Pith. sign in

REVIEW 2 major objections 5 minor 17 references

A Multi-Frequency Helmholtz Solver Based on the WaveHoltz Algorithm

T0 review · 2 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The Multi-Frequency WaveHoltz algorithm computes several Helmholtz solutions from one wave equation solve, with O(N) total cost when frequencies are fixed.

desk verdict A genuine multi-frequency extension of WaveHoltz with clean analysis and good scaling, but the spurious-resonance failure mode is a real caveat that should be flagged more prominently. read the letter →

arxiv 2507.23613 v1 pith:UJBU6B7I submitted 2025-07-31 math.NA cs.NA

Daniel Appelö , Francis Appiah , Jeffrey W. Banks , Cassandra Carrick , William D. Henshaw , Donald W. Schwendeman This is my paper · ORCID
classification math.NAcs.NA MSC 65N2265N5565F1035J05
keywords HelmholtzequationWaveHoltzmulti-frequencytime-filteringfixed-pointiterationGMRESimplicittime-steppingmultigrid
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 presents Multi-Frequency WaveHoltz (MFWH), an algorithm that solves several Helmholtz boundary-value problems at different frequencies and forcings simultaneously by solving a single forced wave equation and applying multiple time filters. It proves that the fixed-point iteration converges exactly when a scalar filter function mu(lambda) satisfies |mu(lambda_nu)| < 1 for every discrete eigenvalue, and that Krylov acceleration handles cases where the plain iteration diverges. With implicit time stepping and an O(N) multigrid solver for the implicit equations, the scheme achieves O(N) total cost for a fixed set of frequencies. Time-discretization errors are removed by using modified frequencies and adjusted quadrature, so the converged iterates match the discrete Helmholtz solutions with the original frequencies. Numerical results in 2D confirm the convergence theory and the O(N) scaling.

What carries the argument

The machinery is the multi-frequency filter function mu($\lambda$) = sum_{m=1}^{N_f} w_m beta_m($\lambda$), where w_m are the column sums of $A^{{-1}}$ and beta_m($\lambda$) is the single-frequency WaveHoltz filter, a combination of three sinc functions centered at $\lambda$ = omega_m, $\lambda$ = -omega_m, and $\lambda$ = 0. The matrix A has entries beta_i(omega_j) and the rank-one matrix B($\lambda$) has entries beta_i($\lambda$); because B is rank one, the iteration matrix M_nu = $A^{{-1}}$B has exactly one nonzero eigenvalue, namely mu($\lambda$). This reduction turns the multi-frequency convergence question into a one-dimensional check: plot mu against the discrete spectrum and see whether any eigenvalue lands where |mu| >= 1. The discrete version uses modified frequencies (Eq. 39) and quadrature weights (Eq. 45) so the fixed point of the filtered iteration is the discrete Helmholtz solution with the original frequencies.

What would settle it

Compute the discrete spectrum of the operator on a specific grid (for example, the Dirichlet Laplacian on a unit square), plot mu_d(lambda) from Eq. (46), and check whether any eigenvalue lambda_{h,nu} distinct from all omega_m has |mu_d| >= 1; the paper's Figure 5 already exhibits a case where the fixed point diverges because the asymptotic convergence rate exceeds one, and Figure 1 shows spurious resonances near lambda = 7 and 11 for omega_1 = 5, omega_2 = 9 with N_p = 1. A direct test is to run GMRES on (I - S_h)V_h = b_h for a configuration with a spurious resonance and observe the residual in the original discretized Helmholtz equations failing to reach machine precision, confirming that A_h is singular or nearly singular.

Watch

Extended reading notes

Core claim

The central claim is that multiple Helmholtz solutions can be obtained from one wave-equation solve by forming a composite forcing sum f^(m)(x) cos(omega_m t) and filtering the wave solution with N_f time integrals, one per frequency. The error in each eigenmode contracts by the scalar mu($\lambda$) = sum w_m beta_m($\lambda$), the unique nonzero eigenvalue of $A^{{-1}}$B($\lambda$); the fixed-point iteration converges iff |mu(lambda_nu)| < 1 for all discrete eigenvalues lambda_nu. When the plain iteration stalls, solving the linear system (I - S_h)V_h = b_h with GMRES converges rapidly, and with implicit time stepping at about ten steps per period plus an O(N) multigrid solve per step, the total work scales as O(N) for fixed frequencies.

