Pith. sign in

REVIEW 5 minor 24 references

Smoothing by, and eccentric smoothing of, compactly supported RBFs

T0 review · 0 major / 5 minor · reviewed 2026-07-11 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Multiplying a compactly supported RBF by a smoother one raises its smoothness away from zero while preserving native-space and Fourier decay.

desk verdict Clean elementary construction that raises eccentric smoothness of compactly supported RBFs without changing the native space, unlocking samplet compression and Hörmander-class mapping theorems. read the letter →

arxiv 2607.04512 v1 pith:NUPRB6FG submitted 2026-07-05 math.NA cs.NA

Thomas Hangelbroek , Christian Rieger This is my paper · ORCID
classification math.NAcs.NA MSC 65D1265F5547G30
keywords compactlysupportedRBFseccentricsmoothinggeneralizedWendlandfunctionssampletcompressionpseudodifferentialoperatorsBesovspacesasymptoticsmoothness
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Compactly supported radial basis functions are valued for sparse collocation matrices, yet their limited smoothness away from the origin blocks both high-order compression and clean mapping properties of the associated integral operators. This paper shows that a simple pointwise multiplication by a second compactly supported positive-definite kernel of higher smoothness produces a new kernel that keeps the original global smoothness and algebraic Fourier decay, while the smoothness away from zero (the eccentric smoothness) can be raised arbitrarily. When that eccentric smoothness becomes infinite the integral operator becomes a properly supported Hörmander-class pseudodifferential operator, so it maps every Besov space $B^s_{p,q}$ continuously into $B^{s+2t}_{p,q}$. The same construction also guarantees the finite-order asymptotic smoothness needed for samplet compression of the collocation matrix, without the conditioning penalty that would accompany a genuine increase in global smoothness. The result therefore supplies a practical route to better-conditioned, compressible kernel matrices and to rigorously controlled smoothing operators that remain compactly supported.

What carries the argument

Proposition 3.1 (eccentric smoothing via pointwise product): the product of two compactly supported positive-definite RBFs inherits the global smoothness and Fourier decay of the less regular factor while adding their eccentric smoothness parameters (or taking the minimum with the smoother factor’s regularity).

What would settle it

Construct an explicit product kernel with the claimed infinite eccentric smoothness, compute its Fourier transform numerically on a large frequency grid, and check whether the lower bound $C \langle \xi \rangle^{-2t}$ still holds; if the lower bound collapses for large $|\xi|$, the pseudodifferential and Besov-mapping claims fail.

Watch

Extended reading notes

Core claim

Eccentric smoothing—pointwise multiplication of a compactly supported RBF of global smoothness $\lambda_1$ by a second compactly supported positive-definite factor of sufficient smoothness—yields a new RBF that retains the original native space $W^t_2$ and the Fourier bounds $C \langle \xi \rangle^{-2t}$, yet possesses arbitrarily high (or infinite) smoothness on $\mathbb{R}^d \setminus \{0\}$. When the eccentric smoothness is infinite the associated integral operator is a properly supported pseudodifferential operator with symbol in the Hörmander class $S^{-2t}_{1,0}$ and therefore acts continuously on every Besov scale.

Load-bearing premise

The lower Fourier bound of the original kernel must survive convolution against the Fourier transform of the smoother factor; the argument needs that Fourier transform to be strictly positive in a neighbourhood of the origin.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

0 major / 5 minor

Summary. The paper introduces eccentric smoothing of compactly supported radial basis functions (RBFs) that satisfy Definition 1 (polyhomogeneous form with global smoothness λ_{1}, eccentric smoothness λ_{2}, and Fourier decay Jξ K^{-2t}). Proposition 3.1 shows that pointwise multiplication by a second compactly supported positive-definite kernel τ of sufficient smoothness produces a new RBF that retains the original global smoothness and Fourier decay while raising the eccentric smoothness (arbitrarily, including to infinity). Corollaries 4.1–4.2 then establish that the resulting kernels are asymptotically smooth of finite order, which makes samplet compression of the collocation matrix feasible while preserving the conditioning controlled by λ_{1}. When the eccentric smoothness is infinite, Proposition 5.1 proves that the integral operator is a properly supported pseudodifferential operator with symbol in the Hörmander class S^{-2t}_{1,0}; Corollaries 5.2–5.3 deduce continuous mapping properties on Besov spaces B^s_{p,q} and on L^p(μ) for finite measures μ.

