Pith. sign in

REVIEW 2 major objections 4 minor 33 references

An Auxiliary Space Preconditioner for Fractional Laplacian of Negative Order

T0 review · 2 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Negative fractional Laplacians can be preconditioned through positive fractional H(div) operators via the discrete gradient.

desk verdict A clean, honest transfer argument for negative-order fractional Laplacian preconditioning, with the additive multigrid leg left conditional on unproven two-level estimates—worth refereeing despite the gap. read the letter →

arxiv 1908.04498 v1 pith:6NME6JVY submitted 2019-08-13 math.NA cs.NA

Trygve B{ae}rland This is my paper · ORCID
classification math.NAcs.NA MSC 65F0865N5565N3035R11
keywords fractionalLaplaciannegativeSobolevspacesauxiliaryspacepreconditioneradditivemultigridH(div)operatorsRaviart-Thomaselementsspectralequivalenceoperatorinterpolation
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

This paper attacks a problem that arises when coupled physics problems are solved with interface conditions posed in fractional Sobolev spaces: solving equations that involve the fractional Laplacian with a negative exponent, $(-\Delta)^{-s}$ for $s\in[0,1]$. Because the offending eigenvectors are the smooth ones, ordinary multigrid smoothing cannot handle the operator by itself; the author instead transfers the problem to a positive-order operator on the divergence side. The central result is that $B_s^h=\nabla_h^* B_{\mathrm{div},h}^{1-s}\nabla_h$ is uniformly spectrally equivalent to $A_h^{-s}$ whenever $B_{\mathrm{div},h}^{1-s}$ is a uniformly spectrally equivalent preconditioner for the fractional $H(\operatorname{div})$ operator $\Lambda_h^{-(1-s)}$, with constants $C_1\beta^{2(1-s)}$ and $C_2$. The paper then constructs $B_{\mathrm{div},h}^{s}$ as an additive multigrid operator built from vertex-patch smoothers, proves the upper bound and a conditional lower bound, and shows numerically that condition numbers stay flat. The stated caveat is that the lower-bound proof rests on two-level error estimates that the author leaves as future work.

What carries the argument

The load-bearing object is the discrete Helmholtz decomposition $V_h=\operatorname{curl}C_h\oplus\nabla_hS_h$ together with the operator $\Lambda_h=I+\nabla_h\nabla_h^*$ on $V_h$. Since $\Lambda_h$ commutes with both subspaces, its fractional power acts as the identity on $\operatorname{curl}C_h$ and like $(I+A_h)^s$ on $\nabla_hS_h$, so preconditioning $\Lambda_h^s$ reduces to a scalar elliptic problem plus a trivial curl part. The auxiliary-space identity $B_s^h=\nabla_h^*B_{\mathrm{div},h}^{1-s}\nabla_h$ transfers the negative fractional Laplacian into that positive fractional $H(\operatorname{div})$ problem. The additive multigrid operator $\sum_k R_k^sQ_k$ with vertex-patch smoothers supplies the $H(\operatorname{div})$ factor, and Lemma 2.1, a Jensen operator inequality for contractions between different Hilbert spaces, carries the bounds from the endpoint cases to all $s\in(0,1)$.

What would settle it

Compute the ratios $\|\nabla_k u\|^2/(h_{k-1}^{2s}(\Lambda_k^s\tau,\tau))$ and $\|q\|/(h_{k-1}\|\tau\|)$ over a sequence of uniformly refined meshes for $\tau\in(I-P^s_{k,k-1})V_k$ with the Helmholtz decomposition (4.19); if either ratio grows without bound as $h_k\to0$, the stable decomposition in Lemma 4.2 fails and the additive multigrid preconditioner cannot be uniformly spectrally equivalent to $\Lambda_h^{-s}$.

Watch

Extended reading notes

Core claim

On a shape-regular triangulation, $S_h$ is the space of piecewise polynomials, $V_h=\mathrm{RT}_r(\mathcal{T}_h)$ is the Raviart-Thomas space, and the discrete gradient $\nabla_h:S_h\to V_h$ is defined by $(\nabla_h u,\tau)=-(u,\operatorname{div}\tau)$. Theorem 3.2 proves the two-sided inequality $\beta^{2(1-s)}A_h^s\le\nabla_h^*\Lambda_h^{-(1-s)}\nabla_h\le A_h^s$, and Corollary 3.1 converts any $B_{\mathrm{div},h}^{1-s}$ satisfying (3.24) into $C_1\beta^{2(1-s)}A_h^s\le B_s^h\le C_2A_h^s$. The proof interpolates between the endpoint cases $s=0$ and $s=1$ through Jensen's operator inequality. For the $H(\operatorname{div})$ factor, the additive multigrid operator $B_{\mathrm{div},h}^s=\sum_{k=1}^J R_k^s Q_k$ uses vertex-patch additive Schwarz smoothers $R_k^s=\sum_\nu\Lambda_{k,\nu}^{-s}Q_{k,\nu}$, and Theorem 4.1 gives spectral equivalence to $\Lambda_h^{-s}$ under the smoother assumptions (4.7) and (4.8). The lower-bound half is conditional: Lemma 4.2 reduces (4.8) to the two-level estimates (4.20), and Remark 3 explicitly states that proving them is left as future work.

