Chiral rank-$k$ truncations for the multigrid preconditioner of Wilson fermions in lattice QCD

Andreas Stathopoulos; Eloy Romero; Travis Whyte

REVIEW 3 major objections 5 minor 26 references

Chiral rank-$k$ truncations for the multigrid preconditioner of Wilson fermions in lattice QCD

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Chiral SVD truncation of a larger test-vector basis lowers multigrid solve times.

desk verdict Useful but underspecified: the chiral rank-k SVD truncation likely works, but the paper never documents the conventional baseline against which it is compared, so the size of the claimed gain is uncertain. read the letter →

arxiv 2502.03091 v1 pith:ZFC4OCPE submitted 2025-02-05 hep-lat cs.NAmath.NA

Travis Whyte , Andreas Stathopoulos , Eloy Romero This is my paper · ORCID

classification hep-latcs.NAmath.NA MSC 65F1065N5515A18 PACS 12.38.Gc

keywords latticeQCDWilsonfermionsmultigridpreconditionersingularvaluedecompositionchiralprojectiontestvectorsanisotropicsetupiterations

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 is trying to establish a simple change in how multigrid preconditioners are built for Wilson fermions in lattice QCD. Instead of stopping at the usual 24--32 smoother-generated test vectors, the setup computes a larger basis (up to $m=128$) and then trims it with a singular value decomposition applied to each chiral half of the vectors, keeping only the $k$ leading left singular vectors per lattice domain. The truncation keeps the coarse-grid matrices small while capturing more of the near-null space that the smoother exposes. On an anisotropic lattice with $m_\pi \approx 239$ MeV and an isotropic lattice with $m_\pi \approx 220$ MeV, the paper reports that this construction lowers the mean time to solve the Wilson linear systems compared with conventional multigrid across every setup-iteration count tested, and that the benefit persists as the lattice volume grows. A reader would care because propagator solves are a dominant cost in many lattice QCD calculations, and the change lives entirely in the setup phase of an existing preconditioner.

What carries the argument

The central object is the chirally split, SVD-truncated test-vector basis. Smoothing on $\mathbf{D}\psi_i\approx 0$ with $m$ random initial guesses produces test vectors $\Psi=[\psi_1,\dots,\psi_m]$; the chiral projectors $(1\pm\gamma_5)/2$ divide their spin components into two sets, and each set is restricted to a lattice domain $\Lambda_j$. The singular value decomposition of each chiral block is the low-rank mechanism: keeping the $k$ largest left singular vectors gives the best rank-$k$ approximation to that block, forming a prolongator $\mathbf{P}$ of dimension $n\times 2kd$, and the matching right factor becomes the coarse-grid test vectors. The chiral projectors enforce the symmetry $\gamma_5\mathbf{P} = \mathbf{P}\sigma_3$, so right singular vectors of $\mathbf{P}$ align with left singular vectors of $\mathbf{R}$, and the SVD's orthonormality removes the conventional Gram-Schmidt step. This mechanism carries the argument because it lets a larger, more informative smoother basis be compressed back to the same coarse-grid size that conventional multigrid uses.

What would settle it

Run a parameter sweep for conventional multigrid on the same two ensembles, varying its test-vector count (for example 24, 48, 96) and its setup-iteration count (100--600) with the same three-level hierarchy and ten right-hand sides; if the best conventional mean solve time matches or beats the best chiral rank-$k$ time, the reported advantage is explained by parameter selection rather than by the SVD truncation.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that a rank-$k$ truncation of the chirally split test-vector basis produces a better preconditioner for the Wilson-Dirac operator than conventional multigrid at comparable coarse-grid cost. The test vectors are generated by smoothing on $\mathbf{D}\psi_i \approx 0$ for $i=1,\ldots,m$ with $m$ larger than standard practice, split with the projectors $(1\pm\gamma_5)/2$, and restricted to each lattice domain $\Lambda_j$. The singular value decomposition of each chiral block is $\mathbf{U}\boldsymbol{\Sigma}\mathbf{V}^{\dagger}$; the $k$ leading left singular vectors become the prolongator columns forming $\mathbf{P}$, $\mathbf{R}=\mathbf{P}^{\dagger}$, and the next-level operator is $\mathbf{D}_{\ell+1}=\mathbf{R}\mathbf{D}_\ell \mathbf{P}$. Because the retained vectors are already orthonormal, the separate orthonormalization used in the conventional setup is unnecessary. The paper reports that this modified setup reduces mean solve time on an anisotropic HSC ensemble with $m_\pi\approx 239$ MeV and an isotropic MILC ensemble with $m_\pi\approx 220$ MeV, with $m=96$ and $k=24$ near optimal, and that the gain appears over a range of setup iterations and across spatial volumes $L_s=24,32,40$.

