pith. sign in

arxiv: 2604.02150 · v2 · submitted 2026-04-02 · 🧮 math.NA · cs.NA· math.PR· math.ST· stat.ML· stat.TH

Samplet limits and multiwavelets

Gianluca Giacchi , Michael Multerer , Jacopo Quizi This is my paper

Pith reviewed 2026-05-13 21:00 UTC · model grok-4.3

classification 🧮 math.NA cs.NAmath.PRmath.STstat.MLstat.TH
keywords sampletsmultiwaveletsvanishing momentshierarchical partitionspolynomial primitivesscattered datamultiresolution analysisprobabilistic limit
0
0 comments X

The pith

Polynomial primitives turn samplet bases into multiwavelets with broken polynomial densities on hierarchical partitions in the infinite-data limit.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that samplets, defined as data-adapted multiresolution analyses of localized discrete signed measures on scattered points, converge when polynomials serve as primitives. In the probabilistic infinite-data limit these bases become signed measures whose densities consist of broken polynomials. The densities act as multiwavelets relative to a hierarchical partition of the domain. The same limit supplies a construction of multiwavelets that admits flexible vanishing moments without restriction to tensor-product forms. For congruent partitions the resulting filters are independent of scale and of the specific partition chosen.

Core claim

In the probabilistic framework, the samplet basis constructed with polynomials as primitives converges to signed measures that possess densities made of broken polynomials. These densities function as multiwavelets with respect to a hierarchical partition of the region containing the data sites. The resulting objects therefore furnish a general construction of multiwavelets that permits arbitrary prescription of vanishing moments beyond tensor-product restrictions. When the partitions are congruent, classical multiwavelets with scale- and partition-independent filter coefficients are recovered.

What carries the argument

The samplet basis with polynomial primitives in the probabilistic infinite-data limit, which produces signed measures whose densities are broken polynomials serving as multiwavelets on a hierarchical partition.

Load-bearing premise

The probabilistic data model together with the choice of polynomials as primitives is sufficient to produce exactly the claimed multiwavelet densities with broken polynomials in the infinite-data limit.

What would settle it

Numerical computation showing whether the samplet basis for successively larger numbers of data points approaches the predicted piecewise-polynomial multiwavelet densities on the hierarchical partition and satisfies the corresponding vanishing-moment conditions.

Figures

Figures reproduced from arXiv: 2604.02150 by Gianluca Giacchi, Jacopo Quizi, Michael Multerer.

Figure 1
Figure 1. Figure 1: Number of samples N and minimum number of samples per tree leaf for random sampling (left, average over 10 runs) and Halton points (right) [PITH_FULL_IMAGE:figures/full_fig_p021_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Visualization of the samplet convergence in d = 1 for increasing num￾ber of samples. A scaling distribution is shown on the left and a samplet on the right [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of the samplet convergence in d = 2 for increasing num￾ber of samples. A scaling distribution is shown on the left and a samplet on the right. To study the convergence, we employ an approximation with a larger number of samples as a reference. In all cases, we approximately use 5.37 · 108 samples for the computation of the references. The exact numbers are given in [PITH_FULL_IMAGE:figures/f… view at source ↗
Figure 4
Figure 4. Figure 4: Average projection error of filter coefficients of scaling distribution (left) and samplets (right) taken over all non-leaf clusters for increasing N and uniformly random points. The error bars denote the standard deviation from 10 runs. rate N −1/2 for d = 1, 2, 3. The standard deviation between the different runs is relatively small [PITH_FULL_IMAGE:figures/full_fig_p023_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: depicts the corresponding errors in case of Halton points. As expected, the convergence 100 101 102 103 104 105 106 10−6 10−4 10−2 minτ∈L(T ) |τ | average projection error d = 1 d = 2 d = 3 100 101 102 103 104 105 106 10−6 10−5 10−4 10−3 10−2 10−1 minτ∈L(T ) |τ | average projection error d = 1 d = 2 d = 3 [PITH_FULL_IMAGE:figures/full_fig_p023_5.png] view at source ↗
read the original abstract

