{"id":"67896eb3-8b4f-4280-869b-db3ddac79863","arxiv_id":"2506.03341","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":2,"one_line_summary":"Generalized Haar coefficients with power-law bounds characterize pixelated Lipschitz functions defined via dyadic pseudo-metrics on non-atomic probability spaces.","lead":"This paper proves that a power law in the size of generalized Haar wavelet coefficients detects a pixelated form of Lipschitz regularity on any non-atomic probability space, even when the dyadic grid has no metric. It opens a wavelet-based route to estimating texture regularity in images and time series without requiring an orthonormal basis.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The converse direction rests entirely on assumption 2.1.e; without it, cube measures along a branch need not vanish and the geometric summation in Theorem 6.6 collapses.","rationale":"The reader's conditional acceptance is reasonable. I read Theorems 6.1, 6.3 and 6.6 as the core and checked the algebra in Lemmas 6.4–6.5 and the telescoping sum in Theorem 6.6; the steps are consistent and the constants check out. The minor gaps the reader flags (Proposition 3.1's attainment of infima, Theorem 6.3 at j=J-1, reproducibility of Section 7) are real but fixable and do not affect the coefficient-to-pixelated-Lipschitz equivalence. My stress-test focuses on the hypothesis 2.1.e, the same structural assumption the reader identified as weakest. I find it is indeed load-bearing: without it, the nested measures along a chain need not tend to zero, so δD need not be a pseudo-metric and the geometric series in Theorem 6.6 has no uniform bound. However, because 2.1.e is explicitly part of Definition 2.1 and Theorem 6.6, this is a limitation of scope rather than an internal flaw. The paper would be improved by stating in the abstract and introduction that the results apply to dyadic families with a uniform inheritance coefficient B, and by noting that dropping 2.1.e can make δD(x,x)>0. I therefore do not change the conditional verdict; the central claim, as stated, is sound. Agreement with reader is partial: we both locate 2.1.e as the weakest point, but the reader treats it as a structural premise while I stress that it is also required for δD to be a pseudo-metric and that explicit families violating it break the framework.","tokens_in":15860,"tokens_out":23302,"duration_ms":279163,"concrete_test":"Build the binary dyadic family on [0,1) defined by Q_{j+1,0}=[0, μ(Q_j)(1-2^{-j})) and Q_{j+1,1}=[μ(Q_j)(1-2^{-j}), μ(Q_j)). Check that 2.1.a–2.1.d hold and 2.1.e fails for every B. (i) Show δD(x,x)=lim μ(Q_j)≈0.288>0 for x in the 'bulk' branch, so Proposition 3.1.1 fails. (ii) Choose Haar coefficients at the maximal magnitude allowed by the power law and compute the pixelated Lipschitz constant from Lemma 6.4's formula over levels 0..J; verify it grows without bound as J increases, whereas Theorem 6.6 would predict a constant independent of J. This settles that 2.1.e is essential and cannot be dropped from the central characterization.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 6.6's converse is only as strong as 2.1.e. Lemma 2.2.3, which gives μ(Q_{j+l,m}) ≤ (1-1/B)^l μ(Q_{j,k}), is proved from 2.1.e and is the sole source of the uniform geometric decay used to sum the telescoping ancestor series with constant B^α/(B^α-(B-1)^α). The same condition is needed in Proposition 3.1 to show μ(Q_{j,k(j,x)})→0, i.e. δD(x,x)=0. If a dyadic family satisfies only 2.1.a–2.1.d, nothing prevents a binary chain with μ(Q_{j+1}) ≥ (1-ε_j) μ(Q_j) and ∑ε_j<∞; then μ(Q_j) tends to a positive limit. Along such a branch δD(x,x)>0, so δD is not a pseudo-metric, and the coefficient power law |⟨f,ψ⟩| ≤ C μ^{α+1/2} does not force a finite uniform Lipschitz constant for the pixelated means, because Lemma 6.4's bound contains the factor sqrt((μ(Q)-μ(\\tilde Q))/(μ(Q)μ(\\tilde Q))) which diverges as μ(\\tilde Q)/μ(Q)→0. Thus 2.1.e is not a harmless normalization but the load-bearing uniformity that makes the measure-theoretic setting work. The paper states it as an axiom but does not flag in the abstract or introduction that the claimed extension to 'non-atomic probability spaces' is restricted to families admitting a fixed B-uniform bound on parent/child measures; in the absence of such a bound the characterization is false in general.