Pith. sign in

REVIEW 3 major objections 6 minor 46 references

Solving Large-Scale QUBO with Transferred Parameters from Multilevel QAOA of low depth

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read A single low-depth QAOA parameter search can drive a multilevel solver for large QUBO problems, because coarsening preserves the structure QAOA depends on.

desk verdict Plausible hybrid multilevel QAOA+QRR+GA pipeline with an honest classical comparison; the parameter-transfer claim is promising but only indirectly supported, and the paper lacks reproducibility basics. read the letter →

arxiv 2505.11464 v1 pith:O7YRSB4F submitted 2025-05-16 quant-ph

classification quant-ph
keywords QAOAmultileveloptimizationparametertransferQUBOMaxCutQuantumRelaxandRoundgeneticalgorithmnear-termdevices
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that a depth-1 QAOA circuit does not need to be re-optimized at each scale of a multilevel solver for large QUBO and MaxCut problems. Its central claim is that relaxation-based coarsening preserves the structural information that fixes QAOA's energy landscape, so the angles found on the coarsest graph remain near-optimal at every finer level. If this is right, a single low-depth quantum parameter search, followed by classical refinement, can seed high-quality solutions for graphs with thousands to tens of thousands of nodes. The paper also reports that warm-starting a classical rank-two SDP solver with those multilevel solutions improves the best objective on most benchmark graphs while using less total runtime.

What carries the argument

The carrying mechanism is the V-cycle multilevel hierarchy with threshold-based coarsening. Nodes are embedded on a sphere and repositioned to maximize weighted distances from their neighbors; pairs are merged only when their sphere distance is below a threshold and their edge weight is non-positive, which keeps the coarse graph faithful to the original QUBO. At the coarsest level, p=1 QAOA expectation values and two-point correlations are obtained analytically, Quantum Relax & Round turns the correlation matrices into candidate bitstrings, and a genetic algorithm selects the best solution. The selected QAOA parameters are then transferred unchanged to each finer level, where the parameter grid is pruned to the top candidates from the previous level.

What would settle it

On a fresh graph, compute the best fine-level p=1 QAOA angles by an exhaustive fine-grid search and compare the objective value they produce with the objective produced by the transferred coarse-level angles; if the transferred angles are substantially worse while the landscape overlap is still near 95%, the similarity measure is not a valid proxy.

Watch

Extended reading notes

Core claim

The paper's central claim, stated on its own terms, is that QAOA parameter transfer across a multilevel hierarchy works because the coarsening phase is information-preserving. For p=1 QAOA, the paper reports that the energy landscapes at consecutive hierarchy levels overlap by roughly 95% in their interesting regions, and that the optimized parameters from the coarsest level, carried to finer levels with a pruned parameter grid, yield the best final solution without further variational optimization. The solver combines analytically computed single-layer QAOA correlations with Quantum Relax & Round to produce candidate bitstrings, uses a genetic algorithm to refine them, and uses the resulting solutions as warm starts for a classical rank-two SDP relaxation. On benchmark graphs with 3,400 to 14,000 nodes, the warm-started solver matches or beats the bare classical solver on most instances and runs at least 800 seconds faster.

Load-bearing premise

The load-bearing premise is that the reported overlap between the interesting regions of QAOA energy landscapes at neighboring levels, measured on grid points whose expectation value exceeds three standard deviations, really means the same parameters stay near-optimal at the finer level.

