A Multi-Frequency Helmholtz Solver Based on the WaveHoltz Algorithm

Daniel Appelö, Francis Appiah, Jeffrey W. Banks, Cassandra Carrick, William D. Henshaw, Donald W. Schwendeman · 2025

Read A Multi-Frequency Helmholtz Solver Based on the WaveHoltz Algorithm by Daniel Appelö, Francis Appiah, Jeffrey W. Banks, Cassandra Carrick, William D. Henshaw, Donald W. Schwendeman (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-23613.txt.

Introduction

Francis Appiah and co-authors present a solver for wave problems at many frequencies in a 2025 paper. The Helmholtz equation governs steady waves in sound, light, and other fields. Solving it again for each frequency and each source wastes effort when many answers are needed.

Their method extends an approach that filters a related wave equation over time. One wave solve, paired with several filters, yields answers for all frequencies at once. A standard speedup method makes the iteration converge faster. The scheme can run with few time steps per wave period and keeps costs in line with grid size. Tests show accurate results at high order in space.

Catalog record

CREATORDaniel Appelö, Francis Appiah, Jeffrey W. Banks, Cassandra Carrick, William D. Henshaw, Donald W. Schwendeman
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-23613

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Multi-Frequency Helmholtz Solver Based on the WaveHoltz Algorithm Daniel Appelö, Francis Appiah, Jeffrey W. Banks, Cassandra Carrick, William D. Henshaw, Donald W. Schwendeman Abstract We develop and analyze a new approach for simultaneously computing multiple solutions to the Helmholtz equation for different frequencies and different forcing functions. The new Multi-Frequency WaveHoltz (MFWH) algorithm is an extension of the original WaveHoltz method and both are based on time-filtering solutions to an associated wave equation. With MFWH, the different Helmholtz solutions are computed simultaneously by solving a single wave equation combined with multiple time filters. The MFWH algorithm defines a fixed-point iteration which can be accelerated with Krylov methods such as GMRES. The solution of the wave equation can be efficiently solved with either explicit time-stepping or implicit time-stepping using as few as five time-steps per period. When combined with an O ⁡ ( N ) O(N) solver for the implicit equations, such a multigrid, the scheme has an O ⁡ ( N ) O(N) solution cost when the frequencies are fixed and the number of grid points N N increases. High-order accurate approximations in space are used together with second-order accurate approximations in time. We show how to remove time discretization errors so that the MFWH solutions converge to the corresponding solutions to the discretized Helmholtz problems. Numerical results are given using second-order accurate and fourth-accurate discretizations to confirm the convergence theory. 1 Introduction MFWH : Multi-Frequency WaveHoltz. {{\mathcal{N}}_{f}} : number of different Helmholtz solutions to compute. u^{(m)}(\mathbf{x}) : continuous Helmholtz solutions, m=1,2,\ldots,{{\mathcal{N}}_{f}} . f^{(m)}(\mathbf{x}),g^{(m)}(\mathbf{x}) : forcing functions for Helmholtz problems (interior and boundary) m=1,2,\ldots,{{\mathcal{N}}_{f}} . {\mathcal{L}} : elliptic operator in Helmholtz equation {\mathcal{L}}u^{(m)}+\omega^{2}u^{(m)}=f^{(m)} . {\mathcal{B}} : boundary condition operator in Helmholtz equation {\mathcal{B}}u^{(m)}=g^{(m)} . (\lambda_{\nu},\phi_{\nu}(\mathbf{x})) : eigenvalues and eigenfunctions of ({\mathcal{L}},{\mathcal{B}}) , \nu=1,2,\ldots . \omega_{m} : frequencies in Helmholtz equations, m=1,2,\ldots,{{\mathcal{N}}_{f}} , with \omega_{1}<\omega_{2}<\ldots . L_{ph} : p-th