{"id":"15ace654-2c15-48b4-a1af-aea25ab9d2c2","arxiv_id":"2607.09118","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For a qubit under lossy continuous monitoring, every impurity moment of order at least 1/2 decays at the same optimized rate 2ηℓ², set by the distinguishability of two ambiguous measurement-record laws.","lead":"A qubit monitored with a lossy detector can only be purified at a rate set by how distinguishable two possible measurement records remain, even when the controller adapts optimally. The authors prove the exact long-time speed limit and show quantum nondemolition monitoring achieves it.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified — the analytic proof of Eq. (4) is internally consistent under its stated control class.","rationale":"The reader's verdict ACCEPT with moderate confidence is well supported. The reader's weakest_assumption identifies the idealization of the loss model and control class as the main limitation. I agree that this is a limitation for direct experimental application, but it is not a load-bearing concern for the mathematical theorem: the proof is valid for the stated model, and the unlimited-bandwidth assumption makes the result stronger, not weaker. My close reading of the technical steps—especially the change-of-measure bound for the high-order plateau and the determinant contraction for the qutrit—found no internal inconsistency or unjustified limit exchange. The only checkable risk is the absence of formal verification (no Lean/Coq), so confidence remains moderate, but no specific error was identified. Therefore the verdict should remain ACCEPT, and the reader's confidence should not be downgraded on the basis of a concrete objection.","tokens_in":24571,"tokens_out":28209,"duration_ms":269830,"concrete_test":"Independently re-derive the high-order upper bound for a qubit using two different routes—the fixed-threshold certificate of SM Sec. VI with ω→0, and the change-of-measure of SM Sec. V with s,γ→0—and verify they both yield the same exponent 2ηℓ² for, e.g., η=0.5, ℓ=1, θ=2. If the two derivations disagree in the exponential constant, the corridor argument or the certificate contains a hidden error.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I reviewed the proof steps behind the main result Eq. (4), focusing on the two most delicate parts: (i) the high-order plateau for θ>1/2, 0<η<1, and (ii) the dimensional embedding and full-rank qutrit separation. The high-order plateau rests on the change-of-measure argument in SM Sec. V. The key steps are the near-pure rigidity bound (SM S39), the corridor construction with χ_u, the exponential-martingale maximal inequality (S46), and the reverse-density estimate (S47). These are applied to a policy-specific measure Q^(t), but the final lower bound (S48) is independent of the policy, so the policy infimum can indeed be taken before t→∞. The order of limits (t→∞ at fixed s,γ, then γ↓0, s↓0) is legitimate because the threshold for 'sufficiently large t' may depend on s,γ but the inequality holds for all larger t. I also checked the quartic inequality (S28), the master inequality (S31), and the determinant-root theorem (S60); their proofs are consistent. The main limitation is the physical idealization of the loss model and unrestricted basis bandwidth, but this is an explicitly stated scope condition, not an internal inconsistency. Moreover, the unrestricted control class makes the result stronger: even with arbitrarily fast causal rotations, no policy can beat the QND rate. Thus I do not find a load-bearing concern that would invalidate the central claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes continuous qubit purification under fixed-spectrum Hermitian monitoring with detection efficiency \\eta. It defines the optimized long-time impurity-moment exponent V_\\theta(t,\\eta) over causal basis policies and claims an exact closed-form spectrum, Eq. (4): for \\theta<1/2 the rate is 8\\theta(1-\\theta)\\eta\\ell^2, for \\theta\\ge 1/2 and 0<\\eta<1 it freezes at 2\\eta\\ell^2, and at \\eta=1 the ideal-detector branch 4\\theta\\ell^2 is restored. The plateau is identified with the Bhattacharyya information rate between the two QND record laws, via the exact half-moment identity Eq. (5). The converse of QND optimality is established by a policy-uniform change-of-measure argument in SM Sec. V, with a qubit-specific event certificate in SM Sec. VI. The paper then extends the mechanism to higher dimensions: rank-two QND embeddings attain the qubit ceiling for 0<\\eta<1, while a determinant-root submartingale gives a stricter upper bound for full-rank qutrits over a finite moment interval, proving boundary\\textendash{}interior