REVIEW 3 major objections 5 minor 50 references
Raising the coarse polynomial degree on R-tree agglomerates keeps multilevel CG iterations stable for high-order continuous finite elements where standard AMG deteriorates.
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 →
T0 review · grok-4.5
2026-07-31 13:44 UTC pith:QBRICDPE
load-bearing objection Solid incremental methods paper: high-order Nicolaides on R-tree boxes works in practice for continuous FE where plain AMG slips; theory is honest two-level with imposed geometry assumptions. the 3 major comments →
R3MG-C: a high-order algebraic-geometric multilevel preconditioner for continuous finite element discretizations
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
An algebraic-geometric V-cycle whose coarse spaces are continuous nodal interpolants of discontinuous tensor-product polynomials of degree p' defined on R-tree bounding boxes of the fine support points yields a high-order Nicolaides correction; under admissible-hierarchy assumptions the two-level error propagator satisfies a bound that improves with p' and that remains controlled when H/h is held fixed, and the same construction produces stable CG iteration counts for high-order continuous elements where classical low-order AMG deteriorates.
What carries the argument
High-order Nicolaides coarse space: on each R-tree agglomerate box B_i one takes the discontinuous space Q_{p'}(B_i), restricts it to the owned support points, and injects the result into the continuous fine space by nodal interpolation; the resulting prolongation produces a nested Galerkin hierarchy without a prescribed mesh hierarchy.
Load-bearing premise
The R-tree must produce boxes whose aspect ratios stay uniformly bounded, whose interiors occupy a non-degenerate connected portion of each box, and whose owned support points remain stable for local polynomial interpolation of degree p'; if the heuristic yields skinny or poorly unisolvent agglomerates the approximation estimate collapses.
What would settle it
Fix a high-order continuous discretization (p ≥ 3) with controlled H/h, raise the coarse degree p' while keeping the same agglomeration parameters, and check whether CG iteration counts stay bounded and better than a standard low-order AMG run on the identical systems; if iterations still grow with p or with H/h the claimed offset fails.
If this is right
- Controlling the agglomerate-to-mesh ratio H/h and choosing p' > 0 yields mesh-independent CG iteration counts for fixed fine degree p.
- The same R-tree construction supplies a black-box hierarchy for both cell-based and point-based agglomeration without any geometric multigrid mesh sequence.
- On unstructured and realistic geometries (ventricle, liver) the enriched coarse space remains competitive with, and often cheaper in iterations than, a production smoothed-aggregation AMG package.
- Increasing p' improves the theoretical constant even when agglomerates are large, giving a practical knob that classical constant-per-aggregate AMG lacks.
Where Pith is reading between the lines
- The same box-polynomial injection idea could be tried for other continuous high-order bases (e.g., serendipity or hierarchical) once unisolvency of the owned nodes is verified.
- If the admissible-hierarchy assumptions can be proved rather than imposed for R*-trees, the method would become a fully rigorous black-box high-order AMG.
- Balancing agglomerate size against p' is itself an optimization problem; an adaptive choice of m and p' per level could further reduce setup cost on very large 3-D meshes.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes R3MG-C, an algebraic-geometric multilevel preconditioner for continuous Lagrangian finite-element discretizations of elliptic problems. Using only support-point coordinates, an R*-tree builds a hierarchy of axis-aligned agglomerates; on each agglomerate box a discontinuous tensor-product space of degree p' is injected into the fine continuous space by nodal interpolation, yielding a high-order Nicolaides-type coarse space and a Galerkin hierarchy. A two-level analysis in the Xu–Zikatanov subspace-correction framework gives an explicit bound on the A-norm of the error propagator in which the factor p'^{-2} offsets large H/h and high fine degree p, under uniform box-regularity, interpolation stability, and stable-decomposition assumptions. Numerical V-cycle–CG experiments in 2D/3D on structured, unstructured, simplicial, and realistic geometries (ventricle, liver) show stable iteration counts and competitiveness with Trilinos ML AMG, including regimes where standard AMG deteriorates.