Load-bearing premise

The method assumes that no discrete eigenvalue of the spatial discretization, other than the target frequencies themselves, makes the multi-frequency filter function equal (or nearly equal) to one; if such a spurious resonance exists, the linear system becomes singular and the Krylov solver will likely fail, with only a posteriori residual checks and changing the number of periods as remedies.

Editorial extensions

If this is right

  • A single wave-equation solve over one common time interval produces N_f Helmholtz solutions at distinct frequencies, so the per-frequency cost is amortized.
  • Because the implicit time-stepping matrix is definite, the indefinite Helmholtz system is never formed; multigrid applied to the wave-equation steps keeps total cost O(N) for fixed frequencies.
  • The fixed-point iteration's convergence rate is exactly predicted by the scalar function mu(lambda), so the number of periods N_p can be tuned by inspecting mu_d against the discrete spectrum.
  • Krylov acceleration converges even in cases where the plain fixed point diverges (for example, N_p = 1), with similar effective cost per grid point.
  • Time-discretization errors are eliminated by solving with modified frequencies and filtered quadrature, so the converged iterates match the discrete Helmholtz solutions with the original frequencies.

Reading between the lines

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

  • The rank-one structure of B(lambda) means the entire multi-frequency convergence analysis reduces to a one-dimensional filter function; the same reduction is likely available for filters built from other quadrature rules or time-stepping schemes.
  • The spurious-resonance failure mode suggests a practical safeguard: before committing to a set of frequencies and N_p, scan mu_d(lambda) over the discrete spectrum at negligible cost and adjust N_p or slightly shift frequencies to keep |mu_d| < 1 away from the target frequencies.
  • The method's current restriction to closed domains with Dirichlet, Neumann, or Robin conditions limits its use in scattering and inverse problems; extending the composite-forcing idea to absorbing boundary conditions or perfectly matched layers would make simultaneous multi-frequency solves viable for those settings.
  • Because the algorithm only needs a time-domain wave solver, existing mature explicit and implicit time-stepping codes could be wrapped to produce frequency-domain solutions without building a new Helmholtz-specific solver.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The manuscript introduces the Multi-Frequency WaveHoltz (MFWH) algorithm for solving N_f Helmholtz boundary value problems with distinct frequencies ω_m and forcings f^(m) by solving a single forced wave equation whose composite forcing is a sum of f^(m) cos(ω_m t). The iteration applies N_f time filters to the wave-equation solution, updates the component iterates through a small N_f × N_f linear system, and can be accelerated by GMRES. The paper derives a rank-one error-propagation matrix whose nonzero eigenvalue μ(λ) is a weighted sum of filter functions, proves μ(ω_m)=1, and defines a discrete version μ_d with modified frequencies to remove time-discretization error. Numerical experiments on the unit square with Dirichlet conditions confirm the predicted fixed-point rates and show GMRES convergence, and scaling tests with implicit time stepping and multigrid indicate O(N) complexity.

Significance. The core idea is attractive and the continuous-level convergence analysis is clean, self-contained, and consistent with the reported residuals. If the discrete time-error correction can be made fully self-contained, the method offers a genuinely different route to multi-frequency Helmholtz solves: one time-domain wave-equation solve with multiple filters, a definite implicit time-stepping matrix amenable to multigrid, and demonstrated O(N) scaling for fixed frequencies. The paper also provides transparent numerical validation and reports timings up to 4.2×10^6 unknowns. However, two caveats limit the strength of the central claim: the acknowledged spurious-resonance failure mode in Section 6, and the deferral of the discrete time-error analysis to the unpublished companion [11]. Neither appears fatal, but both need to be addressed before the advertised performance can be considered robust.

