Pith. sign in

REVIEW 2 major objections 4 minor 42 references

Extending Data to Improve Stability and Error Estimates Using Asymmetric Kansa-like Methods to Solve PDEs

T0 review · 2 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Choosing the test set Y as a norming set for the trial space makes the rectangular Kansa matrix well-posed, and both least-squares and thinned square systems converge with explicit rates on spheres.

desk verdict Solid stability result for oversampled Kansa matrices, but the advertised error rates overreach: the regularity of L(u-I_Xu) is insufficient for the rates in Theorems 6.3 and 7.2. read the letter →

arxiv 2507.15137 v1 pith:KC4CEMZJ submitted 2025-07-20 math.NA cs.NA

Thomas Hangelbroek , Francis J. Narcowich , Joseph D. Ward This is my paper · ORCID
classification math.NAcs.NA MSC 65N3565D1265N12
keywords KansamethodasymmetriccollocationnormingsetsBernsteininequalitysphericalbasisfunctionsrank-revealingQRerrorestimatesmeshlessPDEmethods
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

The paper addresses a known weakness of Kansa's asymmetric collocation method for elliptic PDEs on spheres and other closed manifolds: the collocation matrix is often ill-conditioned or even singular, and stability depends on an unverifiable compatibility between the kernel and the differential operator. The authors propose replacing the test point set $X$ with a larger set $Y$ that is a norming set for the trial space, producing a rectangular Kansa-like matrix. They prove that this matrix is bounded below independently of problem size, with a lower bound proportional to the square root of the oversampling ratio $\kappa = \#Y/\#X$. Under that stability control, they derive two concrete approximation routes with explicit $L^2$ error rates: discrete least squares and a rank-revealing thinning of $Y$ down to size $\#X$. The point is to make meshless collocation reliable for operators and kernels chosen independently of one another.

What carries the argument

The central object is the norming set $Y$ for the trial space $S_X(L\Phi)$: a quasi-uniform point set with $\#Y = \kappa\#X$ for which $\|w\|_{L^2} \leq C M^{-1/2}\|w|_Y\|_{\ell^2(Y)}$ for every $w$ in that space. Its existence follows from a Bernstein inequality for the kernel space, an inequality bounding a higher Sobolev norm of any trial function by $q_X^{-\gamma}$ times its $L^2$ norm, combined with a sampling inequality that bounds an $L^p$ norm by a fill-distance-weighted combination of the Sobolev norm and discrete samples. Carrying the argument is the stability ratio $r_2(X)$ of the basis used for the Kansa matrix; with a Riesz (Lagrange or localized) basis, the lower bound on the rectangular matrix becomes proportional to $\sqrt{\kappa}$. For the thinning route, the mechanism is a strong rank-revealing QR factorization that selects $N$ of the $M$ test points while keeping the smallest singular value of the reduced square matrix at least $C N^{-1}$.

What would settle it

Compute the smallest singular value $\sigma_{\min}(K)$ of the rectangular Kansa matrix for a thin-plate-spline space $S_X(\phi_s)$, a smooth elliptic $L$ satisfying (2.2), and quasi-uniform norming sets $Y$ with fixed oversampling ratio $\kappa$ as $N \to \infty$: the paper predicts $\sigma_{\min}(K) \geq C\sqrt{\kappa}$ with $C$ independent of $N$, so observing $\sigma_{\min}(K)$ decaying like a negative power of $N$ would falsify Theorem 5.2.

