{"id":"c978e40f-fa94-4712-b4b7-eeaf9544b6ec","arxiv_id":"2607.15978","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A reaction-coordinate embedding lets continuously monitored non-Markovian quantum systems be analyzed with standard Bayesian methods, with an analytic long-time Fisher-information formula and a thermometry demonstration.","lead":"Quantum probes that interact strongly with their environment produce measurement records that standard statistical tools cannot interpret. This paper shows how to reformulate such problems so the records become readable, and gives formulas for the precision that can be reached—demonstrated on non-invasive temperature measurement.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"GKLS embedding is benchmarked only on unconditional dynamics; the measured steady state that enters the FI formula is never independently validated.","rationale":"The reader's weakest assumption—that the Born–Markov–secular approximation for the S–RC unit might fail for the measurement-modified dynamics—is precisely the load-bearing concern we identify. No new, independent objection emerged from re-reading the manuscript. The End Matter derivation of the FI is internally consistent, and the unconditional benchmark in Fig. 3 is real but insufficient. The paper already carries a CONDITIONAL verdict, which remains appropriate: the central claim is plausible and the math checks out, but the missing benchmark of the conditional steady state means the main quantitative prediction is not yet fully verified. Our recommendation is therefore UNCHANGED, not because there is no concern, but because the reader already conditioned the verdict on exactly this unresolved issue.","tokens_in":32889,"tokens_out":7924,"duration_ms":92424,"concrete_test":"For the parameters of Fig. 2 (γ=Ω_S=1, g=√10, λ=Ω_S, Λ=1000Ω_S, T sampled from prior), simulate the exact non-Markovian conditional dynamics under homodyne detection—e.g., by truncating the original bath to N=50 modes and propagating the Gaussian conditional covariance without the Markovian embedding—and compare the steady-state covariance with the solution of Eq. (18). Also compare the empirical EMSLE after a long measurement time with the bound from Eq. (27). If they disagree beyond Monte Carlo error, the central claim is not supported in the demonstrated regime; if they agree, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that Eq. (27) gives the asymptotic FI for any differentiable parameter—rests on the accuracy of the GKLS drift and diffusion matrices (C51)–(C70) for the S–RC unit *under continuous measurement*. The only external check is Fig. 3, which compares unmonitored position and momentum variances against the exact quantum Langevin solution for a single parameter set. The measurement-modified steady-state covariance σ_ss, which is the solution of the Lyapunov equation (18) and enters the Kalman gain K(θ) and hence F_total, is never benchmarked. Physical reason for concern: homodyne detection adds a strong dissipative channel (λ=Ω_S in the numerics) that can modify system–residual-bath correlations and memory times, potentially invalidating the Born–Markov–secular treatment in Appendix C2. The measurement also localizes the system, so the conditional state explores regions far from the thermal equilibrium used in the RC derivation. Thus even if the unconditional dynamics is well captured, σ_ss—and therefore the FI—may be inaccurate. This is not an internal inconsistency but an unvalidated assumption in exactly the regime where the central claim is made. The cutoff-dependence issue (δΩ_R² = 2γΛ/π, Appendix A2) compounds this: the finite-Λ numerics silently assume the conditional steady state is cutoff-independent, which is not checked.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a reaction-coordinate (RC) embedding to extend Bayesian parameter estimation from continuous Gaussian measurements to non-Markovian (CP-indivisible) dynamics. The central idea is that by augmenting the original system with one or more collective bath modes, the system–RC unit becomes Markovian and admits a GKLS master equation, so that a quantum Kalman filter (Eqs. (8)–(9)) can be applied. The authors derive an analytic asymptotic Fisher information F_total(θ) = τ λ Tr[LᵀV⁻¹L QΣ_ss Qᵀ] (End Matter, Eq. (27)), where Σ_ss solves the augmented Lyapunov equation (26). They demonstrate the method on thermometry of a bosonic bath with a Lorentzian spectral density, using Monte Carlo simulations of the EMSLE and comparing with Bayesian Cramér–Rao bounds. The unconditional dissipative dynamics is benchmarked against the exact quantum-Langevin solution in Fig. 3.","tokens_in":33062,"tokens_out":6309,"duration_ms":76519,"significance":"If the central claim is correct, the paper makes a valuable contribution: it extends continuous-monitoring