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper defines dyadic families with an inheritance coefficient B on non-atomic probability spaces, constructs the induced pseudo-ultra-metric δ_D from dyadic cube measures, and introduces α-Lipschitz and pixelated α-Lipschitz function classes. The main results (Theorems 6.1, 6.3, 6.6, Corollary 6.7) establish an equivalence between a power-law bound on generalized Haar coefficients, |⟨f, ψ^λ_{j,k}⟩| ≤ C μ_{j,k}^{α+1/2}, and dyadic Lipschitz or pixelated Lipschitz regularity, with an explicit constant 2CB^α√(B(B−1))/(B^α−(B−1)^α) in the converse direction. Section 7 gives numerical illustrations on synthetic images using square, parabolic, and non-metric band dyadic families.","tokens_in":16182,"tokens_out":11193,"duration_ms":115510,"significance":"If the small proof gaps noted below are patched, this is a clean and useful contribution: it extends Haar-type wavelet characterization of Lipschitz regularity to purely measure-theoretic settings, without requiring the dyadic family to be a differentiation basis or the space to carry a metric. The forward and converse directions are proved in a self-contained way, and the converse provides an explicit, parameter-free constant in terms of B and α. The paper does not rely on any circular reasoning or fitted parameters in the proofs; the numerical regressions are clearly illustrative outputs rather than inputs. The main structural caveat is that the converse direction depends critically on the inheritance condition 2.1.e, which is explicitly assumed in the theorems.","major_comments":[{"comment":"The proof of the ultra-metric inequality 3.1.3 assumes that there exist dyadic cubes Q1 and Q2 attaining the infima δ_D(x,y) and δ_D(y,z). This assumption fails when either distance is 0, since no cube of measure 0 exists by 2.1.c. The paper itself notes after Proposition 3.1 that δ_D(x,y)=0 for distinct points can occur (vertical bands example), so this is not a vacuous case. The statement of 3.1.3 is nevertheless true: for positive distances one can take approximating cubes with measure less than the distance plus ε and use the nesting property of dyadic cubes, while the case of a zero distance follows from the fact that two cubes containing a common point are nested. The proof should be repaired accordingly.","section":"Section 3, Proposition 3.1.3"},{"comment":"The proof of the discrete version of Theorem 6.3 handles only the range 0 ≤ j < J−1, via Lemma 6.2 (the proof begins 'For large J and 0 ≤ j < J−1'). The case j = J−1 is not treated, yet the theorem statement claims the inequality for all 0 ≤ j ≤ J−1. For j = J−1, ψ_{J−1,k} is a linear combination of indicators of level-J cubes, and the bound can be obtained by writing ⟨f, ψ⟩ = Σ β_m μ(Q_{J,m}) (f_{Q_{J,m}} − f_{Q_{J−1,k}}), applying the pixelated Lipschitz hypothesis, and using Cauchy–Schwarz. This is a local patch, but as written the proof does not cover the full claimed range.","section":"Section 6, Theorem 6.3"}],"minor_comments":[{"comment":"The abstract and introduction could be read as claiming a characterization for arbitrary non-atomic probability spaces. Since the converse direction (Theorem 6.6) depends essentially on the inheritance condition 2.1.e, please state explicitly in the abstract or introduction that the characterization is for dyadic families satisfying properties 2.1.a–2.1.e with a fixed inheritance coefficient B.","section":"Abstract and Introduction"},{"comment":"In the display after 'taking absolute values and using Lemma 6.5', the two sums are separated by a minus sign; they should be separated by a plus sign, as the next line correctly shows.","section":"Theorem 6.6 proof"},{"comment":"The notation '#{K_j} = K_j' is confusing; the cardinality of the index set should be written, for instance, as #K_j or |K_j|.","section":"Definition 2.1.b"},{"comment":"The assertion '#{O(Q)} ≥ 2' follows from 2.1.b and 2.1.d, not from 2.1.d alone; please cite both properties.","section":"Lemma 2.2.2"},{"comment":"Table 2 is difficult to parse because the α and log C values are interleaved in the same cells. Consider using separate rows or separate columns for the two quantities.","section":"Section 7, Table 2"},{"comment":"There is a typo: 'without loosing generality' should be 'without losing generality'.","section":"Section 2, first paragraph"}],"recommendation":"minor_revision","confidential_remarks":"The manuscript is a solid contribution