Watch

Extended reading notes

Core claim

On the sphere $S^d$ (or a smooth closed Riemannian manifold), let $L$ be a uniformly strongly elliptic second-order operator satisfying the lower bound $\|Lf\|_{L^2(M)} \geq c_L\|f\|_{L^2(M)}$, and let $S_X(\Phi)$ be the kernel trial space generated by thin-plate splines or similar spherical basis functions. The central claim is that if $Y$ is chosen as a quasi-uniform norming set for the space $S_X(L\Phi)$, then the rectangular Kansa matrix $K = (LB_k(y_j))$ built from a Riesz basis $\{B_k\}$ satisfies the inverse inequality $\|K a\|_{\ell^2(Y)} \geq (c_L c_R/C_N)\sqrt{\kappa/2}\,\|a\|_{\ell^2(X)}$ (Theorem 5.2), so it is well-posed and its condition behavior is controlled. From this, the least-squares solution of the overdetermined system $Ka = f|_Y$ converges with rate $\|u - u^*\|_{L^2} \leq C\rho^{2s+d} q_X^{2s+d-2}\|u\|_{H^{2s+d}}$ (Theorem 6.3), and the thinned square system obtained by a strong rank-revealing QR factorization achieves $\|u - u^*\|_{L^2} \leq C\rho^{2s+d} q_X^{2s-2}\|u\|_{H^{2s+d}}$ (Theorem 7.2).

Load-bearing premise

The analysis starts from the standing assumption that the elliptic operator $L$ satisfies $\|Lf\|_{L^2(M)} \geq c_L\|f\|_{L^2(M)}$, so $L$ has no kernel and the two-sided estimate (2.3) holds; if $L$ is, for example, the pure Laplace-Beltrami operator, the stability lower bound and all error estimates that rely on it do not apply.

Editorial extensions

If this is right

  • Kansa-type collocation becomes a provably convergent method for elliptic operators chosen independently of the kernel, provided the operator satisfies the lower-bound condition (2.2).
  • The discrete least-squares route has high-order accuracy: for smooth solutions, $\|u-u^*\|_{L^2} \leq C q_X^{2s+d-2}$, where $s$ is the thin-plate-spline smoothness parameter.
  • The thinning route uses only $\#X$ test points after a rank-revealing QR step and retains essentially the same rate, so the square-system version of Kansa collocation is also provably stable.
  • The stability of the Kansa matrix is comparable to the stability of the elliptic operator acting on the trial space, so ill-conditioning of the kernel basis no longer contaminates the method when a Lagrange or localized basis is used.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The authors do not treat boundary-value problems; the same Bernstein-inequality construction suggests that oversampling by quasi-uniform norming sets should also stabilize Kansa collocation on bounded Euclidean domains with boundary conditions, provided the elliptic operator's kernel is controlled by the boundary data.
  • The $N^{-1}$ loss in the thinned system's lower bound comes entirely from the rank-revealing factorization's factor $q_1(N,M) \sim N$; replacing it with a randomized or greedy subset-selection rule would likely improve the $q_X^{2s-2}$ rate in Theorem 7.2, but the paper does not pursue this.
  • The theory assumes $M \sim N$ with a fixed oversampling ratio $\kappa$; a numerical check that $\sigma_{\min}(K)$ stays bounded below as $N$ grows, for a fixed operator and kernel, would separate the asymptotic claim from the constants.
  • Operators with a nontrivial nullspace, such as pure Laplace-Beltrami, are excluded by (2.2); one plausible extension is to project the equation onto the orthogonal complement of the kernel, which the authors do not address.
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

2 major / 4 minor

Summary. The manuscript develops a theoretical framework for stabilizing the asymmetric Kansa collocation method for elliptic PDEs on spheres and other compact manifolds. It replaces the square test set X by a larger norming set Y for the trial space S_X(LΦ), producing a rectangular Kansa matrix K=(LB_j(y_k)). Under the ellipticity/coercivity assumption (2.2) on L and Bernstein/Riesz-basis hypotheses, Theorem 5.2 gives the lower bound ∥Ka∥_{ℓ2(Y)} ≥ c√κ∥a∥_{ℓ2(X)}. The paper then analyzes two solution strategies: discrete least squares (Theorem 6.3) and an RRQR-thinned square system (Theorem 7.2), claiming L2 error rates q^{2s+d-2} and q^{2s-2} for thin-plate splines. The norming-set construction via Bernstein inequalities (Theorem 4.1) and the stability lower bound are the main novel ingredients.

