{"id":"d8fb708c-7666-4c88-aa84-eb9c17d54288","arxiv_id":"2607.24972","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":4,"one_line_summary":"Haar-averaged Rodeo responses equal the density of states convolved with a kernel fixed by the evolution-time distribution, yielding a single-ancilla quantum DoS estimator.","lead":"Averaging the Rodeo algorithm over random quantum states reconstructs a system's density of states as a smoothed kernel estimate, without diagonalizing the Hamiltonian. This gives near-term quantum hardware a direct analogue of the classical kernel polynomial method for spectral thermodynamics.","discovery_kind":"new_method","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"The core derivation (Eqs. 7–20, App. A) is algebraically sound; the load-bearing gap is empirical: typicality-driven variance suppression and robustness to preparable state ensembles are untested beyond N_s = 243, exactly the regime where the method claims advantage.","rationale":"The reader identified the correct load-bearing assumption — that Haar-like input states are preparable at scales where typicality suppression matters — and I confirm it is the right target rather than a manufactured one. My independent re-derivation of the estimator (kernel structure in Eq. 8, spectral kernel in Eq. 12, Haar average in Eq. 18, and the fourth-moment variance calculation in Appendix A) found no algebraic or statistical error, so there is no internal-inconsistency basis for moving the verdict. The reader's CONDITIONAL verdict already prices in exactly the right conditions: artifact release, validation at sizes beyond trivial enumeration, and quantification of approximate-design effects. I add two refinements that slightly strengthen the paper's position and one that sharpens the caveat. On the strengthening side: unbiasedness requires only that the state ensemble average to the maximally mixed state, a much weaker condition than exact Haar — random product states already satisfy it — so the \"unbiased estimator\" headline is safer than the paper's own limitations paragraph implies; only the variance constant is at stake. On the caveat side: the typicality suppression is an absolute-error statement and does not protect relative error in DoS tails, which is where low-temperature thermodynamics (Fig. 5) draws its weight; the flat-histogram extension the author proposes is not optional garnish but necessary for that use case. Neither refinement changes the verdict calculus: the mathematics is verified, the numerics agree at small N, and the open items are scope and evidence rather than correctness. Hence UNCHANGED, with the proposed scaling study as the single most decisive next check — it is cheap, requires no hardware, and directly settles whether the method's advertised advantage survives contact with the regime it is designed for.","tokens_in":16207,"tokens_out":5173,"duration_ms":182659,"concrete_test":"Purely classical emulation (no hardware needed). For the TFIM at N = 6, 8, 10, 12, 14 (N_s up to 16384, still exactly diagonalizable for ground truth), reconstruct the DoS at fixed R = 10 and N_t = 1000 using (i) exact Haar-random states and (ii) random product states / one-layer random local circuits. For each N and each ensemble, measure (a) the bias |E[ĝ] − (g∗G)| at several energies, and (b) the empirical variance of ĝ versus the Appendix-A prediction. If Haar variance scales as 2^{−N}, product-state variance scales as ~3^{−N} (or another exponential base) with no detectable bias, the typicality argument extends to preparable states and the central claim is secured in the relevant regime; if bias appears or variance flattens with N for shallow ensembles, the practical version of the claim needs the quantified design-depth analysis the paper defers.