Load-bearing premise

The load-bearing premise is the pair of two-level error estimates (4.20): for an error $\tau\in(I-P^s_{k,k-1})V_k$ with discrete Helmholtz decomposition $\tau=\nabla_k u+\operatorname{curl}q$, one must have $\|\nabla_k u\|^2\le c h_{k-1}^{2s}(\Lambda_k^s\tau,\tau)$ and $\|q\|\le ch_{k-1}\|\tau\|$. The paper uses these estimates in Lemma 4.2 to verify the stable decomposition that powers the lower bound, and Remark 3 leaves their proof to future work, so the spectral equivalence of the multigrid preconditioner is conditional on them.

Editorial extensions

If this is right

  • For any $s\in[0,1]$, a preconditioner $B_{\mathrm{div},h}^{1-s}$ satisfying (3.24) lifts to a preconditioner for $A_h^{-s}$ with h-independent constants $C_1\beta^{2(1-s)}$ and $C_2$, so the negative-order problem inherits the quality of the positive-order H(div) preconditioner.
  • If the two-level estimates (4.20) hold, the additive multigrid operator $B_{\mathrm{div},h}^s$ is uniformly spectrally equivalent to $\Lambda_h^{-s}$ for every $s\in[0,1]$ by interpolation between the cases $s=0$ and $s=1$.
  • The numerical experiments keep conjugate-gradient iteration counts and estimated condition numbers flat across four levels of refinement for both $\Lambda_h^s$ and $A_h^{-s}$, consistent with the proven upper bound and the conditional lower bound.
  • Applied to the trace-constrained system (1.2), the new operator supplies the fractional block of an ideal block-diagonal preconditioner without requiring a global fractional power of the Laplacian.

Reading between the lines

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

  • A sharper stress test would fix $J$ and push refinements deeper; the theorem's upper bound scales linearly in $J$, so condition numbers should grow at most linearly in $J$, and the flat table with $J=4$ cannot distinguish the proven upper bound from a failing lower bound.
  • The integral-representation route sketched in Remark 3, if completed, would tie the fractional H(div) two-level estimates to standard estimates for $(I-t^2\Delta)^{-1}$ and would identify the exact domain-regularity condition needed, likely making the proof transfer to nonuniform meshes.
  • Because the argument only uses the inf-sup condition (3.15), a commuting Helmholtz decomposition, and stable vertex-patch decompositions, the same two-stage construction should carry over to higher-order Raviart-Thomas spaces and to three dimensions.
  • For the motivating trace problem, the construction would replace a fractional operator on the interface by standard H(div) multigrid on the interface mesh; the remaining difficulty would then be geometric approximation of the trace space by coarse interface meshes.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper proposes an auxiliary-space preconditioner for discretizations of the fractional Laplacian of negative order. Section 3 develops a clean argument, both continuous and discrete, showing that B_s^h = ∇_h^* B_{div,h}^{1-s} ∇_h is spectrally equivalent to A_h^{-s} provided B_{div,h}^{1-s} satisfies the two-sided bound (3.24). Section 4 attempts to supply such a B_{div,h}^{1-s} as the additive multigrid operator (4.6). Theorem 4.1 gives sufficient conditions on the smoothers, and Lemma 4.2 verifies the main condition under two-level error estimates (4.20) that Remark 3 explicitly leaves unproven. Numerical experiments with a fixed number of levels (J=4) report bounded iteration counts and condition numbers.

