arxiv: 2605.11500 · v1 · submitted 2026-05-12 · 🪐 quant-ph

Recognition: no theorem link

Digital Annealer-Assisted Accuracy-First Quantum Circuit Transpilation with Integrated QUBO Mapping and Routing

Kazuma Watanabe , Hideaki Kawaguchi , Shin Nishio , Takahiko satoh

Authors on Pith no claims yet

Pith reviewed 2026-05-13 02:06 UTC · model grok-4.3

classification 🪐 quant-ph

keywords quantum transpilationdigital annealerQUBO optimizationCNOT minimizationqubit mappingroutingNISQ circuitshybrid optimization

0 comments

The pith

Digital Annealer optimization reduces CNOT gate counts by 13.7 percent on average in quantum circuit transpilation.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a method to use the Digital Annealer for solving optimization problems in transpiling quantum circuits, focusing on reducing the number of CNOT gates that introduce errors on noisy hardware. It offers two approaches: one combining the annealer for global qubit mapping with fast heuristic routing, and another using the annealer for both mapping and routing through iterative QUBO solving. Benchmarks show these methods cut CNOT counts more than standard tools like Qiskit on many circuits, especially those with regular patterns, suggesting a way to get better results when compilation time is not the main concern. This matters because fewer CNOTs can mean higher success rates for quantum algorithms run on current limited hardware.

Core claim

The central claim is that leveraging the Digital Annealer to solve QUBO formulations for qubit mapping and routing in an accuracy-first transpilation framework yields lower CNOT counts than heuristic-based compilers, with a Hybrid strategy achieving 13.7% average reduction and up to 57.4% on structured circuits compared to Qiskit's highest optimization level.

What carries the argument

QUBO formulations for mapping and routing solved by the Digital Annealer, which encode the costs of initial qubit placements and subsequent SWAP insertions to minimize total CNOT operations in the final circuit.

If this is right

Structured circuits like GHZ and ASP see the largest improvements because their connectivity patterns reward careful global layout choices.
The Hybrid approach provides a practical balance, using DA only for mapping while relying on fast heuristics for routing.
Full DA encoding of the entire circuit in one pass can achieve even larger gains on some problems but degrades when connectivity is random or concentrated due to growing problem size.
These global methods incur more compute time than pure heuristics but offer value in high-precision settings where gate noise dominates.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar annealing techniques might apply to other quantum compilation tasks such as gate scheduling or error mitigation strategy selection.
Testing the resulting circuits on actual quantum processors could reveal if the CNOT reductions translate directly to higher fidelity beyond simulation.
Future work could explore scaling the QUBO size or hybridizing with machine learning for faster approximation on larger circuits.

Load-bearing premise

That the QUBO models correctly capture the true costs of mapping and routing decisions and that the Digital Annealer consistently finds higher-quality solutions than heuristics within practical time limits.

What would settle it

Running the same benchmark circuits through the proposed methods and Qiskit on a real NISQ device and finding no statistically significant improvement in output fidelity or success probability despite the reported CNOT reductions.

Figures

Figures reproduced from arXiv: 2605.11500 by Hideaki Kawaguchi, Kazuma Watanabe, Shin Nishio, Takahiko satoh.

**Figure 1.** Figure 1: Proposed transpilation workflow integrating Digital Annealer (DA) optimization. The process begins with initial mapping (Step 2), minimizing initial SWAP requirements by anticipating future dependencies. The subsequent iterative wiring phase (Steps 3–5) resolves non-executable gates by solving combinatorial optimization problems on the DA, identifying efficient qubit movement. or non-executable based on t… view at source ↗

**Figure 3.** Figure 3: Execution time comparison. For DA-based methods, light-shaded bars denote cloud communication overhead, and solid bars show core computation time. V. CONCLUSION This paper proposed an annealing-assisted transpilation framework designed to minimize CNOT counts for circuit partitioning workflows. Our “accuracy-first” approach prioritizes gate reduction over compilation latency, reflecting the reality that t… view at source ↗

