Pith. sign in

REVIEW 3 major objections 6 minor 44 references

A black-box, multilevel algebraic preconditioning framework for conforming finite elements

T0 review · 3 major / 6 minor · reviewed 2026-07-09 · glm-5.2

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

Pith's one-line read Conforming finite elements yield black-box multilevel solvers via Gram equivalence

desk verdict Theoretical framework is sound and genuinely new; practical black-box claim is undermined by heuristic threshold tuning and high operator complexity. read the letter →

arxiv 2607.07485 v1 pith:I3X2AU4S submitted 2026-07-08 math.NA cs.NA

O. A. Krzysik , B. S. Southworth , G. A. Wimmer This is my paper
classification math.NAcs.NA MSC 65F0865N3065N55
keywords algebraicmultigridGramrepresentationSPSDsplittingconformingfiniteelementsspectralcoarsespacesdomaindecompositionoverlappingSchwarzpreconditioning
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 proves that a sparse symmetric positive definite matrix A admits a cover-local Gram representation A=G^T G if and only if it admits an exact cover-local symmetric positive semidefinite (SPSD) splitting. This equivalence is the bridge that turns conforming finite-element discretizations into a natural source of the Gram structure required by the LS-AMG-DD multilevel solver: standard element-by-element assembly already produces element-local SPSD energy contributions, which can be factored into local Gram blocks and stacked into a global factor G. Because Galerkin coarsening preserves the Gram form (A_c = (GP)^T (GP)), the entire multilevel hierarchy inherits the same structure recursively, making the solver essentially black-box: it needs only the assembled matrix and element-level information, without problem-specific near-nullspace vectors, auxiliary-space constructions, or geometric mesh hierarchies. The paper then demonstrates on three notoriously difficult problem classes—grad-div in H(div), anisotropic biharmonic in H^2, and near-incompressible linear elasticity in vector H^1—that this approach converges in tens of iterations where classical AMG methods fail to converge in 10^5 or more, and achieves errors 2–5 orders of magnitude smaller at matched residual tolerances.

What carries the argument

Theorem 3.2 (equivalence of local Gram and local SPSD splitting), Proposition 5.1 (elementwise Gram stacking), the G-row closure construction for overlapping subdomains (Definition 5.4), and the adaptive eigenvalue threshold τ_cut ∝ max_j mult_ω(j).

What would settle it

A conforming finite-element problem for which element matrices are not SPSD (e.g., due to non-coercive facet terms in DG methods), or a problem class where the adaptive threshold produces operator complexities that grow unboundedly with problem size, would break the black-box claim.

Watch

Extended reading notes

Core claim

The central discovery is the equivalence (Theorem 3.2) between local Gram representations and local SPSD splittings of an SPD matrix, combined with the observation that conforming finite-element assembly is a canonical instance of such a splitting. Every element contributes a local SPSD quadratic energy a_T(u,u), and factoring each element block A_T = G_T^T G_T and stacking the embedded factors G_T R_T yields a global Gram factor G satisfying A = G^T G. This factor is local (each row is supported on a single element), and the Gram structure propagates under Galerkin coarsening, enabling a fully recursive multilevel solver. The paper also shows that the algebraic overlapping subdomains usedby

Load-bearing premise

The practical performance depends on an adaptive eigenvalue cutoff threshold whose proportionality constant is chosen empirically (0.75 or 1.0 depending on the problem) without a principled derivation; if this threshold cannot be controlled, the coarse spaces may grow too large to be practical on harder problems.

Editorial extensions

If this is right

  • Any conforming finite-element discretization whose element matrices are SPSD—including high-order elements, vector-valued problems, and problems in H(div), H(curl), or H^2—can be fed into LS-AMG-DD as a black-box solver without requiring near-nullspace vectors, auxiliary spaces, or geometric coarsening hierarchies.
  • The equivalence between local Gram structure and local SPSD splittings means that any discretization framework producing local SPSD energy contributions (not just conforming FE) is a candidate for the same solver pipeline, including certain structure-preserving finite-difference schemes.
  • The fact that Galerkin coarsening preserves the Gram form means the solver does not face the two-level bottleneck that limits many spectral coarse-space methods; the spectral enrichment mechanism recurses naturally on every level.
  • The observation that G-row closures can differ from assembled-matrix graph closures (Table 1) suggests that standard graph-based AMG aggregation may miss structural information that is visible in the unassembled form, with substantial effects on convergence.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This paper establishes that conforming finite-element discretizations naturally admit element-local Gram representations A = G^T G, which is the structure required by the LS-AMG-DD multilevel preconditioner introduced in the authors' companion work [37]. The central theoretical result (Theorem 3.2) proves that, on a prescribed DOF cover C, a C-local Gram representation exists if and only if A admits a C-local SPSD splitting. Proposition 5.1 specializes this to conforming FE assembly, and Theorem 5.7 shows that the algebraic G-row closures recover geometric element-closure structure under Assumption 5.6. Numerical experiments on H(div) grad-div, H^2 anisotropic hyperdiffusion, and vector H^1 elasticity demonstrate robust convergence where classical AMG methods fail.

