Pith. sign in

REVIEW 4 major objections 6 minor 34 references

Scalable Variational Quantum Optimization via Pauli Correlation Encoding: Application to Large-Scale Power Demand Portfolio Optimization

T0 review · 4 major / 6 minor · reviewed 2026-07-31 · grok-4.5

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

Pith's one-line read Pauli correlators let a handful of qubits encode and optimize power-portfolio problems with thousands of binary variables to near-certified optimality.

desk verdict Solid applied PCE stress-test on dense real QUBOs to m~10k; the large-m headline is mostly classical greedy on a flat landscape, and the authors mostly say so. read the letter →

arxiv 2607.24722 v1 pith:25DUXF6O submitted 2026-07-27 quant-ph

Takuya Yoshioka , Keita Sasada , Riku Usuki , Yuichiro Nakano , Keisuke Fujii This is my paper · ORCID
classification quant-ph
keywords PaulicorrelationencodingvariationalquantumalgorithmsQUBOpowerdemandportfoliooptimizationqubit-efficientcontinuousrelaxationtrapped-ionhardwareNISQ
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

Variational quantum optimizers usually need one qubit per binary variable, so large real-world problems do not fit on near-term hardware. This paper shows that encoding those variables as the signs of multi-body Pauli correlators—expectation values of k-body Pauli products—turns a small quantum state into a continuous relaxation of a huge binary portfolio. Applied to electric power demand portfolio selection, the method, with a time-averaged warm start and greedy cleanup after decoding, reaches normalized cost gaps on the order of 10^{-4} versus certified optima for instances from 18 up to 10,296 consumers, using at most 14 qubits. Performance hinges on how finely the correlator distribution resolves the binary decisions: larger problems produce denser, more continuous correlators, so continuous loss improvements more reliably become better discrete portfolios. The same circuits still yield high-quality solutions on a trapped-ion processor after post-processing, despite noise and finite shots.

What carries the argument

Pauli correlation encoding (PCE): classical bits are the signs of expectation values of k-body Pauli products (X/Y/Z on the same k-subset), yielding m=3 C(n,k) continuous variables from n qubits (here k=n/2), optimized via a sigmoid-relaxed loss then threshold-decoded.

What would settle it

On a dense portfolio instance of several thousand variables, run the same depth-5 ansatz and decoding pipeline and check whether the decoded-plus-greedy cost stays within ~10^{-4} of a certified optimum; if continuous loss falls while decoded cost stays at the random-greedy baseline, the claim that correlator resolution transmits improvements fails.

Watch

Extended reading notes

Core claim

Pauli correlation encoding is a qubit-efficient variational framework for dense quadratic binary problems with real nonuniform coefficients: binary portfolio choices are represented by the signs of k-body Pauli correlators generated by a shallow fully connected ansatz, optimized as a continuous relaxation, then decoded and greedily refined. On power-demand portfolio QUBOs this complete workflow achieves near-optimal normalized cost gaps (~10^{-4}) from m=18 to m=10,296 with n≤14 qubits, and the quality is controlled by the effective resolution of the correlator representation—the bridge between continuous loss and discrete cost—which becomes more reliable as system size grows.

Load-bearing premise

The fixed shallow circuit must still generate enough problem-relevant structure in its many correlated Pauli expectation values that lowering the continuous loss actually improves the decoded binary portfolio, even though those correlators are not independent controls.

Editorial extensions

If this is right

  • Dense fully connected QUBOs with real coefficients can be attacked with qubit counts logarithmic in problem size via k~n/2 correlators rather than one qubit per variable.
  • A time-averaged Model-1 warm start plus time-resolved Model-2 refinement is a practical two-stage recipe for operational portfolio problems.
  • As m grows, correlator histograms become continuous and regularization toward 1/2 becomes unnecessary or harmful; hyperparameter policy should track that size transition.
  • Hardware noise mainly corrupts signs near zero; three global bases plus greedy post-processing can still recover near-simulation cost on trapped-ion devices.
  • Overall wall-clock cost remains classical-dominated (O(m^2) QUBO evaluation and shot needs near the decoding boundary), so representational scalability is not end-to-end quantum advantage.