Significance. High-order continuous FE systems remain a practical weak spot for black-box AMG; a construction that needs only support-point coordinates, no prescribed mesh hierarchy, and no local eigenproblems is of clear interest. The high-order Nicolaides enrichment is a clean, inexpensive idea, the two-level analysis is explicit about the p'/H/h tradeoff, and the public polyDEAL implementation plus extensive tables (structured/unstructured, Q/P, 2D/3D, realistic meshes) make the contribution reproducible and falsifiable. Even with a non-sharp bound and heuristic agglomeration, the work is a solid addition to aggregation-based multilevel methods for high-order conforming elements.
major comments (3)
- [Section 5, Assumptions 2] Section 5, Assumptions 2: the two-level rate (Thm. 1 with Thms. 2–3) rests on uniform box aspect ratios, nondegenerate connected interiors with h/H/p/p'-independent Poincaré and approximation constants, O(h) overlap of bounded multiplicity, and uniform stability/unisolvence of interpolation from Q_{p'}(B_i) on the owned nodes P_i. These are imposed, not derived from the R*-tree heuristics (Sec. 3, Remarks 1–2). If an agglomerate is skinny, has a poorly connected interior, or owned nodes fail unisolvence for degree p', the Bramble–Hilbert step that produces the p'^{-2} improvement in μ_c^{-1} fails and the claimed offset of large H/h or high p disappears. Please either (i) state verifiable sufficient conditions on (m,M) and the mesh that imply Assumptions 2, or (ii) add cheap post-agglomeration diagnostics (aspect-ratio histograms, interior volume fraction, local interpolation condition n
- [Section 5.2; Tables 5, 10, 15, 32] The practical p'–H tradeoff is load-bearing for the central claim that raising p' improves robustness, yet it is only noted in passing. Remarks 1–2 and several tables (e.g. Tables 5, 10, 15, 16, 32) show that admissible local interpolation for larger p' forces larger m and thus larger H/h; in some simplicial/realistic cases the iteration count does not improve when p' is raised. The two-level bound has (H/h)^3 in the denominator and only p'^{-2} in the numerator, so the net effect is not automatic. A short systematic discussion (or a small parameter study) of how to choose m relative to p and p' so that the p' enrichment wins would substantially strengthen the paper’s guidance and match the theory to the numerics.
- [Abstract; Section 5; Section 6] All reported solvers are multilevel V-cycles, while the analysis is strictly two-level (Sec. 5). This is common, but the abstract and introduction present the bound as quantifying the method that is actually used. Please either sketch how the two-level estimate extends under nested R-tree levels (or recursive application of the same assumptions), or clearly separate the proven two-level statement from the multilevel numerical method and avoid implying that the displayed rate covers the full V-cycle hierarchy.
minor comments (5)
- [Abstract; Section 1] Abstract and Sec. 1: “support-point are recursively partitioned” → “support points are…”. Several similar small grammar slips (e.g. “degreep'”, missing spaces before math) should be cleaned in a pass.
- [Section 5.3; Remark 6] Lemma 2 and the GLL mass scaling are for tensor-product elements; the simplicial case (Sec. 5.3) correctly weakens the bound but still uses uniformly spaced nodes in deal.II for p≤3. A one-sentence caveat that the reported simplicial constants are for low p only would help.
- [Table 1; Section 6] Table 1: Trilinos ML settings are given; for high-order runs it would help to state the precise meaning of higher_order_elements and whether smoothed aggregation strength/threshold were retuned, so the comparison is fully reproducible.
- [Section 4; Figure 3] Figure 3 caption and the ownership rule are clear; a brief note that the visit order of agglomerates can affect which box owns interface nodes (and thus the sparsity pattern of P) would aid implementers.
- [Section 5; Section 7] The bound’s p^{2d+8} factor is very pessimistic relative to the tables; the authors already say it is not sharp—consider moving the sharpest form of the estimate next to the numerical takeaway in the conclusions so readers are not discouraged by the worst-case powers.