within the scope of math.CA. The self-citations [1] and [3] are appropriate prior art and are not used as inputs to the main arguments. The numerical section is clearly illustrative and does not affect the mathematical claims. The two major comments are local proof gaps that can be repaired without changing the results; I therefore recommend minor revision rather than major revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nThis paper actually delivers a new result: a characterization of dyadic-Lipschitz regularity by power-law decay of Haar-type coefficients on non-atomic probability spaces, without a metric or differentiation basis. The main theorems hold as stated under the 'inheritance coefficient' 2.1.e. The forward directions (Theorems 6.1 and 6.3) are clean; the converse (Theorem 6.6) is the real work, and the finite-dimensional orthogonalization in Lemmas 6.4–6.5 is a genuinely nice trick. The pixelated Lipschitz class is the right notion for non-differentiating families, and the numerical example with non-metric bands is a plausible texture-analysis tool.\n\nSoft spots, in proportion.\n\nProposition 3.1's proof of the ultra-metric inequality assumes cubes that attain the infimum in δD. Those need not exist when δD(x,y)=0—the paper's own band example shows this. The statement is true and an approximation argument fixes it, but the proof has a gap.\n\nTheorem 6.3 states the discrete bound up to j=J-1, but Lemma 6.2, which the proof cites, covers only j<J-1. The endpoint j=J-1 follows from the pixelated condition directly, so it's a presentation slip, not a mathematical hole.\n\nMost importantly, the converse depends entirely on 2.1.e, which forces each child cube to have at least μ(parent)/B mass. Without that, the geometric decay in Lemma 2.2.3 fails: you can construct chains of dyadic cubes with measures converging to a positive limit, δD stops being a pseudo-metric that separates points, and the coefficient power law does not force a finite Lipschitz constant. The paper is not hiding this—2.1.e is in the definition—but the abstract's 'non-atomic probability spaces' is broader than the actual hypothesis. A skimming reader will miss the restriction. That deserves a fix in revision.\n\nThe numerical section is illustrative; no code, no data, under-specified regression. Fine for a math paper, but not reproducible as a standalone claim.\n\nOverall, the mathematics is sound, the literature use is honest, and the self-citations are contextual, not load-bearing. This is a modest but real contribution.\n\nRecommendation: have a competent referee look at it; the main result is correct and the issues are local and patchable. I'd expect minor revisions, not a rejection.\n\nBest.","headline":"A genuine measure-theoretic generalization of Haar-wavelet Lipschitz detection, but the converse hinges on an inheritance coefficient the abstract doesn't mention.","tokens_in":16766,"tokens_out":6679,"would_cite":true,"duration_ms":72612,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["42C15","28A99"],"pacs":[],"model":"deepseek-v4-flash","headline":"On a non-atomic probability space with a dyadic family but no metric, a power-law bound on generalized Haar coefficients is equivalent to α-Lipschitz regularity with respect to the pseudo-metric the family induces.","keywords":["Lipschitz regularity","Haarlet analysis","wavelets","dyadic families","generalized Haar systems","non-atomic probability spaces","pseudo-metric","inheritance coefficient"],"falsifier":"Compute the pixelated Lipschitz seminorm of a single normalized Haar wavelet $\\psi^\\lambda_{j,k}$ with its own coefficient bound $C=(\\mu_{j,k})^{-(\\alpha+1/2)}$ and compare it with $2CB^\\alpha\\sqrt{B(B-1)}/(B^\\alpha-(B-1)^\\alpha)$; if the seminorm exceeds the predicted bound for some dyadic family satisfying the inheritance condition, Theorem 6.6 is wrong.","tokens_in":15576,"feed_emoji":"🌊","tokens_out":19952,"duration_ms":193801,"temperature":0.7,"pith_summary":"The paper proves a measure-theoretic counterpart of the classical wavelet characterization of Lipschitz regularity. On a non-atomic probability space equipped with a dyadic family (a nested sequence of finite measurable partitions, each set split into at least two children), a function is of dyadic $\\alpha$-Lipschitz type precisely when its generalized Haar coefficients obey the power law $|\\langle f,\\psi^\\lambda_{j,k}\\rangle| \\le