{"id":"7adf6ee5-a521-4932-878a-3446fb0f1067","arxiv_id":"2509.05339","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Persistent homology of signed-distance filtrations correlates with and can classify the hyperuniform character of scalar fields, demonstrated on Cahn-Hilliard patterns and Gaussian random fields.","lead":"This paper shows that persistent homology, a tool from topology, can capture the large-scale \"hyperuniform\" order of disordered two-phase patterns by summarizing local shape features. It tests the idea on simulations of the Cahn-Hilliard equation and on Gaussian random fields, and uses the resulting summaries to tell hyperuniform from non-hyperuniform patterns.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"97% classification accuracy is measured against the finite-size proxy eH<0.011, not the prescribed H used to generate the GRFs; unless label noise is quantified, the classifier does not establish inference of true hyperuniformity.","rationale":"The reader's weakest assumption—that the eH<0.011 threshold may mislabel the training data—is precisely the load-bearing concern. The paper's strongest evidence for the central claim is the 97% classifier accuracy in Section 3.3, but that accuracy is computed against the same finite-size proxy used to define labels. Because the GRFs are generated with a known prescribed H, the true labels are available; using eH introduces avoidable label noise. The paper does not quantify how often eH crosses the 0.011 threshold for fields with prescribed H near 0.01, nor does it compare against a classifier using eH directly. Without this, the high accuracy could simply reflect that persistence diagrams correlate with eH, not that they infer actual hyperuniformity. The Wasserstein distance results are suggestive but only show sensitivity to spectral parameters, not selectivity for hyperuniformity. This concern is addressable, and the reader already conditioned the verdict on similar issues, so no change to the CONDITIONAL verdict is needed; the proposed test would make the necessary condition explicit.","tokens_in":15273,"tokens_out":3683,"duration_ms":42934,"concrete_test":"Retrain the Section 3.3 classifier with labels defined by the prescribed parameter H≤10^{-2} (the actual spectral density in eq. 3.2) instead of eH<0.011, using the same 5-fold cross-validation and 30 random initializations. Report the contingency table between prescribed-H labels and eH<0.011 labels, and the classifier accuracy against each label set. If accuracy against prescribed H drops substantially below 97%, or if a nontrivial fraction (>5%) of fields are mislabeled by the eH proxy, the central inference claim is not supported. Additionally, compare with a baseline classifier using eH alone to check whether persistence diagrams provide predictive information beyond the finite-size proxy.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that persistent homology can be used to infer hyperuniformity rests on the classifier in Section 3.3. Training labels are assigned using eH<0.011, where eH is the finite-size estimate (2.5). Section 2.1 explicitly states eH overestimates H. The authors do not report the confusion between the prescribed spectral parameter H (eq. 3.2) and the eH-based labels. If fields with true H just above 0.01 are observed with eH<0.011, or true HU fields with H just below 0.01 are observed with eH≥0.011, the reported 97.3% accuracy measures agreement between topological summaries and the proxy, not with actual hyperuniformity. Since the GRFs are generated with known H, the ground truth is available; using the proxy is unnecessary and undermines the inference claim. The Wasserstein distance results in Fig. 6(a,b) show sensitivity to H, α, K, but do not demonstrate that topological measures select hyperuniform fields better than the spectral measure eH itself; no baseline comparison is provided.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a persistent-homology (PH) characterization of hyperuniform scalar fields. It uses a signed-distance filtration (Eq. 2.12) rather than the usual sublevel filtration, computes Wasserstein distances between persistence diagrams (Eq. 2.13), and applies the framework to two systems: numerical solutions of the Cahn-Hilliard equation, where it checks epsilon-convergence and self-similarity, and Gaussian random fields (GRFs) with prescribed spectral parameters (alpha, H, K). The central claim is that hyperuniform characteristics correlate systematically with topological feature distributions, that Wasserstein distances can find the closest GRF match to a Cahn-Hilliard pattern, and that a neural network trained on binned persistence diagrams can classify HU vs non-HU GRFs with 97.3% accuracy (Table 1). The CH validation and the GRF parameter grid provide substantive numerical evidence, but the inference claim is weakened by the choice of training labels and the absence of baseline comparisons.","tokens_in":15537,"tokens_out":5023,"duration_ms":59143,"significance":"If the central claim holds, the paper would extend TDA-based hyperuniformity analysis from point patterns to scalar fields, offering a local geometric descriptor for two-phase microstructures and a potential screening tool for large datasets. The paper is careful in several places: it explicitly notes that the convergence analysis for the topological measure is out of reach, restricts the inverse approach to GRFs, and acknowledges the finite-size overestimation of the hyperuniformity metric. These self-critical statements are a strength. However, the headline classification result does not currently establish