{"id":"53258d1d-34d2-4b06-897b-4f03e3ede9e7","arxiv_id":"2509.04754","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Quantum state smoothing experimentally beats filtering for an optical parametric oscillator, recovering 10.3% of lost purity and 7.6% of lost squeezing.","lead":"Splitting an optical cavity's output between two detectors, one acting as a hidden observer, the authors experimentally show that quantum state smoothing, using both past and future records, restores 10.3% more purity and 7.6% more squeezing than standard filtering. This is the first demonstration of smoothing for a linear Gaussian system, pointing toward post-processing methods that recover lost quantum resources.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'true' state is an approximation because η_tot≈0.86; the headline recovery percentages inherit an unquantified systematic uncertainty from this assumption.","rationale":"The reader's weakest_assumption identifies the same load-bearing concern: the 'true' state is only an approximation because η_tot≈0.86, and the absolute recovery percentages depend on this assumption. I checked whether there were other, more serious flaws—e.g., correlated measurement noise between Alice and Bob from the shared vacuum port of the beam splitter. That concern does not land: unitarity of the beam splitter makes the two homodyne noises uncorrelated, consistent with Eq. (42). The smoothing-vs-filtering advantage itself is supported by parameter-free theory and multiple independent metrics, so the central qualitative claim is secure. The weakness is specifically the quantitative headline numbers, which lack a systematic uncertainty analysis for the true-state definition. This is best addressed by reanalyzing with an alternative unraveling of the residual loss. The reader's CONDITIONAL verdict already captures this, so no verdict change is needed.","tokens_in":22202,"tokens_out":16411,"duration_ms":173654,"concrete_test":"Recompute the maximum purity and squeezing recoveries from the raw data under an alternative model in which the residual 14% loss is treated as an additional unit-efficiency homodyne record at a fixed phase, and repeat for phases that extremize the recovery. If the reported 10.3% and 7.6% values shift by more than the quoted error bars, the headline percentages are not robust to the unraveling assumption; if they stay within error, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claims (10.3%±1.6% purity recovery and 7.6%±2.6% squeezing recovery) are defined relative to the 'true' state ρ_T, which is the state conditioned on both Alice's and Bob's homodyne records with total efficiency η_tot≈0.86. As the paper concedes, this is not the ideal pure true state of quantum state smoothing because ~14% of the information is lost to unobserved channels. If that residual loss were assigned a different unraveling—e.g., an additional ideal homodyne record at some phase—P(ρ_T) would shift, changing both the denominator P(ρ_T)−P(ρ_F) and, through the nonlinear determinant in Eq. (7), the numerator R_P and R_S. The reported error bars are statistical only and do not include systematic uncertainty in η_tot, the unraveling choice, or the ±2° phase noise. The core smoothing-versus-filtering comparison is robust because both estimates share the same theoretical V_T, but the absolute recovery percentages are not.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports an experimental demonstration of quantum state smoothing for a continuous-variable linear-Gaussian system: an optical parametric oscillator whose output is split into two homodyne channels, one observed by Alice (the estimator) and the other by a hidden observer Bob. Alice's smoothed estimate uses her past and future records, and is compared with her causal filtered estimate. The 'true' state is defined as Bob's state conditioned on both records, with total efficiency η_tot≈0.86. The authors measure purity, average trace-squared deviation (TrSD) from the true state, squeezing, and anti-squeezing as functions of Alice's efficiency and of both homodyne angles. Parameter-free theory curves (from independently measured efficiencies, loss, gain, and angles) agree with the data across four metrics and two parameter scans. The headline results are a maximum purity recovery of 0.016±0.003 (10.3%±1.6% of the total recoverable purity) and a maximum squeezing recovery of 0.006±0.002 (7.6%±2.6% of the total recoverable conditional squeezing). The authors conclude that smoothing yields purer, more squeezed states that are better estimates of the true state than filtering.","tokens_in":22396,"tokens_out":7237,"duration_ms":69418,"significance":"If the results hold, this is the first experimental realization of quantum state smoothing for a linear-Gaussian continuous-variable system, a milestone that had been awaited since the theoretical proposals. The paper's strengths are notable: the theory curves are parameter-free predictions using independently measured quantities; four complementary metrics (purity, TrSD, squeezing, anti-squeezing) are compared; the TrSD is computed independently of the reconstructed covariances as an internal consistency check; and the optimal measurement settings predicted by theory are tested rather than fitted. The qualitative conclusion that smoothing outperforms filtering is robust and well supported. The main weakness is that the absolute recovery percentages in the abstract are defined relative to an approximate 'true' state whose residual impurity is