Significance. The construction is elementary yet useful: it decouples the eccentric smoothness parameter from the native-space index t, so that samplet compression (and the associated O(N log N) multiscale algorithms) becomes available for compactly supported kernels without the usual penalty on the condition number of Φ_Ξ. The infinite-smoothness case further places the associated integral operators inside the classical Hörmander calculus, yielding clean mapping statements on the full scale of Besov spaces that were previously unavailable for compactly supported RBFs. The proofs rely only on the product rule, Leibniz rule, Peetre’s inequality and standard symbol estimates; they are fully written out and appear free of circularity or hidden parameters. The work therefore supplies a practical tool for kernel methods and a clean theoretical bridge between compactly supported RBFs and pseudodifferential operators.

minor comments (5)
  1. In the abstract and the first paragraph of the introduction the phrase “we show that eccentric smoothing makes wavelet-inspired compression technique for the kernel matrix feasible” is slightly ungrammatical; a definite article or plural would improve readability.
  2. Section 4, after Corollary 4.2: the short argument that guarantees positive-definiteness of the compressed matrix K_η for η ≳ q_Ξ^{-(λ_{1}+d+2)/(κ+λ_{2}+1)} is useful but terse; a one-sentence reminder that the Frobenius-to-spectral-norm comparison uses the quasi-uniformity assumption N ∼ q_Ξ^{-d} would help the reader.
  3. Proposition 5.1: the sentence “Fix α It suffices to show…” is missing a period after “α”.
  4. Throughout the manuscript a few compound words are missing spaces (“showthateccentricsmoothing”, “positivedefiniteness”, “Peetre’sinequality”). These are purely typographic and do not affect the mathematics.
  5. Reference [11] is listed as a 2026 preprint; if a stable arXiv identifier is already available it would be helpful to include it for the reader.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the eccentric-smoothing construction and PDO/mapping claims are derived directly from the product definition, Leibniz rule, Fourier convolution and Peetre estimates.

full rationale

The paper’s central results (Proposition 3.1, Corollaries 4.1–4.2, Proposition 5.1 and the Besov mapping corollaries) are obtained by an explicit pointwise product ˜ϕ=τϕ together with elementary product-rule/Leibniz arguments for smoothness, convolution of Fourier transforms for the symbol bounds, and standard Peetre inequalities. These steps are self-contained and do not reduce to any fitted parameter, self-referential normalization, or uniqueness claim imported from the authors’ prior work. Self-citations ([10], [11]) supply only background motivation and known analytic properties of generalized Wendland kernels that are independently available in the literature ([6], [23]); they are not load-bearing for the new estimates. No data-fitting, no “prediction” forced by construction, and no renaming of a known empirical pattern appear. Consequently the derivation chain is free of circularity.

Assumptions & free parameters 0 free parameters · 4 assumptions · 1 invented entities

The paper rests on classical Fourier analysis, the known Fourier asymptotics of generalized Wendland functions, Peetre’s inequality, and the mapping properties of Hörmander-class pseudodifferential operators. No free parameters are fitted; the only new entity is the secondary smoothness parameter itself, which is defined by the construction.

