{"id":"b6ec125f-af75-44d2-8176-d39359a4a894","arxiv_id":"2504.13033","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A statevector-emulated HHL solver applied to Carleman-linearized lattice Boltzmann equations achieves about 10^-3 fidelity on tiny 2D benchmarks, but physical accuracy is dominated by the Carleman truncation error, not the quantum solver.","lead":"An industry lab emulated a quantum solver for the lattice Boltzmann equations on small 2D flows and reported that the quantum part of the pipeline reproduces the linearized equations to about one part in a thousand. The catch is that the larger error comes from the classical linearization step, so the headline accuracy does not measure the true gap to the physical flow.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Spectral transfer from Nx=4 is end-to-end tested only for lattices divisible by 4 (Nx=8,12,20); non-commensurate sizes may break the eigenvalue-pre-evaluation claim.","rationale":"I agree with the reader that spectral transfer is the weakest load-bearing assumption, and I sharpen it: the empirical support is commensurate-only. The paper's own conclusion (Sec. VI) concedes that the Carleman-vs-LBM error is about 5%, larger than the HHL error, so the abstract's 10^-3 should be read as solver fidelity; that is a framing issue, not a reason to reject. The internal HHL emulation uses exact statevector Hamiltonian simulation, so the single- and multi-step fidelity numbers are plausible as solver benchmarks. What would have to be true for the central claim to hold is that the Nx=4 spectrum approximates the large-lattice spectrum well enough for the HHL rotation angles in Eqs. (29)-(30). This is tested only for Nx divisible by 4, where Fourier-mode embedding makes the comparison favorable. A non-commensurate test, especially with an initial condition exciting modes absent from the small lattice, would settle whether the spectral-transfer claim generalizes. Until then, the reader's CONDITIONAL verdict remains appropriate; I would keep it unchanged.","tokens_in":20283,"tokens_out":16057,"duration_ms":164240,"concrete_test":"Repeat the Sec. V E experiment for Nx=10 or Nx=18 (not divisible by 4) at Nt=1 and Nt=7, using both the smooth Kolmogorov initial condition and a high-wavenumber initial condition whose Fourier support is mostly absent from the Nx=4 spectrum. Compute the HHL fidelity residual and success probability using the Nx=4 spectrum versus the full spectrum. Also compute the minimal matching distance between the Nx=4 eigenvalues and the target eigenvalues; if any target eigenvalue with appreciable |<u|b>| weight lies farther than the HHL rotation quantization step (4*lambda_max/2^nc, about 0.027 for nc=7) from the Nx=4 set, or if the fidelity residual exceeds 1e-3, the spectral-transfer claim does not generalize.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central new claim is that the eigenspectrum of a small Nx=4 lattice can be reused for larger lattices in HHL, mitigating eigenvalue pre-evaluation (Secs. V B and V E). The HHL-level evidence for this approximation is Fig. 12 (Nx=20, single step) and Fig. 13 (Nx=12, three steps); every lattice-size scan in Secs. V C and V D uses Nx=4,8,12,16,20. All of these are multiples of the reference size 4. For periodic and bounce-back boundary conditions the streaming operator is translation-invariant, so its eigenmodes are labeled by wavenumbers k=2pi*m/Nx; the Nx=4 modes are embedded in the target spectrum only when Nx is a multiple of 4. Thus the tested cases do not exercise the approximation for non-commensurate lattices such as Nx=6,10,14,18. The zeta metric in Fig. 7 is a binned count of missing bins, not a per-eigenvalue error, and it is not paired with HHL fidelity tests for those sizes. If the missing eigenvalues carry nontrivial weight in b, which the smooth Kolmogorov initial condition may suppress, the rotation angles in Eq. (29) differ and the reported 'negligible' residual could degrade. No error bound or proof is given for spectral transfer, so the central claim rests on interpolation over commensurate toy sizes.