{"id":"1af16289-b19e-4a2f-9ecd-d7f4dcf389b6","arxiv_id":"2607.09893","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":5,"one_line_summary":"For small discrete MRFs with classically enumerable 2^n probabilities, amplitude-encoded i.i.d. sampling shows no wall-clock advantage over inverse-CDF and only modest ESS gains over tuned classical MCMC.","lead":"This paper benchmarks amplitude-encoded quantum sampling of small discrete Markov random fields against classical MCMC and exact inverse-CDF methods. It finds modern classical samplers nearly close the ESS gap and that amortized wall-clock favors classical inverse-CDF once the full distribution is enumerated.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified","rationale":"The strongest claim is numerical and protocol-bound, not a speedup assertion. The manuscript already isolates the i.i.d. structural property (τ≈1), reports the full monotone hierarchy of classical baselines, amortizes preprocessing on both sides, and supplies executable artifacts. The only soft spot the reader flags—backend-specific wall-clock—is disclosed by the authors and does not reverse the relative ESS or amortized-rate conclusions. Because the concern is already internalized and does not threaten the stated claims, no verdict adjustment is warranted. ACCEPT with high confidence remains appropriate for a reproducible negative-result case study.","tokens_in":13783,"tokens_out":494,"duration_ms":4352,"concrete_test":"Independently re-run the 60-instance ESS protocol from the committed JSON (paper_table_metrics.json + parallel_tempering_baseline.json) with the same Geyer IPS estimator on Hamming-weight series; if mean Quantum/baseline ratios shift by more than ~10% or the inverse-CDF amortized ordering reverses, the headline numbers would need revision. Otherwise the claim stands.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper’s central claim is a carefully scoped empirical characterization: under a fixed protocol (60 instances, five graph families, 1k burn-in, 3k retained samples), amplitude-encoded i.i.d. sampling yields mean ESS ratios of 16.35 / 7.29 / 1.82 / 1.79 versus single-site, block, tuned-block Gibbs and parallel tempering, and after amortizing the shared O(2^n) preprocessing, exact inverse-CDF dominates wall-clock ESS/s (36\times mean rate, 153× per-instance). The authors explicitly disclaim quantum advantage, flag that absolute ESS/s figures are backend-specific (local statevector vs. queue-latency cloud ~714 ESS/s; Discussion §V-A, Table II caption), and ship open artifacts that regenerate every table. The reader’s weakest-assumption note about simulator proxy is already acknowledged by the paper and does not undercut the relative ESS hierarchy or the amortized inverse-CDF comparison, both of which rest on the same classical enumeration of Pθ. No load-bearing derivation error, circular construction, or unacknowledged scope violation is present.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper studies amplitude-encoded i.i.d. sampling for small discrete MRFs in the regime where the full 2^n distribution is classically enumerable. After classical precomputation of P_\theta, Qiskit StatePreparation is used to prepare |ψ⟩ = ∑_x √P_\theta(x)|x⟩ and samples are obtained by measurement. Across 60 synthetic instances on five graph families (1k burn-in, 3k retained samples), mean Quantum/classical ESS ratios are 16.35 (single-site Gibbs), 7.29 (block Gibbs), 1.82 (tuned-block), and 1.79 (parallel tempering). Amortizing the shared O(2^n) preprocessing, exact inverse-CDF sampling reaches ~17.7M ESS/s versus ~488k ESS/s for the quantum sampler (36× mean rate, 153× per-instance), so there is no wall-clock advantage. Secondary results include amplitude-encoding verification at n=8,10,12 (F≈1), an MPS scaling study to n=40 (F=0.721±0.059 at χ=32), and a matched-budget VQC-vs-MPS comparison in which shallow hardware-efficient VQCs underperform MPS at every tested size. The authors explicitly disclaim quantum advantage and release open code and artifacts.","tokens_in":14084,"tokens_out":1253,"duration_ms":9161,"significance":"If the reported ESS hierarchy and amortized wall-clock comparison hold, the paper supplies a carefully scoped, reproducible benchmark that quantifies how much of the apparent i.i.d. sampling advantage of amplitude encoding is closed by modern classical MCMC (tuned-block Gibbs and parallel tempering) and by exact inverse-CDF once enumeration is free. The multi-seed MPS ceiling at n=40 and the negative VQC-vs-MPS and mean-field-vs-VQC results are useful reference points for future variational and quantum sampling work. Strengths include open artifacts that regenerate every table, explicit negative results, and a clear statement of the enumerable-regime scope. The contribution is empirical characterization rather than a new algorithm or asymptotic claim, but that characterization is load-bearing for how the community interprets small-n quantum sampling comparisons.","major_comments":[{"comment":"Table I and §III-D / §IV-A: ESS ratios are single-chain point estimates from Hamming-weight series of length 3,000 (Geyer IPS). The paper correctly flags that ratios should be read at the distributional (mean/range) level, but the headline means 16.35 / 7.29 / 1.82 / 1.79 are still the central quantitative claim. A short multi-chain or bootstrap uncertainty on the per-instance ESS (or at least on the family-level means) would make the hierarchy more robust without