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I checked the central construction line by line and it holds. The ancilla expectation (Eq. 7/8) follows from summing the QFT-interfered amplitudes: the clock-operator expectation picks out n−m=1 (mod d_a) pairs, giving exactly the (d_a−1)/d_a e^{−i∆t} + (1/d_a)e^{i(d_a−1)∆t} kernel. Averaging over p(t) via the characteristic function (Eqs. 10–12) and over Haar weights (Eqs. 17–18) correctly yields E[R] = (1/N_s)Tr[e^{−σ²(E−H)²/2}] = (g∗G)(E). The variance derivation in Appendix A uses the correct Haar fourth moment and the resulting Var(R) = Σ(G_i−Ḡ)²/(N_s(N_s+1)) = O(N_s^{−1}) is right. So the claim \"unbiased estimator of g∗G\" is not at risk internally. The load-bearing weakness is the one the reader flagged, and I would sharpen it rather than replace it. First, unbiasedness is more robust than the paper's framing suggests: any input ensemble with E[|ψ⟩⟨ψ|] = I/N_s — including random product states, since local-Haar product states average to the maximally mixed state — yields an unbiased estimator. What degrades with shallow, device-preparable ensembles is the variance: for random product states the participation ratio is ~3^N rather than 2^N, so the exponential suppression survives but with a smaller base, and this has not been quantified. Second, the validation regime (N_s = 32 and 243, R = 10/36 states) is precisely where classical enumeration is trivial and where typicality asymptotics say nothing; the advertised advantage (\"statistical error decreases as the system size increases\") is extrapolated, not demonstrated. Third, and secondary: the O(N_s^{−1/2}) suppression is absolute, not relative per-energy — in DoS tails where Ḡ(E) is exponentially small (the region dominating low-T thermodynamics), relative error scales like (N_s·Ḡ(E))^{−1/2} and the typicality guarantee is vacuous once fewer than O(1) levels sit within a kernel width. The paper implicitly concedes this via the flat-histogram proposal, but Fig. 5's low-T agreement is at N=5, where tail counts O(","agreement_with_reader":"agree"},"referee_report":{"model":"moonshotai/kimi-k3","summary":"The manuscript shows that the Rodeo eigenvalue-location algorithm, when its response is averaged over Haar-random input states, yields an unbiased estimator of the density of states convolved with a spectral kernel G that is fixed entirely by the characteristic function of the temporal sampling distribution p(t): E[R(E,ψ)] = (g∗G)(E) (Eqs. 11–20). Random states thus play the role of stochastic trace estimation and p(t) the role of the KPM damping kernel, a correspondence the author makes explicit in a \"KPM dictionary\" (Table I) and a window-function table (Table II). Appendix A derives Var(R) = Σ_i(G_i − Ḡ)²/[N_s(N_s+1)] = O(N_s^{-1}) from the Haar fourth moment, and Appendix B gives a closed-form Hann-kernel expression with |E|^{-3} side-lobe decay. The method is validated against exact diagonalization on the N=5 transverse-field Ising ring (N_s=32) and the N=5 spin-1 ring (N_s=243), including peak-integrated degeneracies (Table III) and thermodynamic quantities (Fig. 5).","tokens_in":16733,"tokens_out":3507,"duration_ms":110245,"significance":"If the construction holds up at scale, it gives a genuinely simple quantum DoS estimator: the standard single-ancilla Rodeo circuit is used unmodified, the reconstruction kernel is parameter-free in the sense of being fixed once p(t) is chosen, and the estimator is provably unbiased with an exactly computed variance (Eq. A12). The explicit dictionary between classical window design and temporal sampling laws (Tables I–II, Appendix B) is a useful methodological contribution that imports decades of signal-processing knowledge into quantum spectral estimation, and the closed-form Hann kernel with its |E|^{-3} leakage floor is a concrete, checkable result. The numerical agreement with exact diagonalization, including the specific heat (which is variance-sensitive), is encouraging. The work is a clean conceptual unification rather than a demonstrated computational advantage: the central algebraic claims are sound, but the practical case for the regime where the method is \"aimed\" (Hilbert spaces beyond classical enumeration) rests on assumptions about state preparation and evolution-time cost that the manuscript does not yet quantify.","major_comments":[{"comment":"The typicality scaling σ(ĝ)=O(N_s^{-1/2}) (Eq. 21, Eq. A14) is derived from exact Haar second and fourth moments (Eq. A3), yet the paper's stated target is the regime where exact Haar states are exponentially costly to prepare, and the Conclusion asserts that 'a (pseudo)random product of local rotations' will suffice, citing classical KPM practice. This citation does not settle the question: classical KPM random vectors have independent entries (Gaussian or ±1), which is not the same ensemble as shallow-circuit or random-product states on a quantum register. Two things can be stated precisely and would substantially strengthen the paper. (i) Unbiasedness requires only E[|ψ⟩⟨ψ|] = I/N_s, which holds exactly for random product states with local