{"id":"1952c4d3-68af-4c3c-b224-7f98f1146854","arxiv_id":"2607.13641","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"An end-to-end quantum algorithm for PIV-style 2D cross-correlation — zero-mean sparse state preparation plus contracted amplitude amplification — recovers benchmark velocity fields to sub-pixel accuracy in simulated runs.","lead":"This paper designs an end-to-end quantum algorithm, QuPIV, that computes the particle-displacement cross-correlations at the heart of Particle Image Velocimetry using quantum Fourier transforms and a new gate-saving 'contracted' amplitude amplification. On synthetic and benchmark experimental images the simulated pipeline recovers velocity fields to sub-pixel accuracy, though the authors explicitly disclaim any proven speedup over classical computing.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Contracted-amplification decoupling (Eqs. 14-15) is asserted, not proven; S_ref and S_G both mark B=0, so the iteration is not standard Grover, and Fig. 4(c)'s error estimate depends on this unverified claim.","rationale":"The reader's weakest assumption — the contracted-amplification decoupling of Section 3.4 — is indeed the most load-bearing point. The paper's headline numerical claim, that 15 contracted amplification steps plus 500 samples give 99.9% of peak errors below 1 pixel and processing errors below 10^-2 pixels, depends on Eq. (15) being a correct description of the circuit's behavior. That equation is not derived from the circuit in the main text; it is asserted from a decoupling picture and deferred to an unavailable Supplementary Note. Moreover, replacing S0 with a projector that marks the same B=|0> subspace as S_G changes the structure of the iteration from standard Grover to a product of reflections about two subspaces, so the sin((2r+1)theta) growth is not automatic. This is not an external-consensus disagreement; it is an internal correctness risk. I do not see a stronger objection: the QFT-based cross-correlation construction and the state-normalization algebra are internally consistent, and the authors are appropriately candid about the lack of a proven speedup and about image reduction being the dominant error source. The decoupling assumption is the single point where the argument is least secure. A full statevector simulation at accessible sizes would settle it, and the paper should either provide that verification or supply the promised Supplementary analysis before the central claim is accepted. My recommendation therefore leaves the reader's CONDITIONAL verdict unchanged.","tokens_in":15104,"tokens_out":30119,"duration_ms":281197,"concrete_test":"Statevector-simulate the actual contracted-amplification circuit (state preparation, QFTs, cNOTs, S_ref, S_G) for N=8 or N=16 (12-16 qubits total) on several random image pairs. After each query r=0..15, compare the exact probability mass in the B=0 subspace with the prediction of Eq. (15). Also compute the Schmidt decomposition of |Psi_U> across A|B and verify that the branch coefficients c_k and angles theta_k match Eq. (15). If exact success probabilities or branch angles deviate beyond statistical uncertainty, the decoupling assumption fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The end-to-end viability claim rests on Section 3.4's contracted amplitude amplification. Eqs. (14)-(15) assert that replacing S0 by S_ref on register B decomposes the state into 2^kappa independent branches with local growth sin((2r+1)theta_k), c_k=|A_k|, sin(theta_k)=|B_k|. But in this application, S_ref marks B=|0>, which is exactly the target subspace marked by S_G; hence the contracted Grover step is R_{U T} R_T (reflections about U(T) and T), not the standard R_{U|0>} R_T. The statement that 'the state on register A remains unaffected' is insufficient: the cNOTs entangle A and B, and QFT_A does not commute with the cNOT, so U S_ref U† is not block-diagonal in the A basis. The paper defers the proof to the Supplementary Note and says only 'we find it also to hold if the amplification can be decoupled.' The experimental-scale processing error in Fig. 4(c) is computed from Eq. (15), not from a full 24-qubit circuit simulation. If the decoupling fails, both the gate-count saving and the sub-10^-2-pixel processing-error prediction are unsupported. Because the abstract and Introduction call this the central advancement and the basis of the end-to-end claim, this is the most load-bearing assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes QuPIV, an end-to-end quantum algorithm for estimating particle-image displacements in Particle Image Velocimetry. The pipeline is: (i) binarize each interrogation window by selecting the N largest-intensity pixels; (ii) prepare zero-mean sparse states on two quantum registers A and B; (iii) implement 2D cross-correlation via QFT, complex-conjugated QFT, cNOT-based sorting, and inverse QFT; (iv) apply a modified amplitude-amplification scheme called 'contracted amplitude amplification'; and (v) sample the correlation map and locate the peak with three-point interpolation. The authors validate the approach on case F of the 4th PIV Challenge, comparing against the analytical solution and the classical OpenPIV package. They report that, with 15 contracted amplification queries and 500 samples, 99.9% of peak position errors are below 1 pixel, and that the processing error is below 10^-2 pixels. The central theoretical claim is that contracted amplitude amplification—replacing the all-qubit