, which allows researchers to explore a broader spectrum of quench times relevant to non-equilibrium quantum dynamics, such as the Kibble-Zurek mechanism.

Lev: I'm trying to keep framing this in terms of what this means for running things. If the bottleneck shifts, does that imply we can run longer or just larger systems without blowing up the computational resources?

Kai: It implies we can evaluate residual excitation energy across a full range of quench times that classical simulations can manage, which is a key feature they highlight. This allows them to test the Kibble-Zurek power law governing residual energy, Q proportional to t-(d nu+z nu-one)/(1+z nu)a, in three dimensions where d=three.

Mira: That scaling relation is the theoretical underpinning, and their work shows how their Monte Carlo sampling technique allows them to actually compute that residual energy across the relevant time scales. They are using a three dee tensor network ansatz, specifically an iPEPS, for this evolution.

Lev: I gotta ask about the complexity of that iPEPS simulation itself; simulating anything in three dimensions with those network methods usually means massive computational overhead, so how does this Monte Carlo sampling actually tame that growth?

Kai: The paper details using a neighborhood tensor update to manage the exponential growth of the bond dimension during time evolution, truncating it back to a fixed bond dimension D. They also use SVD1 truncations for closed loops in three dee to maintain a Hermitian and non-negative metric tensor for stable truncation back to D.

Mira: That's the engineering trick they employ; managing that exponential growth via truncation is necessary for the time evolution part, and this Monte Carlo method then builds upon that structure to evaluate observables more efficiently than traditional deterministic methods.

Lev: It sounds like a lot of work just to get the simulation running before you even get to the sampling efficiency. I wonder if this method is robust enough for real-world noise or imperfections we see on actual quantum devices.

Kai: The paper tests it on various lattice configurations, including an infinite lattice with periodic nearest-neighbor random Ising couplings, and a finite lattice with open boundary conditions, PBC and OBC. They show how the method performs across these different setups.

Mira: Overall, the importance of this research lies in providing a more computationally feasible path for studying non-equilibrium quantum dynamics in higher dimensions by changing where the computational cost is concentrated.

Lev: So, to sum up: this paper proposes a new way to estimate residual energy using Monte Carlo sampling on projected PEPS to make time evolution simulations for three dee quantum annealing more tractable.

Conclusion: Kai: Considering the title, "Monte Carlo sampling from a projected entangled-pair state in simulations of quantum annealing in the three dimensional random Ising model," it really sounds like this work is fundamentally about finding a better computational tool for probing these specific quantum systems. The authors are Jacek Dziarmaga and colleagues.

Mira: I think the implications are that this research provides a more viable pathway to study the behavior of quantum systems undergoing transitions in three dimensions, especially when those transitions involve non-equilibrium processes like quenching. It moves beyond just finding the ground state to studying how systems evolve during the annealing process itself.

Lev: From an error correction angle, if we can efficiently characterize these dynamics, it gives us better benchmarks for developing error mitigation strategies tailored to these specific three dee models rather than relying on simplified 1D or 2D approximations.

Kai: Exactly. If we can simulate the dynamics of a realistic three dee random Ising model more accurately, it gives us concrete data points that help us understand the physics of noise and how to handle it in quantum annealing experiments.

Mira: The core contribution is shifting the computational bottleneck; they show that observables can be evaluated either deterministically or through this three dee Monte Carlo sampling, which is more efficient for their specific context because it leverages a single-layer structure of the involved tensor networks.

Lev: I see how that single-layer structure mentioned in the paper could translate into tangible computational gains, suggesting that the cost of running these simulations is actually lower than previously estimated when you factor in this sampling method.

Kai: So, what we’re seeing here is a shift in focus: instead of spending all our time just calculating expectation values deterministically, we can afford to focus more on simulating the time evolution itself over a wider range of times.

Mira: That's the practical implication: we gain access to a broader range of non-equilibrium dynamics that were previously computationally prohibitive, which is vital for understanding systems like those in real quantum hardware.

Lev: Ultimately, this work suggests that the simulation techniques can be adapted to provide richer data sets for validating theories about quantum phase transitions in higher dimensions.

Kai: So, to wrap up on "Monte Carlo sampling from a projected entangled-pair state in simulations of quantum annealing in the three dimensional random Ising model," the paper offers a method that makes simulating three dee quantum annealing dynamics more computationally feasible by strategically placing the computational load on time evolution.

