{"id":"6f7c0959-61cc-4eb1-8416-d933fc526c35","arxiv_id":"2411.08214","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For general sequential quantum measurements, the paper derives asymptotic covariance matrices of empirical distributions of outcome substrings and a relative-entropy measure of the influence of correlations.","lead":"The paper derives formulas for the covariance of empirical histograms of measurement outcomes from sequentially measured quantum systems, and introduces a relative-entropy measure for how correlations affect those histograms. It matters because the framework applies to any quantum instrument and gives tools for quantum metrology and full counting statistics.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Drazin-inverse step presupposes a spectral gap |lambda_j|<1, but Appendix A only assumes |Re lambda_j|<1; a cyclic-shift instrument gives eigenvalues i,-i and breaks Eqs. (31)-(32).","rationale":"The reader's weakest assumption identifies the spectral-gap/Gaussian condition, and I agree. The load-bearing step is in Appendix A: the geometric series sum_ell lambda_j^{ell-1} is replaced by (1-lambda_j)^{-1}, which is valid only if |lambda_j|<1. The paper states only |Re lambda_j|<1, which is strictly weaker: a trace-preserving instrument with a cyclic unitary between measurements has eigenvalues i and -i (modulus 1, real part 0) while retaining a unique steady state. For the d=4 shift instrument, the exact Psi_N in Eq. (7) does not have a unique large-N limit (it depends on N mod 4), so Eq. (31) is not an approximation to it, and the empirical distribution does not follow the assumed O(1/N) Gaussian law. This directly removes the validity of Eqs. (31), (32), and hence the large-N usage of Eq. (11) and I_m^L, for a class of stationary instruments the text claims to cover. The fix is to state and verify trivial peripheral spectrum (|lambda_j|<1 for j>=1), or an explicit spectral gap, and restrict the central claims accordingly; with that condition the derivations appear coherent. I also note the paper states I_m^L diverges for L>m+1 but plots finite values for such points in Figs. 5 and 9; this is a secondary presentation inconsistency.","tokens_in":21155,"tokens_out":33494,"duration_ms":349657,"concrete_test":"Use d=4, U the cyclic shift, and instruments M_x rho = |x>=|k+1>, the instrument M = D o U (D = dephasing in the measurement basis) has eigenvalues 1, i, -1, -i and unique steady state I/4, so |Re(i)|=0<1 but |i|=1. For this stationary instrument, p_{y<-x}(ell) is periodic, so Psi_N in Eq. (7) has no unique large-N limit (it differs along N mod 4), and Eq. (31) is not an approximation to it. Consequently Eqs. (31)-(32), the large-N form of the covariance Eq. (11), and the Gaussian relative-entropy construction I_m^L are invalid for this class of instruments that nevertheless satisfies the assumptions as stated in Appendix A. The paper's Sec. V C acknowledges one related failure (conserved total spin), but the required condition is a spectral gap, not merely unique steady state plus |Re lambda_j|<1.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the empirical distributions of length-L outcome blocks generated by a stationary sequence of quantum-instrument measurements. Its central technical results are the large-N covariance formula Eq. (11) with the correlation matrix Psi given by Eq. (31) for L=1 and Eq. (32) for general L, both expressed through the Drazin inverse (1-M)^+ of the unconditional instrument superoperator M. The paper then defines, in Eq. (38), a quantity I_m^L based on the KL divergence between Gaussian covariances, meant to quantify how correlations in the measurement string affect reconstruction under an order-m Markov model. The formalism is illustrated on an amplitude-damped qubit, a boundary-driven XX spin chain, and a periodically measured spin chain, and it is connected to Fisher information for parameter estimation.","tokens_in":21317,"tokens_out":7254,"duration_ms":76874,"significance":"If the large-N Psi formulas are valid, they provide a broadly applicable full-counting-statistics tool for sequential quantum measurements, extending the earlier ED1 treatment of Ref. [34] to general EDL and to quantum jump processes. The derivations in Appendices A-C are mostly coherent, the analytic bit-flip case provides a useful closed-form check, and no quantities are fitted to data: the information measure is compared against an explicit Markov benchmark. The main gap is that the spectral assumptions stated in Appendix A do not justify the geometric-series resummation for all instruments covered by the paper's claims; this is fixable by strengthening the assumption to a spectral gap and restricting the scope accordingly.","major_comments":[{"comment":"The eigendecomposition assumes a unique steady state and |Re lambda_j| < 1, but the geometric-series resummation leading to (1-M)^+ requires |lambda_j| < 1. The weaker condition admits instruments with non-trivial peripheral spectrum. A concrete example is M = D o U, where U is the cyclic shift on a d=4 computational