Table of Contents
Fetching ...

The geometric error is less than the pollution error when solving the high-frequency Helmholtz equation with high-order FEM on curved domains

Théophile Chaumont-Frelet, Euan A. Spence

Abstract

We consider the $h$-version of the finite-element method, where accuracy is increased by decreasing the meshwidth $h$ while keeping the polynomial degree $p$ constant, applied to the Helmholtz equation. Although the question "how quickly must $h$ decrease as the wavenumber $k$ increases to maintain accuracy?" has been studied intensively since the 1990s, none of the existing rigorous wavenumber-explicit analyses take into account the approximation of the geometry. In this paper we prove that for nontrapping problems solved using straight elements the geometric error is order $kh$, which is then less than the pollution error $k(kh)^{2p}$ when $k$ is large; this fact is then illustrated in numerical experiments. More generally, we prove that, even for problems with strong trapping, using degree four (in 2-d) or degree five (in 3-d) polynomials and isoparametric elements ensures that the geometric error is smaller than the pollution error for most large wavenumbers.

The geometric error is less than the pollution error when solving the high-frequency Helmholtz equation with high-order FEM on curved domains

Abstract

We consider the -version of the finite-element method, where accuracy is increased by decreasing the meshwidth while keeping the polynomial degree constant, applied to the Helmholtz equation. Although the question "how quickly must decrease as the wavenumber increases to maintain accuracy?" has been studied intensively since the 1990s, none of the existing rigorous wavenumber-explicit analyses take into account the approximation of the geometry. In this paper we prove that for nontrapping problems solved using straight elements the geometric error is order , which is then less than the pollution error when is large; this fact is then illustrated in numerical experiments. More generally, we prove that, even for problems with strong trapping, using degree four (in 2-d) or degree five (in 3-d) polynomials and isoparametric elements ensures that the geometric error is smaller than the pollution error for most large wavenumbers.
Paper Structure (27 sections, 12 theorems, 194 equations, 4 figures)

This paper contains 27 sections, 12 theorems, 194 equations, 4 figures.

Key Result

theorem 1

Under the assumptions in §sec_assumptions, let $u\in \mathcal{H}$ be the unique solution of the variational problem Let and If $B>0$, then there exists a unique $u_h \in V_h$ such that and the error $u-u_h$ satisfies the bound

Figures (4)

  • Figure 2.1: Numerical example: sound-soft scattering with $p=2$
  • Figure 2.5: Numerical example: penetrable obstacle with $p=2$
  • Figure 4.1: Example of a curved domain $\Omega$ with two subdomains $Q_\pm$. The mesh $\mathcal{T}_h$ consisting of straight elements induces an approximate domain $\Omega^h$ partitioned into $Q_\pm^h$. Here, for each $K \in \mathcal{T}_h$, $\mathcal{F}_K: \widehat{K} \to K$ is the affine map mapping the reference element into $K$. The mapping $\Psi^h$ maps $\Omega^h$ onto $\Omega$, and, in particular, maps each element $K_j$ onto $\widetilde{K}_j$. In this particular example, $\Psi^h|_{K_j}$ is the identity map for $j=2,4,6$ and $8$, and is a non-affine mapping on the remaining elements. $\Phi^h$ correspondingly maps $\widetilde{K}_j$ onto $K_j$.
  • Figure 4.2: Example illustrating Assumption \ref{['assumption_straight_fem']}. The mesh $\mathcal{T}_h$ is depicted on the left. A mapping $\Psi^h$ satisfying Assumption \ref{['assumption_straight_fem']} is represented in the middle. A mapping $\Psi^h$ not satisfying Assumption \ref{['assumption_straight_fem']} is represented on the right.

Theorems & Definitions (34)

  • definition 1: Sound-soft scattering by a 2-d ball
  • definition 2: Scattering by a penetrable 2-d ball
  • definition 3: PML approximation to sound-soft scattering by a 2-d ball
  • definition 4: PML approximation to scattering by a penetrable 2-d ball
  • remark 1: The relationship between \ref{['eq_Garding']} and the standard Gå rding inequality
  • theorem 1: Abstract elliptic-projection-type argument with variational crime
  • lemma 1
  • lemma 2
  • proof : Proof of Theorem \ref{['thm_abs1']} using Lemmas \ref{['lem:abs1']} and \ref{['lem:abs2']}
  • proof : Proof of Lemma \ref{['lem:abs2']}
  • ...and 24 more