PaperlayerView exact revision on arXiv

Source-linked AI summary

Quasi-interpolation for the Helmholtz-Hodge decomposition

Nicholas Fisher, Gregory Fasshauer, Wenwu Gao

arXiv:2412.04600v1math.NAcs.NA

TL;DR

Interpolation-based approaches face numerical instability, and earlier divergence-free quasi-interpolants cannot directly reconstruct a general vector field from arbitrary data. This paper builds polyharmonic-spline matrix kernels and uses scaled convolution and its discretization to construct quasi-interpolants for Helmholtz–Hodge components. The paper reports convergence estimates and numerical simulations supporting vector-field reconstruction and decomposition.

  • Problem

    Earlier divergence-free quasi-interpolants assume divergence-free data and therefore cannot directly reconstruct a general vector field; interpolation-based methods also face numerical instability.

    Source: §1 · p. 2

  • Method

    The paper constructs a polyharmonic-spline matrix kernel with divergence-free, curl-free, and harmonic forms, then discretizes scaled convolutions to form quasi-interpolants.

    Source: §1 · p. 2

  • Results

    Scaled divergence-free or curl-free convolution sequences converge to the corresponding Helmholtz–Hodge components, and the paper reports error estimates and simulations for the quasi-interpolation method.

    Source: §1 · p. 2, §3.3.1 · p. 12, §5 · p. 22

  • Takeaways & Limitations

    The technique approximates a smooth vector field and its individual Helmholtz–Hodge components, with simultaneous error estimates for whole-space and bounded-domain settings.

    Source: §5 · p. 22

    The numerical experiments assume that continuous extensions of the kernel at the origin and possibly other points are handled appropriately.

    Source: §4 · p. 19

Abstract

from arXiv · show

The paper aims at proposing an efficient and stable quasi-interpolation based method for numerically computing the Helmholtz-Hodge decomposition of a vector field. To this end, we first explicitly construct a matrix kernel in a general form from polyharmonic splines such that it includes divergence-free/curl-free/harmonic matrix kernels as special cases. Then we apply the matrix kernel to vector decomposition via the convolution technique together with the Helmholtz-Hodge decomposition. More precisely, we show that if we convolve a vector field with a scaled divergence-free (curl-free) matrix kernel, then the resulting divergence-free (curl-free) convolution sequence converges to the corresponding divergence-free (curl-free) part of the Helmholtz-Hodge decomposition of the field. Finally, by discretizing the convolution sequence via certain quadrature rule, we construct a family of (divergence-free/curl-free) quasi-interpolants for the Helmholtz-Hodge decomposition (defined both in the whole space and over a bounded domain). Corresponding error estimates derived in the paper show that our quasi-interpolation based method yields convergent approximants to both the vector field and its Helmholtz-Hodge decomposition

1. Introduction

The paper addresses limitations of interpolation-based vector decomposition by constructing structure-preserving quasi-interpolants from a generalized matrix kernel. Scaled-kernel convolution targets the Helmholtz–Hodge components, with a boundary-specific construction for bounded domains.

1. Introduction

  • Earlier divergence-free quasi-interpolants assume divergence-free data, while interpolation-based methods can be numerically unstable.This limits their direct use for reconstructing a general vector field.Source: §1 · pp. 1–2
  • The generalized polyharmonic-spline matrix kernel includes divergence-free, curl-free, and harmonic kernels through parameter choices.The kernel’s spatial- and Fourier-domain properties are also studied.Source: §1 · p. 2
  • Convolution with a scaled divergence-free or curl-free kernel converges to the corresponding Helmholtz–Hodge component when its parameters are appropriately chosen.The field is written as a divergence-free part plus a curl-free part.Source: §1 · p. 2
  • Discretizing the convolution sequence yields quasi-interpolants that approximate the corresponding divergence-free or curl-free component of a general vector field.The paper uses a quadrature rule, with the rectangular rule as an example.Source: §1 · pp. 2–3
  • For bounded domains, the method combines interpolation near the boundary with quasi-interpolation in the interior.This adopts an approach proposed to address the boundary problem.Source: §1 · p. 3
  • The paper’s stated contributions are structure-preserving quasi-interpolants and a computationally efficient, numerically stable technique for Helmholtz–Hodge decomposition.The method is developed for vector-valued function approximation and decomposition.Source: §1 · p. 3

2. Preliminaries

The preliminaries establish the spline tools and error estimates used for approximation, then set out the whole-space and bounded-domain Helmholtz–Hodge decompositions. They provide the framework for the paper’s quasi-interpolation approach.