Editorial extensions

If this is right

  • Only the coarsest level needs variational parameter optimization; every finer level reuses the transferred angles, so the quantum cost is essentially one p=1 QAOA parameter search plus correlation-matrix evaluations.
  • Warm-starting a classical rank-two SDP relaxation with multilevel solutions improves the best objective on most tested benchmark graphs and reduces total runtime by at least 800 seconds compared with the classical solver alone.
  • Combining Quantum Relax & Round samples with genetic-algorithm refinement during uncoarsening outperforms either interpolation strategy on its own.
  • The reported roughly 95% landscape similarity across levels suggests the transfer scheme is not limited to the single example shown and should carry over to other instances built with the same coarsening rules.
  • The authors expect that increasing QAOA circuit depth would improve solution quality further within the same framework.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A stronger test of the transfer claim would compare transferred-parameter performance with per-level re-optimization on a held-out set of graphs; the paper's three-standard-deviation overlap measure is only an indirect proxy.
  • If the landscape similarity is generic, the same parameter-transfer scheme could be applied to weighted QUBO families beyond MaxCut, such as portfolio or scheduling problems, without per-instance quantum optimization.
  • The coarsening geometry is a design choice; using different embeddings or merging criteria might change how many levels are needed and how well the angles transfer.
  • The results hint that p=1 QAOA variational parameters are approximately scale-invariant for structured optimization problems, a property that could be studied independently of the multilevel solver.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes a multilevel hybrid quantum-classical solver for large QUBO/MaxCut problems. It coarsens graphs via a relaxation-based sphere embedding with a new threshold-based pairwise merging rule, solves the coarsest level with depth-1 QAOA using analytic expectation values, feeds the resulting spin-correlation matrices into Quantum Relax & Round, and refines candidate solutions with a genetic algorithm as the hierarchy is uncoarsened. The central new claim is that p=1 QAOA parameters optimized at the coarsest level can be carried to finer levels without re-optimization, because the coarsening is said to preserve the structural information needed for QAOA parameterization. This claim is supported primarily by a z-score-based Jaccard similarity between QAOA energy landscapes at consecutive hierarchy levels, and the resulting solver is evaluated by warm-starting the Burer-Monteiro SDP heuristic on Gset and SuiteSparse instances up to 14,000 nodes. The paper states that an analytic proof of the parameter-transfer property remains open and that the current evidence is numerical.

Significance. If the parameter-transfer claim holds, the paper would reduce a major cost of multilevel QAOA, namely repeated variational optimization at every level, and would provide a practical NISQ-era strategy for large MaxCut/QUBO instances. The paper has genuine strengths: it uses the analytical p=1 QAOA correlation formulas to avoid expensive sampling, it combines Quantum Relax & Round with a genetic algorithm in a sensible way, it tests on standard large public benchmarks, and it honestly compares against strong classical solvers while acknowledging that the quantum-only solutions are not competitive. The significance is conditional, however, because the load-bearing transfer claim currently rests on a landscape-similarity proxy that is not directly validated, and the numerical evidence in Table I lacks statistical detail.

