Pith. sign in

REVIEW 3 major objections 22 references

A divide-and-conquer QAOA that partitions satellite co-observation graphs into NISQ-sized subgraphs can design large constellations with coverage competitive with classical solvers where full-graph QAOA cannot run.

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 · grok-4.5

2026-07-10 13:02 UTC pith:D2XN25IK

load-bearing objection Useful applied DC-QAOA package for large satellite MCLP that empirically works under NISQ limits, with an overstated linearization claim and thin classical baselines. the 3 major comments →

arxiv 2607.08102 v1 pith:D2XN25IK submitted 2026-07-09 quant-ph cs.SYeess.SY

Decomposition-Based QAOA for Maximum Coverage Location Problem in Satellite Constellation Design

Divya Sisodiya , Amiratabak Bahengam , Hang Woon Lee , Hao Chen This is my paper
classification quant-ph cs.SYeess.SY
keywords divide-and-conquer algorithmsgraph partitioningquantum approximate optimization algorithmmaximum coverage location problemquadratic unconstrained binary optimizationsatellite constellation designQAOANISQ
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.

Satellite constellation design can be cast as a maximal covering location problem: choose a fixed number of orbital slots to maximize ground-target coverage over time. That problem is NP-hard and grows too large for standard quantum approximate optimization algorithms on near-term hardware. This paper shows that building a co-observation graph from visibility data, cutting it into small balanced subgraphs by spectral bisection, solving each subgraph with a tailored QAOA circuit, and then merging the pieces with quantum reconstruction (especially graph-state reconstruction) yields feasible constellations for hundreds of slots. The resulting coverage ratios stay competitive with an exact classical optimizer while using only a few dozen qubits per subproblem and finishing faster on the harder instances. The practical point is that quantum optimization becomes usable for real constellation sizing under current qubit limits.

Core claim

Large MCLP instances that arise in satellite constellation design can be solved at practical scale by classical spectral decomposition of a co-observation graph into subgraphs small enough for NISQ QAOA, followed by either quantum-state or graph-state reconstruction of the separator slots; the graph-state merge consistently recovers coverage ratios of roughly 80–99 percent of the classical optimum while remaining tractable where full-graph QAOA cannot be simulated.

What carries the argument

The co-observation graph (vertices = orbital slots, edge weights = shared target-time visibility) together with spectral bisection into qubit-bounded subgraphs and the two Hamiltonian-based merge operators QSR (replicated separator qubits plus post-selection) and GSR (single-copy coverage-weighted separator QUBO).

Load-bearing premise

The equal-share linearization of the exact multi-body coverage product, plus local re-optimization only on separator slots, still leaves the recovered coverage close enough to the true MCLP optimum for design use.

What would settle it

Take any visibility matrix from the paper, solve the identical MCLP exactly with a classical MIP solver and also with the full decomposition pipeline (same q_max, same GSR or QSR), and check whether the coverage ratio falls below the reported range (especially for sparse, low-density matrices or larger satellite budgets).

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

If this is right

  • Constellation sizing problems with several hundred orbital slots become executable on present-day quantum processors without requiring full-graph qubit counts.
  • Graph-state reconstruction is the preferred merge: it uses half the qubits of QSR and yields higher, more stable coverage ratios across the tested visibility matrices.
  • Runtime stays bounded by the fixed subgraph size rather than by total slots, so wall-clock time grows roughly with the number of subgraphs rather than exponentially.
  • The same co-observation-graph-plus-decomposition pattern can be reused for other coverage-cardinality combinatorial problems once the Hamiltonian is rewritten for their reward structure.

Where Pith is reading between the lines

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

  • If the linearization error remains negligible under realistic multi-target, multi-plane visibility, the method could be extended to reconfiguration and transfer problems that share the same cardinality-plus-coverage structure.
  • Hardware noise primarily degrades the approximation ratio by roughly ten percentage points on a 12-qubit Heron device; shallow depth (p=3) already limits decoherence, so modest error mitigation would likely close most of the remaining gap.
  • The approach may transfer to other facility-location or sensor-placement settings where demand nodes couple only weakly across distant facilities.

