{"id":"9116e96d-4c41-47d8-ac61-780b4b268e53","arxiv_id":"2608.07868","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":14,"one_line_summary":"SPLICE simultaneously estimates the separation and relative intensity of two unequal incoherent sources with RMSE about 50 percent above the quantum limit and up to six times better than direct imaging.","lead":"This paper shows that a simple phase-sensitive fiber measurement can estimate the separation and brightness ratio of two close, faint light sources at the same time. It gets within about 50 percent of the fundamental quantum limit and beats ordinary imaging by up to sixfold.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The q-estimation claim is confounded by the fitted ε1 background: the experiment reports RMSE below the ideal QCRB at small separations, so the 'near-quantum-limit' comparison for q is not established for the actual SPLICE measurement.","rationale":"The paper is an honest experimental extension of SPLICE to two-parameter estimation, with a thoughtful treatment of crosstalk and a calibration procedure. My concern is not about intent or effort; the authors explicitly flag the sub-QCRB q behavior as parasitic. Rather, that admission shows the experiment is not testing the ideal SPLICE model for q: the QCRB in Eqs. (11)-(12) is derived for two sources with identical Gaussian PSFs, but the two beams in the Sagnac interferometer have different incidence angles (Sec. 3), and the resulting ε1 q term in Eq. (30) changes the physical state. The estimator exploits this term to achieve q RMSE below the ideal QCRB at small separations. Therefore the 'within 50% of quantum limit' statement for q is not supported by the reported comparison; the correct benchmark for this apparatus is the QCRB of the actual imperfect state. The same issue affects the DI comparison: the DI simulation assumes identical PSFs, so the claimed sixfold advantage may partly consist of comparing the ε1 information in SPLICE counts against a DI model that lacks it. This is a correctness risk in the central claim, not merely a transferability limitation. The reader's weakest assumption correctly identifies ε1; I sharpen it to an internal-validity problem. The verdict remains CONDITIONAL: the δ advantage may survive, but acceptance should require a revised analysis using the actual-state QCRB and a DI baseline that includes the same incidence-angle imperfection.","tokens_in":12408,"tokens_out":12632,"duration_ms":144987,"concrete_test":"Using the stored count data, construct the actual two-source state including the measured incidence-angle difference and fitted ε0, ε1, and compute the quantum Fisher information matrix for (δ,q) on that state; then compare the reported experimental q RMSE (and δ RMSE at small separations) against the corrected QCRB. If the experimental RMSE is no longer below the corrected bound, the apparent sub-QCRB q precision is fully explained by the q-dependent crosstalk, and the ideal-state QCRB comparison in Sec. 4 should be replaced by this actual-state bound in the paper's claims.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's headline comparison to the quantum limit is internally compromised for the q parameter. Section 4 reports that the experimental RMSE for q falls below the QCRB of Eqs. (11)-(12) at small separations and attributes this to a 'parasitic effect' of the q-dependent crosstalk ε1 q introduced in Eq. (30). The QCRB is a lower bound for any measurement on the ideal two-Gaussian-source state; an RMSE below it means either the physical state is not that ideal state (the two beams have different incidence angles, so their PSFs are not identical) or the estimator is biased and the CRB comparison is being applied outside its valid regime. In either case, the observed q precision is not evidence that SPLICE operates within roughly 50% of the quantum limit for q; it is evidence that an uncontrolled apparatus degree of freedom is carrying information. Because ε1 is fitted from the same calibration data used by the estimator (Eq. 31), the claimed q RMSE and the small-δ δ RMSE are entangled with this fitted background. The paper's own defense, that δ RMSE at δ/σ>0.2 goes below the background bound, addresses δ only; the q claim, and therefore the 'near-quantum-limit multiparameter estimation' headline, is not established for the actual experiment. The direct-imaging simulation likewise does not include the incidence-angle difference, so the claimed sixfold advantage may partly compare SPLICE's parasitic information against a DI model that lacks it.