major comments (3)
  1. [V-A, Figure 3, and Algorithm 1] The 3-sigma Jaccard landscape-similarity metric does not directly establish that parameters optimal at the coarsest level are near-optimal at finer levels. The metric compares only the sets of grid points whose expectation exceeds a threshold, and a high overlap of broad high-energy regions can coexist with a shifted or narrowed optimum. Because Algorithm 1 prunes the parameter grid to the top s parameter sets from the previous level without re-optimization, this gap is load-bearing for the central claim. I request a direct test: for every benchmark graph and every hierarchy level, report the expectation value obtained by evaluating the coarsest-level optimal parameters at that level, the level-optimized optimum, and a random-parameter baseline, with medians and quantiles across the benchmark set. The reported ~75% similarity at the coarsest level is especially relevant, since that is exactly the level from which transfer originates.
  2. [IV-A and IV-B] The coarsening thresholds Delta_1 and Delta_2, the number of retained parameter sets k, the candidate count s, and all genetic-algorithm hyperparameters are never specified, and the source code and data are currently only promised upon acceptance. These parameters determine the hierarchy that is built, the landscape that is compared, and the transfer search, so the similarity numbers in Section V-A and the results in Table I cannot be reproduced or assessed for sensitivity. Please report the actual values, a sensitivity study for Delta_1 and Delta_2, and a permanent code/data artifact or a detailed instance-level results table.
  3. [V-B, Table I, and Figure 4] The performance claim that warm-starting Burer-Monteiro with multilevel solutions 'provides an advantage in the majority of the tested cases' is not statistically supported. The table contains single objective values per instance, and QAOA+BURER02 exceeds plain BURER02 in 10 of 17 rows, ties in one row, and is worse in six rows, often by small margins. There are no error bars, repeated trials, or per-instance runtime breakdowns. I request means and standard deviations over multiple runs, a statement of how often the warm-started variant beats the baseline by more than a meaningful tolerance, and a per-instance comparison of the total runtime including the multilevel preprocessing.
minor comments (6)
  1. [II-A, Eq. (2)] The constraint in Eq. (2) is garbled: 's.t: p_t_i - 1 = 0' should presumably be the unit-norm constraint ||p_t_i||_2^2 = 1, consistent with the preceding sentence.
  2. [II-A] The sentence 'QAOA withplayers alternatingly apply' should read 'QAOA with p layers'.
  3. [IV-A] The phrase 'separating nodes from their neighbors connected by positive weight while bounding neighbors connected by positive weight' repeats 'positive weight'; the second occurrence should likely be 'negative weight' or otherwise be clarified.
  4. [Table I] The column header 'wE' is unexplained; please define it as the set of edge weights used in the instance.
  5. [Algorithm 1] Lines 7-8 are ambiguous: 'Select top n-th from k x m best candidates' and 'Initialize {(gamma', beta')} with parameters from top n-th candidates' do not specify how n and m are chosen and how the pruned parameter sets are paired with the selected candidate bitstrings.
  6. [Index Terms] The Index Terms use 'Quantum Relax & Rounds' with an inconsistent plural; the main text consistently uses 'Round', so the index term should be singular.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the parameter-transfer claim is supported by fresh numerical landscape comparisons and algorithmic experiments, not by self-citations or by construction.

full rationale

The paper's central claim is that relaxation-based coarsening preserves enough QUBO structure that QAOA parameters optimized at the coarsest level can be transferred to finer levels (Section IV-B). This is presented as the result of an empirical observation ('we numerically observe that for p=1 circuits, the optimal parameters are approximately transferred between the circuits at different levels of the hierarchy', Section IV), and is validated directly in Section V-A by computing p=1 QAOA energy landscapes from the independent analytic expressions of Ozaeta et al. [9] and comparing the regions with expectation above 3 sigma across consecutive levels. Nothing is fitted to the final solver outputs; the similarity metric is a heuristic, and the later solver results (Figure 4, Table I) are produced by executing the proposed algorithm rather than by construction from the metric. The paper explicitly lists an analytic proof of parameter transfer as future work (Section VI), confirming that the empirical evidence is not being masqueraded as a derivation. The self-citations [5], [7], [8], [14] describe the multilevel QAOA framework that the paper extends, but the extension—threshold-based coarsening, QRR+GA hybridization, and parameter transfer—is tested with new experiments against external baselines (Gset, SuiteSparse). No equation equates an input to an output, no fitted parameter is renamed a prediction, and no uniqueness theorem is imported from the authors' prior work. The main weakness is that the 3-sigma Jaccard overlap is only an indirect proxy for near-optimality of transferred parameters, but that is a validity or generalization concern, not circularity.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The method inherits the multilevel framework and p=1 QAOA formulas from prior work and relies on several unproven heuristic assumptions: that the z-score similarity metric predicts parameter transfer, that sphere-embedding coarsening preserves QUBO structure, and that V-cycle refinement is beneficial. The main tuned inputs (coarsening thresholds, grid sizes, genetic algorithm settings) are not reported, which limits independent checking.