Load-bearing premise

The comparison assumes that the conventional multigrid baseline is as well tuned as the new method; the paper does not report the baseline's test-vector count, smoother setup, or cycle parameters, and the new method's $m=96$, $k=24$ settings are chosen by sweeps on the same ensembles used for the headline timings.

Editorial extensions

If this is right

Using $m=96$ smoother vectors truncated to $k=24$ lowers the mean solve time for Wilson linear systems on both tested ensembles relative to conventional multigrid.
Increasing the initial basis size tends to improve the preconditioner up to about $m=96$--$128$, after which added vectors contribute little, so the truncation rank can be chosen near the saturation point.
The truncated preconditioner is less sensitive to the number of setup iterations, meaning fewer smoother sweeps can be used without sacrificing solver performance.
The benefit persists when the spatial lattice volume grows from $L_s=24$ to $L_s=40$, the direction required for production-scale calculations.

Reading between the lines

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

Because the truncation operates on the chiral blocks of a linear-algebra basis, the same recipe should transfer to other fermion discretizations that already use adaptive multigrid, such as domain-wall or twisted-mass formulations; the paper demonstrates Wilson fermions only.
The reduced sensitivity to setup iterations points to a production trade not measured in the paper: replace many smoother sweeps with a larger initial basis and rely on the coarse-grid K-cycle, then compare total wall-clock time including setup.
The rapid decay of the singular spectra suggests an adaptive rule for choosing $k$ from a relative singular-value threshold, which could remove the need for a parameter sweep on new ensembles.

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.

Desk Editor's Note

read the letter

Short version: this is a modest but probably real algorithmic improvement. The authors take an existing trick—rank-k truncation of a large set of smooth aggregates—and apply it to Wilson fermions with chiral splitting. The math is not deep, and the paper admits the extension is straightforward. What is new is the numerical evidence: two physically relevant ensembles, several volumes, and a range of setup iterations, with consistently lower solve times for the truncated method.

I think the results are likely genuine, not fabricated. The cleanest part is Fig. 2: the singular spectra of the chirally split test vectors decay fast, so truncating at k=24 is well motivated. The volume scaling in Fig. 7 uses fixed parameters, which is a fairer test than the earlier sweeps.

The soft spots are real but not fatal. The main one: the paper never tells you what "Regular MG" is. How many test vectors per level? What smoother, what cycle parameters? Without that, the comparison could be against an untuned or deliberately small conventional setup. The authors imply typical choices are 24–32 vectors, but their own method starts with m=96. So the comparison may be measuring the benefit of a larger search space, not of the SVD truncation per se. A proper ablation would use the same m for both, or at least constrain total coarse-grid cost.

Second, the optimal m and k are chosen on the same ensembles used for the headline numbers. The sweeps in Figs. 1 and 3 are sensible engineering, but they mean the reported gain is a selected maximum, not a prediction. There are no error bars anywhere; ten right-hand sides is enough to see a trend, not to quantify a 20–30% improvement.

Third, the notation in Eq. (5) is sloppy, and no code or detailed run scripts are provided, so reproduction would be painful.

