{"id":"cb7a789d-1e02-425c-89f3-451946e9d4a0","arxiv_id":"2509.08807","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":10,"one_line_summary":"A spectral-sparsity-based quantum solver is claimed to solve 2^80-cell Navier-Stokes problems in 42.6 days with 8.71 million physical qubits, a 1,100x speedup over a classical supercomputer.","lead":"The paper claims a full-stack, error-corrected quantum algorithm that solves Navier-Stokes equations exponentially faster than classical computers, estimating 8.71 million physical qubits and 42.6 days for a 2^80-cell grid. A generalist should read it because it is one of the most concrete end-to-end resource proposals for practical quantum advantage in a core engineering PDE.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Tomography sample count N_sample=1000 is fitted only for d_eff≤32 but applied to d_eff≈2.7e5; the paper's own CS bound raises measurements ~10^5-fold, invalidating the 42.6-day claim.","rationale":"The paper's headline feasibility claim rests on the resource estimate of Eq. (F30), which is linearly proportional to N_sample. The value N_sample=1000 is obtained from numerical CS-QST experiments whose largest tested effective dimension is 32, while the actual decoded state used in the resource model has effective dimension ~270,400 by the paper's own Appendix C.5. This is not a matter of disagreeing with an outside consensus; it is an internal inconsistency between the empirical validation regime and the extrapolation target. Theorem C.9, which the paper itself cites, gives m = C r d_eff log^2 d_eff, and plugging in the self-reported effective dimension yields a required measurement count orders of magnitude larger than 1000. Since the time estimate is proportional to N_sample, the central claim of a 42.6-day runtime and a ~1100x speedup collapses. The paper's Discussion contains an explicit limitation statement acknowledging that the tomography sampling number is 'numerically validated' rather than backed by a tight bound for large S, which confirms that this is a known weak point rather than an artifact of the review pipeline. The secondary τ omission in the proof of Theorem 2 reinforces that the asymptotic optimality claim is not established, but it is not the primary reason for rejecting the practical resource claim. The block-encoding and circuit-synthesis contributions may be independently useful, and the low-dimensional numerical experiments appear consistent, but they do not rescue the headline extrapolation.","tokens_in":64382,"tokens_out":5599,"duration_ms":66152,"concrete_test":"Evaluate Theorem C.9 for d_eff = S^3+2S^2+S = 270400 using the regression constant fitted from Fig. 19 (and, as a robustness check, run the same CS-QST protocol at intermediate d_eff=64–256 to verify the scaling). If the predicted N_sample exceeds 1000 by a factor ≳10^4, replace Eq. (F26) with the predicted value and recompute Eq. (F30): the runtime will exceed the classical estimate, invalidating the central claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Eq. (F30) makes the central runtime T_quantum = N_sample × depth × d × T_cycle. N_sample is set to 1000 in Eq. (F26), based on Appendix F's compressed-sensing tests for effective dimensions 4–32 (Fig. 19). But Appendix C.5 computes the decoded state's effective dimension as S^3+2S^2+S, i.e., ~270,400 for the S=64 case used in the resource estimate. The paper's own Theorem C.9 requires m = C r d_eff log^2 d_eff random Pauli expectations; even with the small prefactor implied by Fig. 19 (roughly C≈1), this gives m~10^8 rather than 1000. The manuscript's Discussion explicitly concedes that no tight theoretical bound for large S is known and that the sampling number is only numerically validated. Since Eq. (F30) is linear in N_sample, the 42.6-day headline becomes centuries/millennia once the sampling scaling is applied, eliminating the claimed ~1100x advantage. A secondary issue is that the proof of Theorem 2 in Appendix C.6 derives per-iteration complexity and drops the τ factor present in the theorem statement, so the claimed saturation of the Theorem 1 lower bound is also not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an end-to-end, fault-tolerant quantum algorithm for solving the Navier-Stokes