Pith. sign in

REVIEW 3 major objections 4 minor 56 references

Overlapping Schwarz methods are not anisotropy-robust multigrid smoothers

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Overlapping multiplicative Schwarz smoothers are not anisotropy-robust multigrid smoothers for anisotropic diffusion; blocks must grow as $\epsilon^{-1/2}$ for the smoothing factor to stay bounded away from 1.

desk verdict Solid LFA study showing local overlapping Schwarz smoothers are not anisotropy-robust; the O(epsilon^{-1/2}) block-size scaling is credible, with a disclosed but unproved smoothness assumption as the main caveat. read the letter →

arxiv 2501.00938 v1 pith:DZ6OAUSZ submitted 2025-01-01 math.NA cs.NA

Oliver A. Krzysik , Ben S. Southworth , Bobby Philip This is my paper · ORCID
classification math.NAcs.NA MSC 65N5565N1265F10
keywords overlappingSchwarzlocalFourieranalysisanisotropicdiffusionmultigridsmoothersmoothingfactoranisotropyrobustnesslineblocksmoothers
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 asks whether local overlapping block smoothers can replace global line smoothers in multigrid for strongly anisotropic diffusion, and it answers no. For any fixed subdomain size that stays well short of the whole domain, the smoothing factor of overlapping multiplicative Schwarz tends to 1 as the anisotropy ratio $\epsilon$ tends to 0, no matter how much the blocks overlap. This means the smoother fails to damp exactly the high-frequency error components that a full-coarsening multigrid method cannot represent on the coarse grid. The paper proves a quantitative version of the failure: for the finite-difference discretization the smoothing factor is $1 - \ell(\ell+1)\epsilon + O(\epsilon^2)$ for $\ell \times 1$ blocks, so holding it below a fixed constant forces $\ell$ to grow like $\epsilon^{-1/2}$. A sympathetic reader would care because this closes a question left open by recent successful uses of block smoothers on transport-dominated problems: those successes do not transfer to anisotropic diffusion.

What carries the argument

The central object is the Fourier symbol of the multiplicative Schwarz error propagator for maximally overlapping rectangular subdomains, and its maximum over high-frequency modes, called the smoothing factor. For $\ell\times1$ blocks the paper shows the propagator is diagonalized by Fourier modes, so the symbol can be written as a power series in $\epsilon$; the first-order term is what makes the argument quantitative. The critical mechanism is that the worst mode sits at $\omega_1=0$, where the zeroth-order symbol is exactly 1, and a perturbation lemma transfers the maximization order by order in $\epsilon$. That yields the coefficient $c\ell(\ell+1)$ multiplying $\epsilon$, which directly couples the block length to the anisotropy ratio and produces both the failure at fixed $\ell$ and the $\epsilon^{-1/2}$ cure.

What would settle it

Take the finite-difference discretization of the grid-aligned problem, fix $\ell=8$, and compute the Fourier symbol from the linear system that defines the smoother on a dense grid of high frequencies at $\epsilon=10^{-6}$. If the maximum of the symbol's magnitude differs from $1-\ell(\ell+1)\epsilon$ by more than $O(\epsilon^2)$, or is attained at a point where the symbol is not differentiable, then the smoothness assumption behind the central expansion fails. A simpler solver-level test: if any fixed block size such as $4\times1$ keeps the V-cycle convergence factor strictly below 1 for all $\epsilon\in[10^{-8},1]$, the paper's main negative claim is wrong.