Reading between the lines

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

  • If ansatz expressivity does not scale with correlation order k, PCE may asymptote to a structured random initializer for classical local search rather than a genuine quantum optimizer.
  • The same continuous-to-discrete resolution story should transfer to other dense real-coefficient QUBOs (e.g., finance or logistics) where certified classical solvers still struggle past a few thousand variables.
  • Shot budgets that adaptively increase only for correlators near the sign boundary could cut hardware cost without changing the encoding.
  • Comparing PCE against other compressed encodings on identical power-portfolio instances would isolate whether multi-body correlators, not just any relaxation, drive the observed gaps.
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 / 6 minor

Summary. The manuscript applies Pauli correlation encoding (PCE; Sciorilli et al. 2025) — representing m binary variables via expectation values of k-body Pauli correlators on n qubits, with m = 3*C(n,k) — to electric power demand portfolio optimization, a fully connected QUBO with real, nonuniform coefficients built from open consumption data (plus mixup augmentation for m>2,143). Using a depth-5 Ry–Rz–Rzz ansatz with k=n/2 (n=4 to 14), BFGS optimization of a sigmoid-relaxed loss in exact statevector simulation, sign decoding, and a single-pass greedy post-processing step, the authors report normalized cost gaps of order 1e-4 (down to 1.4e-5) relative to Gurobi-certified optima across m = 18 to 10,296. They analyze the interplay between the continuous relaxed loss and the decoded binary cost via correlator distributions, identifying a size-dependent transition from discrete to effectively continuous representations, and validate robustness on IonQ Forte with fixed optimized parameters and 3x4,001 shots. The paper is candid that at the largest sizes the QPU-derived initializations converge to the greedy(random) baseline after post-processing and that the landscape becomes increasingly degenerate.

Significance. If the results stand, this is a useful and unusually careful empirical study of compressed quantum encodings (PCE/QRAO-like) on dense, real-valued QUBOs at scales (m up to 10,296 variables on 14 qubits) well beyond prior PCE demonstrations. Strengths worth naming: (i) all headline numbers are benchmarked against Gurobi-certified optima (relative MIP gap 1e-4), with normalization by independently computed C_min/C_max; (ii) the size sweep is complete rather than cherry-picked, and loss/cost traces, correlator histograms, match probabilities, and greedy-random distributions are all reported; (iii) the mechanistic analysis linking correlator-distribution resolution to loss-to-decoded-cost transmission is a genuine conceptual contribution to understanding relaxation-based quantum optimization; (iv) fixed-parameter IonQ Forte runs with a fixed shot budget provide an honest hardware robustness data point; and (v) the paper explicitly distinguishes representational from computational scalability and disclaims quantum advantage. Notably, the large-m degeneration of the quantum contribution is itself a scientifically valuable (partly negative) result that the community needs documented.