quantum metrology to a class of non-Markovian dynamics that are otherwise difficult to treat, and it provides an analytic, parameter-free expression for the asymptotic Fisher information. The combination of RC mapping with Gaussian quantum filtering is conceptually clean, the End-Matter derivation is coherent, and the code for the simulations is publicly available. The analytic derivation of the FI relies on standard martingale properties of conditional scores, which is a strength. The main caveat is that the GKLS embedding—the load-bearing assumption—is only validated against the exact solution for the unmonitored system, not for the measured conditional dynamics that enters the FI formula.","major_comments":[{"comment":"The total FI is computed from the steady-state solution Σ_ss of the augmented Lyapunov equation (26), which uses the GKLS drift and diffusion matrices from Appendix C2. The only external benchmark, Fig. 3, compares the unconditional position and momentum variances with the exact quantum-Langevin solution. The conditional steady state σ_ss solving Eq. (18) and the Kalman gain K(θ) entering F_total are never independently validated. Since the measurement strength in Fig. 2 is λ=Ω_S, the measurement-induced dissipation is not a small perturbation and could modify system–residual-bath correlations and memory times, potentially invalidating the Born–Markov–secular approximation in Appendix C2. This is a load-bearing gap: the central claim (Eq. (27)) depends on the accuracy of the GKLS matrices under continuous monitoring. Please provide a benchmark of the monitored steady state or a quantitat","section":"End Matter, Eq. (27); Fig. 3"},{"comment":"The residual-bath representation requires the divergent counter-term δΩ_R² = 2γΛ/π. The numerics set Λ = 1000Ω_S, but no cutoff-sensitivity analysis is reported. The Fisher information (27) depends on the steady-state covariance of the monitored S–RC unit, which may inherit a Λ dependence through the GKLS drift/diffusion matrices. Without evidence that the FI and the measurement-modified steady state are stable as Λ→∞, the claim that the RC embedding bypasses CP-indivisibility in a cutoff-independent way is not fully established for the measured dynamics.","section":"Appendix A2, Eq. (A46)"}],"minor_comments":[{"comment":"The vanishing of the score cross-terms is asserted without derivation. A one-line argument based on the martingale property of conditional scores, or a reference, would make the step easier to check.","section":"End Matter, Eq. (19)"},{"comment":"The sentence 'The bound holds only for unbiased estimators' refers to the TBCRB, but as written it could be misread as applying to the BCRB. Please clarify which bound is meant.","section":"Fig. 2 caption"},{"comment":"The matrix notation in the normal-mode transformation is dense; a sentence specifying the dimensions of the matrices and the ordering of the quadrature vector would improve readability.","section":"Eq. (C41)"},{"comment":"The prior contains a free parameter α, but its role and the value used in Fig. 2 are not stated. Please indicate how α is chosen.","section":"Eq. (15)"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern raised by the skeptic is well founded: the paper's central FI formula rests on the GKLS description of the S–RC unit under continuous measurement, but the only validation is for the unconditional dynamics. This is not an internal inconsistency, but it is a missing check in exactly the regime where the claim is made. If the authors can supply a benchmark or a convincing quantitative argument for the validity of the monitored GKLS dynamics, the paper would be suitable for publication. The cutoff-dependence issue is secondary but should also be addressed explicitly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper delivers something real: it extends continuous-monitoring Bayesian parameter estimation to CP-indivisible dynamics for a concrete subclass (linear Gaussian systems coupled to structured baths), by embedding the system in a Markovian system–reaction-coordinate unit. The key step is recognizing that the sequential likelihood structure survives the embedding, so the standard Kalman–Bayes machinery applies and yields an analytic steady-state Fisher information. I checked the End-Matter derivation and it is sound: the cross-score terms vanish via the martingale property of the conditional scores, and the Lyapunov construction for the augmented system is internally consistent. The simulation parameters are fixed up front, with no fitting of the bounds to the data, and the unconditional dynamics is benchmarked against the exact quantum-Langevin solution (Fig. 3). That is honest, reproducible work.