Significance. The Section 3 transfer argument is elegant, correctly uses interpolation and Jensen's operator inequality, and is of independent interest. If the missing estimates were supplied, the paper would give a useful preconditioning strategy for the fractional interface problems that motivate it. The author is clearly aware of the main theoretical gap and states it honestly. However, as submitted the central construction of B_{div,h}^{1-s} is conditional, and the numerical experiments, while consistent with the theory, do not test the level-dependence that the theory still leaves open.

major comments (2)
  1. [Section 4, Lemma 4.2 and Remark 3] The verification of Assumption (4.8) rests entirely on the two-level estimates (4.20), which are stated as assumptions in Lemma 4.2 and explicitly left unproven in Remark 3. Without (4.20), the stable decomposition (4.21), Assumption (4.8), and therefore the lower bound in Theorem 4.1 are not established. Consequently, the paper does not prove that the additive multigrid operator (4.6) satisfies the two-sided bound (3.24), and Corollary 3.1 cannot be invoked to deliver the advertised spectral equivalence for the negative-order preconditioner. The proposal in Remark 3 to derive (4.20) through the techniques of [9] is a strategy, not a proof; this is a load-bearing gap rather than a presentation issue.
  2. [Theorem 4.1, Eq. (4.9)] Even if (4.20) were available, the upper bound in (4.9) contains the explicit factor J, the number of grid levels. Since J grows as the mesh is refined, the asserted spectral equivalence constants are not independent of the discretization unless this factor is removed or otherwise shown to be harmless. The numerical experiments in Tables 1 and 3 keep J=4 in every run, so they do not exercise the level-dependence of the bound; reporting results with increasing J and a fixed coarsest mesh would be needed to support an h-uniform spectral equivalence claim.
minor comments (4)
  1. [Remark 2, matrix realization] The matrix entries (D_h)_{i,j} = -(φ_i^h, div ψ_i^h) appear to have an index typo; they should presumably involve ψ_j^h so that the matrix maps coefficient vectors on S_h to dual vectors on V_h.
  2. [Section 4 and Section 5] There are several typographical errors, including 'charachteristic' in Section 4, 'adressed' in Section 4, 'Bogovski ˘ ı' in Section 3, and 'preconditioned conjugate method' in Section 5.2.
  3. [Tables 1-3] The captions should state explicitly how the fixed number of levels J=4 is chosen relative to N; if the coarsest mesh changes with N, the experiments do not isolate the level-dependence of the method.
  4. [References] Reference [5] is cited as an arXiv preprint; the authors should update it to the published version if one exists, and should indicate more explicitly which parts of the present analysis rely on results from that preprint.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the auxiliary-space reduction and the additive multigrid construction are genuine two-step arguments, and the unproven two-level estimates are an acknowledged gap rather than an input-output identification.

full rationale

The central derivation is self-contained and does not reduce to its own inputs. Section 3 proves Theorem 3.2 using operator inequalities and interpolation arguments: Lemma 2.1 gives T* A^s T <= (T* A T)^s for contractions, Lemma 3.1 supplies the discrete bounds on the Helmholtz-decomposition left-inverse L, and the inequalities (3.20) follow from these ingredients alone. Corollary 3.1 is a conditional reduction: if B_div^{1-s}_h satisfies the two-sided bound (3.24), then B_h^s = ∇_h^* B_div^{1-s}_h ∇_h inherits spectral equivalence to A_h^{-s} by combining (3.24) with (3.20). This is not circular because the hypothesis is independent of the conclusion; the paper does not define B_div^{1-s}_h in terms of A_h^{-s}. Section 4 attempts to supply such a B_div^{1-s}_h by the additive multigrid operator (4.6). Theorem 4.1 proves spectral equivalence under assumptions (4.7) and (4.8), and Lemma 4.1 verifies (4.7). The remaining assumption (4.8) is addressed in Lemma 4.2 only conditionally on the two-level estimates (4.20), which the paper explicitly leaves unproven: the introduction states that the analysis 'assumes certain two-level error estimates on Λ^{1-s} that will go unproven in this work,' and Remark 3 says verifying the assumption of Lemma 4.2 'falls beyond the scope of this paper.' This is an honest, clearly identified gap in the proof of full spectral equivalence, not a circular step: the result is not obtained by assuming the conclusion. The self-citations to [5] are motivational and not load-bearing; Theorem 4.1's analysis is carried out in the paper, and the proposed route to (4.20) cites [9], not [5]. The numerical experiments are consistent with the theory but use a fixed number of levels J=4, so they do not exercise level-independence; this is a completeness concern, not circularity.