equations (NSE) on very large grids. The claimed architecture combines a spectral-sparsity-based input/output protocol (hierarchy spectral block-encoding and sparse spectral tomography), explicit circuit synthesis (match-and-merge, mask-and-merge), a QLSS based on Ref. [26], and a full error/resource model with surface-code and magic-state overheads. The headline result is that a 2^80-cell two-dimensional NSE problem can be solved in 42.6 days using 8.71 million physical qubits at a physical error rate of 5e-4, giving an approximately 1100x speedup over a state-of-the-art supercomputer. The paper also states an asymptotic end-to-end complexity that it claims saturates a lower bound for iterative QLSS.","tokens_in":64739,"tokens_out":4368,"duration_ms":46926,"significance":"If the resource estimate were correct, this would be a landmark result in practical quantum advantage: it would be the first full-stack, end-to-end estimate for a nonlinear PDE with a concrete fault-tolerant resource count, explicit circuits, and a nontrivial lower bound. The paper genuinely contributes several interesting technical pieces: an explicit hierarchy block-encoding for the structured Jacobian, a spectral I/O scheme that reduces the effective output dimension from N to poly(S), a concrete circuit-synthesis methodology with numerical reductions, and an end-to-end error accounting that separates algorithmic, deployment, and tomography errors. However, the central quantitative claim is invalidated by the tomography sampling estimate, as detailed below. The asymptotic framework and the circuit-synthesis techniques may still be of interest, but the paper's practical-advantage conclusion is not supported by the evidence presented.","major_comments":[{"comment":"The central runtime estimate T_quantum = N_sample × (RD+TD) × d × T_cycle is linear in N_sample, but N_sample = 1000 is not justified for the actual decoded state. The decoded state's effective dimension is computed in Appendix C.5 as S^3 + 2S^2 + S, i.e., about 270,400 for S=64, the value used in the resource estimate. The paper's own compressed-sensing bound (Theorem C.9) requires m = C r d_eff log^2 d_eff random Pauli expectations. The numerical fitting in Appendix F (Fig. 19) is performed only for effective dimensions 4 to 32; even using the small fitted prefactor implied there, one obtains m ~ 10^7-10^8, not 1000. Since Eq. (F30) is linear in N_sample, the 42.6-day headline becomes several centuries to millennia, eliminating the claimed ~1100x advantage. The Discussion explicitly concedes that no tight theoretical bound for large S is known; the empirical extrapolation from d_eff <=","section":"Appendix F, Eqs. (F26) and (F30); Appendix C.5; Theorem C.9"},{"comment":"The proof of Theorem 2 derives an end-to-end time complexity of Õ(S ε^{-2}(κ(S + log log N) + log^2 N)) and does not include the factor τ that appears in the theorem statement, Õ((κDS + log^2 N) τS / ε^2). No argument is given for how τ is absorbed or why the per-iteration cost should be independent of the iteration count. Thus the claimed saturation of the Theorem 1 lower bound is not established. This is not merely a typo: the resource model in Eq. (F30) also contains no τ factor, so the practical time estimate appears to be for a single iteration rather than for the full NSE simulation, unless τ is somehow hidden inside (RD+TD), which is not shown.","section":"Appendix C.6, proof of Theorem 2"},{"comment":"The classical cost estimate appears to count a single linear solve of dimension NLS = 2^82 (via CG or Cholesky), with FLOP counts 9.96e29 and 7.13e27 respectively. The NSE is time-marched with implicit Euler, requiring a linear solve at each step; the number of iterations τ is present in the quantum algorithm (Algorithm 3) and in Theorem 2, but no corresponding multiplication by τ appears in T_ElCapitan64. If τ is large, the classical baseline is undercounted by a factor of τ. This does not rescue the quantum estimate, but it makes the claimed crossover ratio T_quantum/T_classical unquantified.","section":"Appendix F.4, classical baseline"}],"minor_comments":[{"comment":"There are several typographical inconsistencies in the appendices, e.g., Eq. (C24) labels the vertical viscous flux as G_C = ... (should be