Jacek Dziarmaga

Jagiellonian University

quant-ph, cond-mat.dis-nn, cond-mat.str-el

Submitted: 2026-03-17

Updated: 2026-09-28

Comments: 12 pages, 10 figures; v2 with improved presentation accepted in Phys.Rev.B

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 82/100

The gist: Monte Carlo sampling from a projected entangled-pair state in simulations of quantum annealing in the three dimensional random Ising model addresses computational bottlenecks in simulating quantum

Key concepts

Quantum Annealing Model: This describes a 3D random Ising model that is slowly transformed from a transverse field to an Ising Hamiltonian over time. The evolution is controlled by a parameter 's' representing the annealing schedule, allowing researchers to study quantum phase transitions in this system.
Time Evolution via Tensor Networks: The simulation of how the quantum state changes over time uses a 3D tensor network called iPEPS. To keep calculations manageable, a technique called neighborhood tensor update (NTU) is used to limit the network's complexity, preventing it from growing exponentially during this evolution.
Monte Carlo Sampling Approach: Instead of complex deterministic methods, this uses a Monte Carlo algorithm on a projected PEPS. This method samples physical indices directly, which avoids the need for computationally expensive double-layer networks often required by traditional deterministic expectation value calculations.
Computational Bottleneck Shift: The study finds that using Monte Carlo sampling makes evaluating observables more efficient than deterministic methods in this context. This allows researchers to focus computational power on simulating time evolution, which is less restrictive for studying non-equilibrium dynamics.

Terminology

Monte Carlo — 35× in this paper
Phase Transition — 9× in this paper · explained in 2 episodes
Phase Transitions — 6× in this paper · explained in 2 episodes
Singular Value Decomposition (SVD) — 2× in this paper · explained in 2 episodes
Accuracy — 1× in this paper · explained in 1 episode
Benchmark — 1× in this paper · explained in 2 episodes
Computational overhead — 1× in this paper · explained in 2 episodes
Core — 1× in this paper · explained in 2 episodes
Generalization — 1× in this paper · explained in 13 episodes
Matrix Product States (MPS) — 1× in this paper · explained in 2 episodes
Topological Defects — 1× in this paper · explained in 2 episodes
Universality Class — 1× in this paper · explained in 2 episodes

Summary

Monte Carlo sampling from a projected entangled-pair state in simulations of quantum annealing in the three dimensional random Ising model addresses computational bottlenecks in simulating quantum phase transitions by proposing an efficient Monte Carlo approach for evaluating expectation values within 3D tensor network simulations. This method is significant because it shifts the primary computational burden from expectation value evaluation to time-evolution simulation, offering a more computationally feasible path for studying non-equilibrium quantum dynamics like the Kibble-Zurek mechanism (QKZM) in higher dimensions.

The gist: Monte Carlo sampling allows for the estimation of residual excitation energy across a full range of quench times accessible to classical simulations, effectively shifting the computational bottleneck back to time-evolution simulation.

Quantum Annealing Model and Theoretical Framework

The study considers a 3D transverse-field quantum Ising model on a cubic lattice, parameterized for quantum annealing as described by the Hamiltonian:

H(s) = Γ(s) HD + J (s) HI,

where s is a dimensionless time-like parameter ranging from 0 to 1, representing the annealing schedule. The system evolves from a nearly pure transverse-field Hamiltonian to the pure Ising Hamiltonian. The relevant critical exponents for this transition are cited as 1/ν = 1.55 and z = 1.3, with the predicted residual excitation energy scaling as Q∝t-(dν+zν−1)/(1+zν)a, where d=3 is the spatial dimensionality.

Simulation of Time Evolution via Tensor Networks

Time evolution is implemented using a second-order Suzuki-Trotter decomposition on a three-dimensional (3D) tensor network ansatz, specifically an iPEPS (infinite PEPS). To manage the exponential growth of the bond dimension during time evolution, the neighborhood tensor update (NTU) is employed for truncation back to a fixed bond dimension D.

The NTU prevents exponential growth by truncating back to D.

In 3D, closed loops are approximated using SVD1 truncations to maintain a Hermitian and non-negative metric tensor for stable bond-dimension truncation. The simulation is performed for various ramp times on an infinite lattice with periodic randomness, an L3 unit cell, and a finite lattice with open boundary conditions (PBC and OBC).

Deterministic Expectation Value Evaluation

The deterministic method for computing expectation values combines techniques from Ref. 130 and Ref. 131. This approach involves:

Determining the upper and lower double-PEPS boundary states shown in Fig. 2 (c,i).

Subsequently, the resulting effective two-dimensional tensor network is contracted using the double-layer corner transfer matrix renormalization group (CTMRG) method [117, 121]. For an infinite lattice with periodic disorder characterized by an L3 unit cell, there are L distinct upper and lower boundaries obtained via the power method. Each horizontal layer acts as a transfer matrix applied repeatedly until convergence is reached, with boundary bond dimensions truncated back to d using the simple-update scheme.