free parameters (4)
  • Coarsening thresholds Delta_1 and Delta_2 = not reported
    Threshold Pairwise Merging uses max{Delta_1 * mean_distance, Delta_2 * min_distance} to decide which node pairs to merge; values are not reported and likely chosen by hand.
  • Number of retained parameter sets k and candidate count s = not reported
    Refinement keeps the top s QRR candidates from k parameter sets; these values control the runtime-quality tradeoff and are not specified.
  • Coarsest-level QAOA parameters (gamma, beta) = grid-selected, not reported
    A grid optimizer at the coarsest level selects these parameters; the transfer mechanism inherits them, so the result depends on this choice.
  • Genetic algorithm hyperparameters = not reported
    Population size, crossover, mutation, and termination criteria are not given, despite the genetic algorithm being central to the refinement loop.
assumptions (4)
  • domain assumption Analytic p=1 QAOA expectation value and correlation expressions from Ozaeta et al. are correct.
    Section IV-B computes spin correlations from these analytic formulas instead of sampling; an error here would change all QRR candidate solutions.
  • ad hoc to paper The z-score landscape similarity metric defined in Section V-A captures transferable QAOA parameters.
    The paper uses grid points with expectation above 3 sigma and Jaccard-style overlap as evidence for parameter transfer; this is a paper-specific proxy with no proof that it predicts solution quality.
  • domain assumption V-cycle coarsening and uncoarsening with local refinement improves or preserves solution quality.
    Section II-D and Algorithm 1 assume the standard multigrid-style refinement heuristic works for QUBO; the paper does not prove convergence.
  • ad hoc to paper Relaxation-based sphere embedding coarsening preserves QUBO structural information.
    Section IV-A claims preservation of problem-relevant structure, supported only by illustrative examples and benchmark results, not by a formal invariant.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Solving Large-Scale QUBO with Transferred Parameters from Multilevel QAOA of low depth." pith.science (2026). https://pith.science/paper/O7YRSB4F

@misc{pith2026250511464,
  author       = {Pith},
  title        = {Pith review of: Solving Large-Scale QUBO with Transferred Parameters from Multilevel QAOA of low depth},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/O7YRSB4F}},
  note         = {Machine review of arXiv:2505.11464}
}
abstract

The Quantum Approximate Optimization Algorithm (QAOA) is a promising quantum approach for tackling combinatorial optimization problems. However, hardware constraints such as limited scaling and susceptibility to noise pose significant challenges when applying QAOA to large instances. To overcome these limitations, scalable hybrid multilevel strategies have been proposed. In this work, we propose a fast hybrid multilevel algorithm with QAOA parameterization throughout the multilevel hierarchy and its reinforcement with genetic algorithms, which results in a high-quality, low-depth QAOA solver. Notably, we propose parameter transfer from the coarsest level to the finer level, showing that the relaxation-based coarsening preserves the problem structural information needed for QAOA parametrization. Our strategy improves the coarsening phase and leverages both Quantum Relax \& Round and genetic algorithms to incorporate $p=1$ QAOA samples effectively. The results highlight the practical potential of multilevel QAOA as a scalable method for combinatorial optimization on near-term quantum devices.

Figures

Figures reproduced from arXiv: 2505.11464 by the authors.

Figure 1
Figure 1. Example between two merging strategies: (1) Logarithmic Pairwise [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Illustration showing how the p = 1 QAOA landscape changes with coarsening hierarchy levels. The empirically observed similarity between the non-trivial regions of the landscape motivates the parameter transfer between levels. In the plots on the right, we additionally denote the top s QRR solutions (see the main text) as red points. and ¯d (min d) denotes the average (minimum) distance. By merging all nodes paired i… view at source ↗
Figure 3
Figure 3. An example of landscape similarity between consecutive hierarchy [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Example of G1 graph from Gset dataset [41]. Here, we plot the approximation ratio (top) and the runtime (bottom) for each hierarchy level, where level 12 denotes the coarsest level, and level 0 denotes the original graph. Burer-Monteiro implementation to which we compa…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 28 canonical work pages

  1. [1]

    A quantum approximate optimization algorithm.arXiv preprint:1411.4028, 2014

    Edward Farhi, Jeffrey Goldstone, and Sam Gutmann. A quantum approximate optimization algorithm.arXiv preprint:1411.4028, 2014

  2. [2]

