Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

An adaptive delaminating Levin method in two dimensions

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The Levin PDE has a slowly varying approximate solution at every frequency, so bivariate oscillatory integrals can be evaluated rapidly even when stationary and resonance points are present.

desk verdict Rigorous 2D existence theory for the Levin PDE, but the implemented algorithm's k×k grid collocation leaves a real, acknowledged gap between the theorem and the code. read the letter →

arxiv 2506.02424 v1 pith:PCK5VCW6 submitted 2025-06-03 math.NA cs.NA

Shukui Chen , Kirill Serkh , James Bremer , Murdock Aubry This is my paper
classification math.NAcs.NA MSC 65D3265N3565D30
keywords oscillatoryintegralsLevinmethodChebyshevspectralcollocationadaptivesubdivisiondelaminatingstationarypointsresonancemultivariateintegration
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

This paper removes the classical non-resonance barrier for two-dimensional Levin methods. It proves that the Levin PDE $p_x + i g_x p = f$ admits a slowly varying approximate solution both when $g_x$ never vanishes and when $|g_x|$ is small enough that stationary points are allowed, so the approximation cost need not grow with frequency. On that foundation it derives error estimates for solving the PDE by Chebyshev spectral collocation with a truncated singular value decomposition, and it designs an adaptive \"delaminating\" Levin method that solves one univariate problem per fiber and evaluates boundary integrals with a univariate adaptive Levin method, which removes resonance points from consideration. Numerical experiments show accurate, frequency-independent timings on integrals with elementary, Bessel, stationary-point, and resonance-point phases. If correct, this makes multivariate Levin methods practical in exactly the cases previously regarded as impossible.

What carries the argument

The load-bearing objects are the Levin PDE $p_x + i g_x p = f$, the change of variables $u(x,y) = (\frac{1}{W}\int_{-1}^{x} g_x(t,y)\,dt, y)$ that flattens $g_x$ to a constant $W$, and bandlimited approximations built from Fourier-supported tempered distributions that yield slowly varying solutions with controlled bandlimits. The numerical machinery is Chebyshev spectral collocation on tensor-product grids, truncated singular value decompositions for the ill-conditioned collocation systems, and delamination: writing the $k\times k$ collocation system as $k$ independent $k\times k$ problems, one per $y$-fiber, followed by a univariate adaptive Levin method for the boundary integrals that eliminates resonance points.

What would settle it

Take a rectangle with $g_x$ of moderate size, for instance varying between $1/2$ and $2$, choose a smooth $f$ that forces the slowly varying solution to have degree near $k-1$, and compare the delaminating method enforced on the $k\times k$ grid with the method enforced on the $(2k-1)\times(2k-1)$ grid at the same $k$. If the two estimates disagree beyond the truncation tolerance, or if the adaptive subdivision count grows much faster than the analysis predicts, the Section 5 heuristic fails and the implemented algorithm is not covered by the theorem.

Watch

Extended reading notes

Core claim

The central claim is that the first-order Levin PDE $p_x + i g_x p = f$ on $[-1,1]^2$ has an approximate solution $p$ that is slowly varying, with bandlimit and smoothness controlled independently of the magnitude of $g_x$, in two complementary regimes: when $g_x$ is nonvanishing and its ratio of maximum to minimum magnitude is moderate, and when $g_x$ is of small magnitude, a regime that includes stationary points. From these existence results the paper derives a frequency-independent error estimate for the Chebyshev spectral collocation discretization and then introduces an adaptive algorithm that splits the domain, solves the PDE along $x$-fibers independently via truncated SVD, and evaluates the resulting boundary integrals with the univariate adaptive Levin method. The conclusion is that stationary and resonance points of the phase do not prevent rapid and accurate evaluation of bivariate oscillatory integrals over rectangular domains.

Load-bearing premise

The implemented method rests on the unproven claim that after adaptive subdivision, on each rectangle both the PDE solution $p$ and the product $g_x p$ are polynomials of degree below the collocation order $k$, so the $k\times k$ grid used in practice inherits the error analysis proved for the $(2k-1)\times(2k-1)$ grid.