basis, U|k> = |k+1 mod 4>, and D is dephasing in that basis. This instrument is trace-preserving, has unique steady state I/4, and its eigenvalues are 1, i, -1, -i, so |Re lambda_j| < 1 holds for all j >= 1. However, p_{y<-x}(ell) = delta_{y, x+ell mod 4} is periodic, and the sum defining Psi_N in Eq. (7) has no unique large-N limit: along N = 4k the value depends on y-x mod 4, while along other subsequences it differs. Hence Eq. (31) is not an approximation to Psi for this stationary instrument, and the covariance formula Eq. (11) as stated does not apply. The assumption should be strengthened to a spectral gap, |lambda_j| < 1 for all j >= 1, and this restriction should be stated in Sec. III and Appendix A; the discussion in Sec. V C of conserved total spin acknowledges a different failure and does not cover the periodic case.","section":"Appendix A, Eqs. (A5)-(A7) and (A14)-(A15)"},{"comment":"The quantity I_m^L is defined as DKL(Sigma || Sigma^(m,L)), the KL divergence between two Gaussian distributions with the same mean, not as the KL divergence between the process distributions P_L and Q_m^L. Writing DKL(P_L || Q_m^L) is therefore an overloading that is justified only if the empirical distributions are (at least asymptotically) Gaussian and only when m+1 >= L so the means coincide. Equation (35) is likewise a covariance-only divergence. This distinction is not cosmetic because Sec. V C exhibits a stationary process for which the empirical distribution is multimodal; for such processes I_m^L is not a relative entropy of the empirical distributions. Please rename the quantity (for example, 'Gaussian covariance divergence') or verify the Gaussian approximation for the models in which it is interpreted as a relative entropy.","section":"Sec. IV B, Eq. (38)"}],"minor_comments":[{"comment":"'Meteorological' should be 'metrological'.","section":"Sec. IV B, after Eq. (38)"},{"comment":"The symbol M is used both for the alphabet of outcomes and for the unconditional instrument superoperator in Eq. (14); using |M| for alphabet cardinality or another symbol for the superoperator would remove ambiguity.","section":"Secs. II A-II C"},{"comment":"The constraint argument drops the boundary error epsilon_N after noting it is strictly bounded; please clarify explicitly that the support constraints are exact only up to these boundary corrections and that the O(N^{-1}) width claim follows from the bound rather than from a distributional calculation.","section":"Appendix C"},{"comment":"The quantities q_+ and P(q_+) are used in the caption but not defined there; please define them explicitly.","section":"Fig. 8 caption"}],"recommendation":"major_revision","confidential_remarks":"The spectral-gap counterexample is the decisive technical issue: as stated, Appendix A's condition does not imply convergence of Psi_N, so the paper's main claim of applicability to generic quantum instruments overreaches. The fix is straightforward in principle -- require |lambda_j| < 1 and restrict the statements accordingly -- but it changes the scope of the advertised result. The KL-divergence notation issue in Eq. (38) should also be corrected before publication. I do not see grounds for rejection: the derivations are largely sound under the stronger spectral assumption, and the examples are meaningful."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper is a solid extension of the authors' earlier ED1 framework to length-L empirical distributions, with two genuinely new pieces (the overlap-corrected quantum-jump formula for Psi, and the arbitrary-order Markov comparison I_m^L). The appendices are mostly careful, and there's no fitting or circularity.\n\nThe main soft spot is a real but fixable gap in the spectral condition. Appendix A posits |Re lambda_j| < 1, but the geometric series step uses |lambda_j| < 1. The latter is what's needed. A trace-preserving instrument can have a unique steady state and still have eigenvalues on the unit circle (roots of unity); when that happens, the correlations don't decay and the covariance has no large-N limit. The paper's Sec. V C flags a different failure mode (conserved total spin) but not this one. The stress-test example as written includes a -1 eigenvalue, which violates even the stated |Re lambda_j| < 1, so that counterexample doesn't land; a 3-cycle would. The fix is straightforward: replace the condition by |lambda_j| < 1 for j >= 1, or discuss the periodic case separately.\n\nSecond issue: Eq. (38) defines the KL divergence as covariance-only (zero mean difference), yet the text says that for L > m+1 the means differ and I_m^L diverges. A covariance-only KL divergence doesn't diverge from a mean mismatch, so either the definition or that statement is off. The figures imply the mean term is included somewhere; that needs to be made explicit.\n\nThe examples (amplitude-damped qubit, boundary-driven spin chain) are helpful, and the paper is clearly organized. Typos are cosmetic.