major comments (4)
  1. [Abstract; Sec. IV A; Sec. VI; Fig. 2; Fig. 7; Table II; Table III] Attribution of the headline gap at large m. The abstract and Sec. VI state that the PCE workflow achieves 10^{-4}-10^{-5} normalized gaps 'across problem sizes from m=18 to 10,296', but the paper's own data show that at m=2,772 and 10,296 the quantum component contributes no measurable value beyond random initialization: in Fig. 2 and Fig. 7(e,f) the QPU post-processed gaps sit at the mean of the greedy(random) distribution; Table III gives QPU-vs-optimal match probabilities of 0.5191/0.5125 (random agreement for balanced portfolios); and Table II shows that at m=10,296 post-processing alone improves the gap from 6.80e-4 to 1.36e-5 (~50x). The body (Secs. IV A, IV D, V) discloses this honestly, but the abstract, the contribution list in Sec. I, and the conclusion do not. Since the m>=2,772 instances are what carry the 'scalable' claim, the abstract/conclusion should state explicitly that
  2. [Sec. III B 2-3; Table I; Appendix A] Hyperparameter and run selection on the reported metric. Per Table I and Appendix A, alpha_sc and beta are chosen per instance by grid search minimizing the same decoded cost C_T that is then reported as the result; theta* is selected as the trajectory minimum of the decoded cost (Sec. III B 3); and the reported run is the best of 5 random restarts. The reported 10^{-4} gaps are therefore best-case values under selection on the evaluation metric itself, which inflates apparent performance relative to any deployed use. The paper should either adopt a selection rule decoupled from the reported objective (e.g., select on the relaxed loss, or on a held-out time window / separate validation instances generated with different consumer subsets) or report the full distribution (mean +/- std over the 5 runs and over neighboring (alpha_sc, beta) values) alongside the selected best, so the magnitud
  3. [Sec. IV A; Sec. IV D; Fig. 5] Strength of classical baselines. The only classical comparisons are single-pass greedy from the all-zero string and from 1,000 random strings, both with the same O(m) update rule as the PCE post-processing. Given the demonstrated flatness of the landscape at large m (Secs. IV A, IV D; degeneracy of near-optima), these baselines are weak: any reasonable classical heuristic (multi-pass local search, simulated annealing, or multi-start greedy with matched evaluation budget) would plausibly also reach 10^{-4} gaps at m=10,296. This matters because the paper claims 'clear advantages in intermediate regimes' and frames PCE as providing 'structured initialization'. To substantiate that the PCE initialization is doing something a cheap classical method cannot, at least one stronger classical baseline with a matched objective-evaluation budget should be included, particularly at m=756-2,772 where
  4. [Sec. IV B; Fig. 3; Fig. 4; Appendix B 2] The 'effective resolution' mechanism is supported only at two endpoints. The central conceptual claim -- that the transition from discrete to effectively continuous correlator distributions governs how reliably loss improvements transmit to decoded cost -- is illustrated only for m=18 and m=10,296 (Figs. 3-4). A quantitative, size-resolved diagnostic would substantially strengthen it: e.g., the fraction of correlators within the finite-shot resolution 1/sqrt(N_shot) of the decoding boundary, or the empirical probability that a loss-decreasing BFGS step changes the decoded bitstring, plotted across all six problem sizes for simulation and QPU. This would convert the mechanism from a two-point narrative into a tested scaling claim, and would also directly quantify when hardware runs cross into the random-init regime identified in Appendix B.
minor comments (6)
  1. [Fig. 4; Eq. (12)] Fig. 4 caption defines y_i(theta*) = sigmoid(2 alpha_sc <Pi>), but Eq. (12) uses 2 alpha with alpha = alpha_sc n^{floor(k/2)}. Given alpha_sc=1.5 and alpha=6.0 for m=18, one of these is a typo; please make the caption consistent with Eq. (12).
  2. [Sec. III B 2-3] The symbol theta* is used with two different meanings: in Sec. III B 2 it denotes the best of five runs by decoded cost, while in Sec. III B 3 it denotes the trajectory minimum of the decoded cost within a run. Distinct notation (e.g., theta*_run vs theta*_dec) would avoid confusion, especially since Table I's n_iter refers to the former and Figs. 3-4 to the latter.
  3. [Sec. II A] The mixup augmentation used to construct m>2,143 instances (Sec. II A) assumes convex combinations of consumer profiles preserve the relevant covariance structure. Since the two largest instances carry the scalability claim, a brief validation (e.g., comparing the spectrum or typical entries of Cov(p_{t,i},p_{t,j}) for synthetic vs real subsets) would be worthwhile.
  4. [Table III] Table III: please state in the caption that ~0.5 is the expected agreement for balanced random portfolios (P_target = 1/2 sum E[p]), so readers can calibrate the large-m values 0.5191/0.5125 without deriving it themselves.
  5. [Sec. IV C] Hardware details are thin: please specify the IonQ Forte configuration/generation used, whether any error mitigation was applied (apparently none -- worth stating), and the native gate compilation overhead of the fully connected Rzz layer, since this bears on the source of the correlator broadening in Fig. 8.
  6. [Abstract; Appendix B 1; Sec. V; Sec. III B 1; Table I] Typos/presentation: 'ranging fromm= 18' (abstract); 'greedy local greedy search' (Appendix B 1); 'deeper or more expressive ansatz' -> 'ansätze' (Sec. V); the depth l=5 choice is justified only empirically (Sec. III B 1) -- a sentence on the observed sensitivity to l, even if a full study is out of scope, would help; n_params in Table I is never given a formula.