ground-state projector S0 by a projector S_ref on register B—decouples the amplification into 2^κ independent branches, enabling simulation of the processing error without running the full 24-qubit circuit.","tokens_in":15436,"tokens_out":14069,"duration_ms":127598,"significance":"The problem is well motivated, and the paper is unusually complete in treating the input, processing, and output stages of a quantum algorithm for an industrially relevant classical task. The circuit-level design is concrete, the statistical treatment of input, output, and processing errors is thoughtful, and the use of a well-known experimental benchmark (4th PIV Challenge, case F) gives the work a clear falsifiable target. The claimed gate reduction via contracted amplitude amplification, if rigorously justified, could also be of broader interest to amplitude-amplification applications. However, the central theoretical assertion—the decoupling of the amplification into independent register-A branches—is not proved in the main text, and the paper's headline numerical claims depend on that assertion. The significance is therefore conditional on resolving the correctness of Section 3.4.","major_comments":[{"comment":"The contracted-amplification analysis is the load-bearing part of the paper, but the key relation is not established. In the proposed use-case, S_ref and S_G both act on register B and both mark the B=|0> subspace; hence the iteration is U S_ref U† S_G = R_{U(T)} R_T, not the standard Grover iteration R_{U|0>} R_T. The statement that 'the state on register A remains unaffected' is insufficient: the cNOTs entangle A and B, QFT_A does not commute with the cNOTs, and U is not block-diagonal in the A basis. The paper defers the proof to the Supplementary Note and states only 'we find it also to hold if the amplification can be decoupled'. Please provide a full derivation of Eqs. (14)-(15), or a small full-circuit simulation that directly verifies the predicted branch-wise rotation and the formulas c_k=|A_k|, sin(theta_k)=|B_k|.","section":"3.4, Eqs. (14)-(15)"},{"comment":"The processing-error curve in Fig. 4(c) is computed from Eqs. (14)-(15), not from a simulation of the 24-qubit contracted-amplification circuit. Since Eq. (15) is precisely the relation whose validity is at issue, this does not constitute an independent validation. The claims of a sub-10^-2 pixel processing error and of end-to-end viability therefore rest on an untested assumption. A full simulation for at least one representative interrogation window (or a formally derived error bound) is needed to support Fig. 4(c).","section":"4, Fig. 4(c)"},{"comment":"The phrase 'end-to-end quantum algorithm' and the numerical statement '99.9% of all peak position errors remain below 1 pixel' go beyond what is actually demonstrated. The paper simulates components separately—state preparation, sampling, and the predicted processing error—but does not report a full end-to-end circuit simulation or a hardware run. If the three error models are combined, the headline result is conditional on all three being correct. The authors should state explicitly which parts were simulated end-to-end and which were assembled from component models, and should qualify the abstract and Introduction accordingly.","section":"1 and 4"}],"minor_comments":[{"comment":"The thresholding iteration is stated to terminate within 4N^2 comparisons 'on our test data', and the greedy permutation heuristics are said to 'always yield' a valid circuit. These are empirical claims; a worst-case bound or a discussion of possible failure modes would be useful.","section":"3.1"},{"comment":"The complex-conjugated QFT ('QFT*') is a key step. Please give an explicit circuit identity or a concise derivation, since this operation is less standard than the inverse QFT.","section":"3.2"},{"comment":"The statement that code and data 'will be provided in an open-access format once this work is accepted' prevents independent verification. Please make the code and data available with the submission, at least to reviewers.","section":"Data Availability"},{"comment":"The 13×13 vector field in Fig. 3(b) is a decimation of the 121×121 field; please specify the decimation rule (e.g., every 10th vector) and indicate whether any smoothing was applied.","section":"4, Fig. 3"},{"comment":"The convention N=2^n, with an additional ancilla qubit per dimension for linear cross-correlation, should be stated in one place with a consistent notation. The current text switches between n and n+1 qubits per edge in a way that is easy to misread.","section":"Notation"},{"comment":"There is a typo in the Introduction: 'classical comuputing' should read 'classical computing'.","section":"Introduction, typo"},{"comment":"Several key derivations (state-preparation permutation details, the decoupling proof, and the sampling analysis) are deferred to Supplementary Notes that were not included with the manuscript. These should be made available to reviewers, or the essential steps should be moved into the main text.