**Figure 2.** Figure 2: Comprehensive comparison of equivalent CNOT counts across different transpilation methods. (a) Full DA demonstrates superior scaling over ISAAQ for most structured circuits. The ISAAQ value for asp exceeds the plot range (3205, indicated above the bar). (b) The Hybrid approach achieves competitive or lower gate counts compared to industry-standard compilers. overhead of cloud-based global optimization. How… view at source ↗

read the original abstract

In the Noisy Intermediate-Scale Quantum (NISQ) era, limited qubit counts and high gate error rates directly constrain circuit fidelity, making the minimization of CNOT gate counts crucial. While conventional compilers prioritize heuristic efficiency, there is a compelling need for "accuracy-first" transpilation that prioritizes gate reduction over compilation latency. We propose a framework leveraging the Digital Annealer (DA) via two complementary strategies: (1) Hybrid, which uses DA-driven global initial mapping combined with high-speed heuristic routing by Qiskit, and (2) Full DA, which solves mapping and routing as separate DA-assisted QUBO subproblems within an iterative workflow. Benchmarks demonstrate that our Hybrid approach achieves an average CNOT reduction of 13.7 % (up to 57.4 %) compared to Qiskit's highest optimization level, with the largest gains on structured circuits such as GHZ and ASP where the initial layout is decisive. The Full DA approach matches Hybrid on structured circuits and outperforms ISAAQ by 23.1 % on average (maximum 90.8 %), but degrades on circuits with random or concentrated connectivity - exposing a trade-off between QUBO size and solution quality when the entire circuit is encoded in a single annealing pass. Although these global optimizations incur higher computational overhead than pure heuristics, our results indicate that for high-precision workflows where gate noise is the primary bottleneck, DA-assisted global initial placement provides a practical "time-for-quality" trade-off for enhancing the utility of near-term quantum hardware.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper shows DA-driven QUBO mapping and routing can cut CNOT counts versus Qiskit on structured circuits, but gains are uneven and the QUBO details need more scrutiny.

read the letter

The main thing to know is that this work gets measurable CNOT reductions by feeding mapping and routing into a Digital Annealer as QUBO problems, with the hybrid version averaging 13.7 percent fewer CNOTs than Qiskit at high optimization and bigger wins on GHZ and ASP circuits. The full iterative DA version matches that on structured cases and beats ISAAQ by 23.1 percent on average, though it drops off on random-connectivity circuits when the whole problem is solved in one pass. That trade-off between QUBO size and solution quality is the central practical finding. What is new is the specific pairing of DA global initial mapping with either heuristic or DA-based routing, presented as two complementary workflows rather than a pure heuristic or pure ILP approach. The paper does a decent job laying out the accuracy-first motivation and giving named-circuit benchmarks against real tools. The numbers are concrete enough to evaluate the claim that global optimization helps when gate noise dominates. The soft spots sit in the methods and scaling. The abstract supplies no error bars, no circuit-selection protocol, and no comparison of DA outputs to exact solvers on small instances, so it is unclear how much the reported averages depend on particular penalty coefficients or annealing schedules. The degradation on non-structured circuits suggests the QUBO objective may not fully capture cumulative swap and crosstalk costs across layers, or that the annealer simply returns locally good solutions within the time limit. This is aimed at quantum compiler developers and NISQ application teams who already care about gate-count minimization and are willing to spend extra compile time for it. Readers working on heuristic versus global optimization trade-offs will get the most out of the hybrid versus full comparison. It deserves a serious referee because the empirical claims are testable against existing baselines and the hardware (Digital Annealer) is real, even if the QUBO formulation and statistical reporting need tightening. I would send it to review.

Referee Report

3 major / 3 minor

Summary. The paper proposes two Digital Annealer (DA)-assisted strategies for accuracy-first quantum circuit transpilation: a Hybrid approach combining DA-driven global initial mapping with Qiskit heuristic routing, and a Full DA approach solving mapping and routing as separate QUBO subproblems in an iterative workflow. It reports empirical benchmarks showing an average 13.7% CNOT reduction (up to 57.4%) for Hybrid versus Qiskit's highest optimization level (largest gains on structured circuits like GHZ and ASP) and 23.1% average improvement (max 90.8%) for Full DA versus ISAAQ, while noting degradation on random/concentrated-connectivity circuits due to QUBO size versus solution quality trade-offs.