that persistence diagrams infer true hyperuniformity, because the labels are derived from a finite-size proxy rather than the known prescribed parameter H. The contribution is promising but needs revision.","major_comments":[{"comment":"The training labels for HU vs non-HU are assigned using the finite-size estimate eH < 0.011, not the prescribed H in Eq. (3.2). Since the GRFs are generated with known H, the true label is available; using eH introduces label noise. Section 2.1 states that eH overestimates H, and the threshold 0.011 is ad hoc. The reported 97.3% accuracy is therefore an agreement with the proxy, not with actual hyperuniformity. Please either use the prescribed H as ground truth, or report the confusion matrix between eH-based labels and prescribed-H labels and quantify the label-noise rate.","section":"§3.3, Eq. (2.5), Eq. (3.2)"},{"comment":"No baseline classifier is reported. A simple classifier using eH itself or low-order spectral features would likely achieve high accuracy on this parameter grid, so the 97% accuracy does not by itself demonstrate that topological descriptors add value. Please compare against a non-topological baseline, and ideally evaluate on held-out parameter combinations rather than random 5-fold cross-validation, since each (alpha, H, K) combination has three realizations and random splits may leak the same parameter combination into both train and test.","section":"§3.3, Table 1"},{"comment":"The central correlation plots show no error bars or significance measures. The distances are computed from a single reference field and only three realizations per parameter combination; it is unclear whether the small variations in Wasserstein distance for H < 10^-2 are meaningful. Adding error bars, confidence intervals, or a statistical test would strengthen the claim that topological features systematically correlate with the global HU character.","section":"§3.2.2, Fig. 6(a,b)"},{"comment":"There is a definitional inconsistency: Section 2.1 says H < 10^-4 is 'effectively HU' and H < 10^-2 is 'nearly HU', while Section 3.3 defines as HU all fields with H <= 10^-2. This broadens the HU class to include nearly HU fields, so the classification result is not directly about hyperuniformity as defined earlier. Please align the terminology or explicitly justify the threshold choice.","section":"§2.1 vs §3.3"}],"minor_comments":[{"comment":"The displayed formula for the equilibrium profile appears to have a missing closing parenthesis in the tanh argument; please fix the typo.","section":"Eq. (2.7)"},{"comment":"The notation for the Wasserstein distance is inconsistent: Eq. (2.13) defines W^k_{p,q}, while the Figure 3(f) caption uses W_{q,p}. Please use a single convention.","section":"Fig. 3(f) and Eq. (2.13)"},{"comment":"The statement that barcodes/diagrams are 'stable under perturbations of the input' is too broad as written; the stability bound in the following sentence is the precise statement. Consider reformulating.","section":"§2.3"},{"comment":"The subset ranges for the persistence-diagram binning (e.g., [-15,8] x [0,15] for P0) appear without justification. A short explanation of how these ranges were chosen would improve reproducibility.","section":"§3.3"},{"comment":"The Data Availability statement says data will be released only upon acceptance. Since the central inference results are numerical and reproducibility would benefit from immediate release, please consider providing the code/data in a public repository at revision time.","section":"Data Availability"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern is valid and lands: the proxy-label issue directly affects the central inference claim. The paper has solid components—the CH validation and the Wasserstein-distance comparisons are informative—but the classification result needs to be re-grounded on prescribed H and compared with baselines. Group-wise cross-validation should also be addressed. I do not see this as a reject; the fix is within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look. This is a genuine extension of the persistent-homology hyperuniformity framework from point clouds to scalar fields, built on a signed-distance filtration, and the Cahn-Hilliard validation is solid. The main claim—that PH statistics track HU character in scalar fields—is supported by the systematic Wasserstein-distance trends and the high classification accuracy. The paper is honest about limits, noting that a convergence analysis for the topological measures is out of reach and that data/code will only be released upon acceptance.\n\nThe stress-test note about the eH<0.011 label is fair but not fatal. Since eH is the practical finite-size proxy and the prescribed H is a generation parameter, predicting eH is legitimate; however, the authors should report the confusion between prescribed H and eH labels, and ideally test the classifier against prescribed H. They have the ground truth; ignoring it weakens the inference claim. More importantly, there is no baseline comparison: the paper claims the topological approach is more accurate than existing finite-size measures, but never compares against eH or the spectral density directly.\n\nSeveral figures lack error bars. With only three realizations per parameter combination, that matters; Fig. 4 has error bars but Fig. 6 does not. The 'reconstruction' in Fig. 6(c) is really a grid search over the parameter space, not an inversion, and the single-isoline choice (c=0 for GRFs, c=0.5 for CH) is a free parameter that the authors admit was chosen because it worked better in tests. Those are addressable concerns, not fatal ones.