C(\\mu_{j,k})^{\\alpha+1/2}$. The forward direction holds for the usual $\\alpha$-Lipschitz class and for the weaker pixelated class defined by means over dyadic cubes; the converse direction, which needs the inheritance condition $\\mu(Q)\\le B\\mu(\\tilde Q)$ for every child, recovers the pixelated Lipschitz bound by summing a geometric series along ancestor chains. Because the Haar-type system need not be an orthonormal basis or a differentiation basis, the result gives a metric-free route from coefficient decay to smoothness, with consequences for texture estimation in images and time series.","feed_headline":"Power-law Haar coefficients reveal dyadic Lipschitz regularity","feed_subtitle":"A single bound on wavelet coefficients recovers the Lipschitz exponent, with explicit constants, even without a metric.","key_machinery":"The central object is the generalized Haar system $H_{\\mathcal D}=\\{\\psi^\\lambda_{j,k}\\}$ associated with the dyadic family $\\mathcal D$: each wavelet is obtained by orthonormalizing the indicator of a dyadic cube against the indicators of its children, so it has zero mean, unit $L^2$ norm, and support on one cube. The metric-side object is the pseudo-metric $\\delta_{\\mathcal D}(x,y)=\\inf\\{\\mu(Q): x,y\\in Q,\\, Q\\in\\mathcal D\\}$, whose balls are exactly the dyadic cubes and for which the measure of a radius-$r$ ball is comparable to $r$. The load-bearing identity is Lemma 6.4, which writes the mean difference $f_{\\tilde Q}-f_Q$ between a cube and its child as a finite linear combination of the wavelet inner products with coefficient norm $\\sqrt{(\\mu(Q)-\\mu(\\tilde Q))/(\\mu(Q)\\mu(\\tilde Q))}$; the inheritance condition $\\mu(Q)\\le B\\mu(\\tilde Q)$ bounds this by $\\sqrt{B(B-1)}/\\mu(Q)$ and supplies the geometric decay used to sum the ancestor chain in Theorem 6.6.","core_discovery":"The central result is Theorem 6.6: if $f$ is integrable and there is a constant $C>0$ such that $|\\langle f,\\psi^\\lambda_{j,k}\\rangle| \\le C(\\mu_{j,k})^{\\alpha+1/2}$ for every scale $j$, position $k$, and wavelet index $\\lambda$, then every pair of dyadic cubes at the same level satisfies $|f_{Q_{J,m}}-f_{Q_{J,n}}| \\le \\frac{2CB^\\alpha\\sqrt{B(B-1)}}{B^\\alpha-(B-1)^\\alpha}\\,\\delta_D(Q_{J,m},Q_{J,n})^\\alpha$. Thus $f$ belongs to the pixelated $\\alpha$-Lipschitz class $\\Lambda_{\\delta_D}(\\alpha)$ with seminorm bounded by that explicit constant depending only on $B$, $C$, and $\\alpha$. Theorems 6.1 and 6.3 provide the converse implications: functions that are $\\alpha$-Lipschitz, or pixelated $\\alpha$-Lipschitz, always have Haar coefficients with the stated decay. Corollary 6.7 upgrades the pixelated conclusion to pointwise almost-everywhere $\\alpha$-Lipschitz regularity when the dyadic family is also a differentiation basis.","pith_inferences":["The explicit dependence of the constant on the inheritance coefficient $B$ suggests that dyadic families with very unbalanced children require deeper scales before the power-law regime becomes visible, which could guide how many levels a texture-analysis computation needs.","A testable extension is to weaken the uniform inheritance bound to an averaged or logarithmic decay of child measures and check whether the ancestor sums still converge, with a modified constant.","The same coefficient-to-mean-difference mechanism likely yields norm equivalences between the pixelated class $\\Lambda_{\\delta_{\\mathcal D}}(\\alpha)$ and a Besov-type space on the dyadic tree, though the paper does not state this.","The numerical examples indicate that parabolic, anisotropic dyadic families separate directional textures better than square or band families, motivating further non-uniform dyadic decompositions tuned to image features."],"forward_implications":["The dyadic Lipschitz exponent of a function can be read directly from the decay rate of its generalized Haar coefficients, with an explicit Lipschitz constant $2CB^\\alpha\\sqrt{B(B-1)}/(B^\\alpha-(B-1)^\\alpha)$.","The coefficient-to-regularity implication holds even when the Haar system is only an orthonormal set rather than a basis and when the dyadic family is not a differentiation basis; the pixelated $\\alpha$-Lipschitz conclusion is unconditional.","When the dyadic family is a differentiation basis, the same coefficient bound upgrades to pointwise almost-everywhere $\\alpha$-Lipschitz regularity.","The