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 / 0 minor

Summary. The paper formulates satellite constellation design as an MCLP, maps it to a co-observation graph, partitions that graph by recursive spectral bisection into NISQ-sized subgraphs, solves each subgraph with a QAOA circuit on a linearized QUBO Hamiltonian, and reconstructs a global constellation via two quantum merge strategies (QSR and GSR). Across six visibility-matrix instances (J up to 958) and budgets N∈{2,4,6}, GSR-based decomposition reports coverage ratios often in the 80–99% range versus Gurobi optima, with runtimes competitive or better on harder instances, while full-graph QAOA is reported as infeasible under classical simulation. Hardware runs on IBM ibm_kingston (q_max=12, N=4) are included with runtime confidence intervals.

Significance. Applying divide-and-conquer QAOA to coverage-type MCLP (rather than Max-Cut-style graph partitioning) is a genuine and useful extension; the co-observation graph, proportional budget allocation, and especially the GSR merge Hamiltonian are problem-specific contributions that fit satellite design under NISQ constraints. The systematic Table V comparison (coverage, ρ, Δ, runtime) across VM-1–6, N, and q_max, plus real-hardware validation with CIs, is a clear empirical strength. If the approximation and reconstruction gaps are better quantified, the framework would be a solid methods contribution for quantum combinatorial optimization in aerospace applications.

major comments (3)
  1. §V-A4 and Table VI: the linearization error analysis is internally inconsistent and overstates negligibility. Eq. (23) correctly bounds the absolute pointwise error by ½−(½)^|C|≈½ for |C|∈[81,212]. The text then claims this is “negligible in relative terms (<0.001%)” and Table VI lists “Error Bound Magnitude <0.001%,” which does not follow from Eq. (23). With π_tp∈{1,2} the Hamiltonian (Eqs. 20–26) optimizes a soft fractional-coverage surrogate, not the original indicator MCLP. The reported ρ values in Table V are true-coverage evaluations of bit-strings found under that surrogate; without a quantified surrogate-to-true gap (or ranking-preservation argument) for general visibility structure, the central claim that the method solves MCLP competitively rests on unanalyzed approximation error. Please correct the bound interpretation, report true vs. surrogate objective values, and discuss w
  2. Table V and §V-A: the “standard QAOA” baseline is classical AerSimulator timeout / memory exhaustion on 2^J statevectors (J=288–958), not a quantum-hardware or even a fair algorithmic baseline. Labeling these rows “Timeout†” and using them to claim superiority of decomposition is misleading; full-graph QAOA was never executable on the same classical or quantum resources. Either restrict the baseline to instances where non-decomposed QAOA can actually run (or be classically simulated), or replace it with strong classical heuristics / MILP warm-starts so that the comparison isolates the value of decomposition rather than simulation limits.
  3. §III-E (β=50 via grid search over {10,50,100,500}) and §III-F (λ=100): penalty coefficients and other free parameters are selected on the same test instances used for reporting. This is mild but load-bearing hyperparameter circularity for an empirical methods paper. Provide a sensitivity study (or hold-out configurations) showing that coverage ratios and feasibility remain stable under modest β/λ variation, and state clearly that β, λ, α, p, p_merge, and q_max are free parameters of the method.

Circularity Check

1 steps flagged

Mild hyperparameter tuning of penalty coefficients on the same test instances used for reporting; no structural circularity in the MCLP-to-QUBO mapping, decomposition, or external Gurobi evaluation.

specific steps
  1. fitted input called prediction [Section III-E (QUBO Matrix Construction) and III-F (QSR/GSR)]
    "The cardinality penalty coefficient is set to β=50 to balance constraint satisfaction with solution quality. This value is selected via grid search over β∈[10,50,100,500] to minimize constraint violations while maintaining coverage performance across all test instances. … The agreement penalty coefficient is set to λ=100 …"

    Penalty weights that appear inside the Hamiltonians optimized by QAOA are chosen by grid search that explicitly references performance on the same visibility-matrix instances later reported in Table V. The reported coverage ratios therefore reflect a post-hoc tuned surrogate rather than a parameter-free run; the effect is mild optimism bias, not an algebraic identity between input and output.