    Qaoa for max-cut requires hundreds of qubits for quantum speed-up.Scientific reports, 9(1):6903, 2019

    Gian Giacomo Guerreschi and Anne Y Matsuura. Qaoa for max-cut requires hundreds of qubits for quantum speed-up.Scientific reports, 9(1):6903, 2019

  3. [3]

    Filip B Maciejewski, Stuart Hadfield, Benjamin Hall, Mark Hodson, Maxime Dupont, Bram Evert, James Sud, M Sohaib Alam, Zhihui Wang, Stephen Jeffrey, et al. Design and execution of quantum circuits using tens of superconducting qubits and thousands of gates for dense ising optimization problems.Physical Review Applied, 22(4):044074, 2024

  4. [4]

    What limits the simulation of quantum computers?PRX, 10(4):041038, 2020

    Yiqing Zhou, E Miles Stoudenmire, and Xavier Waintal. What limits the simulation of quantum computers?PRX, 10(4):041038, 2020

  5. [5]

    Multilevel combi- natorial optimization across quantum architectures.ACM Transactions on Quantum Computing, 2(1):1–29, 2021

    Hayato Ushijima-Mwesigwa, Ruslan Shaydulin, Christian FA Negre, Susan M Mniszewski, Yuri Alexeev, and Ilya Safro. Multilevel combi- natorial optimization across quantum architectures.ACM Transactions on Quantum Computing, 2(1):1–29, 2021

  6. [6]

    Multigrid solvers and multilevel optimiza- tion strategies.Multilevel optimization in VLSICAD, pages 1–69, 2003

    Achi Brandt and Dorit Ron. Multigrid solvers and multilevel optimiza- tion strategies.Multilevel optimization in VLSICAD, pages 1–69, 2003

  7. [7]

    Mlqaoa: Graph learning ac- celerated hybrid quantum-classical multilevel qaoa

    Bao Bach, Jose Falla, and Ilya Safro. Mlqaoa: Graph learning ac- celerated hybrid quantum-classical multilevel qaoa. In2024 IEEE International Conference on Quantum Computing and Engineering (QCE), volume 1, pages 1–12. IEEE, 2024

  8. [8]

    A multilevel approach for solving large-scale qubo problems with noisy hybrid quantum approximate optimization

    Filip B Maciejewski, Bao G Bach, Maxime Dupont, P Aaron Lott, Bhuvanesh Sundar, David E Bernal Neira, Ilya Safro, and Davide Venturelli. A multilevel approach for solving large-scale qubo problems with noisy hybrid quantum approximate optimization. In2024 IEEE High Performance Extreme Computing Conference (HPEC), pages 1–

Show all 46 references
  1. [9]

    Expectation values from the single-layer quantum approximate optimization algorithm on ising problems.Quantum Science and Technology, 7(4):045036, 2022

    Asier Ozaeta, Wim van Dam, and Peter L McMahon. Expectation values from the single-layer quantum approximate optimization algorithm on ising problems.Quantum Science and Technology, 7(4):045036, 2022

  2. [10]

    Extending relax-and-round combinatorial optimization solvers with quantum correlations.Physical Review A, 109(1):012429, 2024

    Maxime Dupont and Bhuvanesh Sundar. Extending relax-and-round combinatorial optimization solvers with quantum correlations.Physical Review A, 109(1):012429, 2024

  3. [11]

    What works best when? a systematic evaluation of heuristics for max-cut and qubo.INFORMS Journal on Computing, 30(3):608–624, 2018

    Iain Dunning, Swati Gupta, and John Silberholz. What works best when? a systematic evaluation of heuristics for max-cut and qubo.INFORMS Journal on Computing, 30(3):608–624, 2018

  4. [12]

    Optimization via quantum preconditioning.arXiv preprint arXiv:2502.18570, 2025

    Maxime Dupont, Tina Oberoi, and Bhuvanesh Sundar. Optimization via quantum preconditioning.arXiv preprint arXiv:2502.18570, 2025

  5. [13]