G_mu), and 'ad iabetic' appears for 'adiabatic' in Appendix F. These are cosmetic but should be cleaned up.","section":"General / notation"},{"comment":"The notation 'ND j=1 O_SUB(N_j)' is missing the tensor-product symbol; it should read \\bigotimes_{j=1}^D O_SUB(N_j).","section":"Eq. (C109)"},{"comment":"The sentence 'We track the value of T∝e and ρ with varying problem size and find that ...' is repeated almost verbatim two sentences later. Please consolidate.","section":"Appendix F.2, iteration-tolerant error bound"},{"comment":"The caption for Fig. 19 does not state the effective dimensions or the fitted constants for the linear regression used to estimate C in Theorem C.9. Since the sampling number is a key hyperparameter, the figure and the regression details should be fully specified.","section":"Fig. 19 caption / Appendix F"}],"recommendation":"reject","confidential_remarks":"The paper has interesting algorithmic ideas and a serious full-stack resource-analysis structure, but the practical quantum-advantage claim is the paper's raison d'être and it is not supported once the tomography sampling count is corrected using the paper's own bounds. The N_sample issue is not a minor prefactor: it changes the endpoint by five orders of magnitude and erases the claimed crossover. The τ omission in the proof of Theorem 2 and in the resource model is an additional correctness gap. These are load-bearing errors that cannot be fixed by local revisions; the advertised conclusion would need to be substantially weakened. I recommend rejection, while noting that the circuit-synthesis and hierarchy-block-encoding contributions might be publishable as a separate, more modest algorithmic-resource paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a serious, detailed attempt at an end-to-end quantum Navier-Stokes solver, and it deserves a careful referee rather than a desk reject. The new pieces are real: the iterative-QLSS lower bound (Theorem 1), the hierarchy spectral block-encoding, the match-and-merge and mask-and-merge synthesis, and a genuinely end-to-end error model. The authors also give proper credit where the spectral block-encoding core is a reformulation of their prior state-preparation framework and Rosenkranz et al. Those contributions are worth keeping.\n\nThe headline 42.6-day claim, however, does not survive contact with the paper's own math. The resource model sets N_sample = 1000 in Eq. (F26), validated only for effective dimensions 4–32. The decoded state for S=64 has effective dimension S^3 + 2S^2 + S ≈ 2.7e5 by the paper's own Appendix C.5. Theorem C.9 requires m = C r d_eff log^2 d_eff measurements; even with a modest constant this is around 5e7, not 1000. Since T_quantum is linear in N_sample, the 42.6 days becomes millennia. The Discussion explicitly concedes that no tight bound for large S is known and that the sampling number is only numerically validated. This is not a loose prefactor; it is the load-bearing number in the paper. I agree with the stress-test note on this point.\n\nThere is a second formal gap: the proof of Theorem 2 in Appendix C.6 derives per-iteration complexity and drops the tau factor that appears in the theorem statement, so the claimed saturation of the Theorem 1 lower bound is not established. The classical baseline also appears to count a single Cholesky solve rather than tau time steps, which understates the classical work; that direction actually helps the quantum comparison, but it is still an inconsistency in the accounting.\n\nWho should read this? Anyone working on quantum algorithms for nonlinear PDEs or on end-to-end FTQC resource estimation. It is a useful template for what a full-stack analysis should include, and a cautionary example of how empirical fitting can be over-extrapolated. The framework is repairable—fix the sampling scaling, restate Theorem 2 honestly, and the asymptotic and synthesis results may stand on their own.