full rationale

This is an empirical algorithmic methods paper whose central claims are computational (scalability of spectral-bisection + MCLP-QAOA + QSR/GSR under q_max ≤ 20, coverage ratios vs Gurobi on six visibility matrices). The derivation chain (co-observation graph from Φ, spectral min-cut, linearized QUBO Hamiltonian, independent QAOA subproblems, separator reconstruction) is constructive and does not redefine its target quantity as an input. Coverage is always scored with the original indicator (Eq. 43) against an independent classical MILP optimum; the linearization (Eq. 22) is an explicit approximation whose error is bounded (Eq. 23), not a self-definition. The only mild circularity is ordinary hyperparameter selection: eta and λ are chosen by grid search that includes the reported instances. That introduces a small optimism bias common to all tuned solvers and does not force the coverage ratios by construction. No self-citation is load-bearing for uniqueness or ansatz; no fitted constant is renamed a prediction. Score remains 1.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 2 invented entities

The central performance claim rests on standard spectral-graph and QAOA machinery plus several paper-specific modeling choices and hand-tuned penalties. Free parameters (β, λ, α, p, q_max) directly affect feasibility and coverage. Domain assumptions include that co-observation weights and proportional budget allocation preserve MCLP structure under recursive cuts, and that the linearized coverage term is an acceptable stand-in for the exact product. Invented methodological pieces are the co-observation graph encoding for this MCLP and the GSR merge Hamiltonian; neither has independent external validation beyond this study.

free parameters (5)
  • cardinality penalty β = 50
    Set to 50 after grid search over {10,50,100,500} to balance constraint satisfaction and coverage on the test instances; enters QUBO diagonals/off-diagonals and GSR merge.
  • QSR agreement penalty λ = 100
    Set to 100 to force agreement between duplicated separator qubits; controls post-selection feasibility and merge energy landscape.
  • GSR normalization α = 1
    Set to 1 when building GSR QUBO from coverage scores σ_i and co-observation weights; scales relative strength of coverage bias vs coupling.
  • QAOA depth p and merge depth p_merge = 3
    Both fixed at 3 layers for subproblems and merges; depth is a free algorithmic choice that trades expressivity for noise/runtime.
  • max subgraph size q_max = {8,12,20}
    Swept in {8,12,20} as a hardware-facing free design parameter that governs decomposition depth and solution quality/runtime trade-off.
axioms (5)
  • domain assumption Spectral bisection via the Fiedler vector yields balanced partitions with sufficiently small cut weight for co-observation graphs of orbital slots.
    Invoked in Section III-C as the classical decomposition engine; quality of global MCLP solutions depends on this approximation of discrete min-cut.
  • ad hoc to paper Linearizing the multi-body coverage product by equal reward sharing (Eq. 22) preserves ranking of good constellations well enough for reported optimality gaps.
    Introduced to avoid exponential multi-body terms; error bound (23) approaches 1/2 for large |C|, which the paper treats as relatively negligible.
  • ad hoc to paper Proportional satellite budget allocation across subgraphs (Eqs. 12–14) plus separator residual budgets preserves global cardinality without large optimality loss.
    Section III-D; enables independent subproblem solves but is not proven optimal for MCLP coverage objectives.
  • domain assumption QAOA with shallow depth and COBYLA-trained parameters yields useful approximate solutions on the constructed QUBOs under ideal simulation.
    Standard hybrid-quantum assumption used throughout Sections III-E and IV; hardware section shows noise reduces approximation ratio.
  • domain assumption Co-observation frequency weights w_ij = sum Φ_tip Φ_tjp adequately encode the coupling structure needed for coverage-preserving graph cuts.
    Section III-B; the entire divide-and-conquer strategy is only as good as this graph model of MCLP interactions.
invented entities (2)
  • Graph State Reconstruction (GSR) merge Hamiltonian no independent evidence
    purpose: Merge separator decisions with one qubit per separator node using coverage scores, co-observation couplings, and a local cardinality term, avoiding QSR’s 2|S| qubit overhead.
    Presented as a problem-aware alternative to QSR for MCLP; performance claims for GSR superiority are internal to this paper’s instances.
  • Co-observation graph for satellite MCLP no independent evidence
    purpose: Map orbital slots to vertices and simultaneous visibility counts to edge weights so spectral partitioning can split large constellation instances.
    Standard graph modeling idea specialized to visibility tensors; not a physical entity, but a paper-specific intermediate object the pipeline depends on.