assumptions (4)
  • domain assumption Generalized Wendland functions satisfy the polyhomogeneous representation (2.1) and the two-sided Fourier bound (2.2) with t = (λ₁ + d + 1)/2.
    Taken from Chernih–Hubbert and Wendland’s monographs; used as the starting point for Definition 1 and all subsequent estimates.
  • standard math Peetre’s inequality: ∫ JζK^{-N1} Jξ-ζK^{-N2} dζ ≤ C JξK^{-min(N1,N2)}.
    Invoked repeatedly for convolution estimates of Fourier transforms (proof of Prop. 3.1 and Prop. 5.1).
  • standard math Hörmander-class operators with symbols in S^m_{1,0} map Besov spaces B^s_{p,q} continuously into B^{s-m}_{p,q}.
    Cited from Triebel; used to obtain Corollary 5.2 once the symbol class is established.
  • standard math A positive continuous function that is bounded below on a neighborhood of the origin yields a strictly positive convolution lower bound after Peetre.
    Used to preserve the lower Fourier bound under multiplication by τ.
invented entities (1)
  • eccentric smoothness parameter λ₂
    purpose: Secondary regularity index measuring smoothness of the RBF on R^d \ {0}; raised by the product construction without changing the native space.
    Defined in the introduction and Definition 1; the whole paper is devoted to controlling and exploiting this parameter.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Smoothing by, and eccentric smoothing of, compactly supported RBFs." pith.science (2026). https://pith.science/paper/NUPRB6FG

@misc{pith2026260704512,
  author       = {Pith},
  title        = {Pith review of: Smoothing by, and eccentric smoothing of, compactly supported RBFs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NUPRB6FG}},
  note         = {Machine review of arXiv:2607.04512}
}
read the original abstract

We consider compactly supported RBFs having algebraically decaying Fourier transforms. Here we focus especially on generalized Wendland RBFs and their modification by making them smoother away from zero, a process we call eccentric smoothing. Specifically, we consider mapping properties of the integral operators (sometimes known as covariance operator) for these novel type RBFs. Moreover, we show that eccentric smoothing makes wavelet-inspired compression technique for the kernel matrix feasible.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 1 linked inside Pith

  1. [1]

    Amal Al-Rashdan and Michael J. Johnson. Minimal degree univariate piecewise polynomials with prescribed sobolev regularity.Journal of Approximation Theory, 164(1):1–5, 2012

    2012

  2. [2]

    Multiscale Scattered Data Analysis in Samplet Coordinates.SIAM J

    Sara Avesani, Rüdiger Kempf, Michael Multerer, and Holger Wendland. Multiscale Scattered Data Analysis in Samplet Coordinates.SIAM J. Sci. Comput., 47(5):A3038–A3063, 2025

    2025

  3. [3]

    An introduction, volume No

    Jöran Bergh and Jörgen Löfström.Interpolation spaces. An introduction, volume No. 223 of Grundlehren der Mathematischen Wissenschaften. Springer-Verlag, Berlin-New York, 1976

    1976

  4. [4]

    Kluwer Academic Publishers, Boston, MA, 2004

    Alain Berlinet and Christine Thomas-Agnan.Reproducing kernel Hilbert spaces in probability and statistics. Kluwer Academic Publishers, Boston, MA, 2004. With a preface by Persi Diaconis

    2004

  5. [5]

    M. D. Buhmann. Radial functions on compact support.Proc. Edinburgh Math. Soc. (2), 41(1):33–46, 1998

    1998

  6. [6]

    Closed form representations and properties of the generalised Wendland functions.J

    Andrew Chernih and Simon Hubbert. Closed form representations and properties of the generalised Wendland functions.J. Approx. Theory, 177:17–33, 2014

    2014

  7. [7]

    On the mathematical foundations of learning.Bull

    Felipe Cucker and Steve Smale. On the mathematical foundations of learning.Bull. Amer. Math. Soc. (N.S.), 39(1):1–49, 2002

    2002

  8. [8]

    Springer, New York, third edition, 2014

    Loukas Grafakos.Modern Fourier analysis, volume 250 ofGraduate Texts in Mathematics. Springer, New York, third edition, 2014

    2014

Show all 24 references
  1. [9]