Samplets are data adapted multiresolution analyses of localized discrete signed measures. They can be constructed on scattered data sites in arbitrary dimension such that they exhibit vanishing moments with respect to any prescribed set of primitives. We consider the samplet construction in a probabilistic framework and show that, if choosing polynomials as primitives, the resulting samplet basis converges to signed measures with broken polynomial densities in the infinite data limit. These densities amount to multiwavelets with respect to a hierarchical partition of the region containing the data sites. As a byproduct, we therefore obtain a construction of general multiwavelets that allows for a flexible prescription of vanishing moments going beyond tensor product constructions. For congruent partitions we particularly recover classical multiwavelets with scale- and partition- independent filter coefficients. The theoretical findings are complemented by numerical experiments that illustrate the convergence results in case of random as well as low-discrepancy data sites.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper defines samplets as data-adapted multiresolution analyses of localized discrete signed measures on scattered sites in arbitrary dimension, constructed to have vanishing moments with respect to prescribed primitives. In a probabilistic framework with polynomial primitives, it claims that the samplet basis converges in the infinite-data limit to signed measures whose densities are broken polynomials; these limits are asserted to be multiwavelets with respect to the underlying hierarchical partition. The construction yields a flexible method for generating multiwavelets with arbitrary vanishing moments (beyond tensor-product cases) and recovers classical multiwavelets with scale-independent filters on congruent partitions. Numerical experiments on random and low-discrepancy points illustrate the convergence.

Significance. If the convergence and identification with multiwavelets are rigorously established, the work supplies a probabilistic route to multiwavelet constructions that is dimension-independent and allows direct prescription of vanishing moments, which is a genuine advance over existing tensor-product or spline-based approaches. The byproduct construction of general multiwavelets and the recovery of classical filters on congruent partitions are particularly valuable for approximation theory and numerical PDE methods on irregular domains. The numerical illustrations provide supporting evidence but do not substitute for the missing analytic step.

major comments (2)
  1. [§4] §4 (convergence theorem): the proof establishes vanishing moments and support localization for the limiting signed measures but does not derive that the limiting densities satisfy the two-scale refinement relations or the precise broken-polynomial form required to qualify as multiwavelets (beyond the vanishing-moment property). Because multiwavelets are defined by both properties, this verification is load-bearing for the central claim that the limits 'amount to multiwavelets'.
  2. [§3.2] §3.2 (probabilistic limit construction): the argument assumes that the hierarchical partition and filter coefficients converge in a manner that preserves the multiwavelet scaling relations, yet no explicit control (e.g., uniform bounds on the partition-of-unity functions or convergence of the refinement masks) is provided for the infinite-data limit. This step must be supplied to close the identification.
minor comments (2)
  1. Notation for the broken-polynomial densities should be introduced earlier (e.g., before the statement of the main theorem) to improve readability.
  2. The numerical experiments section would benefit from explicit reporting of the discrepancy measures and the precise polynomial degree used in each test case.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. The points raised concern the completeness of the identification of the limiting objects as multiwavelets, and we will revise the paper to supply the missing explicit derivations while preserving the original probabilistic framework.

read point-by-point responses

  1. Referee: [§4] §4 (convergence theorem): the proof establishes vanishing moments and support localization for the limiting signed measures but does not derive that the limiting densities satisfy the two-scale refinement relations or the precise broken-polynomial form required to qualify as multiwavelets (beyond the vanishing-moment property). Because multiwavelets are defined by both properties, this verification is load-bearing for the central claim that the limits 'amount to multiwavelets'.

    Authors: We acknowledge that the current argument in §4 verifies vanishing moments and localization but stops short of explicitly passing to the limit in the discrete two-scale relations. In the revised manuscript we will add a dedicated paragraph deriving the refinement relations for the limiting densities by taking the limit (in the weak-* topology) of the finite-data refinement equations, using the already-established convergence of the samplet coefficients. This step will also confirm that the densities are precisely broken polynomials on the limiting hierarchical partition. revision: yes

  2. Referee: [§3.2] §3.2 (probabilistic limit construction): the argument assumes that the hierarchical partition and filter coefficients converge in a manner that preserves the multiwavelet scaling relations, yet no explicit control (e.g., uniform bounds on the partition-of-unity functions or convergence of the refinement masks) is provided for the infinite-data limit. This step must be supplied to close the identification.