Significance. If the empirical claims hold after proper validation, the work demonstrates a practical time-for-quality trade-off for NISQ transpilation by using global DA optimization on QUBO formulations to reduce CNOT counts, which could improve circuit fidelity where gate noise dominates; the distinction between Hybrid and Full DA strategies usefully highlights when full-circuit encoding is viable.

major comments (3)

[Abstract / Benchmarks] Abstract and benchmarks section: The headline CNOT reductions (13.7% Hybrid average, 23.1% Full DA vs ISAAQ) are reported as concrete percentages without error bars, explicit circuit-selection criteria, number of instances per family, or statistical tests; this is load-bearing for the central claim that DA-driven QUBO solutions are superior to heuristics.
[Methods / QUBO Mapping and Routing] QUBO formulation and methods: No validation is provided that the QUBO objectives exactly encode swap depths, crosstalk, or cumulative routing costs across layers, nor any comparison of DA outputs to exact ILP solutions on small instances to confirm solution quality within allotted annealing times; this directly affects the weakest assumption that DA reliably outperforms heuristics.
[Results / Discussion] Results on non-structured circuits: The noted degradation when encoding the full circuit in one annealing pass for random or concentrated-connectivity circuits is acknowledged but not accompanied by analysis of annealing schedules, time caps, or QUBO penalty coefficient sensitivity, leaving the scalability limitation unquantified.

minor comments (3)

[Methods] The description of the iterative workflow for Full DA could be clarified with pseudocode or a flowchart to distinguish the separate mapping and routing QUBO subproblems.
[Introduction] Missing references to prior QUBO-based transpilation literature and Digital Annealer applications in quantum compilation.
[Results] Benchmark tables or figures lack captions detailing the exact circuit families, qubit counts, and baseline configurations used for comparison.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thorough review and constructive suggestions. We have carefully considered each major comment and provide point-by-point responses below, indicating the revisions we will incorporate in the updated manuscript.

read point-by-point responses

Referee: [Abstract / Benchmarks] Abstract and benchmarks section: The headline CNOT reductions (13.7% Hybrid average, 23.1% Full DA vs ISAAQ) are reported as concrete percentages without error bars, explicit circuit-selection criteria, number of instances per family, or statistical tests; this is load-bearing for the central claim that DA-driven QUBO solutions are superior to heuristics.

Authors: We agree that additional details and statistical support would strengthen the presentation of our empirical results. In the revised version, we will expand the Abstract and Benchmarks section to include: explicit criteria for circuit selection and the number of instances per circuit family; standard deviations or error bars alongside the average CNOT reductions; and appropriate statistical tests (such as Wilcoxon signed-rank tests) to assess the significance of the observed improvements over baselines. These additions will provide a more rigorous foundation for our claims. revision: yes
Referee: [Methods / QUBO Mapping and Routing] QUBO formulation and methods: No validation is provided that the QUBO objectives exactly encode swap depths, crosstalk, or cumulative routing costs across layers, nor any comparison of DA outputs to exact ILP solutions on small instances to confirm solution quality within allotted annealing times; this directly affects the weakest assumption that DA reliably outperforms heuristics.

Authors: We acknowledge this gap in validation. We will add a new subsection to the Methods section that (i) explicitly details how the QUBO objectives encode swap depths, crosstalk penalties, and cumulative routing costs across circuit layers, and (ii) presents a validation study comparing DA solutions against exact ILP solvers on small-scale instances (circuits with up to 8-10 qubits) to verify solution quality within the given annealing time budgets. This will confirm the fidelity of our QUBO formulations. revision: yes
Referee: [Results / Discussion] Results on non-structured circuits: The noted degradation when encoding the full circuit in one annealing pass for random or concentrated-connectivity circuits is acknowledged but not accompanied by analysis of annealing schedules, time caps, or QUBO penalty coefficient sensitivity, leaving the scalability limitation unquantified.