pith-pipeline@v1.1.0-grok45 · 31232 in / 4064 out tokens · 50832 ms · 2026-07-10T13:02:20.164189+00:00 · methodology

0 comments
read the original abstract

An increase in earth observation missions has increased the demand of efficient design and optimization of satellite constellations. Maximizing coverage of the target while effectively utilizing the limited orbital resources is one of the critical design challenges for complex combinatorial optimization problems. The maximal covering location problem (MCLP), serves as a base for orbital coverage modeling, is NP-hard and computationally intractable for large-constellation instances. Using heuristics, metaheuristics, and mixed-integer linear programming, classical solvers have achieved optimal or near-optimal results, yet their scalability is limited as the problem size increases. Quantum computing advancements, including the quantum approximate optimization algorithms, offer a potential solution to NP-hard combinatorial optimization problems. Current quantum hardware limitations, such as low qubit counts and circuit depth, restrict solutions for small-scale instance problems. To address this challenge, this paper proposes a scalable quantum optimization framework for MCLP in satellite constellation design. A decomposition-based quantum methodology is proposed, in which large MCLP instances are partitioned into subgraphs by classical decomposition, optimized independently via quantum optimization circuits, and combined using quantum reconstruction strategies. Computational results across different constellation sizes reveal better scalability in less time while maintaining competitive coverage performance compared to classical solvers.

Figures

Figures reproduced from arXiv: 2607.08102 by Amiratabak Bahengam, Divya Sisodiya, Hang Woon Lee, Hao Chen.

Figure 1
Figure 1. Figure 1: FIGURE 1: Decomposition-Based QAOA Framework [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: illustrates an MCLP-QAOA circuit for a subgraph Gℓ with n qubits and p layers. Each qubit qj represents one orbital slot and xj ∈ {0, 1} indicates whether it is selected. H Gates prepare the uniform superposition |+⟩ ⊗n. Each layer r applies single-qubit Rz(2γrQjj ) gates encoding the coverage reward and cardinality penalty (diagonal QUBO entries Qjj ), two-qubit Rzz(2γrβ) gates on all n 2  qubit pairs en… view at source ↗
Figure 3
Figure 3. Figure 3: shows a QSR circuit with 2|S| qubits and pmerge layers. s (1) i and s (2) i are two copies of separator qubits rep￾resenting S adjacent subgraphs. Each layer pmerge applies the problem unitary e −iγ′ rHQSR and mixer unitary e −iβ′ r P q Xq . The first term P i∈S λ · Zi1Zi2 HQSR enforces agreement between copies via cross-copy CNOT gates, while the sec￾ond term preserves co-observation structure via Rzz(2γ … view at source ↗
Figure 4
Figure 4. Figure 4: shows GSR circuit with |S| qubits and pmerge layers. Each qubit sj represents one separator qubit, requiring half the qubits of QSR with no post-selection. Each layer of r ap￾plies the problem unitary e −iγ′′ r HGSR and mixer e −iβ′′ r P j Xj . Single-qubit Rz(2γ ′′ r Qii) gates encode coverage bias and cardinality penalty, while two-qubit Rzz(2γ ′′ r Qij ) gates en￾code co-observation coupling and cardina… view at source ↗
Figure 5
Figure 5. Figure 5: FIGURE 5: Satellite constellation geometry showing satel [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: The reference visibility profile vt indicates time steps at which the reference ground station is visible to at least one satellite. The orbital slot selection variable xj identifies the four selected orbital slots. The resulting coverage time￾line bt confirms near-continuous coverage across the repeat cycle, with isolated gaps at time steps where no selected 14 [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIGURE 6: Selected orbital slots in RAAN-argument-of [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIGURE 8: Coverage and runtime across VM-1 to VM-6 for [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIGURE 9: Coverage and runtime across VM-1 to VM-6 for [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIGURE 10: Coverage and runtime across VM-1 to VM-6 [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIGURE 11: Coverage (%) versus Maximum Subgraph Size [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FIGURE 12: Approximation ratio [PITH_FULL_IMAGE:figures/full_fig_p018_12.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