    Springer, Heidelberg, 2015

    Wolfgang Hackbusch.Hierarchical matrices: algorithms and analysis, volume 49 ofSpringer Series in Computational Mathematics. Springer, Heidelberg, 2015

    2015

  2. [10]

    Hangelbroek and C

    T. Hangelbroek and C. Rieger. Extending error bounds for radial basis function interpolation to measuring the error in higher order Sobolev norms.Math. Comp., 94(351):381–407, 2025

    2025

  3. [11]

    Hangelbroek, C

    T. Hangelbroek, C. Rieger, and G. B. Wright. Kernel approximation beyond the native space – with applications to approximation on manifolds, 2026. preprint arXiv:2606.28564

    2026 arXiv

  4. [12]

    Harbrecht, M

    H. Harbrecht, M. Multerer, O. Schenk, and Ch. Schwab. Multiresolution kernel matrix algebra. Numer. Math., 156(3):1085–1114, 2024

    2024

  5. [13]

    Samplets: construction and scattered data compression.J

    Helmut Harbrecht and Michael Multerer. Samplets: construction and scattered data compression.J. Comput. Phys., 471:Paper No. 111616, 23, 2022

    2022

  6. [14]

    Closed form representations for the compactly supported radial basis functions of Buhmann, Wendland and Wu.Adv

    Simon Hubbert and Janin Jäger. Closed form representations for the compactly supported radial basis functions of Buhmann, Wendland and Wu.Adv. Comput. Math., 51(6):Paper No. 48, 30, 2025

    2025

  7. [15]

    Distributed learning with regularized least squares.J

    Shao-Bo Lin, Xin Guo, and Ding-Xuan Zhou. Distributed learning with regularized least squares.J. Mach. Learn. Res., 18:Paper No. 92, 31, 2017

    2017

  8. [16]

    Platte.C ∞ compactly supported and positive definite radial kernels.SIAM Journal on Scientific Computing, 37(4):A1934–A1956, 2015

    Rodrigo B. Platte.C ∞ compactly supported and positive definite radial kernels.SIAM Journal on Scientific Computing, 37(4):A1934–A1956, 2015

    2015

  9. [17]

    Schaback

    R. Schaback. Improved error bounds for scattered data interpolation by radial basis functions.Math. Comp., 68(225):201–216, 1999

    1999

  10. [18]

    The missing Wendland functions.Adv

    Robert Schaback. The missing Wendland functions.Adv. Comput. Math., 34(1):67–81, 2011

    2011

  11. [19]

    Kernel construction techniques.Engineering Analysis with Boundary Elements, 188:106782, 2026

    Robert Schaback. Kernel construction techniques.Engineering Analysis with Boundary Elements, 188:106782, 2026

    2026

  12. [20]

    Taylor.Partial differential equations II

    Michael E. Taylor.Partial differential equations II. Qualitative studies of linear equations, volume 116 ofApplied Mathematical Sciences. Springer, Cham, third edition, 2023

    2023

  13. [21]

    II, volume 84 ofMonographs in Mathematics

    Hans Triebel.Theory of function spaces. II, volume 84 ofMonographs in Mathematics. Birkhäuser Verlag, Basel, 1992

    1992

  14. [22]

    Piecewise polynomial, positive definite and compactly supported radial functions of minimal degree.Adv

    Holger Wendland. Piecewise polynomial, positive definite and compactly supported radial functions of minimal degree.Adv. Comput. Math., 4(4):389–396, 1995

    1995

  15. [23]

    Cambridge University Press, Cambridge, 2005

    Holger Wendland.Scattered data approximation, volume 17 ofCambridge Monographs on Applied and Computational Mathematics. Cambridge University Press, Cambridge, 2005

    2005

  16. [24]

    Compactlysupportedpositivedefiniteradialfunctions.Adv

    ZongMinWu. Compactlysupportedpositivedefiniteradialfunctions.Adv. Comput. Math., 4(3):283– 292, 1995. 9

    1995

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.