Pith. sign in

REVIEW 3 major objections 5 minor 40 references

Projecting a chosen constraint penalty onto a quantum annealer's native couplings yields a tuning-free QUBO that often beats denser surrogates.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 00:03 UTC pith:CIKVEEGP

load-bearing objection Clean L2 projection idea with honest limitations, but the feasibility-ordering gap and fragile Table IV statistics make the practical claim conditional. the 3 major comments →

arxiv 2607.26349 v1 pith:CIKVEEGP submitted 2026-07-28 quant-ph math.OC

Hardware-Aware QUBO Reformulation of Constrained Binary Optimization via the Walsh-Fourier Transform

Loong Kuan Lee , Harsha Nagarajan , Thore Gerlach , Sascha M\"ucke , Ragavi Krishnamoorthy , Nico Piatkowski This is my paper
classification quant-ph math.OC MSC 81P6890C2742C10 PACS 03.67.Ac
keywords QUBOquantum annealingWalsh-Fourier transformconstrained binary optimizationinequality constraintspenalty methodshardware topology
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper claims that the vexing part of encoding constrained binary optimization for quantum annealing—turning an inequality constraint into a QUBO penalty without auxiliary slack variables or hand-tuned coefficients—can be solved by one weighted least-squares projection. The user picks any penalty function that captures the constraint and a probability measure saying which slack values matter; the method returns the unique best quadratic approximation using only the couplings the hardware physically supports. Because the surrogate is built on native couplers, no minor-embedding chains are needed, and no per-constraint penalty parameters are tuned. On multidimensional-knapsack benchmarks run on real quantum annealers, this topology-aware surrogate attains the lowest mean objective gap on most instances and the smallest worst-case gap, outperforming a denser full-pairwise projection after embedding.

Core claim

For a user-chosen penalty ψ and a full-support target measure µ, the unique minimizer of ∥ψ−φ∥²_µ over φ in L_E = span{1, z_i, z_i z_j : {i,j}∈E} is the best hardware-admissible quadratic surrogate of ψ, and its coefficients solve the normal equations G θ = c. If the admissible edge set E is induced by a native placement on the annealer's graph, the resulting QUBO is chain-free by construction. The paper further reports that this topology-restricted surrogate, despite living in a strictly smaller approximation space, yields lower mean and worst-case objective gaps than a hardware-blind full-quadratic projection after embedding.

What carries the argument

The Walsh–Fourier basis of the Boolean cube: every pseudo-Boolean function has a unique multilinear expansion in characters χ_S(z)=∏_{i∈S} z_i. Projecting the chosen penalty onto the subspace spanned by the constant, linear, and hardware-admissible quadratic characters, under a user-chosen weighted inner product, gives a unique quadratic surrogate via the normal equations; the weights come from a slack-domain Gaussian target pulled back to the spin cube through a saddlepoint approximation, and self-normalized importance sampling estimates the Gram matrix and correlation vector.

Load-bearing premise

A small weighted least-squares error between the chosen penalty and its quadratic surrogate is assumed to be enough to keep feasible solutions ranked below infeasible ones; the paper states this ordering is not certified by the projection.

What would settle it

Enumerate all 2^n binary vectors for a small multidimensional-knapsack instance, compute the exact projection under the paper's Gaussian target, and check whether any infeasible vector has lower penalized energy than the best feasible vector while the projection error is small; one such counterexample would show the surrogate can change the problem's solution.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Penalty design and hardware topology become separate choices: the user supplies ψ and µ; the device supplies E; the projection reconciles them.
  • When E comes from a native placement, the projected QUBO is chain-free, eliminating chain-strength tuning and chain-break postprocessing.
  • No per-constraint penalty coefficients need tuning; only a single global multiplier per problem family remains.
  • The construction applies to any pseudo-Boolean function, so equality constraints and even higher-order objectives can be projected the same way.
  • On benchmark knapsack instances, the topology-aware surrogate achieves the lowest mean objective gap on most instances and the smallest worst-case gap, while the full-quadratic projection yields more feasible samples.

