Pith. sign in

REVIEW 2 major objections 7 minor 107 references

A geometric multigrid V-cycle that never assembles vectors or hanging-node constraints is algebraically identical to classical local multigrid.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-12 02:39 UTC pith:MFQILFJY

load-bearing objection Solid architectural result: cell-wise unassembled residuals + edge masks make hanging-node multigrid algebraically identical to Janssen–Kanschat without ever building constraints, with clean proofs and competitive A100 numbers. the 2 major comments →

arxiv 2607.03413 v1 pith:MFQILFJY submitted 2026-07-03 math.NA cs.NA

Coalesced Matrix-Free Geometric Multigrid on Persistent Cell-Wise Storage

Micha{\l} Wichrowski This is my paper
classification math.NA cs.NA MSC 65N5565N3065F0865Y1065Y20
keywords geometric multigridadaptive mesh refinementhanging nodesmatrix-freefinite elementsGPUcell-wise storagelocal smoothing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper shows that if every field in a high-order finite-element solver is kept in a redundant, cell-local representation, the usual obstacles of adaptive multigrid simply disappear. Hanging-node constraints need never be built: plain tensor-product transfers applied to the raw, unassembled residual already reproduce the classical constrained restriction, including the action of the transposed constraint matrix. Edge operators of local smoothing reduce to a pointwise mask on the residual, so the level operator never has to be split into interior and edge blocks. The single inter-cell primitive that remains is one structured direct-stiffness summation inside the smoother; that kernel can stay topologically oblivious even on arbitrarily refined meshes. The author proves that the resulting cell-wise V-cycle produces exactly the same iterates as the classical local multigrid method of Janssen and Kanschat, and therefore inherits its level-independent convergence theory. Numerical tests for the Laplace equation confirm that convergence is essentially unaffected by hanging nodes, and that a masked point-Jacobi smoother already delivers GPU throughput competitive with more elaborate patch smoothers.

Core claim

When fields live permanently in redundant cell-wise storage, hanging-node constraints and edge-operator splitting are unnecessary: plain geometric transfers on the unassembled residual, together with a pointwise residual mask, already realize the classical constrained local multigrid V-cycle. The cell-wise algorithm is therefore algebraically identical, iterate by iterate, to the classical method and inherits its convergence theory without any new spectral analysis.

What carries the argument

Three intertwining identities (prolongation commuting with gather, unassembled residual restriction reproducing the transposed constraint matrix, and smoother equivalence under masking) that, by induction on levels, make the cell-wise V-cycle identical to the classical local multigrid V-cycle.

Load-bearing premise

The cell-wise smoother, when fed a continuous iterate and an unassembled residual, must produce a continuous correction that matches exactly what the classical local smoother would produce on the assembled system; if that matching fails, the inheritance of classical convergence theory collapses.

What would settle it

On a sequence of adaptively refined meshes with hanging nodes, run both the cell-wise V-cycle and a classical local multigrid code with the same smoother and check whether the iteration matrices (or successive residual vectors) differ by more than round-off; any systematic discrepancy falsifies the claimed equivalence.

Watch this falsifier — get emailed when new claim-graph text bears on it.

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

2 major / 7 minor

Summary. The paper constructs a geometric multigrid V-cycle for high-order continuous FEM that never forms assembled global vectors, working entirely in the redundant cell-wise storage of the companion framework [37]. The central claim is architectural and algebraic: hanging-node constraints are never assembled; plain tensor-product transfers applied to the unassembled residual reproduce the classical constrained restriction (including the transposed constraint matrix); and local-smoothing edge operators reduce to pointwise residual masking. Equivalence, iterate by iterate, to the classical local multigrid method of Janssen and Kanschat [18] is proved by induction (Theorem 2) from intertwining lemmas for prolongation, unassembled restriction, residual transfer under property (20) of CHN, and a smoother Assumption 1 verified for masked damped Jacobi. As a consequence the structured, topologically oblivious DSS kernel can run unchanged on adaptive level meshes. Numerical experiments for the Laplacian (cube and curved ball; uniform and adaptive) show grid-independent contraction essentially unaffected by hanging nodes, with up to 1.1 GDoF/s per V-cycle and end-to-end solve throughput on par with a recent patch-smoother code on a single A100 in fp64.