Significance. If the error estimates were valid, the paper would be a significant step: it would provide a kernel-independent stabilization mechanism for Kansa with an explicit oversampling factor, and a principled thinning procedure. The stability part (Theorem 5.2) is well argued and rests on a clean combination of norming sets, Riesz bases, and the two-sided estimate (2.3); the norming-set construction is also a useful contribution. The main advertised accuracy claims, however, are not established as written: the proofs use a residual smoothness that exceeds what the thin-plate spline trial space can have. I therefore regard the central error estimates as needing substantial revision, not as ready for publication in their present form.

major comments (2)
  1. [Proposition 6.1 / Eq. (3.10)] The stated interpolation rate is not supported by the kernel's regularity. The Gamma ratio in (3.10) decays as ℓ^{-(2s+d)}, not as the displayed ℓ^{2s+d}; a translate of the thin-plate spline φ_s belongs to H^γ only for γ<2s+d/2, which is exactly the range in the authors' own Bernstein inequality (Corollary 3.5 with β=2s+d). Consequently Proposition 6.1 cannot be true for β>2s+d/2, since u-I_Xu would have infinite H^β norm for a generic smooth u and a generic point set X. The appeal to [34, Theorem A.3] with 2τ=2s+d requires checking that theorem's hypotheses against the actual coefficient decay; as stated, the proposition overstates the smoothness available in the trial space.
  2. [Section 6, Eq. (6.3)] The estimate ∥I_Yg-g∥_{L2} ≤ C q_Y^{2s+d-2} ∥g∥_{H^{2s+d-2}} is applied to g=L(u-I_Xu). From Corollary 3.5 (or directly from (3.10)), I_Xu is at best in H^{2s+d/2-ε}, so g is at best in H^{2s+d/2-2-ε}. Since 2s+d-2 exceeds 2s+d/2-2 for d>0, the norm ∥g∥_{H^{2s+d-2}} is not controlled by the preceding interpolation estimate and is generically infinite for the residual of a thin-plate-spline interpolant. This gap is load-bearing: it is the step that produces (6.4), (6.7), Theorem 6.3, and Theorem 7.2. The proof needs either an improved regularity argument for g or corrected rates consistent with the Bernstein range, together with a numerical rate check on a concrete example (e.g., S^2, s=1) to support the claimed exponents.
minor comments (4)
  1. [Section 2.2, Eq. (2.2)] The assumption (2.2) excludes the pure Laplace-Beltrami operator, which is a standard test case for Kansa methods; the abstract and introduction should state this restriction explicitly so the scope of the theory is not overstated.
  2. [Lemma 4.2 proof] In the proof of Lemma 4.2, after applying (4.7) and (4.6) the term ∥u∥_{W^k_2(Q_j)} should be ∥u∥_{W^k_p(Q_j)}, and the display following 'Finally, the estimates (4.8) and (4.9) provide' repeats the Euclidean inequality from two displays earlier; the proof should be cleaned up.
  3. [Lemma 5.1 and Theorem 5.2] The notation S_Y(L^{(1)}Φ) is undefined and appears where S_X(LΦ) or the image space S_X(LΦ) is intended; the operator notation L^{(1)}Φ should be defined, and the Riesz-basis hypothesis on the family (LB_k) should be stated explicitly rather than by reference to (5.1), which is written for the original basis {B_k}.
  4. [References] Reference [31] appears to duplicate reference [20]; please consolidate the two entries and correct any bibliographic details, since both cite the same Gu-Eisenstat paper.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: stability and error estimates are derived from stated assumptions; apparent concerns reduce to correctness, not circularity.