Watch

Extended reading notes

Core claim

The paper's central discovery is a negative result with a quantitative cure. For the grid-aligned anisotropic diffusion equation $-u_{xx}-\epsilon u_{yy}=f$ with $\epsilon\in(0,1]$, analyzed through local Fourier analysis, the smoothing factor of maximally overlapping multiplicative Schwarz on $\ell\times1$ blocks is $\mu_{\ell,1}(\epsilon)=1-\ell(\ell+1)\epsilon+O(\epsilon^2)$ for the second-order finite-difference discretization, and at least $1-\frac{3}{2}\ell(\ell+1)\epsilon+O(\epsilon^2)$ for the bilinear finite-element discretization. Because the smoothing factor is the worst-case amplification of high-frequency Fourier modes, any fixed $\ell$ yields $\lim_{\epsilon\to0^+}\mu_{\ell,1}(\epsilon)=1$, meaning the smoother leaves the hardest high-frequency modes undamped. Inverting the expansion shows that keeping the smoothing factor below a fixed constant $\mu^*<1$ requires $\ell=O(\epsilon^{-1/2})$, and the paper gives an explicit ceiling formula for the finite-difference case. Numerical V-cycle experiments on finite domains confirm the local Fourier analysis predictions once the block size is bounded well away from the domain size. The conclusion the authors draw is that global line or plane smoothing is necessary for anisotropy-robust multigrid; local overlapping Schwarz blocks are not a substitute, and the recent successes of block smoothers on transport-dominated problems do not carry over to anisotropic diffusion.

Load-bearing premise

The whole expansion, and therefore the $\epsilon^{-1/2}$ threshold, assumes the Fourier symbols are smooth enough that their highest high-frequency mode can be located by a critical-point calculation; the paper does not verify this smoothness explicitly.

Editorial extensions

If this is right

  • For any fixed block size, the multigrid convergence factor using this smoother tends to 1 as $\epsilon\to0$, so the solver is not $\epsilon$-robust for any fixed local block.
  • Maximal overlap is the strongest case; since even that fails, reducing overlap cannot restore anisotropy robustness.
  • Anisotropy-robust smoothing forces block diameter $O(\epsilon^{-1/2})$, so for fusion-relevant anisotropies with $\epsilon\sim10^{-12}$ local blocks are effectively global and line or plane smoothers remain the practical option.
  • For mild or moderate anisotropy, the explicit formula $\ell=\left\lceil\sqrt{1-\mu^*}\,\epsilon^{-1/2}\right\rceil$ gives a concrete block size that should deliver $\epsilon$-independent smoothing in the finite-difference case.
  • Square $\ell\times\ell$ blocks give almost no smoothing benefit over $\ell\times1$ blocks, so the extra work of square blocks is not justified at small $\epsilon$.

Reading between the lines

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

  • An additive Schwarz version of these smoothers is not analyzed here, and since additive variants typically need damping, the same $\epsilon^{-1/2}$ threshold may or may not apply; but the multiplicative analysis gives the strongest case, so the negative conclusion is likely to carry over.
  • The paper proves the threshold only for grid-aligned anisotropy, while the rotated-case numerics show failure in additional regions; the required block size is therefore unlikely to be smaller off-axis.
  • If the smoothness assumption behind the expansion fails, the specific $O(\epsilon^2)$ form could be wrong while the qualitative conclusion that fixed $\ell$ fails might still survive; evaluating the symbol numerically at very small $\epsilon$ would separate those two possibilities.
  • A directly checkable practical consequence is that, for a chosen target smoothing factor $\mu^*$ on the grid-aligned finite-difference problem, setting $\ell=\left\lceil\sqrt{1-\mu^*}\,\epsilon^{-1/2}\right\rceil$ should produce nearly $\epsilon$-independent multigrid convergence, providing a clean test of the paper's scaling law.
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 / 4 minor

Summary. The paper studies overlapping multiplicative Schwarz methods as smoothers in geometric multigrid for two-dimensional anisotropic diffusion. Using local Fourier analysis, the authors derive asymptotic expansions for the smoothing factor of maximally overlapping ℓ×1 subdomain smoothers applied to grid-aligned anisotropic diffusion, for both finite-difference and bilinear finite-element discretizations. They show that for each fixed ℓ the smoothing factor behaves as 1 - c ℓ(ℓ+1) ε + O(ε²), so it tends to 1 as the anisotropy ratio ε tends to 0; they conclude that ε-robust smoothing requires block diameters growing like O(ε^{-1/2}). Numerical V-cycle experiments on finite domains with Dirichlet boundary conditions agree closely with the LFA predictions, and additional experiments for ℓ×ℓ blocks and smaller overlaps corroborate the qualitative conclusion.

