{"id":"3acfeee5-acdf-480c-96e8-127be8639e7b","arxiv_id":"2607.09228","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"M-bounded persistence-diagram functionals of germ-grain models with exponential correlation decay are asymptotically normal, enabling chi-squared GoF tests for spatial interactions.","lead":"The paper proves a central limit theorem for functionals of M-bounded persistence diagrams of germ-grain random sets under exponential decay of correlations. This justifies normality-based goodness-of-fit tests that detect clustering or repulsion and is illustrated on breast histology images.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"Variance lower bound remains the load-bearing gap for Theorem 4.4; score construction for q=1 is only sketched.","rationale":"The Reader correctly isolates the unverified variance lower bound as the weakest assumption of Theorem 4.4. My additional observation is that the score construction for q=1 is only sketched, so the reduction itself is not fully rigorous for loops. Both issues are load-bearing for the strongest claim; neither is fatal once an explicit assignment rule is written down and the variance growth is checked (analytically or by a larger simulation). The rest of the paper—stabilisation for q=0, simulation power, histology application—remains sound. Hence the verdict stays CONDITIONAL, exactly as the Reader concluded.","tokens_in":18216,"tokens_out":596,"duration_ms":7168,"concrete_test":"Construct an explicit, deterministic assignment of each M-bounded 1-feature to a unique germ (e.g., the germ of the younger component at the moment the loop is born, or the germ nearest the birth critical point). Verify that the resulting ξ_1 still has bounded radius of stabilisation and satisfies the p-moment condition under (igstar). Then recompute the empirical variance of ⟨f,PD_{M,1}⟩ on windows of area n=25^{2},50^{2},100^{2} for the Boolean model; if the log-log slope is statistically consistent with 0 the Ω(n^\nu) hypothesis fails and the CLT does not apply.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Theorem 4.4 reduces the CLT for ⟨f,PD_{M,q}(P̃_n)⟩ to the stabilisation CLT of [26] by writing the functional as a sum of scores ξ_q. For q=0 the assignment of births to germs is unambiguous and the radius of stabilisation is clearly ≤M. For q=1 the paper only asserts that “one point can give birth to at most P̃(W_n)-1 holes” and that |ξ_1|≤P̃_n(B_{4r_f}(z))∥f∥_∞; it never exhibits an explicit, measurable rule that selects a unique responsible germ for each M-bounded hole. Without that rule the score is not rigorously defined, so the reduction to Theorem 4.2 is incomplete for dimension 1. Even if the score can be made rigorous, the variance lower bound Var=Ω(n^\nu) is still only assumed (and “suggested by simulation”), exactly as the Reader noted. Both gaps sit inside the strongest claim.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper proves a central limit theorem (Theorem 4.4) for integrals of bounded measurable functions against M-bounded persistence diagrams of germ-grain random sets in R^{2}. The argument reduces the PD functional to a sum of scores with bounded stabilisation radius by assigning each M-bounded feature to a unique germ that “gives birth” to it, then invokes the stabilisation CLT of Yogeshwaran–Błaszczyszyn–Yukich under exponential decay of correlations, a moment condition (⋆), and a variance lower bound Var=Ω(n^ν). The CLT is used to construct χ^{2} goodness-of-fit tests based on rectangular partitions of the PD and on two functional summaries (APF and the support function of the lift zonoid). Simulation studies on Boolean, Quermass, Matérn-cluster, cell and DPP models, and an application to histological images of breast tissue, illustrate the tests’ ability to detect clustering or repulsion.","tokens_in":18538,"tokens_out":903,"duration_ms":9895,"significance":"If the reduction is made fully rigorous and the variance condition is verified for the models of interest, the result supplies the first asymptotic normality theorem for persistence-diagram functionals of germ-grain sets. That would give a solid theoretical foundation for the TDA-based goodness-of-fit procedures already used in spatial statistics and medical imaging, and would place the earlier exploratory work of Gotovac Ðogaš & Mandarić on a firmer footing. The explicit construction of stabilising scores for topological features is a useful technical contribution even if some details remain incomplete.","major_comments":[{"comment":"Theorem 4.4 (and the surrounding argument in §4.2) assumes rather than proves the variance lower bound Var(⟨f,PD_{M,q}(˜P_n)⟩)=Ω(n^ν) for some ν>0. The text only remarks that “our simulation study suggests” the bound