{"id":"4ef48560-200d-46e7-bb82-3d92903864bb","arxiv_id":"2607.08708","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":3,"one_line_summary":"Semiclassical analysis of QAOA on the SK spin glass shows log(p)/p convergence to the Parisi ground state energy, matching classical performance and indicating no quantum advantage.","lead":"This paper uses semiclassical simulations to argue that the quantum approximate optimization algorithm (QAOA) on the Sherrington-Kirkpatrick spin glass converges to the optimal energy at a rate of log(p)/p, matching what a purely classical algorithm can achieve. If correct, it implies no quantum speedup for this benchmark problem and provides a concrete scaling prediction that future experiments can test.","discovery_kind":"unclear","skeptic_critique":{"model":"glm-5.2","headline":"The log(p)/p bound on quantum QAOA rests on an empirical finite-p comparison; no theoretical argument guarantees the semiclassical outperformance persists asymptotically.","rationale":"The reader correctly identifies the extrapolation from finite p to asymptotic p as the load-bearing assumption. I agree this is the central weakness. The paper's physical mechanism for log(p)/p scaling — the competition between initial noise (1/S) and Ehrenfest-time exponential growth (log S) — is clean and well-motivated for the semiclassical system, but it is specific to the classical noisy dynamics. The quantum QAOA at S=1/2 does not have a tunable noise parameter and does not experience the same tradeoff. The only evidence linking the quantum scaling to the semiclassical scaling is the finite-p numerical comparison showing semiclassical slightly outperforms quantum. This is a reasonable empirical observation but not a bound. The paper is honest about this ('We have no reason to believe...'), which is appropriate, but it means the central claim is a conjecture supported by finite-size evidence rather than a proven result. The CONDITIONAL verdict is appropriate: the paper is a valuable contribution with a concrete, falsifiable prediction, but the headline claim of 'absence of quantum advantage' is not yet established. The concrete test of pushing exact quantum calculations to larger p would directly settle whether the extrapolation holds. If the quantum data continues to track log(p)/p and remains above the semiclassical curve, the claim is substantially strengthened; if a crossover occurs, it fails.","tokens_in":7304,"tokens_out":2719,"duration_ms":148241,"concrete_test":"Extend the exact quantum QAOA calculation (using the methods of Ref [9]) to p=100–200 and compare the energy directly with the TWA at optimal S*(p). If the quantum energy drops below the semiclassical energy at any p in this range, the log(p)/p upper bound on QAOA fails. Additionally, check whether the quantum data at p=80–200 is better fit by 1/p^α with α>0.88 (approaching 1) rather than log(p)/p, which would indicate the semiclassical scaling does not govern the quantum convergence rate.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's central claim — that QAOA converges to the Parisi value as log(p)/p and therefore offers no quantum advantage — depends on a single empirical observation: the TWA semiclassical simulation at optimal S* slightly outperforms the exact S=1/2 QAOA for p=10 to 80. From this, the authors extrapolate to all p: 'We have no reason to believe this does not persist to asymptotically large p.' This is the load-bearing assumption, and it is not supported by a theoretical bound. The TWA at large S*~p is not an approximation of the quantum S=1/2 dynamics — it is a different classical dynamical system with tunable Gaussian noise. The TWA is controlled only at large S, and the paper itself notes it fails at small S. There is no established inequality relating the two systems' performances. A crossover at larger p is physically plausible: the TWA cannot capture quantum interference or entanglement effects that may become more effective at greater depth, and the log(p)/p scaling is itself derived from a mechanism (Ehrenfest-time competition between initial noise and exponential growth) specific to the classical noisy dynamics, not to the quantum evolution. The quantum QAOA has fixed fluctuation strength (S=1/2) and does not face the same noise–localization tradeoff that produces the logarithmic correction in the semiclassical analysis. Without a rigorous bound or a mechanism explaining why quantum cannot eventually outperform the semiclassical, the extrapolation from p≤80 to asymptotic