22 extracted references · 22 canonical work pages · 3 internal anchors

  1. [1]

    Optimal Satellite Constellation Config- uration Design:A Collection of Mixed Integer Linear Programs,

    D. O. W. Rogers, D. Won, D. Koh, K. Hong, and H. W. Lee, “Optimal Satellite Constellation Config- uration Design:A Collection of Mixed Integer Linear Programs,”Journal of Spacecraft and Rockets, Mar. 2026

    work page 2026

  2. [2]

    THE MAXIMAL COV- ERING LOCATION PROBLEM,

    R. Church and C. R. Velle, “THE MAXIMAL COV- ERING LOCATION PROBLEM,” en,Papers in Re- gional Science, vol. 32, no. 1, pp. 101–118, Jan. 1974

    work page 1974

  3. [3]

    Optimal constellation design based on satellite ground tracks for Earth observation missions,

    Y . Imoto, S. Satoh, T. Obata, and K. Yamada, “Optimal constellation design based on satellite ground tracks for Earth observation missions,” en,Acta Astronau- tica, vol. 207, pp. 1–9, Jun. 2023

    work page 2023

  4. [4]

    Satellite Constellation Design for Minimum Worst-Case Revisit Time,

    J. Ko, B. Gwon, and J. Ahn, “Satellite Constellation Design for Minimum Worst-Case Revisit Time,” en, International Journal of Aeronautical and Space Sci- ences, vol. 27, no. 1, pp. 916–929, Jan. 2026

    work page 2026

  5. [5]

    Regional Constellation Re- configuration Problem: Integer Linear Programming Formulation and Lagrangian Heuristic Method,

    H. W. Lee and K. Ho, “Regional Constellation Re- configuration Problem: Integer Linear Programming Formulation and Lagrangian Heuristic Method,” en, Journal of Spacecraft and Rockets, vol. 60, no. 6, pp. 1828–1845, Nov. 2023

    work page 2023

  6. [6]

    A decomposition approach for the prob- abilistic maximal covering location-allocation prob- lem,

    F. D. A. Corrêa, L. A. N. Lorena, and G. M. Ribeiro, “A decomposition approach for the prob- abilistic maximal covering location-allocation prob- lem,” en,Computers & Operations Research, vol. 36, no. 10, pp. 2729–2739, Oct. 2009

    work page 2009

  7. [7]

    A Review on Quantum Machine Learning in Applied Systems and Engineering,

    Y .-Y . Hong and D. Josh Domingo Lopez, “A Review on Quantum Machine Learning in Applied Systems and Engineering,”IEEE Access, vol. 13, pp. 144 607– 144 631, 2025

    work page 2025

  8. [8]

    A Quantum Approximate Optimization Algorithm

    E. Farhi, J. Goldstone, and S. Gutmann,A Quantum Approximate Optimization Algorithm, arXiv:1411.4028 [quant-ph], Nov. 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  9. [9]

    Using quantum comput- ing to solve the maximal covering location problem,

    A. Giraldo-Quintero, J. G. Lalinde-Pulido, J. C. Duque, and D. Sierra-Sosa, “Using quantum comput- ing to solve the maximal covering location problem,” en,Computational Urban Science, vol. 2, no. 1, p. 43, Dec. 2022

    work page 2022

  10. [10]

    Two-Step QAOA: Enhancing Quantum Optimization by Decomposing K-hot Constraints in QUBO Formulations

    Y . Minato,Two-Step QAOA: Enhancing Quantum Optimization by Decomposing K-hot Constraints in QUBO F ormulations, arXiv:2408.05383 [quant-ph], Feb. 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  11. [11]

    Quantum Optimization for Closed-Loop Scheduling of Earth Observation Satellite Formation,