Circularity Check

1 steps flagged · score 1.0 of 10

Empirical PCE application benchmarked against external Gurobi optima; only minor circularity from per-instance hyperparameter and θ* selection on the same decoded cost that is reported.

  1. fitted input called prediction [Sec. IV A / Table I; Sec. III B 3 (θ* definition)]
    "The hyperparameters α_sc and β are selected to minimize the total cost C_T via a grid search. ... For each Model 1 instance, the variational optimization is repeated over five independent runs with different random initial circuit parameters. Among these runs, we select the one that yields the minimum decoded cost ... and denote the corresponding circuit parameters by θ*. ... we denote by θ* the variational parameters along the optimization trajectory that minimize the decoded binary cost without post-processing"

    α_sc, β, run selection, and the reported iterate θ* are chosen explicitly to minimize the same decoded binary cost C_T (or C_t) that appears in the headline normalized gaps. The continuous loss is what is optimized, but the published solution is the best decoded cost along the path and across a small grid/restart set. This mildly forces reported performance toward the evaluation metric by construction; it does not, however, define C_min/C_max or the Gurobi baseline, so it does not collapse the main claim.

full rationale

The paper is a methods-and-benchmarks work, not a first-principles derivation. Binary variables are relaxed via Pauli correlators (PCE from Sciorilli et al., external), optimized under a continuous loss, then sign-decoded and greedily refined; performance is scored by normalized gaps against C_min/C_max from a commercial MIP solver (Gurobi) and against greedy baselines. Those external references are independent of the PCE ansatz and of the authors’ prior work, so the central near-optimality claim is not forced by construction. Mild circularity appears only in experimental protocol: α_sc and β are grid-searched to minimize the decoded cost C_T that is later reported, and θ* is chosen along each trajectory as the iterate that minimizes decoded binary cost rather than the continuous loss being optimized. That is standard (if slightly optimistic) hyperparameter/trajectory selection on the evaluation metric, not a self-definitional reduction of the scientific claim. No uniqueness theorem is imported from overlapping authors, no ansatz is smuggled in as a theorem, and no fitted physical constant is renamed a prediction. Score 1 reflects that single minor protocol issue; the derivation/benchmark chain is otherwise self-contained.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

Load-bearing content is mostly standard QUBO modeling plus the imported PCE construction. Free knobs are the relaxation sharpness, regularizer, fixed ansatz depth, and per-size grid picks. No new physical entities; the main modeling risks are mixup augmentation and the unproven adequacy of a shallow ansatz for k=n/2 correlator manifolds.

free parameters (5)
  • α_sc (sigmoid sharpness scale; α=α_sc n^{⌊k/2⌋}) = size-dependent: 1.5, 0.1, 0.1, 0.5, 0.1, 0.1 for m=18…10296
    Grid-searched per problem size to minimize decoded C_T; controls how polarized continuous variables become.
  • β (quadratic regularizer toward y_i=1/2) = 0.1 for m≤210; 0.0 for m≥756
    Grid-searched; finite for small m, set to 0 for large m based on decoded cost.
  • circuit depth l = 5
    Fixed by hand as expressivity/cost tradeoff; not ablated systematically.
  • N_shot per measurement basis = 4001
    Fixed hardware sampling budget; sets correlator sign resolution ~1/sqrt(N_shot).
  • best-of-5 random circuit initializations = 5 seeds; best retained
    Reported Model-1 solution is the run with minimum decoded cost among five seeds.
assumptions (6)
  • domain assumption Binary variables may be represented as signs of k-body Pauli correlators with m=3 C(n,k), and sigmoid-relaxed expectations yield a useful continuous QUBO surrogate (PCE of Sciorilli et al.).
    Imported encoding; entire variational pipeline rests on it (Sec. III).
  • domain assumption Hourly portfolio cost is the quadratic mean-variance form C_t(x)=σ_t(x)^2+(P_t(x)-P_target_t)^2 with P_target_t half the mean available reduction.
    Problem definition Eqs. (1)–(8); standard soft-constraint QUBO modeling choice.
  • ad hoc to paper Mixup convex combinations of consumer time series preserve statistical structure enough that large-m synthetic instances remain valid benchmarks.
    Used for all m>2143 (Sec. II A); no external check that near-optima degeneracy matches real data.
  • ad hoc to paper A depth-5 fully connected Ry/Rz/Rzz ansatz plus BFGS on exact statevector gradients is expressive enough for the correlator manifold at k=n/2 up to n=14.
    Fixed without expressibility proof; Sec. V flags this as open.
  • domain assumption Single-pass greedy bit flips after sign decoding are an acceptable completion of the solver when reporting ‘PCE’ performance.
    Complete workflow definition (Sec. III B 3); comparisons try to isolate init quality but final gaps include greedy.
  • standard math Standard linear algebra / QUBO arithmetic and variational principle for parameterized circuits.
    Background throughout Sec. III.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Scalable Variational Quantum Optimization via Pauli Correlation Encoding: Application to Large-Scale Power Demand Portfolio Optimization." pith.science (2026). https://pith.science/paper/25DUXF6O