Significance. The equivalence in Theorem 3.2 is a clean, self-contained characterization that clarifies when local Gram structure arises and connects it to the well-understood SPSD splitting framework. The specialization to conforming FE (Proposition 5.1) is a constructive observation that makes LS-AMG-DD applicable as a black-box solver without problem-specific near-nullspace vectors or auxiliary-space constructions. The numerical experiments span three distinct PDE classes with systematic parameter sweeps (reaction coefficient, anisotropy ratio, Poisson ratio) and polynomial degree, and the comparisons against RS-AMG and RN-AMG baselines are informative. The Firedrake implementation via broken spaces (Appendix A) provides a reproducible route to constructing G. The paper honestly acknowledges the operator complexity growth limitation in Section 7.

major comments (3)
  1. §2, Eq. (2.10): The adaptive threshold tau_cut ∝ max_j mult_omega(j) is stated as 'motivated' but not rigorously justified. The proportionality constant is varied per problem: 0.75 for H(div) (§6.1) and vector H^1 (§6.3), 1.0 for H^2 (§6.2). The paper itself notes that 'slightly decreasing it can lead to a marked improvement in iteration counts, e.g., by a factor of two.' This per-problem tuning is in tension with the 'essentially black-box' claim in the abstract and title. A more principled derivation, or at minimum a clear default recommendation with sensitivity analysis, would strengthen the practical contribution. This is load-bearing because the threshold directly controls the robustness-complexity trade-off that determines whether the method is usable.
  2. Table 2: Operator complexities (OC) of 4.1–10.8 are very high by AMG standards (typical values 1.2–1.5). The hierarchy is capped at 3 levels with coarsening stopped at 25% nonzero density, making the method effectively a 2–3 level method with expensive coarse levels. While the paper acknowledges this in §7, the claim of being a 'multilevel' solver in the AMG sense needs qualification. The reader needs to understand whether these OC values represent a fundamental limitation of the spectral enrichment approach or an artifact of the current threshold strategy, since this determines whether future work can realistically bring OC into a competitive range.
  3. §6, Figures 4, 7, 10: The AMG baseline comparisons use very small problem sizes (416 DOFs in Fig. 4, 3750 DOFs in Fig. 7, 594 DOFs in Fig. 10), while the LS-AMG-DD results in Figures 3, 6, 9 use much larger systems. The baseline comparisons would be more convincing if conducted at problem sizes comparable to those used for the main LS-AMG-DD results, or if the paper explicitly justified why the small-size comparisons are representative. As stated, the mismatch in scale makes it difficult to assess relative performance fairly.
minor comments (6)
  1. §5.4, Assumption 5.6: The footnote states that Assumption 5.6 'can typically be enforced numerically' by finding an orthogonal matrix U such that the rotated factor satisfies the full-support condition. It would help to briefly indicate how U is found in practice (e.g., is this a search over random rotations, or is there a constructive procedure?), since this is a step in the practical pipeline.
  2. §2, Eq. (2.10): The phrase 'for brevity we do not discuss in more detail here' is unsatisfying given that this threshold is the main free parameter. Even a brief paragraph explaining the intuition behind the proportionality to max_j mult_omega(j) would help the reader.
  3. Table 1 caption: The closure type labeled 'A-closure' in the table rows could be confused with the matrix A. Consider labeling as 'Def. 5.3 (A-closure)' and 'Def. 5.4 (G-row closure)' for clarity.
  4. §6.1: The Nitsche penalty parameter gamma_F = 16*alpha*p^2/h_F is stated, but it would be useful to note whether the SPSD property of the element matrices is verified or assumed to hold for this penalty choice.
  5. References [27], [29], [37] are self-cited and appear to be concurrent or submitted works. The dependency on [27] for the convergence theory and [37] for the LS-AMG-DD construction means the present paper's practical claims rest on results not yet peer-reviewed. This is noted for the editor's awareness; the present paper's own theoretical contributions (Theorem 3.2, Proposition 5.1, Theorem 5.7) are self-contained.
  6. Figure 2 caption: 'Theorem 5.6' should read 'Assumption 5.6' in the sentence beginning 'Theorem 5.6 does not hold automatically...'.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for a careful and constructive report. The referee correctly identifies the central theoretical contribution (Theorem 3.2), the constructive specialization to conforming FE (Proposition 5.1), and the numerical demonstrations across three PDE classes. The three major comments concern: (1) the per-problem tuning of the threshold tau_cut and its tension with the black-box claim, (2) the high operator complexities and whether the method qualifies as multilevel in the AMG sense, and (3) the mismatch in problem sizes between the LS-AMG-DD results and the AMG baseline comparisons. We address each below. We agree that all three points warrant revisions to the manuscript: we will add a sensitivity analysis and default recommendation for tau_cut, qualify the multilevel terminology and discuss the OC limitation more carefully, and add explicit justification for the baseline problem sizes. No standing objections remain.