Significance. If the claims hold, this is a useful and somewhat counterintuitive negative result: local overlapping Schwarz smoothers, despite their reputation as strong smoothers, cannot replace global line smoothers for strongly anisotropic diffusion. The paper's strengths are its explicit LFA symbol derivations, the parameter-free form of the leading-order constants in Theorems 1 and 2, and the systematic numerical verification in Figures 6-10 and B2. The asymptotic predictions are not fitted to the numerical data, and the close quantitative agreement, including the ℓ(ℓ+1) scaling, is convincing evidence. However, two gaps in the proof are load-bearing for the headline scaling: the smoothness assumption in Lemma 6 is explicitly not verified, and the extension of fixed-ℓ asymptotics to the ℓ = ℓ(ε) regime used in Corollary 1 requires uniformity in ℓ that is not established. These issues are fixable within the scope of the manuscript, but they should be addressed before publication.

major comments (3)
  1. [Section 3.6, Theorem 3; Appendix B.3, Lemma 6] The proof of the smoothing-factor expansion μ_{ℓ,1}(ε) = 1 - ℓ(ℓ+1)ε + O(ε²) and its FE counterpart relies on Lemma 6, which requires the real-valued functions g0 and g1 obtained from |e_s|² to have bounded derivatives up to third and second order, respectively, on the high-frequency domain. As the authors state in footnote 9 and immediately before Lemma 6, these smoothness conditions are not verified. The functions in (B31) and (B34) are rational in e^{iω}, but their denominators can vanish or fail to be smooth at isolated points for some parameter values, and no argument rules this out. Figure B2 samples the true and linearized symbols on a 64×64 grid and demonstrates pointwise O(ε²) closeness, but it does not estimate derivatives of the remainder. Since the maximum of |e_s|² could in principle be attained at a non-smooth point away from the perturbed critical point of g0, the first-order formula that drives Corollary 1 is not fully proven. Please either verify the derivative bounds analytically, verify them computationally with rigorous interval or certified methods, or prove the maximum-location statement by a different argument.
  2. [Corollary 1, claims 2 and 3; Theorem 3] Corollary 1 draws conclusions for ℓ = ℓ(ε) growing as ε^{-1/2}, but Theorem 3 is an asymptotic statement for fixed ℓ as ε → 0. The O(ε²) remainder in (B36) may depend on ℓ, and no uniformity in ℓ is proved. Claim 3 explicitly assumes uniformity, but claim 2, stated as a necessary condition, does not. If the remainder grows like, say, C ℓ^p ε² with p ≥ 2, then for ℓ ∼ ε^{-1/2} the remainder could be O(ε^{2-p/2}) and could dominate the leading-order term for p > 2. Thus the inference that robustness forces ℓ = O(ε^{-1/2}) requires a two-parameter asymptotic estimate or an explicit bound on the remainder uniform over the relevant range of ℓ. This is a load-bearing point because the O(ε^{-1/2}) block-size scaling is the paper's main quantitative conclusion. Please either supply such a bound or state the conclusion as a conjecture supported by the numerics, with the necessary-condition claim appropriately qualified.
  3. [Abstract and Section 4] The abstract and conclusions state the result for 'any fixed block size,' but the rigorous LFA results cover maximally overlapping ℓ×1 subdomains (Theorems 1-3 and Corollary 1) and 2×2 subdomains (Lemma 1). The ℓ×ℓ case with ℓ > 2 is treated only numerically in Section 4, Figure 9. The numerical evidence is consistent with the ℓ×1 analysis and is persuasive, but the stated scope of the theoretical claim is broader than what is proved. Please state precisely which subdomain geometries and overlap patterns are covered by the proof and which are supported only by numerical experiments.
minor comments (4)
  1. [Introduction, Section 1] The phrase 'pit falls' should be 'pitfalls'.
  2. [Remark 2, Section 4] The interval notation 'θ ∈ (π/4π/2]' appears to be a typo; it should read 'θ ∈ (π/4, π/2]'.
  3. [Figure 10 caption] The line break in 'ra-tio' should be removed: 'anisotropy ratio'.
  4. [Section 3.6, proof of Theorem 3] The sentence 'the function −2ℓ(ℓ+1)(1−cos ω2) is maximized at either end point where cos ω2 = 0' is slightly imprecise: the maximum over ω2 ∈ [π/2, 3π/2] is attained at both endpoints, and the value is −2ℓ(ℓ+1). The subsequent substitution is clear, but the wording could be tightened.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the LFA symbols are derived from the Schwarz correction equations, and the smoothing-factor asymptotics follow from those symbols rather than from fitted data or load-bearing self-citations.