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents a statevector-emulated hybrid quantum-classical pipeline for two-dimensional D2Q9 lattice Boltzmann simulations. The nonlinear collision step is linearized with a first-order Carleman expansion, and the resulting multi-step linear system of the form \\tilde{A} x = b (Eqs. 21-27) is solved with HHL, using exact spectra of the linear operator A. The authors benchmark three cases (periodic, bounce-back, and lid-driven cavity), quantify the Carleman truncation error against classical LBM (Sec. V A), and report HHL fidelity and success probability as functions of clock qubits, lattice size, and time steps (Secs. V C-D). They further propose using the eigenspectrum of an Nx=4 lattice for larger Nx, claiming negligible fidelity loss (Sec. V E). The headline results are median error fidelities on the order of 10^-3 and success probabilities described as sufficient for practical quantum state sampling.","tokens_in":20537,"tokens_out":6073,"duration_ms":58669,"significance":"The paper is a useful engineering-oriented numerical study: it separates Carleman linearization error from HHL implementation error, tests several boundary conditions, and provides explicit spectral data. If the spectral-transfer result holds for general lattice sizes, it would be a concrete step toward reducing HHL eigenvalue-estimation cost, a known bottleneck. However, the reported 10^-3 fidelity measures agreement with the exact solution of the linear system HHL is designed to solve, not with the physical LBM flow; the physical-model error is the Carleman RMSE, which reaches about 5% (Figs. 4-5). The spectral-transfer claim is supported only by commensurate lattices and lacks an error bound. The paper is honest about the Carleman error and about using exact statevector emulation, but the abstract's wording overstates the level of validation.","major_comments":[{"comment":"The 'error fidelity' in Eq. (34) compares the HHL output with the exact solution of the Carleman linear system A x = b, not with the physical LBM solution. The abstract's claim of 'median error fidelities on the order of 10^-3' is therefore a statement about circuit-level consistency, while the physical accuracy is governed by the Carleman truncation error reported separately in Sec. V A, which reaches about 5% (Figs. 4 and 5). Please state this distinction explicitly in the abstract and conclusions, and avoid presenting the internal fidelity as evidence of physical validity.","section":"Abstract; Sec. V C, Eq. (34)"},{"comment":"The spectral-transfer claim is tested only for lattices with Nx divisible by 4 (Nx=8, 12, 20; Secs. V B and V E). For the periodic and bounce-back streaming operators the eigenmodes are labeled by wavenumbers 2\\pi m/Nx, so the Nx=4 modes are embedded in the target spectrum only when Nx is a multiple of 4. Non-commensurate sizes such as Nx=6, 10, 14, 18 are never tested, and the zeta metric in Eq. (33) is a binned count of missing bins rather than a per-eigenvalue error weighted by the initial vector b. An explicit test on non-commensurate lattices, or an error bound for the spectrum-reuse approximation, is needed before the claimed mitigation of eigenvalue pre-evaluation can be regarded as established.","section":"Sec. V B, Sec. V E, Eq. (33), Figs. 12-13"},{"comment":"The statement that the observed success probabilities are 'sufficient for practical quantum state sampling' is not backed by a sampling analysis. With correct-state probabilities around 10^-3, a single accepted sample requires roughly 10^3 measurement repetitions, and recovering macroscopic flow fields would require many such samples; the paper does not provide a shot budget or discuss amplitude amplification. Please either justify the 'sufficient' wording quantitatively or soften it to a statement about the raw success probability.","section":"Sec. V C, Fig. 8; abstract"}],"minor_comments":[{"comment":"The caption of Fig. 6 lists Nx=16 while the legend shows Nx=4, 8, 12; the caption should be corrected to match the plotted data.","section":"Sec. V B, Fig. 6"},{"comment":"Equation (32) defines a relative error 1 - f_Car/f_LBM, yet the text calls it RMSE; rename it relative RMSE or define a true RMSE.","section":"Sec. V A, Eq. (32)"},{"comment":"The text says 'Figure 12 summarizes the results' for the three-step Nx=12 evolution, but this appears to refer to Fig. 13, which is the multi-step figure; please fix the cross-reference.","section":"Sec. V E"},{"comment":"The zeta metric depends on the histogram bin width; please state the binning procedure and show how zeta changes with bin width.","section":"Sec. V B, Eq. (33)"},{"comment":"Typos