PaperlayerView exact revision on arXiv

Source-linked AI summary

Asymmetric quantum error correction efficiently tackles application-specific noise effects

Abhishek Yadav, Peter K. Schuhmacher, Michael Epping

arXiv:2609.07235v1quant-phcs.ET

TL;DR

The paper asks how bit-flip and phase-flip noise affect a QAOA application differently, and whether error correction can be tailored to that impact. It compares noise models and circuit decompositions, then illustrates asymmetric correction with a quantum parity code. Bit-flip noise causes the larger degradation in the studied FGA instance, and a smaller asymmetric code performs nearly as well as a larger symmetric choice.

  • Problem

    The work examines whether bit-flip and phase-flip errors affect QAOA applications differently, since this asymmetry may matter when choosing resource-efficient error correction.

    Source: §I · pp. 1–2

  • Method

    The study compares Pauli noise in gate-level and layer-level models across circuit decompositions, then uses a quantum parity code to illustrate asymmetric error correction.

    Source: §I · p. 2, D. Gate-Level and Layer-Level Noise… · pp. 9–10, §VI · p. 18

  • Results

    For the studied FGA instance, bit-flip noise degrades optimized QAOA cost more than phase-flip noise, most clearly in the layer-level and native-RZZ simulations.

    Source: C. Results · p. 13, §VI · p. 18

  • Takeaways & Limitations

    The (5,3)-QPC performs as well as the (5,5)-QPC while using ten fewer qubits, supporting error-correction resource allocation by application impact as well as physical error probability.

    Source: §VI · p. 18

Abstract

from arXiv · show

Noise is a major challenge for current quantum computers. It can be broadly categorized into bit-flip and phase-flip errors. These two types do not necessarily affect the executed algorithm, thus also the application, in the same way. We illustrate this general effect for the example of the quantum approximate optimization algorithm (QAOA) applied to a small instance of the flight-gate assignment (FGA) problem. We compare bit-flip and phase-flip Pauli noise under both layer-level and gate-level noise models, using two circuit decompositions of the same ideal QAOA unitary: a CNOT-based decomposition and a native-$R_{ZZ}$ decomposition. In the simulations, bit-flip noise produces the larger degradation in the performance of the quantum optimization. The asymmetry is most visible in the layer-level and native-$R_{ZZ}$ simulations. We explain this by how the errors affect mixing, final measurements, and how they propagate inside the circuit. We then exploit these insights to tackle noise particularly efficiently using asymmetric error-correcting codes. As an illustration, we use the quantum parity code (QPC), a generalization of the 9-qubit Shor code, and show that a smaller asymmetric code can achieve nearly the same improvement as a larger symmetric choice. This demonstrates that error-correction resources should be assigned not only according to physical error rates, but also according to how strongly each error channel affects the application. As a result, asymmetric quantum error correction proves useful even in cases where the noise model is symmetric. Finally, we discuss how information about the noise obtained through calibration can be exploited in our approach.

I. INTRODUCTION

The paper examines how bit-flip and phase-flip noise affect QAOA differently, and whether asymmetric error correction can address the more damaging channel efficiently. It studies this question using multiple circuit and noise-model descriptions and a flight-gate assignment example.

II. BACKGROUND

  • QAOA alternates cost- and mixer-Hamiltonian evolutions, with variational angles optimized using measurements from a finite-depth circuit.The cost Hamiltonian encodes candidate solution quality; the mixer enables transitions between assignment strings.Source: §I · p. 1

I. INTRODUCTION

  • The study compares bit-flip and phase-flip noise across gate-level and layer-level models and two gate sets, then uses the observed asymmetry to motivate asymmetric error correction.The FGA problem is the application example.Source: §I · pp. 2–3

A. Instance Definition

  • In the FGA example, assignments minimize passenger commuting time while assigning each flight to one gate and preventing overlapping flights from sharing a gate.The problem is formulated as a QUBO and mapped to an Ising Hamiltonian for QAOA.Source: A. Instance Definition · p. 3, §II · p. 3

1. Objective Function

The section defines passenger walking and transfer time as the FGA objective, then incorporates assignment feasibility into an unconstrained QUBO through penalties.

1. Objective Function

  • The objective adds gate-specific walking times for departing and arriving passengers to quadratic transfer-time contributions between flights.The transfer term depends on the gates assigned to both flights.Source: §1 · p. 3
  • Feasibility requires each flight to use exactly one gate and conflicting flights not to share a gate.The unpenalized passenger-time objective alone does not enforce these constraints.Source: §1 · pp. 3–4

B. QUBO Formulation

  • The constrained FGA problem becomes a QUBO by adding positive penalties for violations of the unique-gate and conflict constraints.The penalty weights must make infeasible assignments less favorable than feasible assignments.Source: B. QUBO Formulation · p. 4

1. Penalty Weight Estimation

Penalty weights encode FGA feasibility in the QUBO, which is then mapped to an Ising cost Hamiltonian and paired with a transverse-field mixer for QAOA.