not propagated into systematic uncertainties.","major_comments":[{"comment":"The headline quantitative claims (abstract: 10.3%±1.6% purity recovery, 7.6%±2.6% squeezing recovery) are defined relative to the 'true' state ρ_T, which is conditioned on both Alice's and Bob's homodyne records with η_tot≈0.86. The paper concedes that η_tot is not unity, so ρ_T is not the ideal pure true state of quantum state smoothing. The reported error bars are statistical only and do not include the systematic effect of the residual ~14% loss, the choice of unraveling for that loss, or the ±2° phase noise. A different modeling of the unobserved loss would shift P(ρ_T) and therefore both the numerator and denominator of the recovery fractions. This does not undermine the qualitative smoothing-vs-filtering comparison, which is robust, but it does affect the absolute percentages featured in the abstract. Please quantify this systematic uncertainty or rephrase the headline claim to avo","section":"Results – Experimental setup; Discussion"},{"comment":"In Eq. (7), the reconstructed covariances \\tilde V_F and \\tilde V_S are obtained by adding the theoretically calculated V_T to the experimental mean-square-error matrices. This choice protects the filtering-vs-smoothing comparison from common noise in V_T, but the absolute purities and squeezing values used in Figs. 4–6, and especially the ratios R_P/[P(ρ_T)-P(ρ_F)] and R_S/[S(ρ_F)-S(ρ_T)], inherit any model error in V_T. Since V_T also enters the denominators, the percentage recoveries are not independent of the theoretical model. A systematic propagation of uncertainties in η_T, η_B, ξ, and V_T should be reported, or the claims should be restricted to the directly measured differences rather than the normalized percentages.","section":"Methods – Reconstruction of covariance, Eq. (7)"},{"comment":"In the phase-scan measurements (Fig. 5 and Fig. 6), θ_B is swept at ~140 Hz and the data are binned into 1° blocks, with a stated control-loop phase noise of ±2°. These uncertainties are not propagated into the error bars shown in Fig. 5. Given that the theoretical predictions are evaluated at nominal angles, a brief statement that the agreement is robust to these phase uncertainties—or inclusion of them in the error bars—would strengthen the quantitative comparison.","section":"Data acquisition (phase scanning)"}],"minor_comments":[{"comment":"The x-axis label reads 'Measurement efficiency, η_A', but the tick values are 0, 45, 90, 135, 180 degrees and the caption describes a scan over θ_A. This appears to be a copy-paste from Fig. 4 and should be changed to 'Measurement angle, θ_A (deg)'.","section":"Fig. 6 caption and axis"},{"comment":"The covariance decomposition is misstated. The text says V_unc = V_S + Cov(⟨x̂⟩_unc, ⟨x̂⟩_S) = V_S + Cov(⟨x̂⟩_S,0) since ⟨x̂⟩_unc=0. As written, Cov(⟨x̂⟩_S,0)=0, which would imply V_unc=V_S and contradict the data. The intended relation is V_unc = V_S + Cov(⟨x̂⟩_S,⟨x̂⟩_S). Similarly, Eq. (8) should be \\tilde V_T = V_unc − Cov(⟨x̂⟩_T,⟨x̂⟩_T), not Cov(⟨x̂⟩_T,0).","section":"Main text near Fig. 3; Methods Eq. (8)"},{"comment":"The legend entry 'Unconditioning' should probably be 'Unconditional' or 'Unconditional state' for consistency with the text.","section":"Fig. 4 legend"}],"recommendation":"major_revision","confidential_remarks":"The paper is a strong experimental demonstration with parameter-free theory and multiple independent metrics; the core qualitative claim is solid. My concerns are about the systematic uncertainty in the absolute recovery percentages and the approximate 'true' state; these are addressable in revision. I do not see grounds for rejection. The self-citation to the authors' earlier theory papers is appropriate here. The paper is well within the scope of the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper for one reason: it is the first experimental demonstration of quantum state smoothing, not in a toy system but in a properly characterized OPO with two homodyne channels. The experiment is careful, the theory curves are parameter-free, and the data agree across four metrics (purity, TrSD, squeezing, anti-squeezing) and two parameter scans. The independent computation of TrSD without reconstructing covariances is a nice internal check that the estimation pipeline is consistent.\n\nWhat is genuinely new here is the realization itself. The theory (largely by the same group, Refs 26, 27, 37) predicted smoothing would beat filtering for linear Gaussian systems; this paper does it, quantifies the gains (10.3% purity recovery, 7.6% squeezing recovery), and maps the measurement-angle dependence including the optimal Bob angle. The authors also show that smoothed states are better estimates of the hidden true state, not just purer. That is a clean, non-obvious result.\n\nNow the soft spots. The 'true' state used as ground truth is the state conditioned on both Alice's and Bob's homodyne records, with total efficiency about 0.86. The paper says that is 'close enough' to pure to use the term 'true'. That is a reasonable practical choice, but the absolute recovery percentages depend on this benchmark, and the reported error bars are statistical only. A more careful treatment would propagate the uncertainty in eta_tot and the phase noise into the recovery numbers. This is a real caveat, but it does not undermine the core smoothing-versus-filtering comparison, because both estimates share the same theoretical covariance V_T. The worry about 'different unraveling' of the residual loss does not really land: the unobserved loss is not measured, and the conditional states are defined relative to the actual records, not to a hypothetical record from the loss channel. The limitation is more mundane: the benchmark is slightly mixed, so the absolute numbers are approximate.