Significance. If the equivalence and the adaptive DSS argument hold, the paper removes a genuine implementation bottleneck for matrix-free adaptive multigrid on GPUs: constraint matrices, edge/interior operator splits, and topology-dependent residual weights. The contribution is cleanly scoped as architectural rather than spectral (p-robustness is not claimed; Jacobi is the studied smoother), and the inheritance of [18] via exact operator identities is the right way to transfer convergence theory. Strengths include explicit intertwining lemmas, an induction proof of Theorem 2, verification of Assumption 1 for the Jacobi construction (Lemma 5), and adaptive experiments that keep iteration counts flat in refinement depth. The observation that the same edge mask required by local smoothing also licenses a topologically oblivious structured DSS kernel is a useful conceptual point. Competitive throughput with only a masked point-Jacobi smoother, against a patch-smoother baseline driven to a looser tolerance, makes the engineering claim concrete.

major comments (2)
  1. §4.3.1 and Algorithm 9: The argument that incorrect structured-DSS sums on hanging interfaces are harmless is written for residual-side DSS (Algorithm 6): dual mask zeros E_ℓ before DSS, and the correction on E_ℓ is claimed to vanish. The performance-critical realization is the deferred-assembly sweep (Algorithm 9 / identity (32)), which applies averaging DSS S̄ to the updated iterate. If the cascade writes corrupted values onto E_ℓ when exchanging toward an empty pocket (Figure 2: “stale or zero”), S̄ can alter edge DoFs that local smoothing must leave equal to the prolonged coarse field. Please spell out why Algorithm 9 preserves E_ℓ values under the same incorrect cascade—e.g., explicit re-mask after S̄, pocket contents, or a proof that S̄δ vanishes on E_ℓ whenever δ does—so that the central “topologically oblivious kernel on adaptive meshes” claim covers the implemented smoother, not
  2. §3.2.3 Assumption 1 and §4: Theorem 2 imports the full convergence theory of [18] only through smoother intertwining. Lemma 5 verifies this for the abstract masked Jacobi (26) assuming the algebraic DSS form (28). The manuscript should state as a precise hypothesis what is taken from the companion [37] (that the implemented structured/unstructured cascade realizes (28) and the averaging form (30) on interior interfaces of each level mesh after edge masking) versus what is proved here. Without that boundary, the load-bearing external condition for both equivalence and the adaptive performance claim is harder to audit than the transfer lemmas.
minor comments (7)
  1. §6 / Table 1 vs [11]: The comparison is useful but should state more prominently in the table caption or main text that [11] stops at residual reduction 10^{-9} while this work uses energy-norm reduction 10^{-14}, so the near-parity already favors the present solver on tolerance; the current discussion in the prose is easy to miss.
  2. §6: The fixed under-relaxation (ω=0.7 cube, 0.6 ball) and the exclusion of p=1 because that single choice is “mildly too large” should be flagged earlier when the smoother is introduced (§3.3), not only in the experiments section.
  3. References [22] and [23] appear to be duplicate entries of the same Kronbichler–Sashko–Munch paper; please deduplicate.
  4. Figure 4: The caption correctly notes that the energy measure vanishes at iteration zero for u=0; consider starting the solid curves at cycle 1 in the plot itself to avoid a visual gap that readers may misread as missing data.
  5. Notation: V_cell vs V_{cell,ℓ} vs V^{act}_{cell,ℓ} and the several inclusion maps (ι_ℓ, ι_{Sℓ}, ι^*†_ℓ) are introduced densely in §3.2; a short notation table or a one-paragraph recap before Algorithm 4 would help readers who skip the companion paper.
  6. Author line: “Micha l Wichrowski” appears to have a spurious space ( Michał ).
  7. §5 “Detection of the refinement edge”: the one-time DSS-based edge detection is elegant; a sentence on cost relative to a single V-cycle would reassure readers that setup is negligible for the reported solve throughputs.