major comments (2)
  1. [Section 6, Eq. (48)] The statement that the Krylov-accelerated MFWH solves A_h V_h = b_h is load-bearing, but the paper explicitly acknowledges that a discrete eigenvalue λ_{h,ν} distinct from all ω_m with μ_d(λ_{h,ν})=1 (or nearly 1) makes A_h singular (or nearly singular) and will likely cause the Krylov solver to fail. No sufficient condition, parameter choice, or automatic detection is provided; the only proposed remedies are a posteriori residual checks and manually changing N_p. This is not a hypothetical edge case: the N_f=2, ω=(5,9), N_p=1 case plotted in the left panel of Figure 1 shows spurious crossings near λ≈7, and the unit-square Dirichlet Laplacian has an eigenvalue λ=π√5≈7.025, so a standard geometry can fall at or near such a crossing. Since the abstract and conclusions present MFWH as a general simultaneous solver without this caveat, the central claim needs to be either restricted or supported by a condition that rules out spurious resonances, or by a robust automatic remedy.
  2. [Section 5, Eqs. (39)-(45); Section 5.3] The proof that the discrete iterates converge to the discrete Helmholtz solutions with pristine frequencies ω_m (Eq. (41)) is not self-contained. The modified-frequency definition (39), the discrete filter (42), the choice of α_m in (45), and the time-step recipe are stated, but the derivation that these choices exactly remove the time-discretization error is deferred to the unpublished manuscript [11]. The sentence "following the recipe in [11]" in Section 5.3 delegates a central step on which the numerical claims and the O(N) scaling argument rest. The authors should supply a self-contained derivation or a rigorous error estimate for the discrete filter, or at minimum make the argument reproducible within this paper.
minor comments (5)
  1. [Section 1] The phrase "we restrict out attention" should be "we restrict our attention."
  2. [Section 7, Eq. (52)] The notation V^{m,k}_j should be V^{(m,k)}_j to match Table 1, and the residual defined in Eq. (52) is an iteration residual (difference of successive iterates) rather than a Helmholtz residual; this should be stated explicitly to avoid confusion.
  3. [Section 7, Eq. (49)] The ACR reported in Figures 3-6 is computed from the same discrete filter function μ_d that defines the iteration, so the match between predicted and observed convergence is partly a consistency check. An independent validation would be a quantitative comparison of the converged MFWH solutions against direct Helmholtz solves for several discretizations; the paper mentions a relative error of about 1e-12 but does not provide a table.
  4. [Section 4.2, Eq. (37)] The formula for μ in Eq. (37) is given only for N_f=2; a general expression for the weights w_m in Eq. (28) would help readers implement the ACR prediction for larger N_f.
  5. [References] References [8] and [11] are marked as "submitted" or "unpublished"; the manuscript should indicate how these can be accessed or should cite published versions if available.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the convergence analysis is derived from the stated filter definitions, and the numerical ACR is a consistency check rather than a fitted prediction; the self-citation to [11] is minor and not load-bearing.

full rationale

The derivation chain in Sections 3-4 is self-contained. Equation (21)-(22) expresses the error iteration as e^{k+1} = A^{-1}B(lambda)e^k, with A and B built from the filter functions beta_m(lambda) defined in Eq. (10) for the prescribed target frequencies; the target frequencies are inputs, never outputs inferred from the iteration. Theorem 2's mu(lambda) follows algebraically, and Theorem 3 is a direct check that mu(omega_m)=1, not a fitted condition. The numerical ACR in Eq. (49) is a direct evaluation of the analytic mu_d at the known discrete eigenvalues, so comparing it with the measured CR in Eq. (51) is a consistency check of the eigenfunction analysis, not a curve-fit or a renamed prediction. The time-discretization correction in Eqs. (39) and (45) is stated explicitly in the paper even though some details are delegated to the same-authors' unpublished manuscript [11]; this is an omitted-proof/self-citation point, but the formulas are given and the paper's own scaling runs use 10 steps per period, so the 'five time-steps per period' claim in [11] is not load-bearing for the demonstrated results. Section 6 honestly acknowledges that a spurious resonance can make A_h singular and that the Krylov solver may fail, with only a posteriori residual checks and a change of N_p as remedies; this is a real correctness limitation on the abstract's unqualified claim, but it is a robustness gap, not a circular reduction of the result to its own inputs. No fitted parameter is renamed as a prediction and no uniqueness theorem is imported from the authors' prior work.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central claim rests on eigenfunction analysis, superposition for the linear wave equation, absence of resonance, and a discrete time-error correction carried over from the submitted single-frequency WaveHoltz paper. User-selected parameters Np, time-steps per period, and CFL safety factor affect convergence and cost but are not fitted to data. No new physical entities are introduced.

