⚛️ quantum physics

Utility-Scale Quantum State Preparation: Classical Training using Pauli Path Simulation

This paper demonstrates the use of Pauli Path simulation to classically train parametrized circuits for preparing ground states of large-scale quantum many-body Hamiltonians, successfully validating these states against classical benchmarks and experimentally realizing them on Quantinuum's H2 quantum computer to achieve low energy errors and demonstrate anyon braiding.

Original authors: Cheng-Ju Lin, Hrant Gharibyan, Vincent P. Su

Published 2026-03-11

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Cheng-Ju Lin, Hrant Gharibyan, Vincent P. Su

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 find the lowest point in a vast, foggy mountain range. This mountain represents a complex quantum system (like a collection of interacting atoms), and the lowest point is its "ground state"—the most stable, energy-efficient configuration.

In the world of quantum computing, finding this lowest point is crucial for solving problems in materials science, chemistry, and physics. However, the mountain is so huge and the fog so thick that standard maps (classical computers) can't see the whole thing, and the terrain is too rough for hikers (quantum computers) to navigate without getting lost or exhausted.

This paper introduces a clever new strategy to solve this problem, acting as a bridge between classical supercomputers and noisy quantum machines. Here is the breakdown using simple analogies:

1. The Problem: The "Foggy Mountain"

The Challenge: To simulate a quantum system with 100+ atoms (qubits), you need to track an astronomical number of possibilities. It's like trying to count every grain of sand on a beach while blindfolded.
The Limitation:
- Classical Computers: They are too slow to simulate the whole mountain at once. They run out of memory.
- Quantum Computers: They are fast, but they are currently "noisy." It's like trying to hike in a blizzard; the wind (noise) knocks you off course, making it hard to reach the true bottom.

2. The Solution: "Pauli Path Simulation" (The Smart GPS)

The authors developed a method called Pauli Path Simulation (PPS). Think of this as a Smart GPS that doesn't try to map every single grain of sand.

How it works: Instead of tracking every single possibility, the GPS looks at the "main roads" (the most important paths) and ignores the tiny, irrelevant dirt trails. It uses a "truncation" technique—essentially saying, "If a path is too small to matter, let's cut it out to save time."
The Result: This allows a classical computer to simulate a 100-qubit system (which is usually impossible) by focusing only on the most significant parts of the quantum state. It's like navigating a city by only looking at the major highways and ignoring the alleyways.

3. The Strategy: "Classical Training, Quantum Execution"

The paper proposes a Hybrid Approach:

Classical Training: Use the "Smart GPS" (PPS) on a classical computer to figure out the perfect route (the optimal settings for the quantum circuit) to reach the bottom of the mountain. This is done in a "noise-free" simulation.
Quantum Execution: Take those perfect settings and load them onto a real quantum computer. The quantum computer doesn't have to "learn" or "search"; it just follows the pre-planned route.

Analogy: Imagine you are training a race car driver (the quantum computer). Instead of letting them practice on a dangerous, rainy track (noisy hardware) and hoping they learn the best line, you use a super-accurate simulator (PPS) to calculate the perfect racing line. You then give those exact coordinates to the driver, who just has to execute the turn.

4. The Results: Testing the Map

The authors tested this method on three different "mountains" (quantum models):

The Ising Model: A standard test for magnetic materials.
The Kitaev Honeycomb Model: A complex model known for "exotic" particles called anyons.

The Findings:

Accuracy: The "Smart GPS" found routes that were incredibly close to the true bottom of the mountain. In some cases, it even found a lower point than the best existing classical maps (DMRG).
Real-World Test: They took the settings for a 48-qubit system and ran them on a real quantum computer (Quantinuum's H2). Even without fixing the "noise" (the blizzard), the quantum computer reached a state with only a 5% error. That is remarkably good for current technology!
Topological Magic: They didn't just find the bottom; they proved the quantum state had "topological" properties. They successfully performed "anyon braiding" (twisting particles around each other like shoelaces), which is a signature of exotic quantum matter. This proves the quantum computer wasn't just guessing; it was holding a genuine, complex quantum state.

5. Why This Matters

This paper is a game-changer because it changes the role of the quantum computer.

Before: We asked quantum computers to do everything, including the hard work of learning the solution, which they struggled with due to noise.
Now: We use classical computers to do the heavy lifting of "training" the solution, and we use quantum computers to do what they are best at: executing the complex, pre-calculated steps.

The Bottom Line:
This method is like building a bridge. It allows us to use the power of classical supercomputers to prepare the ground, so that even our current, imperfect quantum computers can walk across and perform tasks that were previously thought impossible. It paves the way for "utility-scale" quantum computing, where we can solve real-world problems today, not just in the distant future.

1. Problem Statement

The preparation of high-quality ground states for quantum many-body systems is a fundamental prerequisite for quantum simulations (e.g., quench dynamics, scattering). While Variational Quantum Algorithms (VQAs) are the standard approach for near-term devices, they face significant challenges:

Resource Intensity: Training VQAs requires repeated evaluation of cost functions and gradients on noisy quantum hardware, which is slow and error-prone.
Classical Simulation Limits: Exact state-vector simulations are limited to small system sizes ( $\sim$ 50 qubits), while tensor network methods (like DMRG) struggle with 2D systems or high entanglement.
The "Utility Scale" Gap: There is a need for methods that can handle system sizes of $\sim$ 100 qubits (beyond exact simulation but within reach of current quantum hardware) to generate parametrized circuits that can be executed on quantum devices.