\n\nThe soft spots are real but proportionate. The main one is that the measurement-modified steady state—the covariance that enters the Kalman gain and hence the FI—is never benchmarked. The only external check is for unmonitored dynamics. Homodyne detection at λ = Ω_S adds a strong dissipative channel and can localize the state far from the thermal equilibrium used in the RC master-equation derivation, so the assumption that the Born–Markov–secular treatment in Appendix C2 still holds in the measured regime is plausible but unverified. This is exactly the regime where the central claim lives. Second, the conclusion says \"any probe—regardless of non-Markovianity,\" which overreaches; the text elsewhere correctly limits the method to appropriate parameter ranges and linear Gaussian systems. Third, the FI bounds are written with natural logarithms while the EMSLE is defined with log2; unless the bounds are converted, a factor (ln 2)^2 is missing. That is minor and easy to fix, but it will confuse readers if left. Also minor: no error bars on the Monte Carlo points in Fig. 2, and the \"publicly available repository\" has no URL in the text. The cutoff issue (δΩ_R² = 2γΛ/π) means the finite-Λ numerics silently assume cutoff independence for the conditional steady state; not checked, but not obviously fatal.\n\nWho gets value: anyone working on continuous-measurement metrology, Gaussian quantum information, or reaction-coordinate methods. This is a solid contribution to a real open problem, and the derivation is carefully done. It deserves a serious referee. I would send it to peer review with requests to benchmark (or at least discuss) the measured steady state, fix the log factor, and temper the generality claim. If those checks come back clean, I'd expect it to become a standard reference for non-Markovian continuous-monitoring estimation.","headline":"A genuinely useful extension of continuous-monitoring Bayesian estimation to non-Markovian dynamics via reaction coordinates; the core math holds, but the measured steady state is only spot-checked and the conclusions overreach a bit.","tokens_in":33748,"tokens_out":1627,"would_cite":true,"duration_ms":19910,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81S22","81P50","81P15"],"pacs":["03.65.Ta","03.65.Yz","03.67.-a","05.30.-d"],"model":"deepseek-v4-flash","headline":"A reaction-coordinate embedding lets Bayesian thermometry work on continuously monitored non-Markovian quantum systems, with analytic Fisher information and 1/√τ scaling.","keywords":["continuous monitoring","non-Markovian dynamics","reaction coordinate mapping","Bayesian estimation","Fisher information","quantum thermometry","Gaussian measurements","Kalman filter"],"falsifier":"Compute the exact quantum-Langevin conditional steady-state covariance of the system for a parameter set outside the single benchmarked point (e.g., stronger system–bath coupling g or lower cutoff Λ) and compare it with the steady state predicted by the RC master equation and used in the Kalman gain. A mismatch at the level of the system's conditional covariance—rather than just the unconditional variances—would invalidate the Fisher-information formula.","tokens_in":32612,"feed_emoji":"🌡️","tokens_out":1404,"duration_ms":18310,"temperature":0.7,"pith_summary":"The paper tackles the open problem of estimating unknown parameters from a continuously monitored quantum system when the system's dynamics are non-Markovian (CP-indivisible). It argues that the obstacle can be bypassed by embedding the system in a larger 'system-plus-reaction-coordinate' unit whose dynamics are Markovian. The paper then develops a Bayesian estimation framework for this augmented unit, derives an analytic expression for the Fisher information of the continuous measurement record, and shows that estimation precision improves as the inverse square root of measurement time. The method is demonstrated by using homodyne detection on a Brownian oscillator to estimate the temperature of a strongly coupled bosonic bath, with numerical Monte Carlo verification.","feed_headline":"Non-Markovian quantum thermometry gets a Markovian workaround","feed_subtitle":"A reaction-coordinate embedding lets continuous homodyne monitoring estimate bath temperatures with 1/√τ precision.","key_machinery":"The central object is the reaction-coordinate mapping, which replaces the original bath by a collective mode (the reaction coordinate) coupled to a residual bath. When enough reaction coordinates are extracted, the residual bath becomes effectively unstructured and weakly coupled, so the system–RC unit obeys a global GKLS master equation. The conditional dynamics of this unit under a Gaussian measurement are then described by a quantum Kalman filter, with the measurement outcome likelihood given by a Gaussian whose covariance is set by the detector environment and measurement strength. The Fisher information is computed by tracking the sensitivity of the conditional mean