\n\nWho this is for: people working on full counting statistics, quantum metrology with sequential measurements, and thermodynamic uncertainty relations. They will find the formulas directly useful. I'd send it to a serious referee; the central results are likely correct for mixing instruments and the weaknesses are fixable in revision.","headline":"Sound extension of the ED1 formalism to length-L sequences, but the spectral-gap condition and the covariance-only KL divergence need fixing.","tokens_in":21942,"tokens_out":4492,"would_cite":true,"duration_ms":44002,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P15","81P45","62B10"],"pacs":["03.65.Ta","02.50.Cw","05.40.-a"],"model":"deepseek-v4-flash","headline":"This paper derives closed-form covariance formulas for histograms of finite outcome strings from generic quantum instruments and introduces a relative-entropy measure that quantifies how much Markov-model compression loses.","keywords":["quantum instruments","empirical distributions","measurement correlations","quantum jumps","covariance matrix","relative entropy","Fisher information","full counting statistics"],"falsifier":"Simulate the spin-conserving periodic spin chain of Sec. V C (κ=0) with random projective measurements and compute the empirical distribution of q_+ for N=$10^{5}$: the multimodal structure visible in Fig. 8 for rapid measurements would persist rather than become Gaussian, directly violating the Gaussian assumption that underlies Eqs. (11) and (38).","tokens_in":20840,"feed_emoji":"📊","tokens_out":8200,"duration_ms":72799,"temperature":0.7,"pith_summary":"This paper establishes that for a stationary sequence of outcomes produced by repeatedly measuring a quantum system with any quantum instrument, the covariance matrix of the empirical distribution of length-L substrings has a closed-form expression in the large-data limit, Eq. (11), with the correlation matrix Ψ given explicitly in terms of the instrument superoperator and its Drazin inverse, Eq. (32). This reduces the complicated many-body problem of correlated quantum measurement records to linear algebra, making the Fisher information and other statistical quantities directly computable from the instrument alone. The paper further introduces a relative-entropy measure I_m^L that quantifies how much information is lost when the true string statistics are compared with an order-m Markov-model approximation, thereby giving a quantitative signature of non-Markovian correlations. A sympathetic reader would care because these formulas turn full counting statistics and quantum metrology for a wide class of measurement schemes into a single tractable calculation.","feed_headline":"One matrix captures all correlations in quantum measurement strings","feed_subtitle":"New formulas give covariances and a Markov-comparison entropy for histograms of any outcome-string length.","key_machinery":"The load-bearing object is the instrument superoperator M = \\sum_x M_x together with its Drazin inverse (1-M)^+, the inverse defined on the subspace orthogonal to the steady-state eigenspace. The correlation matrix Ψ constructed from these objects, Eq. (32), compresses the entire infinite tail of measurement correlations into a single finite matrix entry that appears in the covariance formula, so that quantum backaction and memory effects are fully captured by linear algebra. For L > 1, the extra overlap terms p_{y←x}(\\ell)-p_y, \\ell=1,...,L-1, handle the fact that a substring of length L is counted multiple times when it overlaps its predecessor, which is a purely combinatorial correction independent of quantum dynamics.","core_discovery":"The central claim is that for a stationary quantum-instrument process with unique steady state π, the covariance of the empirical distribution of length-L substrings is, for large N, Σ_{xy} = 1/(N−L+1)[p_x(δ_{xy}−p_y)+Ψ_{yx}p_x+Ψ_{xy}p_y], where Ψ_{yx} = \\sum_{\\ell=1}^{L-1}[p_{y←x}(\\ell)-p_y] + (1/p_x)\\langle\\langle 1|M_y(1-M)^+M_x|\\pi\\rangle\\rangle. Here M = \\sum_x M_x is the completely positive instrument superoperator, (1-M)^+ its Drazin inverse, and the overlap terms p_{y←x}(\\ell)-p_y account for the counting of overlapping substrings. The paper proves this in Appendix A by eigendecomposing M and summing the geometric series of transient eigenvalues, and it specializes the result to quantum jumps, where M_x = -J_x $L_0^{{-1}}$ is built from the jump and no-jump superoperators. It then defines the empirical-distribution mutual information I_m^L = D_{KL}(Σ\\|$Σ^{{(m,L)}}$), the KL divergence between the true covariance and that of the process marginalized to an order-m Markov model, and shows this quantity is zero for Markov processes and independent of N when the means coincide.","pith_inferences":["The same Ψ-based formulas should, if the spectral-gap assumption holds, apply to any rapidly measured open system, including weak continuous measurements treated as a high-rate instrument; a direct Monte Carlo test of Eq. (11) for a weakly measured cavity would cleanly separate the validity of the Gaussian assumption from the spectral-gap condition.","Because I_m^L is independent of N when the first moments match, one could