Circularity Check
No circularity: two-level rate is standard subspace-correction plus approximation theory under explicit assumptions, not forced by definition or self-citation.
full rationale
The central claim is the two-level bound (Thm. 1 combined with Thms. 2–3) on ∥E∥²_A. It is obtained from the Xu–Zikatanov abstract estimate by bounding C1, C2, μ_c via inverse inequalities, Poincaré/Bramble–Hilbert, and a stable nodal partition of unity; the p′^{-2} improvement appears only when the local L2 projection onto polynomials of degree p′ is inserted into those estimates. None of these steps equates the claimed rate to its own definition, nor is any constant fitted to data and then re-presented as a prediction. The R*-tree agglomeration is taken from the authors’ prior DG paper [21], but that citation supplies only the geometric partitioning tool; the continuous nodal-interpolation coarse space, the Galerkin hierarchy, and the entire convergence analysis are developed in the present manuscript under stated Assumptions 1–2. Numerical comparisons use an external Trilinos ML baseline. Soft spots (admissible-hierarchy assumptions not proved for every R*-tree output) affect correctness risk, not circularity. Hence score 0 with empty steps.
Axiom & Free-Parameter Ledger
free parameters (3)
- R-tree order parameters (m, M=2m), separately for point- and cell-based agglomeration =
Typical values m∈{2,4,6,7} depending on mesh and p′ (Tables 3–32)
- Coarse polynomial degree p′ and number of multigrid levels retained =
p′∈{0,1,2}; levels chosen bottom-up, often skipping R-tree leaves
- Chebyshev–Jacobi smoother parameters (degree 3, 2 pre/post sweeps) and CG tolerances =
degree 3; 2+2 sweeps; abs 1e-9 or 6-order residual drop
axioms (5)
- standard math Xu–Zikatanov two-level subspace-correction convergence theorem (Theorem 1): rate controlled by μ_c/(C1 C2 c_D) under stable decomposition and strengthened Cauchy–Schwarz-type bounds.
- domain assumption Assumptions 2 (admissible hierarchy): uniform box aspect ratios, nondegenerate connected interiors with mesh-independent Poincaré/approximation constants, O(h) overlap layers, stable unisolvence of owned nodes for Q_{p′} interpolation.
- standard math hp-FEM inverse inequalities and GLL/Fekete basis L2 scaling (Lemmas 2–6) and Bramble–Hilbert/Aubin–Nitsche approximation on agglomerate interiors for p′≥1.
- ad hoc to paper R*-tree ownership yields an algebraic partition of unity on unconstrained DoFs (Remark 5), enabling the Nicolaides-type stable decomposition.
- domain assumption Model problem is Poisson with homogeneous Dirichlet data on shape-regular quasi-uniform meshes; multilinear geometry maps for the sharpest bounds.
invented entities (1)
-
R3MG-C high-order Nicolaides coarse space (continuous nodal interpolant of discontinuous Q_{p′} on R-tree bounding boxes)
independent evidence
read the original abstract
Algebraic multigrid (AMG) methods are robust and efficient black-box preconditioners for linear systems arising from low-order discretizations of elliptic partial differential equations, but their performance often deteriorates for high-order methods. We introduce an algebraic-geometric multilevel preconditioner for continuous lagrangian finite element discretizations. The method automatically constructs prolongation operators and a Galerkin hierarchy using only the coordinates of finite element support points, without requiring a prescribed mesh hierarchy or domain decomposition. The support-point are recursively partitioned by an R-tree algorithm based on axis-aligned bounding boxes, producing a hierarchy of agglomerates. In contrast to classical AMG, which typically uses piecewise-constant aggregate coarse spaces, our method embeds local discontinuous polynomial spaces of degree $p'>0$, defined on the agglomerate boxes, through continuous nodal interpolation. This yields a high-order extension of the Nicolaides coarse space. A two-level analysis quantifies how the coarse polynomial degree offsets the effects of large agglomerates and high-order fine discretizations, under uniform box-regularity, interpolation-stability, and stable-decomposition assumptions. Numerical experiments in two and three dimensions show that the resulting V-cycle, used as a conjugate-gradient preconditioner, maintains stable iteration counts and remains effective in regimes where standard AMG deteriorates.