Haar-random (or even discrete 1-design) single-site rotations — so the estimator's mean is safe under far weaker assumptions than the text suggests","section":"Sec. II.C, Appendix A, Sec. IV (limitations paragraph)"},{"comment":"There is an unresolved internal tension between spectral resolution and simulation cost that bears directly on the method's viability. Narrowing the Gaussian kernel requires large σ (Eq. 16), and the sampling law p(t)=N(0,σ) has unbounded support, so typical draws have |t|~σ and tails to several σ. With the first-order Trotter choice r=t²/δ (Eq. 27), the σ=200 refinement in Fig. 3 implies draws with t~200–600, i.e. r up to ~10^6 Trotter steps per controlled evolution at δ=0.05 — many orders of magnitude beyond any near-term hardware, and the dominant cost of the whole protocol. The paper acknowledges the trade-off qualitatively but never aggregates it: there is no estimate of total evolution time (or total controlled-U count) per reconstructed DoS point, and no comparison of that cost against the sparse matvec count of classical KPM for the same resolution. Since the abstract and Conclus","section":"Secs. II.B, II.D.2, III.B (Fig. 3), Eq. (27)"},{"comment":"All validation is at N_s=32 and N_s=243, where exact diagonalization is trivial and where the O(N_s^{-1/2}) typicality asymptotics say nothing — with R=10–36 states the observed errors are dominated by the R^{-1/2} factor, not by the dimension suppression that is the paper's headline mechanism. This is acknowledged only obliquely. The paper would be considerably more convincing with one scaling demonstration: a sequence of system sizes (e.g. N=4,6,8,10 for the spin-1/2 TFIM, still classically checkable) showing the predicted N_s^{-1/2} decay of the single-state variance and the quality of the R-averaged estimator as N_s grows. This is cheap classically, directly tests Eq. A12 rather than merely illustrating the estimator, and would substantiate the claim that the method improves precisely where classical enumeration becomes prohibitive.","section":"Sec. III (Figs. 2–5, Table III)"}],"minor_comments":[{"comment":"The definition 'N_s = d^N_s' (rendered as N s =d N s) is typographically ambiguous; please write N_s = d_s^N explicitly.","section":"Sec. II.A, Eq. (6)"},{"comment":"The symbol σ is overloaded: it is the width of the Gaussian sampling law (Eq. 13) and is reused in Eq. (23) for the total standard deviation of the estimator. Please use a distinct symbol for the latter.","section":"Sec. II.D.1, Eq. (23)"},{"comment":"r = t²/δ is presented as a definition ('we consider the time dependence...') rather than a consequence of the first-order Trotter error bound; please state the assumption (norm of the commutator sum absorbed into δ, or cite the standard scaling, e.g. Ref. [27]) and justify the choice δ=0.05 used throughout Sec. III.","section":"Sec. II.D.2, Eq. (27)"},{"comment":"Entries such as ĝ=0.01(1) imply a ~100% relative uncertainty, yet a relative deviation ∆rg=0.26 is quoted to two digits without comment. Please clarify how the uncertainties in parentheses were propagated and whether the ∆rg values are meaningful given those error bars.","section":"Table III"},{"comment":"The phase-shift gate is labeled P(ϕ) in the figure but defined as P(E,t) in Eq. (3); please harmonize the notation.","section":"Fig. 1 vs. Eq. (3)"},{"comment":"The Gaussian is introduced with mean µ which is then set to zero; consider defining the zero-mean law directly and mentioning the µ≠0 phase issue in one sentence.","section":"Sec. II.B, Eqs. (13)–(14)"},{"comment":"Typo: 'RESUL TS' in the section heading.","section":"Sec. III header"},{"comment":"The repository is listed as 'in preparation'. Given that the numerical comparisons are a central part of the validation, the code and raw data should be deposited (with a citable DOI) before publication.","section":"Data Availability"},{"comment":"The 0.1-width binning of exact eigenvalues is described as 'for visualization purposes only', but the red reference curve/markers in Fig. 4 should be explicitly identified as binned in the caption so readers do not mistake bin height for degeneracy.","section":"Sec. III.C, Fig. 4"}],"recommendation":"major_revision","confidential_remarks":"The derivation is single-author and leans heavily on the author's own Rodeo-kernel papers [19–21]; in particular the single-shot response formula and the ancilla-qudit construction are imported from Ref. [21], which is itself a 2026 arXiv preprint (arXiv:2603.16049) that does not appear to be peer-reviewed yet. The present manuscript is internally self-contained enough that this is not disqualifying, but the editor may wish to confirm that [21] is publicly available in a stable form. The LLM-use disclosure in the acknowledgments is appropriately specific. Scope fit for a quantum-information journal is good; the main question is whether the revisions requested (preparable-ensemble variance, resource accounting, scaling numerics) are completed, since they are what stands between this being a neat observation and a demonstrated method."