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

The central derivation rests on standard interpolation and operator inequalities, plus two independent stability results for Raviart-Thomas spaces. The main addition to this ledger is the unproven two-level error estimate (4.20), which the paper explicitly leaves for future work.

assumptions (5)
  • standard math Interpolation theory and operator monotonicity (Löwner-Heinz, Jensen's operator inequality, Lemma 2.1)
    Used throughout Section 3 to establish spectral equivalences; cited to [22], [18], [19].
  • standard math Bogovskii operator T exists with mapping properties (3.6) on star-shaped (or Lipschitz) domains
    Used in proof of Theorem 3.1; established in [15].
  • standard math Discrete inf-sup condition (3.15) for Raviart-Thomas spaces with β independent of h
    Used to show A_h is SPD and to prove Lemma 3.1; from [8].
  • standard math L2-stable decomposition of V_k into vertex patches (4.15)
    Used in definition and analysis of additive Schwarz smoothers; from [2,3].
  • ad hoc to paper Two-level error estimates (4.20) for fractional H(div) problems
    Assumed in Lemma 4.2 to prove Assumption 4.8; explicitly left unproven in Remark 3. This is the load-bearing gap of the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An Auxiliary Space Preconditioner for Fractional Laplacian of Negative Order." pith.science (2026). https://pith.science/paper/6NME6JVY

@misc{pith2026190804498,
  author       = {Pith},
  title        = {Pith review of: An Auxiliary Space Preconditioner for Fractional Laplacian of Negative Order},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6NME6JVY}},
  note         = {Machine review of arXiv:1908.04498}
}
abstract

Coupled multiphysics problems often give rise to interface conditions naturally formulated in fractional Sobolev spaces. Here, both positive and negative fractionality are common. When designing efficient solvers for discretizations of such problems it would then be useful to have a preconditioner for the fractional Laplacian, $(-\Delta)^s$, with $s \in [-1,1]$. Previously, additive multigrid preconditioners for the case when $s \geq 0$ have been proposed. In this work we complement this construction with auxiliary space preconditioners suitable when $s \leq 0$. These preconditioners are shown to be spectrally equivalent to $(-\Delta)^{-s}$, but requires preconditioners for fractional $H(\operatorname{div})$ operators with positive fractionality. We design such operators based on an additive multigrid approach. We finish with some numerical experiments, verifying the theoretical results.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 32 canonical work pages

  1. [9]

    Bonito and J

    A. Bonito and J. E. Pasciak. Numerical approximation of f ractional powers of elliptic operators. Math. Comp. , 84(295):2083–2110, 2015

    work page 2015

  2. [1]

    Arioli and D

    M. Arioli and D. Loghin. Discrete interpolation norms wi th applications. SIAM J. Numer. Anal. , 47(4):2924– 2951, 2009

    work page 2009

  3. [2]

    D. N. Arnold, R. S. Falk, and R. Winther. Preconditioning in H(div) and applications. Math. Comp., 66(219):957– 984, 1997

    work page 1997

  4. [3]

    D. N. Arnold, R. S. Falk, and R. Winther. Multigrid in H(div) and H(curl). Numer. Math. , 85(2):197–217, 2000

    work page 2000

  5. [4]

    I. Babuška. The finite element method with Lagrangian mul tipliers. Numer. Math. , 20:179–192, 1972/73

    work page 1972

  6. [5]

    Multigrid Methods for Discrete Fractional Sobolev Spaces

    T. Bærland, M. Kuchta, and K.-A. Mardal. Multigrid metho ds for discrete fractional sobolev spaces. arXiv preprint arXiv:1806.00222, 2018

    work page Pith review arXiv 2018

  7. [6]

    Bertoluzza, V

    S. Bertoluzza, V. Chabannes, C. Prud’homme, and M. Szopo s. Boundary conditions involving pressure for the Stokes problem and applications in computational hemod ynamics. Comput. Methods Appl. Mech. Engrg. , 322:58–80, 2017

    work page 2017

  8. [7]

    M. S. Birman and M. Z. Solomjak. Spectral theory of selfadjoint operators in Hilbert space . Mathematics and its Applications (Soviet Series). D. Reidel Publishing Co. , Dordrecht, 1987. Translated from the 1980 Russian original by S. Khrushchëv and V. Peller

    work page 1987

Show all 33 references
  1. [8]

    D. Boffi, F. Brezzi, and M. Fortin. Mixed finite element methods and applications , volume 44 of Springer Series in Computational Mathematics . Springer, Heidelberg, 2013

    2013

  2. [10]

    J. H. Bramble. Multigrid methods , volume 294 of Pitman Research Notes in Mathematics Series . Longman Scientific & Technical, Harlow; copublished in the United St ates with John Wiley & Sons, Inc., New York, 1993

    1993

  3. [11]

    J. H. Bramble, Z. Leyk, and J. E. Pasciak. The analysis of multigrid algorithms for pseudodifferential operators of order minus one. Math. Comp. , 63(208):461–478, 1994

    1994

  4. [12]

    J. H. Bramble, J. E. Pasciak, and P. S. Vassilevski. Comp utational scales of Sobolev norms with application to preconditioning. Math. Comp. , 69(230):463–480, 2000