Editorial extensions

If this is right

  • The classical non-resonance condition is unnecessary: stationary points inside the domain no longer need to be located or excluded before applying a Levin method.
  • For rectangular domains, the cost of evaluating a bivariate oscillatory integral can be bounded independently of frequency $\lambda$, with only a mild logarithmic subdivision count in the moderate-frequency regime.
  • Resonance points at subrectangle boundaries are handled by the univariate adaptive Levin method rather than by extra two-dimensional subdivision, so the rectangle count stays low.
  • The same tensor-product analysis carries to dimensions three and higher, making volumetric Levin methods feasible without a non-resonance assumption.
  • Phase functions for special functions, such as the Bessel-phase construction used for Hankel-function integrands, can be plugged directly into the method.

Reading between the lines

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

  • The paper's Section 5 heuristic, that after subdivision the product $g_x p$ has degree below $k$ on each rectangle so the $k\times k$ grid inherits the $(2k-1)\times(2k-1)$ error analysis, is unproven; if it fails in the moderate-frequency regime, the implemented algorithm is not covered by the stated theorem.
  • The choice of delamination direction is presented as arbitrary, but the error bound involves $G_1/G_0$; an adaptive choice comparing both partial derivatives, when numerically well-conditioned, could reduce subdivisions for anisotropic phases.
  • The proof strategy appears to extend naturally to domains meshed with smooth curved quadrilateral elements and to dimensions three and higher, although the paper only sketches these directions.
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

3 major / 5 minor

Summary. The paper studies the numerical evaluation of bivariate oscillatory integrals over rectangles by Levin-type methods. It proves two existence theorems for slowly-varying approximate solutions of the Levin PDE p_x + i g_x p = f: Theorem 3.3 covers the case where g_x is nonvanishing with small relative variation, and Theorem 3.4 covers the case where |g_x| is uniformly small, thereby including stationary points. Section 4 derives error estimates for a Chebyshev spectral collocation discretization on a (2k-1)x(2k-1) tensor product grid with truncated SVD solution. Section 5 presents an adaptive delaminating Levin method that uses a kxk collocation grid and evaluates the boundary integrals by a univariate adaptive Levin method. Numerical experiments in Sections 6.1-6.5 demonstrate high accuracy and speed across a wide frequency range, including examples with stationary and resonance points.

Significance. The existence results of Section 3, together with the Section 4 error estimates, would remove the classical non-resonance barrier for two-dimensional Levin methods on rectangular domains, which is a substantial advance. The proofs are technically careful, using bandlimited approximations and distribution-theoretic arguments, and the numerical experiments are broad and convincing as empirical evidence. However, the implemented algorithm of Section 5 is not covered by the Section 4 analysis: the theory applies to a (2k-1)x(2k-1) collocation grid, while the code uses a kxk grid, and the boundary integrals are not computed exactly as assumed in the analysis. The theoretical core is a meaningful contribution even if the algorithmic gap is not fully closed, but the paper's strongest computational claim currently rests on an unproven heuristic.

