⚛️ quantum physics

Efficient classical simulation of two-dimensional long-range systems: Rydberg arrays and beyond

This paper introduces a new framework that reduces the computational cost of variational Monte Carlo simulations for two-dimensional long-range quantum systems from $O(N^3)$ to $O(N)$ , enabling accurate classical benchmarking of large-scale Rydberg array experiments and resolving previous limitations in tensor network state evolution.

Original authors: Jia-Lin Chan, Tao Xiang, Yantao Wu

Published 2026-07-07

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Jia-Lin Chan, Tao Xiang, Yantao Wu

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

Imagine you are trying to predict how a massive crowd of people will move and interact in a giant square room. In the world of quantum physics, this "crowd" is made of atoms, and the "room" is a grid of 100 spots (a 10x10 array). These atoms are special: they can talk to every other atom in the room, not just their immediate neighbors. This is like if everyone in a stadium could instantly whisper a secret to everyone else, no matter how far away they are.

Scientists have built real machines (called Rydberg simulators) to do this, but to trust what they see, they need a "classical" computer to double-check the results. The problem is, calculating how these 100 atoms interact is incredibly hard for normal computers. It's like trying to solve a puzzle where every piece affects every other piece simultaneously.

The Problem: The "Cubic" Bottleneck

In the past, trying to simulate this on a computer was like trying to count every possible handshake in a room of 100 people. If you have $N$ atoms, the math required to check the energy of the system grew so fast (specifically, $N^3$ ) that it became impossible for computers to handle large groups. It was a "computational bottleneck" that stopped scientists from verifying the experiments on the 10x10 grid.

The Solution: A New Way to Look at the Crowd

The authors of this paper invented a clever new method to speed this up. They realized that instead of trying to calculate everything at once in one "language" (or basis), they could break the problem down into smaller, easier chunks by changing the "language" they used to look at the atoms.

Think of it like this:

The Old Way: Trying to describe a complex dance by writing down every single move for every dancer at the same time. It takes forever.
The New Way: The authors realized they could split the dance into two simpler parts. First, they look at the dancers moving left and right (the "X" direction). Then, they switch their view to look at the dancers moving up and down (the "Y" direction).
By switching between these two views (bases), they found that the complicated "long-distance" whispers between atoms became simple, local whispers. This reduced the math from a massive, impossible task ( $N^3$ ) to a manageable one ( $N$ ). Now, the computer can handle the 100-atom crowd easily.

The Experiment: Testing the "Adiabatic" Ramp

The team used their new tool to simulate a specific experiment that was recently done with real atoms. In this experiment, scientists started with a perfect checkerboard pattern of atoms (some pointing up, some down) and slowly changed the rules to see how the atoms settled into a new state. This process is called an "adiabatic ramp."

Previously, there was a mystery:

The real experiment showed weak connections between atoms far apart.
But a different type of computer calculation (using a method called MPS) suggested the atoms should have strong connections.

No one knew if the real experiment was "broken" or if the computer calculation was wrong.

The Results: Solving the Mystery

Using their new, fast method, the authors ran a perfect simulation of the experiment on a 10x10 grid.

The Verdict: Their simulation showed that if you do the experiment perfectly, the atoms do form strong, long-distance connections, just like the MPS calculation predicted.
The Conclusion: This means the "weak connections" seen in the real experiment weren't because the physics was wrong. Instead, it was likely due to small imperfections in the actual machine (like not having enough laser power to set the rules perfectly).

Why It Matters

This paper gives scientists a powerful new "ruler" to measure their quantum machines.

It's Fast: It turns a calculation that was impossible into one that takes a normal amount of time.
It's Accurate: It can handle the tricky "long-range" interactions that other methods struggle with.
It Fixes Bugs: It helps scientists figure out if a real-world experiment failed because of the physics or because of a glitch in the machine.

In short, the authors built a new, efficient engine that allows classical computers to keep up with the most advanced quantum simulators, helping us understand how these strange quantum systems really work.

Technical Summary: Efficient Classical Simulation of Two-Dimensional Long-Range Systems

Problem Statement
Variational Monte Carlo (VMC) is a primary tool for benchmarking quantum simulators, particularly for two-dimensional (2D) systems with long-range interactions (e.g., Rydberg atom arrays with $1/r^3$ or $1/r^6$ couplings). However, a significant computational bottleneck exists in evaluating the "local energy" ( $H_{\text{loc}}$ ) for systems with arbitrary all-to-all two-body interactions.

In standard VMC, while sampling costs are $O(N)$ for Tensor Network States (TNS) and $O(N^2)$ for Neural Quantum States (NQS), the evaluation of $H_{\text{loc}}$ for long-range off-diagonal couplings requires summing over $O(N^2)$ non-local terms.
Since each term requires a full network evaluation taking $O(N)$ time, the total cost per Monte Carlo (MC) sweep scales as $O(N^3)$ . This cubic scaling renders accurate classical simulation of large 2D long-range systems (e.g., $10 \times 10$ lattices) intractable.
Additionally, a technical pathology arises when evolving TNS from product states (common in Rydberg experiments): at exact product states, the Quantum Geometry Tensor (QGT) develops null directions if the bond dimension is greater than one, causing the Time-Dependent Variational Principle (TDVP) equations to become singular.

