{"id":"d8923238-c35c-4103-82bc-68931c804636","arxiv_id":"2505.13512","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A dimensionless convergence metric derived from Gaussian Process Regression predicts sufficient measurement duration in time-resolved SANS, but the claimed universal power-law scaling largely follows from Poisson statistics and a data-reduction error floor.","lead":"This paper introduces a dimensionless metric, built on Gaussian Process Regression, for deciding when a time-resolved neutron scattering measurement has run long enough. The metric decays as a power law with time, and after per-system normalization it looks similar across several soft-matter systems, prompting the authors to propose early forecasting of sufficient exposure.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. 4 treats cumulative profiles I(t_i) and I(t_{i-1}) as independent, but Section II defines them as cumulative; the variance sum is incorrect, so the t^-2 scaling and universal exponent are not established.","rationale":"The reader correctly identifies the zero-count error floor as a fragile piece of the exponent derivation. My stress-test goes further: the paper's own cumulative definition makes Eq. 4 internally inconsistent, because the variance of a difference of cumulative totals is not the sum of their variances. This is a more direct and more load-bearing flaw than the instrument-specific zero-count convention, since even with the zero-count floor the predicted scaling changes once the covariance is accounted for. I therefore keep the CONDITIONAL verdict, but the required condition should be an explicit re-derivation or simulation of Eq. 2 under cumulative Poisson statistics; if that check fails, the universal power-law claim should be downgraded. The practical GPR-denosing and early-stopping workflow may still be useful, and I am not questioning the empirical data, only the theoretical explanation that underlies the universality claim.","tokens_in":7297,"tokens_out":11200,"duration_ms":114024,"concrete_test":"Simulate the EQSANS reduction as described: draw Poisson counts per pixel from a fixed intensity profile, form cumulative profiles I(t_i) at the paper's exposure times, assign zero-count bins error 1/t_i, and compute Eq. 2. Then re-derive the expected scaling using Var[I(t_i)-I(t_{i-1})] = lambda(1/t_{i-1} - 1/t_i) instead of Eq. 4. If the simulated log-log slope departs from the reported -2 (or from the range [-2,-1]) while the paper's independence-based prediction is used, the t^-2 scaling and universal exponent claim are not supported by the stated statistics.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section II defines I(t_i) as the cumulative scattering intensity profile at total exposure time t_i and bases the metric Eq. 1 on I(t_i)-I(t_{i-1}). Yet the derivation after Eq. 3 states: \"consider two successive measurements I(t_i) and I(t_{i-1}), modeled as independent random variables\" and writes Var[I(t_i)-I(t_{i-1})] = Var[I(t_i)] + Var[I(t_{i-1})] (Eq. 4). If I(t) is cumulative, this variance identity is false: the two profiles share all counts up to t_{i-1}, so Cov[I(t_i), I(t_{i-1})] is not zero. With Poisson increments and normalized intensity I(t)=N(t)/t, the correct variance is Var[I(t_i)-I(t_{i-1})] = lambda(1/t_{i-1} - 1/t_i), not lambda(1/t_i^2 + 1/t_{i-1}^2). In the zero-count-floor regime the denominator in Eq. 2 scales as Delta I(t_i) proportional to 1/t_i, so the ratio in Eq. 2 need not scale as t^-2; for equal time steps it would be approximately constant, and for the paper's growing intervals it scales differently. Thus the predicted exponent range (-2,-1), and with it the claimed universal collapse, does not follow from the stated statistical model. This is an internal inconsistency, not merely a question of the zero-count convention. The empirical collapse in Fig. 3 may still be real, but it currently lacks a correct theoretical explanation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a dimensionless metric, Eqs. (1)-(2), for judging when a time-resolved SANS measurement has converged, based on successive cumulative intensity profiles processed with Gaussian Process Regression. The authors apply the metric to EQSANS data from CTAB/NaSal and several other soft-matter systems, observe an approximately linear log-log decrease of the metric with exposure time, and claim that after rescaling time by a system-specific characteristic time t* the