major comments (3)
  1. [Section 5, Step 2 (Eqs. (173)-(177)) vs. Section 4 (Eqs. (148)-(149))] The implemented delaminating algorithm enforces the Levin PDE only on the kxk tensor product Chebyshev grid, whereas the error analysis of Section 4 requires the (2k-1)x(2k-1) grid to establish the polynomial identity (148) and the residual bound (149). On the kxk grid, a residual polynomial of degree up to 2k-2 can vanish at all collocation nodes while being large elsewhere, so the computed bp is not controlled by the stated theorem. The paragraph beginning "When the delaminating Levin method is used adaptively" offers only a heuristic justification, asserting that adaptive subdivision over-discretizes f and g_x so that g_x bp has degree below k; no proof is given that accepted subrectangles satisfy this degree condition, nor that the moderate-frequency regime requires only a bounded number of subdivisions. This gap concerns the central claim that the adaptive delaminating Levin method delivers guaranteed high accuracy, and it must be addressed either by modifying the algorithm to match the analysis or by supplying a rigorous proof of the degree heuristic.
  2. [Section 4, Eq. (151) and Section 5, Step 3] The integral error estimate (153) assumes that the univariate integrals in (151) are computed exactly. In the algorithm, these integrals are evaluated by the univariate adaptive Levin method with a tolerance epsilon_1D = beta max{||bp||_inf/||f||_inf, 1} epsilon, and no error bound for this univariate computation is incorporated into the analysis. The choice of epsilon_1D is justified only by a heuristic convergence argument. Consequently, the total error of the implemented adaptive delaminating method is not bounded by the theorems of Section 4.
  3. [Theorem 3.4 (Section 3, inequalities (109)-(111))] The proof of Theorem 3.4 states that the key inequalities (109)-(111) follow "by the same derivation presented in the proof of Theorem 5 in [9]" without providing the derivation. Since these inequalities are essential to the small-g_x existence result, the paper should either reproduce the argument in the bivariate setting with y as a parameter or give a precise statement that the proof from [9] applies verbatim. In particular, the bandlimit propagation for the integral operator A_delta in equations (112)-(114) requires careful verification, as A_delta involves an indefinite integral and the composition with (g_x)_b,delta.
minor comments (5)
  1. [Section 5, Step 4] The acceptance condition "val0 - (val1 + val2 + val3 + val4) < epsilon" compares a complex number with a real tolerance; it should presumably be |val0 - (val1 + val2 + val3 + val4)| < epsilon, or a relative-error variant.
  2. [Section 4.2] The phrase "bandlimied function" is a typo for "bandlimited function."
  3. [Section 3, Lemma 3.1] The reference "Lemma (2.3)" should be "Lemma 2.3."
  4. [Section 6.2] The text states that the reported times account for both the evaluation of the phase functions and the execution of the method, but immediately adds that the time taken to construct the phase function is not included; please clarify whether "evaluation" refers to the per-integral use of already-constructed phase functions, with construction as a separate one-time cost.
  5. [Section 2.1] The notation for C^k(U) and the subspace of functions whose derivatives extend continuously to the closure appears garbled; the overline on the closure is missing in places, making the definition ambiguous.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the PDE existence and collocation error analysis are self-contained; the k-by-k grid gap is a coverage issue, not a circular reduction.

full rationale

The paper's central results are derived from an explicit Fourier-domain construction, not from a fitted parameter or from an equation that equals its own input. In Lemma 3.1, the approximate solution is defined by p_b,delta = inverse Fourier transform of f_hat/(i(xi_1 + W)), and the residual bound follows from the bandlimited approximation error of f. The constants c_f(epsilon), c_gx(epsilon), and c_h(epsilon) are infima of bandlimits of epsilon-bandlimited approximations, shown to be finite by Theorem 2.7; they characterize approximation complexity and are not fitted to the target integrals. Section 4 derives error estimates from Chebyshev interpolation inequalities, Lemma 2.1, and the truncated SVD perturbation theorem (Theorem 2.10), with the collocation degree k chosen from approximation properties of p_b, g_x, and f, independently of the magnitude of g_x. Those estimates are not obtained by assuming the desired accuracy. The reliance on the univariate adaptive Levin method from [9] for the boundary integrals is an external, overlapping-author dependency: [9] is a published peer-reviewed result with its own proof, and the present two-dimensional analysis does not reduce the bivariate problem to that theorem by definition. Section 5 enforces the PDE on the k-by-k grid, whereas Section 4 analyzes the (2k-1)-by-(2k-1) grid; the paper explicitly labels the justification as a heuristic expectation about over-discretization under adaptive subdivision. This is an implementation-coverage gap and a potential correctness risk, but it is not a case where a prediction is defined in terms of its input or where a fitted value is renamed as a prediction. The numerical experiments are validated against explicit formulas and an independent adaptive Gauss-Legendre reference, so the claims are not supported solely by self-citation. No circular step meeting the quoted-reduction standard is present.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The mathematical sections rely on standard distribution theory and smoothness assumptions. The load-bearing heuristic content appears in the two ad hoc assumptions about adaptive subdivision and polynomial degree, which connect the proven estimates to the implemented algorithm. No new physical or mathematical entities such as particles, forces, or conserved quantities are introduced.