and wording issues include 'staisfy' (Sec. V B), 'quibts' (Sec. V C), 'spcecifically' (Sec. IV), 'employes' (Sec. III), 'I In contrast' (Sec. V C), and the duplicated phrase in Sec. II ('cannot be directly implemented on a quantum computer, preventing its direct implementation').","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper's central novelty is an empirical spectral-transfer observation; the main risk is that this observation may be an artifact of commensurate lattice sizes and a binned spectrum comparison. I would encourage the editor to treat the abstract's 'validity' claim with caution until the fidelity metric is disambiguated and the spectral transfer is tested on non-commensurate sizes."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Reading this paper you should keep one distinction in hand: the 10^-3 fidelity in the abstract measures HHL against the exact solution of the same linear system HHL is built to solve. That is a circuit-level check, and it passes. The physical modeling error — Carleman truncation versus LBM — is reported separately and runs up to about 5%. The body is candid about this split; the abstract is not. \"Confirm the validity of the approach\" glosses over the difference between solving a linear system accurately and modeling a flow accurately.\n\nWhat is actually new: the benchmark set itself (periodic, bounce-back, and lid-driven cavity boundary conditions on tiny 2D lattices), the clock-qubit versus success-probability trade-off scan, and the spectral-transfer observation. The eigenspectrum of the A matrix for Nx=4 looks nearly identical to those for Nx=8 and Nx=12, and substituting the small spectrum into the rotation-angle computation for Nx=12 and Nx=20 changes fidelity by less than 10^-4. If that trick holds at scale, it genuinely eases one of HHL's practical bottlenecks. The HHL numbers are internally consistent, the multi-time-step metrics are disclosed (the known t=t0 states are discarded, which is defensible), and the conclusion explicitly separates the 5% Carleman error from the sub-3% HHL error.\n\nThe soft spots, in proportion. The spectral-transfer claim is the load-bearing new result, and the evidence is thinner than the text suggests. Every lattice tested — 4, 8, 12, 16, 20 — is a multiple of the reference size 4. That is exactly the set for which the Nx=4 streaming wavenumbers embed in the larger spectrum. Non-commensurate sizes like 6, 10, 14 introduce wavenumbers the Nx=4 spectrum never sees, and the paper tests none of them. The zeta metric is a binned count of missing histogram bins, not a per-eigenvalue error, and it is not paired with HHL fidelity tests on the sizes where the approximation would actually strain. No proof or error bound is offered, so the central claim rests on interpolation over commensurate toy sizes. That is a real gap, though an honest one: the paper presents the transfer as an empirical observation, not a theorem. Two smaller items: no code or data released, which limits how much others can build on the benchmark, and the title promises \"industrial CFD\" while the content is 4-to-20-point lattices.\n\nWho this is for: people working on quantum linear solvers for PDEs and the QCFD subfield. The paper deserves a serious referee. The right referee would push for a reframed abstract, a test on non-commensurate lattice sizes, and release of the simulation data. With those changes this becomes a solid benchmark paper. I would send it out rather than desk-reject it.","headline":"A careful statevector-emulation benchmark of the Carleman-LBM plus HHL pipeline on three toy 2D flows; the 10^-3 fidelity is a circuit check, the Carleman error is ~5%, and the spectral-transfer claim is real but only validated on lattices commensurate with 4.","tokens_in":21140,"tokens_out":8034,"would_cite":false,"duration_ms":77782,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","76M28"],"pacs":["03.67.Ac","47.11.