data collapse onto a universal power law with exponent between -2 and -1. They further claim that linear fits using only the first ten time steps are stable enough to forecast the required measurement duration.","tokens_in":7614,"tokens_out":10357,"duration_ms":103879,"significance":"The practical goal is valuable: a data-driven stopping rule for neutron scattering experiments, especially at low-flux facilities, would improve beam utilization, and the integration of GPR into the EQSANS workflow is a useful engineering contribution. The paper also tests the metric on several experimental datasets and demonstrates a forecasting protocol in Fig. 4. However, the two headline claims—the derivation of the t^{-2} short-time scaling and the universality of the collapse—are not supported as presented. The derivation mis-treats cumulative profiles as independent, and the collapse is produced by a per-system normalization that pins all curves to a common point by construction. These issues are central, not cosmetic.","major_comments":[{"comment":"The derivation after Eq. (3) models I(t_i) and I(t_{i-1}) as independent random variables and writes Var[I(t_i)-I(t_{i-1})]=Var[I(t_i)]+Var[I(t_{i-1})]. This is inconsistent with the definition of I(t) as the cumulative intensity profile. If N(t) is a Poisson process with mean λt and I(t)=N(t)/t, then Cov[N(t_i),N(t_{i-1})]=Var[N(t_{i-1})]=λt_{i-1}, and the correct variance is Var[I(t_i)-I(t_{i-1})]=λ(1/t_{i-1}-1/t_i), not λ(1/t_i^2+1/t_{i-1}^2). Consequently the proportionality leading to Eq. (6) does not follow, and the predicted t^{-2} short-time scaling is not established. The manuscript must either use the correct covariance or explicitly redefine I(t_i) as independent repeated measurements, which would contradict the cumulative-exposure description in the text and Fig. 1.","section":"II, Eq. (4)"},{"comment":"The universal collapse in Fig. 3(a) is enforced by construction. The text states that t* is defined so that the mean relative variation of the raw data equals 1 at t=t*; rescaling each system by its own t* forces all raw-data curves through the point (1,1). The overlap at the reference point therefore cannot be used as evidence for a universal curve. What remains testable is the slope of each curve and whether the slopes agree. The authors should report per-system slopes with uncertainties and test the collapse after removing the trivial pinning, for example by fixing t* from an independent physical timescale or by comparing residuals of the fitted slopes.","section":"II, paragraph before Fig. 3(a)"},{"comment":"The claimed exponent range (-2,-1) is derived from the model, not discovered from the data. Because the short-time end of the derivation rests on the flawed independence assumption (Major Comment 1) and on the instrument-specific zero-count floor (the assignment of unit uncertainty to empty bins), the statement that 'the extracted scaling exponents cluster around -2' has no valid theoretical grounding. If the authors wish to retain the universal-exponent claim, they need a corrected derivation that accounts for the cumulative structure of the profiles. If such a derivation is not possible, the paper should be reframed as reporting an empirical trend rather than a universal law.","section":"II, Eqs. (6)-(8)"},{"comment":"The zero-count bin convention ΔI ∝ 1/t is load-bearing: it is the sole source of the t^{-2} branch. The authors should test whether the claimed scaling and collapse survive alternative error assignments, for example a floor of 0.5 or 2, or a proper Poisson treatment of zero counts. Without such a sensitivity test, the universality claim depends on a particular data-reduction choice rather than on the physics of the measurement.","section":"II, paragraph after Eq. (3)"}],"minor_comments":[{"comment":"The phrase 'this behavior has a well-defined statistical interpretation' begins with a lowercase letter, and the following sentence is a run-on; also, the claimed scaling of the l-th cumulant as O(n^{1-l}) is not defined. Please define n and l and connect that statement to the rest of the argument.","section":"II, paragraph after Eq. (3)"},{"comment":"The notation with angle brackets subscripted by Q after the squared term, together with the separate factor 1/\\langle[I(t_i)]^2\\rangle_Q, is easy to misread; use explicit brackets, e.g. \\left\\langle [\\cdots]^2\\right\\rangle_Q.","section":"II, Eq. (2)"},{"comment":"The text says 