free parameters (3)
  • Np (integration window in periods of the longest frequency) = 1, 2, 3, 6 in the numerical examples
    Chosen by the user; larger Np narrows the filter peaks and improves fixed-point convergence but costs more wave solves. Not fitted to data.
  • time-steps per period for implicit stepping = 10 in tests, minimum 5 claimed
    Determines Delta t; the five-steps-per-period rule is taken from the prior WaveHoltz paper [11] and is not re-derived here.
  • CFL safety factor for explicit stepping = 0.9
    Hand-set in Section 5.3 to stay below the von Neumann stability limit; does not affect the central convergence claim.
assumptions (6)
  • domain assumption The eigenvalue BVP L phi = -lambda^2 phi, B phi = 0 has a complete set of linearly independent eigenfunctions on Omega.
    Invoked in Section 4.1 to expand Helmholtz and wave solutions; standard for Dirichlet/Neumann on bounded domains.
  • domain assumption The frequencies omega_m are positive, distinct, and none equals an eigenvalue lambda_nu (no resonance).
    Stated in Section 2; required for the coefficient formula (16) and for mu(omega_m)=1 with invertible A.
  • domain assumption The analysis section assumes zero boundary forcing g^(m)=0; numerical tests use homogeneous Dirichlet boundary conditions.
    Section 4.1 says boundary forcings are set to zero to simplify; non-homogeneous boundaries are not analyzed.
  • standard math The wave equation is linear, so the solution with composite forcing and summed initial data is the sum of the per-frequency component solutions.
    Used implicitly in the eigenfunction ODE solution (17d) and in the filter update (7).
  • domain assumption The modified frequencies (39) and discrete quadrature (42) with alpha_m from (45) remove time-discretization error so iterates converge to the discrete Helmholtz solutions with pristine frequencies.
    The paper states this and verifies numerically, but the full proof is adapted from the single-frequency WaveHoltz analysis in the submitted paper [11], not fully reproduced here.
  • domain assumption The implicit time-stepping matrix I - alpha_I Delta t^2 L_h is positive definite and solvable by an O(N) multigrid method.
    Required for the claimed O(N) complexity; the paper relies on the Ogmg solver [15-17] and does not prove multigrid convergence for this system.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Multi-Frequency Helmholtz Solver Based on the WaveHoltz Algorithm." pith.science (2026). https://pith.science/paper/UJBU6B7I

@misc{pith2026250723613,
  author       = {Pith},
  title        = {Pith review of: A Multi-Frequency Helmholtz Solver Based on the WaveHoltz Algorithm},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UJBU6B7I}},
  note         = {Machine review of arXiv:2507.23613}
}
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 $O(N)$ 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. 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.

Figures

Figures reproduced from arXiv: 2507.23613 by the authors.

Figure 1
Figure 1. Multi-frequency WaveHoltz filter function for two frequencies, [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Multi-frequency WaveHoltz filter function for two frequencies, [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Multi-frequency algorithm, square128, order 4, [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Multi-frequency algorithm, square128, order 4, [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Multi-frequency algorithm, square128, order 4, [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Multi-frequency algorithm, square128, order 4, [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Solving a three frequency MFWH problem with implicit time-stepping and multigrid. Left: Solvers statistics for grids [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

17 extracted references · 14 canonical work pages

  1. [11]

    Appel¨ o, J

    D. Appel¨ o, J. W. Banks, W. D. Henshaw, D. W. Schwendeman, An optimal O(N) Helmholtz solver using WaveHoltz and overset grids, submitted (2025)

    work page 2025

  2. [1]