-j"],"model":"deepseek-v4-flash","headline":"The paper claims that an HHL quantum linear solver applied to a Carleman-linearized lattice Boltzmann system solves benchmark two-dimensional flows with median error fidelities near $10^{-3}$, and that small-lattice eigenvalue spectra can…","keywords":["quantum computational fluid dynamics","Lattice Boltzmann method","Carleman linearization","HHL algorithm","state-vector emulation","eigenvalue pre-evaluation","lid-driven cavity","bounce-back boundary condition"],"falsifier":"Run the same pipeline on a 32-by-32 or larger lattice (or with more than seven time steps), compute the true spectrum of $A$ for that lattice, and compare HHL solutions obtained with the true spectrum against those obtained with the 4-by-4 spectrum; if the bin-wise discrepancy $\\zeta$ grows beyond a few percent or the fidelity residual rises above $10^{-4}$, the spectral-transfer shortcut fails. The paper's own $\\zeta$ metric and residual plots make this a direct check.","tokens_in":19990,"feed_emoji":"🌊","tokens_out":10551,"duration_ms":101561,"temperature":0.7,"pith_summary":"This paper sets out to show that the Lattice Boltzmann equations used in industrial computational fluid dynamics can be solved by a hybrid quantum-classical pipeline: a Carleman expansion linearizes the nonlinear collision step, several time steps are packed into one linear system, and the HHL quantum algorithm solves that system. The evidence comes from state-vector emulations of three benchmark flows — an open periodic flow, a closed box with bounce-back walls, and a lid-driven cavity — for which the paper reports median error fidelities on the order of $10^{-3}$ and success probabilities high enough to sample the answer in practice. The paper also argues that the eigenvalue spectrum of the linear system for a small lattice closely matches the spectrum for larger lattices, so the costly eigenvalue pre-evaluation of HHL can be done once on a tiny lattice and reused. If this holds, the main practical bottleneck of HHL would be softened for a structured, industrially relevant problem class, and quantum CFD would have a concrete numerical starting point.","feed_headline":"Quantum fluid pipeline hits 1e-3 median error","feed_subtitle":"Carleman-linearized lattice Boltzmann flows solved via HHL; 4-by-4 spectra stand in for larger lattices.","key_machinery":"The load-bearing object is the matrix $A$ that encodes the Carleman-linearized lattice Boltzmann evolution. Carleman linearization introduces auxiliary variables such as $g_{ij}(t,n,m)=f_i(t,n)f_j(t,m)$, turning the nonlinear BGK collision step into a linear update $\\varphi(t+dt)=C\\varphi(t)$; stacking $N_t$ such updates gives a block-bidiagonal linear system $\\tilde A x=b$ whose solution contains the whole time history, and HHL solves the Hermitian form $A$ obtained by placing $\\tilde A$ and its adjoint off-diagonal. The spectral-transfer argument is about this matrix: with the number of time steps fixed, the positive eigenvalue density of $A$ is nearly independent of lattice size, so a spectrum obtained once on a 4-by-4 lattice calibrates the controlled rotations of HHL for larger lattices.","core_discovery":"On the paper's own terms, the central discovery is that the Carleman–HHL combination works numerically for all three boundary-condition classes. For single-step evolution the HHL solution matches the exact Carleman solution with fidelity error typically between $10^{-4}$ and $10^{-2}$ depending on clock qubits and relaxation frequency; for multi-step evolution the median error stays near $10^{-3}$ while the success probability grows. The spectral-transfer observation is the second finding: for a fixed number of time steps, the positive part of the spectrum of the matrix $A$ is nearly identical for $N_x = 4,8,12$ after one step, and still within a few percent bin-wise discrepancy after seven steps. Using the 4-by-4 spectrum in place of the true 12-by-12 or 20-by-20 spectrum leaves the fidelity error nearly unchanged, with residuals below $10^{-4}$. This is presented as evidence that eigenvalue pre-evaluation, normally one of HHL's serious bottlenecks, can be amortized over a family of lattice sizes.","pith_inferences":["The spectral transfer result suggests a production workflow the authors do not spell out: eigenvalue spectra could be precomputed once per boundary-condition class and Reynolds number, making HHL's eigen-decomposition an amortized pre-processing cost rather than a per-simulation cost.","The near size-independence of the spectrum is plausibly inherited from the local block structure of the Carleman matrix, in which the same small block is repeated across lattice sites; this points toward a rigorous bound on the discrepancy $\\zeta$ in terms of $N_x$ and boundary-condition effects, which the paper does not provide.","The