full rationale

I walked the derivation chain and found no step in which an output quantity is equivalent to an input by construction. The smoothing factor mu_{ell,1}(epsilon) is not fitted; it is defined as the maximum of the Fourier symbol modulus (Definition 1), and the symbol itself is obtained by solving the exact Schwarz correction linear system (24)/(31) for the 2x2 and ell x 1 subdomain cases, respectively. Theorems 1 and 2 are explicit asymptotic expansions of those symbols in powers of epsilon, derived algebraically from the stencil coefficients and the Sherman-Morrison formula in Appendix B; there is no fitted constant hidden in the derivation. Theorem 3 then maximizes |es|^2 over the high-frequency domain using the critical-point Lemma 6, and Corollary 1 algebraically solves 1 - c ell(ell+1)epsilon = mu* for ell, which is a necessary-condition statement rather than a prediction forced by data. The external result used for Fourier-mode invariance is MacLachlan and Oosterlee [24], which is not authored by the present authors; moreover, Lemma 2 verifies the invariance for the ell x 1 case, and the resulting symbols and smoothing factors are checked against independent numerical solutions of (31)/(14) in Figures 6, 7, and B2. The self-citations that do appear, such as [44] for a specialized fusion discretization and [43] for upwind stabilization, are background or motivation and do not carry the main theorem. The one explicit limitation is the unverified smoothness hypothesis in Theorem 3: the paper states, "We have no reason to believe that these functions do not meet the smoothness conditions, but we do not explicitly verify due the complexity or non-analytic form of the functions" (Section 3.6, Theorem 3; Appendix B.3). This is a genuine assumption gap that could invalidate the critical-point argument if the symbols fail to be sufficiently smooth, but it is a correctness risk, not circularity: the theorem would fail to be proved, not reduce to its own inputs. I therefore assign score 0, since no self-definitional, fitted-input-as-prediction, or self-citation-chain circularity is exhibited.

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

No parameters are fitted to data; the central results follow from closed-form LFA symbol derivations. The main axioms are the LFA infinite-grid and invariance framework plus two unproved technical premises: symbol smoothness and monotonicity of smoother strength in overlap. No new physical or mathematical entities are introduced.

assumptions (5)
  • domain assumption The discrete operator and the Schwarz smoother are locally constant-coefficient and invariant with respect to Fourier modes on an infinite periodic grid (LFA framework).
    Sections 3.1 and 3.5; this is the standard LFA assumption that boundary effects and coefficient variation are ignored. It is needed to define scalar Fourier symbols and the smoothing factor.
  • standard math The error propagator for maximally overlapping multiplicative Schwarz on 2x2 and ell x 1 subdomains is invariant on Fourier-mode subspaces.
    Section 3.2 and Lemma 2; the invariance is inherited from MacLachlan and Oosterlee [24] by direct application, and the paper's Lemma 2 is a proof sketch rather than a fully self-contained proof.
  • ad hoc to paper The linearized Fourier symbols es0 and es1 are sufficiently smooth with bounded derivatives on the high-frequency domain.
    Assumed in Theorem 3 to apply Lemma 6; the authors explicitly state they do not verify it. This is the weakest technical premise of the central scaling result.
  • ad hoc to paper More overlap gives a stronger smoother, so non-robustness of the maximally overlapping smoother implies non-robustness for all smaller overlaps.
    Stated in Section 4 as 'clearly, the smoother is strongest when maximally overlapped' and used to extend the abstract's 'irrespective of overlap' claim; not proven, though supported by Figure 10.
  • ad hoc to paper The asymptotic expansions in epsilon are valid and the O(epsilon^2) remainders are uniformly small for the epsilon range and ell values studied.
    Corollary 1(3) requires this uniformity; the paper notes good agreement numerically in Figure B2 but does not prove uniform bounds.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Overlapping Schwarz methods are not anisotropy-robust multigrid smoothers." pith.science (2026). https://pith.science/paper/DZ6OAUSZ