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports an extension of the SPLICE technique to simultaneous estimation of the separation δ and relative intensity q of two incoherent, unequal-intensity sources. After deriving quantum and classical Cramér-Rao bounds for Gaussian point-spread functions, the authors implement two phase-shifted fiber-mode projectors approximating the iHG basis and perform a single-photon experiment with 810 nm light. A polynomial response model with seven fitted coefficients per projector (Eq. 31) is calibrated on random halves of the data and tested on the remaining halves. The paper reports RMSE within about 50% of the quantum limit and up to a sixfold improvement over direct imaging, with the advantage growing for smaller separations and larger intensity imbalances.","tokens_in":12916,"tokens_out":8690,"duration_ms":92199,"significance":"If the central claims survive scrutiny, the work provides a comparatively simple experimental route to near-quantum-limited multiparameter estimation for unbalanced sources, with concrete relevance to exoplanet imaging and microscopy. Strengths include held-out random-partition evaluation, Monte Carlo simulations for the biased regime, and explicit modeling of crosstalk. The demonstration that a fiber collimator and phase shifter can jointly estimate δ and q is interesting and experimentally valuable. The main caveat is that the relative-intensity claim is currently entangled with a fitted parasitic background, and the direct-imaging comparison does not include that same background.","major_comments":[{"comment":"The q RMSE claim is not established as a near-quantum-limit result. The paper reports experimental RMSE for q below the ideal QCRB at small separations and attributes this to the ε1 q background in Eq. (30). Since the QCRB in Eqs. (11)-(12) is a lower bound for unbiased estimators on the ideal two-Gaussian-source state, an RMSE below it implies either a different physical state (e.g., the different incidence angles mentioned in the text) or a biased estimator; in either case the comparison is outside its valid regime. The fitted ε1 is part of the same calibration procedure, so the apparent q precision is carried by an apparatus degree of freedom rather than by the SPLICE projection. The paper's defense at δ/σ>0.2 is formulated for the separation estimator, not for q, and no analogous regime separation is shown for q. Please re-analyze the q data with the parasitic background modeled explicitly, or restrict the quantum-limit claim to δ and to simulations.","section":"Section 4, Eq. (30)"},{"comment":"The claimed sixfold improvement over direct imaging may be inflated by an asymmetry in the comparison. The DI simulation uses the ideal intensity profile of Eq. (23) and does not include the incidence-angle mismatch or the q-dependent crosstalk that the SPLICE experiment enjoys through ε1 in Eq. (30). Thus the comparison partly credits SPLICE with information that comes from an uncontrolled, fitted background rather than from the SPLICE measurement principle. A fair comparison would include the same crosstalk and calibration model in the DI simulation, or compare against an actual DI measurement on the same apparatus, before concluding that SPLICE itself provides the sixfold improvement.","section":"Section 4, DI Monte Carlo simulation"},{"comment":"The calibration model contains fourteen free parameters (seven per projector) fitted to ten calibration trials per setting, but only the χ² values (1.46 and 1.93) are reported. The stability of the fitted coefficients, especially ε1, is load-bearing because ε1 is exactly the term that produces the sub-QCRB q RMSE. Please report the fitted coefficients with uncertainties and correlations, and show that the held-out RMSE is robust to the random-partition procedure, e.g., by reporting the spread across the 20 partitions or a bootstrap analysis.","section":"Section 4, Eq. (31)"}],"minor_comments":[{"comment":"The balanced-source limit is written as (Q^{-1})_{δδ}=4σ²; it should include the factor 1/N from Eq. (11). This typo appears in a central bound and should be corrected.","section":"Section 2.1, after Eq. (11)"},{"comment":"The projector positions g=±σ and f=±2σ are said to be optimal by reference to thesis [51]; since this optimality underpins the comparison with the QCRB, a derivation or a peer-reviewed reference would make the paper more self-contained.","section":"Section 2.2"},{"comment":"The definition of M and the