Circularity Check

0 steps flagged

No significant circularity: the cell-wise/classical equivalence is an algebraic intertwining proof against the external local multigrid of Janssen–Kanschat [18]; companion [37] is only the storage/DSS substrate.

full rationale

The load-bearing claim (Theorem 2) is that the cell-wise V-cycle of Algorithm 5 is iterate-identical to the classical local multigrid of Janssen and Kanschat once Assumption 1 holds. The proof is by induction on level, using three intertwining identities established in this paper: prolongation commutativity (Lemma 1 / Eq. 10), unassembled residual restriction (Lemma 2 / Eq. 11 and Lemma 4 / Eq. 21, which recovers the transposed hanging-node constraint without ever forming it), and smoother equivalence (Assumption 1, verified for masked damped Jacobi in Lemma 5). Continuity of the prolonged field across refinement edges (Lemma 3) supplies the only property of C_HN that is used (Eq. 20). Convergence estimates are then imported from the external reference [18], not re-derived or fitted. The free parameter ω is an ordinary under-relaxation choice, not a fit that forces a later prediction. The sole self-citation of the companion paper [37] supplies the cell-wise storage format and the algebraic form of DSS (S = G I_Riesz G^T and its averaging variant); that is infrastructure reuse, not a uniqueness theorem or ansatz that forces the multigrid claim. No step reduces a claimed prediction to its own input by construction, so the circularity score is at most 1.

Axiom & Free-Parameter Ledger

1 free parameters · 5 axioms · 0 invented entities

The paper's load-bearing content is an algebraic reformulation of classical local multigrid in cell-wise storage. It imports multilevel subspace-correction convergence from Janssen–Kanschat and the cell-wise primal-dual/DSS substrate from the companion paper. The only free numerical knobs in the experiments are fixed under-relaxation factors. No new physical entities are postulated.

free parameters (1)
  • Jacobi under-relaxation ω = 0.7 (cube), 0.6 (ball)
    Fixed once per geometry (ω=0.7 cube, ω=0.6 ball), not tuned per degree; mildly too large for p=1 so that degree is excluded. Affects contraction rates and reported iteration counts.
axioms (5)
  • domain assumption Classical local multigrid of Janssen and Kanschat [18] converges level-independently under standard smoother hypotheses (e.g. ω ≲ 1/λmax(D^{-1}A) for damped Jacobi).
    Theorem 2 and Corollary 1 transfer convergence by equivalence rather than re-deriving spectral bounds; the paper explicitly imports this theory.
  • domain assumption Cell-wise storage, gather G, unassembled dual residuals, pairing identity (3), and structured DSS of companion paper [37] are correct.
    Entire construction lives in that framework; Section 2 defers proofs to [37].
  • domain assumption Hanging-node edge-constraint operator CHN satisfies CHN P_CG = P_CG on prolonged continuous coarse fields (property (20)).
    Lemma 4 uses only this property to equate unassembled cell-wise restriction with classical constrained residual transfer.
  • domain assumption Nested hierarchical meshes with isotropic 2^d refinement and continuous tensor-product Qp elements on quads/hexes.
    Standard geometric multigrid setting stated in Sections 2–3; transfers are Kronecker products of 1D prolongations.
  • standard math Standard finite-element duality, adjoints of prolongation as restriction, and multilevel subspace-correction framework [4,5].
    Used throughout Lemmas 1–2 and the induction in Theorem 2.

pith-pipeline@v1.1.0-grok45 · 32696 in / 3197 out tokens · 28906 ms · 2026-07-12T02:39:18.674865+00:00 · methodology

0 comments
read the original abstract

We present a geometric multigrid preconditioner for high-order continuous finite elements that operates entirely on redundant, cell-wise stored vectors: the assembled global vector is never formed on any level of the hierarchy. In this storage paradigm the machinery that classically complicates adaptive multigrid dissolves. Hanging-node constraints are never assembled: we prove that the plain tensor-product transfer operators, applied to the \emph{unassembled} residual, algebraically reproduce the classical constrained restriction, including the action of the transposed constraint matrix, and the edge operators of local smoothing reduce to a pointwise masking of the residual, with no splitting of the level operator into interior and edge blocks. As a consequence, the single inter-cell primitive of the whole V-cycle can use a topologically oblivious structured kernel even on adaptively refined meshes. We prove that the resulting cell-wise V-cycle is equivalent, iterate by iterate, to the classical local multigrid method, and therefore inherits its convergence theory. Numerical experiments for the Laplace operator confirm grid-independent convergence that is essentially unaffected by local refinement; on a single GPU, using nothing more than a masked point-Jacobi smoother, the solver sustains up to $1.1$\, GDoF/s per V-cycle in double precision and reaches end-to-end solve throughput on par with patch-smoother-based solvers.

Figures

Figures reproduced from arXiv: 2607.03413 by Micha{\l} Wichrowski.