@misc{pith2026250100938,
  author       = {Pith},
  title        = {Pith review of: Overlapping Schwarz methods are not anisotropy-robust multigrid smoothers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DZ6OAUSZ}},
  note         = {Machine review of arXiv:2501.00938}
}
abstract

We analyze overlapping multiplicative Schwarz methods as smoothers in the geometric multigrid solution of two-dimensional anisotropic diffusion problems. For diffusion equations, it is well known that the smoothing properties of point-wise smoothers, such as Gauss--Seidel, rapidly deteriorate as the strength of anisotropy increases. On the other hand, global smoothers based on line smoothing are known to generally provide good smoothing for diffusion problems, independent of the anisotropy strength. A natural question is whether global methods are really necessary to achieve good smoothing in such problems, or whether it can be obtained with locally overlapping block smoothers using sufficiently large blocks and overlap. Through local Fourier analysis and careful numerical experimentation, we show that global methods are indeed necessary to achieve anisotropy-robust smoothing. Specifically, for any fixed block size bounded sufficiently far away from the global domain size, we find that the smoothing properties of overlapping multiplicative Schwarz rapidly deteriorate with increasing anisotropy, irrespective of the amount of overlap between blocks. Moreover, our results indicate that anisotropy-robust smoothing requires blocks of diameter ${\cal O}(\epsilon^{-1/2})$ for anisotropy ratio $\epsilon \in (0,1]$.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 56 canonical work pages

  1. [1]

    Abu-Labdeh, S

    R. Abu-Labdeh, S. MacLachlan, and P. E. Farrell,Monolithic multigrid for implicit Runge–Kutta discretizations of incompressible fluid flow, Journal of Computational Physics 478 (2023), 111961

    work page 2023

  2. [2]

    J. H. Adler, T. R. Benson, E. C. Cyr, P. E. Farrell, S. P. MacLachlan, and R. S. Tuminaro,Monolithic multigrid methods for magnetohydrodynamics, SIAM Journal on Scientific Computing 43 (2021), no. 5, S70–S91

    work page 2021

  3. [3]

    R. E. Alcouffe, A. Brandt, J. E. Dendy, Jr, and J. W. Painter, The multi-grid method for the diffusion equation with strongly discontinuous coefficients, SIAM Journal on Scientific and Statistical Computing 2 (1981), no. 4, 430–454

    work page 1981

  4. [4]

    Rodrigo, and F

    Álvaro Pé de la Riva, C. Rodrigo, and F. J. Gaspar,A two-level method for isogeometric discretizations based on multiplicative Schwarz iterations, Computers & Mathematics with Applications 100 (2021), 41–50

    work page 2021

  5. [5]

    Rodrigo, F

    Álvaro Pé de la Riva, C. Rodrigo, F. J. Gaspar, J. H. Adler, X. Hu, and L. Zikatanov,A local Fourier analysis for additive Schwarz smoothers, Computers & Mathematics with Applications 158 (2024), 13–20

    work page 2024

  6. [6]

    N. Bell, L. N. Olson, J. Schroder, and B. Southworth, PyAMG: Algebraic Multigrid Solvers in Python, Journal of Open Source Software 8 (2023), no. 87, 5495

    work page 2023

  7. [7]

    Brandt, Multi-level adaptive solutions to boundary-value problems, Math

    A. Brandt, Multi-level adaptive solutions to boundary-value problems, Math. Comp. 31 (1977), no. 138

    work page 1977

  8. [8]

    Algebraic distance for anisotropic diffusion problems: multilevel results

    A. Brandt, J. Brannick, K. Kahl, and I. Livshits,Algebraic distance for anisotropic diffusion problems: multilevel results(2014). ArXiv:1409.4702

    work page Pith review arXiv 2014

Show all 56 references
  1. [9]

