Pith. sign in

REVIEW 4 major objections 4 minor 52 references

Compensating connectivity restrictions in quantum annealers via splitting and linearization techniques

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read By splitting dense QUBO couplings into hardware-native quadratic parts and linearized leftovers, this paper argues that a quantum annealer can solve large problems without minor embedding.

desk verdict Genuinely new heuristic for embedding-free QUBO annealing; theorem covers a regime the experiments don't use, but the idea is plausible and honestly presented. read the letter →

arxiv 2507.12536 v1 pith:WKMHERW7 submitted 2025-07-16 quant-ph

Marcel Seelbach Benkner , Zorah Lähner , Vladislav Golyanik , Martin Kliesch , Michael Moeller This is my paper · ORCID
classification quant-ph
keywords quantumannealingQUBOIsingmodelminorembeddingconnectivityrestrictionsplittingandlinearizationlargeneighborhoodlocalsearchweakmonotonicity
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 is trying to establish that the connectivity restrictions of quantum annealers can be compensated without minor embedding, the standard but expensive trick of merging several physical qubits into one logical qubit. The proposed iteration splits the QUBO/Ising coupling matrix into a quadratic part that matches a randomly permuted copy of the hardware graph and a linearized leftover, then solves the resulting hardware-compatible subproblem with a damping term that keeps the next spin configuration near the current one. A weak monotonicity theorem shows that a sufficiently large damping term makes the energy non-increasing at every step, so the finite search space guarantees convergence of the energy. In experiments on regular spin-glass and Max-Cut benchmarks, the method is competitive with large-neighborhood local search and can be applied to dense problems that are too large to be minor-embedded at all; proof-of-concept runs on a real quantum annealer confirm that the energy decreases on Max-Cut instances with $\pm1$ coefficients. This matters because it offers a route to larger dense problems on the same hardware and removes the time-consuming embedding step from the solve loop.

What carries the argument

The load-bearing object is the split $A = A_{\rm quad}^{(k)} + A_{\rm lin}^{(k)}$ obtained by masking $A$ with a randomly permuted hardware adjacency matrix, combined with a proximal-gradient-style update that treats the leftover $A_{\rm lin}$ as a linear term and adds the damping term $-\lambda\langle s, s^{(k)}\rangle$. The mask keeps only couplings native to the hardware graph, so each subproblem can be handed to the annealer as a quadratic program; the random permutation cycles which edges are native, so edges missing in one iteration become quadratic in later iterations; and the damping term, inherited from proximal gradient descent, controls the error of the linearization. The weak monotonicity proof works by showing that for $\lambda \ge 2\|A_{\rm lin}\|$ the surrogate $E^{(k)}$ dominates the true energy $E$, giving $E(s^{(k+1)}) \le E^{(k)}(s^{(k+1)}) \le E^{(k)}(s^{(k)}) = E(s^{(k)})$ at every step.

What would settle it

Run the method on dense random $\pm1$ Max-Cut instances of about 200 variables with a Pegasus-style mask, using the paper's $\lambda$ schedule below $2\|A_{\rm lin}\|$; if from many random starts the best energy after 25 iterations is no better than optimizing a single fixed mask once, or if the energy sequence rises on any run, the practical claim fails. The theoretical monotonicity claim would be contradicted only by an iterate with $\lambda \ge 2\|A_{\rm lin}\|$ that increases the energy.

Watch

Extended reading notes

Core claim

At the paper's core is the claim that a hardware-restricted annealer can be used directly on dense Ising/QUBO problems by iterating a mask-and-linearize step. Writing the objective as $s^T A s$, the method draws a random permutation $P^{(k)}$ of the hardware graph's adjacency matrix $M$, defines $A_{\rm quad}^{(k)} = ((P^{(k)})^T M P^{(k)}) \odot A$ and $A_{\rm lin}^{(k)} = A - A_{\rm quad}^{(k)}$, and computes $s^{(k+1)} = \arg\min_{s\in\{-1,1\}^n} s^T A_{\rm quad}^{(k)} s + \langle 2 A_{\rm lin}^{(k)} s^{(k)}, s\rangle - \lambda \langle s, s^{(k)}\rangle$. Because the quadratic part uses only edges present in the current permuted hardware graph, the annealer can solve the subproblem natively. The paper proves that for $\lambda \ge 2\|A_{\rm lin}\|$ the surrogate objective majorizes the true one, which forces $E(s^{(k+1)}) \le E(s^{(k)})$ and therefore convergence of the energy sequence on the finite set of binary vectors. The practical claim is that iterating fresh random permutations with experimentally chosen smaller $\lambda$ values yields a heuristic competitive with large-neighborhood local search on the tested spin-glass and Max-Cut benchmarks, scales beyond minor-embedding capacity, and succeeds on low-precision Max-Cut instances on actual annealing hardware.