use the measured histograms from an experimental record to fit the smallest Markov order m that makes I_m^L statistically indistinguishable from zero, turning the measure into a data-driven model-selection tool that requires no knowledge of the underlying instrument.","The same matrices that appear in the covariance also determine the Fisher information, suggesting that the formalism could be used to quantify how much estimation error is avoided by keeping correlations in the data, i.e., the gap between the true estimator variance and the iid-optimal bound; the paper touches on this via Eq. (30) but does not fully exploit the comparison."],"forward_implications":["For any instrument-based measurement sequence, the Fisher information for estimating a parameter from the empirical distribution is F = N(\\partial_\\theta p)^T(P+ΨP+PΨ^T)^{-1}\\partial_\\theta p, so metrological error bars can be computed from the instrument superoperators alone, without Monte Carlo simulation of the record.","The measure I_m^L vanishes exactly when the underlying process is Markov of order at most m (for L ≥ m+1), so it functions as a quantitative non-Markovianity witness for sequential quantum measurements.","Even for instruments whose individual outcomes are independent, the ED_L covariance contains overlap contributions from the second line of Eq. (32), so higher-order empirical distributions carry information about sequence structure that ED1 cannot see.","In the quantum-jump case, substituting M_x=-J_x L_0^{-1} into Eq. (32) gives a closed expression for Ψ in terms of the Lindblad generator, making the covariance and the information measure computable directly from the master equation and jump operators."],"supporting_citations":[{"why":"Supplies the original ED1 covariance formula, the Gaussian approximation, and the Ψ-matrix definition that this paper extends to arbitrary sequence length L.","marker":"[34]"},{"why":"Defines the jump-channel steady state and the quantum-jump instrument construction M_x = -J_x L_0^{-1} used to specialize the general formulas to quantum jumps.","marker":"[7]"},{"why":"Provides the Drazin inverse conventions and the full-counting-statistics framework used in Eqs. (31)-(32) to evaluate the correlation matrix.","marker":"[4]"},{"why":"Supplies the general quantum-instrument formalism in which the whole paper is cast, giving the most general type of sequential measurement.","marker":"[31]"},{"why":"Background for quantum jump processes and the dissipator/jump superoperator notation that underlies the quantum-jump examples.","marker":"[1]"}],"fun_headline_variants":["New covariance formula for quantum measurement strings","How correlations shape empirical distributions in quantum measurements","Entropy measure quantifies correlation range in quantum strings","One matrix predicts empirical distribution correlations in quantum data"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole approach presumes that the empirical distribution becomes Gaussian in the large-N limit, which in turn requires that all transient eigenvalues of the instrument superoperator M have modulus strictly less than one so that correlations decay fast enough; if some eigenvalue sits on the unit circle, as in the conserved-spin chain of Sec. V C, the covariance formula and the information measure no longer apply.","fun_headline_variants_meta":{"raw":{"variants":["New covariance formula for quantum measurement strings","How correlations shape empirical distributions in quantum measurements","Entropy measure quantifies correlation range in quantum strings","One matrix predicts empirical distribution correlations in quantum data"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000797,"raw_usage":{"total_tokens":3502,"prompt_tokens":932,"completion_tokens":2570,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":548,"completion_tokens_details":{"reasoning_tokens":2513}},"tokens_in":548,"tokens_out":2570,"duration_ms":21755,"temperature":1.0,"reasoning_tokens":2513,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T21:51:28.091078+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the spin-conserving periodic spin chain of Sec. V C (κ=0) with random projective measurements and compute the empirical distribution of q_+ for N=$10^{5}$: the multimodal structure visible in Fig. 8 for rapid measurements would persist rather than become Gaussian, directly violating the Gaussian assumption that underlies Eqs. (11) and (38).","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the original ED1 covariance formula, the Gaussian approximation, and the Ψ-matrix definition that this paper extends to arbitrary sequence length L."},{"cited_title":"Milz and K","cited_arxiv_id":null,"evidence_quote":"Supplies the general quantum-instrument formalism in which the whole paper is cast, giving the most general type of sequential measurement."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Background for quantum jump processes and the dissipator/jump superoperator notation that underlies the quantum-jump examples."}],"review_version":1}