per-photon normalization of J should be stated explicitly; currently N only appears later in Eqs. (20)-(21), which makes the formula easy to misread.","section":"Section 2.2, Eq. (18)"},{"comment":"The phrase 'an root mean squared error' should be 'a root-mean-square error'; the abstract and text use 'root mean squared error' inconsistently.","section":"Section 5"},{"comment":"The statement that data are not publicly available will hinder verification; for a metrology experiment, releasing the raw count data and fitted coefficients would materially strengthen reproducibility.","section":"Data Availability"}],"recommendation":"major_revision","confidential_remarks":"I am not recommending rejection because the experimental design and held-out evaluation are sound and the central δ-advantage claim is credible. The main issue is that the headline claim for q relies on a fitted parasitic background, which I believe can be addressed with a more careful analysis and a restricted or corrected claim. I would also encourage the editor to ask the authors to report the fitted coefficients with uncertainties and to consider making the calibration data available. The paper relies heavily on Ref. [36] for SPLICE theory and Ref. [51] for projector optimality; these are defensible citations, but the authors should be asked whether the key optimality claim can be made self-contained."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth reading if you work on sub-Rayleigh imaging. The new thing is the first SPLICE implementation that simultaneously estimates separation and relative intensity of unequal-brightness sources. The authors do careful held-out calibration, Monte Carlo simulations for DI and SPLICE, and they are transparent about the crosstalk model. That is real experimental work.\n\nThe main soft spot is the quantum-limit comparison itself. In Figure 6, the measured RMSE for q drops below the QCRB at small separations, which is impossible on the ideal two-Gaussian-source state. The authors attribute this to a q-dependent crosstalk background (ε1 q in Eq. 30) caused by slightly different incidence angles of the two beams. That means the q precision is not coming from SPLICE; it's coming from an extra apparatus degree of freedom. So the 'near-quantum-limit' and the '50% larger than the quantum limit' claims cannot be asserted for q. The same parasitic background likely contributes to the claimed sixfold advantage over DI, because the DI simulation uses ideal PSFs with no such background. The paper's own defense works for δ for δ/σ > 0.2, but the abstract and conclusion make a broader claim.\n\nMinor issues: Eq. 11's balanced-source limit is missing a factor of N (should read 4σ²/N). The optimal projector positions are justified by a thesis citation, not reproduced in the paper. The calibration model has 14 free parameters; held-out evaluation helps, but the data are not public, so independent checks aren't possible.\n\nOverall, this is a credible experimental extension with a clear statement of its own limitation. But the headline claim overreaches. A referee should ask them to separate the parasitic contribution from the SPLICE contribution and to restrict the quantum-limit comparison to the regime where the ideal model actually applies. That's a major revision, not a rejection.\n\nI'd send it to peer review. For a reading group, it's a useful case study of how experimental imperfections can create apparent sub-QCRB performance.","headline":"First SPLICE experiment for two-parameter estimation, but the 'near-quantum-limit' q claim is undermined by a parasitic crosstalk that the authors themselves document.","tokens_in":13457,"tokens_out":5867,"would_cite":true,"duration_ms":53968,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A phase-sensitive SPLICE measurement estimates both the separation and relative intensity of two unresolved, unequal-intensity sources with RMSE about 50 percent above the quantum limit and up to sixfold better than direct imaging.","keywords":["sub-Rayleigh imaging","spatial-mode demultiplexing","SPLICE","multiparameter estimation","quantum Fisher information","Cramér-Rao bound","unequal-intensity sources","single-photon regime"],"falsifier":"Repeat the RMSE measurement at $\\delta/\\sigma \\approx 0.1$ and $q \\approx 0.1$ with the two beams' incidence angles matched so that $\\epsilon_1 \\approx 0$ (removing the $q$-dependent background in Eq. (30)); if the separation RMSE no longer beats direct imaging by the predicted factor, or if the $q$ RMSE jumps