p is the weakest link.","agreement_with_reader":"agree"},"referee_report":{"model":"glm-5.2","summary":"The manuscript studies the Quantum Approximate Optimization Algorithm (QAOA) applied to the Sherrington-Kirkpatrick (SK) spin glass using the Truncated Wigner Approximation (TWA). By treating the spin magnitude S as a tunable parameter controlling initial quantum noise, the authors identify two competing effects: at small S, excessive initial noise degrades performance; at large S, exponential growth of fluctuations (governed by the Ehrenfest time) limits the dynamics. The optimal balance occurs at S* ~ p, yielding a residual energy scaling of log(p)/p above the Parisi value. The key claim is that the semiclassical simulation slightly outperforms the exact S=1/2 QAOA for p=10 to 80, from which the authors infer that the exact QAOA also converges as log(p)/p, implying no quantum advantage. They further show that removing initial noise and re-optimizing parameters yields 1/p convergence.","tokens_in":7593,"tokens_out":1394,"duration_ms":172755,"significance":"The question of whether QAOA offers a quantum advantage for spin glass optimization is of considerable interest. The paper provides a concrete, falsifiable prediction for the scaling of QAOA performance on the SK model (log(p)/p) and a physical mechanism (Ehrenfest-time competition) explaining the logarithmic correction. The observation that a noiseless classical variant achieves 1/p convergence is a notable result. However, the central claim regarding the exact quantum QAOA's asymptotic scaling rests on an extrapolation from finite p, which limits the significance of the strongest conclusions.","major_comments":[{"comment":"The central claim that the exact quantum S=1/2 QAOA converges as log(p)/p rests on the empirical observation that the TWA simulation at optimal S* slightly outperforms the exact QAOA for p=10 to 80 (Fig. 3). The authors state: 'We have no reason to believe this does not persist to asymptotically large p.' This is the load-bearing assumption of the paper's main conclusion, and it is not supported by a rigorous bound or a theoretical argument. The TWA at S*~p is a classical dynamical system with tunable Gaussian noise, not an approximation of the quantum S=1/2 dynamics (the TWA is controlled at large S and the paper itself notes it fails at small S). There is no established inequality relating the performance of the two systems. A crossover at larger p is physically plausible, as the quantum QAOA has fixed fluctuation strength and does not face the same noise-localization tradeoff. The log","section":null},{"comment":"The log(p)/p scaling for the semiclassical simulation is derived from the heuristic model in Eq. (7): delta(S,p) ~ a/S + c*log(S)/p. While this fits the data well (Fig. 1), the model itself is an ansatz, not derived from first principles. The two terms correspond to two distinct mechanisms (initial noise and Ehrenfest-time growth), but their additive combination and the specific functional forms are assumed. The claim that 'the QAOA can at best converge to the Parisi value like log(p)/p' (page 4) is presented with more certainty than the heuristic derivation supports. The authors should more clearly distinguish between the empirically observed scaling of the TWA simulation and the inferred scaling of the exact QAOA, and should acknowledge that the latter is a conjecture supported by finite-size evidence rather than a proven bound.","section":null},{"comment":"The complexity argument on page 4 states that achieving a 1-epsilon approximation requires depth p = log(1/epsilon)/epsilon, yielding O(N^2 log(1/epsilon)/epsilon) complexity. However, this complexity is derived from the log(p)/p scaling, which is itself the extrapolated result. If the exact QAOA were to eventually outperform the TWA at larger p (as raised in comment 1), the complexity could be better. The complexity claim is only as strong as the scaling claim, and this dependency should be stated explicitly.","section":null}],"minor_comments":[{"comment":"In Eq. (7), the constant is labeled 'c' in the equation but referred to as 'b' in the text on page 3 ('delta ~ b*log(S)'). The fit function in Fig. 1 caption uses 'b' as well. Please make the notation consistent.","section":null},{"comment":"The abstract states 'convergence of the final energy to the Parisi value like log(p)/p' without qualification. Given that this is an extrapolation for the exact QAOA (see major comments), the abstract should reflect the inferential