⚛️ quantum physics

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

This paper simulates quantum annealing in the three-dimensional random Ising model using a projected entangled-pair state (PEPS) tensor network, demonstrating that the residual energy follows the Kibble-Zurek power law for both infinite and finite lattices by employing deterministic and Monte Carlo sampling methods, respectively.

Original authors: Jacek Dziarmaga

Published 2026-03-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jacek Dziarmaga

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: A Race Against Time

Imagine you are trying to organize a chaotic room (a quantum system) into a perfectly tidy state (the ground state). You have a set of rules (the laws of physics) that tell you how to arrange the furniture.

In this paper, the author, Jacek Dziarmaga, is studying what happens when you try to tidy up this room very quickly.

The Setup: The room starts in a state of total chaos (paramagnetic phase), where everything is jumbled.
The Goal: You want to reach a state of perfect order (spin-glass phase), where every piece of furniture is in its specific, correct spot.
The Problem: If you move too slowly, you can do it perfectly. But if you rush (which is what "Quantum Annealing" does), you inevitably knock over a few chairs or leave a book on the floor. These mistakes are called defects or residual energy.

The paper asks: "How many mistakes will we make if we rush at different speeds?"

The Cast of Characters

The Quantum Annealer (The D-Wave Machine): Think of this as a super-fast, high-tech robot arm trying to tidy the room. It's real hardware that exists in the lab.
The Simulation (The Virtual Twin): Before we trust the robot, we want to simulate the process on a supercomputer to see what should happen. This is where the math gets heavy.
The PEPS (The 3D Puzzle): To simulate the room, the author uses a mathematical tool called a Projected Entangled-Pair State (PEPS).
- Analogy: Imagine the room is a giant 3D jigsaw puzzle. Each piece of the puzzle is connected to its neighbors. To understand the whole room, you have to look at how all these pieces fit together.
- The Challenge: In 2D (a flat floor), this puzzle is manageable. But in 3D (a full room with height), the connections become a tangled web of spaghetti. Calculating the final state of this 3D puzzle is incredibly hard for computers.

The Two Main Problems

The paper tackles two specific hurdles in simulating this 3D puzzle:

1. The "Double-Layer" Bottleneck (The Old Way)

To check if the robot did a good job, the old method required building two identical 3D puzzles side-by-side: one representing the "real" state and one representing the "mirror image" (to calculate probabilities).

The Metaphor: Imagine trying to weigh a heavy box by building a second, identical box out of gold bricks and comparing them. It's accurate, but it's incredibly expensive and slow.
The Result: This method worked for small, infinite grids, but it was too slow to test many different speeds.

2. The "Monte Carlo" Solution (The New Way)

The author introduces a new, smarter way to check the work using Monte Carlo sampling.

The Metaphor: Instead of building the whole second gold box, imagine you are a detective. You don't check every single brick in the room. Instead, you randomly pick a few spots, check them, and use statistics to guess the condition of the whole room.
How it works: The computer "samples" the 3D puzzle layer by layer. It makes a guess, checks if it's likely, and moves on. By doing this thousands of times, it builds a very accurate picture of the final energy without needing to solve the impossible math of the whole 3D web at once.
The Benefit: This is like switching from weighing the whole box to taking a quick, smart sample. It's much faster and allows the computer to test a much wider range of speeds.

The Discovery: The "Kibble-Zurek" Law

The author ran these simulations to test a famous theory called the Kibble-Zurek (KZ) mechanism.

The Theory: This theory predicts a specific relationship between speed and mistakes. It says: "If you slow down your annealing time by a certain factor, the number of mistakes (residual energy) will drop by a predictable power."
The Analogy: Think of driving a car over a bumpy road. If you drive at 100 mph, you get thrown around a lot (many defects). If you drive at 50 mph, you get thrown around less. The KZ law gives you the exact formula for how much smoother the ride gets as you slow down.

The Results

The Infinite Lattice (The Perfect World): Using the old "deterministic" method, the author confirmed that for an infinite grid, the results follow the KZ law perfectly.
The Finite Lattice (The Real World): Using the new "Monte Carlo" method, the author simulated a finite grid (a room with walls). Even though this method involves some randomness (sampling), the results still matched the KZ law.
The Conclusion: As the annealing time gets longer (the robot moves slower), the "residual energy" (the mess left behind) drops exactly as the theory predicted.

Why This Matters

This paper is a victory for computational efficiency.

Before this, simulating 3D quantum systems was like trying to solve a Rubik's cube that is 1,000 times bigger than a normal one.
By using the Monte Carlo sampling trick, the author showed we can solve these 3D puzzles much faster.
This allows scientists to benchmark real quantum computers (like D-Wave) against classical simulations more effectively. It proves that the quantum computer is behaving "quantumly" and following the laws of physics, rather than just acting like a noisy classical machine.

In a nutshell: The author found a faster, smarter way to simulate a 3D quantum puzzle. By using a "sampling" technique instead of a "brute force" calculation, they proved that the rules of quantum speed limits (Kibble-Zurek) hold true, even in complex 3D environments. This helps us trust and understand the next generation of quantum computers.

1. Problem Statement

The paper addresses the computational bottleneck in simulating Quantum Annealing (QA) for the 3D random Ising model using Projected Entangled-Pair States (PEPS), a tensor network ansatz suitable for higher dimensions.

