arxiv: 2603.28413 · v3 · submitted 2026-03-30 · 🪐 quant-ph

Recognition: no theorem link

Resource-efficient quantum approximate optimization algorithm via Bayesian optimization and maximum-probability evaluation

Siran Zhang , Shuming Cheng

Authors on Pith no claims yet

Pith reviewed 2026-05-14 21:50 UTC · model grok-4.3

classification 🪐 quant-ph

keywords QAOAMaxCutBayesian optimizationadaptive shot allocationquantum optimizationvariational quantum algorithmscombinatorial optimization

0 comments

The pith

QAOA reaches comparable MaxCut quality with fewer shots by optimizing the cut value of the single most probable bitstring.

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

The paper replaces the standard expectation-value objective in QAOA with the cut value of the highest-probability measured bitstring. It pairs this objective with Bayesian optimization for parameter search and an adaptive shot allocator that uses mode and cut-value variance to decide how many measurements to take at each step. On 3-regular MaxCut graphs, both weighted and unweighted, the resulting discrete solutions match the quality of conventional QAOA while usually requiring fewer total shots to reach the same final accuracy. A reader cares because near-term quantum hardware has tight limits on the number of circuit executions it can afford before noise or time budgets are exhausted, so any method that trims the measurement count without sacrificing solution quality makes the algorithm more practical.

Core claim

By using the cut value of the most probable measured bitstring as the optimization objective, combined with Bayesian optimization and dual-criterion adaptive shot allocation based on mode and normalized cut-value variance, the framework produces discrete solutions of quality comparable to expectation-based QAOA while typically requiring fewer total shots to reach the same final mode accuracy on 3-regular MaxCut instances.

What carries the argument

The maximum-probability bitstring cut value, serving as the objective function and evaluated under adaptive shot allocation driven by mode and cut-value variance.

If this is right

The new objective allows QAOA to converge to high-quality MaxCut solutions under tighter measurement budgets than the expectation approach.
Adaptive allocation based on mode stops wasting shots once the dominant bitstring is sufficiently reliable.
The efficiency gain appears for both unweighted and weighted 3-regular graphs under identical parameter-search and allocation rules.
Bayesian optimization integrates directly with the max-probability objective, enabling efficient parameter tuning with the reduced shot counts.

Where Pith is reading between the lines

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

The same max-probability objective could be tested in other variational quantum algorithms that return discrete samples rather than expectation values.
Pairing the method with device-specific noise models or error mitigation might extend the shot savings to real hardware.
Scaling tests on graphs larger than the 3-regular cases studied would reveal whether the shot reduction persists as problem size increases.

Load-bearing premise

That optimizing the cut value of the single most probable bitstring reliably yields high-quality discrete solutions without bias introduced by the adaptive shot-allocation rules.

What would settle it

A side-by-side run of the new method and standard expectation-based QAOA on the same larger MaxCut instances, counting total shots each needs to hit a fixed target mode accuracy; reversal of the claimed efficiency would falsify the result.

Figures

Figures reproduced from arXiv: 2603.28413 by Shuming Cheng, Siran Zhang.

**Figure 1.** Figure 1: FIG. 1: Comparison between the conventional expectation and the mode-based objective on unweighted 3-regular [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗

**Figure 2.** Figure 2: FIG. 2: Comparison between the conventional expectation and the mode-based objective on weighted 3-regular [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

**Figure 4.** Figure 4: FIG. 4: Resource saving rate relative to conventional [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗

**Figure 6.** Figure 6: FIG. 6: Final mode accuracy at different qubit scales. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗

**Figure 5.** Figure 5: FIG. 5: Accuracy–resource Pareto comparison in the [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗

**Figure 7.** Figure 7: FIG. 7: Results of the circuit-depth-scaling experiment. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗

**Figure 8.** Figure 8: FIG. 8: Accuracy–resource Pareto comparison under the [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗

**Figure 9.** Figure 9: FIG. 9: Auxiliary results for the noise sweep. The left [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗

read the original abstract

The quantum approximate optimization algorithm (QAOA) is a leading variational approach to combinatorial optimization, but its practical performance depends strongly on objective design, parameter search, and shot allocation. We present a resource-efficient QAOA framework that uses the cut value of the most probable measured bitstring as the optimization objective, combines it with Bayesian optimization, and adaptively allocates shots using dual criteria based on mode confidence and normalized cut-value variance. Numerical experiments on 3-regular MaxCut show that, for both unweighted and weighted instances, the proposed scheme achieves discrete-solution quality comparable to that of the conventional expectation-based objective while typically requiring fewer total shots to reach the same final mode accuracy. These results indicate that reorganizing QAOA around the maximum-probability bitstring provides an effective route to improving practical performance under limited measurement budgets.

