{"id":"1f699b93-fd6b-4fa6-9cf0-cf26144323c0","arxiv_id":"2607.17740","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Local noise in shallow randomized measurements damps Pauli observables exponentially with operator size, with slope and intercept set by the operator's light cone, enabling small-string calibration of large-string estimates.","lead":"Randomized measurement protocols that estimate quantum states are biased by circuit noise; this paper derives how that bias grows with the size of the measured operator in shallow 1D circuits. It shows the bias follows a simple exponential law, which can be calibrated on small operators and used to correct larger ones.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (7) rests on an unproven extensivity claim for cumulants of the activated-noise sum; the path-measure tilt and finite-depth correlations could produce k-dependent corrections, and the numerical fits use short k-windows without error bars.","rationale":"The paper's goal is to derive and use a light-cone damping law. For that law to hold, the cumulants of X_k must be extensive in k. The main text asserts this rather than proving it, and the supplement is not available for review. I examined whether the bulk light-cone volume O(kd) plus independent local noise is sufficient. It is not automatic: the path measure is tilted by the final support size, so the expectation in Eq. (5) is not the bare support distribution; correlations in the tilted path ensemble can contribute to cumulants in a k-dependent way. The authors' own caveat about common-mode layer fluctuations shows that extensivity depends on noise correlations. Since the central calibration protocol extrapolates from small k to large k, a non-extensive correction would directly invalidate the protocol. The numerical data is genuine support: transfer-matrix simulations reproduce linear log eta for two gate ensembles and three noise models, with high R^2. But the fits use a narrow k range (max k=14, d<=6) and no error bars or window-splitting checks, so they cannot distinguish small curvature or a crossover from true extensivity. This does not make the claim implausible; it makes it insufficiently supported. The reader's CONDITIONAL verdict captures this. I agree with the identified weakest assumption; my concrete test would settle whether the extensivity assumption holds.","tokens_in":13318,"tokens_out":8315,"duration_ms":107540,"concrete_test":"Run the transfer-matrix code with an auxiliary source s coupled to X_k, compute log E_nu e^{-s X_k} for s in {0,0.01,0.02,0.05}, and extract kappa_2 and kappa_3 by finite differences at s=0. Then test whether kappa_n(k)-kappa_n(k-1) is constant over k=4..14 for d=4 and d=6 (and d=8 with N=40). If the increments are constant to numerical precision, extensivity is confirmed; if they drift with k, Eq. (7) needs correction and the calibration protocol loses its stated predictive power.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is Eq. (7): log eta_k = -alpha(d,Lambda)k - beta(d,Lambda) for contiguous Pauli strings. The derivation in the main text reduces this to the assertion, immediately after Eq. (6), that under the measurement-weighted path measure nu(gamma|P_k)=Pr(gamma)3^{-m(gamma)}/w_0(P_k), every cumulant of X_k = sum_a I_a(gamma)lambda_a(gamma) is affine in k: kappa_n(X_k) = k a_n(d) + b_n(d). This is the load-bearing step. It is not a trivial consequence of the light-cone volume O(kd): nu is not a product measure—the 3^{-m(gamma)} final-support tilt couples the whole path—and the indicators I_a are correlated along the support path. Connected cumulants of a sum over the light cone need only be linear if the tilted path ensemble has sufficiently short-ranged correlations in the spatial direction; that is exactly what is asserted and deferred to [40]. The paper itself acknowledges that a layer-correlated common-mode fluctuation produces nonlinear corrections in k, so the regime boundary is real. The numerical evidence consists of transfer-matrix fits at d=2,4,6 over even-k windows (k=4..14 for Clifford, k=6..14 for iSWAP) with R^2 reported but no error bars and no k-window stability analysis; the calibration in Fig. 4 uses a single 10^7-sample realization with circuit-global p~N(0.03,0.005^2). These fits are consistent with Eq. (7) but do not independently establish the cumulant extensivity needed for large-k extrapolation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a microscopic, path-integral description of how local in-circuit noise biases classical-shadow estimators based on shallow locally scrambled circuits. Under independent local twirling, noise reduces to stochastic Pauli damping, and a noise event contributes only when it intersects the Heisenberg support path of the measured Pauli operator. This leads to a measurement-weighted path-average formula for the noisy Pauli coefficient, Eq. (4). For contiguous Pauli strings in 1D shallow circuits, the light-cone volume argument yields an exponential damping law log η_k = -α(d,Λ)k - β(d,Λ), Eq. (7), whose slope and offset are determined by bulk