Figures
Reference graph
Works this paper leans on
-
[1]
Adams, M
M. Adams, M. Brezina, J. Hu, and R. Tuminaro. Parallel multigrid smoothing: polynomial versus Gauss-Seidel.Journal of Computational Physics, 188(2):593–610, 2003
2003
-
[2]
Amestoy, I
P. Amestoy, I. Duff, and J.-Y. L’Excellent. Multifrontal parallel distributed sym- metric and unsymmetric solvers.Computer Methods in Applied Mechanics and En- gineering, 184(2):501–520, 2000
2000
-
[3]
P. F. Antonietti, M. Sarti, and M. Verani. Multigrid algorithms forhp-discontinuous Galerkin discretizations of elliptic problems.SIAM Journal on Numerical Analysis, 53(1):598–618, 2015
2015
-
[4]
Arndt, W
D. Arndt, W. Bangerth, M. Bergbauer, M. Feder, M. Fehling, J. Heinz, T. Heister, L. Heltai, M. Kronbichler, M. Maier, P. Munch, J.-P. Pelteret, B. Turcksin, D. Wells, and S.Zampini. Thedeal.IIlibrary, version 9.5.Journal of Numerical Mathematics, 31(3):231–246, 2023
2023
-
[5]
Pelteret, B
D.Arndt, W.Bangerth, D.Davydov, T.Heister, L.Heltai, M.Kronbichler, M.Maier, J.-P. Pelteret, B. Turcksin, and D. Wells. The deal.II finite element library: Design, features, and insights.Computers & Mathematics with Applications, 81:407–422, Jan. 2021
2021
-
[6]
Babuška and M
I. Babuška and M. Suri. The h-p version of the finite element method with quasiuni- form meshes.ESAIM : Modélisation mathématique et analyse numérique, 21(2):199– 238, 1987
1987
-
[7]
Babuška and M
I. Babuška and M. Suri. The p and h-p versions of the finite element method, basic principles and properties.SIAM Review, 36(4):578–632, 1994
1994
-
[8]
Babuška and M
I. Babuška and M. Suri. The p- and h-p versions of the finite element method, an overview.Computer Methods in Applied Mechanics and Engineering, 80(1):5–26, 1990
1990
-
[9]
Beckmann, H.-P
N. Beckmann, H.-P. Kriegel, R. Schneider, and B. Seeger. The R*-tree: an efficient and robust access method for points and rectangles.SIGMOD Rec., 19(2):322–331, 1990
1990
-
[10]
Boost C++ libraries.https://www.boost.org/, 2024
Boost. Boost C++ libraries.https://www.boost.org/, 2024. 42
2024
-
[11]
L. Bos. Bounding the lebesgue function for lagrange interpolation in a simplex. Journal of Approximation Theory, 38(1):43–59, 1983
1983
-
[12]
L. Bos, M. A. Taylor, and B. A. Wingate. Tensor product gauss-lobatto points are fekete points for the cube.Math. Comput., 70(236):1543–1547, Oct. 2001
2001
-
[13]
A. Brandt. Algebraic multigrid theory: The symmetric case.Applied Mathematics and Computation, 19(1–4):23–56, 1986
1986
-
[14]
S. C. Brenner and L. R. Scott.The Mathematical Theory of Finite Element Methods, volume 15 ofTexts in Applied Mathematics. Springer, New York, NY, 3 edition, 2008
2008
-
[15]
Brezina, A
M. Brezina, A. J. Cleary, R. D. Falgout, V. E. Henson, J. E. Jones, T. A. Manteuffel, S.F.McCormick, andJ.W.Ruge. Algebraicmultigridbasedonelementinterpolation (AMGe).SIAM Journal on Scientific Computing, 22(5):1570–1592, 2001
2001
-
[16]
Canuto, M
C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang.Spectral Methods: Fun- damentals in Single Domains. Scientific Computation. Springer Berlin, Heidelberg, 2006
2006
-