through an augmented","core_discovery":"The central claim is that the reaction-coordinate (RC) mapping, a standard technique for Markovian embedding of structured environments, can restore complete positive divisibility in a suitably augmented system–RC unit even when the bare system dynamics are strongly non-Markovian. Because this augmented unit evolves under a GKLS master equation, the usual sequential Bayesian update of conditional Gaussian states applies. The paper derives the conditional dynamics via a quantum Kalman filter (Eqs. (8)–(9)) and shows that, in the long-time limit, the total Fisher information of the measurement record is given by F_total(θ) = τ λ Tr[LᵀV⁻¹L Q Σ_ss Qᵀ], where Σ_ss solves an augmented Lyapunov equ","pith_inferences":["The same machinery likely generalizes to parameter estimation in non-Markovian Gaussian systems beyond thermometry, such as estimating coupling strengths or bath spectral parameters, provided the RC construction remains valid.","A testable extension is to apply the framework to non-Gaussian measurements or non-linear dynamics; the RC embedding may still hold, but the Gaussian Kalman filter would need replacement by a more general filtering equation.","The analytic Fisher information could be used to design adaptive measurement strategies that choose the measurement strength λ or the detector state in real time to maximize information gain, a direction the paper mentions but does not pursue.","The calibration of the RC master equation against exact quantum-Langevin dynamics for a single parameter set suggests the method should be validated across a wider range of parameters, especially where the Markovian approximation for the S–RC unit is stressed."],"forward_implications":["Any differentiable parameter of a linear Gaussian continuously monitored system with an RC-embeddable non-Markovian bath can be estimated with error ∝ 1/√τ, matching the standard quantum-limited scaling despite strong non-Markovianity.","The method provides an analytic formula for the Fisher information (Eq. (27)) that can be used to optimize Gaussian measurements (e.g., homodyne vs. heterodyne) and detector-environment states for a given estimation task.","The Bayesian framework directly applies to thermometry, enabling accurate temperature estimates from a continuous homodyne record even when the probe is strongly coupled to a structured bath.","The approach can be iterated: extracting multiple reaction coordinates extends the Markovian embedding to more complex spectral densities, broadening the class of non-Markovian systems amenable to continuous-monitoring metrology.","The derived Cramér–Rao bounds provide benchmarks for practical estimators and quantify the information-theoretic limits of continuously monitored non-Markovian sensors."],"fun_headline_variants":["Reaction coordinates tame non-Markovian quantum parameter estimation","Bayesian quantum thermometry with homodyne detection beyond Markovian limit","Non-Markovian monitoring meets Kalman filter: precision 1/√τ","Quantum Kalman filter tames non-Markovian noise for parameter estimation","Reaction-coordinate trick unlocks non-Markovian quantum metrology"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the system–reaction-coordinate unit truly undergoes Markovian (CP-divisible) dynamics under the Born–Markov–secular approximation in the eigenbasis of H_S–RC; if this effective Markovianity fails, the Kalman filter and Fisher-information formula no longer apply.","fun_headline_variants_meta":{"raw":{"variants":["Reaction coordinates tame non-Markovian quantum parameter estimation","Bayesian quantum thermometry with homodyne detection beyond Markovian limit","Non-Markovian monitoring meets Kalman filter: precision 1/√τ","Quantum Kalman filter tames non-Markovian noise for parameter estimation","Reaction-coordinate trick unlocks non-Markovian quantum metrology"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000167,"raw_usage":{"total_tokens":1073,"prompt_tokens":705,"completion_tokens":368,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":449,"completion_tokens_details":{"reasoning_tokens":270}},"tokens_in":449,"tokens_out":368,"duration_ms":4423,"temperature":1.0,"reasoning_tokens":270,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T21:46:09.734722+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the exact quantum-Langevin conditional steady-state covariance of the system for a parameter set outside the single benchmarked point (e.g., stronger system–bath coupling g or lower cutoff Λ) and compare it with the steady state predicted by the RC master equation and used in the Kalman gain. A mismatch at the level of the system's conditional covariance—rather than just the unconditional variances—would invalidate the Fisher-information formula.","supporting_citations":[],"review_version":1}