Load-bearing premise

The practical effectiveness rests on the unproven heuristic that, at the regularization strengths actually used in the experiments (chosen below the theoretically safe bound), re-masking with fresh random permutations and linearizing the missing edges keeps producing subproblem solutions that lower the original energy.

Editorial extensions

If this is right

  • Dense QUBO instances that cannot be minor-embedded at all become approachable, since the method never needs to map the full problem graph into the hardware graph.
  • The expensive embedding-search step is removed from the solve loop; on large regular spin-glass instances the paper finds that minor-embedding time dominates the wall-clock time of the comparison methods.
  • The weak monotonicity result is tunable: choosing $\lambda$ large enough guarantees the energy never rises, at the cost of possible stagnation, while smaller $\lambda$ values turn the iteration into a practical heuristic without a guarantee.
  • The method's behavior interpolates with large-neighborhood local search, and which variant wins depends on subproblem size, so users can trade quality against iteration cost by choosing the mask schedule and $\lambda$.
  • For real annealing hardware, coefficient precision is a limiting factor: the proof of concept works on Max-Cut instances with $\pm1$ coefficients, while some float-valued instances fail to improve.

Reading between the lines

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

  • A deterministic schedule of masks that guarantees each problem edge appears in the quadratic part within a fixed number of iterations could be tested as a replacement for random permutations; if coverage drives progress, it should match or beat the randomized version with reproducible behavior.
  • The same mask-and-linearize recipe could be applied to any hardware-restricted quadratic solver, not only quantum annealers, as long as the solver can natively handle the masked quadratic subproblem with a damping term.
  • The simulated-annealing-like regularization proposed in Section 3.8 is explicitly left numerically unverified by the authors; implementing it with a proper temperature schedule would turn the monotone iteration into a global-search heuristic and directly extend the paper's evidence.
  • The observed failures on high-precision coefficients suggest a preprocessing step that rounds or compresses QUBO weights before splitting, or a split that keeps high-precision edges inside the quadratic part, as a way to broaden the method's hardware applicability.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. Seelbach Benkner et al. propose an iterative 'splitting' method for solving dense QUBO/Ising problems on connectivity-restricted quantum annealers without minor embedding. At each outer iteration a random permutation P^k is applied to the hardware adjacency matrix M, the coupling matrix A is split into A_quad (entries supported on the permuted hardware graph) and A_lin (the complement), and a subproblem is solved in which A_quad is kept quadratic while A_lin is linearized around the current iterate and a damping term λ⟨s,s^k⟩ is added. The authors provide a weak-monotonicity proof for the proximal-gradient-style iteration when λ is at least the Lipschitz constant of the linearized part, relate the method to projected gradient descent and large-neighborhood local search, propose a heuristic λ-selection scheme, and benchmark the method on regular spin-glass instances and MQLib Max-Cut instances with simulated annealing as the subproblem solver, plus a D-Wave Advantage proof-of-concept on 13 instances with ±1 coefficients. They also include an appendix proof of a ground-state conjecture for the regular spin-glass family and an ablation study on the damping term.

Significance. If the method performed as claimed, it would be a valuable practical tool: it bypasses the NP-hard minor-embedding step, can in principle address problems larger than the embedding capacity, and the idea of periodically randomizing which edges are treated quadratically is natural and easy to implement. The paper is transparent about several limitations (λ values below the monotonicity threshold, failure cases on non-±1 coefficients, code not yet released), and the appendix proof of the spin-glass ground-state conjecture and the damping ablation are useful contributions. However, the present manuscript does not yet establish that the algorithm as benchmarked is the algorithm analyzed: the theoretical guarantee holds only for λ ≥ 2||A_lin||, the λ-selection rule in Algorithm 2 is under-specified, and the algorithmic listing appears to omit a factor of 2 in the linearization term. The experimental evidence is averaged without error bars and the hardware demonstration is narrow. The work is a plausible proof-of-concept whose central claims need tightening before publication.