changing the experimental design.","section":null},{"comment":"Table II and Discussion §V-A: Absolute ESS/s figures for the quantum row use local statevector sampling of StatePreparation; the paper itself notes that BlueQubit cloud execution is queue-latency bound (~714 ESS/s) and that timings are backend-specific. The amortized inverse-CDF comparison remains fair because both sides share the O(2^n) enumeration, but the manuscript should state more prominently (e.g., in the abstract or Table II caption) that the quantum ESS/s is a simulator proxy and not a claim about hardware or circuit-depth cost of amplitude encoding.","section":null}],"minor_comments":[{"comment":"Abstract vs. Table IV: abstract rounds VQC/MPS fidelities to (0.31, 0.99), (0.21, 0.96), (0.17, 0.88); body reports (0.306, 0.990), (0.210, 0.958), (0.165, 0.878). Align rounding or cite the table.","section":null},{"comment":"§IV-C and §IV-E: The two VQC fidelity series (fixed-budget Table IV vs. unconstrained Table VI) answer different questions; the explicit flag is good, but a single sentence in the abstract or introduction would prevent readers from treating them as interchangeable.","section":null},{"comment":"Fig. 2: R² ≈ 0.00036 for ESS ratio vs. mixing-difficulty proxy is effectively null; the caption could state more directly that topology/family effects dominate any single spectral-gap predictor.","section":null},{"comment":"Notation: H_\theta is introduced as a diagonal Hamiltonian (Eq. 2) but is used only to obtain the classical diagonal of unnormalized probabilities; a brief remark that no quantum Hamiltonian simulation is performed would reduce possible confusion with QCGM.","section":null},{"comment":"Related work: the distinction from Piatkowski & Zoufal (QCGM) is clear; a one-sentence pointer to other quantum-enhanced MCMC (Layden et al., Ferguson & Wallden) already present could be tightened to emphasize that those works target non-enumerable regimes.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The manuscript is a solid, reproducible case study rather than a high-novelty algorithmic contribution. Fit for a methods/benchmarks venue or a quantum-ML special issue is good; for a top-tier theory journal the empirical scope may be borderline. The open artifact and honest negative results are genuine strengths and should weigh in favor of acceptance after minor revision."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"This is a carefully scoped empirical package, not a speedup claim. The author deliberately stays inside the enumerable regime (n=8–12 for the main comparisons) so that every circuit sample is independent (τ≈1) and can be compared cleanly to classical MCMC under a fixed protocol: 60 instances, five graph families, 1k burn-in, 3k retained samples. The headline numbers are the ESS ratios—16.35 / 7.29 / 1.82 / 1.79 versus single-site, block, tuned-block Gibbs and parallel tempering—and the amortized wall-clock result that exact inverse-CDF reaches ~17.7 M ESS/s against ~488 k for the quantum sampler. Those measurements, plus the multi-seed MPS curve to n=40 and the matched-budget VQC-vs-MPS negative result, are the actual new content.\n\nWhat the paper does well is honesty and reproducibility. Negative results are stated without hedging (shallow VQC loses to MPS at every tested size; mean-field beats VQC on global fidelity at n=8,10). Scope is explicit: no QCGM, no hardware noise, no exponential advantage. Artifacts regenerate every table. The citation pattern is appropriate; the math is standard and the ESS protocol is conventional.\n\nSoft spots are real but secondary. ESS ratios are single-chain point estimates, so treat them distributionally. Absolute ESS/s figures are backend-specific (local statevector vs. queue-latency cloud); the paper itself flags this in the discussion and Table II caption, so it does not undercut the relative hierarchy or the amortized inverse-CDF comparison, both of which rest on the same classical Pθ. The VQC depth and iteration budgets are modest; that is already framed as a negative result for shallow hardware-efficient ansätze rather than a claim of optimality.\n\nThis is for people who work on quantum sampling, Born machines, or MCMC for graphical models and want a clean reference for when amplitude encoding does and does not help. It deserves a serious referee. I would accept it for peer review and would cite the ESS hierarchy and the amortized wall-clock comparison when I need a controlled baseline.","headline":"Honest, reproducible case study that quantifies how much of the ESS edge of amplitude-encoded sampling vanishes against modern classical baselines and shows amortized wall-clock favors inverse-CDF.","tokens_in":14700,"tokens_out":549,"would_cite":true,"duration_ms":5492,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"For small discrete MRFs with fully precomputed probabilities, modern classical samplers nearly erase the ESS edge of amplitude-encoded quantum sampling, and inverse-CDF wins on amortized wall-clock.","keywords":["amplitude encoding","Markov random fields","quantum sampling","Monte Carlo methods","effective sample size","variational quantum circuits","matrix product states","hardware-efficient ansatz"],"falsifier":"Re-run the identical sixty-instance protocol with a non-simulated quantum device (or a different statevector backend) and check whether the amortized Quantum/inverse-CDF ESS-per-second ratio remains near 1/150 and the ESS ratios versus tuned-block and