Who should read this: lattice QCD people doing Wilson or Clover propagators, and AMG specialists curious about the chiral-split SVD. The paper deserves a serious referee—the idea is useful and the evidence, while incomplete, is consistent. A referee should ask for the baseline configuration, error bars or at least a statement of timing variance, and perhaps a matched-m comparison. As a proceedings contribution it is acceptable; as a journal paper it would need more work.

Recommendation: send it to peer review. The concerns are fixable with more information, not with new theory.

Referee Report

3 major / 5 minor

Summary. The paper proposes a modification to the setup of the multigrid preconditioner for Wilson fermions. Instead of using the conventional set of smoother-generated test vectors directly, the authors generate a larger basis (size m) and then, separately for each chiral component and each lattice domain, compute an SVD and retain only the k leading left singular vectors to form the prolongator and restriction operators. Numerical experiments on one anisotropic HSC ensemble and one isotropic MILC ensemble, together with a volume-scaling study, report lower mean solve times and iteration counts for the proposed 'CSVD MG' preconditioner than for a 'Regular MG' baseline.

Significance. The idea is simple, well motivated by the observed rapid decay of the singular spectra in Fig. 2, and potentially useful because it decouples the cost of the coarse-grid operator from the size of the search space used during setup. The paper reports consistent qualitative trends across two different fermion actions, several lattice volumes, and many setup-iteration counts. However, the central quantitative claim—that the chiral rank-k truncation improves convergence relative to conventional multigrid—is not yet established because the baseline is not specified and because the effect of the truncation is not isolated from the effect of using a larger initial basis. The evidence is therefore suggestive rather than conclusive in its current form.

major comments (3)

[Section 4.3 (Figs. 4-5)] The 'Regular MG' baseline is never specified. The paper does not state the number of test vectors per level, the smoother type or number of smoothing iterations, the cycle parameters, or whether the same setup-iteration counts are used for both methods. Since the headline claim is that CSVD MG reduces solve time for all setup-iteration counts, the comparison cannot be assessed without this information. Please provide a complete configuration for the conventional multigrid hierarchy, ideally from the same code with the truncation disabled.
[Sections 4.1-4.3] The paper does not isolate the effect of the SVD truncation from the enlarged basis size. All CSVD runs use m=96 with k=24, while the conventional baseline appears to operate at the typical 24-32 test vectors mentioned in Sec. 4.1. Because Fig. 1 shows that increasing m alone lowers solve time for the CSVD method, the observed improvement over 'Regular MG' could be due to the larger generation space rather than to the rank-k truncation. A controlled experiment comparing conventional MG with m=96 (or several values of m) against CSVD with the same m and varying k is needed to support the claim that the truncation itself improves convergence.
[Section 4.2 and Figs. 1, 3, 4, 5, 7] The optimal values m=96 and k=24 are selected from sweeps on the same ensembles (Figs. 1 and 3) that are then used for the headline comparisons in Sec. 4.3, and no statistical uncertainties are reported anywhere in the paper. Timing measurements in high-performance solvers have run-to-run variability, and selecting parameters on the test data introduces selection bias. Please report error bars or ranges (for instance over the ten right-hand sides), and preferably demonstrate that the relative improvement persists for parameter choices that were not used in tuning, e.g., on a separate validation set or by reporting the analogous sweep for the conventional baseline.

minor comments (5)