2.2. Polyharmonic splines satisfying higher-order Strang–Fix conditions

  • Higher-order polyharmonic splines satisfy Strang–Fix conditions of order 2k, unlike the basic spline’s order-two condition, supporting higher-order Schoenberg quasi-interpolation.The basic spline’s approximation order is two; Rabut’s construction can reach order 2ℓ.Source: §2.2 · pp. 4–5, §2.1 · p. 4
  • A level-kℓ spline is formed by truncating p_d,ℓ,k to order ℓ+k, and the construction includes the basic eΔ^ℓφℓ spline when k = 1.The fundamental solution φℓ satisfies Δℓφℓ = δ.Source: §2.1 · p. 3, §2.2 · p. 4

Then we have error estimates

  • The quasi-interpolant provides simultaneous approximation to the function and its derivatives, with ∥DαQh f − Dα f∥p,Rd = O(h^(2k−|α|)) when 0 ≤ |α| < min{2k, r−2k, 2ℓ−d−1}.Source: Then we have error estimates · p. 5

2.3. The Helmholtz–Hodge decomposition

  • The whole-space decomposition is a unique pair of divergence-free and curl-free projections that are orthogonal in L2(Rd); the bounded-domain version has additional boundary conditions.For the bounded-domain Leray projector, the stated special case is g = 0.Source: §2.3 · p. 6
  • For a bounded connected Lipschitz domain, the decomposition specifies the divergence-free component’s boundary normal value g, while the potential is unique up to a constant.When g = 0, this decomposition gives the Leray projector.Source: §2.3 · p. 6
  • The section introduces the projection framework that the paper uses to build quasi-interpolation methods for decompositions in both the whole space and bounded domains.The whole-space projections yield a unique orthogonal decomposition in L2(Rd).Source: §2.3 · p. 6

3. The Main Results

The main results construct matrix kernels from polyharmonic splines, including divergence-free, curl-free, and harmonic cases, and use their convolution with a general vector field to approximate its Helmholtz–Hodge components. The paper also gives simultaneous approximation estimates for the kernels and their projected components.

3. The Main Results

  • The method’s stated scope includes quasi-interpolation for vector decomposition in both the whole space and bounded domains.The main-results section presents the kernel construction and its application to Helmholtz–Hodge decomposition.Source: §3 · p. 6

3.1. Divergence-free, curl-free, and harmonic matrix kernels

  • A general matrix-kernel construction from polyharmonic splines encompasses divergence-free, curl-free, and harmonic kernels through parameter choices.The construction is illustrated with polyharmonic splines, while the paper notes that its basic idea also applies to other functions.Source: §3.1 · pp. 7–8
  • For vector fields matching a kernel’s divergence-free, curl-free, or harmonic property, the corresponding scaled kernel gives simultaneous approximation estimates for the field and its derivatives.The estimates hold for 0 ≤ |α| ≤ min{2k, r−2k}.Source: §3.1 · p. 8

3.2. Kernel projections

  • Convolution with the divergence-free kernel returns the divergence-free component, while convolution with the curl-free kernel returns the curl-free component.The paper derives simultaneous error estimates for these projected components and their derivatives.Source: §3.2 · pp. 9–10

2. For the curl-free projection

The section develops quasi-interpolation schemes by discretizing convolution approximations, then establishes that these schemes approximate vector fields and their Helmholtz–Hodge components in the whole space.