free parameters (4)
  • collocation order k = 7
    Algorithm parameter used in all numerical experiments. Not fitted to target integrals; chosen to balance cost and accuracy.
  • subdivision tolerance epsilon = 1e-12
    User-specified accuracy target for the adaptive scheme. Not fitted to data.
  • truncation tolerance scale beta0 = 1/2
    Sets the truncation tolerance epsilon0 = beta0 * epsilon / ||f||_infty. Hand-chosen.
  • safety factor beta = 0.1
    Scales the univariate subdivision tolerance epsilon_1D. Hand-chosen.
assumptions (5)
  • domain assumption f and g admit infinitely differentiable extensions to an open neighborhood of [-1,1]^2.
    Invoked throughout Sections 3 and 4 to justify bandlimited approximation, change of variables, and Chebyshev convergence.
  • standard math Standard tempered distribution theory, including Fourier transform, Paley-Wiener, and Radon measure representation of compactly supported distributions.
    Used in Lemma 3.1, Corollary 3.2, and Theorem 3.4 to construct approximate solutions with controlled Fourier support.
  • standard math The univariate adaptive Levin theorem of Chen, Serkh, and Bremer [9] is correct.
    Used for boundary integral evaluation and for the deferred inequalities (109)-(111) in Theorem 3.4. The result is peer-reviewed and published, though authored by overlapping authors.
  • ad hoc to paper Adaptive subdivision eventually places every subrectangle in one of the two analyzed regimes: g_x nonvanishing with small G1/G0, or G1 <= 1/4.
    Section 5 assumes this rather than proving termination or complexity for arbitrary smooth phases.
  • ad hoc to paper On adaptively produced subrectangles, bp and g_x bp are representable by polynomials of degree below k.
    This heuristic in Section 5 bridges the analyzed (2k-1) grid to the implemented k by k grid. No proof is supplied.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An adaptive delaminating Levin method in two dimensions." pith.science (2026). https://pith.science/paper/PCK5VCW6

@misc{pith2026250602424,
  author       = {Pith},
  title        = {Pith review of: An adaptive delaminating Levin method in two dimensions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PCK5VCW6}},
  note         = {Machine review of arXiv:2506.02424}
}
read the original abstract

We present an adaptive delaminating Levin method for evaluating bivariate oscillatory integrals over rectangular domains. Whereas previous analyses of Levin methods impose non-resonance conditions that exclude stationary and resonance points, we rigorously establish the existence of a slowly-varying, approximate solution to the Levin PDE across all frequency regimes, even when the non-resonance condition is violated. This allows us to derive error estimates for the numerical solution of the Levin PDE via the Chebyshev spectral collocation method, and for the evaluation of the corresponding oscillatory integrals, showing that high accuracy can be achieved regardless of whether or not stationary and resonance points are present. We then present a Levin method incorporating adaptive subdivision in both two and one dimensions, as well as delaminating Chebyshev spectral collocation, which is effective in both the presence and absence of stationary and resonance points. We demonstrate the effectiveness of our algorithm with a number of numerical experiments.

Figures

Figures reproduced from arXiv: 2506.02424 by the authors.

Figure 1
Figure 1. The results of the experiments of Section 6.1. The first row of plots pertains to the integral I1(λ), and the second pertains to I2(λ). In each row, the plot on the left gives the absolute error in the calculation of the integral as a function of λ, and the plot on the right shows the running time in milliseconds as a function of λ. 6.1 Integrals involving elementary functions (with explicit formulas) In the experim… view at source ↗
Figure 2
Figure 2. The results of the experiments of Section 6.2. The first row of plots pertains to the integral I3(λ), and the second pertains to I4(λ). In each row, the plot on the left gives the absolute error in the obtained value of the integral as a function of λ, and the plot on the right shows the running time in milliseconds as a function of λ. 6.2 Integrals involving the Bessel functions (with explicit formulas) In the expe… view at source ↗
Figure 3
Figure 3. The results of the experiments of Section 6.3. The first row of plots shows the absolute error in the calculated value of I5(λ, n) as a function of λ for the values of n considered. The second row of plots shows the running time of the adaptive delaminating Levin method in milliseconds as a function of λ for n = 2, 3, 4, 5, 6, 7, 8, and 9. h can be represented by a Chebyshev expansion of small fixed order. Thus, we … view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The results of the experiments of Section 6.4. The first row of plots shows the absolute error in the calculated value of the integral I6(λ) and the running time of the adaptive delaminating Levin method in milliseconds as a function of λ. The second row of plots compa…
Figure 5
Figure 5. Figure 5: The results of the experiments of Section 6.5. The plots in the first row give the number of rectangles in the adaptively determined discretization of [0, 1]2 used to compute the integral I7(λ, m) as a function of λ for the values of m considered. The plots in the seco…
Figure 6
Figure 6. Figure 6: The plots show the adaptive discretization of the domain produced by the adaptive delaminating Levin method for computing the integral I5(300, n) with n = 2, 3, 4, and 5. The subrectangles are colored according to the ratio of the maximum to minimum value of the partia…
Figure 7
Figure 7. Figure 7: The plots in the column on the left and right show the adaptive discretization of the domain produced by the adaptive method that solves the rectangular systems without delamination, and by the adaptive delaminating Levin method, respectively, for computing the integra…