Context: Quantum annealing devices (like D-Wave) are used to study the Kibble-Zurek Mechanism (KZM), which predicts how topological defects (or residual energy) scale with the quench rate when a system crosses a quantum critical point.
The Challenge: While PEPS can represent 3D quantum states efficiently, evaluating expectation values (specifically the residual excitation energy) on these networks is computationally expensive.
- In previous work (Ref. 101), the bottleneck was identified as the evaluation of observables on a double-layer tensor network (required for deterministic expectation values).
- This limitation restricted the ability to simulate long annealing times and test the KZ power law scaling ( $Q \propto t_a^{-0.965}$ ) over a broad range of parameters.

2. Methodology

The author employs a 3D tensor network to simulate the time evolution of a cubic lattice Ising model driven across a quantum phase transition (from paramagnetic to spin-glass). The methodology is divided into time evolution and two distinct approaches for evaluating observables:

A. Time Evolution

Model: A 3D transverse-field Ising model with random nearest-neighbor couplings ( $J_{ij} \in [-1, 1]$ ).
Algorithm: Time evolution is implemented using a second-order Suzuki-Trotter decomposition.
Truncation: To prevent the exponential growth of the bond dimension ( $D$ $D$ ) during evolution, the Neighborhood Tensor Update (NTU) method is used.
- In 3D, neighborhoods containing closed loops are approximated using Singular Value Decomposition (SVD) truncations (specifically retaining only the leading singular value, SVD1) to manage computational cost while maintaining a stable, non-negative metric tensor.

B. Observable Evaluation: Two Approaches

The paper compares two methods for calculating the residual energy $Q$ at the end of the annealing ramp:

Deterministic Method (Periodic Boundary Conditions - PBC):
- Used for infinite lattices with periodic disorder (unit cell $L^3$ ).
- Relies on contracting a double-layer PEPS network (state $\times$ conjugate).
- Uses the Corner Transfer Matrix Renormalization Group (CTMRG) to contract the effective 2D boundary networks.
- Limitation: Computationally heavy due to large singular value decompositions required for the double-layer structure.
Monte Carlo (MC) Sampling Method (Open Boundary Conditions - OBC):
- Core Innovation: Instead of contracting the full double-layer network, the algorithm operates on a projected PEPS where physical indices are fixed to specific values.
- Sampling: Uses the Metropolis-Hastings algorithm to sample configurations of physical indices ( $\sigma^z$ basis).
- Efficiency:
  - Eliminates the need for a double-layer network, reducing the problem to a single-layer contraction.
  - Utilizes 2D PEPS boundaries and 1D Matrix Product State (MPS) boundaries updated via a "zipper" procedure (moving an orthogonality center across the lattice).
  - The bond dimension $d$ of the boundaries can be chosen significantly smaller than in the deterministic double-layer case, drastically reducing contraction costs.
- Trade-off: Requires multiple sampling sweeps to estimate observables with statistical accuracy, but the per-sweep cost is much lower.

3. Key Contributions

Algorithmic Breakthrough: The introduction of a 3D Monte Carlo sampling scheme for PEPS that avoids the double-layer contraction bottleneck. This shifts the computational bottleneck from "evaluating observables" back to "time evolution," allowing for broader exploration of parameter space.
Validation of KZM in 3D: Successfully simulates the Kibble-Zurek scaling in a 3D random Ising model, confirming the predicted power law $Q \propto t_a^{-0.965}$ (derived from critical exponents $\nu=1.55, z=1.3, d=3$ ).
Benchmarking: Provides a rigorous classical benchmark for quantum annealers (D-Wave Advantage) by extending the range of accessible annealing times in classical simulations.

4. Results

Scaling Verification:
- Deterministic (PBC): For an infinite lattice with $L^3=8^3$ , the residual energy approaches the KZ power law slope as annealing time $t_a$ increases (up to $\approx 3$ ns).
- Monte Carlo (OBC): For a finite lattice with open boundaries ( $L=8$ ), the MC method successfully reproduces the KZ scaling for longer annealing times (up to $\approx 5$ ns).
Performance:
- The MC method allows for the estimation of residual energy across the full range of quench times accessible to classical simulations.
- While the MC method requires many sweeps (increasing from $16 \times 6$ to $420 \times 6$ sweeps for slower ramps) to maintain a relative error of $\delta Q/Q = 0.01$ , the computational gain from the single-layer structure outweighs the cost of sampling.
Error Analysis: The autocorrelation time is low (a few sweeps), and statistical errors are estimated using block averaging.

5. Significance

Overcoming Dimensionality Limits: This work demonstrates that PEPS simulations in 3D are not limited to small systems or short times if efficient sampling techniques are employed. It effectively extends the reach of classical tensor network simulations.
Hardware Validation: By extending the range of simulated annealing times, the study provides a more robust test for the "quantumness" of D-Wave hardware. If the quantum device deviates from the KZ scaling predicted by these advanced classical simulations, it indicates non-adiabatic effects or thermal noise.
General Applicability: The Monte Carlo sampling scheme is not limited to annealing; it can be applied to other problems involving PEPS, such as computing ground state properties or thermal states, particularly where open boundary conditions are relevant.
Future Direction: The paper suggests that the remaining bottleneck is the time evolution itself, implying that future improvements should focus on optimizing the Trotter steps or the NTU truncation for 3D systems.

In summary, this paper presents a significant methodological advance in tensor network simulations by replacing expensive deterministic contractions with efficient Monte Carlo sampling, thereby enabling high-precision tests of quantum critical dynamics in three dimensions.