full rationale

The derivation chain is a sequence of theorems with explicit hypotheses. The stability lower bound (5.3) follows from the norming-set inequality (4.1), the elliptic two-sided estimate (2.3), and the Riesz-basis stability ratio; the nontrivial content is the construction of such a norming set in Theorem 4.1, not a fitted parameter. The error estimates in Theorems 6.3 and 7.2 likewise chain from interpolation estimates, the stability bound, and previously published kernel bounds; no quantity is fitted to data and then renamed as a prediction. Several cited results ([30, Thm 6.1], [16], [34, Thm A.3]) are from the authors' own prior work, but they are external published theorems with stated, parameter-free assumptions that do not include the target result, so under the stated rules they count as independent evidence rather than circularity. One genuine concern is correctness, not circularity: the displayed asymptotic in (3.10), \hat\phi_s(\ell)=C\Gamma(\ell-s)/\Gamma(\ell+s+d), decays like \ell^{-(2s+d)}, so applying [34, Thm A.3] with 2\tau=2s+d to obtain Prop. 6.1's rate may be unsupported; that is a mathematical-error risk rather than a self-referential reduction. Remark 7.3 also openly flags the thinning cost that limits Theorem 7.2. Accordingly, no circular step is exhibited.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a chain of established results: Bernstein inequalities for SBF networks, Riesz basis properties of Lagrange bases, the Euclidean sampling inequality, and strong RRQR factorization. The only domain-specific assumption is the coercivity-style lower bound (2.2) on the elliptic operator L. No free parameters are fitted, and no new entities are introduced.

assumptions (5)
  • domain assumption The operator L satisfies the lower bound ∥Lf∥_{L2(M)} ≥ c_L ∥f∥_{L2(M)} (2.2), and the resulting Sobolev equivalences (2.3).
    Invoked in Proposition 2.1 and throughout (e.g., Lemma 5.1, Section 6) to relate L2 and H2 norms of the trial functions. Excludes operators with nontrivial nullspaces, e.g., the pure Laplace-Beltrami operator.
  • standard math The SBF network S_X(Ψ_β) satisfies the Bernstein inequality (3.5) and its strengthened version Proposition 3.1 from [30].
    Used in Section 4.3 to prove the Bernstein inequality (4.12) for S_X(Lφ_s), which is the key input for Theorem 4.1.
  • standard math The Lagrange basis {χ_ξ} for S_X(φ_s) is a Riesz basis satisfying (4.10), as shown in [16].
    Used to control the stability ratio r_2(X) and to pass between ∥a∥_{ℓ2} and ∥Σ a_k χ_k∥_{L2} in Sections 5 and 6.
  • standard math The strong rank-revealing QR factorization (7.3) exists with q_1(N,M)=√(1+N(M−N)) from [20].
    Basis of the thinning algorithm in Section 7.
  • standard math The Euclidean sampling inequality [29, Theorem 3.5] used in the proof of Lemma 4.2.
    The manifold sampling estimate is reduced to the Euclidean estimate via normal coordinates.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Extending Data to Improve Stability and Error Estimates Using Asymmetric Kansa-like Methods to Solve PDEs." pith.science (2026). https://pith.science/paper/KC4CEMZJ

@misc{pith2026250715137,
  author       = {Pith},
  title        = {Pith review of: Extending Data to Improve Stability and Error Estimates Using Asymmetric Kansa-like Methods to Solve PDEs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KC4CEMZJ}},
  note         = {Machine review of arXiv:2507.15137}
}
abstract

In this paper, a theoretical framework is presented for the use of a Kansa-like method to numerically solve elliptic partial differential equations on spheres and other manifolds. The theory addresses both the stability of the method and provides error estimates for two different approximation methods. A Kansa-like matrix is obtained by replacing the test point set $X$, used in the traditional Kansa method, by a larger set $Y$, which is a norming set for the underlying trial space. This gives rise to a rectangular matrix. In addition, if a basis of Lagrange (or local Lagrange) functions is used for the trial space, then it is shown that the stability of the matrix is comparable to the stability of the elliptic operator acting on the trial space. Finally, two different types of error estimates are given. Discrete least squares estimates of very high accuracy are obtained for solutions that are sufficiently smooth. The second method, giving similar error estimates, uses a rank revealing factorization to create a ``thinning algorithm'' that reduces $\#Y$ to $\#X$. In practice, this algorithm doesn't need $Y$ to be a norming set.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 41 canonical work pages