    V . Marchioli, M. Boggio, D. V olpe, L. Massotti, and C. Novara, “Quantum Optimization for Closed-Loop Scheduling of Earth Observation Satellite Formation,” en,SN Computer Science, vol. 6, no. 6, p. 739, Aug. 2025

    work page 2025

  12. [12]

    COMPARATIVE STUDY OF TWO DIVIDE AND CONQUER SORT- ING ALGORITHMS: QUICKSORT AND MERGE- SORT,

    O. Esau Taiwo, A. O. Christianah, A. N. Oluwatobi, K. A. Aderonke, and A. J. Kehinde, “COMPARATIVE STUDY OF TWO DIVIDE AND CONQUER SORT- ING ALGORITHMS: QUICKSORT AND MERGE- SORT,” en,Procedia Computer Science, vol. 171, pp. 2532–2540, 2020

    work page 2020

  13. [13]

    Large-Scale Quantum Approximate Optimization via Divide-and-Conquer,

    J. Li, M. Alam, and S. Ghosh, “Large-Scale Quantum Approximate Optimization via Divide-and-Conquer,” IEEE Transactions on Computer-Aided Design of In- tegrated Circuits and Systems, vol. 42, no. 6, pp. 1852– 1860, Jun. 2023

    work page 2023

  14. [14]

    QAOA-in- QAOA: Solving Large-Scale MaxCut Problems on 20 Small Quantum Machines,

    Z. Zhou, Y . Du, X. Tian, and D. Tao, “QAOA-in- QAOA: Solving Large-Scale MaxCut Problems on 20 Small Quantum Machines,” en,Physical Review Ap- plied, vol. 19, no. 2, p. 024 027, Feb. 2023

    work page 2023

  15. [15]

    Graph decomposition techniques for solving combinatorial optimization problems with variational quantum algo- rithms,

    M. Ponce, R. Herrman, P. C. Lotshaw, et al., “Graph decomposition techniques for solving combinatorial optimization problems with variational quantum algo- rithms,” en,Quantum Information Processing, vol. 24, no. 2, p. 60, Feb. 2025

    work page 2025

  16. [16]

    A Decomposition Method for the Hybrid Quantum-Classical Solution of the Number Partitioning Problem

    Z. Li, T. Seidel, D. Leib, M. Bortz, and R. Heese, “Ef- ficient solution of the number partitioning problem on a quantum annealer: A hybrid quantum-classical de- composition approach,”Journal of Heuristics, vol. 31, no. 2, p. 21, Apr. 2025, arXiv:2312.08940 [math]

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  17. [17]

    TensorHyper-VQC: A tensor-train-guided hypernet- work for robust and scalable variational quantum com- puting,

    J. Qi, C.-H. Huck Yang, P.-Y . Chen, and M.-H. Hsieh, “TensorHyper-VQC: A tensor-train-guided hypernet- work for robust and scalable variational quantum com- puting,” vol. 12, Feb. 2026

    work page 2026

  18. [18]

    Algebraic connectivity of graphs,

    M. Fiedler, “Algebraic connectivity of graphs,” en, Czechoslovak Mathematical Journal, vol. 23, no. 2, pp. 298–305, 1973

    work page 1973

  19. [19]

    A tutorial on spectral clustering,

    U. V on Luxburg, “A tutorial on spectral clustering,” en,Statistics and Computing, vol. 17, no. 4, pp. 395– 416, Dec. 2007

    work page 2007

  20. [20]

    Spectral par- titioning works: Planar graphs and finite element meshes,

    D. A. Spielman and Shang-Hua Teng, “Spectral par- titioning works: Planar graphs and finite element meshes,” inProceedings of 37th Conference on F oun- dations of Computer Science, Burlington, VT, USA: IEEE, 1996, pp. 96–105

    work page 1996

  21. [21]

    Ising formulations of many NP problems,

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

    work page 2014

  22. [22]

    From the Quantum Approximate Optimization Algorithm to a Quantum Alternating Operator Ansatz,

    S. Hadfield, Z. Wang, B. O’Gorman, et al., “From the Quantum Approximate Optimization Algorithm to a Quantum Alternating Operator Ansatz,” en,Algo- rithms, vol. 12, no. 2, p. 34, Feb. 2019. 21

    work page 2019