Monte Carlo Sampling Approach

The Monte Carlo (MC) Metropolis-Hastings algorithm is adapted to evaluate expectation values on a finite lattice. The main advantage of this approach is that it operates on a projected PEPS where physical indices are fixed, eliminating the need for a double-layer network required by deterministic evaluations.

This eliminates the need for a double-layer network.

The sampling proceeds by updating pairs of physical indices, using upper and lower projected PEPS boundaries to compute probability amplitudes Akl. The process involves:

Defining a triple tensor at each site, which may or may not include the physical index depending on whether the full or projected PEPS tensor is considered.
Converting the triple-layer network into a planar network (Fig. 6(a)) by defining planar tensors as in Fig. 5(c).
Updating upper and lower MPS boundaries using the zipper method [151].
Using singular value decomposition (SVD) to move the orthogonality center one site to the right, interpolating between mixed boundaries to generate new intermediate boundaries.

The residual energy is evaluated at s = 0.6, when the transverse field is negligible for PBC and Monte Carlo methods are used for OBC. The computational gain from this single-layer structure outweighs the cost of multiple Monte Carlo sweeps required to achieve desired error bars.

Conclusion on Computational Bottleneck Shift

The research concludes that observables can be evaluated either deterministically or via 3D Monte Carlo sampling, which is more efficient in this context. The computational bottleneck shifts to the simulation of time evolution, which is less restrictive for imaginary-time evolution toward the ground state where the Monte Carlo approach can be particularly advantageous. This advantage stems from the single-layer structure of the involved tensor networks, drastically reducing contraction costs compared to double-layer networks used in deterministic evaluations.

Improvements for AI systems

Here are specific improvements to AI systems based on this scientific paper, along with what these improved systems could achieve:

Improved Simulation of Quantum Annealing Dynamics (QKZM) via 3D Tensor Networks: The system can now accurately simulate the non-equilibrium dynamics (quenches) of complex quantum many-body systems, such as those encountered in quantum annealing (e.g., the random Ising model), across three spatial dimensions using Projected Entangled-Pair States (PEPS).
Accurate Prediction of Topological Defect Scaling: The system can now predict the residual excitation energy scaling with annealing time, governed by the Kibble-Zurek mechanism (KZ) power law, for 3D systems. This allows for a quantitative probe of whether quantum hardware exhibits expected critical behavior versus classical limitations.
Efficient Evaluation of High-Dimensional Expectation Values: The system can compute expectation values in 3D tensor networks using a hybrid approach: deterministic methods (like CTMRG) for periodic boundary conditions and Monte Carlo sampling for open boundary conditions. This overcomes the computational bottleneck previously associated with full double-layer PEPS simulations.
Robust Benchmarking Against Classical Methods: The improved system can serve as a rigorous benchmark to test the performance of classical tensor network approaches (like MPS) when applied to higher-dimensional problems, specifically by quantifying the advantage (or disadvantage) of quantum simulation techniques across different lattice geometries and boundary conditions.
Optimized Sampling for Complex States: By employing Monte Carlo sampling on projected PEPS boundaries, the system can efficiently estimate observables in high-dimensional quantum states where deterministic contraction methods are prohibitively expensive, enabling broader exploration of the state space relevant to quantum annealing.

These improvements enable AI systems to perform:

High-fidelity simulation and characterization of complex physical phenomena occurring at phase transitions in 3D quantum materials or condensed matter models relevant to hardware design.
Predictive modeling for quantum computing architectures by assessing the quantumness or limitations of hardware (like D-Wave systems) through observable scaling laws derived from non-equilibrium dynamics.
More scalable and resource-efficient methods for calculating properties of large, high-dimensional quantum states that are intractable for standard classical tensor network solvers.

Abstract

Quantum annealing with the D-Wave Advantage system in the random Ising model on a cubic lattice is simulated using a three-dimensional (3D) tensor network. The Hamiltonian is driven across a quantum phase transition from a paramagnetic phase to a spin-glass phase. The network is represented as a tensor product state, also known-particularly in two dimensions-as a projected entangled-pair state (PEPS). The annealing procedure is repeated for a range of annealing times in order to test the Kibble-Zurek (KZ) power law governing the residual energy at the end of the annealing ramp. For an infinite lattice with periodic nearest-neighbor random Ising couplings, the final energy is evaluated using a deterministic method. For a finite lattice with open boundaries, we introduce a more efficient Monte Carlo sampling approach. In both cases, the residual energy as a function of annealing time approaches the KZ power law as the annealing time increases.