Authors: We will enhance the Results and Discussion sections with a dedicated analysis of the scalability limitations. Specifically, we will include sensitivity studies varying annealing schedules, time caps, and penalty coefficients for the Full DA approach on random and concentrated-connectivity circuits. This will quantify the trade-offs and provide clearer guidelines on the applicability of the method, addressing the unquantified aspects noted. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical benchmarks against external tools with no self-referential derivation.

full rationale

The paper presents a practical framework for DA-assisted QUBO-based mapping and routing in quantum transpilation, with claims resting entirely on empirical CNOT-count comparisons to Qiskit (level 3) and ISAAQ across benchmark circuits. No equations, uniqueness theorems, or ansatzes are derived or invoked that reduce by construction to the paper's own fitted parameters, self-citations, or input data. The reported averages (13.7 % Hybrid, 23.1 % Full DA) are direct measurement outcomes, not predictions forced by the method's internal definitions. The approach is self-contained against external baselines.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Abstract-only review yields minimal ledger entries; the framework implicitly assumes standard QUBO encoding of swap costs and that the annealer hardware behaves as a reliable global optimizer.

free parameters (1)

QUBO penalty coefficients
Weights balancing mapping and routing objectives are required for any QUBO formulation but are not reported in the abstract.

axioms (1)

domain assumption Digital Annealer returns high-quality solutions to the formulated QUBO instances within practical time limits
This assumption underpins the claim that DA-assisted placement outperforms heuristic routing.

pith-pipeline@v0.9.0 · 5590 in / 1446 out tokens · 48396 ms · 2026-05-13T02:06:05.139191+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages · 1 internal anchor

[1]

Simulating large quantum circuits on a small quantum computer,

T. Peng, A. W. Harrow, M. Ozols, and X. Wu, “Simulating large quantum circuits on a small quantum computer,”Physical Review Letters, vol. 125, no. 15, Oct. 2020. [Online]. Available: http://dx.doi.org/10.1103/PhysRevLett.125.150504

work page doi:10.1103/physrevlett.125.150504 2020
[2]

Constructing a virtual two-qubit gate by sampling single-qubit operations,

K. Mitarai and K. Fujii, “Constructing a virtual two-qubit gate by sampling single-qubit operations,”New Journal of Physics, vol. 23, no. 2, p. 023021, feb 2021. [Online]. Available: https: //doi.org/10.1088/1367-2630/abd7bc

work page doi:10.1088/1367-2630/abd7bc 2021
[3]

Overhead for simulating a non-local channel with local channels by quasiprobability sampling,

——, “Overhead for simulating a non-local channel with local channels by quasiprobability sampling,”Quantum, vol. 5, p. 388, Jan. 2021. [Online]. Available: https://doi.org/10.22331/q-2021-01-28-388

work page doi:10.22331/q-2021-01-28-388 2021
[4]

Approximation and hardness for token swapping,

T. Miltzow, L. Narins, Y . Okamoto, G. Rote, A. Thomas, and T. Uno, “Approximation and hardness for token swapping,” 2016. [Online]. Available: https://arxiv.org/abs/1602.05150

work page arXiv 2016
[5]

On the complexity of quantum circuit compilation,

A. Botea, A. Kishimoto, and R. Marinescu, “On the complexity of quantum circuit compilation,” inProceedings of the Eleventh International Symposium on Combinatorial Search (SOCS 2018). AAAI Press, 2018, pp. 138–142. [Online]. Available: https://ojs.aaai. org/index.php/SOCS/article/view/18463

work page 2018
[6]

Qiskit/qiskit: Qiskit 2.0.1,

M. Treinishet al., “Qiskit/qiskit: Qiskit 2.0.1,” may 2025. [Online]. Available: https://doi.org/10.5281/zenodo.15442200

work page doi:10.5281/zenodo.15442200 2025
[7]

Quantum computing with Qiskit

A. Javadi-Abhari, M. Treinish, K. Krsulich, C. J. Wood, J. Lishman, J. Gacon, S. Martiel, P. D. Nation, L. S. Bishop, A. W. Cross, B. R. Johnson, and J. M. Gambetta, “Quantum computing with Qiskit,” 2024. [Online]. Available: https://arxiv.org/abs/2405.08810

work page internal anchor Pith review Pith/arXiv arXiv 2024
[8]

In: Proceedings of the Twenty-Fourth International Conference on Architectural Support for Programming Languages and Operating Systems