[17]
Chartier, R
T. Chartier, R. D. Falgout, V. E. Henson, J. E. Jones, T. A. Manteuffel, S. F. McCormick, J. W. Ruge, and P. S. Vassilevski. Spectral AMGe (ρAMGe).SIAM Journal on Scientific Computing, 25(1):1–26, 2003
2003
-
[18]
Dolean, P
V. Dolean, P. Jolivet, and F. Nataf.An Introduction to Domain Decomposition Methods. Society for Industrial and Applied Mathematics, Philadelphia, PA, 2015
2015
-
[19]
Dolean, F
V. Dolean, F. Nataf, R. Scheichl, and N. Spillane. Analysis of a two-level schwarz method with coarse spaces based on local dirichlet-to-neumann maps.Computational Methods in Applied Mathematics, 12(4):391–414, 2012
2012
-
[20]
Efendiev, J
Y. Efendiev, J. Galvis, R. Lazarov, and J. Willems. Robust domain decomposi- tion preconditioners for abstract symmetric positive definite bilinear forms.ESAIM: Mathematical Modelling and Numerical Analysis, 46(5):1175–1199, 2012
2012
-
[21]
Feder, A
M. Feder, A. Cangiani, and L. Heltai. R3MG: R-tree based agglomeration of poly- topal grids with applications to multilevel methods.Journal of Computational Physics, 526:113773, 2025
2025
-
[22]
Feder, L
M. Feder, L. Heltai, and A. Cangiani. polyDEAL.https://github.com/fdrmrc/ Polydeal
-
[23]
M. Fekete. Über die verteilung der wurzeln bei gewissen algebraischen gleichungen mit ganzzahligen koeffizienten.Mathematische Zeitschrift, 17(1):228–249, Dec. 1923
1923
-
[24]
Galvis and Y
J. Galvis and Y. Efendiev. Domain decomposition preconditioners for multiscale flows in high contrast media: Reduced dimension coarse spaces.Multiscale Modeling & Simulation, 8(5):1621–1644, 2010. 43
2010
-
[25]
A. Guttman. R-trees: a dynamic index structure for spatial searching.SIGMOD Rec., 14(2):47–57, 1984
1984
-
[26]
M. A. Heroux, R. A. Bartlett, V. E. Howle, R. J. Hoekstra, J. J. Hu, T. G. Kolda, R. B. Lehoucq, K. R. Long, R. P. Pawlowski, E. T. Phipps, A. G. Salinger, H. K. Thornquist, R. S. Tuminaro, J. M. Willenbring, A. Williams, and K. S. Stanley. An overview of the trilinos project.ACM Transactions on Mathematical Software, 31(3):397––423, sep 2005
2005
-
[27]
M. R. Hestenes and E. Stiefel. Methods of conjugate gradients for solving linear systems.Journal of research of the National Bureau of Standards, 49:409–435, 1952
1952
-
[28]
J. Heys, T. Manteuffel, S. McCormick, and L. Olson. Algebraic multigrid for higher- order finite elements.Journal of Computational Physics, 204(2):520–532, 2005
2005
-
[29]
J. E. Jones and P. S. Vassilevski. AMGe based on element agglomeration.SIAM Journal on Scientific Computing, 23(1):109–133, 2001
2001
-
[30]
Manolopoulos, A
Y. Manolopoulos, A. Nanopoulos, A. N. Papadopoulos, and Y. Theodoridis.R-Trees: Theory and Applications. Advanced Information and Knowledge Processing. Springer London, London, 1 edition, 2006
2006
-
[31]
Nataf, H
F. Nataf, H. Xiang, V. Dolean, and N. Spillane. A coarse space construction based on local dirichlet-to-neumann maps.SIAM Journal on Scientific Computing, 33(4):1623–1642, 2011
2011
-
[32]
R. A. Nicolaides. Deflation of conjugate gradients with applications to boundary value problems.SIAM Journal on Numerical Analysis, 24(2):355–365, 1987