major comments (4)
  1. [§3.1.1, §3.2, §3.5, §3.6] The weak-monotonicity guarantee is stated for λ ≥ c with c = ||∇²E₂|| = 2||A_lin|| for symmetric A_lin. Section 3.5 explicitly says that in experiments the authors 'allow for λ parameters smaller than ||A_lin|| from section 3.1.1', and Algorithm 2 selects λ from S_sign-change without any check against this bound. Consequently the convergence guarantee does not cover the algorithm as evaluated; every reported simulation and D-Wave run uses λ values for which monotonicity is unproven. Please either present experiments in the λ ≥ 2||A_lin^(k)|| regime (and report the energy trajectory), or explicitly state that the practical method is a heuristic whose monotonicity guarantee is not asserted for the tested parameters.
  2. [§3.6, Algorithm 2, Eqs. (28)-(29)] The λ-selection rule is not well-defined: L in Eq. (28) is defined using the same λ that Eq. (29) is supposed to select, so S_sign-change depends on the unknown λ. Algorithm 2 line 1 gives no convention for resolving this circularity. Please specify a complete selection rule (e.g., a fixed reference λ or a definition of L that does not involve the variable being selected) so that the experiments are reproducible.
  3. [Algorithm 1, Eq. (27)] Algorithm 1 line 7 does not implement the iteration in Eq. (27). Substituting z = P_k^T x_A and x_Aold = P_k z_old, the linear coefficient passed to S is A_lin z_old − λ z_old, whereas Eq. (27) requires 2 A_lin z_old − λ z_old (the gradient of s^T A_lin s is 2 A_lin s). The missing factor of 2 changes the method: the algorithm as written is not the proximal-gradient iteration analyzed in Section 3.1.1. Please correct the algorithm listing, or if the factor-2 omission is intentional, justify the modified update and re-derive the monotonicity statement for it.
  4. [§4, Figs. 5-10] The benchmark evidence is too thin to support the broad practical claim. Results are reported only as averaged approximation ratios without error bars, standard deviations, or per-instance distributions; the D-Wave validation uses 13 instances with coefficients restricted to ±1, and Section 4.3.2 explicitly reports failure to improve or worsening on float-valued and integer-valued instances. Please add statistical detail (variance or per-instance plots) and either extend the hardware experiments to a wider coefficient range or limit the conclusion to the tested setting.
minor comments (4)
  1. [§3.1.1, Eq. (14)] Equation (14) is missing a minus sign: Taylor expansion of −E₂ gives −(1/2)⟨∇²E₂(ξ)d,d⟩, not +(1/2)⟨...⟩. The inequality on the following line uses the negative sign, so the intended bound is correct after fixing this typo.
  2. [§3.2] The condition 'λ greater or equal than the largest eigenvalue of A_lin' should refer to the spectral norm (largest absolute eigenvalue), since the bound involves ||2A_lin||; for indefinite A_lin the largest eigenvalue can be much smaller than the norm.
  3. [§4.2] Typos: 'regular spinglas' should be 'regular spin glass', and 'Watz–Strogatz' should be 'Watts–Strogatz'; the dataset name 'MQLib' is spelled inconsistently as 'MQlib' in several places.
  4. [§5 and repository [52]] The source code is only announced for future publication. For reproducibility of the central experiments, please provide the code or a complete parameter specification (including the exact λ selection implementation and the number of random permutations per instance) at submission time.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the splitting iteration and weak-monotonicity proof derive from a standard proximal-gradient majorization argument, and the benchmarks are evaluated against external instances.