Where Pith is reading between the lines

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

  • A natural testable extension is to add a post-projection verification step that checks, by enumeration on small instances, whether the surrogate preserves feasible/infeasible ordering; the authors flag that small L2 error does not certify this ordering.
  • The Gaussian target centered at mid-feasible slack is presented as one heuristic choice; one could adapt the target measure per constraint or optimize it to enforce a margin between feasible and infeasible energies.
  • The framework suggests judging QUBO reformulations by embedded performance rather than by fit alone, which could change how future constraint-encoding methods are benchmarked.
  • For sparse constraints the projection loses fidelity, so the method is most promising for dense constraints; an intermediate feature set could trade off fidelity against hardware efficiency.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces TAWP (Topology-Aware Walsh-Fourier Penalization), a method for converting constrained binary optimization problems into QUBO form without slack variables. A user-specified penalty (e.g., Heaviside) and a target probability measure are projected in weighted L2 onto the span of constant, linear, and hardware-admissible quadratic Walsh characters. The coefficients are obtained from normal equations estimated by self-normalized importance sampling. The authors argue the resulting penalty requires no per-constraint tuning, is chain-free by construction for native placements, and empirically outperforms unbalanced penalization and full pairwise projections on D-Wave hardware.

Significance. The projection/normal-equations derivation (§III-B) is mathematically clean, and the paper is commendably explicit about its limitations. If the feasible/infeasible ordering transfer were established, TAWP would be a practically valuable QUBO reformulation tool: it replaces manually tuned per-constraint penalties with a unique weighted least-squares surrogate and makes the approximation error explicit. The availability of source code and the guarded interpretation of results are additional strengths.

major comments (3)
  1. [§VI, Eq. (22), Fig. 4] The central encoding claim—that the projected quadratic surrogate acts as a valid penalty, i.e., feasible solutions are ranked below infeasible ones in the final QUBO—does not follow from the projection theorem in §III-B. The paper concedes in §VI that 'a small weighted least-squares error does not by itself certify correct feasible/infeasible ordering.' Eq. (21) concentrates target mass on the feasible side, so the infeasible side is deliberately underweighted, and Fig. 4 is a single-constraint illustration rather than a general argument. If ordering fails, the annealer solves a different problem than the original constrained optimization. Please either derive conditions on (ψ, μ_Z, E) that guarantee a feasible/infeasible margin, or provide a systematic numerical study over all benchmark constraints showing that the final Hamiltonians separate feasible from infeasible energy levels with
  2. [§V-C, Table IV] The hardware headline rests on point estimates without repeated runs, error bars, or significance statements, and several cells are based on a single feasible sample: pb5 TAWP(topology) on Advantage reports mean=min=max=9.35 from 1/2500 feasible reads, and pb2 TAWP(topology) reports 15.29 from 1/2500; the corresponding Advantage2 rows contain 2/2500. A mean computed from one feasible draw is not evidence of distributional advantage, yet it contributes to the claim of 'lowest mean gap on 8 of 9 instances.' Please report repeated anneals with different seeds, confidence intervals or exact feasible-sample distributions, and a paired comparison test; at minimum, exclude or clearly flag single-sample cells from the headline statistics.
  3. [§V-A, Table III, Fig. 5] The relative CoP comparisons underlying the 'topology-aware projection regularizes MDKP but hurts MIS' interpretation are hard to evaluate because several absolute CoP values are zero or nearly zero (e.g., MDKP 20 all zeros; MDKP 15 embedding TAWP 18.02 vs logical 0). It is unclear whether zero CoP means no optimal solution was found or no feasible solution, and a relative change from zero is not meaningful. Please clarify the meaning of these entries and report the underlying feasible/optimal counts.
minor comments (5)
  1. [Alg. 1 and §III-C] The ratio w_k = μ_Z(z^{(k)})/ν(z^{(k)}) is introduced, but the SNIS estimators in (13)–(14) use a self-normalizing denominator Σ w_k. It would help to state explicitly that w_k are unnormalized and that normalization is part of the estimator; Algorithm 1 also omits the bin-mapping notation from §III-D.
  2. [Eq. (20), Table II] The hat notation for unbalanced-penalization shape parameters \hatλ_1,\hatλ_2 is visually close to the Walsh coefficient vector \hatθ; consider using a different symbol.
  3. [Fig. 5] The caption 'Cells labeled 0/0 have zero mean CoP for both methods' is cryptic; clarify whether this means no feasible/optimal solutions were found and how such cells are treated in aggregate plots.
  4. [§V-C / Table IV] The paper says optima are obtained with CPLEX but does not state the time limit or optimality tolerance, or whether every benchmark instance was solved to proven optimality. This matters because the gap metric (23) is relative to the true optimum.
  5. [Throughout] Several formatting artifacts remain (e.g., 'W alsh–Hadamard' in Fig. 2, 'isuniquelyandoptimallydetermined' in the abstract). These should be corrected in the camera-ready version.