read point-by-point responses

  1. Referee: §2, Eq. (2.10): The adaptive threshold tau_cut ∝ max_j mult_omega(j) is stated as 'motivated' but not rigorously justified. The proportionality constant is varied per problem: 0.75 for H(div) (§6.1) and vector H^1 (§6.3), 1.0 for H^2 (§6.2). The paper itself notes that 'slightly decreasing it can lead to a marked improvement in iteration counts, e.g., by a factor of two.' This per-problem tuning is in tension with the 'essentially black-box' claim in the abstract and title. A more principled derivation, or at minimum a clear default recommendation with sensitivity analysis, would strengthen the practical contribution. This is load-bearing because the threshold directly controls the robustness-complexity trade-off that determines whether the method is usable.

    Authors: The referee is correct that the current treatment of tau_cut is insufficiently justified and that the per-problem variation of the proportionality constant is in tension with the black-box framing. We accept this point and will revise the manuscript accordingly. Specifically, we will: (1) provide a sensitivity analysis showing how iteration counts and operator complexities vary as the proportionality constant ranges over, e.g., {0.5, 0.75, 1.0, 1.25, 1.5} for each problem class, demonstrating that performance is not highly sensitive to this parameter; (2) recommend a single default value (we propose 0.75) with a clear statement that this default was used for two of the three problem classes without modification and yields comparable results for the third; (3) soften the 'essentially black-box' language to clarify that the threshold is a single scalar parameter with a recommended default, analogous to strength-of-connection thresholds in classical AMG, rather than requiring problem-specific tuning. We note that the convergence theory in [27] already establishes that the convergence rate is controlled by tau_max (the realized largest discarded eigenvalue), with tau_max <= tau_cut, so the threshold has a theoretical grounding through the weak approximation property; what is missing is the explicit connection between the proportionality constant and tau_max, which we will add. We agree that a fully principled, parameter-free derivation of the optimal constant would be stronger, but this is beyond the scope of the current paper and is a natural topic for future work. revision: yes

  2. Referee: Table 2: Operator complexities (OC) of 4.1–10.8 are very high by AMG standards (typical values 1.2–1.5). The hierarchy is capped at 3 levels with coarsening stopped at 25% nonzero density, making the method effectively a 2–3 level method with expensive coarse levels. While the paper acknowledges this in §7, the claim of being a 'multilevel' solver in the AMG sense needs qualification. The reader needs to understand whether these OC values represent a fundamental limitation of the spectral enrichment approach or an artifact of the current threshold strategy, since this determines whether future work can realistically bring OC into a competitive range.