full rationale

The paper's derivation chain is self-contained. Equation (9) is obtained by applying the proximal gradient scheme (7) with f(x)=s^T A_lin s and g(x)=s^T A_quad s plus the indicator of {-1,1}^n; the update is a direct algebraic linearization, not an ansatz whose conclusion is hidden in its premise. The weak monotonicity result in Section 3.1.1 is the standard majorization argument: for lambda >= ||2 A_lin||, the surrogate E^(k) upper-bounds E, so E(x^(k+1)) <= E^(k)(x^(k+1)) <= E^(k)(x^(k)) = E(x^(k)). This is an inequality derived from Taylor expansion with the Hessian bound stated explicitly; no fitted parameter is used to define the objective or to force the convergence claim. The random permutation step in Section 3.5 changes which edges are quadratic and which are linearized, but it does not presuppose the energy decrease it is used to test. Lambda, maxsubiter, and chain strength are scheduling hyperparameters, not quantities fitted to the benchmark optima and then reported as predictions. The self-citations (Refs. [31] and [32]) appear only in the related-work discussion of QUBO formulations and are not load-bearing for the new method or its proof. The paper explicitly acknowledges that the experiments use lambda values smaller than the monotonicity bound to avoid stagnation, and it reports failure cases on the D-Wave hardware; these are honest scope limitations, not circular reasoning. The under-specification in Eqs. (28)-(29) of how Lambda enters S_sign-change is an algorithmic ambiguity, not an equivalence between an output and an input. Overall, no step reduces by construction to its own inputs, so the correct finding is no significant circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central algorithm relies on standard QUBO math, the assumed quality of the restricted solver, and an unproven small-λ heuristic; no new physical entities are introduced. Hyperparameters λ, iteration limits, and chain strength are hand-selected experimental choices.

free parameters (3)
  • Regularization parameter λ = chosen from S_lambda (eq. 29), typically below ||A_lin||
    Controls damping in eq. (27); monotonicity bound requires λ ≥ 2||A_lin||, but experiments use smaller values (Section 3.5, Section 3.6).
  • Iteration limits (maxiter, maxsubiter) = maxiter=120, maxsubiter=15
    Reported in Section 4.2 as the run configuration; affects all comparisons.
  • Chain strength factor for D-Wave = 1.001 times the largest coupling or bias
    Chosen by trying instances before the reported runs (Section 4.3.2); a hand-selected experimental parameter.
assumptions (4)
  • standard math The QUBO/Ising equivalence with symmetric Q is taken as background; Eqs. (1) to (2) and the gradient rule in Eq. (10) rely on it.
    This is standard linear algebra for binary quadratic problems and does not add assumptions.
  • domain assumption For the iterate sequence to converge, every binary vector is assumed to have a distinct energy (Section 3.2).
    The authors state this assumption explicitly; it is not guaranteed and does not affect the practical algorithm.
  • domain assumption The hardware-restricted QUBO solver S returns sufficiently low-energy states for the subproblems in each iteration.
    The method's progress depends on subproblem solver quality; hardware failures on 'imgseg 271031' and 'gka.5f' show the assumption is not universal (Section 4.3.2).
  • ad hoc to paper Useful progress with λ values below the monotonicity threshold is assumed in the experiments.
    Sections 3.5 and 3.6 choose λ from a sign-change candidate set, which can be smaller than the bound λ ≥ 2||A_lin||; this is an unproven heuristic that the practical algorithm relies on.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Compensating connectivity restrictions in quantum annealers via splitting and linearization techniques." pith.science (2026). https://pith.science/paper/WKMHERW7

@misc{pith2026250712536,
  author       = {Pith},
  title        = {Pith review of: Compensating connectivity restrictions in quantum annealers via splitting and linearization techniques},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WKMHERW7}},
  note         = {Machine review of arXiv:2507.12536}
}
read the original abstract

Current quantum annealing experiments often suffer from restrictions in connectivity in the sense that only certain qubits can be coupled to each other. The most common strategy to overcome connectivity restrictions so far is by combining multiple physical qubits into a logical qubit with higher connectivity, which is achieved by adding terms to the Hamiltonian. Practically, this strategy is implemented by finding a so-called minor embedding, which is in itself an NP-hard problem. In this work, we present an iterative algorithm that does not need additional qubits but instead efficiently uses the available connectivity for different parts of the problem graph in every step. We present a weak monotonicity proof and benchmark our algorithm against the default minor-embedding algorithm on the D-Wave quantum annealer and multiple simple local search variants. While most of the experiments to compare the different iterative methods are performed with simulated annealing solvers, we also confirm the practicality of our method with experiments on the D-Wave Advantage quantum annealer.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 48 canonical work pages

  1. [1]

    Oxford University Press (2020)

    Grant, E.K., Humble, T.S.: Adiabatic Quantum Computing and Quantum Anneal- ing. Oxford University Press (2020). https://oxfordre.com/physics/view/10.1093/ acrefore/9780190871994.001.0001/acrefore-9780190871994-e-32

    arXiv 2020

  2. [2]

    SIAM review50(4) (2008)