    Quantum computing: progress and prospects

    Mark Horowitz and Emily Grumbling. Quantum computing: progress and prospects. 2019

  6. [14]

    Hybrid quantum-classical multilevel approach for maximum cuts on graphs

    Anthony Angone, Xiaoyuan Liu, Ruslan Shaydulin, and Ilya Safro. Hybrid quantum-classical multilevel approach for maximum cuts on graphs. In2023 IEEE High Performance Extreme Computing Confer- ence (HPEC), pages 1–7. IEEE, 2023

  7. [15]

    Algebraic distance on graphs.SIAM Journal on Scientific Computing, 33(6):3468–3490, 2011

    Jie Chen and Ilya Safro. Algebraic distance on graphs.SIAM Journal on Scientific Computing, 33(6):3468–3490, 2011

  8. [16]

    Evidence of scaling advantage for the quantum approximate optimization algorithm on a classically intractable problem.Science Advances, 10(22):eadm6761, 2024

    Ruslan Shaydulin, Changhao Li, Shouvanik Chakrabarti, Matthew De- Cross, Dylan Herman, Niraj Kumar, Jeffrey Larson, Danylo Lykov, Pierre Minssen, Yue Sun, et al. Evidence of scaling advantage for the quantum approximate optimization algorithm on a classically intractable probl...

  9. [17]

    Quantum supremacy through the quantum approximate optimization algorithm.arXiv preprint arXiv:1602.07674, 2016

    Edward Farhi and Aram W Harrow. Quantum supremacy through the quantum approximate optimization algorithm.arXiv preprint arXiv:1602.07674, 2016

  10. [18]

    Improving quantum approximate optimization by noise- directed adaptive remapping.arXiv preprint arXiv:2404.01412, 2024

    Filip B Maciejewski, Jacob Biamonte, Stuart Hadfield, and Davide Venturelli. Improving quantum approximate optimization by noise- directed adaptive remapping.arXiv preprint arXiv:2404.01412, 2024

  11. [19]

    End-to-end protocol for high-quality qaoa parameters with few shots.arXiv preprint arXiv:2408.00557, 2024

    Tianyi Hao, Zichang He, Ruslan Shaydulin, Jeffrey Larson, and Marco Pistoia. End-to-end protocol for high-quality qaoa parameters with few shots.arXiv preprint arXiv:2408.00557, 2024

  12. [20]

    Towards large-scale quantum optimization solvers with few qubits.Nature Communications, 16(1):476, 2025

    Marco Sciorilli, Lucas Borges, Taylor L Patti, Diego Garc ´ıa-Mart´ın, Giancarlo Camilo, Anima Anandkumar, and Leandro Aolita. Towards large-scale quantum optimization solvers with few qubits.Nature Communications, 16(1):476, 2025

  13. [21]

    Approximate solutions of combinatorial problems via quantum relaxations.IEEE Transactions on Quantum Engineering, 2024

    Bryce Fuller, Charles Hadfield, Jennifer R Glick, Takashi Imamichi, Toshinari Itoko, Richard J Thompson, Yang Jiao, Marna M Kagele, Adriana W Blom-Schieber, Rudy Raymond, et al. Approximate solutions of combinatorial problems via quantum relaxations.IEEE Transactions on Quantu...

  14. [22]

    Qubit-efficient quantum combinatorial optimization solver.arXiv preprint:2407.15539, 2024

    Bhuvanesh Sundar and Maxime Dupont. Qubit-efficient quantum combinatorial optimization solver.arXiv preprint:2407.15539, 2024

  15. [23]

    Graph decom- position techniques for solving combinatorial optimization problems with variational quantum algorithms.Quantum Information Processing, 24(2):60, 2025

    Moises Ponce, Rebekah Herrman, Phillip C Lotshaw, Sarah Powers, George Siopsis, Travis Humble, and James Ostrowski. Graph decom- position techniques for solving combinatorial optimization problems with variational quantum algorithms.Quantum Information Processing, 24(2):60, 2025

  16. [24]

    Decomposition pipeline for large-scale portfolio optimization with applications to near-term quantum computing.arXiv preprint arXiv:2409.10301, 2024

    Atithi Acharya, Romina Yalovetzky, Pierre Minssen, Shouvanik Chakrabarti, Ruslan Shaydulin, Rudy Raymond, Yue Sun, Dylan Her- man, Ruben S Andrist, Grant Salton, et al. Decomposition pipeline for large-scale portfolio optimization with applications to near-term quantum computi...