    Authors: The referee raises a valid and important point. We agree that the OC values are high by standard AMG standards and that the 'multilevel' terminology requires qualification. We will revise the manuscript to: (1) explicitly qualify the multilevel claim, noting that the current implementation is effectively a 2–3 level method and that the coarsest levels are expensive; (2) discuss more transparently whether the high OC is fundamental or an artifact. On this latter point, we believe it is a combination of both: the spectral enrichment that gives robustness inherently increases coarse-space dimension (this is a fundamental trade-off shared by all spectral coarse-space methods, including GenEO-type constructions), but the current threshold strategy and the lack of aggressive coarsening on deeper levels are artifacts that can be improved. The 25% nonzero density cap is a practical stopping criterion, not a fundamental limitation. We note that the grid complexities (GC) in Table 2 are 1.1–1.3, which are competitive; the high OC is driven by fill-in on the coarse levels, not by an excessively large number of coarse DOFs per se. Future work on sparsification of coarse operators and more aggressive coarsening strategies (e.g., reducing the spectral threshold on deeper levels) could plausibly bring OC into a more competitive range, though we cannot guarantee this at present. We will add this discussion to Section 7 rather than overclaiming in the abstract and title. revision: yes

  3. Referee: §6, Figures 4, 7, 10: The AMG baseline comparisons use very small problem sizes (416 DOFs in Fig. 4, 3750 DOFs in Fig. 7, 594 DOFs in Fig. 10), while the LS-AMG-DD results in Figures 3, 6, 9 use much larger systems. The baseline comparisons would be more convincing if conducted at problem sizes comparable to those used for the main LS-AMG-DD results, or if the paper explicitly justified why the small-size comparisons are representative. As stated, the mismatch in scale makes it difficult to assess relative performance fairly.

    Authors: The referee is correct that the mismatch in problem sizes between the baseline comparisons and the main LS-AMG-DD results is not adequately justified. We will address this in two ways. First, we will add explicit justification: the purpose of the baseline comparisons is to demonstrate that the chosen problem classes are genuinely hard for standard black-box AMG, not to provide a head-to-head performance benchmark at matched sizes. The baselines already fail dramatically (10^4–10^6 iterations) at the small sizes shown, and their convergence degrades with mesh refinement, so running them at larger sizes would only show worse failure. Second, we will add at least one baseline comparison at a problem size comparable to the smallest LS-AMG-DD result for each problem class, to demonstrate that the failure persists at larger scales. We note that the RS-AMG and RN-AMG baselines were run as two-level methods (as stated in the text), and their failure at small sizes with degradation under refinement is representative of their behavior at larger sizes. However, we agree that making this explicit in the manuscript, rather than leaving it implicit, is necessary for the reader to fairly assess the comparison. revision: yes

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; theoretical core is self-contained with minor self-citations for methodology.

full rationale

The paper's central theoretical results—Theorem 3.2 (equivalence of C-local Gram and C-local SPSD splitting), Proposition 5.1 (elementwise Gram from conforming assembly), and Theorem 5.7 (G-row closure equals element closure under Assumption 5.6)—are all proven from first principles within the paper. Theorem 3.2's proof is a clean regrouping argument: direction 1 factors each SPSD block and stacks; direction 2 assigns each row outer product to a cover member and reassembles. Neither direction reduces to its conclusion by definition. Proposition 5.1 is an explicit specialization of Theorem 3.2 to the standard FE assembly A = Σ R_T^T A_T R_T, and the construction G = stack(G_T R_T) is presented as a construction, not a prediction. The self-citations [37] (LS-AMG-DD solver) and [27] (convergence theory) are methodological tools applied in this paper, not results whose validity circularly depends on the present paper's claims. The convergence bound from [27] (K_TG ≤ λ_max(...) τ_max) is used to motivate the heuristic threshold τ_cut ∝ max_j mult_ω(j), but the paper explicitly acknowledges this as 'motivated' rather than rigorously derived, and varies the proportionality constant per problem. This is a heuristic parameter choice, not a circular derivation. The SPSD splitting (2.7) is constructed directly from G by definition of A = G^T G, but this is stated as a construction (Remark 3.4 confirms it instantiates Lemma 3.3's framework), not claimed as a derived prediction. No step in the derivation chain reduces to its inputs by construction in a way that would constitute circularity. The minor self-citations are standard methodological references and do not undermine the independent content of the theoretical framework presented here.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

No new physical entities, particles, forces, or dimensions are introduced. The framework operates entirely within standard linear algebra and finite-element methodology. The free parameters are practical tuning constants for the solver, not fundamental postulates. The axioms are standard domain assumptions (SPD assembly, Nitsche boundary conditions) or a mild technical condition (Assumption 5.6) that is shown to be enforceable.