  1. [1]

    An extension of a bound for functions in sobolev spaces, with applications to (m, s)-spline interpolation and smoothing

    Mar ´ ıa Cruz L´ opez de Silanes Arcang´ eli, R´ emi and Juan Jos´ e Torrens. An extension of a bound for functions in sobolev spaces, with applications to (m, s)-spline interpolation and smoothing. Numerische Mathematik, 107(2):181–211, 2007

    work page 2007

  2. [2]

    A Generalized Randomized Rank-Revealing Factorization

    Grey Ballard, James Demmel, Ioana Dumitriu, and Alexander Rusciano. A Generalized Randomized Rank-Revealing Factorization. arXiv e-prints , page arXiv:1909.06524, September 2019

    work page Pith review arXiv 1909

  3. [3]

    Brad J. C. Baxter and Simon Hubbert. Radial basis functions for the sphere. In Recent progress in multivariate approximation (Witten-Bommerholz, 2000), volume 137 of Internat. Ser. Numer. Math. , pages 33–47. Birkh¨ auser, Basel, 2001

    work page 2000

  4. [4]

    On selecting a maximum volume sub-matrix of a matrix and related problems

    Ali C ¸ ivril and Malik Magdon-Ismail. On selecting a maximum volume sub-matrix of a matrix and related problems. Theoretical Computer Science, 410(47-49):4801–4811, 2009

    work page 2009

  5. [5]

    Extrinsic meshless collocation methods for PDEs on manifolds

    Meng Chen and Leevan Ling. Extrinsic meshless collocation methods for PDEs on manifolds. SIAM J. Numer. Anal., 58(2):988–1007, 2020

    work page 2020

  6. [6]

    Proving the stability estimates of variational least-squares kernel-based methods

    Meng Chen, Leevan Ling, and Dongfang Yun. Proving the stability estimates of variational least-squares kernel-based methods. Comput. Math. Appl. , 180:46–60, 2025

    work page 2025

  7. [7]

    A kernel-based embedding method and convergence analysis for surfaces PDEs

    Ka Chun Cheung and Leevan Ling. A kernel-based embedding method and convergence analysis for surfaces PDEs. SIAM J. Sci. Comput. , 40(1):A266–A287, 2018

    work page 2018

  8. [8]

    H 2-convergence of least-squares kernel collocation methods

    Ka Chun Cheung, Leevan Ling, and Robert Schaback. H 2-convergence of least-squares kernel collocation methods. SIAM J. Numer. Anal. , 56(1):614–633, 2018

    work page 2018

Show all 42 references
  1. [9]

    Kernel-Based Galerkin Methods on Compact Manifolds Without Boundary, with an Em- phasis on SO(3)

    Patrick Collins. Kernel-Based Galerkin Methods on Compact Manifolds Without Boundary, with an Em- phasis on SO(3) . Phd thesis, University of Hawai’i at Manoa, 2021

    2021

  2. [10]

    F. Dai, A. Prymak, A. Shadrin, V. N. Temlyakov, and S. Tikhonov. On the cardinality of lower sets and universal discretization. J. Complexity , 76:Paper No. 101726, 16, 2023

    2023

  3. [11]

    Error bounds for a least squares meshless finite difference method on closed manifolds

    Oleg Davydov. Error bounds for a least squares meshless finite difference method on closed manifolds. Adv. Comput. Math. , 49(4):Paper No. 48, 42, 2023

    2023

  4. [12]