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 a mode-based objective for QAOA plus Bayesian optimization and dual adaptive shots can match standard performance on MaxCut with lower total shots, but the baseline comparison may be confounded by allocation rules.

read the letter

The core idea is to replace the usual expectation-value objective in QAOA with the cut value of the single most probable bitstring, then tune parameters via Bayesian optimization and allocate shots adaptively using mode and variance. On 3-regular MaxCut the numerical runs reach similar final mode accuracy with fewer shots overall for both weighted and unweighted graphs. That directly targets the limited-shot regime that matters on current hardware. The adaptive rules based on mode and normalized variance are a practical addition that focuses measurements where uncertainty is high. Bayesian optimization is applied in a standard way but fits the setting without extra overhead. The experiments are on standard benchmarks, which makes the results easy to check. The soft spot is the baseline. The abstract does not state whether the conventional expectation runs used identical dual-criteria adaptive allocation. If they used fixed or uniform shots instead, the reported savings could come from the allocation policy rather than the mode objective itself. A matched control or explicit statement in the full text would remove that ambiguity. Instance sizes, number of trials, and error bars are also not detailed in the summary, so the statistical strength is hard to gauge. This paper is aimed at people who implement QAOA on near-term devices and need to stretch limited measurement budgets. A reader looking for concrete tweaks to variational algorithms will find usable ideas here. I would send it to peer review. The empirical focus on shot efficiency is worth referee time even if the comparison needs tightening.

Referee Report

1 major / 2 minor

Summary. The paper introduces a resource-efficient QAOA framework that optimizes the cut value of the most probable measured bitstring as the objective function, integrates Bayesian optimization for parameter search, and uses adaptive shot allocation based on mode confidence and normalized cut-value variance. Numerical experiments on 3-regular MaxCut instances demonstrate that this approach achieves discrete solution quality comparable to conventional expectation-based QAOA while requiring fewer total shots to reach the same mode accuracy.

Significance. If the comparison to the baseline is fair and the efficiency gains are attributable to the maximum-probability objective rather than the adaptive allocation alone, this work could offer a practical improvement for QAOA under limited measurement resources, potentially advancing variational quantum algorithms for combinatorial optimization.

major comments (1)

[Numerical experiments] Numerical experiments section: the description of the conventional 'expectation-based' baseline does not explicitly state whether it employs the same adaptive shot allocation rules (dual criteria based on mode confidence and normalized cut-value variance) as the proposed method. This omission leaves open the possibility that the reported reduction in total shots is due to the allocation heuristic rather than the change in optimization objective, undermining the central efficiency claim.

minor comments (2)

[Abstract] The abstract lacks details on the sizes of the 3-regular MaxCut instances tested, the number of independent runs, statistical significance measures, exact baseline implementations, and error bars on the reported shot counts and solution qualities.
[Abstract] It would be helpful to clarify if the Bayesian optimization is applied identically in both the proposed and baseline methods to isolate the effect of the objective function.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for highlighting the need for greater clarity in the description of the baseline. We agree that the current text leaves ambiguity regarding the shot allocation strategy used for the conventional expectation-based QAOA. We address this point below and will revise the manuscript to remove any such ambiguity.

read point-by-point responses

Referee: [Numerical experiments] Numerical experiments section: the description of the conventional 'expectation-based' baseline does not explicitly state whether it employs the same adaptive shot allocation rules (dual criteria based on mode confidence and normalized cut-value variance) as the proposed method. This omission leaves open the possibility that the reported reduction in total shots is due to the allocation heuristic rather than the change in optimization objective, undermining the central efficiency claim.

Authors: We thank the referee for this observation. The manuscript does not explicitly state that the conventional expectation-based baseline employs the same dual-criteria adaptive shot allocation. In the numerical experiments, both the proposed maximum-probability objective and the expectation-based baseline were run with identical adaptive shot allocation rules (mode confidence and normalized cut-value variance thresholds) to ensure that any observed reduction in total shots could be attributed to the change in optimization objective. We will revise the Numerical experiments section to state this explicitly, including a brief description of the shared allocation procedure and a note that the comparison isolates the effect of the objective function. revision: yes