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Gravitational radiation from Kerr black holes using the Sasaki-Nakamura formalism: Waveforms and fluxes at infinity

    gr-qc 2025-11 conditional novelty 7.0 of 10

    A new integration-by-parts scheme computes Sasaki-Nakamura waveforms directly from the Teukolsky source term, bypassing the standard extra radial integration for bound orbits.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages · cited by 1 Pith paper

  1. [9]

    On the adaptive Levin method

    Shukui Chen, Kirill Serkh, and James Bremer. On the adaptive Levin method. Numer. Math., 156(6): 1927–1985, 2024

    work page 1927

  2. [1]

    SIAM, 2018

    Alfredo Dea˜ no, Daan Huybrechs, and Arieh Iserles.Computing Highly Oscillatory Integrals. SIAM, 2018

    work page 2018

  3. [2]

    Procedures for computing one- and two-dimensional integrals of functions with rapid irregular oscillations.Math

    David Levin. Procedures for computing one- and two-dimensional integrals of functions with rapid irregular oscillations.Math. Comp., 38(158): 531–538, 1982

    work page 1982

  4. [3]

    On the Quadrature of Multivariate Highly Oscillatory Integrals Over Non-polytope Domains.Numer

    Sheehan Olver. On the Quadrature of Multivariate Highly Oscillatory Integrals Over Non-polytope Domains.Numer. Math., 103(4): 643–665, 2006

    work page 2006

  5. [4]

    Fast, numerically stable computation of oscillatory integrals with stationary points.BIT Numer

    Sheehan Olver. Fast, numerically stable computation of oscillatory integrals with stationary points.BIT Numer. Math., 50(1): 149–171, 2010

    work page 2010

  6. [5]

    Fast and stable augmented Levin methods for highly oscillatory and singular integrals.Math

    Yinkun Wang and Shuhuang Xiang. Fast and stable augmented Levin methods for highly oscillatory and singular integrals.Math. Comp., 91(336): 1893–1923, 2022

    work page 1923

  7. [6]

    A universal solution to one-dimensional oscillatory integrals.Sci

    JianBing Li, XueSong Wang, and Tao Wang. A universal solution to one-dimensional oscillatory integrals.Sci. China Ser. F-Inf. Sci., 51(10): 1614–1622, 2008

    work page 2008

  8. [7]

    An improved Levin quadrature method for highly oscillatory integrals.Appl

    Jianbing Li, Xuesong Wang, Tao Wang, and Shunping Xiao. An improved Levin quadrature method for highly oscillatory integrals.Appl. Numer. Math., 60(8): 833–842, 2010

    work page 2010

Show all 26 references
  1. [8]

    PhD thesis

    Andrew James Moylan.Highly Oscillatory Integration, Numerical Wave Optics, and the Gravitational Lensing of Gravitational Waves. PhD thesis. The Australian National University, 2008

    2008

  2. [10]

    Cauchy data for Levin’s method.IMA J

    Anthony Ashton. Cauchy data for Levin’s method.IMA J. Numer. Anal., 45(1): 87–125, 2025

    2025

  3. [11]

    Delaminating quadrature method for multi-dimensional highly oscillatory integrals.Appl