    Bayliss, C

    A. Bayliss, C. Goldstein, E. Turkel, The numerical solution of the Helmholtz equation for wave prop- agation problems in underwater acoustics, Computers & Mathematics with Applications 11 (7) (1985) 655–665, special Issue Computational Ocean Acoustics

    work page 1985

  3. [2]

    Ihlenburg, I

    F. Ihlenburg, I. Babuˇ ska, Finite element solution of the Helmholtz equation with high wave number Part I: The h-version of the FEM, Computers & Mathematics with Applications 30 (9) (1995) 9–37

    work page 1995

  4. [3]

    Appel¨ o, J

    D. Appel¨ o, J. W. Banks, W. D. Henshaw, D. W. Schwendeman, A rule of thumb for choosing points- per-wavelength for finite difference approximations of Helmholtz problems, submitted (2025)

    work page 2025

  5. [4]

    Ernst, M

    O. Ernst, M. Gander, Why it is difficult to solve Helmholtz problems with classical iterative methods, in: Numerical analysis of multiscale problems, Springer, 2012, pp. 325–363. 17

    work page 2012

  6. [5]

    Erlangga, Advances in iterative methods and preconditioners for the Helmholtz equation, Archives of Computational Methods in Engineering 15 (1) (2008) 37–66

    Y. Erlangga, Advances in iterative methods and preconditioners for the Helmholtz equation, Archives of Computational Methods in Engineering 15 (1) (2008) 37–66

    work page 2008

  7. [6]

    Lahaye, J

    D. Lahaye, J. Tang, K. Vuik (Eds.), Modern Solvers for Helmholtz Problems, Birkh¨ auser, 2017

    work page 2017

  8. [7]

    Appel¨ o, F

    D. Appel¨ o, F. Garcia, O. Runborg, WaveHoltz: Iterative solution of the Helmholtz equation via the wave equation, SIAM Journal on Scientific Computing 42 (4) (2020) A1950–A1983

    work page 2020

Show all 17 references
  1. [8]

    Z. Peng, D. Appel¨ o, EM-WaveHoltz: A flexible frequency-domain method built from time-domain solvers, IEEE Transactions on Antennas & Propagation

  2. [9]

    Appel¨ o, F

    D. Appel¨ o, F. Garcia, A. A. Loya, O. Runborg, El-WaveHoltz: A time-domain iterative solver for time-harmonic elastic waves, Computer Methods in Applied Mechanics and Engineering 401 (2022) 115603

    2022

  3. [10]

    Rotem, O

    A. Rotem, O. Runborg, D. Appel¨ o, Convergence of the semi-discrete WaveHoltz iteration, arXiv:2407.06929

  4. [12]

    W. A. Strauss, Partial differential equations: an introduction, 2nd Edition, Wiley, United States of America, 2008

    2008

  5. [13]

    N. G. Al Hassanieh, J. W. Banks, W. D. Henshaw, D. W. Schwendeman, Local compatibility boundary conditions for high-order accurate finite-difference approximations of PDEs, SIAM J. Sci. Comput. 44 (2022) A3645–A3672

    2022

  6. [14]

    A. M. Carson, J. W. Banks, W. D. Henshaw, D. W. Schwendeman, High-order accurate implicit-explicit time-stepping schemes for wave equations on overset grids, Journal of Computational Physics 520 (2025) 113513

    2025

  7. [15]

    W. D. Henshaw, Ogmg: A multigrid solver for Overture, user guide, version 1.00, Research Report UCRL-MA-134446, Lawrence Livermore National Laboratory (1999)

    1999

  8. [16]

    W. D. Henshaw, On multigrid for overlapping grids, SIAM J. Sci. Comput. 26 (5) (2005) 1547–1572

    2005

  9. [17]

    C. Liu, W. D. Henshaw, Multigrid with nonstandard coarse-level operators and coarsening factors, Journal of Scientific Computing 94 (58) (2023) 1–27. 18

    2023

Pith tools

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