free parameters (2)
  • tau_cut proportionality constant = 0.75 or 1.0 (varies by problem)
    The adaptive local eigenvalue threshold tau_cut = c * max_j mult_omega(j) uses an order-one constant c that is varied between problems (0.75 for H(div) and elasticity, 1.0 for H^2) without a principled derivation.
  • maximum number of levels = 3
    Capped at three levels; coarsening stops below 10 DOFs or above 25% nonzero density. These are practical heuristics, not derived from theory.
assumptions (4)
  • domain assumption A is SPD and admits the assembly A = sum_T R_T^T A_T R_T with A_T >= 0
    Section 4, Eq. (4.1) and Section 5, Eq. (5.1). This holds for conforming FE discretizations of coercive problems but not for all discretizations (e.g., DG interior penalty).
  • ad hoc to paper Assumption 5.6 (full-support local row per element): each element's local Gram factor has at least one row nonzero in every local DOF
    Section 5.4, Assumption 5.6. Needed for Theorem 5.7 (G-row closure = element closure). Authors note it can typically be enforced numerically via orthogonal transformation but do not prove it always holds.
  • standard math Convergence theory of LS-AMG-DD from companion paper [27]
    Section 2 cites [27] for the two-level convergence bound K_TG <= lambda_max(M^{-1}_omega J fM) * tau_max. The present paper relies on this result but does not reprove it.
  • domain assumption Essential boundary conditions imposed weakly via symmetric Nitsche method with sufficient penalty
    Section 5.1. Required for local SPSD structure to be preserved on boundary elements. Strong imposition requires extra bookkeeping not implemented.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A black-box, multilevel algebraic preconditioning framework for conforming finite elements." pith.science (2026). https://pith.science/paper/I3X2AU4S

@misc{pith2026260707485,
  author       = {Pith},
  title        = {Pith review of: A black-box, multilevel algebraic preconditioning framework for conforming finite elements},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I3X2AU4S}},
  note         = {Machine review of arXiv:2607.07485}
}
abstract

Recently we introduced the least-squares algebraic-multigrid domain-decomposition (LS-AMG-DD) method as a multilevel, algebraic preconditioner for sparse symmetric positive definite (SPD) matrices that admit a Gram representation \(A=G^{\top}G\) \cite{southworth2026lsamgdd}. The factor \(G\) induces a local symmetric positive semidefinite (SPSD) splitting of \(A\) used to define local spectral problems from which an interpolation $P$ is built, and a coarse-level Gram operator induced under Galerkin coarsening, \(A_c=G_c^\top G_c\), for \(G_c:=GP\). This paper clarifies when this Gram structure arises, showing that, on a prescribed degree-of-freedom cover \({\cal C}\), a \({\cal C}\)-local Gram representation of $A$ exists if and only if \(A\) admits a \({\cal C}\)-local SPSD splitting. We then connect this viewpoint to conforming finite-element discretizations, where bilinear forms are naturally assembled from elementwise SPSD energies and therefore admit element-local Gram representations after choosing local factors (e.g., via algebraic factorizations of element blocks). Taken together, these observations provide an essentially black-box route for applying LS-AMG-DD to conforming finite-element problems. Numerical tests illustrate the robustness of the method on several problems for which classical AMG methods require more than $10^5$ iterations to converge, including high-order discretizations of grad--div in \(\hdiv\), anisotropic hyperdiffusion in $H^2$, and linear elasticity in vector \(H^1\). Moreover, in some comparisons with existing AMG methods, LS-AMG-DD produces errors that are 2--5 orders of magnitude smaller, even when all methods are stopped at the same relative residual tolerance.

Figures

Figures reproduced from arXiv: 2607.07485 by the authors.

Figure 1
Figure 1. Element-local unassembled versus assembled perspectives in conforming [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Illustration of Theorem 5.7 for a conforming [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Two-dimensional H(div; D) grad–div results for RT and BDM discretizations. The columns correspond to α = 10−3 , 1, 103 . The top row shows iteration counts and the bottom row shows operator complexity [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Comparison of LS-AMG-DD with two baseline AMG solvers for the grad–div problem [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: Solutions of the anisotropic biharmonic equation (6.4) in [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: Results for the H2 problem (6.4) with the two b fields in (6.5). The top row shows iteration counts and the bottom row shows operator complexity. DOF count of 3750 is still significantly smaller than the smallest Bell DOF count of 15606 shown in [PITH_FULL_IMAGE:figur…
Figure 7
Figure 7. Figure 7: Comparison of LS-AMG-DD with the two baseline AMG solvers for the anisotropic [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Displaced cantilever-beam solutions for the vector [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: Results for linear elasticity (6.7) discretized in [PITH_FULL_IMAGE:figures/full_fig_p021_9.png]
Figure 10
Figure 10. Figure 10: Comparison of LS-AMG-DD with the two baseline AMG solvers for the linear [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages

  1. [37]

    B. S. Southworth, H. A. Daas, G. A. Wimmer, and E. Threlfall , Algebraic Multigrid with Overlapping Schwarz Smoothers and Local Spectral Coarse Grids for Least Squares Problems , arXiv preprint arXiv:2601.04112, (2026), https://arxiv.org/abs/2601.04112, https://arxiv.org/abs/2601.04112

    work page arXiv 2026

  2. [27]

    O. A. Krzysik, B. S. Southworth, and H. Al Daas , Two-level convergence of Algebraic Multigrid with Overlapping Smoothers and Spectral Coarse Grids , arXiv preprint arXiv:2606.18404, (2026)

    work page Pith review arXiv 2026

  3. [1]

    Anderson, J

    R. Anderson, J. Andrej, A. Barker, J. Bramwell, J.-S. Camier, J. Cerveny, V. Dobrev, Y. Dudouit, A. Fisher, T. Kolev, et al. , MFEM: A modular finite element methods library , Computers & Mathematics with Applications, 81 (2021), pp. 42--74

    work page 2021

  4. [2]

    D. N. Arnold, R. S. Falk, and R. Winther , Multigrid in H( div ) and H( curl ) , Numerische Mathematik, 85 (2000), pp. 197--217

    work page 2000

  5. [3]

    A. H. Baker, T. V. Kolev, and U. M. Yang , Improving algebraic multigrid interpolation operators for linear elasticity problems , Numerical Linear Algebra with Applications, 17 (2010), pp. 495--517

    work page 2010

  6. [4]

    N. Bell, L. N. Olson, J. Schroder, and B. Southworth , Pyamg: algebraic multigrid solvers in python , Journal of Open Source Software, 8 (2023), p. 5495

    work page 2023

  7. [5]

    P. B. Bochev, C. J. Garasi, J. J. Hu, A. C. Robinson, and R. S. Tuminaro , An improved algebraic multigrid method for solving maxwell's equations , SIAM Journal on Scientific Computing, 25 (2003), pp. 623--642

    work page 2003

  8. [6]

    Brezina, A

    M. Brezina, A. J. Cleary, R. D. Falgout, V. E. Henson, J. E. Jones, T. A. Manteuffel, S. F. McCormick, and J. W. Ruge , Algebraic Multigrid Based on Element Interpolation (AMGe) , SIAM Journal on Scientific Computing, 22 (2001), pp. 1570--1592, https://doi.org/10.1137/S1064827598344303

    work page doi:10.1137/s1064827598344303 2001

Show all 44 references
  1. [7]

    Chartier, R

    T. Chartier, R. D. Falgout, V. Henson, J. Jones, T. Manteuffel, S. McCormick, J. Ruge, and P. S. Vassilevski , Spectral amge ( amge) , SIAM Journal on Scientific Computing, 25 (2003), pp. 1--26

    2003

  2. [8]

    Claus , Multigrid smoothers for saddle point systems , PhD thesis, University of Wuppertal, 2019

    L. Claus , Multigrid smoothers for saddle point systems , PhD thesis, University of Wuppertal, 2019

    2019

  3. [9]

    Clees , AMG strategies for PDE systems with applications in industrial semiconductor simulation , Shaker Herzogenrath, Germany, 2005

    T. Clees , AMG strategies for PDE systems with applications in industrial semiconductor simulation , Shaker Herzogenrath, Germany, 2005

    2005

  4. [10]

    L. B. da Veiga, K. Lipnikov, and G. Manzini , The Mimetic Finite Difference Method for Elliptic Problems , MS&A, Springer International Publishing, Cham, 2014, https://doi.org/10.1007/978-3-319-02663-3

    2014 doi

  5. [11]