1987
-
[33]
Y. Notay. Aggregation-based algebraic multilevel preconditioning.SIAM Journal on Matrix Analysis and Applications, 27(4):998–1018, 2006
2006
-
[34]
S. V. Parter. On the legendre–gauss–lobatto points and weights.Journal of Scientific Computing, 14(4):347–355, 1999
1999
-
[35]
E. M. Rønquist and A. T. Patera. Spectral element multigrid. i. formulation and numerical results.Journal of Scientific Computing, 2:389–406, 1987
1987
-
[36]
J. W. Ruge and K. Stüben. Algebraic multigrid. In S. F. McCormick, editor,Multi- grid Methods, volume 3 ofFrontiers in Applied Mathematics, pages 73–130. SIAM, Philadelphia, PA, 1987
1987
-
[37]
Schwab.p- and hp-Finite Element Methods
C. Schwab.p- and hp-Finite Element Methods. Oxford Science Publications, 1998
1998
-
[38]
J. Shen, T. Tang, and L.-L. Wang.Spectral Methods: Algorithms, Analysis and Applications, volume 41 ofSpringer Series in Computational Mathematics. Springer Berlin, Heidelberg, 2011. 44
2011
-
[39]
Spillane, V
N. Spillane, V. Dolean, P. Hauret, F. Nataf, C. Pechstein, and R. Scheichl. A robust two-level domain decomposition preconditioner for systems of pdes.Comptes Rendus Mathematique, 349(23):1255–1259, 2011
2011
-
[40]
Spillane, V
N. Spillane, V. Dolean, P. Hauret, F. Nataf, C. Pechstein, and R. Scheichl. Abstract robustcoarsespacesforsystemsofpdesviageneralizedeigenproblemsintheoverlaps. Numerische Mathematik, 126(4):741–770, 2014
2014
-
[41]
K. Stüben. A review of algebraic multigrid.Journal of Computational and Applied Mathematics, 128(1–2):281–309, 2001
2001
-
[42]
Sundar, G
H. Sundar, G. Stadler, and G. Biros. Comparison of multigrid algorithms for high- order continuous finite element discretizations.Numerical Linear Algebra with Ap- plications, 22(4):664–680, 2015
2015
-
[43]
M. A. Taylor, B. A. Wingate, and R. E. Vincent. An algorithm for computing fekete points in the triangle.SIAM Journal on Numerical Analysis, 38(5):1707–1720, 2000
2000
-
[44]
Toselli and O
A. Toselli and O. B. Widlund.Domain Decomposition Methods – Algorithms and Theory, volume 34 ofSpringer Series in Computational Mathematics. Springer, 2005
2005
-
[45]
van Lent, R
J. van Lent, R. Scheichl, and I. G. Graham. Energy-minimizing coarse spaces for two-level schwarz methods for multiscale pdes.Numerical Linear Algebra with Ap- plications, 16:775–799, 2009
2009
-
[46]
Vaněk, M
P. Vaněk, M. Brezina, and J. Mandel. Convergence of algebraic multigrid based on smoothed aggregation.Numerische Mathematik, 88:559–579, 2001
2001
-
[47]
Vaněk, J
P. Vaněk, J. Mandel, and M. Brezina. Algebraic multigrid by smoothed aggregation for second and fourth order elliptic problems.Computing, 56:179–196, 1996
1996
-
[48]
J. Xu. Iterative methods by space decomposition and subspace correction.SIAM Review, 34(4):581–613, 1992
1992
-
[49]
Xu and L
J. Xu and L. T. Zikatanov. Algebraic multigrid methods.Acta Numerica, 26:591–721, 2017
2017
-
[50]
L. T. Zikatanov. Two-sided bounds on the convergence rate of two-level methods. Numerical Linear Algebra with Applications, 15(5):439–454, 2008. 45
2008
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.