    W. L. Briggs, V . E. Henson, and S. F. McCormick,A Multigrid Tutorial, Second Edition, 2nd edn., Society for Industrial and Applied Mathematics, 2000,

    2000

  2. [10]

    Chacón and G

    L. Chacón and G. Di Giannatale, An asymptotic-preserving semi-lagrangian algorithm for the anisotropic heat transport equation with arbitrary magnetic fields, Journal of Computational Physics 516 (2024), 113368

    2024

  3. [11]

    A. P. de la Riva, C. Rodrigo, and F. J. Gaspar,A robust multigrid solver for isogeometric analysis based on multiplicative Schwarz smoothers, SIAM Journal on Scientific Computing 41 (2019), no. 5, S321–S345

    2019

  4. [12]

    P. E. Farrell, Y . He, and S. P. MacLachlan,A local Fourier analysis of additive Vanka relaxation for the Stokes equations, Numerical Linear Algebra with Applications 28 (2021), no. 3, e2306

    2021

  5. [13]

    P. E. Farrell, L. Mitchell, and F. Wechsung, An augmented Lagrangian preconditioner for the 3d stationary incompressible Navier–Stokes equations at high Reynolds number, SIAM Journal on Scientific Computing 41 (2019), no. 5, A3073–A3096

    2019

  6. [14]

    M. W. Gee, J. J. Hu, and R. S. Tuminaro,A new smoothed aggregation multigrid method for anisotropic problems, Numerical Linear Algebra with Applications 16 (2009), no. 1, 19–37

    2009

  7. [15]

    Green, X

    D. Green, X. Hu, J. Lore, L. Mu, and M. L. Stowell, An efficient high-order solver for diffusion equations with strong anisotropy on non- anisotropy-aligned meshes, SIAM Journal on Scientific Computing 46 (2024), no. 2, S199–S222

    2024

  8. [16]

    Greif and Y

    C. Greif and Y . He,A closed-form multigrid smoothing factor for an additive Vanka-type smoother applied to the poisson equation, Numerical Linear Algebra with Applications 30 (2023), no. 5, e2500

    2023

  9. [17]

    R. D. Guy, B. Philip, and B. E. Griffith, Geometric multigrid for an implicit-time immersed boundary method , Advances in Computational Mathematics 41 (2015), 635–662

    2015

  10. [18]

    Hackbusch, Multi-grid methods and applications, vol

    W. Hackbusch, Multi-grid methods and applications, vol. 4, Springer Berlin, Heidelberg, 2003

    2003

  11. [19]

    M. Hoelzl et al., The JOREK non-linear extended MHD code and applications to large-scale instabilities and their control in magnetically confined fusion plasmas, Nuclear Fusion 61 (2021), no. 6, 065001

    2021

  12. [20]

    Hofhaus and E

    J. Hofhaus and E. F. Van de Velde,Alternating-direction line-relaxation methods on multicomputers, SIAM Journal on Scientific Computing 17 (1996), no. 2, 454–478

    1996

  13. [21]

    C. T. Kelley, Iterative Methods for Linear and Nonlinear Equations, Society for Industrial and Applied Mathematics, 1995,

    1995

  14. [22]

    Laakmann, P

    F. Laakmann, P. E. Farrell, and L. Mitchell,An augmented Lagrangian preconditioner for the magnetohydrodynamics equations at high Reynolds and coupling numbers, SIAM Journal on Scientific Computing 44 (2022), no. 4, B1018–B1044

    2022

  15. [23]

    I. M. Llorente and N. D. Melson, Robust multigrid smoothers for three dimensional elliptic equations with strong anisotropies, NASA/CR-1998- 208700, 1998

    1998

  16. [24]

    S. P. MacLachlan and C. W. Oosterlee,Local fourier analysis for multigrid with overlapping smoothers applied to systems of pdes, Numerical Linear Algebra with Applications 18 (2011), no. 4, 751–774

    2011

  17. [25]

    T. A. Manteuffel, J. Ruge, and B. S. Southworth, Nonsymmetric algebraic multigrid based on local approximate ideal restriction (lAIR), SIAM Journal on Scientific Computing 40 (2018), no. 6, A4105–A4130