\n\nThe other practical issue is that no code or processed data are provided, so independent verification of the estimation pipeline is not possible from the paper alone. That is increasingly expected in modern experimental papers, but it is not fatal here given the internal consistency checks.\n\nOverall, this is a solid paper that deserves a serious referee. It will be of real interest to anyone working on quantum trajectories, continuous measurement, or post-processing in quantum optics. I would recommend sending it to peer review; the main revisions I would ask for are to quantify the systematic uncertainty in the recovery percentages and to clarify that the 'true' state is an approximation, ideally with a short discussion of how the numbers would shift if the benchmark were pure. After that, it should be publishable.","headline":"First experimental realization of quantum state smoothing in a linear Gaussian optical system; the data strongly support the theory, and the central result is solid, though the absolute recovery percentages inherit an unquantified systematic caveat from the approximate 'true' state.","tokens_in":22935,"tokens_out":2532,"would_cite":true,"duration_ms":28203,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P16","81P15","81V80"],"pacs":["03.65.Ta","42.50.Dv","42.50.Lc"],"model":"deepseek-v4-flash","headline":"Using both past and future measurement records, quantum state smoothing experimentally recovers more purity and squeezing in a continuously monitored optical cavity than conventional filtering, and gives a better estimate of the hidden true","keywords":["quantum state smoothing","quantum filtering","linear Gaussian systems","optical parametric oscillator","continuous measurement","squeezing","purity recovery","homodyne detection"],"falsifier":"In a repeat run, deliberately mis-set the hidden detector's angle by about 20 degrees and see whether the purity recovery drops in the direction and magnitude predicted by the smoothing equations. If a mis-set angle leaves the recovery unchanged—or if smoothing stops beating filtering when the residual loss is assigned to a different measurement model—the assumption-dependence of the central claim collapses.","tokens_in":22102,"feed_emoji":"⚛","tokens_out":8268,"duration_ms":78330,"temperature":0.7,"pith_summary":"Quantum state filtering, the standard way to track a monitored quantum system, uses only past records and leaves the estimated state impure when information escapes into unobserved channels. This paper experimentally shows that quantum state smoothing, which also uses future records and assumes a specific hidden-observer measurement on the lost channel, retroactively recovers some of that lost purity and squeezing. In an optical parametric oscillator whose output is split between two homodyne detectors, the smoothed state is purer, more squeezed, and on average closer to the dual-detector hidden state than the filtered state is. The largest measured improvements are 10.3%±1.6% of the recoverable purity and 7.6%±2.6% of the recoverable squeezing. The result makes smoothing a practical post-processing tool for quantum information tasks that can tolerate delayed estimates.","feed_headline":"Smoothing recovers purity and squeezing that filtering loses","feed_subtitle":"Future homodyne records retroactively restore 10.3% of lost purity and 7.6% of lost squeezing.","key_machinery":"The load-bearing mechanism is the linear-Gaussian smoothing identity that combines a filtered estimate from the past record with a retrofiltered element from the future record: V_S = V_T + (V_F - V_T)[I + Λ_R(V_F - V_T)]^{-1}, with Λ_R = (V_R + V_T)^{-1} built from the future record and the assumption that the hidden channel was a homodyne measurement at a specified angle. This identity converts a causal trajectory into a retroactive estimate whose covariance is smaller than the filtered covariance, which is why purity and squeezing recover, and why the smoothed mean lands closer to the hidden true state.","core_discovery":"The paper's central claim is that the acausal estimator called quantum state smoothing outperforms causal quantum filtering on three useful metrics in a real, linear-Gaussian optical system. The experiment continuously monitors an optical parametric oscillator through two homodyne detectors; one measurement record is used for estimation, while the other detector's record is withheld. The state conditioned on both records plays the role of the true state. Against that benchmark, the smoothed state—formed from the past record, the future record, and the assumption that the hidden channel was a homodyne measurement at a known angle—has greater purity, greater squeezing, and a smaller average tr","pith_inferences":["If the claim generalises, a natural test the paper does not report is to vary the total efficiency downward and check that the recovered-fraction numbers move as the 0.86-efficiency benchmark predicts; if they do not, the absolute percentages reflect the benchmark rather than a universal feature.","If the claim transfers to other linear-Gaussian platforms—mechanical resonators, circuit-QED readout, atomic ensembles—the same covariance equations would make this demonstration a generic recipe