'the solid line representing the overall best-fit trend' but the figure caption refers to 'vertical lines'; make the description consistent and add error bars or confidence intervals to the individual exponents.","section":"Fig. 3(b)"},{"comment":"The statement that forecast uncertainty 'falls below a level of approximately 2' is unexplained; specify what quantity is plotted and how the threshold is chosen.","section":"Fig. 4"},{"comment":"The GPR kernel and its hyperparameters are not described here. Since the metric is computed from GPR-inferred profiles, provide the implementation details or explicitly state that they are fully given in Ref. 9.","section":"II"},{"comment":"Typo: 'Thist−1 scaling' should be 'This t^{-1} scaling'.","section":"II, paragraph after Eq. (8)"}],"recommendation":"reject","confidential_remarks":"For the editor: the manuscript has an internally inconsistent statistical derivation (Eq. 4) at the heart of the universal-scaling claim, and the t* normalization in Fig. 3(a) makes the collapse a construction rather than a standalone finding. I would not consider acceptance until both issues are resolved. The practical forecasting tool might be publishable as a substantially revised, more limited empirical study with a corrected statistical model and a non-circular collapse test."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, this is a practical paper: the dimensionless metric and the early forecasting are useful for SANS beam time, especially at low-flux sources. Second, the central theoretical claim — the universal power law with exponent between -2 and -1 — is not supported by the paper's own derivation. Equation 4 assumes I(t_i) and I(t_{i-1}) are independent, but Section II defines them as cumulative totals. They share all counts up to t_{i-1}; the variance of their difference is λ(1/t_{i-1} - 1/t_i), not the sum of 1/t_i^2 and 1/t_{i-1}^2. For the increasing step sizes used here, that gives t^{-1} scaling, not t^{-2}. The claimed exponent range collapses with the math.\n\nThe empirical collapse in Fig. 3 might be real, but it isn't evidence for the theory. Each system is rescaled by a fitted t*, so all curves are forced through the same reference point; that by itself makes curves look similar. The slopes are fitted too, so the 'universal' exponent is a fit, not a prediction. The zero-count error floor changes the story further.\n\nWhat the paper does well: the metric requires no ground truth, is easy to compute, and the forecast in Fig. 4 stabilizes quickly. That could genuinely save beam time. The GPR component is incremental but sensible, building on their prior work.\n\nThe paper deserves peer review, not a desk reject, but it needs major revision. The authors should correct the variance calculation, test the collapse without per-system fitting (e.g., using a fixed t* from one system to predict others), and release data/code. If the collapse survives, it's a solid methods paper. If not, the early-forecast still has practical value.\n\nI'd bring it to the reading group; the error in Eq. 4 is instructive. I wouldn't cite it in its current form.","headline":"A practically motivated stopping-rule metric for time-resolved SANS, but the universal power-law claim rests on a mistaken variance calculation and a per-system fit that forces the collapse.","tokens_in":8159,"tokens_out":9737,"would_cite":false,"duration_ms":88035,"reading_group":"yes","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A dimensionless metric predicts when time-resolved neutron scattering data are sufficient, with a universal power-law exponent between -2 and -1.","keywords":["time-resolved SANS","Gaussian process regression","measurement sufficiency","dimensionless convergence metric","universal power-law scaling","early-time forecasting","Poisson statistics","central limit theorem"],"falsifier":"Reanalyze one of the paper's time-resolved datasets using an alternative zero-count treatment, for example discarding empty bins or assigning them a fractional uncertainty, and recompute the dimensionless metric. If the early-time slope moves away from $-2$ or the rescaled curves for different systems stop collapsing, the universal range is an artifact of that convention. The same test can be run on synthetic Poisson data where the ground truth is known.","tokens_in":7087,"feed_emoji":"⚛️","tokens_out":10744,"duration_ms":90622,"temperature":0.7,"pith_summary":"This paper claims that the time