    2018

  18. [26]

    T. A. Manteuffel, L. N. Olson, J. B. Schroder, and B. S. Southworth, A root-node–based algebraic multigrid method, SIAM Journal on Scientific Computing 39 (2017), no. 5, S723–S756

    2017

  19. [27]

    T. A. Manteuffel, S. Münzenmaier, J. Ruge, and B. Southworth,Nonsymmetric reduction-based algebraic multigrid, SIAM Journal on Scientific Computing 41 (2019), no. 5, S242–S268

    2019

  20. [28]

    W. A. Mulder, A new multigrid approach to convection problems, Journal of Computational Physics 83 (1989), no. 2, 303–323

    1989

  21. [29]

    W. A. Mulder,A note on the use of symmetric line Gauss–Seidel for the steady upwind differenced euler equations, SIAM journal on scientific and statistical computing 11 (1990), no. 2, 389–397

    1990

  22. [30]

    Oliveira, M

    F. Oliveira, M. Pinto, C. Marchi, and L. Araki, Optimized partial semicoarsening multigrid algorithm for heat diffusion problems and anisotropic grids, Applied Mathematical Modelling 36 (2012), no. 10, 4665–4676

    2012

  23. [31]

    C. W. Oosterlee, The convergence of parallel multiblock multigrid methods, Applied numerical mathematics 19 (1995), no. 1-2, 115–128

    1995

  24. [32]

    Philip and T

    B. Philip and T. P. Chartier, Adaptive algebraic smoothers, Journal of Computational and Applied Mathematics 236 (2012), no. 9, 2277–2297. 22 KRZYSIK ET AL

    2012

  25. [33]

    Prieto, R

    M. Prieto, R. Santiago, D. Espadas, I. M. Llorente, and F. Tirado,Parallel multigrid for anisotropic elliptic equations, Journal of Parallel and Distributed Computing 61 (2001), no. 1, 96–114

    2001

  26. [34]

    Reisner, M

    A. Reisner, M. Berndt, J. D. Moulton, and L. N. Olson, Scalable line and plane relaxation in a parallel structured multigrid solver, Parallel Computing 100 (2020), 102705

    2020

  27. [35]

    J. B. Schroder, Smoothed aggregation solvers for anisotropic diffusion, Numerical Linear Algebra with Applications 19 (2012), no. 2, 296–312

    2012

  28. [36]

    A. A. Sivas, B. S. Southworth, and S. Rhebergen, Air algebraic multigrid for a space-time hybridizable discontinuous galerkin discretization of advection (-diffusion), SIAM Journal on Scientific Computing 43 (2021), no. 5, A3393–A3416

    2021

  29. [37]

    Trottenberg, C

    U. Trottenberg, C. W. Oosterlee, and A. Schuller, Multigrid, Academic press, 2001

    2001

  30. [38]

    S. P. Vanka,Block-implicit multigrid solution of Navier-Stokes equations in primitive variables, Journal of Computational Physics 65 (1986), no. 1, 138–158

    1986

  31. [39]

    Virtanen et al., SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python, Nature Methods 17 (2020), 261–272

    P. Virtanen et al., SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python, Nature Methods 17 (2020), 261–272

    2020

  32. [40]

    C. J. V ogl, I. Joseph, and M. Holec,Mesh refinement for anisotropic diffusion in magnetized plasmas, Computers & Mathematics with Applications 145 (2023), 159–174

    2023

  33. [41]

    Wesseling, An introduction to multigrid methods, Wiley, 1992

    P. Wesseling, An introduction to multigrid methods, Wiley, 1992

    1992

  34. [42]

    R. R. Wienands and W. W. Joppich, Practical Fourier analysis for multigrid methods, Chapman & Hall/CRC, Boca Raton, FL, 2005

    2005

  35. [43]

    G. A. Wimmer, B. S. Southworth, and X.-Z. Tang,An accurate supg stabilized continuous galerkin discretization for highly anisotropic heat flux, (in review) (2024)

    2024

  36. [44]

