⚛️ quantum physics

Hamiltonian-Dependent Compressibility and Physical Preservation in Approximate Quantum Compilation using Tensor Networks

This study demonstrates that the compressibility of approximate quantum compilation using tensor networks is governed by Hamiltonian commutator structures and target-state complexity, enabling a 45.9% reduction in circuit depth for an 80-qubit Heisenberg chain simulation while preserving high fidelity and physically relevant observables.

Original authors: Dohyun Kim, Namshik Han

Published 2026-10-14✓ Author reviewed ⓘ

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Dohyun Kim, Namshik Han

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 by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Quantum computers promise to solve problems that are impossible for today's machines, particularly in simulating how particles interact in complex materials. To do this, scientists program the computer to mimic the passage of time for a specific system, a process that requires a long sequence of tiny, precise instructions called gates. However, current quantum hardware is fragile; the longer the sequence of instructions, the more likely the machine is to make mistakes due to noise and interference. This creates a difficult trade-off: to get an accurate result, you need a long, complex circuit, but to get a result before the machine breaks down, you need a short, simple one. Researchers have developed a method called approximate quantum compilation to bridge this gap. Instead of running the long, perfect sequence, this method tries to find a much shorter sequence of gates that produces nearly the same final state. It is like trying to find a shortcut through a dense forest that leads to the same destination as the long, winding trail, but without getting lost.

In a recent study, researchers tested this shortcutting technique on a large-scale quantum simulation involving eighty qubits, the basic units of information in a quantum computer. They focused on a specific model of magnetic particles arranged in a line, known as an anisotropic XYZ Heisenberg chain, which is a standard way to study how spins interact in materials. The team used a computer algorithm based on tensor networks, a mathematical tool that helps manage the complexity of many interacting parts, to compress a simulation that originally required ten repetitions of a complex gate sequence down to just five repetitions. The goal was to see if this compressed version could still capture the essential physics of the system while using significantly fewer resources.

The results showed that the compression was highly successful. By reducing the circuit from ten repetitions to five, the researchers cut the total depth of the instructions by nearly forty-six percent and reduced the number of two-qubit gates, which are the most error-prone operations, by almost forty-eight percent. Despite this drastic reduction, the compressed circuit retained a fidelity of 0.995474 with the original target state, meaning it remained extremely close to the intended result. When the team ran both the original and the compressed circuits on actual quantum hardware at IBM, they found that the shorter circuit preserved the key physical features of the simulation. Measurements of magnetization, the alignment of neighboring particles, and the total energy of the system all matched the expected values just as well as the longer circuit did, and in some cases, the shorter circuit actually accumulated less error from the hardware.

Beyond simply proving that the shortcut worked, the study investigated why some simulations are easier to compress than others. The researchers discovered that the difficulty of finding a good shortcut depends heavily on the specific arrangement of the magnetic interactions and the direction of the external magnetic field applied to the system. They found a strong link between the compressibility of the circuit and the mathematical structure of the Hamiltonian, which is the equation describing the system's energy. Specifically, the ease of compression correlated with a measure of how much the different parts of the energy equation "disagree" with each other when applied in different orders. When this disagreement, or non-commutativity, was high, the compression was harder. This suggests that the underlying structure of the physical system itself dictates how much a simulation can be simplified, rather than just the complexity of the final state.

The team also looked deeper than just the final numbers to ensure the compressed circuit was truly behaving like the original. They used a technique called local reduced-state tomography to examine small groups of four qubits at different locations along the chain. Even though the entire eighty-qubit system is too large to map out completely, these local snapshots revealed that the compressed circuit maintained the same internal correlations and purity as the original circuit across the system. While the hardware introduced some noise that lowered the overall quality of the data compared to a perfect simulation, the compressed circuit performed just as well as the original one in preserving these local details. This confirms that the shortcut did not just guess the right answer but actually followed a path that remained physically consistent with the original, longer evolution.

These findings offer a practical path forward for running meaningful simulations on today's imperfect quantum machines. By showing that a fifty percent reduction in circuit resources can be achieved without losing the physical essence of the simulation, the work demonstrates that we can extract valuable scientific data from hardware that would otherwise be too noisy to use. The study also provides a new way to predict which simulations will be amenable to this kind of compression, based on the specific magnetic properties of the system being studied. While the research was limited to a one-dimensional chain and a single type of magnetic interaction, it establishes a clear connection between the mathematical structure of a physical problem and the feasibility of simplifying its quantum simulation. This insight helps researchers know when to expect a shortcut to work and when a system is too complex to compress, guiding the design of future experiments on larger and more challenging quantum systems.