log-linear relation between average wavelet coefficients and scale gives a computable estimator of the regularity exponent for images and time series, applicable to square, parabolic, and non-metric band decompositions.","Classical Haar characterizations on the real line and in metric measure spaces are special cases of the same mechanism."],"supporting_citations":[{"why":"Establishes the Haar-wavelet characterization of dyadic Lipschitz regularity that this paper extends to non-atomic probability spaces.","marker":"[1]"},{"why":"Provides the Haarlet characterization of Lipschitz regularity in metric measure spaces, whose metric-guided dyadic families are recovered as a special case.","marker":"[3]"},{"why":"Supplies the notion of differentiation basis used to pass from pixelated to pointwise Lipschitz regularity in Corollary 6.7 and Proposition 5.1.3.","marker":"[6]"},{"why":"Supports the construction of diverse dyadic families on any non-atomic separable probability space.","marker":"[8]"},{"why":"Motivates the power-law criterion by characterizing α-Lipschitz regularity through the continuous wavelet transform.","marker":"[9]"},{"why":"Supplies the companion continuous-wavelet pointwise-regularity result that anchors the coefficient-decay approach.","marker":"[10]"}],"fun_headline_variants":["Haar wavelet decay yields Lipschitz exponent, no metric required","Wavelet bound recovers Lipschitz regularity on abstract spaces","Explicit Lipschitz constants from Haar coefficient bounds","Power-law Haar decay implies Lipschitz even without metric"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Every dyadic child cube must carry at least a fixed fraction $1/B$ of its parent's measure, uniformly across the whole family; if children can shrink arbitrarily compared with their parents, the geometric-summing step in the converse direction can fail and a coefficient power law would no longer force a finite Lipschitz bound.","fun_headline_variants_meta":{"raw":{"variants":["Haar wavelet decay yields Lipschitz exponent, no metric required","Wavelet bound recovers Lipschitz regularity on abstract spaces","Explicit Lipschitz constants from Haar coefficient bounds","Power-law Haar decay implies Lipschitz even without metric"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000221,"raw_usage":{"total_tokens":1387,"prompt_tokens":820,"completion_tokens":567,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":436,"completion_tokens_details":{"reasoning_tokens":498}},"tokens_in":436,"tokens_out":567,"duration_ms":5425,"temperature":1.0,"reasoning_tokens":498,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:06:24.037113+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the pixelated Lipschitz seminorm of a single normalized Haar wavelet $\\psi^\\lambda_{j,k}$ with its own coefficient bound $C=(\\mu_{j,k})^{-(\\alpha+1/2)}$ and compare it with $2CB^\\alpha\\sqrt{B(B-1)}/(B^\\alpha-(B-1)^\\alpha)$; if the seminorm exceeds the predicted bound for some dyadic family satisfying the inheritance condition, Theorem 6.6 is wrong.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supports the construction of diverse dyadic families on any non-atomic separable probability space."},{"cited_title":"Aimar, E","cited_arxiv_id":null,"evidence_quote":"Establishes the Haar-wavelet characterization of dyadic Lipschitz regularity that this paper extends to non-atomic probability spaces."},{"cited_title":"Aimar, A","cited_arxiv_id":null,"evidence_quote":"Provides the Haarlet characterization of Lipschitz regularity in metric measure spaces, whose metric-guided dyadic families are recovered as a special case."},{"cited_title":"de Guzm´ an, Real variable methods in Fourier analysis , North-Holland Mathematics Studies, 46, North-Holland Publishing Co., Amsterdam-New York, 1981","cited_arxiv_id":null,"evidence_quote":"Supplies the notion of differentiation basis used to pass from pixelated to pointwise Lipschitz regularity in Corollary 6.7 and Proposition 5.1.3."},{"cited_title":"non-diff´ erentiable","cited_arxiv_id":null,"evidence_quote":"Motivates the power-law criterion by characterizing α-Lipschitz regularity through the continuous wavelet transform."},{"cited_title":"nondifferentiable","cited_arxiv_id":null,"evidence_quote":"Supplies the companion continuous-wavelet pointwise-regularity result that anchors the coefficient-decay approach."}],"review_version":1}