  17. [25]

    Large-scale quantum approximate optimization via divide-and-conquer.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2022

    Junde Li, Mahabubul Alam, and Swaroop Ghosh. Large-scale quantum approximate optimization via divide-and-conquer.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2022

  18. [26]

    Divide and conquer for combinatorial optimization and distributed quantum computation

    Teague Tomesh, Zain H Saleem, Michael A Perlin, Pranav Gokhale, Martin Suchara, and Margaret Martonosi. Divide and conquer for combinatorial optimization and distributed quantum computation. In 2023 IEEE International Conference on Quantum Computing and En- gineering (QCE), vo...

  19. [27]

    Scal- ing up the quantum divide and conquer algorithm for combinatorial optimization.arXiv preprint arXiv:2405.00861, 2024

    Ibrahim Cameron, Teague Tomesh, Zain Saleem, and Ilya Safro. Scal- ing up the quantum divide and conquer algorithm for combinatorial optimization.arXiv preprint arXiv:2405.00861, 2024

  20. [28]

    Investigating the effect of circuit cutting in qaoa for the maxcut problem on nisq devices

    Marvin Bechtold, Johanna Barzen, Frank Leymann, Alexander Mandl, Julian Obst, Felix Truger, and Benjamin Weder. Investigating the effect of circuit cutting in qaoa for the maxcut problem on nisq devices. Quantum Science and Technology, 8(4):045022, 2023

  21. [29]

    Quantum circuit cutting with maximum-likelihood tomography.npj Quantum Information, 7(1):64, 2021

    Michael A Perlin, Zain H Saleem, Martin Suchara, and James C Osborn. Quantum circuit cutting with maximum-likelihood tomography.npj Quantum Information, 7(1):64, 2021

  22. [30]

    Benchmarking quan- tum optimization for the maximum-cut problem on a superconducting quantum computer.Physical Review Applied, 23(1):014045, 2025

    Maxime Dupont, Bhuvanesh Sundar, Bram Evert, David E Bernal Neira, Zedong Peng, Stephen Jeffrey, and Mark J Hodson. Benchmarking quan- tum optimization for the maximum-cut problem on a superconducting quantum computer.Physical Review Applied, 23(1):014045, 2025

  23. [31]

    Training variational quantum algo- rithms is np-hard.Physical review letters, 127(12):120502, 2021

    Lennart Bittel and Martin Kliesch. Training variational quantum algo- rithms is np-hard.Physical review letters, 127(12):120502, 2021

  24. [32]

    A review of barren plateaus in variational quantum computing.arXiv preprint arXiv:2405.00781, 2024

    Martin Larocca, Supanut Thanasilp, Samson Wang, Kunal Sharma, Jacob Biamonte, Patrick J Coles, Lukasz Cincio, Jarrod R McClean, Zo¨e Holmes, and M Cerezo. A review of barren plateaus in variational quantum computing.arXiv preprint arXiv:2405.00781, 2024

  25. [33]

    Trainability barriers in low-depth qaoa landscapes

    Joel Rajakumar, John Golden, Andreas B ¨artschi, and Stephan Eidenbenz. Trainability barriers in low-depth qaoa landscapes. InProceedings of the 21st ACM International Conference on Computing Frontiers, 2024

  26. [34]

    Multistart methods for quantum approximate optimization

    Ruslan Shaydulin, Ilya Safro, and Jeffrey Larson. Multistart methods for quantum approximate optimization. In2019 IEEE high performance extreme computing conference (HPEC), pages 1–8. IEEE, 2019