\n\nThe central argument holds up. The paper will be useful to the TDA and hyperuniformity communities, especially as a bridge between point-cloud and scalar-field analysis. It deserves a serious referee. For revision: release code and data, add error bars, compare against spectral baselines, and clarify the label issue.\n\nSend it out.","headline":"Solid numerical extension of TDA hyperuniformity to scalar fields with a defensible central claim, but the label proxy and missing baselines need addressing before I'd trust the inversion claims.","tokens_in":16025,"tokens_out":3167,"would_cite":true,"duration_ms":37253,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35B36","55N31"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper shows that local topological summaries of a field—persistence diagrams of a signed-distance filtration—carry enough information to determine whether the field is hyperuniform, and that the link can be inverted to classify fields a","keywords":["Hyperuniformity","Persistent homology","Cahn-Hilliard equation","Gaussian random fields","Topological data analysis","Wasserstein distance","Spinodal decomposition","Scalar fields"],"falsifier":"Generate Gaussian random fields with exactly known asymptotic exponent alpha on a sequence of increasingly large domains, compute persistence diagrams from the signed-distance filtration, and check whether Wasserstein distances and the trained classifier reproduce the true alpha labels rather than the finite-size eH labels. If the ranking shifts with box size, or changes when the isoline level c is varied, the reported correlation is an artifact of the proxy or the filtration choice.","tokens_in":15135,"feed_emoji":"🌀","tokens_out":7566,"duration_ms":79437,"temperature":0.7,"pith_summary":"The paper tries to establish that persistent homology—specifically, persistence diagrams built from a signed-distance filtration around a chosen isoline—captures enough local geometric information about a disordered two-phase scalar field to characterize its global hyperuniformity. The authors test this on numerical solutions of the Cahn-Hilliard equation, showing that Wasserstein distances between persistence diagrams converge as the diffuse interface width shrinks and reproduce the expected coarsening behavior and self-similarity. They then generate Gaussian random fields with prescribed spectral parameters (alpha, H, K) and show that Wasserstein distances between persistence diagrams vary systematically with these parameters. A neural network trained on binned persistence diagrams then separates hyperuniform from non-hyperuniform fields with 97.3% accuracy, suggesting the global property can be inferred from local topological statistics.","feed_headline":"Topology exposes hyperuniformity hidden in disordered fields","feed_subtitle":"Comparing persistence diagrams recovers the spectral parameters behind hyperuniformity and classifies random fields at 97% accuracy.","key_machinery":"The signed-distance filtration (2.12), X^iso_{r,c} = {x in Ω : D(φ(x),c) ≤ r}, where D is the signed Euclidean distance to the c-isoline of the field. Sweeping r from negative to positive builds sublevel sets that trace valleys, the interface, and hills, and persistent homology of this filtration yields persistence diagrams P0 and P1 recording births and deaths of connected components and loops. The workhorse is the total Wasserstein distance W_{p,q} = W0_{p,q} + W1_{p,q} between diagrams: it is stable with respect to perturbations of the field and serves both as a similarity measure between patterns and, after binning, as the feature vector for the classifier.","core_discovery":"The central claim is that in disordered correlated scalar fields, the distribution of local topological features—encoded as points in persistence diagrams from the signed-distance filtration (2.12)—is systematically tied to the global spectral behavior that defines hyperuniformity. The paper demonstrates this by showing that the Wasserstein distance between persistence diagrams changes predictably with the spectral parameters alpha, H, and K in Gaussian random fields, and that the persistence diagram of a Cahn-Hilliard solution is closest to a Gaussian random field with alpha about 4 and K about 0.64, matching the known k^4 spectrum. The classification experiment turns the correlation into a","pith_inferences":["If the correlation holds beyond the Gaussian-random-field family, persistence-diagram distances could be inverted for design: specifying a target persistence diagram would constrain the spectral parameters of a generated structure, a route the paper only gestures at.","The single-isoline choice is a free parameter; testing several levels c in the signed-distance filtration would show whether the Wasserstein ranking of hyperuniform classes is stable or depends on the chosen contour.","Because eH overestimates H, the 97.3% accuracy is a statement about the finite-size proxy; a stricter check would relabel the data using a rigorous finite-size hyperuniformity test and re-measure classification accuracy.","A theoretical bound connecting persistence-diagram Wasserstein distances to sharp-interface convergence would turn the fitted sublinear epsilon-convergence exponent into a provable rate, linking topological data analysis stability to the epsilon→0 limit."],"forward_implications":["Persistence diagrams give a finite-size, local measure of hyperuniformity that does not require resolving the k→0 limit directly.","Wasserstein distances can rank finite patterns by hyperuniform character; the Cahn-Hilliard example shows they identify the