Gate Teleportation vs Circuit Cutting in Distributed Quantum Computing

Technical Summary: Gate Teleportation vs. Circuit Cutting in Distributed Quantum Computing

Problem Statement
Distributed quantum computing aims to overcome the qubit count limitations of monolithic processors by interconnecting multiple quantum processor modules. While current near-term approaches rely heavily on circuit cutting (a classical post-processing technique that partitions circuits into sub-circuits), this method incurs an exponential growth in sampling and classical post-processing overhead as the number of cuts increases. Conversely, gate teleportation (using quantum links to perform remote gates) does not inherently suffer from exponential sampling overhead but requires high-fidelity quantum interconnects. Currently, optical links for superconducting qubits—mediated by microwave-to-optical (M2O) transducers—are noisy and probabilistic. The central problem addressed is determining the hardware metrics (specifically transducer noise) required for noisy remote gates to achieve parity with, or outperform, circuit-cutting techniques for generating multipartite entangled states.

Methodology
The authors perform a comparative simulation of Greenberger–Horne–Zeilinger (GHZ) states generated between superconducting qubit modules using two distinct approaches:

1. Remote Gates (Quantum Links): The authors model a distributed architecture where modules are connected via optical fibers. Entanglement is established between communication qubits using M2O transducers. The primary noise source is the "noise added" ($N_{add}$) by the transducer during Bell pair generation. The simulation employs a "two-click" heralded entanglement protocol, which is more robust to noise than one-click protocols. The noisy Bell pair density matrix is injected into a TeleGate primitive to execute remote CNOT gates between data qubits. Local gate errors are modeled as a depolarization channel.
2. Circuit Cutting (Classical Links): The authors simulate the partitioning of the same GHZ circuits using gate cuts. This approach relies on quasiprobability decomposition (QPD) of non-local gates. The simulation calculates the Hellinger fidelity as a function of the number of shots ($N_{shot}$) per sub-circuit, accounting for the exponential sampling cost ($O(9^N)$ for $N$ cuts) required to reconstruct the original circuit statistics.

The study varies the transducer noise ($N_{add}$) for remote gates and the shot budget ($N_{shot}$) for circuit cuts to identify "break-even" points where the fidelities of both approaches converge.

Key Contributions

• Noise Modeling: The paper integrates the specific physical noise characteristics of M2O transducers (specifically $N_{add}$) directly into the Bell pair preparation circuit, rather than using generic depolarization channels.
• Comparative Fidelity Analysis: It provides a direct comparison of Hellinger fidelity between remote teleported gates and gate-cuts for GHZ state generation across varying circuit sizes (2–5 qubits).
• Threshold Identification: The work identifies specific noise thresholds where remote gates become advantageous over circuit cutting, particularly for larger multipartite states.
• Hybrid Algorithm Proposal: The authors propose a "network-aware greedy algorithm" that dynamically selects between gate cuts and remote gates based on available Bell pairs and shot budgets to minimize total runtime.

Results

• Fidelity Regimes: The simulation reveals two distinct regimes. In the low-noise/high-shot regime, local gate errors dominate. In the intermediate regime, remote gate fidelity is limited by transducer noise ($N_{add}$), while circuit-cut fidelity is limited by sampling error (shot budget).
• Break-Even Points: For a fixed shot budget, remote gates can achieve fidelity comparable to circuit cuts if the transducer noise added ($N_{add}$) is below a certain threshold. This threshold increases with the size of the GHZ circuit (number of CNOT gates).
• Quantitative Thresholds: The study finds that for $N_{add}$ values between 0.01 and 0.1, remote gates can achieve parity with circuit-cutting approaches for generating multipartite entangled states. This range is within 10-fold of the current state-of-the-art transducer noise figures ($N_{add} \sim 0.12\text{--}0.14$).
• Scaling Behavior: As the number of CNOT gates increases, circuit-cutting fidelity degrades rapidly due to the exponential sampling overhead. In contrast, remote gate fidelity degrades more slowly with circuit size, provided $N_{add}$ remains low. Consequently, for larger entangled states, the tolerance for transducer noise is relaxed, making remote gates the preferable approach even with current hardware limitations if $N_{add}$ can be reduced by an order of magnitude.

Significance and Claims
The paper claims that its work informs the near-term hardware metrics required for optical interconnects. It argues that a 10-fold reduction in current M2O transducer noise figures would favor the generation of multipartite entangled states with remote gates over circuit cutting due to the latter's exponential sampling overhead. This finding motivates a shift toward a network-aware hybrid quantum-classical distributed computation approach, where sparse, noisy quantum links are utilized in conjunction with circuit-cutting techniques. The authors assert that this hybrid strategy can minimize quantum runtime by leveraging the strengths of both methods: using quantum links where they offer fidelity advantages (large, distributed entangled states) and falling back to circuit cuts where quantum links are unavailable or too noisy. The work does not claim immediate experimental realization but provides a theoretical roadmap for hardware development and compiler design to achieve utility in distributed quantum computing.