2. Methodology: Pauli Path Simulation (PPS)

The authors propose a hybrid approach where the training of the VQA is performed entirely classically using Coefficient-Truncation Pauli Path Simulation (PPS), also known as sparse Pauli dynamics.

Core Mechanism: Instead of evolving the quantum state in the Schrödinger picture, PPS evolves the observable (Hamiltonian) in the Heisenberg picture. The operator is expanded in the Pauli-string basis.
Truncation Scheme: As the operator evolves through the circuit gates, the number of Pauli terms grows exponentially. PPS mitigates this by discarding Pauli strings with coefficients below a threshold $\delta_c$ . This keeps the computational cost polynomial.
Optimization Loop:
1. A parameterized ansatz (Hamiltonian Variational Ansatz) is defined.
2. The cost function (energy expectation value) and its gradient are estimated classically using PPS.
3. Parameters are updated using a Simultaneous Perturbation Stochastic Approximation (SPSA) combined with ADAM (adaptive moment estimation) to handle high-dimensional parameter spaces efficiently.
4. Once optimized, the resulting circuit parameters are transferred to a quantum device for execution.

3. Key Contributions

Utility-Scale Training: Successfully trained VQAs for systems with 100+ qubits (1D and 2D Ising models, Kitaev honeycomb model), a scale where exact state-vector simulation is impossible.
Benchmarking Against Exact/DMRG: Validated the PPS-trained states against exact solutions (1D Ising, Kitaev) and Density Matrix Renormalization Group (DMRG) results (2D Ising).
Surpassing DMRG in 2D: Demonstrated that for the 2D Ising model on a heavy-hex lattice (matching IBM hardware connectivity), the PPS variational energy was lower than DMRG results in certain parameter regimes, suggesting PPS as a powerful "quantum-inspired" classical numerical method.
Hardware Validation: Successfully executed the PPS-trained circuits on Quantinuum's System Model H2 (48 qubits) to prepare ground states of the Kitaev honeycomb model.
Topological Verification: Demonstrated anyon braiding on the quantum hardware, confirming that the classically trained states preserve topological properties (Abelian anyons) beyond fixed-point models.

4. Key Results

A. Classical Simulation Performance

1D Quantum Ising Model (100 qubits):
- Achieved relative energy errors < 0.5% compared to exact solutions.
- Magnetization ( $M_z, M_x$ ) showed excellent agreement with DMRG, except near critical points where deeper circuits were needed.
2D Quantum Ising Model (Square & Heavy-Hex Lattices, ~100-127 qubits):
- Square Lattice: Relative energy errors < 0.25% against DMRG.
- Heavy-Hex Lattice: In specific parameter regimes, PPS achieved lower variational energies than DMRG (using comparable computational resources). This suggests PPS can outperform standard tensor network methods for certain 2D geometries.
Kitaev Honeycomb Model (100 qubits):
- Achieved relative energy errors < 0.5% in the gapped phase.
- Correctly captured the Topological Entanglement Entropy ( $S_{topo} \approx -1$ ), confirming the state's topological order.

B. Quantum Hardware Execution (Quantinuum H2)

System: 48-qubit Kitaev honeycomb model (gapped and gapless regimes).
Energy Accuracy: Achieved relative energy errors of ~5% without any error mitigation.
- $J=0.3$ : Error $\approx 4.45\%$
- $J=0.6$ : Error $\approx 5.11\%$
Topological Properties: Successfully performed anyon braiding (e-m, e- $\psi$ $ψ$ , m- $\psi$ $ψ$ ) on the prepared states.
- Measured braiding phases were close to the theoretical value of $-1 $(e.g.,$ \alpha_{em} \approx -0.84$ for $J=0.3$ ), demonstrating that the classically trained circuits preserve the topological nature of the ground state even on noisy hardware.

5. Significance and Implications

Bridge Between Classical and Quantum: The method establishes a workflow where classical simulation handles the heavy lifting of parameter optimization, while quantum hardware is reserved for tasks it excels at (e.g., simulating dynamics from the prepared state).
New Classical Numerical Method: PPS emerges as a competitive classical algorithm for 2D condensed matter problems, potentially outperforming DMRG in specific geometries and handling models where Quantum Monte Carlo suffers from the sign problem.
Hardware Efficiency: By providing "warm-start" parameters, PPS significantly reduces the runtime and resource demands on quantum devices, making VQAs more viable for near-term hardware.
Beyond Fixed-Point Models: The successful demonstration of anyon braiding on approximate ground states (not just ideal fixed-point states) marks a significant step toward realizing topological quantum computing on current devices.

Conclusion

This work demonstrates that Pauli Path Simulation is a scalable, effective tool for training variational quantum circuits at the utility scale. It not only produces high-fidelity ground states for complex 2D systems but also serves as a robust pre-training mechanism for quantum hardware, enabling the demonstration of non-trivial topological phenomena like anyon braiding on current noisy quantum processors.