closest spectral parameters (alpha ≈ 4) without fitting the spectrum.","Deviations from self-similarity in coarsening patterns can be quantified for finite interface width, not just in the sharp-interface limit.","The same signed-distance filtration transfers to other interface and free-boundary problems, since it only needs an isoline and a distance field.","Binned persistence diagrams can serve as features for surrogate models, enabling screening of large libraries of correlated scalar fields."],"supporting_citations":[{"why":"Supplies the definition of hyperuniformity, its classes (I, II, III), and the metric H used throughout.","marker":"[45]"},{"why":"Establishes that Cahn-Hilliard solutions are hyperuniform random scalar fields and provides the finite-size estimate eH.","marker":"[25]"},{"why":"Introduces the Cahn-Hilliard equation, the model system whose solutions are the paper's main testbed.","marker":"[5]"},{"why":"Explains the k^4 low-k scaling and its link to isotropy and t^{1/3} growth, the validation target for the topological measures.","marker":"[8]"},{"why":"Provides the Wasserstein stability bound for persistence diagrams, justifying the distance metric's use.","marker":"[42]"},{"why":"Extends persistent-homology statistics to hyperuniform point clouds; this paper aims to make the analogous extension to scalar fields.","marker":"[38]"},{"why":"Introduces inference of hyperuniformity traits from persistence diagrams with machine learning, the template for the binned-diagram classifier.","marker":"[29]"},{"why":"Supplies the semi-implicit Fourier-spectral scheme used to integrate the Cahn-Hilliard equation.","marker":"[6]"},{"why":"Documents convergence of the Cahn-Hilliard equation to the Hele-Shaw problem, the analytical reference for the epsilon-convergence validation.","marker":"[2]"},{"why":"Proposes a rigorous significance test for hyperuniformity, relevant to separating true hyperuniform behavior from finite-size proxies.","marker":"[19]"}],"fun_headline_variants":["Persistence diagrams decode hyperuniformity in random fields","Topological fingerprints classify hyperuniform fields at 97% accuracy","Wasserstein distances reveal hyperuniformity hidden in Cahn-Hilliard patterns","Local topology reads global spectral order in scalar fields","Persistence homology pinpoints hyperuniformity in disordered systems"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The load-bearing premise is that the finite-size estimate eH—computed at the smallest accessible wavevector and thresholded at eH < 0.011—truly separates hyperuniform from non-hyperuniform fields; all reported correlations and the 97.3% classification are measured against this proxy, not against the actual k→0 limit.","fun_headline_variants_meta":{"raw":{"variants":["Persistence diagrams decode hyperuniformity in random fields","Topological fingerprints classify hyperuniform fields at 97% accuracy","Wasserstein distances reveal hyperuniformity hidden in Cahn-Hilliard patterns","Local topology reads global spectral order in scalar fields","Persistence homology pinpoints hyperuniformity in disordered systems"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001193,"raw_usage":{"total_tokens":4740,"prompt_tokens":711,"completion_tokens":4029,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":455,"completion_tokens_details":{"reasoning_tokens":3945}},"tokens_in":455,"tokens_out":4029,"duration_ms":33111,"temperature":1.0,"reasoning_tokens":3945,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T12:27:30.351071+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Generate Gaussian random fields with exactly known asymptotic exponent alpha on a sequence of increasingly large domains, compute persistence diagrams from the signed-distance filtration, and check whether Wasserstein distances and the trained classifier reproduce the true alpha labels rather than the finite-size eH labels. If the ranking shifts with box size, or changes when the isoline level c is varied, the reported correlation is an artifact of the proxy or the filtration choice.","supporting_citations":[{"cited_title":"Torquato, Hyperuniform states of matter.Physics Reports745(2018), 1–95","cited_arxiv_id":null,"evidence_quote":"Supplies the definition of hyperuniformity, its classes (I, II, III), and the metric H used throughout."},{"cited_title":"Ma and S","cited_arxiv_id":null,"evidence_quote":"Establishes that Cahn-Hilliard solutions are hyperuniform random scalar fields and provides the finite-size estimate eH."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the Cahn-Hilliard equation, the model system whose solutions are the paper's main testbed."},{"cited_title":"De Luca, X","cited_arxiv_id":null,"evidence_quote":"Explains the k^4 low-k scaling and its link to isotropy and t^{1/3} growth, the validation target for the topological measures."},{"cited_title":"Salvalaglio, D","cited_arxiv_id":null,"evidence_quote":"Extends persistent-homology statistics to hyperuniform point clouds; this paper aims to make the analogous extension to scalar fields."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces inference of hyperuniformity traits from persistence diagrams with machine learning, the template for the binned-diagram classifier."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the semi-implicit Fourier-spectral scheme used to integrate the Cahn-Hilliard equation."},{"cited_title":"Alikakos, P","cited_arxiv_id":null,"evidence_quote":"Documents convergence of the Cahn-Hilliard equation to the Hele-Shaw problem, the analytical reference for the epsilon-convergence validation."}],"review_version":1}