    Authors: The referee correctly notes the absence of explicit uniform bounds. We will insert into §3.2 a new lemma establishing uniform bounds on the partition-of-unity functions that follow directly from the probabilistic assumptions on the data sites (via concentration inequalities already used elsewhere in the section). We will also prove convergence of the refinement masks in the sup-norm, which together close the passage to the limit in the scaling relations. revision: yes

Circularity Check

0 steps flagged

No circularity: limit theorem derived independently from samplet construction

full rationale

The central claim is a convergence result showing that samplet bases with polynomial primitives converge in the infinite-data probabilistic limit to signed measures whose densities are multiwavelets on a hierarchical partition. This follows from the samplet vanishing-moment construction and standard probabilistic arguments on scattered data; no equation reduces the limiting densities to a fitted parameter, a self-defined quantity, or a load-bearing self-citation. The derivation chain is self-contained against external benchmarks and does not invoke uniqueness theorems or ansatzes from prior author work to force the multiwavelet form. Minor self-citations, if present, are not load-bearing for the limit statement.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The result rests on the definition of samplets, the probabilistic model for the data sites, and the choice of polynomials as primitives; no free parameters are fitted inside the limit statement itself.

axioms (2)
  • domain assumption The data sites are drawn from a probabilistic model whose density is positive on the domain.
    Invoked to obtain the infinite-data limit that produces the broken polynomial densities.
  • domain assumption Polynomials are admissible primitives for the vanishing-moment condition.
    The paper restricts the general samplet construction to polynomials to reach the multiwavelet limit.

pith-pipeline@v0.9.0 · 5461 in / 1416 out tokens · 29221 ms · 2026-05-13T21:00:46.388651+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages

  1. [1]

    B. K. Alpert. A class of bases inL2 for the sparse representation of integral operators.SIAM J. Math. Anal., 24(1):246–262, 1993

    work page 1993

  2. [2]

    B. K. Alpert, G. Beylkin, R. R. Coifman, and V. Rokhlin. Wavelet-like bases for the fast solution of second-kind integral equations.SIAM J. Sci. Comput., 14(1):159–184, 1993

    work page 1993

  3. [3]

    A. I. Aptekarev. Multiple orthogonal polynomials.J. Comput. Appl. Math., 99(1):423–447, 1998

    work page 1998

  4. [4]

    Balazs and M

    P. Balazs and M. Multerer. Construction of generalized samplets in Banach spaces.arXiv, 2025. Preprint arXiv:2412.00954

    work page arXiv 2025

  5. [5]

    Basu and A

    K. Basu and A. B. Owen. Transformations and Hardy–Krause variation.SIAM J. Numer. Anal., 54(3):1946– 1966, 2016

    work page 1946

  6. [6]

    G. Beylkin. On the representation of operators in bases of compactly supported wavelets.SIAM J. Numer. Anal., 29(6):1716–1740, 1992

    work page 1992

  7. [7]

    Beylkin, R

    G. Beylkin, R. Coifman, and V. Rokhlin. Fast wavelet transforms and numerical algorithms I.Comm. Pure Appl. Math., 44(2):141–183, 1991

    work page 1991

  8. [8]

    J. H. Bramble and S. R. Hilbert. Estimation of linear functionals on Sobolev spaces with application to Fourier transforms and spline interpolation.SIAM J. Numer. Anal., 7:112–124, 1970

    work page 1970

  9. [9]

    R. E. Caflisch. Monte Carlo and quasi-Monte Carlo methods. InActa Numerica, 1998, volume 7 ofActa Numer., pages 1–49. Cambridge Univ. Press, Cambridge, 1998

    work page 1998

  10. [10]

    T. S. Chihara.An Introduction to Orthogonal Polynomials. Mathematics and its Applications, Vol. 13. Gordon and Breach Science Publishers, New York-London-Paris, 1978

    work page 1978

  11. [11]

    Chkifa, A

    A. Chkifa, A. Cohen, and C. Schwab. High-dimensional adaptive sparse polynomial interpolation and applica- tions to parametric PDEs.Found. Comput. Math., 14(4):601–633, 2014

    work page 2014

  12. [12]

    Cohen.Numerical Analysis of Wavelet Methods, volume 32 ofStudies in Mathematics and its Applications