Circularity Check

0 steps flagged

No circularity: the TAWP coefficients are genuine weighted least-squares projections, and the paper explicitly disclaims any proof that small L2 error preserves feasible/infeasible ordering.

full rationale

The derivation chain is self-contained. Given a user-chosen penalty ψ, a probability measure μ, and an admissible coupling set E, the normal equations (11) define the unique minimizer of ||ψ-φ||²_μ over L_E by the Hilbert projection theorem; this is a mathematical construction, not a prediction fitted to benchmark outcomes. The penalty coefficients are not fitted to the target objective or to the reported gaps; the only tuned quantities (global λ per family and the UP shape parameters) are explicitly acknowledged in Sections IV-C, IV-D, and VI. The central empirical claim—that TAWP(topology) often attains the lowest mean/worst-case gap on D-Wave hardware—is a measured outcome, not a consequence of the projection theorem, and the authors state the limit: 'a small weighted least-squares error does not by itself certify correct feasible/infeasible ordering' (Section VI). No uniqueness theorem or ansatz is imported from the authors' prior work; the only self-citation ([39]) is a normalization step in Eq. (22) and is not load-bearing for the main result. The heuristic choices of ψ and μ^Z are explicitly described as design freedom, not as derived predictions. Thus no central step reduces by construction to its own inputs. Fragile aspects such as single-feasible-sample rows in Table IV are correctness/statistical concerns, not circularity.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 0 invented entities

The paper contributes a method, not new physical entities. Its load-bearing inputs are the Heaviside origin penalty, the Gaussian slack target, the global multiplier, and the greedy placement; each is chosen by hand or tuned. The mathematical axioms are standard linear algebra and saddlepoint approximation, plus the unproven but explicitly acknowledged assumption that L2 closeness preserves feasibility ordering.

free parameters (5)
  • Global penalty multiplier λ (per problem family) = MDKP 0.1306 full, 0.0621–0.0660 topologies; MIS 0.2167-ish
    Tuned by Nelder–Mead to minimize the optimality gap on fixed problem sizes (MDKP n=15, MIS n=16). This global multiplier remains in TAWP despite the 'no per-constraint tuning' claim.
  • Target density bandwidth σ = not reported per instance
    Chosen per constraint so that at least 95% of the Gaussian mass lies on the feasible slack range; it controls concentration of the error budget and affects all projection coefficients.
  • Target center c = c = smax/2
    Hand-fixed at half the maximum feasible slack to create a 'mid-feasible' bowl; the paper calls this 'one simple choice rather than a uniquely principled one'.
  • SNIS implementation hyperparameters = N=min(2^{n-1},2^15), ε=1e-8, p_i=0.5
    Proposal distribution, sample count, and ridge parameter chosen for the experiments; no optimality criterion selects them.
  • Greedy placement parameters = documented only in code
    The logical-to-physical mapping (Appendix B) has scoring, tie-breaking, and swap-search parameters; exact values are deferred to the repository, so the topology-aware projections are not fully reproducible from the text.
axioms (6)
  • standard math Walsh characters are linearly independent and form an orthonormal basis under the uniform measure; full support of μ makes the Gram matrix positive definite.
    Used in Section III-B to assert that the normal equations have a unique global minimizer.
  • standard math Hilbert projection theorem gives the minimizer of the weighted L2 error.
    Used in Section III-B, Eq. (12)–(15), to justify that the solution of the normal equations is the best approximation.
  • domain assumption The Lugannani–Rice saddlepoint formula yields sufficiently accurate approximations of the proposal slack bin probabilities for SNIS.
    Section III-D relies on the saddlepoint CDF to set weights; the approximation error is not bounded.
  • ad hoc to paper The Gaussian target centered at half max feasible slack encodes the correct notion of penalty error.
    Section IV-C; validated only empirically for one constraint in Figure 4, not proven generally.
  • ad hoc to paper Small projected-penalty error preserves feasible/infeasible ordering.
    Section VI explicitly says this is not certified; the method's practical usefulness depends on this unproven property.
  • domain assumption SQA and D-Wave sampling return low-energy samples representative enough for CoP and objective-gap comparisons.
    Section V; all conclusions about solution quality rely on the sampling distribution being informative about the QUBO energy landscape.