holds. Without a proof or a verifiable criterion for the models listed in §5 (Boolean, Quermass, Matérn cluster, etc.), the CLT remains conditional and the subsequent χ^{2} tests lack asymptotic justification.","section":null},{"comment":"For q=1 the score ξ_1 is only sketched (§4.2, after the display of ξ_1). The paper asserts that “one point can give birth to at most ˜P(W_n)-1 holes” and that |ξ_1|≤˜P_n(B_{4r_f}(z))∥f∥_∞, but never supplies an explicit, measurable rule that selects a unique responsible germ for each M-bounded hole. Without such a rule the score is not rigorously defined, so the reduction to the stabilisation theorem (Theorem 4.2) is incomplete for dimension 1.","section":null}],"minor_comments":[{"comment":"The free parameters M, r_f, R_max and the rectangular cut-points used in §5.1 are chosen after visual inspection of the diagrams; a short sensitivity analysis or a data-driven selection rule would strengthen the simulation claims.","section":null},{"comment":"Figures 7–8 crop the boundary rectangles “for clearer presentation,” yet the numerical values inside those rectangles are part of the test statistic; the full partition should be shown or tabulated.","section":null},{"comment":"Typographical inconsistencies appear throughout (e.g., “Mbounded”, “4=30”, “reffered”, “dimen-sion”). A careful proof-reading pass is needed.","section":null},{"comment":"The real-data analysis (§6) reports that the functional summaries fail to be approximately normal; this limitation should be stated more prominently in the abstract and discussion.","section":null}],"recommendation":"major_revision","confidential_remarks":"The two gaps identified above are genuine but appear fixable within the present framework (an explicit birth-assignment rule for holes and either a proof or a literature citation for the variance lower bound). I therefore recommend major revision rather than rejection. The paper is a natural sequel to the authors’ earlier TDA work and fits the journal’s scope."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The new result is Theorem 4.4: under exponential decay of correlations, the (⋆) moment condition, and the variance lower bound Var=Ω(n^ν), the integral of any bounded measurable f against the M-bounded persistence diagram of a germ-grain model is asymptotically normal as the window grows. That is a genuine extension of the Yogeshwaran–Błaszczyszyn–Yukich stabilisation toolkit to TDA functionals of random sets; the earlier exploratory paper by the same authors did not have this theorem.\n\nWhat they do well is the reduction for dimension 0. They assign each birth to the germ whose inball contains the local minimum of the signed distance function (largest-radius tie-break), get a score with radius of stabilisation ≤ M, and the p-moment bound falls out immediately from bounded f and (⋆). The simulation study is thorough: Boolean, Quermass cluster/repulsive, Matérn cluster, cell process, DPP, ellipse grains; both rectangular partitions and the APF / lift-zonoid support functions show useful power against clustering and repulsion. The histology application is honest about where normality fails.\n\nThe soft spots are real but limited. For q=1 the score is only sketched: they bound the number of holes a germ can “give birth to” and the moment, but never write an explicit measurable selection rule that picks a unique responsible germ for each M-bounded hole. That can almost certainly be fixed (e.g., by a deterministic ordering of the two components that merge), yet as written the reduction to the external CLT is incomplete for loops. The variance lower bound is assumed rather than proved and only “suggested by simulation”; if it fails the CLT fails. Partition cut-points and evaluation points for the functionals are chosen after looking at the diagrams. Code is not released. None of these is load-bearing enough to kill the paper; they are the natural revision targets.\n\nThis is for people who already work with germ-grain models or TDA-based spatial statistics and want asymptotic justification for their tests. It is not a broad probability advance, but it is clean enough and useful enough that a serious editor should send it to referees. I would cite the CLT statement once the q=1 score is made rigorous and the variance condition is either proved for the main examples or clearly flagged as an open hypothesis.","headline":"Solid, usable CLT for M-bounded PD functionals of germ-grain models via stabilisation; variance lower bound and q=1 score construction remain the only real soft spots, neither fatal for peer review.","tokens_in":19095,"tokens_out":580,"would_cite":true,"duration_ms":6929,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60D05","62R40","55N31"],"pacs":[],"model":"grok-4.5","headline":"Functionals