[Section 3, Eq. (5)] The quantity P_alpha^j is not defined before Eq. (5), and the approximation U_k Sigma_k V_k^dagger approximately equal to P_alpha^j Psi^(alpha,l+1)(x_c) is unclear: the left side is the truncated SVD of the domain-restricted test-vector matrix, while the right side appears to mix the prolongator block with a coarse-grid vector. Please rewrite this equation using explicit domain-local notation.
[Section 4.3] The statement that the CSVD preconditioner is 'less sensitive to the number of setup iterations' is supported only by visual inspection of Figs. 4 and 5. A quantitative measure, such as the slope of mean solve time versus setup iterations, would make the claim precise.
[Section 4.4] The 'near optimal setup parameters' used for the volume-scaling runs are not listed. The paper should state the values of m, k, and the setup-iteration counts for both the CSVD and the conventional method for all three volumes.
[Section 3] The relation to prior rank-truncation work in Refs. [22,23] is stated only briefly. A sentence explaining what is new in applying the SVD separately to the chiral components of Wilson-fermion test vectors would help position the contribution.
[Figures 1, 3, 4, 5, 7] None of the timing plots show error bars or scatter of the ten right-hand-side measurements. At minimum, the captions should state whether the plotted means are stable across the ten solves; ideally, error bars should be included.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity found; the chiral rank-k truncation is a construction defined by SVD, and the improvement claim rests on direct numerical experiments rather than on a self-referential derivation.

full rationale

The paper's derivation chain is method construction plus empirical benchmarking, not circular reasoning. Equation (5) defines the SVD truncation and Eq. (6) explicitly constructs the prolongator from the left singular vectors; this is a definitional construction, not a claim that derives its own conclusion. The reported speedups come from direct measurements against conventional multigrid (Figs. 4-5, 7), not from an equation whose output is identical to a fitted input. The choices m=96 and k=24 are tuned on the same ensembles in Sections 4.1-4.2, but this is parameter selection, which the reviewing rules assign to selection bias and soundness rather than to circularity; no 'prediction' is made from those fitted values. The rank-k truncation idea is explicitly attributed to Ref. [22] and its extension to Wilson fermions is presented as straightforward; chiral splitting is credited to established multigrid literature [1]. None of these citations are self-citations that smuggle in the target result. The uncharacterized conventional-multigrid baseline is a legitimate fairness concern, but it is not a circularity: the comparison could be unfair or under-tuned without the paper's argument reducing to its own assumptions. No self-definitional, fitted-input-as-prediction, self-citation-load-bearing, or uniqueness-imported-from-authors step is present.

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

The numerical claims rest on standard SVD theory and on domain assumptions about Wilson fermion multigrid. The main tuned inputs are the basis size m and truncation rank k, plus the number of setup iterations, all chosen empirically. No new physical or mathematical entities are postulated.

free parameters (5)

Initial basis size m on level 0 (m0) = 96
Chosen from the solve-time sweep in Fig. 1; larger values give diminishing returns beyond m=96.
Initial basis size m on level 1 (m1) = 128
Used in Figs. 3-7 as the level-1 basis size for the chiral rank-k method.
Truncation rank k on level 0 (k0) = 24
Selected as the optimal truncation in Fig. 3 for both ensembles.
Truncation rank k on level 1 (k1) = 24
Reported as the optimal truncation in Fig. 3, though Fig. 1 used k1=32 as a starting point.
Setup iterations per level = not stated
The paper varies setup iterations from 0 to 600 and reports improvements across the range, but the exact value used in the volume-scaling runs is described only as 'near optimal' and is not specified.

assumptions (5)

standard math SVD provides the best rank-k approximation to the test-vector basis on each domain.
Used in Section 3, Eq. (5), to justify replacing the full m-vector basis with k left singular vectors.
domain assumption Smooth test vectors generated by the smoother expose the near null space of the Wilson-Dirac operator (local coherence).
Stated in Section 2, citing Refs. [13,14]; the entire multigrid construction depends on this.
domain assumption Chiral splitting with projectors (1 +/- gamma5) preserves the symmetry needed for a valid multigrid hierarchy.
Section 2: the projector enforces gamma5 P = P sigma3, which is required for the coarse-grid correction to be consistent for Wilson fermions.
domain assumption The singular spectrum of the first domain is representative of all domains, justifying a uniform truncation rank k.
Section 4.2 states the first domain is representative of other domains; if false, a fixed k may over- or under-resolve other domains.
domain assumption The two ensembles (HSC and MILC) are representative testbeds for practical lattice QCD multigrid.
The conclusions generalize from these ensembles to other gauge actions and volumes.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Chiral rank-$k$ truncations for the multigrid preconditioner of Wilson fermions in lattice QCD." pith.science (2026). https://pith.science/paper/ZFC4OCPE