    A. Cohen.Numerical Analysis of Wavelet Methods, volume 32 ofStudies in Mathematics and its Applications. North-Holland Publishing Co., Amsterdam, 2003

    work page 2003

  13. [13]

    W. Dahmen. Multiscale and wavelet methods for operator equations.Acta Numer., 6:55–228, 1997

    work page 1997

  14. [14]

    Daubechies.Ten Lectures on Wavelets

    I. Daubechies.Ten Lectures on Wavelets. SIAM, Philadelphia, PA, 1992

    work page 1992

  15. [15]

    Dekel and D

    S. Dekel and D. Leviatan. The Bramble-Hilbert lemma for convex domains.SIAM J. Math. Anal., 35(5):1203– 1212, 2004

    work page 2004

  16. [16]

    R. A. DeVore. Nonlinear approximation.Acta Numer., 7:51–150, 1998. SAMPLET LIMITS AND MULTIW A VELETS 25

    work page 1998

  17. [17]

    D. L. Donoho, N. Dyn, D. Levin, and T. P.-Y. Yu. Smooth multiwavelet duals of Alpert bases by moment- interpolating refinement.Appl. Comput. Harmon. Anal., 9(2):166–203, 2000

    work page 2000

  18. [18]

    R. M. Dudley.Real Analysis and Probability. Cambridge University Press, Cambridge, 2002

    work page 2002

  19. [19]

    Elefante, G

    G. Elefante, G. Giacchi, M. Multerer, and J. Quizi. Bespoke multiresolution analysis of graph signals.arXiv,

    work page

  20. [20]

    Preprint arXiv:2507.19181

    work page arXiv

  21. [21]

    J. S. Geronimo, P. Iliev, and W. Van Assche. Alpert multiwavelets and Legendre-Angelesco multiple orthogonal polynomials.SIAM J. Math. Anal., 49(1):626–645, 2017

    work page 2017

  22. [22]

    Gerstner and M

    T. Gerstner and M. Griebel. Dimension-adaptive tensor-product quadrature.Computing, 71(1):65–87, 2003

    work page 2003

  23. [23]

    Harbrecht and M

    H. Harbrecht and M. Multerer. Samplets: Construction and scattered data compression.J. Comput. Phys., 471:111616, 2022

    work page 2022

  24. [24]

    Klenke.Probability Theory - A Comprehensive Course

    A. Klenke.Probability Theory - A Comprehensive Course. Universitext. Springer, Cham, 2020. Third edition

    work page 2020

  25. [25]

    Mallat.A Wavelet Tour of Signal Processing: The Sparse Way

    S. Mallat.A Wavelet Tour of Signal Processing: The Sparse Way. Academic Press, Burlington, MA, 3rd edition, 2008

    work page 2008

  26. [26]

    Meyer.Wavelets and Operators, volume 37 ofCamb

    Y. Meyer.Wavelets and Operators, volume 37 ofCamb. Stud. Adv. Math.Cambridge University Press, Cam- bridge, 1993

    work page 1993

  27. [27]

    B. S. Mityagin. The zero set of a real analytic function.Mat. Zametki, 107(3):473–475, 2020

    work page 2020

  28. [28]

    Rioul and M

    O. Rioul and M. Vetterli. Wavelets and signal processing.IEEE Signal Process. Mag., 8(4):14–38, 1991

    work page 1991

  29. [29]

    Szegő.Orthogonal Polynomials

    G. Szegő.Orthogonal Polynomials. American Mathematical Society Colloquium Publications, Vol. XXIII. American Mathematical Society, Providence, RI, fourth edition, 1975

    work page 1975

  30. [30]

    V. S. Varadarajan. On the convergence of sample probability distributions.Sankhya, 19:23–26, 1958

    work page 1958

  31. [31]

    Wendland.Scattered Data Approximation, volume 17 ofCambridge Monographs on Applied and Computa- tional Mathematics

    H. Wendland.Scattered Data Approximation, volume 17 ofCambridge Monographs on Applied and Computa- tional Mathematics. Cambridge University Press, Cambridge, 2005. AppendixA.Rates, discrepancy, and deterministic bounds A.1.Star discrepancy and the Koksma-Hlawka inequality.LetD= [0,1] d andX N = {x1, . . . ,xN } ⊂D. The star discrepancy ofX N is defined as...

    work page 2005