separation.","tokens_in":24858,"tokens_out":43089,"duration_ms":385285,"significance":"If correct, the result closes a gap in the theory of feedback purification under detector loss: it gives the first exact, parameter-free moment spectrum optimized over adaptive basis control, and it identifies the rare-record Bhattacharyya rate as the physical resource controlling high-order purification. The derivation is internally consistent and unusually complete for a Letter: an exact QND kernel identity, a half-moment Bhattacharyya identity, sharp trace inequalities with equality conditions, a horizon-uniform change-of-measure converse, and a determinant-root submartingale all fit together. The paper also makes falsifiable predictions (QND threshold exponent, sampling complexity, qutrit slope separation). The main caveat is that the model assumes ideal loss (full backaction with only the \\sqrt{\\eta}-scaled innovation available) and unrestricted basis bandwidth; the paper states this model but does not critically examine how additional decoherence or hidden channels would modify the exact freeze. This is a scope condition, not an internal inconsistency.","major_comments":[],"minor_comments":[{"comment":"The first inequality in Eq. (7), \\Lambda^{(d)}_\\star \\le \\Lambda^{(2)}_\\star, is not a consequence of determinant contraction; it follows from the state-space-uniform ceiling of SM Eq. (S38). The determinant calculation yields the second inequality. Please reword so the attribution is correct.","section":"Main text, Eq. (7)"},{"comment":"In Eq. (2) the exponent \\Lambda^{(2)}_\\star is written as an ordinary limit before existence is established; the SM defines it as a liminf. Define the object initially as a liminf/limsup and state that the limit is subsequently proven.","section":"Eq. (2) and SM Eq. (S1)"},{"comment":"The notation “\\rho=\\mathrm{diag}(-,0,+)(0.50,0.30,0.20)” is ambiguous. If the state is \\rho=\\mathrm{diag}(0.50,0.30,0.20) in the eigenbasis of L, write that explicitly.","section":"Fig. 3(c) caption"},{"comment":"The phrase “high-impurity records” in the Rényi discussion can be misread; clarify that it means highly mixed (large S) records, not records with high state purity.","section":"Rényi transfer discussion"},{"comment":"There are minor typographical issues: “wheredW^2_t=dt” in the introduction lacks a space, and a few equations have cramped subscripts. These do not affect the mathematics.","section":"Main text and SM typos"},{"comment":"The ideal-loss model and unrestricted basis bandwidth are stated but not discussed as limitations. A sentence noting that additional decoherence, hidden measurement channels, or finite feedback bandwidth would modify the exact 2\\eta\\ell^2 freeze would help readers gauge the physical scope.","section":"Model and control class"}],"recommendation":"minor_revision","confidential_remarks":"This is a strong Letter with a rigorous and detailed proof structure. The central claim is sound and the Supplemental Material is unusually complete. I found no load-bearing error. The minor revision is requested mainly to correct the attribution of the first inequality in Eq. (7), to make the limit definitions precise, and to clean up a few presentation issues. I do not see a need for additional technical work."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main thing you should know: this is a rigorous analytic result, not simulation numerology. It proves the exact long-time impurity-moment exponent optimized over all causal basis policies for a qubit under lossy monitoring. Equation (4) is the centerpiece, and the genuinely new part is the plateau at 2ηℓ² for all θ ≥ 1/2 when η < 1. The mechanism is rare QND records with nearly balanced evidence: they dominate every moment of order at least half, and no adaptive policy can beat that rate. I read the supplemental carefully around the change-of-measure argument (S39–S48) and the determinant contraction, and the reasoning is internally consistent. The policy infimum is taken before t→∞, the order of limits is legitimate, and I did not find a circular step or a load-bearing gap.\n\nWhat the paper does well: the half-moment identity (5) is exact and gives the plateau a clean information-theoretic meaning as the Bhattacharyya rate between the two QND record laws. The qubit reduction to a single orientation variable and the trace inequalities are used sharply. The qutrit boundary–interior separation is a nice bonus: it is honestly stated as an upper bound below an attainable rank-two ceiling, not overclaimed as a full higher-dimensional solution. The derivations are detailed enough to be checked line by line even without machine-checked proofs or public code.