\n\nMy recommendation: accept for peer review with major revision. The flaws are specific, identifiable, and fixable, and the paper has enough original content to justify referee time even though the central practical claim is currently unsupported.","headline":"A genuinely full-stack quantum CFD feasibility study whose headline resource estimate is undone by a five-order-of-magnitude tomography sampling undercount.","tokens_in":65325,"tokens_out":3682,"would_cite":false,"duration_ms":44023,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper's central claim: a fault-tolerant quantum computer with 8.71 million physical qubits solves a 2^80-cell Navier-Stokes problem in 42.6 days, where a state-of-the-art supercomputer would need over a century.","keywords":["quantum advantage","Navier-Stokes equations","fault-tolerant quantum computing","quantum linear system solver","spectral sparsity","hierarchy block-encoding","compressed-sensing quantum state tomography","circuit synthesis"],"falsifier":"Take a random S-sparse vector with S=64 on a grid of 2^80 cells, Fourier-transform it, and run the paper's compressed-sensing tomography at effective dimension d_eff = S^3+2S^2+S ≈ 2.7×10^5; if the measured sample count needed to reach fidelity 6×10^-5 tracks m = C·r·d_eff·log^2(d_eff) with the paper's fitted C, the 42.6-day runtime stands, and if the fit stays at the values of the paper's Fig. 19, the runtime grows by about five orders of magnitude. A second check: re-derive Theorem 2 from the Appendix C.6 proof and confirm the τ factor in the stated bound Õ((κDS+log^2 N)τS/ε^2) appears in th","tokens_in":64221,"feed_emoji":"⚛️","tokens_out":15288,"duration_ms":136989,"temperature":0.7,"pith_summary":"The paper attempts to show that the Navier-Stokes equations—nonlinear, input/output-heavy, and a cornerstone of computational fluid dynamics—can be solved on a fault-tolerant quantum computer with an exponential end-to-end speedup over classical simulation, and that the speedup survives a full accounting of circuit depth, error correction, and readout. Its central move is the claim that a fluid field with small spectral sparsity S can be fed into a quantum linear-system solver, evolved through nonlinearity by classical iteration, and read back out, all while touching only O(S)-dimensional data instead of the full N-cell grid. If the resource estimates hold, a 2^80-grid compressible-flow simulation would run in about six weeks on 8.71 million physical qubits at a 5×10^-4 error rate, against an estimated 130 years for the most powerful current supercomputer. The authors further claim the algorithm's time complexity matches a lower bound they prove for any iterative quantum linear-system solver, so the exponential speedup is not an artifact of one implementation.","feed_headline":"42.6 days on 8.7M qubits claimed to outrun a century of supercomputing","feed_subtitle":"Compressing fluid fields to a small spectral subspace dodges the bottleneck blocking quantum fluid dynamics.","key_machinery":"Three devices carry the argument. The iteration-tolerant bandwidth S—the minimum number of independent classical components that must pass between iterations of a quantum linear-system solver—is the subject of Theorem 1, which bounds any such solver's time by roughly τκ·Poly(S,1/ε). The hierarchy spectral block-encoding decomposes the flux Jacobian into a diagonal level, an inter-block level of 4×4 flux blocks, and an element level of diagonal functions, encoding each entry from S spectral coefficients rather than N point values; non-polynomial terms enter via polynomial approximation plus quantum singular-value transformation. Sparse spectral decoding applies a parallel quantum Fourier tran","core_discovery":"Spectral sparsity removes the input, nonlinearity, and readout bottlenecks of quantum fluid simulation. The iteration-tolerant bandwidth S—the fewest classical components an iterative quantum linear-system solver must carry—is proved to force Ω(τκ Poly(S,1/ε)) time, so exponential speedup needs S logarithmic in N. The hierarchy spectral block-encoding assembles the flux Jacobian in three levels and encodes each variable from S spectral coefficients in O(S log N) depth; sparse spectral decoding collapses the dense solution to effective dimension S^3+2S^2+S before compressed-sensing tomography. Payoff: a 2^80-cell Navier-Stokes problem in 42.6 days on 8.71 million physical qubits, versus 130 y","pith_inferences":["If one applies the paper's own compressed-sensing bound at the decoded effective dimension it computes (S^3+2S^2+S ≈ 2.7×10^5 for S=64) instead of the fitted