nature of this claim.","section":null},{"comment":"On page 2, the text states 'the QAOA parameters {gamma},{beta} are taken directly from Ref. [9].' It would be helpful to briefly state what system size N was used for the TWA simulations (the figure caption of Fig. 3 mentions N=16384, but this should be stated in the main text for clarity).","section":null},{"comment":"The reference to 'Ref. [18]' for the noiseless classical results and the re-optimized parameters is cited as a companion paper. Since the 1/p convergence result for the noiseless case is a significant part of the paper's narrative, a brief summary of the method used in Ref. [18] would help the reader assess this claim within the present manuscript.","section":null},{"comment":"Fig. 2: The caption mentions 'Different curves show different spin-S, which can be identified from the initial variance.' It would be clearer to explicitly state the range of S values shown or add a legend.","section":null}],"recommendation":"major_revision","confidential_remarks":"The paper presents interesting numerical observations and a plausible physical mechanism, but the leap from 'TWA slightly outperforms QAOA at p=10-80' to 'QAOA converges as log(p)/p and there is no quantum advantage' is too large for the current evidence. The authors need to either (a) provide a more rigorous argument for why the TWA upper-bounds the QAOA performance asymptotically, or (b) substantially soften the claims to reflect that the log(p)/p scaling is a conjecture supported by finite-size numerics. The paper is suitable for publication if the claims are appropriately qualified and the distinction between the TWA simulation's scaling and the exact QAOA's inferred scaling is made clear."},"author_rebuttal":null,"desk_editor":{"model":"glm-5.2","letter":"The headline: Sels and Morone use a truncated Wigner (semiclassical) analysis of QAOA on the Sherrington-Kirkpatrick model to predict that QAOA converges to the Parisi ground-state energy at rate log(p)/p, and therefore offers no quantum advantage over classical approaches. The log(p)/p is not a theorem — it's a scaling argument backed by numerics for p up to 80 — but the physical mechanism is specific and the argument is worth taking seriously.","headline":"Sels–Morone predicts log(p)/p convergence for QAOA on SK; the bound is heuristic, not rigorous, but the physical mechanism is concrete and the numerics are consistent.","tokens_in":8102,"tokens_out":183,"would_cite":true,"duration_ms":87960,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["75.50.Lk","03.67.Ac","05.30.-d"],"model":"glm-5.2","headline":"Quantum algorithm offers no edge over classical on spin glass","keywords":[],"falsifier":"If exact quantum QAOA at depths well beyond p = 80 were found to converge faster than log(p)/p — for instance, as 1/p — the central claim of no quantum advantage would be directly undermined, since the semiclassical upper bound would no longer be tight.","tokens_in":7377,"feed_emoji":"🎲","tokens_out":2700,"duration_ms":125305,"temperature":0.7,"pith_summary":"The authors apply the Truncated Wigner Approximation — a semiclassical method where the effective spin S controls the level of quantum noise — to simulate QAOA on the Sherrington-Kirkpatrick spin glass, a benchmark optimization problem with a known optimal energy (the Parisi value). They find that the final energy depends non-monotonically on S: at small S, excessive initial noise prevents spins from correlating with the couplings; at large S, the near-zero-noise initial state is an unstable fixed point of the classical dynamics, so tiny fluctuations grow exponentially at a rate set by the Lyapunov exponent, saturating only after the Ehrenfest time t_E ~ log(S). The optimal balance occurs at S* proportional to p (the circuit depth), yielding a residual energy above the Parisi value that scales as log(p)/p. Crucially, the semiclassical simulation slightly outperforms the exact quantum spin-1/2 QAOA across all depths studied (p = 10 to 80). The authors infer from this that the quantum algorithm converges to the Parisi value no faster than log(p)/p, matching the classical rate and implying no quantum advantage. They further show that the logarithmic correction is caused by entropy in the quantum initial state: removing all initial noise and re-optimizing the rotation angles (going fully classical) eliminates it, achieving 1/p convergence.","feed_headline":"Quantum algorithm offers no edge over classical on spin glass","feed_subtitle":"Semiclassical