@misc{pith2026250203091,
  author       = {Pith},
  title        = {Pith review of: Chiral rank-$k$ truncations for the multigrid preconditioner of Wilson fermions in lattice QCD},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZFC4OCPE}},
  note         = {Machine review of arXiv:2502.03091}
}

abstract

We present a modification to the setup algorithm for the multigrid preconditioner of Wilson fermions in lattice QCD. A larger number of test vectors than that used in conventional multigrid is generated by the smoother. This set of test vectors is then truncated by a singular value decomposition on the chiral components of the test vectors, which are subsequently used to form the prolongation and restriction matrices of the multigrid hierarchy. This modification is demonstrated to improve the convergence of linear equations on an anisotropic lattice with $m_{\pi} \approx 239$ MeV from the Hadron Spectrum Collaboration and an isotropic lattice with $m_{\pi} \approx 220$ MeV from the MILC Collaboration. The lattice volume dependence of the method is also examined.

Figures

Figures reproduced from arXiv: 2502.03091 by the authors.

**Figure 1.** The mean solve time as a function of the initial basis size for Ensemble A (left) and Ensemble B (right). The bottom and top 𝑦-axis displays 𝑚 for levels ℓ = 0, 1, respectively. shows the mean execution time for solving the systems of linear equations for both Ensemble A and Ensemble B. The basis is truncated at 𝑘 = 24, 32 for levels ℓ = 0, 1, respectively. A general trend is observed for both ensembles: as 𝑚 is inc… view at source ↗

**Figure 2.** The singular spectrum for the chirally split test vectors on the first domain for level ℓ = 0 (top) and ℓ = 1 (bottom) of Ensemble A (left) and Ensemble B (right). 8 10 12 14 16 18 12 16 24 32 48 16 24 32 48 64 Mean Solve Time (s) k0 with m0 = 96 k1 with m1 = 128 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 12 16 24 32 48 16 24 32 48 64 Mean Solve Time (s) k0 with m0 = 96 k1 with m1 = 128 [PITH_FULL_IMAGE:figures/full… view at source ↗

**Figure 3.** The mean execution time for the system of linear equations when the degree of the truncation is varied using an initial basis size of 𝑚 = 96 for Ensemble A (left) and Ensemble B (right). The bottom and top 𝑦-axis displays 𝑘 for levels ℓ = 0, 1, respectively [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: The total number of iterations on ℓ = 0 (upper left), ℓ = 1 (upper right), ℓ = 2 (lower left) and the mean solve time of the system of linear equations for Ensemble A. 0 2 4 6 8 10 12 14 0 100 200 300 400 500 600 Total Mean Iterations Setup Iterations Level 0 Regular M…

**Figure 5.** Figure 5: As [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

**Figure 6.** Figure 6: The normalized number of iterations on level ℓ = 1 (top) and ℓ = 2 (bottom) for Ensemble A (left) and Ensemble B (right). While the chiral rank-𝑘 truncation setup method is observed to decrease the total cost of solving the linear equations, it is important to quantify…

**Figure 7.** Figure 7: The performance of both MG preconditioners as the lattice volume is increased. 5. Summary We have presented a modification to the setup algorithm for the multigrid preconditioner for systems of linear equations of Wilson fermions. The chiral rank-k truncation utilizes …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 10 canonical work pages

[1]

Brannick et al.,Adaptive Multigrid Algorithm for Lattice QCD,Phys

J. Brannick et al.,Adaptive Multigrid Algorithm for Lattice QCD,Phys. Rev. Lett.100 (2008) 041601 [0707.4018]

arXiv 2008
[2]

RichardC.Broweretal., Multigridforchirallatticefermions: Domainwall ,Phys.Rev.D 102 (2020) 094517 [2004.07732]

arXiv 2020
[3]