@misc{pith2026260724722,
  author       = {Pith},
  title        = {Pith review of: Scalable Variational Quantum Optimization via Pauli Correlation Encoding: Application to Large-Scale Power Demand Portfolio Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/25DUXF6O}},
  note         = {Machine review of arXiv:2607.24722}
}
abstract

Variational quantum algorithms offer a promising route to combinatorial optimization, but their applicability is limited by the challenge of encoding large-scale problems within restricted qubit resources. In this work, we introduce a scalable variational framework based on Pauli correlation encoding (PCE) and apply it to electric power demand portfolio optimization. Binary variables are represented through expectation values of Pauli correlation operators, which encode multi-body correlations of the quantum state and provide a continuous relaxation enabling compact representations with few qubits. We further propose a two-stage hybrid formulation, in which a time-averaged problem provides initialization for a time-resolved optimization. Numerical simulations demonstrate near-optimal performance across problem sizes ranging from $m$=18 to 10,296, with normalized cost gaps on the order of $10^{-4}$ relative to solutions with certified optimality. We show that the performance is governed by the interplay between continuous relaxation and discretization: the effective resolution of the correlator representation determines how reliably improvements in the continuous loss translate into better discrete solutions, with larger systems exhibiting more consistent behavior. Finally, we demonstrate robustness on a trapped-ion quantum processor, where high-quality solutions are obtained despite noise and finite sampling. These results establish PCE as a physically motivated and qubit-efficient framework for large-scale combinatorial optimization.

Figures

Figures reproduced from arXiv: 2607.24722 by the authors.

Figure 1
Figure 1. FIG. 1. Maximum representable problem size [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Normalized cost gap as a function of the problem [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Iteration dependence of the normalized loss [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Frequency distributions of the sets [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Time-resolved deviation from the target power and coefficient of variation for three problem sizes ( [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. Normalized cost gap ∆ [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. Distributions of normalized cost gaps ∆ [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8. Histograms of the sets [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: FIG. 9. Power procurement profiles for the time-resolved formulation (Model 2) with [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 5
Figure 5. Figure 5: For the largest instance (m = 10,296), a similar stacked visualization is not shown, as the large number of con￾sumers leads to substantial visual crowding. Instead, the aggregate performance is summarized through the devi￾ation and CV metrics in [PITH_FULL_IMAGE:figu…
Figure 10
Figure 10. Figure 10: FIG. 10. Same as Fig. 9, but for the instance with [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 1 canonical work pages

  1. [1]

    Model 1: Time-Averaged Portfolio Optimization Model 1 is a time-averaged formulation in which a com- mon portfolio is selected over a time window of length nT . This model is introduced to capture the dominant statistical structure of uncertainty in available demand reduction and to provide effective initial parameters for the subsequent time-resolved opt...

  2. [2]

    (1) is minimized independently at each time stept

    Model 2: Time-Resolved Portfolio Optimization Model 2 corresponds to the fully time-resolved formu- lation, in which the hourly costC t(x) defined in Eq. (1) is minimized independently at each time stept. This formulation represents the practically motivated opera- tional model used to evaluate procurement performance under realistic fluctuations in avail...