Circularity Check

0 steps flagged

No significant circularity; algorithmic framework is self-contained

full rationale

The paper introduces an algorithmic QAOA variant that replaces the expectation objective with the cut value of the most probable bitstring, optimizes parameters via Bayesian optimization, and allocates shots adaptively using mode confidence plus normalized variance. No equations, derivations, or self-citations reduce the claimed shot savings or solution quality to fitted parameters by construction. Numerical comparisons to conventional expectation-based QAOA are performed on external 3-regular MaxCut instances and do not rely on any tautological redefinition of the objective or allocation rules. The framework therefore stands as an independent proposal whose performance claims rest on empirical evaluation rather than internal re-labeling of inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; Bayesian optimization hyperparameters and shot-allocation thresholds are implicit but unspecified.

pith-pipeline@v0.9.0 · 5430 in / 1022 out tokens · 38012 ms · 2026-05-14T21:50:10.593703+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

15 extracted references · 15 canonical work pages · 1 internal anchor

[1]

We propose a mode-based objective that shifts the optimization target of QAOA from the distributional mean to the quality of the most likely output solution, thereby improving the alignment between the training objective and the output form of discrete optimization tasks

work page
[2]

We develop a unified framework that integrates Bayesian optimization, mode-based evaluation, adaptive sampling, and early stopping, enabling objective design, parameter search, and measurement resource allocation to work cooperatively under a finite-shot setting

work page
[3]

how to estimate the expectation value more accurately

Beyond the main Stage-1 framework, we discuss an optional engineering refinement to improve the output stability of the target solution after a promising solution has been found, highlighting the practical deployment potential of the proposed approach. Compared with existing works that mainly focus on expectation-based objectives, a single classical optim...

work page
[4]

A Quantum Approximate Optimization Algorithm

E. Farhi, J. Goldstone, and S. Gutmann, arXiv preprint arXiv:1411.4028 (2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014
[5]

Blekos, D

K. Blekos, D. Brand, A. Ceschini, C.-H. Chou, R.-H. Li, K. Pandya, and A. Summer, Physics Reports1068, 1 (2024)

work page 2024
[6]

Tilly, H

J. Tilly, H. Chen, S. Cao, D. Picozzi, K. Setia, Y. Li, E. Grant, L. Wossnig, I. Rungger, G. H. Booth, et al., Physics Reports986, 1 (2022)

work page 2022
[7]

Bittel and M

L. Bittel and M. Kliesch, Physical review letters127, 120502 (2021)

work page 2021
[8]

L. Zhu, S. Liang, C. Yang, and X. Li, Journal of Chemical Theory and Computation20, 2390 (2024)

work page 2024
[9]

Larkin, M

J. Larkin, M. Jonsson, D. Justice, and G. G. Guer- reschi, Quantum Science and Technology7, 045014 (2022), ISSN 2058-9565, URLhttp://dx.doi.org/10. 1088/2058-9565/ac6973

work page 2022
[10]

Fern´ andez-Pend´ as, E

M. Fern´ andez-Pend´ as, E. F. Combarro, S. Vallecorsa, J. Ranilla, and I. F. R´ ua, Journal of Computational and Applied Mathematics404, 113388 (2022), ISSN 0377- 0427, URLhttp://dx.doi.org/10.1016/j.cam.2021. 113388

work page doi:10.1016/j.cam.2021 2022
[11]

D. R. Jones, M. Schonlau, and W. J. Welch, Jour- nal of Global Optimization13, 455–492 (1998), ISSN 1573-2916, URLhttp://dx.doi.org/10.1023/A: 1008306431147

work page doi:10.1023/a: 1998
[12]

Tibaldi, D

S. Tibaldi, D. Vodola, E. Tignone, and E. Ercolessi, IEEE Transactions on Quantum Engineering4, 1 (2023)

work page 2023
[13]

Sweke, F

R. Sweke, F. Wilde, J. Meyer, M. Schuld, P. K. Faehrmann, B. Meynard-Piganeau, and J. Eisert, Quan- tum4, 314 (2020), ISSN 2521-327X, URLhttp://dx. doi.org/10.22331/q-2020-08-31-314

work page doi:10.22331/q-2020-08-31-314 2020
[14]

A. Gu, A. Lowe, P. A. Dub, P. J. Coles, and A. Ar- rasmith (2021), 2108.10434, URLhttps://arxiv.org/ abs/2108.10434

work page arXiv 2021
[15]

X. Xu, J. Cui, Z. Cui, R. He, Q. Li, X. Li, Y. Lin, J. Liu, W. Liu, J. Lu, et al. (2024), 2406.17248, URLhttps: //arxiv.org/abs/2406.17248

work page arXiv 2024