above the quantum bound, the advantage was carried by parasitic crosstalk rather than by the SPLICE projection itself.","tokens_in":12251,"feed_emoji":"🔭","tokens_out":11565,"duration_ms":105463,"temperature":0.7,"pith_summary":"Two incoherent point sources that sit closer than the Rayleigh–Abbe limit are normally measured by fitting their blurred intensity profile, a method whose error explodes as the separation shrinks and the brightnesses become unequal. This paper claims that a phase-sensitive scheme called SPLICE—a fiber collimator plus a phase-shifting glass slide—can estimate both the separation and the relative intensity at the same time, reaching within about 50 percent of the quantum Cramér-Rao bound and beating direct imaging by up to a factor of six in the tested range. The practical payoff is that near-optimal superresolution does not require a custom mode sorter; a simple interferometric edge projection is enough. The paper also identifies cross-talk in the projector as the quantity that caps the achievable advantage, predicting that better extinction ratios would directly translate into larger gains for dim companions such as exoplanets.","feed_headline":"Phase-shifter imaging beats direct imaging sixfold on dim sources","feed_subtitle":"A fiber collimator and glass slide extract near-quantum-limit information about close, dim sources.","key_machinery":"The load-bearing object is the SPLICE mode projector, whose transverse profile is a Gaussian centered at fiber position $f$ with a sign flip at phase-shifter edge $g$: $\\phi(x) = \\mathrm{sgn}(x-g)\\, e^{-(x-f)^2/(4\\sigma^2)}/(2\\pi\\sigma^2)^{1/4}$. Two such projectors, with $(g,f) = (\\pm\\sigma, \\pm 2\\sigma)$, approximate the ideal interferometric Hermite-Gauss modes $(HG_1 \\pm HG_2)/\\sqrt{2}$; the paper shows their detection probabilities equal the ideal ones up to a factor $4/(\\pi e)$, so the scheme extracts about 46 percent of the total quantum Fisher information. The argument is carried by the Fisher information matrix of these two-outcome projectors and by a calibration model, Eq. (31), that fits seven coefficients $\\epsilon_0, \\epsilon_1, c_2, c_3, c_{40}, c_{41}, c_{42}$ to account for cross-talk from the $HG_0$ mode and other imperfections. The same matrix gives the comparison to direct imaging and the cross-talk ceiling $\\sqrt{3/(2e\\pi^2 \\epsilon)} \\approx 7.5$ on the achievable advantage.","core_discovery":"The central claim is that the two-parameter estimation problem for a Gaussian point-spread function is solved, to within a factor $4/(\\pi e) \\approx 0.46$ of the ideal information, by two SPLICE projectors: an edge phase shift at $x = \\pm\\sigma$ followed by a single-mode fiber at $x = \\pm 2\\sigma$. With those projectors the photon probabilities are $P_{\\phi\\pm} = q(1-q)\\delta^2/(2\\pi e \\sigma^2) \\pm q(1-q)(1-2q)\\delta^3/(2\\pi e \\sigma^3) + O(\\delta^4)$, mirroring the ideal interferometric Hermite-Gauss probabilities. In experiment, using roughly 320,000 single-photon detections per estimate and a seven-coefficient polynomial calibration model, the separation and relative-intensity RMSEs track the SPLICE Cramér-Rao bound, which lies below the direct-imaging bound; the reported RMSE is about 50 percent above the quantum limit and up to sixfold below direct imaging. The paper further reports that the $q$-dependent cross-talk term $\\epsilon_1 q$, arising from slightly different beam incidence angles, is exploited by the least-squares estimator and makes the experimental $q$ RMSE appear better than the quantum bound at small separations, while the separation advantage at $\\delta/\\sigma > 0.2$ is attributed to SPLICE itself.","pith_inferences":["The paper treats $\\epsilon_1 q$ as a nuisance, yet shows it improves $q$ estimation at small separations; a controlled version of this $q$-dependent background could be engineered as an extra information channel, a testable modification the paper does not pursue.","Because the advantage ceiling is set by $\\epsilon$, swapping the cover slips for a higher-extinction mode sorter is a direct way to test the predicted scaling: if the RMSE reduction does not approach about 40 at $\\epsilon \\approx 10^{-6}$, the crosstalk model would need revision.","The paper assumes a known center of mass; a natural extension is a third parameter (the centroid), where the same Fisher-matrix machinery would need additional projectors, and the ratio-symmetric structure of the probabilities suggests how they should be placed."],"forward_implications":["At