    H. A. Daas and L. Grigori , A Class of Efficient Locally Constructed Preconditioners Based on Coarse Spaces , SIAM Journal on Matrix Analysis and Applications, 40 (2019), pp. 66--91, https://doi.org/10.1137/18M1194365

    2019 doi

  6. [12]

    H. A. Daas, L. Grigori, P. Jolivet, and P.-H. Tournier , A Multilevel Schwarz Preconditioner Based on a Hierarchy of Robust Coarse Spaces , SIAM Journal on Scientific Computing, 43 (2021), pp. A1907--A1928, https://doi.org/10.1137/19M1266964

    2021 doi

  7. [13]

    H. A. Daas, P. Jolivet, and J. A. Scott , A Robust Algebraic Domain Decomposition Preconditioner for Sparse Normal Equations , SIAM Journal on Scientific Computing, 44 (2022), pp. A1047--A1068, https://doi.org/10.1137/21M1434891

    2022 doi

  8. [14]

    Dobrev, T

    V. Dobrev, T. Kolev, C. S. Lee, V. Tomov, and P. S. Vassilevski , Algebraic hybridization and static condensation with application to scalable h (div) preconditioning , SIAM Journal on Scientific Computing, 41 (2019), pp. B425--B447

    2019

  9. [15]

    Efendiev, J

    Y. Efendiev, J. Galvis, and T. Y. Hou , Generalized Multiscale Finite Element Methods ( GMsFEM ) , Journal of Computational Physics, 251 (2013), pp. 116--135, https://doi.org/10.1016/j.jcp.2013.04.045

    2013 doi

  10. [16]

    R. D. Falgout, P. S. Vassilevski, and L. T. Zikatanov , On Two-Grid Convergence Estimates , Numerical Linear Algebra with Applications, 12 (2005), pp. 471--494, https://doi.org/10.1002/nla.437

    2005 doi

  11. [17]

    P. E. Farrell, M. G. Knepley, L. Mitchell, and F. Wechsung , Pcpatch: software for the topological construction of multigrid relaxation methods , ACM Transactions on Mathematical Software (TOMS), 47 (2021), pp. 1--22

    2021

  12. [18]

    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), pp. S199--S222

    2024

  13. [19]

    D. A. Ham, P. H. J. Kelly, L. Mitchell, C. J. Cotter, R. C. Kirby, K. Sagiyama, N. Bouziani, S. Vorderwuelbecke, T. J. Gregory, J. Betteridge, D. R. Shapero, R. W. Nixon-Hill, C. J. Ward, P. E. Farrell, P. D. Brubeck, I. Marsden, T. H. Gibson, M. Homolya, T. Sun, A. T. T. McRa...

    2023 doi

  14. [20]

    Hiptmair , Multigrid method for H( div ) in three dimensions , Electronic Transactions on Numerical Analysis, 6 (1997), pp

    R. Hiptmair , Multigrid method for H( div ) in three dimensions , Electronic Transactions on Numerical Analysis, 6 (1997), pp. 133--152

    1997

  15. [21]

    Hiptmair , Multigrid method for maxwell's equations , SIAM Journal on Numerical Analysis, 36 (1998), pp

    R. Hiptmair , Multigrid method for maxwell's equations , SIAM Journal on Numerical Analysis, 36 (1998), pp. 204--225

    1998

  16. [22]

    Hiptmair and J

    R. Hiptmair and J. Xu , Nodal auxiliary space preconditioning in H ( curl ) and H ( div ) spaces , SIAM Journal on Numerical Analysis, 45 (2007), pp. 2483--2509, https://doi.org/10.1137/060660588

    2007 doi

  17. [23]

    J. J. Hu, R. S. Tuminaro, P. B. Bochev, C. J. Garasi, and A. C. Robinson , Toward an h-independent algebraic multigrid method for maxwell's equations , SIAM Journal on Scientific Computing, 27 (2006), pp. 1669--1688

    2006

  18. [24]

    R. C. Kirby and L. Mitchell , Code generation for generally mapped finite elements , ACM Transactions on Mathematical Software (TOMS), 45 (2019), pp. 1--23

    2019

  19. [25]

    T. V. Kolev and P. S. Vassilevski , Parallel auxiliary space amg for H ( curl ) problems , Journal of Computational Mathematics, 27 (2009), pp. 604--623, https://doi.org/10.4208/jcm.2009.27.5.013

    2009 doi

  20. [26]