\n\nSoft spots, in proportion: the main physical assumption is the standard beam-splitter loss model—full backaction with only the √η-scaled innovation available. If real inefficiency includes extra decoherence or hidden measurement channels, the exact freeze at 2ηℓ² need not hold. The authors state this scope condition but do not critically examine it as a limitation. That is worth saying in the paper, but it is a limitation, not an error. The unrestricted basis bandwidth is an idealization; it makes the upper bound stronger, not weaker. The lack of public code and data is minor for an analytic theorem. My confidence is high but not certain: the stochastic calculus is long and unformalized, and I could not machine-check every step.\n\nWho this is for: quantum control theorists and experimentalists working on rapid state preparation, plus anyone interested in large deviations in continuous measurement. It deserves a serious referee. If I were the referee, I would ask for a short discussion of model robustness and perhaps a numerical check of the finite-time prefactor, but I would not ask for a rewrite of the math. Send it to peer review.","headline":"An exact optimized qubit purification spectrum under detector loss, with a rare-record plateau at the Bhattacharyya rate, and the proof holds up on inspection.","tokens_in":25364,"tokens_out":1546,"would_cite":true,"duration_ms":20306,"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":"For a qubit monitored through a lossy detector, the paper establishes the exact optimized long-time impurity-moment spectrum: rates 8θ(1−θ)ηℓ² for orders below half, freezing at the record-distinguishability rate 2ηℓ² for every order at and","keywords":["quantum purification","continuous measurement","quantum feedback","detector inefficiency","quantum nondemolition","Bhattacharyya rate","rare trajectories","impurity moments"],"falsifier":"Run a continuous homodyne measurement on a qubit at known efficiency η<1 aligned with a fixed observable, calibrate the measurement contrast ℓ from the bare backaction, and estimate the long-time slopes of -ln E[√S_t] and of -ln E[S^θ_t] for θ=1 and 2. Equation (4) predicts all three slopes equal 2ηℓ². If the θ≥1/2 slopes differ from 2ηℓ², or if any adaptive policy is found whose limiting slope exceeds 2ηℓ², the plateau claim fails.","tokens_in":24441,"feed_emoji":"⚛️","tokens_out":7687,"duration_ms":72344,"temperature":0.7,"pith_summary":"Quantum purification — turning a mixed state into a pure one by continuous measurement — is limited by how much of the measurement record actually reaches the feedback controller. The paper asks what the fastest possible exponential purification is for a qubit when the detector keeps only a fraction η of the record and the controller may choose any causal sequence of measurement orientations. It answers with an exact spectrum: for impurity moments E[S^θ] with θ<1/2, the optimal long-time rate is 8θ(1−θ)ηℓ²; for every θ≥1/2 and η<1, the rate freezes at 2ηℓ², independent of θ. The freezing mechanism is the rare trajectory whose measurement evidence nearly cancels, leaving two plausible state assignments; its probability decays at the Bhattacharyya information rate between two quantum-nondemolition record laws, a rate that no basis-feedback policy can beat. If true, the result gives a parameter-free speed limit for lossy quantum monitoring and shows that the same ceiling persists in all dimensions, with aligned QND monitoring attaining it.","feed_headline":"Lossy monitoring freezes qubit purification at rate 2ηℓ²","feed_subtitle":"Rare nearly-balanced records set the same speed limit for every high impurity moment, so smarter feedback cannot beat it.","key_machinery":"The load-bearing object is the binary log-likelihood of the two measurement eigenstates, y=(1/2) ln(p_+/p_-), with impurity S=(1/2) sech² y. Under aligned QND monitoring y evolves as a diffusion with drift 4ηℓ² tanh y; rare records keep y at O(1), and the probability of such bounded-evidence records falls at rate 2ηℓ² per unit time. Conjugating the QND generator with cosh y turns it into free heat flow with a spectral shift 2ηℓ², making the half-order identity E[√S_t]/√S_0 = e^{-2ηℓ²t} exact and transferring it to all higher moments. A policy-uniform converse follows from two sharp spectral inequalities, |T3|≤ℓS and T2S+T3²≤ℓ²S², together with a horizon-indexed change of measure on a near-pu","core_discovery":"The central claim is Eq. (4): for every mixed initial qubit under fixed-spectrum Hermitian monitoring, the optimized long-time impurity-moment exponent exists and equals 8θ(1−θ)ηℓ² for 0<θ<1/2, 2ηℓ² for θ≥1/2 with 0<η<1, and 4θℓ² for θ≥1/2 with η=1. The half-moment order θ=1/2 is the transition: below