and boundary cumulant contributions. The authors verify the law numerically for random Clifford and locally scrambled iSWAP circuits of depth d=2,4,6 under uniform, spatially fluctuating, and temporally drifting noise, and use it in a small-string calibration protocol to predict larger-string damping and to correct cluster-state expectation values.","tokens_in":13766,"tokens_out":2906,"duration_ms":39326,"significance":"If the central exponential law is correct, this work gives a conceptually clean and practically useful connection between operator spreading and noise exposure in randomized measurements. It avoids learning the full noisy measurement channel, replacing it with a few geometry-resolved damping parameters. The numerical transfer-matrix verification is a strength: it covers two different gate ensembles, anisotropic noise, spatial fluctuations, and temporal drift, with high reported R² values. The derivation of Eq. (7) from a cumulant expansion is also not a fit to the target result; it is a parameter-free prediction up to the cumulant-extensivity step, and the heuristic equilibrium estimate provides a useful benchmark. The main open point is the missing proof or precise condition for the cumulant extensivity assertion, which is load-bearing for the large-k extrapolation underlying the calibration protocol.","major_comments":[{"comment":"The central result Eq. (7) rests on the assertion that every cumulant of X_k is extensive in k: κ_n(X_k) = k a_n(d) + b_n(d). This is not a trivial consequence of the light-cone volume O(kd). The measurement-weighted path measure ν(γ|P_k) is non-product because of the 3^{-m(γ)} final-support tilt, and the indicators I_a are correlated along the support path. Connected cumulants of the sum over the light cone are affine in k only if the tilted path ensemble has sufficiently short-ranged spatial correlations; that condition is neither stated nor proved in the main text and is deferred to [40]. Since the calibration protocol extrapolates from fitted small-k parameters to large-k observables, this step is load-bearing. Please provide the proof or a precise mixing/clustering condition, and show that it holds for the Clifford and iSWAP ensembles considered. The paper's own caveat that a layer-","section":"Light-cone volume law, after Eq. (6)"},{"comment":"The numerical evidence for Eq. (7) consists of linear fits over short even-k windows: k=4..14 for Clifford and k=6..14 for iSWAP, with reported R² in (0.99,0.99999). No error bars, residuals, or k-window stability analysis are given. Because the predictive content of the protocol is exactly the extrapolation of the fitted line to larger k, the reader needs to see (i) how α_p and β_p change with the fitting window, (ii) whether the linear form holds for k>14 in the transfer-matrix data, and (iii) the statistical uncertainty of the slope and intercept. Please add this analysis or explicitly state that the fits are single representative realizations.","section":"Figs. 2 and 3"},{"comment":"The calibration demonstration uses a single 10^7-sample realization with circuit-global p~N(0.03,0.005^2). There is no sample-to-sample spread, and the error bars on the fitted α_p,β_p are not given. Since the final claim is that small-string fits predict larger strings, the uncertainty in that prediction should be quantified. A single realization cannot establish the reliability of the extrapolation; please report statistics over multiple noise realizations or over bootstrap resamples of the calibration data.","section":"Fig. 4(a)"}],"minor_comments":[{"comment":"Typos: 'iSW AP' should be 'iSWAP' in the abstract and in the Fig. 2 caption.","section":"Abstract and Fig. 2 caption"},{"comment":"Notation is inconsistent: 'Eq. (2)' appears as 'eq. (2)' in the text, and the noiseless coefficient is written w_EU,0, w0, and w_0 in different places; please unify.","section":"Calibration section"},{"comment":"The definition of the cluster-state string O_ZXZ uses 'Z_9 Y_10 X_11 ...' but earlier text describes it as 'Z_1 Y_2 X_3 ...'; please make the indexing consistent or explain the shift.","section":"Fig. 4(b)"},{"comment":"The symbol d is used for circuit depth and appears in the same context as the operator size k; also the 'ZXZ state' is not defined. A brief definition would help.","section":"General notation"}],"recommendation":"major_revision","confidential_remarks":"The manuscript has a plausible central claim, strong numerical support, and a useful practical protocol. The main blocker is the unproved cumulant-extensivity assertion; if the supplemental material contains a rigorous derivation, the paper could be acceptable after the authors surface that argument in the main text and add uncertainty quantification to the numerical fits. I would not reject at this stage, but the proof or precise condition is essential."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper gives a new, plausible microscopic account of how in-circuit noise biases shallow randomized measurements, and the numerical evidence is strong. The main thing to check before trusting it is the cumulant-extensivity step that yields the linear law — it's asserted in the main text and deferred to the Supplemental Material.