parallel tempering stay near 1.8.","tokens_in":14626,"feed_emoji":"⚛️","tokens_out":1014,"duration_ms":12291,"temperature":0.7,"pith_summary":"This paper asks a precise, scoped question: once the full 2^n target distribution of a small discrete Markov random field is already enumerated classically, how much does the structural independence of amplitude-encoded quantum samples (τ ≈ 1) still buy you against carefully chosen classical MCMC? Across sixty instances on five graph families, the mean effective-sample-size ratios fall from roughly 16\times versus single-site Gibbs to about 1.8\times versus tuned-block Gibbs and parallel tempering. When the shared O(2^n) preprocessing cost is amortized into wall-clock time, exact inverse-CDF sampling delivers tens of millions of ESS per second while the quantum path remains hundreds of thousands, confirming no practical speed advantage in this regime. Secondary measurements show shallow hardware-efficient variational circuits lag matched-budget matrix-product states at every tested size, while an MPS scaling curve reaches fidelity about 0.72 at n = 40 with bond dimension 32. The work therefore supplies a clean autocorrelation benchmark and a reproducible ceiling rather than a quantum-advantage claim.","feed_headline":"Classical samplers nearly erase quantum ESS edge on small MRFs","feed_subtitle":"Once 2^n probabilities are precomputed, inverse-CDF is ~150\times faster on amortized wall-clock; modern MCMC closes most of the gap.","key_machinery":"Amplitude encoding: after classical enumeration of P_θ, prepare the state |ψ⟩ = Σ_x √P_θ(x)|x⟩ with a state-preparation primitive so that each measurement returns an independent sample (τ ≈ 1). This isolates the structural independence property for clean comparison against MCMC autocorrelation.","core_discovery":"In the enumerable regime where the full target distribution can be precomputed, amplitude-encoded i.i.d. quantum sampling retains only a modest ESS advantage over modern classical samplers (mean ratios falling from 16.35 versus single-site Gibbs to 1.79 versus parallel tempering), and that advantage disappears on amortized wall-clock once inverse-CDF sampling is allowed to use the same precomputed distribution.","pith_inferences":["The same amortization logic likely applies to any quantum state-preparation pipeline that still requires classical enumeration of an exponentially large diagonal; the bottleneck is shared, not quantum-specific.","The reported MPS fidelity curve at fixed bond dimension gives a concrete target that any future variational or quantum method at n ≈ 40 must beat to claim compression superiority.","Because family/topology effects dominate a single spectral-gap predictor, graph-aware classical block designs may continue to close residual ESS gaps faster than deeper circuits."],"forward_implications":["When full enumeration is feasible, classical inverse-CDF is the practical default for pure sampling throughput.","Claims of quantum sampling advantage on small MRFs must be tested against tuned-block Gibbs or parallel tempering, not only single-site Gibbs.","Shallow hardware-efficient variational circuits are not recommended for full-distributional MRF sampling at n ≤ 12; MPS supplies a higher classical fidelity ceiling.","Future work that wants advantage must leave the enumerable regime (full QCGM-style constructions, larger n, or coherent downstream use of the prepared state)."],"fun_headline_variants":["Modern classical MCMC closes most of quantum's ESS lead on small MRFs","Quantum retains only 1.8x ESS edge over parallel tempering once 2^n precomputed","Amortized inverse-CDF beats amplitude-encoded quantum 36x on wall-clock ESS rate","Enumerable MRFs: classical samplers erase quantum wall-clock advantage","MPS state prep far outpaces VQC; quantum ESS gains vanish vs tuned MCMC"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The wall-clock and ESS-per-second numbers treat local statevector simulation of the preparation circuit as a fair stand-in for the quantum sampler itself; the paper notes those timings are backend-specific and cloud execution is queue-latency bound.","fun_headline_variants_meta":{"raw":{"variants":["Modern classical MCMC closes most of quantum's ESS lead on small MRFs","Quantum retains only 1.8x ESS edge over parallel tempering once 2^n precomputed","Amortized inverse-CDF beats amplitude-encoded quantum 36x on wall-clock ESS rate","Enumerable MRFs: classical samplers erase quantum wall-clock advantage","MPS state prep far outpaces VQC; quantum ESS gains vanish vs tuned MCMC"]},"model":"grok-4.5","effort":"low","cost_usd":0.006346,"raw_usage":{"total_tokens":1683,"prompt_tokens":938,"num_sources_used":0,"completion_tokens":112,"cost_in_usd_ticks":63460000,"prompt_tokens_details":{"text_tokens":938,"audio_tokens":0,"image_tokens":0,"cached_tokens":0},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":633,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":938,"tokens_out":112,"duration_ms":5627,"temperature":1.0,"reasoning_tokens":633,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-14T14:45:54.532351+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Re-run the identical sixty-instance protocol with a non-simulated quantum device (or a different statevector backend) and check whether the amortized Quantum/inverse-CDF ESS-per-second ratio remains near 1/150 and the ESS ratios versus tuned-block and parallel tempering stay near 1.8.","supporting_citations":[],"review_version":1}