order accurate approximation to {\mathcal{L}} . B_{ph} : p-th order accurate boundary conditions. (\lambda_{h,\nu},\boldsymbol{\Phi}_{\nu}) : discrete eigenvalues and eigenvectors of (L_{ph},B_{ph}) , \nu=1,2,\ldots,N_{h} . w(\mathbf{x},t) : solution to the continuous wave equation. W_{\mathbf{j}}^{n} : discrete solution W_{\mathbf{j}}^{n}\approx w(\mathbf{x}_{\mathbf{j}},t^{n}) , for t^{n}=n\Delta t , and grid index \mathbf{j}=[j_{1},j_{2},j_{3}] . F_{\mathbf{j}}^{n} : composite forcing, sum of f^{(m)}(\mathbf{x}_{\mathbf{j}})\cos(\omega_{m}t^{n}) , m=1,2,\ldots,{{\mathcal{N}}_{f}} . G_{\mathbf{j}}^{n} : composite boundary forcing, sum of g^{(m)}(\mathbf{x}_{\mathbf{j}})\cos(\omega_{m}t^{n}) , m=1,2,\ldots,{{\mathcal{N}}_{f}} . T_{m}=(2\pi)/\omega_{m} : period corresponding to \omega_{m} . N_{p,m} : number of time periods over which the filter for solution m is integrated. T_{f,m}=N_{p,m}T_{m} : final time for the filter for solution m . v^{(m,k)}(\mathbf{x}) : continuous approximation to u^{(m)} at iteration k=0,1,2,\ldots V_{\mathbf{j}}^{(m,k)} : discrete approximation to v^{(m,k)}(\mathbf{x}) , V_{\mathbf{j}}^{(m,k)}\approx v^{(m,k)}(\mathbf{x}) . \mathbf{V}_{h}^{(m,k)} : vector holding all grid point values of V_{\mathbf{j}}^{(m,k)} . FPI : MFWH fixed point iteration. \beta_{m}(\lambda) : component WaveHoltz filter function for \omega_{m} . \beta_{d}^{(m)}(\lambda) : time-discrete WaveHoltz filter using an adjusted Trapezoidal rule. \hat{u}_{\nu}^{(m)} : coefficient in the eigenvector expansion of u^{(m)}(\mathbf{x}) , \nu=0,1,2,\ldots . \hat{f}_{\nu}^{(m)} : coefficient in the eigenvector expansion of f^{(m)}(\mathbf{x}) , \nu=0,1,2,\ldots . \hat{v}_{\nu}^{(m,k)} : coefficient in the eigenvector expansion of v^{(m,k)}(\mathbf{x}) , \nu=0,1,2,\ldots . \hat{w}_{\nu}^{(k)}(t) : coefficient in the eigenvector expansion of w^{(k)}(\mathbf{x},t) , \nu=0,1,2,\ldots . \mu(\lambda) : multi-frequency filter function for the implicit filter (determines convergence rate for the FPI). (\alpha^{\scriptscriptstyle I},\beta^{\scriptscriptstyle I}) : weightings in implicit time-stepping ( \beta^{\scriptscriptstyle I}=0 : trapezoidal rule, \beta^{\scriptscriptstyle I}=1/2 : full weighting). {\mathcal{W}}_{h} : MFWH iteration operator (affine), \mathbf{V}^{(k+1)}={\mathcal{W}}_{h}\mathbf{V}^{(k)} . S_{h} : MFWH iteration operator (linear), \mathbf{V}^{(k+1)}=S_{h}\mathbf{V}^{(k)}+\mathbf{b}_{h} , \mathbf{b}_{h}\overset{{\rm def}}{=}{\mathcal{W}}_{h}\mathbf{0} . A_{h}=I-S_{h} : MFWH matrix used with matrix-free GMRES to solve A_{h}\mathbf{V}_{h}=\mathbf{b}_{h} . Table 1: Nomenclature Developing efficient solvers for the Helmholtz equation is an important and challenging topic with wide applicability in the applied sciences. Two main challenges in solving Helmholtz equations numerically are the resolution requirements that arise from pollution errors BaylissGoldsteinTurkel1985 , IlenburgBabuska1995 , overHoltzPartTwo , and the highly indefinite character of the discretized system of equations which can cause great difficulty for iterative methods ernst2012difficult . There are many direct and iterative algorithms that have been developed to solve Helmholtz problems and we refer the reader to erlangga2008advances , ModernSolversForHelmholtz2017 and the references therein for further information. It is often the case in practice that one desires frequency domain solutions for many different frequencies or different forcings. In this article we develop and analyse a new approach for simultaneously computing multiple solutions to the Helmholtz equation. The scheme is an extension of the WaveHoltz algorithm which solves the Helmholtz equation by time-filtering solutions to the wave equation