3.3. Construction of quasi-interpolation schemes

  • Scaled divergence-free or curl-free matrix-kernel convolution approximates the corresponding Helmholtz–Hodge component, and discretizing these sequences yields quasi-interpolation schemes.The construction also covers whole-space approximation and extends to bounded domains.Source: §3.3 · p. 10, §3.3.1 · p. 12

3.3.1. Quasi-interpolation defined in the whole space Rd

  • For whole-space data sampled at equidistant centers, the convolution sequence is discretized using a rectangular rule to construct the quasi-interpolant.The resulting scheme is expressed in Schoenberg form.Source: §3.3.1 · p. 11
  • The whole-space simultaneous error estimates give an approximation order of O(h^(2k−|α|)) for derivatives of order |α|.The stated range is 0 ≤ |α| < min{2k, r−2k, 2ℓ−d−1}.Source: §3.3.1 · pp. 12–13
  • The whole-space quasi-interpolant approximates the field and its derivatives, or the corresponding divergence-free or curl-free component and its derivatives.The matrix kernel is chosen to match the property of the target field or component.Source: §3.3.1 · pp. 11–12

2. For the curl-free projection

For bounded domains, the method combines interpolation near the boundary with quasi-interpolation in the interior, then derives error bounds for the resulting Helmholtz–Hodge approximation.

3.3.2. Quasi-interpolation defined over a bounded domain

  • Direct quasi-interpolation on bounded-domain samples can approximate poorly at the boundary, so the method combines interpolation near the boundary with quasi-interpolation in the interior.The approach follows the boundary treatment proposed in.Source: §3.3.2 · p. 14
  • The construction interpolates sampled data outside an interior region, subtracts that interpolant, and applies divergence-free quasi-interpolation to the residual and the interpolated field’s divergence-free part.The interpolant is chosen so the residual components satisfy the stated smoothness condition.Source: §3.3.2 · pp. 14–15
  • The bounded-domain error bounds rely on an interpolant with approximation order at least O(h^n) in the region outside the interior domain and on the conditions specified for the domains and functions.The theorem’s error range also depends on the smoothness and kernel parameters.Source: §3.3.2 · pp. 16, 18
  • An optimal choice of the scale parameter H is identified for the bounded-domain approximation error estimate.The supplied passage does not state the scale formula or error value.Source: §3.3.2 · p. 18
  • The bounded-domain analysis also derives corresponding results for the curl-free component, and the section concludes with convergence analysis for whole-space and bounded-domain schemes.The stated curl-free construction uses counterparts of the interpolant and quasi-interpolant.Source: §3.3.2 · p. 18

4. Numerical Simulations

Numerical experiments report the expected asymptotic convergence rates on the whole space and bounded domains, and show approximations of the field and its Helmholtz–Hodge components.

4.1. Approximations on Rd

  • O(h^4) for α = 0 and O(h^3) for α = 1 are obtained asymptotically in the whole-space experiments.The slopes are computed from the last 10 observed errors.Source: §4.1 · p. 19, Figure 1 · p. 20

4.2. Approximation on bounded domains

  • O(h^(2/3)) for k = ℓ = 2 and O(h^(3/4)) for k = ℓ = 3 are obtained asymptotically on the bounded domain.The rates are estimated using the last five available errors.Source: §4.2 · p. 20, Figure 2 · p. 21
  • The bounded-domain quasi-interpolant provides approximations to the full field and its divergence-free and curl-free shares.Figure 3 presents observed data, the computed Leray projection, the computed curl-free component, and their sum.Source: §4.2 · p. 20, Figure 3 · p. 21

5. Conclusions and Discussions

The paper presents a quasi-interpolation technique for smooth vector fields whose reconstructed components approximate the corresponding Helmholtz–Hodge parts, with simultaneous error estimates in whole and bounded domains.

5. Conclusions and Discussions

  • The technique approximates smooth vector fields and their individual Helmholtz–Hodge components, with simultaneous error estimates in the whole space and over bounded domains.The reconstruction separates divergence-free and curl-free parts.Source: §5 · p. 22
  • Numerical simulations are reported to show computational efficiency and numerical stability for vector-field reconstruction and Helmholtz–Hodge decomposition.Source: §5 · p. 22
  • Future work will focus on extending the technique to vector-field approximation on manifolds.Source: §5 · p. 22
Loading 2412.04600v1…