    Aharonov, D., Van Dam, W., Kempe, J., Landau, Z., Lloyd, S., Regev, O.: Adiabatic quantum computation is equivalent to standard quantum computation. SIAM review50(4) (2008)

    work page 2008

  3. [3]

    Das, A., Chakrabarti, B.K.: Quantum Annealing and Related Optimization Methods vol. 679. Springer, Heidelberg, Germany (2005)

    work page 2005

  4. [4]

    Denchev, V.S., Boixo, S., Isakov, S.V., Ding, N., Babbush, R., Smelyanskiy, V., Martinis, J., Neven, H.: What is the computational value of finite-range tunneling? Phys. Rev. X6(2016)

    work page 2016

  5. [5]

    Physical Review A65(4) (2002)

    Roland, J., Cerf, N.J.: Quantum search by local adiabatic evolution. Physical Review A65(4) (2002)

    work page 2002

  6. [6]

    Synthesis Lectures on Quantum Computing5(2) (2014) 25

    McGeoch, C.C.: Adiabatic quantum computation and quantum annealing: Theory and practice. Synthesis Lectures on Quantum Computing5(2) (2014) 25

    work page 2014

  7. [7]

    https://docs.dwavesys.com/docs/latest/handbook embedding.html

    QPU Solvers: Minor-Embedding; D-Wave System Documentation documenta- tion — docs.dwavesys.com. https://docs.dwavesys.com/docs/latest/handbook embedding.html. [Accessed 20-08-2024]

    work page 2024

  8. [8]

    arXiv e-prints (2014)

    Cai, J., Macready, W.G., Roy, A.: A practical heuristic for finding graph minors. arXiv e-prints (2014)

    work page 2014

Show all 52 references
  1. [9]

    ACM Transactions on Quantum Computing 4(3) (2023)

    Raymond, J., Stevanovic, R., Bernoudy, W., Boothby, K., McGeoch, C.C., Berkley, A.J., Farr´ e, P., Pasvolsky, J., King, A.D.: Hybrid quantum annealing for larger- than-qpu lattice-structured problems. ACM Transactions on Quantum Computing 4(3) (2023)

    2023

  2. [10]

    Quantum Information Processing17(5) (2018)

    Goodrich, T.D., Sullivan, B.D., Humble, T.S.: Optimizing adiabatic quantum program compilation using a graph-theoretic framework. Quantum Information Processing17(5) (2018)

    2018

  3. [11]

    Quantum Information Processing18(4) (2019)

    Patton, R., Schuman, C., Potok, T., et al.: Efficiently embedding qubo problems on adiabatic quantum computers. Quantum Information Processing18(4) (2019)

    2019

  4. [12]

    Theoretical Computer Science989(2024)

    Lobe, E., Lutz, A.: Minor embedding in broken chimera and derived graphs is np-complete. Theoretical Computer Science989(2024)

    2024

  5. [13]

    the disjoint paths problem

    Robertson, N., Seymour, P.D.: Graph minors .xiii. the disjoint paths problem. Journal of Combinatorial Theory, Series B63(1) (1995)

    1995

  6. [14]

    Quantum Information Processing15(2016)

    Boothby, T., King, A.D., Roy, A.: Fast clique minor generation in chimera qubit connectivity graphs. Quantum Information Processing15(2016)

    2016

  7. [15]

    arXiv preprint arXiv:2009.12479 (2020)

    King, A.D., Bernoudy, W.: Performance benefits of increased qubit connectivity in quantum annealing 3-dimensional spin glasses. arXiv preprint arXiv:2009.12479 (2020)

    2020 arXiv

  8. [16]

    arXiv preprint arXiv:2101.07813 (2021)

    Guerreschi, G.G.: Solving quadratic unconstrained binary optimization with divide-and-conquer and quantum algorithms. arXiv preprint arXiv:2101.07813 (2021)

    2021 arXiv

  9. [17]

    Ponce, M., Herrman, R., Lotshaw, P.C., Powers, S., Siopsis, G., Humble, T., Ostrowski, J.: Graph decomposition techniques for solving combinatorial optimization problems with variational quantum algorithms (2023)

    2023

  10. [18]

    In: 2022 IEEE International Conference on Quantum Computing and Engineering (QCE) (2022)

    Thai, P., Thai, M.T., Vu, T., Dinh, T.N.: Fasthare: fast hamiltonian reduction for large-scale quantum annealing. In: 2022 IEEE International Conference on Quantum Computing and Engineering (QCE) (2022). IEEE