WaveHoltz1 , peng2021emwaveholtz , ElWaveHoltz , WaveHoltzSemi , overHoltzPartOne . The different Helmholtz solutions, which may have different forcings and different frequencies, are computed simultaneously by solving a single wave equation combined with multiple filters. The new scheme, called the Multi-Frequency WaveHoltz (MFWH) algorithm, defines a fixed-point iteration which can be accelerated with Krylov methods such as GMRES. The solution of the wave equation can be solved efficiently either with explicit time-stepping, or with implicit time-stepping using as few as five time-steps per period. The approach we develop here has the advantage that it avoids the need to invert an large indefinite matrix for a shifted Laplacian; the implicit time-stepping matrix is definite and amenable to fast solution methods such as multigrid. When combined with an O(N) solver for the implicit equations such a multigrid, the scheme has an O(N) solution cost when the frequencies are fixed and the number of grid points N increases, as shown for the single frequency case in overHoltzPartOne . Resolution requirements to overcome pollution errors are addressed through the use of high-order accurate methods. In this article we restrict out attention to simple geometries, and to energy conserving problems on closed domains using Dirichlet or Neumann boundary conditions. The extension to Helmholtz problems on complex geometries, and for open domains where radiation or absorbing boundary conditions are used for the wave equation, are left to the future. For reference, Table 1 provides a summary of some of the symbols and notation that will be introduced in subsequent sections. 2 Governing equations and problem definition In many applications, one wishes to find solutions to Helmholtz boundary-value problems for many different frequencies and/or many different forcing functions. In this case, solutions labeled u^{(m)}(\mathbf{x}) , m=1,2,\ldots,{{\mathcal{N}}_{f}} , are sought that satisfy \displaystyle{\mathcal{L}}u^{(m)}+\omega_{m}^{2}\,u^{(m)}=f^{(m)}(\mathbf{x}),\qquad \displaystyle\mathbf{x}\in\Omega,\qquad (1a) \displaystyle{\mathcal{B}}u^{(m)}=g^{(m)}(\mathbf{x}),\qquad \displaystyle\mathbf{x}\in\partial\Omega, (1b) for different frequencies \omega_{m}\in{\mathbb{R}} , different volume forcings f^{(m)} , and different boundary data g^{(m)} . \Omega denotes the bounded domain of interest, and \partial\Omega the boundary. Here {\mathcal{L}}=c^{2}\Delta , {\mathcal{B}} denotes a boundary condition operator, taken here as Dirichlet, Neumann, or Robin 1 1 1 Note that the approach described here can be extended to more general elliptic operators {\mathcal{L}} and more general boundary conditions. . Note that the operators {\mathcal{L}} and {\mathcal{B}} are assumed to be the same for each m . We further assume that the frequencies \omega_{m} , m=1,2,\ldots,{{\mathcal{N}}_{f}} , are positive and distinct, are not at resonance, and have been ordered from smallest to largest, \displaystyle 0<\omega_{1}<\omega_{2}<\ldots<\omega_{{{\mathcal{N}}_{f}}}. (2) The Multi-Frequency WaveHoltz (MFWH) algorithm, as described in Section 3 , is based on solving the single forced wave equation, \displaystyle\partial_{t}^{2}w={\mathcal{L}}w-F(\mathbf{x},t),\qquad \displaystyle\mathbf{x}\in\Omega,\quad t\in[0,{T_{f}}], (3a) \displaystyle{\mathcal{B}}w=G(\mathbf{x},t),\qquad \displaystyle\mathbf{x}\in\partial\Omega, (3b) \displaystyle w(\mathbf{x},0)=w_{0}(\mathbf{x}),\qquad \displaystyle\mathbf{x}\in\Omega, (3c) \displaystyle\partial_{t}w(\mathbf{x},0)=0,\qquad \displaystyle\mathbf{x}\in\Omega, (3d) where the composite forcing function F is the sum of the component forcing functions, and the

Read the complete text