  3. [3]

    The parameterized quantum state|ψ(θ)⟩is pre- pared by a layered circuit ansatz

    Parameterized Quantum Circuit In the present work, the quantum circuit is used as a learning-based parametric model that generates Pauli correlators, rather than as a standalone quantum algo- rithm. The parameterized quantum state|ψ(θ)⟩is pre- pared by a layered circuit ansatz. Each layer consists of the sequential application of (i)R y rotations on all q...

  4. [4]

    Continuous Relaxation and Loss Function To enable variational quantum optimization, the bi- nary variablesxare relaxed to continuous variables y(θ)∈[0,1] m constructed from the Pauli correlators gen- erated by the parameterized circuit. Specifically, each relaxed variable is defined as yi(θ) =ς 2α⟨Π(k) i ⟩θ ,(12) whereς(x) = [1 +e −x]−1 = [tanh(x/2) + 1]/...

  5. [5]

    com- plete PCE workflow

    Decoding and Post-Processing After optimization, binary portfolio selections are ob- tained by threshold-based decoding of the correlators: xi(θ) := 1 2 h sign ⟨Π(k) i ⟩θ + 1 i .(14) In realistic quantum hardware, the estimation of expec- tation values⟨Π (k) i ⟩θ is fundamentally limited by finite- shot projection noise. For a finite number of measure- me...

    2000

  6. [6]

    These distributions correspond to the data used to com- pute the mean values and standard deviations of the greedy (random) results shown in Fig

    Cost Distributions Figure 7 shows the distributions of normalized cost gaps obtained by applying greedy post-processing to ran- domly initialized bit strings for different problem sizes. These distributions correspond to the data used to com- pute the mean values and standard deviations of the greedy (random) results shown in Fig. 2, where they are summar...

  7. [7]

    These distri- butions are particularly useful for interpreting the behav- ior of the QPU results relative to the greedy (random) baseline in Fig

    Correlator Distributions Figure 8 presents the distributions of Pauli correlators ⟨Π(k) i ⟩θ across different problem sizes, comparing state- vector simulations and QPU experiments. These distri- butions are particularly useful for interpreting the behav- ior of the QPU results relative to the greedy (random) baseline in Fig. 2, since they reveal how much...

  8. [8]

    Peruzzo, J

    A. Peruzzo, J. McClean, P. Shadbolt, M.-H. Yung, X.-Q. Zhou, P. J. Love, A. Aspuru-Guzik, and J. L. O’Brien, Nature Communications5, 4213 (2014)

    2014

Show all 34 references
  1. [9]

    Farhi, J

    E. Farhi, J. Goldstone, and S. Gutmann, A quan- tum approximate optimization algorithm (2014), arXiv:1411.4028 [quant-ph]

    2014 arXiv

  2. [10]

    Biamonte, P

    J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, Nature549, 195 (2017)

    2017

  3. [11]

    Mitarai, M

    K. Mitarai, M. Negoro, M. Kitagawa, and K. Fujii, Phys. Rev. A98, 032309 (2018)

    2018

  4. [12]

    Cerezo, A

    M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii, J. R. McClean, K. Mitarai, X. Yuan, L. Cincio, and P. J. Coles, Nature Reviews Physics3, 625 (2021)

    2021

  5. [13]

    Hadfield, Z

    S. Hadfield, Z. Wang, B. O’Gorman, E. G. Rief- fel, D. Venturelli, and R. Biswas, Algorithms12, 10.3390/a12020034 (2019)

    2019 doi

  6. [14]

    Tan, M.-A

    B. Tan, M.-A. Lemonde, S. Thanasilp, J. Tangpanitanon, and D. G. Angelakis, Quantum5, 454 (2021)

    2021

  7. [15]

    A. Glos, A. Krawiec, and Z. Zimbor´ as, npj Quantum In- formation8, 39 (2022)

    2022

  8. [16]

    Teramoto, R

    K. Teramoto, R. Raymond, E. Wakakuwa, and H. Imai, Quantum-relaxation based optimization algo- rithms: Theoretical extensions (2023), arXiv:2302.09481 [quant-ph]

    2023 arXiv

  9. [17]