\n\nWhat's actually new: the activated path-average formula (Eq. 4) and the resulting exponential damping law in k (Eq. 7) for contiguous Pauli strings. Existing noisy-shadow work treats noise as an effective channel; this paper instead resolves which local noise events matter along the Heisenberg light cone. That's a fresh angle, and the calibration protocol that extrapolates from small strings to large ones is genuinely practical.\n\nThe numerics are a real strength. The transfer-matrix verification covers random Clifford and locally scrambled iSWAP gates, with uniform, spatially fluctuating, and drifting noise, and the R^2 values are high. The fact that the fitted slope disagrees with the naive equilibrium estimate (Fig. 3) shows the law is not trivial. The paper is also honest about the regime boundary: layer-correlated common-mode noise is said to produce nonlinear corrections.\n\nThe soft spots are the ones the stress-test flagged. The derivation of Eq. (7) reduces to the claim that every cumulant of X_k is affine in k. That does not follow automatically from the light-cone volume, because the measurement-weighted path measure ν is tilted by 3^{-m(γ)} and the activation indicators are correlated along the path. The paper states this extensivity and points to the SM; if the SM contains a rigorous proof, the paper is in good shape. If it's only a heuristic, then Eq. (7) is effectively a conjecture with very good numerical support, and the calibration protocol's large-k extrapolation is not fully justified. The reported fits also use short k-windows without error bars or window-stability checks — minor, but worth asking for.\n\nOverall: this deserves a serious referee. I'd send it out, with a request that the referee see the SM and that the fits include a sensitivity analysis. If the extensivity step holds up, the result is a useful contribution to shadow estimation and noise characterization.","headline":"A plausible new light-cone damping law for noisy shadows, numerically well-supported, but the central extensivity proof is deferred to the SM.","tokens_in":14178,"tokens_out":2266,"would_cite":true,"duration_ms":27486,"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":"Noise bias in randomized measurements follows an exponential size law.","keywords":["randomized measurements","classical shadows","noise bias","locally scrambled circuits","Pauli twirling","light cone","operator spreading","quantum error mitigation"],"falsifier":"Measure log eta_k versus k for a shallow 1D circuit in which the same noise value is applied to every gate in a layer (common-mode noise). If the curve exhibits curvature or a k-independent offset that breaks the linear fit within the predicted window, the extensivity of cumulants fails. Alternatively, on hardware, fit alpha and beta from small strings on a product state, then prepare a state with known large-string expectation values (e.g., a cluster state) and compare predicted versus measured damping at larger k; a systematic mismatch would falsify the transfer.","tokens_in":13239,"feed_emoji":"⚛️","tokens_out":4235,"duration_ms":47388,"temperature":0.7,"pith_summary":"This paper establishes that, in one-dimensional shallow locally scrambled circuits, gate noise does not act as a flat channel error: a local noise event contributes only when it overlaps the backward Heisenberg evolution of the measured Pauli operator. For a contiguous string of k Pauli operators, the activated light cone contains O(kd) bulk gates and O(d^2) boundary gates, so the logarithm of the damping ratio is linear in k, log eta_k = -alpha(d, Lambda) k - beta(d, Lambda). The authors derive this exponential law from a path-average formula and verify it numerically for random Clifford and locally scrambled iSWAP circuits with uniform, spatially fluctuating, and temporally drifting noise. The practical payoff is a small-string calibration protocol: measuring a few short strings on a product state fixes alpha and beta, and the law then predicts the damping of larger strings with the same geometry without learning the full noisy measurement channel. A sympathetic reader would care because this connects noise bias in shadow tomography to operator-spreading dynamics and offers a cheap, geometry-resolved error-mitigation strategy.","feed_headline":"Exponential law governs noise bias in shallow shadow circuits","feed_subtitle":"Two parameters learned from small strings predict how gate noise damps larger observables, no full channel reconstruction needed.","key_machinery":"The key object is the activated path-average formula (Eq. 4): w_Lambda(P) = sum_gamma Pr(gamma|P, E_U) 3^{-m(gamma)} exp(-sum_a I_a(gamma) lambda_a(gamma)). Local twirling reduces CPTP noise to Pauli damping, and a noise location a is 'activated' (I_a = 1) only when it overlaps the instantaneous Heisenberg support of the measured operator. Dividing by the ideal path average defines the measurement-weighted path measure nu(gamma|P), which