pith-pipeline@v1.3.0-alltime-deepseek · 16292 in / 13857 out tokens · 143467 ms · 2026-08-01T00:03:38.178614+00:00 · methodology

0 comments
read the original abstract

We present a novel slack-free, penalty-based framework for reformulating constrained binary optimization as Quadratic Unconstrained Binary Optimization (QUBO) on near-term quantum annealing hardware. Given a user-chosen penalty function that most naturally captures a constraint---typically non-quadratic, such as a Heaviside-function surrogate---and a target probability measure over the Boolean hypercube, our method returns the weighted least-squares projection of the chosen penalty function onto the subspace spanned by linear and quadratic Walsh--Fourier characters that correspond to physically realizable couplings on the target hardware graph. Within this restricted family, the resulting quadratic surrogate is uniquely and optimally determined by the normal equations: unlike state-of-the-art approaches, it introduces no per-constraint penalty coefficients to tune and avoids dense all-pairs couplings by construction. Two practical consequences follow. First, the projected penalty respects device connectivity, reducing chain lengths and physical-qubit overhead after minor embedding. Second, we show empirically that this hardware-native surrogate can outperform denser full-pairwise projections, despite being drawn from a strictly smaller approximation space. This advantage widens once the QUBO is embedded and sampled on quantum annealers, yielding samples with the lowest worst-case and mean objective gaps compared to unbalanced penalization and a hardware-blind projection onto all quadratic terms.

Figures

Figures reproduced from arXiv: 2607.26349 by Harsha Nagarajan, Loong Kuan Lee, Nico Piatkowski, Ragavi Krishnamoorthy, Sascha M\"ucke, Thore Gerlach.

Figure 1
Figure 1. Figure 1: Distribution of slack values sj (x) = a⊤ j x−bj under uniform measure on {0, 1} n, for two constraints of the SAC-94 pet3 MDKP instance [35]. C. Limitations of the Uniform Measure Generally, for any penalty ψ that depends on x only through the slack sj (x) = a ⊤ j x − bj and any full-support measure µ, the squared error of a surrogate ϕ splits across the attainable slack values, writing sj := sj (x) via ab… view at source ↗
Figure 2
Figure 2. Figure 2: Connectivity-aware penalty approximation: origin penalty [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Exact target bin masses ul vs. their importance-sampling estimates for constraint j = 1 of pet3; estimates track the exact masses closely. domain first, namely µ˜sj , and then construct the corre￾sponding spin-domain measure µ (j) Z . To address both issues, we draw samples z (k) ∼ ν, k = 1, 2, . . . , N, from a simple proposal distribution ν. Then, using SNIS, we estimate G and c by Gˆ := 1 P k wk X N k=1… view at source ↗
Figure 4
Figure 4. Figure 4: Exact projection of Heaviside penalty ψ onto full quadratic subspace for constraint j = 1 of pet3. Lines: mean projected penalty per slack value; bands: min–max range. TABLE II TUNED GLOBAL PENALTY MULTIPLIERS PER PROBLEM FAMILY AND METHOD; λˆ 1, λˆ 2 ARE THE ADDITIONAL SHAPE PARAMETERS OF UNBALANCED PENALIZATION. Family TAWP Unbalanced P. Full Chim. Peg. Zeph. λ λˆ 1 λˆ 2 MDKP 0.1306 0.0621 0.0632 0.0660 … view at source ↗
Figure 5
Figure 5. Figure 5: Mean relative Coefficient of Performance (CoP) difference against the unbalanced-penalization base; negative values favor unbalanced penalization. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Relative CoP difference before and after embedding on each D-Wave [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