    Jianbing Li, Xuesong Wang, Tao Wang, and Chun Shen. Delaminating quadrature method for multi-dimensional highly oscillatory integrals.Appl. Math. Comput., 209(2): 327–338, 2009

    2009

  4. [12]

    On the numerical solution of second order ordinary differential equations in the high-frequency regime.Appl

    James Bremer. On the numerical solution of second order ordinary differential equations in the high-frequency regime.Appl. Comput. Harmon. Anal., 44(2): 312–349, 2018

    2018

  5. [13]

    Phase function methods for second order linear ordinary differential equations with turning points.Appl

    James Bremer. Phase function methods for second order linear ordinary differential equations with turning points.Appl. Comput. Harmon. Anal., 65: 137–169, 2023

    2023

  6. [14]

    A solver for linear scalar ordinary differential equations whose running time is bounded independent of frequency.arXiv preprint, arXiv:2311.08578, 2023

    Murdock Aubry and James Bremer. A solver for linear scalar ordinary differential equations whose running time is bounded independent of frequency.arXiv preprint, arXiv:2311.08578, 2023

    2023 arXiv

  7. [15]

    Properties of Some Weighted Sobolev Spaces and Application to Spectral Approximations.SIAM J

    Christine Bernardi and Yvon Maday. Properties of Some Weighted Sobolev Spaces and Application to Spectral Approximations.SIAM J. Numer. Anal., 26(4): 769–829, 1989. 43

    1989

  8. [16]

    Folland.Real Analysis: Modern Techniques and Their Applications

    Gerald B. Folland.Real Analysis: Modern Techniques and Their Applications. Wiley, 1999

    1999

  9. [17]

    Ian Richards and Heekyung Youn.Theory of Distributions: A Non-Technical Introduction

    J. Ian Richards and Heekyung Youn.Theory of Distributions: A Non-Technical Introduction. Cambridge University Press, 1995

    1995

  10. [18]

    Stein.Singular Integrals and Differentiability Properties of Functions

    Elias M. Stein.Singular Integrals and Differentiability Properties of Functions. Princeton Mathematical Series Vol. 30. Princeton University Press, 2016

    2016

  11. [19]

    John P. Boyd. Approximation of an analytic function on a finite real interval by a bandlimited function and conjectures on properties of prolate spheroidal functions. Appl. Comput. Harmon. Anal., 15(2): 168–176, 2003

    2003

  12. [20]

    Frank W. J. Olver, Daniel W. Lozier, Ronald F. Boisvert, and Charles W. Clark, editors.NIST Handbook of Mathematical Functions. Cambridge University Press, 2010

    2010

  13. [21]

    On the approximation of singular functions by series of noninteger powers.IMA J

    Mohan Zhao and Kirill Serkh. On the approximation of singular functions by series of noninteger powers.IMA J. Numer. Anal., 2025

    2025

  14. [22]

    Dover Publications, 2006

    Fran¸ cois Tr` eves.Topological Vector Spaces, Distributions and Kernels. Dover Publications, 2006

    2006

  15. [23]

    AMS, 2010

    Victor Guillemin and Alan Pollack.Differential Topology. AMS, 2010

    2010

  16. [24]

    Mason and David C

    John C. Mason and David C. Handscomb.Chebyshev Polynomials. CRC, 2003

    2003

  17. [25]

    Note on the homotopy perturbation method for multivariate vector-value oscillatory integrals.Appl

    Ruyun Chen and Shuhuang Xiang. Note on the homotopy perturbation method for multivariate vector-value oscillatory integrals.Appl. Math. Comput., 215(1): 78–84, 2009

    2009

  18. [26]

    The Construction of cubature rules for multivariate highly oscillatory integrals.Math

    Daan Huybrechs and Stefan Vandewalle. The Construction of cubature rules for multivariate highly oscillatory integrals.Math. Comp., 76(260): 1955–1981, 2007. 44 1.0 0.5 0.0 0.5 1.0 Number of rectangles: 712 1.0 0.5 0.0 0.5 1.0 Number of rectangles: 1192 1.0 0.5 0.0 0.5 1.0 Num...

    1955

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.