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The central concern is the unproved decoupling assumption in Section 3.4. If the Supplementary Note contains a rigorous proof, it should be moved into the main text or at least provided to reviewers; otherwise the paper's main numerical claims are unsupported. Also, the data-availability statement is an obstacle to reproducibility and should be addressed before acceptance. I do not see grounds for rejection, because the issue may be fixable with a proof or a targeted full-circuit simulation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a coherent proof-of-concept for an end-to-end quantum cross-correlation pipeline applied to PIV, and it deserves a serious referee. The genuinely new pieces are the zero-mean sparse state preparation (Section 3.1), the 2D QFT cross-correlation with complex conjugation (Section 3.2), and the contracted-amplification error model (Eq. 15). I checked the state-prep normalization and zero-mean sums, the cNOT-sorting construction for the correlation map, and the internal consistency of Eq. 15; they are fine. The authors are unusually candid: the Discussion explicitly says the goal is not to prove an asymptotic quantum speedup, acknowledges that binary input reduces to an N^2 pixel-pair problem classically, and names image reduction as the dominant error source. It is also not a fitting exercise — the amplification query counts are computed from the input spectra to hit a fixed 0.5 target, not matched to the known displacement.\n\nThe soft spots are concentrated in Section 3.4. Replacing S0 by S_ref on register B is the load-bearing step, and the paper does not prove the decoupling that Eq. (14) requires. The stress-test point lands: in this use case S_ref and S_G both mark |0>_B, so the iteration is U(I-2P_B0)U†(I-2P_B0), not the standard Grover reflection about U|0>. The statement that the register-A state remains unaffected is insufficient because the cNOTs entangle the registers and QFT_A does not commute with them. The paper defers the proof to a Supplementary Note and says only 'we find it also to hold if the amplification can be decoupled.' Since the gate-count saving and the sub-10^-2-pixel processing-error prediction both rest on that assumption, the end-to-end claim is conditional on it. That is the thing a referee should push on hardest — it is not a fatal flaw as long as the authors can supply the missing argument, but the argument is not in this manuscript.\n\nTwo lesser issues: the arXiv artifact ships no code or data, and the Supplementary Notes are referenced but not available, so the simulator runs cannot be independently checked; and the abstract's 'replace the classical computation' framing outruns the Discussion's own disclaimers. Both are fixable in revision.\n\nWorth a serious referee. Send it to review, and make the decoupling proof and the supplementary material conditions of acceptance. I would not cite the contracted-amplification claim in its current form, but the state preparation and correlation construction are solid enough to reuse.","headline":"Worth a serious referee: a coherent and candid end-to-end quantum PIV pipeline, but the contracted amplitude amplification it rests on is asserted rather than proven, so the end-to-end claim is conditional.","tokens_in":16002,"tokens_out":4515,"would_cite":false,"duration_ms":43239,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"An end-to-end quantum algorithm, QuPIV, performs Particle Image Velocimetry's displacement estimation via two-dimensional quantum Fourier cross-correlation and contracted amplitude amplification, reproducing a benchmark flow's velocity fiel","keywords":["quantum algorithm","particle image velocimetry","cross-correlation","quantum Fourier transform","amplitude amplification","quantum state preparation","sub-pixel displacement estimation","velocity field measurement"],"falsifier":"Run the complete QuPIV circuit at the largest size a full statevector simulator can handle (for example N=8, 12 qubits) and compare the measured amplified correlation amplitudes against the prediction of Eq. (15) using the same input spectra. A deviation larger than the Monte-Carlo sampling error—or a change in B-register target amplitudes when register A is measured—would show that the decoupling of registers A and B inside contracted amplification does not hold, breaking both the gate-reduction claim and the error model. A direct hardware test on a few qubits with the same comparison would s","tokens_in":14902,"feed_emoji":"🌊","tokens_out":6207,"duration_ms":62042,"temperature":0.7,"pith_summary":"This paper claims that the displacement-estimation step of Particle Image Velocimetry—the repeated cross-correlation of image pairs that yields local velocity vectors—can be run end to end on a quantum computer. The authors assemble a pipeline: sparse binary image preparation, a two-dimensional quantum Fourier cross-correlation with complex conjugation, a cNOT-based sorting step, and a contracted form of amplitude amplification that cuts the circuit depth by marking the ground state on only one of the two image registers. They validate it on experimental images of a rotating flow with known ground truth, reporting peak-position errors below 1 pixel for 99.9% of interrogation windows and predicted processing errors below 10^-2 pixels. If the claims hold, PIV's massive load of repeated correlations becomes a candidate quantum workload, and the contracted-amplification idea generalizes as a resource-saving tool for any amplitude-amplification circuit.","feed_headline":"Quantum algorithm maps PIV flow fields at sub-pixel error","feed_subtitle":"A full quantum pipeline from sparse image loading to peak sampling reproduces a benchmark rotating flow within 1 pixel.","key_machinery":"Contracted amplitude amplification, denoted S_ref, is the central mechanism: it replaces the 4n-qubit ground-state projector S0 with a projector that marks