40 extracted references · 5 linked inside Pith

  1. [1]

    Leveraging quantum comput- ing for accelerated classical algorithms in power systems optimization,

    R. Barrass, H. Nagarajan, and C. Coffrin, “Leveraging quantum comput- ing for accelerated classical algorithms in power systems optimization,” inIntegration of Constraint Programming, Artificial Intelligence, and Operations Research (CPAIOR), ser. Lecture Notes in Computer Sci- ence, vol. 15762. Springer, 2025, pp. 52–67

    2025

  2. [2]

    Hybrid quantum-classical multi-cut Benders approach with a power system application,

    N. G. Paterakis, “Hybrid quantum-classical multi-cut Benders approach with a power system application,”Computers & Chemical Engineering, vol. 172, p. 108161, 2023

    2023

  3. [3]

    Portfolio optimisation using the D- Wave quantum annealer,

    F. Phillipson and H. S. Bhatia, “Portfolio optimisation using the D- Wave quantum annealer,” inInternational Conference on Computational Science (ICCS 2021). Springer, 2021, pp. 45–59

    2021

  4. [4]

    Traffic flow optimization using a quantum annealer,

    F. Neukart, G. Compostella, C. Seidel, D. von Dollen, S. Yarkoni, and B. Parney, “Traffic flow optimization using a quantum annealer,” Frontiers in ICT, vol. 4, p. 29, 2017

    2017

  5. [5]

    Hybrid quantum-classical multi-agent pathfinding,

    T. Gerlach, L. K. Lee, F. Barbaresco, and N. Piatkowski, “Hybrid quantum-classical multi-agent pathfinding,” inInternational Conference on Machine Learning. PMLR, 2025, pp. 19 161–19 171

    2025

  6. [6]

    Ising formulations of many NP problems,

    A. Lucas, “Ising formulations of many NP problems,”Frontiers in Physics, vol. 2, p. 5, 2014

    2014

  7. [7]

    Quantum bridge analytics I: a tutorial on formulating and using QUBO models,

    F. Glover, G. Kochenberger, R. Hennig, and Y . Du, “Quantum bridge analytics I: a tutorial on formulating and using QUBO models,”Annals of Operations Research, vol. 314, no. 1, pp. 141–183, 2022

    2022

  8. [8]

    L. A. Wolsey,Integer Programming. John Wiley & Sons, 2020

    2020

  9. [9]

    On the equal-subset-sum problem,

    G. J. Woeginger and Z. Yu, “On the equal-subset-sum problem,” Information Processing Letters, vol. 42, no. 6, pp. 299–302, 1992

    1992

  10. [10]

    Quantum annealing in the transverse Ising model,

    T. Kadowaki and H. Nishimori, “Quantum annealing in the transverse Ising model,”Physical Review E, vol. 58, no. 5, pp. 5355–5363, 1998

    1998

  11. [11]

    A quantum approximate optimization algorithm,

    E. Farhi, J. Goldstone, and S. Gutmann, “A quantum approximate optimization algorithm,”arXiv preprint arXiv:1411.4028, 2014

    Pith/arXiv arXiv 2014

  12. [12]

    From the quantum approximate optimization algorithm to a quantum alternating operator ansatz,

    S. Hadfield, Z. Wang, B. O’Gorman, E. G. Rieffel, D. Venturelli, and R. Biswas, “From the quantum approximate optimization algorithm to a quantum alternating operator ansatz,”Algorithms, vol. 12, no. 2, p. 34, 2019

    2019

  13. [13]

    Evaluating Ising processing units with integer programming,

    C. Coffrin, H. Nagarajan, and R. Bent, “Evaluating Ising processing units with integer programming,” inInternational Conference on Integration of Constraint Programming, Artificial Intelligence, and Operations Re- search (CPAIOR). Springer, 2019, pp. 163–181

    2019

  14. [14]

    Unbalanced penalization: A new approach to encode inequality constraints of combinatorial problems for quantum optimiza- tion algorithms,

    J. A. Monta ˜nez-Barrera, D. Willsch, A. Maldonado-Romo, and K. Michielsen, “Unbalanced penalization: A new approach to encode inequality constraints of combinatorial problems for quantum optimiza- tion algorithms,”Quantum Science and Technology, vol. 9, no. 2, p. 025022, Apr. 2024