the tested separations ($\\delta/\\sigma$ down to about 0.02–0.1) and brightness ratios ($q$ from 0.1 to 0.5), the reported RMSE for both parameters roughly follows the SPLICE bound and beats the direct-imaging bound.","For an intensity imbalance of 100:1, the predicted RMSE reduction over direct imaging is about 60 times; for $10^8$:1, about $6\\times 10^4$, if crosstalk is low.","Improving the projector extinction ratio from $10^3$ to $10^6$ would raise the maximum advantage from about 7.5 to about 40.","For equal-intensity sources, SPLICE still works, but its Cramér-Rao bound is $2/(\\pi e)$ times the quantum limit because the projectors are optimized for unequal intensities."],"supporting_citations":[{"why":"Establishes the quantum limit for estimating the separation of two incoherent point sources, the baseline against which SPLICE is compared.","marker":"[4]"},{"why":"Identifies the interferometric Hermite-Gauss basis as optimal for Gaussian point-spread functions, the target the SPLICE projectors approximate.","marker":"[19]"},{"why":"Introduces SPLICE and demonstrates sub-Rayleigh separation estimation with phase information, the method this paper extends to two parameters.","marker":"[24]"},{"why":"Supplies the multiparameter quantum Cramér-Rao formulas for separation and relative intensity used in Section 2.1.","marker":"[35]"},{"why":"Gives the SPLICE projector optimization and the $4/(\\pi e)$ information factor that the experimental scheme is based on.","marker":"[36]"},{"why":"Recent experiment estimating separation and relative intensity with a known center of mass, providing the closest prior demonstration and comparison point.","marker":"[44]"},{"why":"Recent experiment estimating separation, relative intensity and centroid together, the neighbouring multiparameter problem this paper builds toward.","marker":"[45]"}],"fun_headline_variants":["SPLICE imaging nears quantum limit for unequal sources","Sixfold gain over direct imaging for sub-Rayleigh sources","Near-quantum-limit multiparameter imaging of dim sources","Unequal-source imaging nears quantum limit with SPLICE"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The near-quantum-limit RMSE depends on a seven-coefficient polynomial response model, Eq. (31), fit to calibration trials; if the detector's true response, particularly the $q$-dependent cross-talk term $\\epsilon_1$, is not stable between calibration and use, the reported advantage over direct imaging and the comparison to the quantum limit will not transfer to another apparatus.","fun_headline_variants_meta":{"raw":{"variants":["SPLICE imaging nears quantum limit for unequal sources","Sixfold gain over direct imaging for sub-Rayleigh sources","Near-quantum-limit multiparameter imaging of dim sources","Unequal-source imaging nears quantum limit with SPLICE"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000249,"raw_usage":{"total_tokens":1598,"prompt_tokens":1038,"completion_tokens":560,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":654,"completion_tokens_details":{"reasoning_tokens":490}},"tokens_in":654,"tokens_out":560,"duration_ms":6032,"temperature":1.0,"reasoning_tokens":490,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T00:45:52.182387+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the RMSE measurement at $\\delta/\\sigma \\approx 0.1$ and $q \\approx 0.1$ with the two beams' incidence angles matched so that $\\epsilon_1 \\approx 0$ (removing the $q$-dependent background in Eq. (30)); if the separation RMSE no longer beats direct imaging by the predicted factor, or if the $q$ RMSE jumps above the quantum bound, the advantage was carried by parasitic crosstalk rather than by the SPLICE projection itself.","supporting_citations":[{"cited_title":"Optimal measurements for resolution beyond the rayleigh limit,","cited_arxiv_id":null,"evidence_quote":"Identifies the interferometric Hermite-Gauss basis as optimal for Gaussian point-spread functions, the target the SPLICE projectors approximate."},{"cited_title":"Beating rayleigh’s curse by imaging using phase information,","cited_arxiv_id":null,"evidence_quote":"Introduces SPLICE and demonstrates sub-Rayleigh separation estimation with phase information, the method this paper extends to two parameters."},{"cited_title":"Realistic sub-rayleigh imaging with phase-sensitive measurements,","cited_arxiv_id":null,"evidence_quote":"Gives the SPLICE projector optimization and the $4/(\\pi e)$ information factor that the experimental scheme is based on."}],"review_version":1}