favors small final supports and is the distribution over which the noise is averaged. The supporting identity is the light-cone volume count: for a contiguous string away from boundaries, the bulk contains O(kd) gates and the two fronts O(d^2), giving linear","core_discovery":"For a locally scrambled shallow circuit with independent local twirling, the noisy measurement channel is Pauli-diagonal, and the damping ratio of a contiguous size-k Pauli string is eta_k = E_nu e^{-X_k}, where X_k sums local stochastic Pauli damping rates over space-time locations that overlap the Heisenberg support of the evolving operator. Because the light-cone bulk is extensive in k while the two fronts contribute only O(d^2), each cumulant of X_k is k a_n(d) + b_n(d); exponentiation of the cumulant expansion yields log eta_k = -alpha(d, Lambda) k - beta(d, Lambda). The paper verifies this exponential scaling for two-qubit random Clifford and locally scrambled iSWAP brick-wall circuits","pith_inferences":["If the exponential law persists beyond the simulated ranges, hardware could calibrate large-string shadow estimates from a single product-state run plus a noiseless transfer-matrix computation, turning operator-spreading dynamics into a practical noise model.","The damping-slope extraction suggests a converse diagnostic: measuring how alpha varies with circuit depth and observable geometry could serve as an experimental probe of operator-spreading velocities in locally scrambled circuits.","The linear-law assumption is directly testable for correlated noise: adding a shared common-mode fluctuation to every gate in a layer should produce curvature in log eta_k versus k, providing a clean experimental signature of such correlated errors.","A similar light-cone volume argument may extend to other randomized-measurement families with known Heisenberg support evolution, such as locally entangled or dual-unitary circuits, where the same geometry-to-noise relation could hold."],"forward_implications":["For any observable whose light cone is a single contiguous string, the damping ratio is determined by two scalar parameters alpha(d, Lambda) and beta(d, Lambda) instead of the full noisy measurement channel.","Small-string measurements on a known product state suffice to fit these parameters; the same fitted law predicts the coefficient of larger strings with matching support geometry.","The calibration corrects the bias of noisy shadow estimates, as demonstrated for cluster-state (ZXZ) strings, and the resulting sample variance follows the noisy shadow norm eta^{-2} w_0(P).","The slope alpha carries information about operator spreading: the heuristic equilibrium estimate alpha_eq = 15 d lambda / 32 does not reproduce finite-depth slopes, showing that activation has not relaxed to equilibrium and that Jensen's inequality matters.","Noncontiguous observables and higher-dimensional circuits obey the same path-average principle but acquire different volume scalings, so they must be calibrated with reference observables of the same light-cone geometry."],"fun_headline_variants":["Exponential noise decay for shallow shadow measurements","Small strings predict large-observable noise in shallow circuits","Exponential damping law for noise in shallow shadow circuits","Light-cone scaling gives exponential noise damping in shadows"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The exponential law rests on the assertion that every cumulant of the activated noise is linear in the string length k, i.e., that noise events at different light-cone locations are effectively independent under the measurement-weighted path measure; if gates in a layer share a common-mode fluctuation, this linearity fails.","fun_headline_variants_meta":{"raw":{"variants":["Exponential noise decay for shallow shadow measurements","Small strings predict large-observable noise in shallow circuits","Exponential damping law for noise in shallow shadow circuits","Light-cone scaling gives exponential noise damping in shadows"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00088,"raw_usage":{"total_tokens":3623,"prompt_tokens":708,"completion_tokens":2915,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":452,"completion_tokens_details":{"reasoning_tokens":2865}},"tokens_in":452,"tokens_out":2915,"duration_ms":20276,"temperature":1.0,"reasoning_tokens":2865,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T17:03:32.556319+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure log eta_k versus k for a shallow 1D circuit in which the same noise value is applied to every gate in a layer (common-mode noise). If the curve exhibits curvature or a k-independent offset that breaks the linear fit within the predicted window, the extensivity of cumulants fails. Alternatively, on hardware, fit alpha and beta from small strings on a product state, then prepare a state with known large-string expectation values (e.g., a cluster state) and compare predicted versus measured damping at larger k; a systematic mismatch would falsify the transfer.","supporting_citations":[],"review_version":1}