1. Penalty Weight Estimation

  • Penalty weights must enforce feasibility without distorting the optimization landscape unnecessarily; the study uses weights from prior DLR studies of FGA.Too-small weights may favor infeasible assignments, while too-large weights can make the landscape harder to explore.Source: §1 · pp. 4–5

C. Ising Hamiltonians of the QAOA

  • The QUBO is mapped through binary-to-spin variables into a diagonal Ising cost Hamiltonian whose ground state encodes the penalized assignment optimum.The simulated instance uses three flights and two gates, requiring six binary variables and six qubits.Source: C. Ising Hamiltonians of the QAOA · p. 5
  • The Hamiltonian terms encode passenger-time costs and assignment penalties, while the transverse-field mixer enables transitions between assignment bit strings.Constraint terms penalize missing or multiple gate assignments and conflicting flights sharing a gate.Source: C. Ising Hamiltonians of the QAOA · pp. 5–6

III. QAOA CIRCUIT DECOMPOSITION AND NOISE MODELS

The section distinguishes ideal QAOA evolution from its noisy gate implementation. It introduces two decompositions and noise-insertion models for the simulations.

III. QAOA CIRCUIT DECOMPOSITION AND NOISE MODELS

  • At the Hamiltonian level, alternating cost and mixer evolutions define the ideal QAOA unitary.The unitary maps the initial state to the final variational state in the noiseless circuit.Source: §III · p. 6
  • A noisy simulation implements that logical evolution with elementary gates and inserts noise channels at a selected circuit level.Source: §III · p. 6
  • The section introduces two gate-set decompositions and Pauli noise channels for gate-level and layer-level simulations.Source: §III · p. 6

A. Ideal QAOA Ansatz as a Unitary Evolution

The ideal depth-d QAOA ansatz alternates cost and mixer evolutions from a uniform-superposition initial state. Its variational objective is the final state's expected cost, which the classical optimizer minimizes.

A. Ideal QAOA Ansatz as a Unitary Evolution

B. Gate-Set Decompositions of the QAOA Layer

The same ideal QAOA layer can be compiled using a CNOT-based gate set or native RZZ interactions. These decompositions differ in their gate sequences and therefore in the locations exposed to gate-level noise.

B. Gate-Set Decompositions of the QAOA Layer

C. Pauli Noise Channels

The simulations use Pauli bit-flip and phase-flip channels with independently varied probabilities to compare their effects on QAOA. Their relevance follows from the different Pauli structure of the cost and mixer Hamiltonians.

C. Pauli Noise Channels

  • Single-qubit bit-flip and phase-flip channels are generated by Pauli-X and Pauli-Z errors, respectively, and reduce to the identity when their error probability is zero.Source: C. Pauli Noise Channels · p. 8
  • Bit-flip and phase-flip probabilities are varied independently, allowing comparisons of X-type and Z-type errors on the same QAOA circuit.Source: C. Pauli Noise Channels · pp. 8–9
  • Two-qubit channels distribute each error probability across errors on either qubit or both; successive bit- and phase-flip channels also include mixed events at order p_X p_Z.Source: C. Pauli Noise Channels · p. 9
  • The cost Hamiltonian is built from Pauli-Z operators, whereas the mixer is built from Pauli-X operators, so the two error types need not disturb QAOA evolution in the same way.Source: C. Pauli Noise Channels · p. 9

D. Gate-Level and Layer-Level Noise Insertion

The paper compares gate-level noise, inserted after each elementary gate, with layer-level noise, applied after a complete QAOA layer. The models represent different assumptions about noise accumulation and whether compilation details are resolved.

D. Gate-Level and Layer-Level Noise Insertion

IV. NOISE SIMULATION

The simulations compare ideal and noisy QAOA on a six-qubit flight-gate assignment instance, using exact density-matrix evolution and repeated optimization restarts. They evaluate performance across gate-level and layer-level noise models and two circuit decompositions, without finite-shot sampling noise.

IV. NOISE SIMULATION

  • The six-qubit FGA instance is formulated as a QUBO and mapped to an Ising Hamiltonian before QAOA circuits and noise models are simulated.Quark, PennyLane, and Qiskit Aer are used for formulation, Hamiltonian and mixer handling, and exact noisy dynamics, respectively.Source: §IV · p. 10, A. Simulation Parameters · p. 10

A. Simulation Parameters

  • Exact density-matrix evolution evaluates the cost for the ideal or noisy state, excluding finite-shot sampling fluctuations.The noisy cost is attributed to the applied bit-flip and phase-flip channels rather than measurement statistics.Source: A. Simulation Parameters · p. 10
  • Several independent random initial parameter vectors are optimized because the non-convex QUBO may yield initialization-dependent local optima.The best optimized cost represents performance, while the best-to-worst interval shows variation across restarts.Source: A. Simulation Parameters · p. 10
  • The protocol compares ideal and noisy circuits under gate-level and layer-level Pauli noise, using CNOT- and native-RZZ-based gate sets.The cost Hamiltonian, initialization rule, and optimization protocol are held fixed across cases.Source: A. Simulation Parameters · p. 10

B. Performance Metric

The section defines the residual-cost metric and compares bit-flip and phase-flip noise across layer-level and gate-level QAOA simulations. Bit-flip noise causes the larger degradation, with the asymmetry clearest in the layer-level and native-RZZ simulations.