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The one thing worth knowing is that averaging the standard Rodeo response over Haar (or any 1-design) inputs really is just stochastic-trace estimation of the DoS smoothed by the characteristic function of p(t). That identification, the uncertainty formulas, and the window-function dictionary are the new content. Everything else is Rodeo plus classical KPM lore.\n\nThe derivation is clean. Eqs. 7–12 and 17–20 do what they claim: clock expectation gives the two-tone kernel, time average gives G = Φ-determined filter, Haar second moment turns the sum into Tr[G(E−H)]/N_s. Appendix A’s fourth-moment variance is correct and gives the O(N_s^{−1/2}) typicality scaling. The KPM table and the Hann/sinc appendix are useful and not just decoration. Small-system checks (TFIM and spin-1, N=5) sit on top of exact diagonalization and look honest; thermodynamics from the reconstructed DoS match within the reported bars.\n\nSoft spots, in proportion. Validation never leaves the regime where classical enumeration is trivial (N_s ≤ 243). The advertised “error falls as Hilbert space grows” is therefore an extrapolation, not a demonstration. Exact Haar is exponentially costly; the paper flags this and defers approximate designs, which is the right call, but it means the method’s practical edge is still open. Unbiasedness actually survives any ensemble with E[|ψ⟩⟨ψ|] = I/N_s (product states included); what changes is the variance base, and that is not quantified. Trotter error is discussed but not folded into the plotted uncertainties, and code/data are “in preparation.” Relative error in the spectral tails is also weaker than the absolute typicality bound suggests—the flat-histogram remark in the conclusion is the honest acknowledgment.\n\nNone of that breaks the central claim. This is a solid methods note for people already using Rodeo or thinking about near-term spectral density. It deserves a serious referee, ideally with a request for released artifacts and at least one larger-N or approximate-state check. I would bring it to reading group as a short methods discussion, not as a centerpiece.","headline":"Clean KPM-style reading of Rodeo that is algebraically solid; the gap is that typicality advantage and preparable states are untested past N=5.","tokens_in":17672,"tokens_out":562,"would_cite":false,"duration_ms":19770,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Averaging the Rodeo response over random states estimates the density of states as a convolution with a kernel fixed by the evolution-time distribution.","keywords":["density of states","Rodeo algorithm","kernel polynomial method","quantum typicality","spectral kernel","Haar-random states","transverse-field Ising model","near-term quantum algorithms"],"falsifier":"On a system large enough that exact diagonalization is still possible, prepare approximate random product states, run the Rodeo estimator with a chosen p(t), and check whether the reconstructed DoS (and thermodynamics derived from it) match the exact spectrum within the predicted error bars; a systematic bias that does not shrink with Hilbert-space dimension would falsify the claim.","tokens_in":17236,"feed_emoji":"⚛️","tokens_out":914,"duration_ms":17829,"temperature":0.7,"pith_summary":"The density of states of a quantum many-body system is hard to rebuild once the Hilbert space is too large to diagonalize. This paper shows that the Rodeo algorithm—a simple single-ancilla circuit already used to locate eigenvalues—becomes a direct quantum analogue of the classical kernel polynomial method when its input is a Haar-random state. Averaging the Rodeo signal over such states yields the density of states smoothed by a spectral kernel whose shape is set entirely by how evolution times are sampled. Random states replace