    T. V. Kolev and P. S. Vassilevski , Parallel auxiliary space amg solver for H ( div ) problems , SIAM Journal on Scientific Computing, 34 (2012), pp. A3079--A3098, https://doi.org/10.1137/110859361

    2012 doi

  21. [28]

    O. A. Krzysik, B. S. Southworth, and B. Philip , Overlapping schwarz methods are not anisotropy-robust multigrid smoothers , Numerical Linear Algebra with Applications, 33 (2026), p. e70070

    2026

  22. [29]

    O. A. Krzysik, B. S. Southworth, and G. Wimmer , Broken-space Additive Schwarz Mass Inverse Approximations and (Block) Preconditioning . Submited, 2026

    2026

  23. [30]

    Magri, N

    V. Magri, N. Castelletto, D. Osei-Kuffuor, R. Settgast, M. Cusini, W. Tobin, R. Aronson, and P. Tomin , A multigrid reduction framework for efficient and scalable multiphysics simulations , in SPE Reservoir Simulation Conference, SPE, 2025, p. D021S009R002

    2025

  24. [31]

    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), pp. S723--S756

    2017

  25. [32]

    Morel, R

    J. Morel, R. M. Roberts, and M. J. Shashkov , A local support-operators diffusion discretization scheme for quadrilateral r-z meshes , Journal of Computational Physics, 144 (1998), pp. 17--51

    1998

  26. [33]

    Pazner , Efficient low-order refined preconditioners for high-order matrix-free continuous and discontinuous galerkin methods , SIAM Journal on Scientific Computing, 42 (2020), pp

    W. Pazner , Efficient low-order refined preconditioners for high-order matrix-free continuous and discontinuous galerkin methods , SIAM Journal on Scientific Computing, 42 (2020), pp. A3055--A3083

    2020

  27. [34]

    Reitzinger and J

    S. Reitzinger and J. Sch \"o berl , An algebraic multigrid method for finite element discretizations with edge elements , Numerical linear algebra with applications, 9 (2002), pp. 223--238

    2002

  28. [35]

    E. M. R nquist and A. T. Patera , Spectral element multigrid. i. formulation and numerical results , Journal of Scientific Computing, 2 (1987), pp. 389--406

    1987

  29. [36]

    T. Shen, J. Brannick, R. Falgout, K. Kahl, and J. Schroder , Nodal coarsening and sparse ideal interpolation for h (curl) problems in algebraic multigrid , arXiv preprint arXiv:2602.23613, (2026)

    2026

  30. [38]

    Spillane, V

    N. Spillane, V. Dolean, P. Hauret, F. Nataf, C. Pechstein, and R. Scheichl , Abstract Robust Coarse Spaces for Systems of PDEs via Generalized Eigenproblems in the Overlaps , Numerische Mathematik, 126 (2014), pp. 741--770, https://doi.org/10.1007/s00211-013-0576-y

    2014 doi

  31. [39]

    Stenberg , On some techniques for approximating boundary conditions in the finite element method , Journal of Computational and applied Mathematics, 63 (1995), pp

    R. Stenberg , On some techniques for approximating boundary conditions in the finite element method , Journal of Computational and applied Mathematics, 63 (1995), pp. 139--148

    1995

  32. [40]

    Tuminaro and C

    R. Tuminaro and C. Glusa , A structure preserving h-curl algebraic multigrid method for the eddy current equations , arXiv preprint arXiv:2506.08284, (2025)

    2025 arXiv

  33. [41]

    Van e k, J

    P. Van e k, J. Mandel, and M. Brezina , Algebraic Multigrid by Smoothed Aggregation for Second and Fourth Order Elliptic Problems , Computing, 56 (1996), pp. 179--196, https://doi.org/10.1007/BF02238511

    1996 doi

  34. [42]

    Van e k, J

    P. Van e k, J. Mandel, and M. Brezina , Algebraic multigrid by smoothed aggregation for second and fourth order elliptic problems , Computing, 56 (1996), pp. 179--196

    1996

  35. [43]

    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), pp. A1821--A1849

    2024

  36. [44]

    Xu and L

    J. Xu and L. Zikatanov , Algebraic Multigrid Methods , Acta Numerica, 26 (2017), pp. 591--721, https://doi.org/10.1017/S0962492917000083

    2017 doi

Pith tools

Reviewed July 9, 2026 · model on record in the stance chip above.