    Z. He, R. Raymond, R. Shaydulin, and M. Pistoia, Sci- entific Reports15, 29191 (2025)

    2025

  10. [18]

    Sharma, Y

    M. Sharma, Y. Jin, H. C. Lau, and R. Raymond, in2024 IEEE International Conference on Quantum Computing and Engineering (QCE), Vol. 01 (2024) pp. 692–698

    2024

  11. [19]

    Matsuyama, W.-h

    H. Matsuyama, W.-h. Huang, K. Nishimura, and Y. Ya- mashiro, in2024 IEEE International Conference on Quantum Computing and Engineering (QCE), Vol. 01 (2024) pp. 470–480

    2024

  12. [20]

    Sciorilli, L

    M. Sciorilli, L. Borges, T. L. Patti, D. Garc ´ ıa-Mart ´ ın, G. Camilo, A. Anandkumar, and L. Aolita, Nature Com- munications16, 476 (2025)

    2025

  13. [21]

    Sciorilli, G

    M. Sciorilli, G. Camilo, T. O. Maciel, and A. Canabarro, A competitive nisq and qubit-efficient solver for the labs problem (2025), arXiv:2506.17391 [quant-ph]

    2025

  14. [22]

    Sharma and H

    M. Sharma and H. C. Lau, A comparative study of quantum optimization techniques for solving com- binatorial optimization benchmark problems (2025), arXiv:2503.12121 [quant-ph]

    2025 arXiv

  15. [23]

    V. P. Soloviev and M. Krompiec, Sci. Rep. 10.1038/s41598-026-54244-2 (2026), in press

    2026 doi

  16. [24]

    Pad ´ ın-Mart ´ ınez, V

    J. Pad ´ ın-Mart ´ ınez, V. P. Soloviev, A. Borrallo- Rentero, A. Rodr ´ ıguez-Otero, R. Alfonso-Rodr ´ ıguez, and M. Krompiec, Pauli correlation encoding for budget- constrained optimization (2026), arXiv:2602.17479 [quant-ph]

    2026 arXiv

  17. [25]

    Gurobi Optimization, LLC, Gurobi optimizer reference manual,https://www.gurobi.com(2026)

    2026

  18. [26]

    Albadi and E

    M. Albadi and E. El-Saadany, Electric Power Systems Research78, 1989 (2008)

    1989

  19. [27]

    Palensky and D

    P. Palensky and D. Dietrich, IEEE Transactions on In- dustrial Informatics7, 381 (2011)

    2011

  20. [28]

    Tsurumi and K

    T. Tsurumi and K. Syoji,A design method for elec- tric power demand portfolio and its basic investigation – Proposition of a method for a resource aggregator to procure negawatt effectively –, Tech. Rep. C18005 (Cen- tral Research Institute of Electric Power Industry, Tokyo, J...

    2019

  21. [29]

    R. Faia, T. Pinto, Z. Vale, and J. M. Corchado, Interna- tional Journal of Electrical Power and Energy Systems 129, 106739 (2021)

    2021

  22. [30]

    Ajagekar and F

    A. Ajagekar and F. You, Energy179, 76 (2019)

    2019

  23. [31]

    Yoshioka, K

    T. Yoshioka, K. Sasada, Y. Nakano, and K. Fujii, in2024 IEEE International Conference on Quantum Computing and Engineering (QCE), Vol. 01 (2024) pp. 469–475

    2024

  24. [32]

    Sustainable open Innovation Initiative, Energy manage- ment system,https://www.ems-opendata.jp(2026)

    2026

  25. [33]

    Suzuki, Y

    Y. Suzuki, Y. Kawase, Y. Masumura, Y. Hiraga, M. Nakadai, J. Chen, K. M. Nakanishi, K. Mitarai, R. Imai, S. Tamiya, T. Yamamoto, T. Yan, T. Kawakubo, Y. O. Nakagawa, Y. Ibe, Y. Zhang, H. Yamashita, H. Yoshimura, A. Hayashi, and K. Fujii, Quantum5, 559 (2021)

    2021

  26. [34]

    Amazon Web Services, Amazon braket documentation, https://docs.aws.amazon.com/braket/(2026)

    2026

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.