it a moving large-deviation saddle selects steadily purifying records and the rate 8θ(1−θ)ηℓ²; at and above it, the selected records have O(1) net evidence, so the conditional state remains appreciably mixed for exponentially long times and every high order costs the same 2ηℓ². This plateau is exactly the decay of the Bhattacharyya coefficient between the two QND record laws conditioned on the","pith_inferences":["The paper leaves implicit that detector efficiency enters purification only through the retained distinguishability of the record laws; a similar high-order plateau should appear for any lossy or coarse-grained measurement channel whose record pair has a computable Bhattacharyya rate, even if the prefactor sector differs.","Because the plateau makes all high moments asymptotically identical, long-time exponential rates cannot rank feedback strategies in the lossy regime; protocol comparisons should shift to finite-time prefactors, threshold first-passage times, or low-order moments, where the spectrum is order-dependent.","The qutrit boundary–interior gap suggests an operational precept for lossy state preparation: keep the conditional support on the two extremal eigenstates, since full-rank initial states are provably slower for intermediate moment orders; whether adaptive rotations can exploit this beyond the proved ceilings is a testable extension.","A direct experimental falsifier is feasible: calibrate ℓ and η, implement aligned QND monitoring, and compare the long-time slope of -ln E[√S_t] with 2ηℓ²; a mismatch beyond known backaction would indicate that the ideal-loss model omits additional decoherence or hidden measurement channels."],"forward_implications":["No causal basis-feedback policy can make any impurity moment of order θ≥1/2 decay faster than e^{-2ηℓ²t} when η<1; repeatedly measuring one fixed observable attains this rate.","The half moment is an order-parameter transition: changing the moment order below 1/2 changes the rate, while at and above 1/2 the rate is frozen, so all Rényi entropies of order α≥1/2 share the same plateau.","Perfect detection is a singular limit: the plateau turns into the rising branch 4θℓ² driven by deterministic unbiased flow, with a logarithmically wide crossover window as η→1.","In any d-dimensional system the same 2ηℓ² high-order ceiling is attainable by an extremal rank-two QND mixture supported on the monitor's smallest and largest eigenstates; full-rank states obey stricter determinant-based bounds, yielding a boundary–interior separation for qutrits.","The predicted variance-to-mean ratio Var(S^θ)/E[S^θ]^2 ≍ t^{1/2}e^{2ηℓ²t} quantifies how rare the bottleneck records are and sets the sampling requirements for experimental verification."],"fun_headline_variants":["Lossy monitoring caps qubit purification speed at 2ηℓ²","Rare balanced records impose speed limit 2ηℓ² on purification","Purification high moments freeze at 2ηℓ² under record loss","Record loss sets quantum purification speed ceiling 2ηℓ²","Universal 2ηℓ² rate bounds high-moment purification"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is the ideal beam-splitter loss model: the qubit feels the full measurement backaction while only the √η-scaled detected innovation is available, with no extra decoherence, hidden channels, feedback latency, or orientation-bandwidth restrictions; if real inefficiency adds any of these, the exact freeze at 2ηℓ² need not survive.","fun_headline_variants_meta":{"raw":{"variants":["Lossy monitoring caps qubit purification speed at 2ηℓ²","Rare balanced records impose speed limit 2ηℓ² on purification","Purification high moments freeze at 2ηℓ² under record loss","Record loss sets quantum purification speed ceiling 2ηℓ²","Universal 2ηℓ² rate bounds high-moment purification"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000496,"raw_usage":{"total_tokens":2263,"prompt_tokens":731,"completion_tokens":1532,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":475,"completion_tokens_details":{"reasoning_tokens":1450}},"tokens_in":475,"tokens_out":1532,"duration_ms":12729,"temperature":1.0,"reasoning_tokens":1450,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T07:40:50.235872+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a continuous homodyne measurement on a qubit at known efficiency η<1 aligned with a fixed observable, calibrate the measurement contrast ℓ from the bare backaction, and estimate the long-time slopes of -ln E[√S_t] and of -ln E[S^θ_t] for θ=1 and 2. Equation (4) predicts all three slopes equal 2ηℓ². If the θ≥1/2 slopes differ from 2ηℓ², or if any adaptive policy is found whose limiting slope exceeds 2ηℓ², the plateau claim fails.","supporting_citations":[],"review_version":2}