G. Li, Y . Ding, and Y . Xie, “Tackling the qubit mapping problem for NISQ-Era quantum devices,” inProceedings of the Twenty-Fourth International Conference on Architectural Support for Programming Languages and Operating Systems, ser. ASPLOS ’19. New York, NY , USA: Association for Computing Machinery, 2019, p. 1001–1014. [Online]. Available: https://doi...

work page doi:10.1145/3297858.3304023 2019
[9]

Quantum Science and Technol- ogy6(1), 014003 (2020).https://doi.org/10.1088/2058-9565/ab8e92 A Notation T able 1.Notation used in the manuscript

S. Sivarajah, S. Dilkes, A. Cowtan, W. Simmons, A. Edgington, and R. Duncan, “t|ket⟩: a retargetable compiler for nisq devices,”Quantum Science and Technology, vol. 6, no. 1, p. 014003, Nov. 2020. [Online]. Available: http://dx.doi.org/10.1088/2058-9565/ab8e92

work page doi:10.1088/2058-9565/ab8e92 2020
[10]

Digital annealer for high-speed solving of combinatorial optimization problems and its applications,

S. Matsubara, M. Takatsu, T. Miyazawa, T. Shibasaki, Y . Watanabe, K. Takemoto, and H. Tamura, “Digital annealer for high-speed solving of combinatorial optimization problems and its applications,” in2020 25th Asia and South Pacific Design Automation Conference (ASP-DAC), 2020, pp. 667–672. [Online]. Available: https://ieeexplore. ieee.org/stamp/stamp.jsp...

work page 2020
[11]

ISAAQ: Ising machine assisted quantum compiler,

S. Naito, Y . Hasegawa, Y . Matsuda, and S. Tanaka, “ISAAQ: Ising machine assisted quantum compiler,” 2023. [Online]. Available: https://arxiv.org/abs/2303.02830

work page arXiv 2023
[12]

Fujitsu and RIKEN develop su- perconducting quantum computer at the RIKEN RQC-Fujitsu col- laboration center, paving the way for platform for hybrid quan- tum computing,

Fujitsu Limited and RIKEN, “Fujitsu and RIKEN develop su- perconducting quantum computer at the RIKEN RQC-Fujitsu col- laboration center, paving the way for platform for hybrid quan- tum computing,” https://www.fujitsu.com/global/about/resources/news/ press-releases/2023/1005-01.html, Oct. 2023, accessed: 2025-12-31

work page 2023
[13]

Japanese joint research group launches quantum com- puting cloud service,

Fujitsu Limited, “Japanese joint research group launches quantum com- puting cloud service,” https://www.fujitsu.com/global/about/resources/ news/press-releases/2023/0324-01.html, Mar. 2023, accessed: 2025-12- 31

work page 2023
[14]

QASMBench: A low-level quantum assembly benchmark suite,

A. Li, “QASMBench: A low-level quantum assembly benchmark suite,” https://github.com/pnnl/QASMBench, 2020, accessed: 2025-12-31

work page 2020
[15]

QASMBench: A low-level quantum assembly benchmark suite for NISQ evaluation and simulation,

A. Li, S. Krishnamoorthy, C. Xie, S. Boettcher, R. J. Williams, and K. Barker, “QASMBench: A low-level quantum assembly benchmark suite for NISQ evaluation and simulation,” in2021 IEEE International Parallel and Distributed Processing Symposium (IPDPS), 2021, pp. 642– 651

work page 2021
[16]

influence

H. Zou, M. Treinish, K. Hartman, A. Ivrii, and J. Lishman, “Lightsabre: A lightweight and enhanced sabre algorithm,” 2024. [Online]. Available: https://arxiv.org/abs/2409.08368

work page arXiv 2024
[17]

Fixstars amplify sdk,

Fixstars Amplify, “Fixstars amplify sdk,” https://amplify.fixstars.com/, 2021

work page 2021
[18]

ABCI-Q: Quantum-Classical Hy- brid Computing Infrastructure,

National Institute of Advanced Industrial Science and Technology (AIST), Global Research and Development Center for Business by Quantum-AI Technology (G-QuAT), “ABCI-Q: Quantum-Classical Hy- brid Computing Infrastructure,” https://unit.aist.go.jp/g-quat/HowToUse/ en/abci q/index.html, 2024, accessed: 2026-01-15

work page 2024