only the ground state of the B register, so the forward and inverse base circuits need not re-prepare register A in each query. The formal identity carries the argument: because the sorting cNOTs permute elements on B only, the amplified state splits into 2^kappa independent local problems with local amplitudes c_k sin((2r+1)theta_k), where the c_k reduce to the spectral magnitudes |A_hat_k| and |B_hat_k|. This gives an explicit error model for the processing stage and is what makes the end-to-end circuit depth tractable. The supporting p","core_discovery":"The central claim is that the cross-correlation map underlying PIV can be produced by a quantum circuit whose only nonlinear step—the component-wise spectral product—is handled by loading the two images into separate registers and using the tensor-product structure itself. After two-dimensional QFTs and complex conjugation of register B's spectrum, the circuit sorts the pairwise products with cNOTs so that an inverse QFT on register A yields the correlation map whenever register B is in |0>. The paper's main theoretical contribution is contracted amplitude amplification: replacing the full ground-state projector over all qubits with a projector S_ref on the B register alone, justified by the","pith_inferences":["Because the binary sparse input removes the classical FFT advantage (the paper itself notes all pair-wise particle shifts can be enumerated classically), the practical payoff of QuPIV would have to come from a constant-factor speedup over many repeated interrogation-window correlations, not from asymptotic scaling—a hardware benchmark would need to show this.","The decoupling that justifies contracted amplification is an architectural claim; on real devices, residual crosstalk or imperfect cNOTs could reintroduce A-B correlations. A small hardware experiment comparing Eq. (15) predictions with measured peak probabilities would be the natural next test.","The same pattern—sparse encoding, Fourier-domain correlation, and sampling only the peak—carries over to other image-registration and template-matching problems whose output is a single displacement or match location, not a full correlation map.","Contracting S_G instead of S_0 is left open; if a state preparation ended with local row/column operations, the paper's own reasoning suggests the QFTs and cNOTs inside U†S_GU could cancel, further reducing circuit depth."],"forward_implications":["QuPIV recovers the velocity field of a rotating benchmark flow in agreement with both the analytical solution and classical correlation-based PIV, with no systematic deviations from the ground truth.","With 15 contracted amplification queries and 500 samples per correlation map, 99.9% of all peak-position errors remain below 1 pixel; the predicted processing error is below 10^-2 pixels for the median and 5–95% range.","Larger interrogation windows (edge length 128 pixels or more) are more favorable: half as many active pixels suffice for sub-pixel accuracy, about 100 measurement shots are enough, and errors under contracted amplification remain consistently low.","The base-circuit success probability scales as 1/N^2, so the optimal number of amplification queries grows linearly with image edge length, and a query count found for one image is reusable for similar subsequent images.","The contracted-amplification error model (built from Eq. 14 and Eq. 15) predicts processing errors without simulating deep full circuits, which becomes important as image sizes grow beyond classical simulation reach."],"fun_headline_variants":["Quantum PIV maps flows with sub-pixel accuracy","QuPIV: quantum algorithm for PIV flow vectors","Quantum Fourier transforms enable end-to-end PIV","Quantum circuit computes velocity fields from images"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that contracting the ground-state projector to the B register keeps the two registers effectively decoupled during amplification, so the local-angle decomposition and the spectrum-based error prediction stay valid; if residual correlations between registers A and B persist, both the gate savings and the error estimates collapse.","fun_headline_variants_meta":{"raw":{"variants":["Quantum PIV maps flows with sub-pixel accuracy","QuPIV: quantum algorithm for PIV flow vectors","Quantum Fourier transforms enable end-to-end PIV","Quantum circuit computes velocity fields from images"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000339,"raw_usage":{"total_tokens":1661,"prompt_tokens":649,"completion_tokens":1012,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":393,"completion_tokens_details":{"reasoning_tokens":952}},"tokens_in":393,"tokens_out":1012,"duration_ms":9779,"temperature":1.0,"reasoning_tokens":952,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T04:37:55.828291+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the complete QuPIV circuit at the largest size a full statevector simulator can handle (for example N=8, 12 qubits) and compare the measured amplified correlation amplitudes against the prediction of Eq. (15) using the same input spectra. A deviation larger than the Monte-Carlo sampling error—or a change in B-register target amplitudes when register A is measured—would show that the decoupling of registers A and B inside contracted amplification does not hold, breaking both the gate-reduction claim and the error model. A direct hardware test on a few qubits with the same comparison would s","supporting_citations":[],"review_version":1}