evolution of a small-angle neutron scattering profile encodes its own stopping rule. By comparing each time frame with the previous frame through a dimensionless quantity that divides the change in intensity by the corresponding uncertainty and average signal, the method defines a metric that needs no model of the sample and no knowledge of the final profile. Across several soft-matter systems the metric collapses onto a single power-law curve once time is rescaled by a system-specific characteristic time $t^{\\star}$, with an exponent between $-2$ and $-1$. The trend stabilizes within the first ten time steps, so the metric can forecast when a measurement will be statistically sufficient and can terminate data collection early. If correct, this would make low-flux neutron and X-ray scattering experiments considerably more efficient.","feed_headline":"Universal power law sets neutron measurement time from first ten steps","feed_subtitle":"Dimensionless curve with exponent between -2 and -1 forecasts when scattering profiles stop changing, saving time.","key_machinery":"The load-bearing object is the dimensionless convergence metric of Eq. (2): the $Q$-average of the squared change in inferred intensity between successive time frames, divided by the uncertainty ratio $\\Delta I(t_i)/\\sqrt{\\langle[\\Delta I(t_i)]^2\\rangle_Q}$ and by the average squared intensity $\\langle[I(t_i)]^2\\rangle_Q$. This quantity requires no ground truth and can be computed online during an experiment. Its scaling behavior is governed by two mechanisms: the zero-count error floor that makes the short-time metric decay as $t^{-2}$, and the Poisson/central-limit behavior of a normal counting regime that gives a $t^{-1}$ decay; the observed exponent therefore lies between $-2$ and $-1$. A Gaussian-process regression kernel supplies the denoised profiles whose faster convergence enables early-time forecasting.","core_discovery":"The central discovery is that the $Q$-averaged mean relative variation between successive scattering profiles decays as a universal power law in measurement time. The paper defines the variation as the squared difference between intensity profiles at consecutive time steps, normalized by the experimental uncertainty and by the average squared intensity, so that the quantity is dimensionless and computable from the data alone. When time is divided by a system-specific characteristic time $t^{\\star}$ (the time at which the raw-data variation equals unity), the curves for different systems collapse onto a single trend with an exponent between $-2$ and $-1$. The exponent is derived from two statistical regimes: at short times a zero-count error floor gives $\\Delta I \\propto 1/t$ and hence a $t^{-2}$ decay, while at long times Poisson/central-limit statistics give $\\Delta I \\propto 1/\\sqrt{t}$ and a $t^{-1}$ decay. The GPR-denoised version of the same metric reaches statistical stability about an order of magnitude earlier than the raw data, making early forecasting practical.","pith_inferences":["The fitted exponent can serve as a live regime diagnostic: a slope near $-2$ means the experiment is still dominated by the zero-count error floor, while a slope drifting toward $-1$ indicates enough counts for normal Poisson averaging.","The same adjacent-frame comparison should apply to any time-binned counting experiment, such as X-ray photon correlation, dynamic light scattering, or electron microscopy frame stacks, provided the zero-count error convention is handled explicitly; this is a testable generalization the paper does not develop.","If the collapse is truly universal, defining $t^{\\star}$ from the long-time asymptotic behavior instead of from the variation-equals-unity point should yield the same master curve; checking this would directly test whether the normalization is arbitrary."],"forward_implications":["An experimenter can decide within the first ten time steps how much longer to collect data, terminating the measurement once the forecast variation falls below a chosen threshold.","Low-flux instruments, including compact accelerator-based neutron sources, could obtain statistically sufficient profiles in a fraction of the usual beam time.","Because the metric is dimensionless and model-free, it transfers directly to any SANS instrument with a two-dimensional detector, and the paper argues it extends to laboratory SAXS and X-ray scattering.","GPR-denoised profiles reach convergence