    Duersch and Ming Gu

    Jed A. Duersch and Ming Gu. Randomized projection for rank-revealing matrix factorizations and low-rank approximations. SIAM Review, 62(3):661–682, 2020

    2020

  5. [13]

    W. Erb, T. Hangelbroek, F. J. Narcowich, C. Rieger, and J. D. Ward. Highly localized rbf lagrange functions for finite difference methods on spheres. BIT Numerical Mathematics , 64(2):16, 2024

    2024

  6. [14]

    Fasshauer

    Gregory E. Fasshauer. Hermite interpolation with radial basis functions on spheres. Adv. Comput. Math. , 10(1):81–96, 1999. 18

    1999

  7. [15]

    Fasshauer

    Gregory E. Fasshauer. Meshfree approximation methods with MATLAB , volume 6 of Interdisciplinary Mathematical Sciences. World Scientific Publishing Co. Pte. Ltd., Hackensack, NJ, 2007. With 1 CD-ROM (Windows, Macintosh and UNIX)

    2007

  8. [16]

    Fuselier, T

    E. Fuselier, T. Hangelbroek, F. J. Narcowich, J. D. Ward, and G. B. Wright. Localized bases for kernel spaces on the unit sphere. SIAM J. Numer. Anal. , 51(5):2538–2562, 2013

    2013

  9. [17]

    Gonzalez

    Teofilo F. Gonzalez. Clustering to minimize the maximum intercluster distance. Theoret. Comput. Sci. , 38(2-3):293–306, 1985

    1985

  10. [18]

    Gonzalez

    Teofilo F. Gonzalez. Covering a set of points in multidimensional space. Inform. Process. Lett., 40(4):181– 188, 1991

    1991

  11. [19]

    Bounding homotopy types by geometry

    Karsten Grove and Peter Petersen, V. Bounding homotopy types by geometry. Ann. of Math. (2) , 128(1):195–206, 1988

    1988

  12. [20]

    Eisenstat

    Ming Gu and Stanley C. Eisenstat. Efficient algorithms for computing a strong rank-revealing QR factor- ization. SIAM J. Sci. Comput. , 17(4):848–869, 1996

    1996

  13. [21]

    Kernel-based greedy approximation of parametric elliptic boundary value problems

    Bernard Haasdonk, Gabriele Santin, and Tizian Wenzel. Kernel-based greedy approximation of parametric elliptic boundary value problems. Preprint, arXiv:2507.06731, 2025

    2025 arXiv

  14. [22]

    Hangelbroek, F

    T. Hangelbroek, F. J. Narcowich, X. Sun, and J. D. Ward. Kernel approximation on manifolds II: the L∞ norm of the L2 projector. SIAM J. Math. Anal. , 43(2):662–684, 2011

    2011

  15. [23]

    Hangelbroek, F

    T. Hangelbroek, F. J. Narcowich, and J. D. Ward. Kernel approximation on manifolds I: Bounding the lebesgue constant. SIAM J. Math. Anal. , 42(4):1732–1760, 2010

    2010

  16. [24]

    Hangelbroek, C

    T. Hangelbroek, C. Rieger, and G. B. Wright. Spectral stability and perturbation results for kernel differ- entiation matrices on the sphere. Numer. Math. , 157(2):505–538, 2025

    2025

  17. [25]

    Narcowich, Christian Rieger, and Joseph D

    Thomas Hangelbroek, Francis J. Narcowich, Christian Rieger, and Joseph D. Ward. Direct and Inverse Re- sults on Bounded Domains for Meshless Methods via Localized Bases on Manifolds. In J. Dick, F. Kuo, and H. Wozniakowski, editors, Contemporary Computational Mathematics - a c...