Methodology
The authors introduce an adaptive-basis tensor-network tVMC framework designed to reduce the computational complexity of local energy evaluation from $O(N^3)$ to $O(N)$ while retaining the ability to handle general inhomogeneous couplings and realistic geometries.

Adaptive Basis Sampling:
- Instead of sampling in a single fixed basis (e.g., the $z$ -basis), the Hamiltonian is decomposed into components that are diagonal in specific bases.
- For the dipolar XY model ( $H = H_{XY} + H_Z$ ), the authors split the evolution into $H_X$ (diagonal in the $x$ -basis) and $H_Y + H_Z$ (diagonal in the $y$ -basis).
- Two independent Markov chains are maintained: one sampling from $|\langle x|\psi_\theta\rangle|^2$ and another from $|\langle y|U|\psi_\theta\rangle|^2$ , where $U$ is the basis-change operator.
- Since the off-diagonal long-range couplings vanish within each specific basis, the local energy evaluation for each component becomes $O(N)$ .
- The parameter updates are computed separately for each basis ( $\dot{\theta}_X$ and $\dot{\theta}_Y$ ) and combined via $\dot{\theta} = \dot{\theta}_X + U^\dagger \dot{\theta}_Y$ .
Handling Product State Initial Conditions:
- To resolve the singularity in the QGT when starting from a product state (where bond dimension $D > 1$ ), the authors employ a Projected tVMC (p-tVMC) approach for the initial time step.
- Instead of solving the singular TDVP equation, the first step ( $\delta t$ ) is determined by maximizing the fidelity $|\langle \psi_\theta | V | \psi(0) \rangle|^2$ , where $V$ is the second-order expansion of the time-evolution operator.
- This optimization is efficient because the initial state is a product state, making the sampling distribution $|\langle s|V|\psi(0)\rangle|^2$ computationally tractable. After this initial step, standard tVMC is resumed.
Symmetry Exploitation:
- The framework explicitly imposes reflection symmetry along the lattice diagonals in the TNS ansatz. This reduces the number of independent variational parameters by a factor of four, significantly alleviating GPU memory requirements and reducing the variance of parameter updates.

Key Results
The method was applied to simulate the adiabatic real-time protocol of a $10 \times 10$ dipolar XY model realized in a recent Rydberg experiment (Chen et al., Nature 2023), a system size previously beyond the reach of classical simulation.

Benchmarking: On a $4 \times 4$ system, the TNS-tVMC results (with bond dimensions $D=4, 5$ ) systematically converged to exact diagonalization results, accurately reproducing driven relaxation dynamics and fast oscillations.
$10 \times 10$ Simulation:
- The simulation tracked the energy evolution, showing the system relaxing toward a near-steady plateau slightly above the ground-state energy.
- Antiferromagnetic (AFM) Correlations: The study addressed a discrepancy in the original experiment, where the dynamically prepared state showed suppressed long-range AFM correlations compared to ground-state Matrix Product State (MPS) calculations.
- The authors' ideal dynamics simulation (starting from a perfect Néel state with a larger detuning $\delta_0 = 19.48$ ) demonstrated that the prescribed adiabatic ramp does produce spatially extended AFM correlations consistent with the variational ground state.
- This suggests the suppression observed in the experiment is likely due to experimental imperfections (such as limited laser power preventing the use of the optimal $\delta_0$ on the $10 \times 10$ array) rather than an intrinsic failure of the adiabatic protocol.
Convergence: The results showed robust convergence with respect to sample size ( $N_s = 81,920$ vs. $163,840$), confirming the reliability of the adaptive-basis approach.

Significance and Claims
The paper claims to provide a scalable classical benchmarking tool for long-range quantum platforms, specifically Rydberg and ion-trap simulators, which are currently in urgent need of such tools.

Technical Achievement: The work resolves the $O(N^3)$ bottleneck, enabling accurate simulation of real-time dynamics and ground states for 2D long-range systems with $O(N)$ cost per MC sweep.
Resolution of Pathology: It offers a robust solution for evolving TNS from product states, a common scenario in experimental protocols.
Scientific Impact: By providing a direct classical reference for the $10 \times 10$ Rydberg experiment, the work clarifies the nature of the discrepancy between experimental results and theoretical ground-state predictions. It demonstrates that the ideal adiabatic protocol is capable of generating the expected long-range order, shifting the focus of experimental optimization to realization imperfections.
Future Outlook: The authors position this method as a bridge for classical-quantum collaboration, where classical simulations can guide the identification of interesting physical setups before entanglement growth becomes too severe for tensor networks, at which point the dynamics can be handed over to quantum simulators. The method is also noted as a new avenue for studying 2D long-range systems with sign problems, where other numerical methods struggle.