Brower et al.,Multigrid algorithm for staggered lattice fermions, Phys

Richard C. Brower et al.,Multigrid algorithm for staggered lattice fermions, Phys. Rev. D97 (2018) 114513 [1801.07823]

arXiv 2018
[4]

Math.132 (2016) 463 [1410.7170]

James Brannick et al.,Multigrid Preconditioning for the Overlap Operator in Lattice QCD, Numer. Math.132 (2016) 463 [1410.7170]

arXiv 2016
[5]

Frommer et al.,Adaptive Aggregation-Based Domain Decomposition Multigrid for the Lattice Wilson–Dirac Operator,SIAM J

A. Frommer et al.,Adaptive Aggregation-Based Domain Decomposition Multigrid for the Lattice Wilson–Dirac Operator,SIAM J. Sci. Comput.36 (2014) A1581 [1303.1377]

arXiv 2014
[6]

S.Bacchioetal., DDalphaAMGforTwistedMassFermions ,PoS LATTICE2016(2016)259 [1611.01034]

work page Pith review arXiv 2016
[7]

J.BrannickandK.Kahl, BootstrapAlgebraicMultigridforthe2DWilsonDiracsystem ,SIAM J. Sci. Comput.36 (2014) B321 [1308.5992]

work page Pith review arXiv 2014
[8]

Deflated GMRES with Multigrid for Lattice QCD

Travis Whyte and Walter Wilcox and Ronald B. Morgan,Deflated GMRES with Multigrid for Lattice QCD, Phys. Lett. B803 (2020) 135281 [1912.02868]

work page Pith review arXiv 2020

Show all 26 references

[9]

Morgan et

Ronald B. Morgan et. al., Two-Grid Deflated Krylov Methods for Linear Equations , 2005.03070

2005 arXiv
[10]

Edwards and Balint Joo,The Chroma software system for lattice QCD,Nucl

Robert G. Edwards and Balint Joo,The Chroma software system for lattice QCD,Nucl. Phys. Proc. Suppl.140 (2005) 832 [hep-lat/0409003]

2005 arXiv
[11]

BalintJooetal., LatticeQCDonIntel ®XeonPhiTMCoprocessors ,Supercomputing,Lecture Notes in Computer Science, 7905 40-54 2013 Springer Berlin Heidelberg

2013
[12]

Balint Joo,MG_PROTO: A Multigrid Library for QCD
[13]

Martin Lüscher, Deflation acceleration of lattice QCD simulations, JHEP 12 (2007) 011 [0710.5417]

2007 arXiv
[14]

MartinLüscher, LocalcoherenceanddeflationofthelowquarkmodesinlatticeQCD .JHEP 07 (2007) 081 [0706.2298]

2007 arXiv
[15]

Rajan Gupta et al.,Isovector Charges of the Nucleon from 2+1+1-flavor Lattice QCD, Phys. Rev. D98 (2018) 034503 [1806.09006]

2018 arXiv
[16]

Anna Hasenfratz and Francesco Knechtli,Flavor symmetry and the static potential with hypercubic blocking, Phys. Rev. D64 (2001) 034504 [hep-lat/0103029]. 9 Chiral rank-𝑘 truncations for multigrid in LQCD Travis Whyte

2001 arXiv
[17]

RobertG.EdwardsandBalintJooandHuey-WenLin, TuningforThree-flavorsofAnisotropic Clover Fermions with Stout-link Smearing, Phys. Rev. D78 (2008) 054501 [0803.3960]

2008 arXiv
[18]

Huey-WenLinetal., Firstresultsfrom2+1dynamicalquarkflavorsonananisotropiclattice: Light-hadronspectroscopyandsettingthestrange-quarkmass ,Phys.Rev.D 79(2009)034502 [0810.3588]

2009 arXiv
[19]

DavidJ.Wilsonetal., Thequark-massdependenceofelastic 𝜋𝐾 scatteringfromQCD ,Phys. Rev. Lett.123 (2019) 042002 [1904.03188]