    2018

  18. [26]

    Y. C. Hon and R. Schaback. On unsymmetric collocation by radial basis functions. J. Appl. Math. Comp. , 119:177–186, 2001

    2001

  19. [27]

    Limonova Kashin, B

    I. Limonova Kashin, B. E. Kosov and V. Temlyakov. Sampling discretization and related problems. Journal of Complexity , 71(101653):55, 2022

    2022

  20. [28]

    On sampling discretization in l2

    Irina Limonova and Vladimir Temlyakov. On sampling discretization in l2. Journal of Mathematical Analysis and Applications , 515(2):14, 2022

    2022

  21. [29]

    W. R. Madych. An estimate for multivariate interpolation. II. J. Approx. Theory , 142(2):116–128, 2006

    2006

  22. [30]

    H. N. Mhaskar, F. J. Narcowich, J. Prestin, and J. D. Ward. Lp Bernstein estimates and approximation by spherical basis functions. Math. Comp. , 79(271):1647–1679, 2010

    2010

  23. [31]

    Efficient algorithms for computing a strong rank-revealing QR factor- ization

    Gu Ming and Stanley C Eisenstat. Efficient algorithms for computing a strong rank-revealing QR factor- ization. SIAM Journal on Scientific Computing , 17(4):848–869, 1996. 19

    1996

  24. [32]

    Springer-Verlag, Berlin- New York, 1966

    Claus M¨ uller.Spherical harmonics, volume 17 of Lecture Notes in Mathematics . Springer-Verlag, Berlin- New York, 1966

    1966

  25. [33]

    Narcowich

    Francis J. Narcowich. Generalized Hermite interpolation and positive definite kernels on a Riemannian manifold. J. Math. Anal. Appl. , 190(1):165–193, 1995

    1995

  26. [34]

    Narcowich, Stephen T

    Francis J. Narcowich, Stephen T. Rowe, and Joseph D. Ward. A novel Galerkin method for solving PDES on the sphere using highly localized kernel bases. Math. Comp. , 86(303):197–231, 2017

    2017

  27. [35]

    Narcowich, Xinping Sun, and Joseph D

    Francis J. Narcowich, Xinping Sun, and Joseph D. Ward. Approximation power of RBFs and their associ- ated SBFs: a connection. Adv. Comput. Math. , 27(1):107–124, 2007

    2007

  28. [36]

    Stein and Guido Weiss

    Elias M. Stein and Guido Weiss. Introduction to Fourier analysis on Euclidean spaces , volume No. 32 of Princeton Mathematical Series . Princeton University Press, Princeton, NJ, 1971

    1971

  29. [37]

    Strichartz

    Robert S. Strichartz. Analysis of the Laplacian on the complete Riemannian manifold. J. Functional Analysis, 52(1):48–79, 1983

    1983

  30. [38]

    A least squares radial basis function finite difference method with improved stability properties

    Igor Tominec, Elisabeth Larsson, and Alfa Heryudono. A least squares radial basis function finite difference method with improved stability properties. SIAM J. Sci. Comput. , 43(2):A1441–A1471, 2021

    2021

  31. [39]

    Theory of function spaces

    Hans Triebel. Theory of function spaces. II , volume 84 of Monographs in Mathematics. Birkh¨ auser Verlag, Basel, 1992

    1992

  32. [40]

    C. F. Van Loan and N. Pitsianis. Approximation with Kronecker products. In Linear algebra for large scale and real-time applications (Leuven, 1992) , volume 232 of NATO Adv. Sci. Inst. Ser. E: Appl. Sci. , pages 293–314. Kluwer Acad. Publ., Dordrecht, 1993

    1992

  33. [41]

    Wendland

    H. Wendland. Scattered Data Approximation. Cambridge University Press, Cambridge, UK, 2005

    2005

  34. [42]

    Sharp inverse estimates for radial basis function interpolation: One-to-one correspondence between smoothness and approximation rates

    Tizian Wenzel. Sharp inverse estimates for radial basis function interpolation: One-to-one correspondence between smoothness and approximation rates. Math. Comp. , Published electronically Apr. 9, 2025. 20

    2025

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.