    G. A. Wimmer, B. S. Southworth, T. J. Gregory, and X.-Z. Tang, A fast algebraic multigrid solver and accurate discretization for highly anisotropic heat flux i: Open field lines, SIAM Journal on Scientific Computing 46 (2024), no. 3, A1821–A1849

    2024

  37. [45]

    Yavneh, On red-black SOR smoothing in multigrid, SIAM Journal on Scientific Computing 17 (1996), no

    I. Yavneh, On red-black SOR smoothing in multigrid, SIAM Journal on Scientific Computing 17 (1996), no. 1, 180–192

    1996

  38. [46]

    Yavneh, Coarse-grid correction for nonelliptic and singular perturbation problems, SIAM J

    I. Yavneh, Coarse-grid correction for nonelliptic and singular perturbation problems, SIAM J. Sci. Comput. 19 (1998), no. 5, 1682–1699. APPENDIX A DERIV ATIONS OF 2 × 2 THEORETICAL RESULTS Before proving the lemma below (a copy of Lemma 1 from Section 3.3) we introduce some si...

    1998

  39. [47]

    + O(ϵ2), (B29) Overlapping Schwarz methods are not anisotropy-robust multigrid smoothers 29 and x∗ 0 is a critical point of g0, i.e., g′ 0(x∗

  40. [48]

    = 0. Proof. Evaluating the derivative of g at the critical point gives 0 = g′(x∗) = g′ 0(x∗) + ϵg′ 1(x∗) + O(ϵ2), = g′ 0(x∗ 0 + ϵx∗ 1 + O(ϵ2)) + ϵg′ 1(x∗ 0 + ϵx∗ 1 + O(ϵ2)) + O(ϵ2), = g′ 0(x∗

  41. [49]

    Setting this last expression equal to zero and equating powers of ϵ we get the system of equations: g′ 0(x∗

    + g′ 1(x∗ 0) + O(ϵ2), with the last equality following from the Taylor expansion q(x + δ) = q(x) + δq′(x) + O(δ2) for small δ and continuously differentiable q with bounded derivative. Setting this last expression equal to zero and equating powers of ϵ we get the system of equ...

  42. [50]

    = 0 and x∗ 1g′′ 0 (x∗

  43. [51]

    Evidently, x∗ 0 is a critical point of g0 as claimed

    = 0. Evidently, x∗ 0 is a critical point of g0 as claimed. Evaluating g at the critical point gives g(x∗) = g0(x∗) + ϵg1(x∗) + O(ϵ2), = g0(x∗ 0 + ϵx∗ 1 + O(ϵ2)) + ϵg1(x∗ 0 + ϵx∗ 1 + O(ϵ2)) + O(ϵ2), = g0(x∗

  44. [52]

    Substitutingg′(x∗

    + g1(x∗ 0) + O(ϵ2), with the last equality again following from Taylor expansion. Substitutingg′(x∗

  45. [53]

    B.3 Main theoretical results Now we prove our main theoretical results presented in Section 3.6

    = 0 gives (B29). B.3 Main theoretical results Now we prove our main theoretical results presented in Section 3.6. To help with readability we repeat the theorem statements from Section 3.6. Since the results are rather technical to derive, we provide some supporting numerical ...

  46. [54]

    For any fixed ℓ ∈ N the smoother is not ϵ-robust, with limϵ→0+ µℓ,1 = 1 in the FD case, and limϵ→0+ µℓ,1 ≥ 1 in the FE case

  47. [55]

    A necessary condition for ϵ-robustness of this class is that ℓ increases at least as fast as O(ϵ–1/2) as ϵ → 0+

  48. [56]

    –1 ± s 1 – 4 µ∗ – 1 + O(ϵ2) cϵ # = 1 2

    Consider the FD case only. Let µ∗ ∈ (0, 1) be a smoothing factor for which the O(ϵ2) terms in (B36) are uniformly small with respect to ℓ, i.e., |µ∗ – [1 – ℓ(ℓ + 1)ϵ]| < 1, for all ℓ ∈ N. Then, for the FD discretization this class of smoothers has an ϵ-independent smoothing fa...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.