B. Performance Metric

  • ∆ measures optimized residual cost against the exact ground-state energy, with smaller values indicating better performance.The shaded spread reflects sensitivity to random parameter initialization; asymmetric sweeps vary pX and pZ independently.Source: B. Performance Metric · p. 11

C. Results

  • In QAOA, phase flips commute with cost evolution but not the mixer; bit flips commute with the mixer, not cost, and alone can spread through multi-qubit cost interactions.At computational-basis measurement, only bit flips directly change assignment populations and cost; phase flips act mainly through subsequent mixer rotations.Source: C. Results · p. 11
  • In layer-level runs, ∆ rises with noise but rises considerably more under bit flips, while residual costs remain smallest among the models.Noise is applied once after each complete QAOA layer, rather than after every elementary gate.Source: C. Results · p. 11, Figure 2 · p. 12
  • Across the simulations, bit-flip noise consistently causes larger degradation in ∆; the asymmetry is clearest at layer level and with native-RZZ gates.In the CNOT-based gate-level model, bit flips remain dominant, but phase-flip noise also becomes significant.Source: Figure 2 · p. 12, Figure 3 · p. 13, C. Results · p. 13
  • With CNOT-RZ-CNOT ZZ interactions, bit flips remain more damaging, but phase flips can propagate from target to control and become significant.The CNOT-based curve separation is less pronounced than with native-RZZ gates.Source: C. Results · p. 12, Figure 3 · p. 13
  • Native-RZZ yields the clearest asymmetry: bit-flip noise raises ∆ more than phase-flip noise, especially at larger depths and noise probabilities.Each two-qubit Ising interaction uses one RZZ gate instead of a longer CNOT-based sequence, reducing noisy two-qubit locations.Source: C. Results · pp. 12–13, Figure 4 · p. 14

V. TAILORED ASYMMETRICAL QUANTUM ERROR CORRECTION

For the studied FGA instance, bit-flip noise degrades QAOA more than phase-flip noise, motivating error correction weighted toward the channel that most harms the objective. The QPC illustration shows that an asymmetric code can match a larger code with fewer physical qubits.

V. TAILORED ASYMMETRICAL QUANTUM ERROR CORRECTION

  • Bit-flip noise harms the FGA QAOA objective more than phase-flip noise, so code resources should target the error channel that most degrades application performance, not simply the most probable one.The paper describes this asymmetry as relevant even when the physical noise is symmetric.Source: §V · pp. 14–15
  • The QPC concatenates an inner repetition code of length n2 and an outer repetition code of length n1, using N = n1n2 physical qubits to encode one logical qubit.The code is chosen to illustrate asymmetry, although better error-correction codes are known.Source: §V · p. 16
  • The analysis estimates QEC performance by replacing physical error rates with logical error rates in a polynomial fit to the simulated QAOA results.This lets the authors analyze application-specific QEC effects without rerunning the simulation.Source: §V · pp. 15–16
  • The (5,3)-QPC achieves the same results as the (5,5)-QPC using 15 rather than 25 physical qubits.The (5,5) code also beats the standard 9-qubit Shor code (3,3) in the plotted layer-level, depth-2 case.Source: §V · p. 17, Figure 9 · p. 18
  • In the targeted early fault-tolerant regime, syndrome-extraction gadgets add considerable noise, which ideally should be accounted for when choosing code asymmetry.Calibration can estimate noise effects on stabilizer expectation values and inform adjustments at runtime.Source: §V · p. 17

VI. CONCLUSION

For a six-qubit FGA instance, the study compares bit-flip and phase-flip noise in QAOA and finds bit-flip noise more damaging to the optimized cost. It then shows that an asymmetric QPC can match a larger code with fewer qubits, supporting application-informed allocation of error-correction resources.

VI. CONCLUSION

  • Bit-flip noise degrades the optimized QAOA cost more than phase-flip noise for the studied FGA instance, most clearly in the layer-level and native-RZZ simulations.In the CNOT-based decomposition, additional gates for ZZ interactions introduce more noisy locations and allow phase-flip errors to propagate, though bit-flip noise remains dominant.Source: §VI · p. 18
  • With both error types present, residual cost rises as either probability increases; layer-level noise causes less overall degradation than gate-level noise, and sensitivity depends on circuit compilation.The layer-level model inserts noise once per QAOA layer, while gate-level models insert it after each elementary gate.Source: §VI · p. 18
  • The (5,3)-QPC performs as well as the (5,5)-QPC while using ten fewer qubits.This supports assigning correction resources according to which error channel most harms the objective, rather than only by physical error probability.Source: §VI · p. 18
  • The authors identify combining their analysis with calibration information, and testing asymmetric QEC on real devices and larger instances, as future directions.Calibration experiments such as the method introduced in are one possible source of noise information.Source: §VI · pp. 18–19
Loading 2609.07235v1…