Figure 1
Figure 1. Figure 1: Level-wise decomposition of a locally refined active mesh. Left: the composed active [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Interaction of the structured DSS cascade with an adaptively refined region, shown on [PITH_FULL_IMAGE:figures/full_fig_p021_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Two 2D examples of test geometries with 4 adaptive refinement toward the domain [PITH_FULL_IMAGE:figures/full_fig_p025_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: V-cycle convergence of the DSS multigrid solver, MG only (A100 80 GB SXM, [PITH_FULL_IMAGE:figures/full_fig_p026_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Roofline of the multigrid building blocks on the A100 80 GB SXM ( [PITH_FULL_IMAGE:figures/full_fig_p029_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

107 extracted references · 4 linked inside Pith

  1. [1]

    arXiv preprint arXiv:2607.02335 , year=

    Coalesced Matrix-Free Finite Elements in Cell-Wise Storage , author=. arXiv preprint arXiv:2607.02335 , year=

    Pith/arXiv arXiv

  2. [2]

    arXiv preprint , year =

    Coalesced Matrix-Free Geometric Multigrid on Persistent Cell-Wise Storage , author =. arXiv preprint , year =

  3. [3]

    and Falk, Richard S

    Arnold, Douglas N. and Falk, Richard S. and Winther, R. , TITLE =. Math. Comput. , VOLUME =. 1997 , NUMBER =

    1997

  4. [4]

    and Falk, Richard S

    Arnold, Douglas N. and Falk, Richard S. and Winther, Ragnar , TITLE =. Numer. Math. , FJOURNAL =

  5. [5]

    Multigrid Methods , author =

  6. [6]

    p4est: Scalable algorithms for parallel adaptive mesh refinement on forests of octrees , author=. SIAM J. Sci. Comput. , volume=. 2011 , publisher=

    2011

  7. [7]

    ACM Trans

    Algorithms and data structures for massively parallel generic adaptive finite element codes , author=. ACM Trans. Math. Softw. , volume=. 2012 , publisher=

    2012

  8. [8]

    Daniel Arndt and Wolfgang Bangerth and Denis Davydov and Timo Heister and Luca Heltai and Martin Kronbichler and Matthias Maier and Jean-Paul Pelteret and Bruno Turcksin and David Wells , journal =. The. 2021 , DOI =

    2021

  9. [9]

    Daniel Arndt and Wolfgang Bangerth and Maximilian Bergbauer and Bruno Blais and Marc Fehling and Rene Gassmöller and Timo Heister and Luca Heltai and Martin Kronbichler and Matthias Maier and Peter Munch and Sam Scheuerman and Bruno Turcksin and Siarhei Uzunbajakau and David Wells and Michał Wichrowski , journal =. The. doi:doi:10.1515/jnma-2025-0115 , year =

    doi:10.1515/jnma-2025-0115 2025

  10. [10]

    and Farrell, Patrick E

    Brubeck, Pablo D. and Farrell, Patrick E. , title =. SIAM J. Sci. Comput. , volume =. 2022 , doi =

    2022

  11. [11]

    The Finite Element Method for Elliptic Problems , author =

  12. [12]

    Multi-grid Methods and Applications , author =

  13. [13]

    and Kanschat, G

    Janssen, B. and Kanschat, G. , title =. SIAM J. Sci. Comput. , year = 2011, volume = 33, number = 4, pages =

    2011

  14. [14]

    , title =

    Kanschat, G. , title =. J. Comput. Appl. Math. , year = 2008, volume = 218, pages =

    2008

  15. [15]

    and Mao, Y

    Kanschat, G. and Mao, Y. , title =. J. Numer. Math. , year = 2015, volume = 23, number = 1, pages =

    2015

  16. [16]

    2012 , doi =

    Kronbichler, Martin and Kormann, Katharina , journal =. 2012 , doi =

    2012

  17. [17]

    ACM Trans

    Kronbichler, Martin and Kormann, Katharina , title =. ACM Trans. Math. Softw. , pages =. 2019 , volume =

    2019

  18. [18]

    Enhancing data locality of the conjugate gradient method for high-order matrix-free finite-element implementations , author =. Int. J. High-Performance Computing Appl. , year =

  19. [19]

    Lucero Lorca, J. P. and G. Kanschat , title =. Electron. Trans. Numer. Anal. , volume =. 2021 , pages =

    2021

  20. [20]

    Lynch, R. E. and Rice, J. R. and Thomas, D. H. , title =. Numer. Math. , volume = 6, pages =

  21. [21]