success probabilities are obtained under exact state-vector emulation; the CNOT-count estimates in the appendix indicate that circuit depth, not the spectral shortcut, is likely to decide practicality on near-term quantum hardware, so the next test should be noise-aware."],"forward_implications":["First-order Carleman truncation is enough for the tested flows: it matches the second-order accuracy in most parameter regimes while costing far less, so the cheaper pipeline can be used in practice.","Spectral transfer directly attacks one of HHL's bottlenecks: a spectrum computed on a 4-by-4 lattice can be reused for 12-by-12 and 20-by-20 lattices with fidelity residuals below $10^{-4}$.","Encoding more time steps into a single HHL solve is the right strategy: success probability rises with the number of encoded time steps while fidelity stays near $10^{-3}$, so reusing the same circuit for long times is worse than adding time-qubits.","The dominant error in the combined method is the Carleman truncation itself (about 5%) rather than the HHL solve (under 3%), so raising the truncation order is the next accuracy lever, and its qubit cost grows only linearly in the Carleman order."],"supporting_citations":[{"why":"Supplies the multi-time-step linear-system form $\\tilde A x = b$ that encodes the whole time history in one HHL solve.","marker":"[16]"},{"why":"Established the quantum Carleman lattice Boltzmann simulation approach that this work builds on.","marker":"[21]"},{"why":"Provides the original Carleman linearization scheme used to treat the nonlinear collision term.","marker":"[27]"},{"why":"Provides the Carleman-lattice-Boltzmann quantum circuit with matrix access oracles that informs the circuit implementation.","marker":"[29]"},{"why":"Original HHL algorithm whose controlled rotations and eigenvalue calibration the paper implements.","marker":"[33]"},{"why":"BGK collision operator is the nonlinear term that Carleman linearization must handle.","marker":"[36]"},{"why":"Defines the RMSE used to assess Carleman error and lays out the Lattice Boltzmann-Carleman quantum algorithm.","marker":"[37]"}],"fun_headline_variants":["Quantum fluid solver hits 1e-3 error","Carleman-HHL flow: 1e-3 error across BCs","Small lattices predict big ones in quantum flow","Quantum LBM: 1e-3 error, spectral shortcut"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the eigenvalue pattern of a tiny 4-by-4 lattice stays representative of larger lattices; the paper verifies this numerically only up to a 20-by-20 lattice and seven time steps, and it supplies no proof or error bound for the transfer.","fun_headline_variants_meta":{"raw":{"variants":["Quantum fluid solver hits 1e-3 error","Carleman-HHL flow: 1e-3 error across BCs","Small lattices predict big ones in quantum flow","Quantum LBM: 1e-3 error, spectral shortcut"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000338,"raw_usage":{"total_tokens":1871,"prompt_tokens":952,"completion_tokens":919,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":568,"completion_tokens_details":{"reasoning_tokens":848}},"tokens_in":568,"tokens_out":919,"duration_ms":9345,"temperature":1.0,"reasoning_tokens":848,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T12:16:59.196687+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same pipeline on a 32-by-32 or larger lattice (or with more than seven time steps), compute the true spectrum of $A$ for that lattice, and compare HHL solutions obtained with the true spectrum against those obtained with the 4-by-4 spectrum; if the bin-wise discrepancy $\\zeta$ grows beyond a few percent or the fidelity residual rises above $10^{-4}$, the spectral-transfer shortcut fails. The paper's own $\\zeta$ metric and residual plots make this a direct check.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the multi-time-step linear-system form $\\tilde A x = b$ that encodes the whole time history in one HHL solve."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Established the quantum Carleman lattice Boltzmann simulation approach that this work builds on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Carleman-lattice-Boltzmann quantum circuit with matrix access oracles that informs the circuit implementation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"BGK collision operator is the nonlinear term that Carleman linearization must handle."}],"review_version":1}