    2024

  15. [15]

    Embedding algorithms for quantum annealers with Chimera and Pegasus connection topologies,

    S. Zbinden, A. B ¨artschi, H. Djidjev, and S. Eidenbenz, “Embedding algorithms for quantum annealers with Chimera and Pegasus connection topologies,” inInternational Conference on High Performance Comput- ing (ISC 2020). Springer, 2020, pp. 187–206

    2020

  16. [16]

    Improving performance in combinatorial optimization problems with inequality constraints: An evaluation of the unbalanced penalization method on D-Wave Advantage,

    J. A. Monta ˜nez-Barrera, P. van den Heuvel, D. Willsch, and K. Michielsen, “Improving performance in combinatorial optimization problems with inequality constraints: An evaluation of the unbalanced penalization method on D-Wave Advantage,” inIEEE International Conference on Quantum Computing and Engineering (QCE), vol. 1. IEEE, 2023, pp. 535–542

    2023

  17. [17]

    Sparse QUBO Formulation for Efficient Embedding via Network-Based Decomposition of Equality and Inequality Constraints,

    K. Suda, S. Naito, and Y . Hasegawa, “Sparse QUBO Formulation for Efficient Embedding via Network-Based Decomposition of Equality and Inequality Constraints,” arXiv preprint arXiv:2601.18108, Jan. 2026

    arXiv 2026

  18. [18]

    Penalty-free approach to accelerating constrained quantum optimization,

    D. Bucher, J. Stein, S. Feld, and C. Linnhoff-Popien, “Penalty-free approach to accelerating constrained quantum optimization,”Physical Review A, vol. 112, no. 6, p. 062605, 2025

    2025

  19. [19]

    Efficient QAOA architecture for solving multi-constrained optimization problems,

    D. Bucher, D. Porawski, M. Janetschek, J. Stein, C. O’Meara, G. Cor- tiana, and C. Linnhoff-Popien, “Efficient QAOA architecture for solving multi-constrained optimization problems,” in2025 IEEE International Conference on Quantum Computing and Engineering (QCE), vol. 1. IEEE, 2025, pp. 356–367

    2025

  20. [20]

    Quantum tree generator improves QAOA state-of-the-art for the knap- sack problem,

    P. J. Christiansen, L. Binkowski, D. Ramacciotti, and S. Wilkening, “Quantum tree generator improves QAOA state-of-the-art for the knap- sack problem,” in2025 IEEE International Conference on Quantum Computing and Engineering (QCE), vol. 1. IEEE, 2025, pp. 1–10

    2025

  21. [21]

    Solving inequality- constrained binary optimization problems on quantum annealer,

    K. Yonaga, M. J. Miyama, and M. Ohzeki, “Solving inequality- constrained binary optimization problems on quantum annealer,”arXiv preprint arXiv:2012.06119, 2020

    Pith/arXiv arXiv 2012

  22. [22]

    Efficient light source placement using quantum computing,

    S. M ¨ucke and T. T. Gerlach, “Efficient light source placement using quantum computing,” inLernen, Wissen, Daten, Analysen (LWDA), 2023

    2023

  23. [23]

    Quantum annealing with inequality constraints: the set cover problem,

    H. N. Djidjev, “Quantum annealing with inequality constraints: the set cover problem,”Advanced Quantum Technologies, vol. 6, no. 11, p. 2300104, 2023

    2023

  24. [24]

    Cutting Slack: Quantum Optimization with Slack-Free Methods for Combinatorial Benchmarks,

    M. Sharma and H. C. Lau, “Cutting Slack: Quantum Optimization with Slack-Free Methods for Combinatorial Benchmarks,” arXiv preprint arXiv:2507.12159, 2025

    Pith/arXiv arXiv 2025

  25. [25]

    Subgradient Method Using Quantum Annealing for Inequality-Constrained Binary Optimization Problems,

    T. Takabayashi, T. Goto, and M. Ohzeki, “Subgradient Method Using Quantum Annealing for Inequality-Constrained Binary Optimization Problems,”Journal of the Physical Society of Japan, vol. 94, no. 5, p. 054003, May 2025