    ACM Trans

    Multidimensional intratile parallelization for memory-starved stencil computations , author=. ACM Trans. Parallel Comput. , volume=. 2017 , doi=

    2017

  22. [22]

    Markus Melenk and Klaus Gerdes and Christoph Schwab

    J. Markus Melenk and Klaus Gerdes and Christoph Schwab. Fully discrete hp-finite elements: fast quadrature. Comp. Meth. Appl. Mech. Engrg. 2001

    2001

  23. [23]

    , journal =

    Orszag, Steven A. , journal =. Spectral methods for problems in complex geometries , volume =

  24. [24]

    Trojak and R

    W. Trojak and R. Watson and F.D. Witherden , title =. doi:10.1016/j.cpc.2021.108235 , year =

    doi:10.1016/j.cpc.2021.108235 2021

  25. [25]

    Proceedings of the 7th IEEE International Conference on eScience , year =

    Kormann, Katharina and Kronbichler, Martin , title =. Proceedings of the 7th IEEE International Conference on eScience , year =

  26. [26]

    and Arndt, D

    Witte, J. and Arndt, D. and Kanschat, G. , title =. Comput. Meth. Appl. Math. , number =. 2021 , pages =

    2021

  27. [27]

    doi:10.3390/axioms7030063 , year =

    Michael Dumbser and Francesco Fambri and Maurizio Tavelli and Michael Bader and Tobias Weinzierl , title =. doi:10.3390/axioms7030063 , year =

    doi:10.3390/axioms7030063

  28. [28]

    Textbook efficiency: massively parallel matrix-free multigrid for the

    Kohl, Nils and R\"ude, Ulrich , journal=. Textbook efficiency: massively parallel matrix-free multigrid for the. 2022 , doi=

    2022

  29. [29]

    ACM Trans

    Multigrid for matrix-free high-order finite element computations on graphics processors , author=. ACM Trans. Parallel Comput. , volume=. 2019 , doi=

    2019

  30. [30]

    An overlapping

    Fischer, Paul F and Miller, Neil I and Tufo, Henry M , journal =. An overlapping. 2000 , publisher=

    2000

  31. [31]

    A massively parallel nonoverlapping additive Schwarz method for discontinuous

    Dryja, Maksymilian and Krzy. A massively parallel nonoverlapping additive Schwarz method for discontinuous. Numer. Math. , volume=. 2016 , publisher=

    2016

  32. [32]

    From h to p efficiently: Implementing finite and spectral/hp element methods to achieve optimal performance for low-and high-order discretisations , author=. J. Comput. Phys. , volume=

  33. [33]

    Implementation of the

    Bauer, Petr and Klement, Vladimir and Oberhuber, Tom. Implementation of the. Computer Phys. Comm. , volume=

  34. [34]

    High-performance implementation of matrix-free high-order discontinuous

    M. High-performance implementation of matrix-free high-order discontinuous. arXiv preprint arXiv:1711.10885 , year=

    Pith/arXiv arXiv

  35. [35]

    Textbook multigrid efficiency for fluid simulations , author=. Annu. Rev. Fluid Mech. , volume=. 2003 , doi=

    2003

  36. [36]

    A matrix-free high-order discontinuous

    Fehn, Niklas and Wall, Wolfgang A and Kronbichler, Martin , journal=. A matrix-free high-order discontinuous. 2019 , publisher=

    2019

  37. [37]

    A robust multigrid method for discontinuous

    Hong, Qingguo and Kraus, Johannes and Xu, Jinchao and Zikatanov, Ludmil , journal=. A robust multigrid method for discontinuous. 2016 , publisher=

    2016

  38. [38]

    A study of vectorization for matrix-free finite element methods , author=. Int. J. High Perf. Comput. Appl. , volume=. 2020 , publisher=

    2020

  39. [39]

    doi:10.1145/3126908.3126963 , year =

    Ken Raffenetti and Abdelhalim Amer and Lena Oden and Charles Archer and Wesley Bland and Hajime Fujita and Yanfei Guo and Tomislav Janjusic and Dmitry Durnov and Michael Blocksome and Min Si and Sangmin Seo and Akhil Langer and Gengbin Zheng and Masamichi Takagi and Paul Coffman and Jithin Jose and Sayantan Sur and Alexander Sannikov and Sergey Oblomov an...

    doi:10.1145/3126908.3126963

  40. [40]