range 4–32, the required measurements rise to about 10^7, which under the paper's time formula would multiply the 42.6-day estimate by roughly five orders of magnitude; this arithmetic is mine, not the paper's.","The iteration-tolerant bandwidth gives problem selectors a concrete diagnostic: measure the spectral sparsity of the intermediate fields of a candidate fluid solver, and if it grows with grid resolution, the exponential advantage is excluded by the paper's own Theorem 1.","The hierarchy block-encoding is a transferable construction for any locally connected discrete PDE—elasticity, electromagnetics, geophysics—so the framework's method, not just its fluid result, is portable to neighboring simulation problems.","The reported κ-saturation (condition number bounded once S is fixed, observed in the benchmark vortex flows studied numerically) implies a scaling law between spectral sparsity and linear-system conditioning that, if confirmed for other flows, would let the quantum cost model's dominant parameter be read off from classical spectra."],"forward_implications":["A fault-tolerant machine near 8.7 million physical qubits at 5×10^-4 physical error rate would outperform a top supercomputer on a 2^80-grid compressible Navier-Stokes problem by a factor of roughly 1,100 in wall-clock time, per the paper's estimates.","Because of the lower bound, any iterative quantum linear-system solver for a problem whose intermediate solutions are spectrally dense is provably limited to polynomial speedup; spectral sparsity is the decisive problem feature.","The qRAM-free input/output design means the grid size N enters the cost only logarithmically (through the quantum Fourier transform and shift arithmetic), so the framework scales to finer grids without an exponential qubit or gate overhead.","The logical-resource reductions (about 23× Toffoli depth, 50× Toffoli count, 5× rotation count and depth) and the 3.5× magic-state-factory cut are what turn an asymptotically efficient algorithm into a specific hardware budget."],"fun_headline_variants":["Quantum speeds Navier-Stokes: 42.6 days vs a century on 8.7M qubits","Spectral trick shrinks quantum fluid sim to 8.7M qubits, 42.6 days","Exponential speedup for Navier-Stokes with spectral sparsity","Quantum advantage in fluid dynamics: 2^80-grid in 42.6 days","8.71M qubits solve Navier-Stokes 1000x faster than supercomputer"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The feasibility window rests on one empirical number: tomography sampling was fitted on states of effective dimension 4 to 32 and then applied to a decoded state whose effective dimension the paper itself computes as roughly 270,000, where its own compressed-sensing formula would require tens of millions of measurements rather than 1,000.","fun_headline_variants_meta":{"raw":{"variants":["Quantum speeds Navier-Stokes: 42.6 days vs a century on 8.7M qubits","Spectral trick shrinks quantum fluid sim to 8.7M qubits, 42.6 days","Exponential speedup for Navier-Stokes with spectral sparsity","Quantum advantage in fluid dynamics: 2^80-grid in 42.6 days","8.71M qubits solve Navier-Stokes 1000x faster than supercomputer"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001205,"raw_usage":{"total_tokens":4811,"prompt_tokens":763,"completion_tokens":4048,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":507,"completion_tokens_details":{"reasoning_tokens":3928}},"tokens_in":507,"tokens_out":4048,"duration_ms":25755,"temperature":1.0,"reasoning_tokens":3928,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T20:07:23.284919+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a random S-sparse vector with S=64 on a grid of 2^80 cells, Fourier-transform it, and run the paper's compressed-sensing tomography at effective dimension d_eff = S^3+2S^2+S ≈ 2.7×10^5; if the measured sample count needed to reach fidelity 6×10^-5 tracks m = C·r·d_eff·log^2(d_eff) with the paper's fitted C, the 42.6-day runtime stands, and if the fit stays at the values of the paper's Fig. 19, the runtime grows by about five orders of magnitude. A second check: re-derive Theorem 2 from the Appendix C.6 proof and confirm the τ factor in the stated bound Õ((κDS+log^2 N)τS/ε^2) appears in th","supporting_citations":[],"review_version":1}