roughly ten times faster than raw profiles, widening the time window in which kinetic processes can be resolved.","The same early-time regression can provide forecast confidence intervals for the required measurement duration, supporting adaptive real-time experimental control."],"supporting_citations":[{"why":"Supplies the Gaussian-process regression method that reconstructs SANS profiles with quantified uncertainty from sparse, noisy frames; all inferred-profile results depend on this.","marker":"[9]"},{"why":"Provides the CTAB/NaSal micellar solution whose time-resolved EQSANS data are the main experimental case study.","marker":"[12]"},{"why":"Provides a related CTAB/NaSal system with different solvent contrast, used as one of the soft-matter datasets in the multi-system collapse.","marker":"[13]"},{"why":"Supplies the Poisson-counting statistics behind the uncertainty model and the zero-count error-floor argument.","marker":"[14]"},{"why":"Supplies the central-limit theorem reasoning that predicts the long-time $t^{-1}$ scaling of the metric.","marker":"[15]"},{"why":"Gives the statistical-physics reference for the central-limit behavior used in the long-time scaling argument.","marker":"[16]"}],"fun_headline_variants":["Universal power law from first ten steps sets neutron measurement time","First ten time steps predict when neutron data stop changing","Exponent -2 to -1: early data forecast neutron experiment duration","Dimensionless metric collapses neutron scattering curves onto one power law"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the data-reduction convention that detector bins registering zero neutrons are assigned an uncertainty of one; if a different zero-count treatment is used, the short-time $t^{-2}$ branch, and with it the claimed universal exponent range between $-2$ and $-1$, would no longer follow.","fun_headline_variants_meta":{"raw":{"variants":["Universal power law from first ten steps sets neutron measurement time","First ten time steps predict when neutron data stop changing","Exponent -2 to -1: early data forecast neutron experiment duration","Dimensionless metric collapses neutron scattering curves onto one power law"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000527,"raw_usage":{"total_tokens":2535,"prompt_tokens":928,"completion_tokens":1607,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":544,"completion_tokens_details":{"reasoning_tokens":1538}},"tokens_in":544,"tokens_out":1607,"duration_ms":13246,"temperature":1.0,"reasoning_tokens":1538,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:47:50.266933+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Reanalyze one of the paper's time-resolved datasets using an alternative zero-count treatment, for example discarding empty bins or assigning them a fractional uncertainty, and recompute the dimensionless metric. If the early-time slope moves away from $-2$ or the rescaled curves for different systems stop collapsing, the universal range is an artifact of that convention. The same test can be run on synthetic Poisson data where the ground truth is known.","supporting_citations":[{"cited_title":"Gelman , author J","cited_arxiv_id":null,"evidence_quote":"Gives the statistical-physics reference for the central-limit behavior used in the long-time scaling argument."},{"cited_title":"F \\\"o rster \\ and\\ author M","cited_arxiv_id":null,"evidence_quote":"Supplies the Gaussian-process regression method that reconstructs SANS profiles with quantified uncertainty from sparse, noisy frames; all inferred-profile results depend on this."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the CTAB/NaSal micellar solution whose time-resolved EQSANS data are the main experimental case study."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides a related CTAB/NaSal system with different solvent contrast, used as one of the soft-matter datasets in the multi-system collapse."},{"cited_title":"\\ Tung , author S","cited_arxiv_id":null,"evidence_quote":"Supplies the Poisson-counting statistics behind the uncertainty model and the zero-count error-floor argument."},{"cited_title":"Jeffreys ,\\ @noop title Theory of Probability ,\\ edition 3rd \\ ed.\\ ( publisher Oxford University Press ,\\ address Oxford ,\\ year 1984 ) NoStop","cited_arxiv_id":null,"evidence_quote":"Supplies the central-limit theorem reasoning that predicts the long-time $t^{-1}$ scaling of the metric."}],"review_version":1}