for recovering purity from stored continuous-measurement records.","Beyond the paper's stated scope, the experiment implies a diagnostic use: the hidden-observer assumption that yields the best smoothed estimate is the unraveling actually operative in the loss channel, so smoothing could be used to characterise loss channels rather than only to compensate for them."],"forward_implications":["Stored measurement records can be reprocessed after the fact to retroactively purify squeezed states, so applications that keep records—such as measurement-based quantum computing—can use smoothing as a post-processing resource-recovery step.","At every tested efficiency and angle combination, smoothing performs at least as well as filtering, so switching from filtering to smoothing carries no penalty when real-time tracking is not required.","The amount of recovered purity and squeezing is controlled jointly by both measurement settings, and the optima for purity and squeezing are not reached simultaneously, so applications must choose which resource to favour.","Smoothed estimates are not merely cleaner-looking states: they reduce the average squared deviation from the hidden true state, giving a statistically better estimate of the system's underlying dynamics."],"supporting_citations":[{"why":"Introduces quantum state smoothing and the proposal that it gives purer estimates than filtering for continuously monitored quantum systems.","marker":"[25]"},{"why":"Supplies the linear-Gaussian quantum state smoothing equations used to process the homodyne records in this experiment.","marker":"[26]"},{"why":"Calculates purity recovery for a similar system and frames the optimal-unraveling interpretation the paper extends.","marker":"[27]"},{"why":"Provides the Gaussian-state formalism used to describe every estimated state by a mean and covariance matrix.","marker":"[34]"},{"why":"Gives the input-output model of the optical parametric oscillator, the source of the squeezed states studied here.","marker":"[35]"},{"why":"Unifying theory showing that filtering and smoothing estimates minimise the expected trace-squared-deviation cost, justifying the paper's comparison metric.","marker":"[36]"},{"why":"Proves that average trace-squared deviation reduces to a purity difference, the identity used for the theoretical TrSD curves.","marker":"[37]"}],"fun_headline_variants":["Quantum smoothing recovers purity and squeezing that filtering loses","Retroactive quantum tracking restores lost purity and squeezing","Smoothing beats filtering: 10.3% purity, 7.6% squeezing restored","Acausal smoothing recovers more quantum info than causal filtering"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The 'true' state used to score the estimates is itself only an estimate: it conditions on both homodyne records but still assumes the residual 14% loss can be neglected, so a different model of that loss could shift the reported recovery percentages.","fun_headline_variants_meta":{"raw":{"variants":["Quantum smoothing recovers purity and squeezing that filtering loses","Retroactive quantum tracking restores lost purity and squeezing","Smoothing beats filtering: 10.3% purity, 7.6% squeezing restored","Acausal smoothing recovers more quantum info than causal filtering"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000176,"raw_usage":{"total_tokens":1125,"prompt_tokens":741,"completion_tokens":384,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":485,"completion_tokens_details":{"reasoning_tokens":321}},"tokens_in":485,"tokens_out":384,"duration_ms":4060,"temperature":1.0,"reasoning_tokens":321,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T05:56:09.172166+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In a repeat run, deliberately mis-set the hidden detector's angle by about 20 degrees and see whether the purity recovery drops in the direction and magnitude predicted by the smoothing equations. If a mis-set angle leaves the recovery unchanged—or if smoothing stops beating filtering when the residual loss is assigned to a different measurement model—the assumption-dependence of the central claim collapses.","supporting_citations":[{"cited_title":"& Wiseman, H","cited_arxiv_id":null,"evidence_quote":"Introduces quantum state smoothing and the proposal that it gives purer estimates than filtering for continuously monitored quantum systems."},{"cited_title":"T., Chantasri, A","cited_arxiv_id":null,"evidence_quote":"Supplies the linear-Gaussian quantum state smoothing equations used to process the homodyne records in this experiment."},{"cited_title":"T., Chantasri, A","cited_arxiv_id":null,"evidence_quote":"Calculates purity recovery for a similar system and frames the optimal-unraveling interpretation the paper extends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Gaussian-state formalism used to describe every estimated state by a mean and covariance matrix."},{"cited_title":"& Gardiner, C","cited_arxiv_id":null,"evidence_quote":"Gives the input-output model of the optical parametric oscillator, the source of the squeezed states studied here."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Unifying theory showing that filtering and smoothing estimates minimise the expected trace-squared-deviation cost, justifying the paper's comparison metric."},{"cited_title":"T., Guevara, I","cited_arxiv_id":null,"evidence_quote":"Proves that average trace-squared deviation reduces to a purity difference, the identity used for the theoretical TrSD curves."}],"review_version":1}