simulation slightly outperforms exact QAOA; both converge to the optimum as log(p)/p, with no quantum speedup.","key_machinery":"The Truncated Wigner Approximation (TWA): a semiclassical method that replaces quantum dynamics with classical rotation of spin vectors, sampling initial conditions from the Wigner distribution of the initial quantum state. The effective spin S controls the width of this distribution (the quantum noise), with variance 1/(2S). The classical map alternates z-rotations (set by instantaneous local fields) with uniform x-rotations, mirroring the QAOA circuit. Two competing effects — entropy at small S and exponential growth of fluctuations at large S via the Lyapunov instability of the initial state — produce a non-monotonic energy landscape with an optimum at S* ~ p.","core_discovery":"A semiclassical simulation of QAOA on the SK spin glass, where the effective spin S tunes the quantum noise level, slightly outperforms the exact quantum spin-1/2 QAOA at every depth studied. The optimal noise level scales linearly with depth p, and the resulting convergence to the Parisi optimum goes as log(p)/p. Since the semiclassical method is an upper bound on quantum performance, the quantum algorithm can do no better — it converges at the same log(p)/p rate, offering no advantage over classical approaches. The logarithmic penalty traces to quantum entropy in the initial superposition state and can be removed entirely by going noiseless with re-optimized parameters.","pith_inferences":[],"forward_implications":["If the log(p)/p scaling holds asymptotically, achieving a (1-ε) approximation to the SK ground state via QAOA requires depth p = log(1/ε)/ε, giving total complexity O(N² log(1/ε)/ε) — the same order as classical message-passing algorithms up to logarithmic factors.","The quantum circuit's apparent O(N) per-layer advantage over the classical O(N²) comes only from parallelizing spin-spin interactions, which classical computation can also parallelize, neutralizing the quantum speedup in wall-clock terms.","The logarithmic slowdown is specifically caused by quantum entropy in the initial superposition; a fully classical noiseless protocol with re-optimized parameters achieves 1/p convergence, suggesting the quantum initial state is actively harmful for this problem class.","The TWA-as-upper-bound technique may extend to other combinatorial optimization problems where QAOA is applied, potentially revealing whether the absence of quantum advantage is generic to mean-field spin glasses."],"fun_headline_variants":["No quantum edge found for spin glass optimization","Semiclassical beats quantum on spin glass problem","Quantum optimization matches classical on spin glass","No quantum speedup for approximate spin glass optimization","Semiclassical method edges out QAOA on spin glass"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The claim that the exact quantum QAOA converges as log(p)/p rests on the observation that the semiclassical simulation slightly outperforms it over the finite range p = 10 to 80, with no theoretical proof that this relationship persists at arbitrarily large depth. The log(p)/p scaling itself is inferred from fitting this limited data rather than derived from first principles.","fun_headline_variants_meta":{"raw":{"variants":["No quantum edge found for spin glass optimization","Semiclassical beats quantum on spin glass problem","Quantum optimization matches classical on spin glass","No quantum speedup for approximate spin glass optimization","Semiclassical method edges out QAOA on spin glass","Quantum noise caps QAOA performance on spin glass","Spin glass QAOA shows no quantum advantage","Classical simulation outperforms quantum on spin glass","QAOA gains no edge over semiclassical spin glass solver","Log penalty from quantum noise limits spin glass optimization"]},"model":"glm-5.2","effort":"low","cost_usd":0.0,"raw_usage":{"total_tokens":1054,"prompt_tokens":507,"completion_tokens":547,"prompt_tokens_details":null},"tokens_in":507,"tokens_out":547,"duration_ms":18407,"temperature":1.0,"reasoning_tokens":417,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-10T02:36:11.541197+00:00","model_set":{"reader":"glm-5.2"},"falsifier":"If exact quantum QAOA at depths well beyond p = 80 were found to converge faster than log(p)/p — for instance, as 1/p — the central claim of no quantum advantage would be directly undermined, since the semiclassical upper bound would no longer be tight.","supporting_citations":[],"review_version":1}