    2022

  11. [19]

    INFORMS Journal on Computing 30(3) (2018) 26

    Dunning, I., Gupta, S., Silberholz, J.: What works best when? a systematic evaluation of heuristics for max-cut and QUBO. INFORMS Journal on Computing 30(3) (2018) 26

    2018

  12. [20]

    arXiv preprint arXiv:2404.17579 (2024)

    Dupont, M., Sundar, B., Evert, B., Neira, D.E.B., Peng, Z., Jeffrey, S., Hodson, M.J.: Quantum optimization for the maximum cut problem on a superconducting quantum computer. arXiv preprint arXiv:2404.17579 (2024)

    2024 arXiv

  13. [21]

    arXiv preprint arXiv:1411.4028 (2014)

    Farhi, E., Goldstone, J., Gutmann, S.: A quantum approximate optimization algorithm. arXiv preprint arXiv:1411.4028 (2014)

    2014 arXiv

  14. [22]

    Scientific Reports12(1) (2022)

    Pelofske, E., Hahn, G., Djidjev, H.N.: Parallel quantum annealing. Scientific Reports12(1) (2022)

    2022

  15. [23]

    Quantum Machine Intelligence 3(1) (2021)

    Bass, G., Henderson, M., Heath, J., Dulny III, J.: Optimizing the optimizer: decomposition techniques for quantum annealing. Quantum Machine Intelligence 3(1) (2021)

    2021

  16. [24]

    Foundations and trends®in Optimization1(3) (2014)

    Parikh, N., Boyd, S., et al.: Proximal algorithms. Foundations and trends®in Optimization1(3) (2014)

    2014

  17. [25]

    IEEE transactions on image processing18(11) (2009)

    Beck, A., Teboulle, M.: Fast gradient-based algorithms for constrained total variation image denoising and deblurring problems. IEEE transactions on image processing18(11) (2009)

    2009

  18. [26]

    SIAM Journal on Optimization29(4) (2019)

    Ochs, P., Pock, T.: Adaptive fista for nonconvex optimization. SIAM Journal on Optimization29(4) (2019)

    2019

  19. [27]

    IEEE Transactions on Quantum Engineering1(2020)

    Gambella, C., Simonetto, A.: Multiblock admm heuristics for mixed-binary opti- mization on classical and quantum computers. IEEE Transactions on Quantum Engineering1(2020)

    2020

  20. [28]

    IEEE transactions on pattern analysis and machine intelligence41(7) (2018)

    Wu, B., Ghanem, B.:ℓp-box admm: A versatile framework for integer programming. IEEE transactions on pattern analysis and machine intelligence41(7) (2018)

    2018

  21. [29]

    In: European Conference on Computer Vision (2022)

    Yurtsever, A., Birdal, T., Golyanik, V.: Q-fw: A hybrid classical-quantum frank- wolfe for quadratic binary optimization. In: European Conference on Computer Vision (2022). Springer

    2022

  22. [30]

    PhD thesis, University of Auckland (2017)

    Fowler, A.: Improved qubo formulations for d-wave quantum computing. PhD thesis, University of Auckland (2017)

    2017

  23. [31]

    In: 2020 International Conference on 3D Vision (3DV) (2020)

    Seelbach Benkner, M., Golyanik, V., Theobalt, C., Moeller, M.: Adiabatic quantum graph matching with permutation matrix constraints. In: 2020 International Conference on 3D Vision (3DV) (2020). IEEE

    2020

  24. [32]

    In: Computer Vision and Pattern Recognition (CVPR) (2025)

    Meli, N.K., Golyanik, V., Benkner, M.S., Moeller, M.: Qucoop: A versatile framework for solving composite and binary-parametrised problems on quantum annealers. In: Computer Vision and Pattern Recognition (CVPR) (2025)

    2025

  25. [33]

    Computer Vision and Image 27 Understanding112(1) (2008)

    Olsson, C., Eriksson, A.P., Kahl, F.: Improved spectral relaxation methods for binary quadratic optimization problems. Computer Vision and Image 27 Understanding112(1) (2008)

    2008

  26. [34]

    Journal of heuristics8(2002)

    Merz, P., Freisleben, B.: Greedy and local search heuristics for unconstrained binary quadratic programming. Journal of heuristics8(2002)

    2002

  27. [35]