2019 arXiv
[20]

Bazovov et al.,Scaling studies of QCD with the dynamical HISQ action, Phys

A. Bazovov et al.,Scaling studies of QCD with the dynamical HISQ action, Phys. Rev. D82 (2010) 074501 [1004.0342]

2010 arXiv
[21]

Bazovov et al.,Lattice QCD Ensembles with Four Flavors of Highly Improved Staggered Quarks,Phys

A. Bazovov et al.,Lattice QCD Ensembles with Four Flavors of Highly Improved Staggered Quarks,Phys. Rev. D87 (2013) 054505 [1212.4768]

2013 arXiv
[22]

Edmond Chow,An Aggregation Multilevel Method Using Smooth Error Vectors,SIAM J. Sci. Comput. 27 (2006) 1727

2006
[23]

Vassilevski,Improving solve time of aggregation-based adaptive AMG, Num

Pasqua D’Ambra and Panayot S. Vassilevski,Improving solve time of aggregation-based adaptive AMG, Num. Lin. Alg. with App.26 (2019) 2269

2019
[24]

Brezina et al.,Adaptive Smoothed Aggregation (𝛼SA), SIAM J

M. Brezina et al.,Adaptive Smoothed Aggregation (𝛼SA), SIAM J. Sci. Comput.25 (2004) 1896

2004
[25]

Vassilevski,Recursive Krylov-based multigrid cycles, Num

Yvan Notay and Panayot S. Vassilevski,Recursive Krylov-based multigrid cycles, Num. Lin. Alg. with App.15 (2008) 473

2008
[26]

Colin Morningstar and Mike Peardon,Analytic smearing of SU(3) link variables in lattice QCD, Phys. Rev. D69 (2004) 054501 [hep-lat/0311018]. 10

2004 arXiv

Pith tools

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

[1] [1]

Brannick et al.,Adaptive Multigrid Algorithm for Lattice QCD,Phys

J. Brannick et al.,Adaptive Multigrid Algorithm for Lattice QCD,Phys. Rev. Lett.100 (2008) 041601 [0707.4018]

arXiv 2008

[2] [2]

RichardC.Broweretal., Multigridforchirallatticefermions: Domainwall ,Phys.Rev.D 102 (2020) 094517 [2004.07732]

arXiv 2020

[3] [3]

Brower et al.,Multigrid algorithm for staggered lattice fermions, Phys

Richard C. Brower et al.,Multigrid algorithm for staggered lattice fermions, Phys. Rev. D97 (2018) 114513 [1801.07823]

arXiv 2018

[4] [4]

Math.132 (2016) 463 [1410.7170]

James Brannick et al.,Multigrid Preconditioning for the Overlap Operator in Lattice QCD, Numer. Math.132 (2016) 463 [1410.7170]

arXiv 2016

[5] [5]

Frommer et al.,Adaptive Aggregation-Based Domain Decomposition Multigrid for the Lattice Wilson–Dirac Operator,SIAM J

A. Frommer et al.,Adaptive Aggregation-Based Domain Decomposition Multigrid for the Lattice Wilson–Dirac Operator,SIAM J. Sci. Comput.36 (2014) A1581 [1303.1377]

arXiv 2014

[6] [6]

S.Bacchioetal., DDalphaAMGforTwistedMassFermions ,PoS LATTICE2016(2016)259 [1611.01034]

work page Pith review arXiv 2016

[7] [7]

J.BrannickandK.Kahl, BootstrapAlgebraicMultigridforthe2DWilsonDiracsystem ,SIAM J. Sci. Comput.36 (2014) B321 [1308.5992]

work page Pith review arXiv 2014

[8] [8]

Deflated GMRES with Multigrid for Lattice QCD

Travis Whyte and Walter Wilcox and Ronald B. Morgan,Deflated GMRES with Multigrid for Lattice QCD, Phys. Lett. B803 (2020) 135281 [1912.02868]

work page Pith review arXiv 2020