    Peter Munch and Martin Kronbichler , title =. Int. J. High Perf. Comput. Appl. , year = 2023, note =

    2023

  41. [41]

    Wall and Martin Kronbichler , title =

    Niklas Fehn and Peter Munch and Wolfgang A. Wall and Martin Kronbichler , title =. doi:10.1016/j.jcp.2020.109538 , year =

    doi:10.1016/j.jcp.2020.109538 2020

  42. [42]

    The International Journal of High Performance Computing Applications , volume=

    Enhancing data locality of the conjugate gradient method for high-order matrix-free finite-element implementations , author=. The International Journal of High Performance Computing Applications , volume=. 2023 , publisher=

    2023

  43. [43]

    SIAM Journal on Scientific Computing , volume=

    Smoothers with Localized Residual Computations for Geometric Multigrid Methods for Higher-Order Finite Elements , author=. SIAM Journal on Scientific Computing , volume=. 2025 , publisher=

    2025

  44. [44]

    A matrix-free multilevel preconditioner for the generalized

    Wichrowski, Micha. A matrix-free multilevel preconditioner for the generalized. Journal of Computational Science , volume=. 2022 , publisher=

    2022

  45. [45]

    An implementation of tensor product patch smoothers on

    Cui, Cu and Grosse-Bley, Paul and Kanschat, Guido and Strzodka, Robert , journal=. An implementation of tensor product patch smoothers on. 2025 , publisher=

    2025

  46. [46]

    Computational Methods in Applied Mathematics , number=

    Tensor-product vertex patch smoothers for biharmonic problems , author=. Computational Methods in Applied Mathematics , number=. 2025 , publisher=

    2025

  47. [47]

    Matrix-free multigrid block-preconditioners for higher order discontinuous

    Bastian, Peter and M. Matrix-free multigrid block-preconditioners for higher order discontinuous. Journal of Computational Physics , volume=. 2019 , publisher=

    2019

  48. [48]

    Numerische Mathematik , volume=

    Additive Schwarz methods for the p-version finite element method , author=. Numerische Mathematik , volume=. 1993 , publisher=

    1993

  49. [49]

    Numerical methods for partial differential equations , pages=

    Spectral methods for problems in complex geometrics , author=. Numerical methods for partial differential equations , pages=. 1979 , publisher=

    1979

  50. [50]

    Journal of Computational Physics , volume=

    Scaling to the stars--a linearly scaling elliptic solver for p-multigrid , author=. Journal of Computational Physics , volume=. 2019 , publisher=

    2019

  51. [51]

    Approximate tensor-product preconditioners for very high order discontinuous

    Pazner, Will and Persson, Per-Olof , journal=. Approximate tensor-product preconditioners for very high order discontinuous. 2018 , publisher=

    2018

  52. [52]

    2016 , publisher=

    Remacle, J-F and Gandham, Rajesh and Warburton, Tim , journal=. 2016 , publisher=

    2016

  53. [53]

    Efficient low-order refined preconditioners for high-order matrix-free continuous and discontinuous

    Pazner, Will , journal=. Efficient low-order refined preconditioners for high-order matrix-free continuous and discontinuous. 2020 , publisher=

    2020

  54. [54]

    SIAM Journal on Scientific Computing , volume=

    Scalable low-order finite element preconditioners for high-order spectral element Poisson solvers , author=. SIAM Journal on Scientific Computing , volume=. 2019 , publisher=

    2019

  55. [55]

    An hp multigrid approach for tensor-product space-time finite element discretizations of the

    Margenberg, Nils and Bause, Markus and Munch, Peter , journal=. An hp multigrid approach for tensor-product space-time finite element discretizations of the

  56. [56]

    An efficient smoother for the

    Braess, Dietrich and Sarazin, Regina , journal=. An efficient smoother for the. 1997 , publisher=

    1997

  57. [57]

    Matrix-free monolithic multigrid methods for

    Jodlbauer, Daniel and Langer, Ulrich and Wick, Thomas and Zulehner, Walter , journal=. Matrix-free monolithic multigrid methods for. 2024 , publisher=

    2024

  58. [58]

    Computing , volume=

    A class of smoothers for saddle point problems , author=. Computing , volume=. 2000 , publisher=

    2000

  59. [59]

    Monolithic multigrid preconditioners for high-order discretizations of

    Voronin, Alexey and Harper, Graham and MacLachlan, Scott and Olson, Luke N and Tuminaro, Raymond S , journal=. Monolithic multigrid preconditioners for high-order discretizations of. 2025 , publisher=