  27. [35]

    Warm-starting quantum optimization.Quantum, 5:479, 2021

    Daniel J Egger, Jakub Mare ˇcek, and Stefan Woerner. Warm-starting quantum optimization.Quantum, 5:479, 2021

  28. [36]

    Parameter transfer for quantum approximate optimization of weighted maxcut.ACM Transactions on Quantum Computing, 4(3):1–15, 2023

    Ruslan Shaydulin, Phillip C Lotshaw, Jeffrey Larson, James Ostrowski, and Travis S Humble. Parameter transfer for quantum approximate optimization of weighted maxcut.ACM Transactions on Quantum Computing, 4(3):1–15, 2023

  29. [37]

    Similarity-based parameter transferability in the quantum approximate optimization algorithm.Frontiers in Quantum Science and Technology, 2:1200975, 2023

    Alexey Galda, Eesh Gupta, Jose Falla, Xiaoyuan Liu, Danylo Lykov, Yuri Alexeev, and Ilya Safro. Similarity-based parameter transferability in the quantum approximate optimization algorithm.Frontiers in Quantum Science and Technology, 2:1200975, 2023

  30. [38]

    Cross-problem parameter transfer in quantum approximate optimization algorithm: A machine learning approach.arXiv preprint arXiv:2504.10733, 2025

    Kien X Nguyen, Bao Bach, and Ilya Safro. Cross-problem parameter transfer in quantum approximate optimization algorithm: A machine learning approach.arXiv preprint arXiv:2504.10733, 2025

  31. [39]

    An algorithm for finding best matches in logarithmic expected time.ACM Transactions on Mathematical Software (TOMS), 3(3):209–226, 1977

    Jerome H Friedman, Jon Louis Bentley, and Raphael Ari Finkel. An algorithm for finding best matches in logarithmic expected time.ACM Transactions on Mathematical Software (TOMS), 3(3):209–226, 1977

  32. [40]

    Rank-two relax- ation heuristics for max-cut and other binary quadratic programs.SIAM Journal on Optimization, 12(2):503–521, 2002

    Samuel Burer, Renato DC Monteiro, and Yin Zhang. Rank-two relax- ation heuristics for max-cut and other binary quadratic programs.SIAM Journal on Optimization, 12(2):503–521, 2002

  33. [41]

    Gset - a suite-style benchmark for graph processing systems

    Yuan Ye. Gset - a suite-style benchmark for graph processing systems. https://web.stanford.edu/∼yyye/yyye/Gset/, 2003

  34. [42]

    The university of florida sparse matrix collection.ACM TOMS, 38(1):1–25, 2011

    Timothy A Davis and Yifan Hu. The university of florida sparse matrix collection.ACM TOMS, 38(1):1–25, 2011

  35. [43]

    Pygad: An intuitive genetic algorithm python library.Multimedia tools and applications, 83(20):58029–58042, 2024

    Ahmed Fawzy Gad. Pygad: An intuitive genetic algorithm python library.Multimedia tools and applications, 83(20):58029–58042, 2024

  36. [44]

    Randomized heuristics for the max-cut problem.Optimization methods and software, 17(6):1033–1058, 2002

    Paola Festa, Panos M Pardalos, Mauricio GC Resende, and Celso C Ribeiro. Randomized heuristics for the max-cut problem.Optimization methods and software, 17(6):1033–1058, 2002

  37. [45]

    A low-level hybridization between memetic algorithm and vns for the max-cut problem

    Abraham Duarte, Angel Sanchez, Felipe Fern ´andez, and Ra ´ul Cabido. A low-level hybridization between memetic algorithm and vns for the max-cut problem. pages 999–1006, 06 2005

  38. [46]

    Diversification-driven tabu search for unconstrained binary quadratic problems.4OR, 2010

    Fred Glover, Zhipeng L ¨u, and Jin-Kao Hao. Diversification-driven tabu search for unconstrained binary quadratic problems.4OR, 2010

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.