    2025

  26. [26]

    Efficient Construction of Feasible Solutions in Column Generation Using Quantum Annealing,

    T. Takabayashi, N. Maruyama, T. Yoshihara, R. Haba, and M. Ohzeki, “Efficient Construction of Feasible Solutions in Column Generation Using Quantum Annealing,”Journal of the Physical Society of Japan, vol. 94, no. 9, p. 094005, Sep. 2025

    2025

  27. [27]

    Implementing Slack-Free Custom Penalty Function for QUBO on Gate-Based Quantum Computers,

    X. W. Lee and H. C. Lau, “Implementing Slack-Free Custom Penalty Function for QUBO on Gate-Based Quantum Computers,” in2025 IEEE International Conference on Quantum Computing and Engineering (QCE). Albuquerque, NM, USA: IEEE, Aug. 2025, pp. 2112–2119

    2025

  28. [28]

    Next-generation topol- ogy of D-Wave quantum processors,

    K. Boothby, P. Bunyk, J. Raymond, and A. Roy, “Next-generation topol- ogy of D-Wave quantum processors,”arXiv preprint arXiv:2003.00133, 2020

    Pith/arXiv arXiv 2003

  29. [29]

    A practical heuristic for finding graph minors,

    J. Cai, W. G. Macready, and A. Roy, “A practical heuristic for finding graph minors,”arXiv preprint arXiv:1406.2741, 2014

    Pith/arXiv arXiv 2014

  30. [30]

    A Closed Set of Normal Orthogonal Functions,

    J. L. Walsh, “A Closed Set of Normal Orthogonal Functions,”American Journal of Mathematics, vol. 45, no. 1, pp. 5–24, Jan. 1923

    1923

  31. [31]

    On Walsh-Fourier series,

    G. W. Morgenthaler, “On Walsh-Fourier series,”Transactions of the American Mathematical Society, vol. 84, no. 2, pp. 472–507, 1957

    1957

  32. [32]

    O’Donnell,Analysis of Boolean Functions

    R. O’Donnell,Analysis of Boolean Functions. Cambridge: Cambridge University Press, 2014

    2014

  33. [33]

    Rudin,Functional Analysis

    W. Rudin,Functional Analysis. New York: McGraw-Hill, 1991

    1991

  34. [34]

    Kreyszig,Introductory Functional Analysis with Applications

    E. Kreyszig,Introductory Functional Analysis with Applications. New York: John Wiley & Sons, 1978

    1978

  35. [35]

    Benchmark instances for the Multidimensional Knapsack Problem,

    J. H. Drake, “Benchmark instances for the Multidimensional Knapsack Problem,” DOI: 10.13140/2.1.3578.9122, 2015

    arXiv 2015

  36. [36]

    Boyd and L

    S. Boyd and L. Vandenberghe,Convex Optimization. Cambridge: Cambridge University Press, 2004

    2004

  37. [37]

    Saddle point approximation for the distribu- tion of the sum of independent random variables,

    R. Lugannani and S. Rice, “Saddle point approximation for the distribu- tion of the sum of independent random variables,”Advances in Applied Probability, vol. 12, no. 2, pp. 475–490, Jun. 1980

    1980

  38. [38]

    Challenge Problems: Independent Sets in Graphs,

    N. J. A. Sloane, “Challenge Problems: Independent Sets in Graphs,” https://oeis.org/A265032/a265032.html, Oct. 2000

    2000

  39. [39]

    Standardization of Multi- Objective QUBOs,

    L. K. Lee, T. Gerlach, and N. Piatkowski, “Standardization of Multi- Objective QUBOs,” in2025 IEEE International Conference on Quantum Computing and Engineering (QCE). Albuquerque, NM, USA: IEEE, Aug. 2025, pp. 58–64

    2025

  40. [40]

    Defining and detecting quantum speedup,

    T. F. Rønnow, Z. Wang, J. Job, S. Boixo, S. V . Isakov, D. Wecker, J. M. Martinis, D. A. Lidar, and M. Troyer, “Defining and detecting quantum speedup,”Science, vol. 345, no. 6195, pp. 420–424, 2014

    2014