    2000

  5. [13]

    J. H. Bramble, J. E. Pasciak, and J. Xu. Parallel multile vel preconditioners. Math. Comp. , 55(191):1–22, 1990

    1990

  6. [14]

    J. H. Bramble, J. E. Pasciak, and J. Xu. The analysis of mu ltigrid algorithms with nonnested spaces or nonin- herited quadratic forms. Math. Comp. , 56(193):1–34, 1991

    1991

  7. [15]

    Costabel and A

    M. Costabel and A. McIntosh. On Bogovski ˘ ı and regulari zed Poincaré integral operators for de Rham complexes on Lipschitz domains. Math. Z. , 265(2):297–320, 2010

    2010

  8. [16]

    C. Davis. A Schwarz inequality for convex operator func tions. Proc. Amer. Math. Soc. , 8:42–44, 1957

    1957

  9. [17]

    S. A. Funken and E. P. Stephan. The bpx preconditioner fo r the single layer potential operator. Applicable Analysis, 67(3-4):327–340, 1997

    1997

  10. [18]

    Hansen and G

    F. Hansen and G. Kjærgå rd Pedersen. Jensen’s inequalit y for operators and Löwner’s theorem. Math. Ann. , 258(3):229–241, 1982

    1982

  11. [19]

    Hansen and G

    F. Hansen and G. K. Pedersen. Jensen’s operator inequal ity. Bull. London Math. Soc. , 35(4):553–564, 2003

    2003

  12. [20]

    Hiptmair

    R. Hiptmair. Multigrid method for H(div) in three dimensions. Electron. Trans. Numer. Anal., 6(Dec.):133–152,

  13. [21]

    Hiptmair and J

    R. Hiptmair and J. Xu. Nodal auxiliary space preconditi oning in H(curl) and H(div) spaces. SIAM J. Numer. Anal., 45(6):2483–2509, 2007

    2007

  14. [22]

    T. Kato. Notes on some inequalities for linear operator s. Math. Ann. , 125:208–212, 1952

    1952

  15. [23]

    T. V. Kolev and P. S. Vassilevski. Parallel auxiliary sp ace AMG solver for H(div) problems. SIAM J. Sci. Comput., 34(6):A3079–A3098, 2012

    2012

  16. [24]

    Kuchta, K.-A

    M. Kuchta, K.-A. Mardal, and M. Mortensen. Preconditio ning trace coupled 3d-1d systems using fractional Laplacian. arXiv preprint arXiv:1612.03574 , 2016

    2016 arXiv

  17. [25]

    Kuchta, M

    M. Kuchta, M. Nordaas, J. C. G. Verschaeve, M. Mortensen , and K.-A. Mardal. Preconditioners for saddle point systems with trace constraints coupling 2d and 1d domains. SIAM J. Sci. Comput. , 38(6):B962–B987, 2016

    2016

  18. [26]

    I. V. Lashuk and P. S. Vassilevski. The construction of t he coarse de Rham complexes with improved approxi- mation properties. Comput. Methods Appl. Math. , 14(2):257–303, 2014

    2014

  19. [27]

    Lions and E

    J.-L. Lions and E. Magenes. Non-homogeneous boundary value problems and applications . Vol. I . Springer-Verlag, New York-Heidelberg, 1972. Translated from the French by P. Kenneth, Die Grundlehren der mathematischen Wissenschaften, Band 181

    1972

  20. [28]

    Mardal and R

    K.-A. Mardal and R. Winther. Preconditioning discreti zations of systems of partial differential equations. Numer. Linear Algebra Appl. , 18(1):1–40, 2011

    2011

  21. [29]

    P. Oswald. Multilevel norms for H-1/2. Computing, 61(3):235–255, 1998

    1998

  22. [30]

    Stevenson and R

    R. Stevenson and R. van Venetië. Optimal preconditioni ng for problems of negative order. arXiv preprint arXiv:1803.05226, 2018. 20

    2018 arXiv

  23. [31]

    Tveito, K

    A. Tveito, K. H. Jæger, M. Kuchta, K.-A. Mardal, and M. E. Rognes. A cell-based framework for numerical modeling of electrical conduction in cardiac tissue. Front. Phys., 5:48, 2017

    2017

  24. [32]

    J. Xu. The auxiliary space method and optimal multigrid preconditioning techniques for unstructured grids. Computing, 56(3):215–235, 1996. International GAMM-Workshop on Mul ti-level Methods (Meisdorf, 1994). 21

    1996

  25. [1997]

    Special issue on multilevel methods (Copper Mountain , CO, 1997)

    1997

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.