    Annals of operations research63(1) (1996)

    Martin, O.C., Otto, S.W.: Combining simulated annealing with local search heuristics. Annals of operations research63(1) (1996)

    1996

  28. [36]

    SIAM journal on imaging sciences2(1) (2009)

    Beck, A., Teboulle, M.: A fast iterative shrinkage-thresholding algorithm for linear inverse problems. SIAM journal on imaging sciences2(1) (2009)

    2009

  29. [37]

    Working Paper, The Management School, Imperial College, London, England (1998)

    Beasley, J.E.: Heuristic algorithms for the unconstrained binary quadratic pro- gramming problem. Working Paper, The Management School, Imperial College, London, England (1998)

    1998

  30. [38]

    Computational Optimization and Applications65, 845–869 (2016)

    Rosenberg, G., Vazifeh, M., Woods, B., Haber, E.: Building an iterative heuristic solver for a quantum annealer. Computational Optimization and Applications65, 845–869 (2016)

    2016

  31. [39]

    Chapman and Hall/CRC, Boca Raton, Florida (2007)

    Gonzalez, T.F.: Handbook of Approximation Algorithms and Metaheuristics. Chapman and Hall/CRC, Boca Raton, Florida (2007)

    2007

  32. [40]

    arXiv preprint arXiv:2003.00133 (2020)

    Boothby, K., Bunyk, P., Raymond, J., Roy, A.: Next-generation topology of d-wave quantum processors. arXiv preprint arXiv:2003.00133 (2020)

    2020 arXiv

  33. [41]

    https://github

    D-Wave Systems, Inc.: Simulated Annealing Sampler. https://github. com/dwavesystems/dwave-neal/blob/master/docs/reference/sampler.rst. online; accessed on the 25.02.2021 (2021)

    2021

  34. [42]

    Frontiers in Physics (2022)

    Mehta, V., Jin, F., Michielsen, K., De Raedt, H.: On the hardness of quadratic unconstrained binary optimization problems. Frontiers in Physics (2022)

    2022

  35. [43]

    https://docs.ocean.dwavesys

    qbsolv - D-Wave Ocean Software Documentation. https://docs.ocean.dwavesys. com/projects/qbsolv/en/latest/intro.html. [Accessed 11-07-2024]

    2024

  36. [44]

    Physical review letters35(26), 1792 (1975)

    Sherrington, D., Kirkpatrick, S.: Solvable model of a spin-glass. Physical review letters35(26), 1792 (1975)

    1975

  37. [45]

    https://github.com/MQLib/MQLib/files/10934880/equivalence.pdf

    Dunning, I., Gupta, S., Silberholz, J.: Equivalence of QUBO and MAXCUT. https://github.com/MQLib/MQLib/files/10934880/equivalence.pdf. [Accessed 11-07-2024]

    2024

  38. [46]

    science 286(5439) (1999)

    Barab´ asi, A.-L., Albert, R.: Emergence of scaling in random networks. science 286(5439) (1999)

    1999

  39. [47]

    Physical review E65(2) (2002) 28

    Holme, P., Kim, B.J.: Growing scale-free networks with tunable clustering. Physical review E65(2) (2002) 28

    2002

  40. [48]

    Networks: An International Journal32(3) (1998)

    Koch, T., Martin, A.: Solving steiner tree problems in graphs to optimality. Networks: An International Journal32(3) (1998)

    1998

  41. [49]

    Springer, New York City, New York (2004)

    Resende, M.G., Sousa, J.P., Rosseti, I., Arag˜ ao, M.P., Ribeiro, C.C., Uchoa, E., Werneck, R.F.: New Benchmark Instances for the Steiner Problem in Graphs. Springer, New York City, New York (2004)

    2004

  42. [50]

    Physical Review E—Statistical, Nonlinear, and Soft Matter Physics86(2) (2012)

    Biazzo, I., Braunstein, A., Zecchina, R.: Performance of a cavity-method-based algorithm for the prize-collecting steiner tree problem on graphs. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics86(2) (2012)

    2012

  43. [51]

    M¨ ucke, S., Gerlach, T., Piatkowski, N.: Optimum-Preserving QUBO Parameter Compression (2023)

    2023

  44. [52]

    https://github.com/MSeelbach/SplittingForQA

    Source Code. https://github.com/MSeelbach/SplittingForQA. [Accessed 11-07- 2024] 29

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.