of M-bounded persistence diagrams from germ-grain random sets become asymptotically normal under exponential decay of correlations, justifying topological goodness-of-fit tests for clustering and repulsion.","keywords":["germ-grain model","persistence diagram","central limit theorem","exponential decay of correlations","goodness-of-fit","topological data analysis","stabilisation","M-bounded features"],"falsifier":"Compute the sample variance of a fixed functional (for example the total lifetime of 0-dimensional features) on successive windows of area n=100,400,1600,… for a Boolean model with fixed intensity and radius distribution; if the variance remains bounded or grows slower than any positive power of n, the claimed CLT cannot hold.","tokens_in":19138,"feed_emoji":"📐","tokens_out":668,"duration_ms":6018,"temperature":0.7,"pith_summary":"The paper proves that many useful summaries of the topology of a germ-grain random set—counts of long-lived components or holes, weighted lifetimes inside rectangles of the persistence diagram, the accumulated persistence function, and support functions of the lift zonoid—obey a central limit theorem once the observation window grows. The argument works by rewriting each summary as a sum of local scores attached to the underlying marked points, then invoking stabilisation theory for processes whose correlations decay exponentially. Because the limiting distribution is known to be Gaussian, the authors can build practical chi-squared goodness-of-fit tests that detect clustering or repulsion and that separate several classical models (Boolean, Matérn cluster, Quermass-interaction, etc.). The same tests applied to histological images of breast tissue distinguish malignant from benign samples at high rates. The result therefore supplies the asymptotic justification that topological data analysis has long needed for inference on random sets.","feed_headline":"Topology of random sets turns Gaussian for large windows","feed_subtitle":"CLT for persistence-diagram functionals justifies fast tests that detect clustering and separate breast-tissue images","key_machinery":"Score functions that assign each marked point the contribution of the single topological feature (component or hole) for which it is responsible; because only M-bounded features are retained, these scores have uniformly bounded radius of stabilisation and therefore fall under existing CLTs for stabilising functionals of marked point processes.","core_discovery":"Under exponential decay of correlations, a mild moment condition, and a variance lower bound of order n^ν (ν>0), the integral of any bounded measurable function against the M-bounded persistence diagram of a germ-grain model, after centering and scaling by its standard deviation, converges in distribution to a standard normal as the window area n tends to infinity.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["CLT for persistence diagrams of germ-grain random sets","Persistence diagram functionals go normal in large windows","Germ-grain topology summaries obey asymptotic normality","Bounded PD functionals of random sets converge to Gaussians","Large windows make germ-grain persistence diagrams Gaussian"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"The variance of the topological functional is assumed to grow at least like a positive power of the window size; the paper never proves this growth and only checks it by simulation for the models it studies.","fun_headline_variants_meta":{"raw":{"variants":["CLT for persistence diagrams of germ-grain random sets","Persistence diagram functionals go normal in large windows","Germ-grain topology summaries obey asymptotic normality","Bounded PD functionals of random sets converge to Gaussians","Large windows make germ-grain persistence diagrams Gaussian"]},"model":"grok-4.5","effort":"low","cost_usd":0.003648,"raw_usage":{"total_tokens":1118,"prompt_tokens":669,"num_sources_used":0,"completion_tokens":74,"cost_in_usd_ticks":36480000,"prompt_tokens_details":{"text_tokens":669,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":375,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":669,"tokens_out":74,"duration_ms":4182,"temperature":1.0,"reasoning_tokens":375,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-13T04:32:44.513615+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Compute the sample variance of a fixed functional (for example the total lifetime of 0-dimensional features) on successive windows of area n=100,400,1600,… for a Boolean model with fixed intensity and radius distribution; if the variance remains bounded or grows slower than any positive power of n, the claimed CLT cannot hold.","supporting_citations":[],"review_version":1}