    2025

  60. [60]

    An energy-efficient

    Anselmann, Mathias and Bause, Markus and Margenberg, Nils and Shamko, Pavel , journal=. An energy-efficient. 2024 , publisher=

    2024

  61. [61]

    A Geometric Multigrid Preconditioner for Discontinuous

    Wichrowski, Michal , journal=. A Geometric Multigrid Preconditioner for Discontinuous

  62. [62]

    arXiv preprint arXiv:2507.17053 , year=

    Matrix-Free Evaluation of High-Order Shifted Boundary Finite Element Operators , author=. arXiv preprint arXiv:2507.17053 , year=

    Pith/arXiv arXiv

  63. [63]

    A multigrid method for

    Cui, Cu and Kanschat, Guido , journal=. A multigrid method for

  64. [64]

    SIAM Journal on Scientific Computing , volume=

    High-performance matrix-free unfitted finite element operator evaluation , author=. SIAM Journal on Scientific Computing , volume=. 2025 , publisher=

    2025

  65. [65]

    Differencing of the diffusion equation in

    Kershaw, David S , journal=. Differencing of the diffusion equation in. 1981 , publisher=

    1981

  66. [66]

    SIAM Journal on Scientific Computing , volume=

    Flexible conjugate gradients , author=. SIAM Journal on Scientific Computing , volume=. 2000 , publisher=

    2000

  67. [67]

    2002 , publisher=

    High-Order Methods for Incompressible Fluid Flow , author=. 2002 , publisher=

    2002

  68. [68]

    Journal of Open Source Software , volume=

    libCEED: Fast algebra for high-order element-based discretizations , author=. Journal of Open Source Software , volume=

  69. [69]

    Parallel Computing , volume=

    NekRS, a GPU-accelerated spectral element Navier--Stokes solver , author=. Parallel Computing , volume=. 2022 , publisher=

    2022

  70. [70]

    International Journal for Numerical Methods in Engineering , volume=

    A method of finite element tearing and interconnecting and its parallel solution algorithm , author=. International Journal for Numerical Methods in Engineering , volume=. 1991 , publisher=

    1991

  71. [71]

    SIAM Journal on Scientific Computing , volume=

    A preconditioner for substructuring based on constrained energy minimization , author=. SIAM Journal on Scientific Computing , volume=. 2003 , publisher=

    2003

  72. [72]

    Computers & Mathematics with Applications , volume=

    MFEM: A modular finite element methods library , author=. Computers & Mathematics with Applications , volume=. 2021 , publisher=

    2021

  73. [73]

    2020 , institution=

    Nek5000 developments in support of industry and the NRC , author=. 2020 , institution=

    2020

  74. [74]

    and Fischer, Paul F

    Lottes, James W. and Fischer, Paul F. , title =

  75. [75]

    Bradbury, James and Frostig, Roy and Hawkins, Peter and Johnson, Matthew James and Leary, Chris and Maclaurin, Dougal and Necula, George and Paszke, Adam and Vander

  76. [76]

    Proceedings of the 3rd ACM SIGPLAN International Workshop on Machine Learning and Programming Languages (MAPL) , pages=

    Triton: an intermediate language and compiler for tiled neural network computations , author=. Proceedings of the 3rd ACM SIGPLAN International Workshop on Machine Learning and Programming Languages (MAPL) , pages=

  77. [77]

    arXiv preprint arXiv:2508.00441 , year=

    DGEMM without FP64 Arithmetic-Using FP64 Emulation and FP8 Tensor Cores with Ozaki Scheme , author=. arXiv preprint arXiv:2508.00441 , year=

    arXiv

  78. [78]

    arXiv preprint arXiv:2504.19442 , year=

    Triton-distributed: Programming overlapping kernels on distributed ai systems with the triton compiler , author=. arXiv preprint arXiv:2504.19442 , year=

    Pith/arXiv arXiv

  79. [79]

    Proceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 1 , pages=

    Linear Layouts: Robust Code Generation of Efficient Tensor Computation Using F\_2 , author=. Proceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 1 , pages=

  80. [80]

    arXiv preprint arXiv:2512.18134 , year=

    Optimal Software Pipelining and Warp Specialization for Tensor Core GPUs , author=. arXiv preprint arXiv:2512.18134 , year=

    arXiv

Showing first 80 references.