Technical Summary: Hamiltonian-Dependent Compressibility and Physical Preservation in Approximate Quantum Compilation using Tensor Networks

Problem Statement
Quantum simulation of many-body systems on gate-based processors requires approximating continuous time evolution, $U(t) = \exp(-iHt)$ , using finite sequences of quantum gates. Standard product-formula methods (e.g., Suzuki–Trotter decompositions) improve accuracy by increasing the number of time steps, but this trade-off results in deeper circuits and higher two-qubit gate counts, which accumulate hardware errors and limit practical simulation depth. Approximate Quantum Compilation (AQC) offers a strategy to replace deep target circuits with shallower variational circuits optimized to reproduce the target output state. While tensor-network-based AQC (specifically AQCtensor) has demonstrated the ability to compress circuits for large systems, the factors governing why certain many-body evolutions are more compressible than others remain poorly understood. Furthermore, the extent to which compressed circuits retain physically relevant information on actual quantum hardware, beyond simple state fidelity, requires systematic characterization.

Methodology
The authors investigated these questions using an 80-qubit one-dimensional anisotropic XYZ Heisenberg chain initialized in a N´eel state. The study employed two distinct simulation settings:

Compressibility Analysis: A systematic sweep of Hamiltonian parameters (exchange anisotropy $J_x, J_y, J_z$ and external magnetic field strength/direction) using 20-repetition target circuits to map compression difficulty.
Hardware Validation: A specific configuration ( $J=(0.5\pi, 1.0\pi, 1.5\pi)$ , $h=(0.5\pi, 1.0\pi, 2.0\pi)$ ) with a 10-repetition target circuit executed on IBM Quantum's ibm_boston backend (Heron r3 processor).

The AQCtensor workflow represented target and variational states as Matrix Product States (MPS). A shallower Trotter-based ansatz was optimized against the target evolution using L-BFGS-B with JAX automatic differentiation, terminating when a target-state fidelity $F \ge 0.99$ was achieved or 200 iterations were reached.

To evaluate physical preservation, the authors utilized:

Observable-level benchmarks: Mean longitudinal/transverse magnetization ( $\langle Z \rangle, \langle X \rangle$ ), nearest-neighbor correlations ( $C_{ZZ}$ ), and Hamiltonian expectation values ( $\langle H \rangle$ ), measured with and without error mitigation (readout mitigation, dynamical decoupling, twirling, zero-noise extrapolation).
Process-level diagnostics: Mirror-circuit fidelity estimation (MCFE) on the relative circuit $W = U^\dagger_{\text{original}} U_{\text{AQC}}$ to estimate process fidelity.
Local-state characterization: Reduced-state tomography on eight selected four-qubit regions to compare local density matrices, purity, and fidelity between original and compressed circuits.

Key Contributions and Results

Hamiltonian-Dependent Compressibility: The study found that compression performance varies systematically with the Hamiltonian's anisotropic coupling and field orientation. These variations are strongly correlated with a nested-commutator measure ( $C_2$ ) related to the second-order Trotter error. Specifically, a higher $C_2$ (indicating stronger non-commutativity among Hamiltonian components) correlates with higher fixed-depth infidelity and greater difficulty in compression (Pearson $r=0.95$ ). This relationship holds even when controlling for field strength.
Link to Target-State Complexity: The directional trends in compressibility (e.g., differences between x- and y-field responses) are also reflected in the target-state MPS bond dimension. However, the correlation between the commutator measure and AQC performance is stronger than that between MPS bond dimension and performance, suggesting that Hamiltonian structure influences compressibility independently of, or in addition to, the complexity of the target state.
Circuit Reduction on 80 Qubits: For the hardware validation case, AQCtensor successfully compressed a 10-repetition target evolution into a 5-repetition circuit while maintaining a target-state fidelity of 0.995474. This resulted in a 45.9% reduction in total circuit depth and a 47.6% reduction in two-qubit gate count.
Physical Preservation: Despite the significant resource reduction, the compressed circuit retained physically relevant features:
- Observables: After error mitigation, the compressed circuit showed comparable agreement with reference values for magnetization, spin correlations, and energy conservation as the original circuit.
- Process Diagnostics: Mirror-circuit fidelity estimation showed lower raw error accumulation for the compressed circuit in certain benchmarks.
- Local States: Reduced-state tomography revealed that the hardware-executed compressed circuit maintained local-state fidelity and purity comparable to the original circuit across sampled regions, despite both deviating from the noiseless baseline due to hardware noise.

Significance
The paper demonstrates that AQCtensor can substantially reduce the hardware cost of large-scale quantum simulations (specifically 80-qubit systems) while preserving physically meaningful observables and local state structures. Crucially, it establishes a link between compressibility and the underlying Hamiltonian commutator structure, suggesting that the non-commutativity of Hamiltonian terms (which governs Trotter errors) is a key structural factor determining how easily a deep evolution can be approximated by a shallow variational ansatz. This work moves beyond simple fidelity metrics to show that compressed circuits retain the essential physics of the original dynamics on real quantum hardware, providing a pathway for more efficient quantum simulation on near-term devices.

Technical Summary: Hamiltonian-Dependent Compressibility and Physical Preservation in Approximate Quantum Compilation using Tensor Networks

More like this