stochastic trace estimation; the time distribution replaces the classical damping kernel. Quantum typicality makes the statistical error shrink as the system grows, and the method needs no new circuit hardware. The author derives the estimator and its uncertainties, maps classical window functions onto quantum kernels, and checks the construction on small transverse-field Ising and spin-1 chains.","feed_headline":"Rodeo circuit estimates density of states via random inputs","feed_subtitle":"Haar-averaged Rodeo signals recover the DoS smoothed by a kernel fixed only by evolution-time sampling","key_machinery":"The spectral kernel G, equal to the characteristic function of the evolution-time distribution p(t) (with a small qudit correction). When the Rodeo response is Haar-averaged, this kernel turns the measured signal into the convolution (g ∗ G)(E), which is the DoS estimator.","core_discovery":"Averaging the Rodeo spectral amplitude over Haar-random input states produces an unbiased estimator of the density of states convolved with a spectral kernel G fixed solely by the temporal sampling distribution p(t). Random states play the role of stochastic trace estimation and p(t) the role of the classical damping kernel, all inside the ordinary single-ancilla Rodeo circuit.","pith_inferences":["Shallow random-product or unitary-design state preparation, already common in classical stochastic trace estimation, is the practical bottleneck that will decide whether the method scales beyond toy models.","Pairing the estimator with flat-histogram Monte Carlo (as the author briefly suggests) could extend dynamic range for quantities that need log g(E) over many decades.","Hardware noise and decoherence will reshape the effective kernel; characterising that distortion is a natural next experimental target."],"forward_implications":["Any classical window function (Gaussian, Hann, Hamming, Blackman, Kaiser) becomes a ready-made quantum reconstruction kernel by choosing the matching p(t).","Spectral resolution is tuned by the width of p(t); level degeneracies are recovered by integrating reconstructed peaks.","Thermodynamic quantities (free energy, mean energy, specific heat) can be read off from the estimated DoS without diagonalizing H.","The same single-ancilla circuit already used for eigenvalue location can be reused unchanged for full DoS reconstruction.","In the large-system limit the typicality error vanishes, leaving temporal sampling and Trotter error as the dominant uncertainties."],"fun_headline_variants":["Haar-averaged Rodeo yields kernel-smoothed density of states","Rodeo mirrors classical KPM to estimate quantum DoS","Random inputs turn single-ancilla Rodeo into DoS estimator","Temporal sampling alone fixes the kernel in Rodeo DoS recovery","Quantum typicality lets Rodeo estimate DoS without full diagonalization"],"cache_read_input_tokens":128,"weakest_assumption_plain":"That Haar-random or sufficiently design-like input states can actually be prepared at the system sizes where the method is supposed to beat classical enumeration, so that typicality really suppresses the variance.","fun_headline_variants_meta":{"raw":{"variants":["Haar-averaged Rodeo yields kernel-smoothed density of states","Rodeo mirrors classical KPM to estimate quantum DoS","Random inputs turn single-ancilla Rodeo into DoS estimator","Temporal sampling alone fixes the kernel in Rodeo DoS recovery","Quantum typicality lets Rodeo estimate DoS without full diagonalization"]},"model":"grok-4.5","effort":"low","cost_usd":0.004822,"raw_usage":{"total_tokens":1334,"prompt_tokens":741,"num_sources_used":0,"completion_tokens":95,"cost_in_usd_ticks":48224000,"prompt_tokens_details":{"text_tokens":741,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":498,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":741,"tokens_out":95,"duration_ms":7056,"temperature":1.0,"reasoning_tokens":498,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T04:34:37.542083+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"On a system large enough that exact diagonalization is still possible, prepare approximate random product states, run the Rodeo estimator with a chosen p(t), and check whether the reconstructed DoS (and thermodynamics derived from it) match the exact spectrum within the predicted error bars; a systematic bias that does not shrink with Hilbert-space dimension would falsify the claim.","supporting_citations":[],"review_version":1}