Pith. sign in

REVIEW 2 major objections 4 minor 36 references

Coarse space preconditioning for Generalized Optimized Schwarz Methods. Part I: continuous case

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 The Generalized Optimized Schwarz Method's scattering operator is a compact perturbation of i times the identity, so its coarse-space-preconditioned GMRES converges superlinearly.

desk verdict First coarse space for GOSM, built on a genuinely new compactness observation, but the central hypothesis on T is unverified and fails for the standard local impedance, so the scope is narrower than advertised. read the letter →

arxiv 2608.08505 v1 pith:DOTM34WN submitted 2026-08-09 math.NA cs.NA

Xavier Claeys This is my paper · ORCID
classification math.NAcs.NA MSC 65N5565F1035J0547B06
keywords generalizedoptimizedSchwarzmethodcoarsespacedomaindecompositionHelmholtzequationGMREScompactperturbationSchattenclassparametrix
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 aims to show that the Generalized Optimized Schwarz Method (GOSM), a non-overlapping domain decomposition method for harmonic wave propagation, is low-rank compressible in a precise sense: its local scattering operator S differs from i times the identity by a compact operator. Working entirely at the continuous level, the author proves that the preconditioned skeleton equation is identity plus compact, so GMRES converges superlinearly with a rate controlled by the singular values of S−iId. The coarse space is the span of the leading singular modes of S−iId, and the Woodbury formula makes the coarse solve practical. If the argument holds, convergence of the preconditioned iteration depends on how fast those singular values decay, not directly on the number of subdomains. This first part is continuous; a discrete counterpart is announced in the paper.

What carries the argument

The object doing the work is S−iId, the deviation of the subdomain scattering operator from i times the identity. Lemma 5.1 shows that replacing A and T by the reference operators A+ and T+ makes S exactly iId; Lemma 5.2 then shows that the same formula with the true operators is a compact perturbation. Corollary 5.3 lifts this to compactness of (Id+ΠS)^4+4Id using Π²=Id, which is what fits Moret's GMRES theorem. The refined preconditioner is built from the truncated singular value decomposition of S−iId, and the Woodbury formula (45) inverts the resulting rank-r update. The single identity that carries the argument is (Id+ΠS)^4 = −4(Id+X)^4 with X = (Π−iId)(S−iId)/2, which turns the compact tail of S−iId into a compact tail of the preconditioned operator.

What would settle it

Take a two-subdomain Helmholtz problem in a square, choose T so that T−T+ is a rank-one smoothing perturbation, compute the singular values σ_j(S−iId) and the GMRES residuals for the preconditioned equation (46), and compare with the product bound (47): the central claim fails if residuals do not outpace the Elman geometric bound (27) while the singular-value product tends to zero. Conversely, taking T as a local Robin impedance (multiplication by a constant) should make σ_j(S−iId) plateau away from zero, showing that the compact-perturbation premise is what the construction needs.

Watch

Extended reading notes

Core claim

The central claim is Lemma 5.2: under the hypothesis that the impedance operator T used in the transmission condition differs from the reference positive-definite Dirichlet-to-Neumann operator T+ by a compact perturbation, S−iId is compact as a map from H(Σ)′ to H(Σ)′. The proof factors S−iId as 2iTB(A−iB*TB)^{-1}(A+−A−iB*(T+−T)B)(A+−iB*T+B)^{-1}B*, using Rellich compactness for A−A+ and the assumed compactness of T−T+. Because Π²=Id, this gives compactness of (Id+ΠS)^4+4Id, and via Moret's singular-value theorem it yields superlinear GMRES convergence for the unpreconditioned equation. The refined preconditioner replaces S by Sr=iId+Kr, where Kr is a rank-r truncation of S−iId; the preconditioned operator (Id+ΠSr)^{-1}(Id+ΠS) is then identity plus a residual whose GMRES factors are σ_{r+j}(S−iId)/γ⋆. The coarse space of dimension r is therefore the span of the leading singular modes of S−iId, and Proposition 6.4 converts this tail into a convergence bound.

Load-bearing premise

The load-bearing premise is that the impedance operator T can be chosen so that T−T+ is compact (and Schatten-p for the quantitative rates), and that the coercivity constant γ⋆ from (25) is strictly positive; the paper assumes, rather than constructs, such a T for the Helmholtz problem (1). If T−T+ is not compact, S−iId need not have a decaying singular-value tail, and all the superlinear estimates lose their engine.

Editorial extensions

If this is right

  • GMRES on the unpreconditioned GOSM skeleton equation converges superlinearly, at a rate governed by the singular values of (Id+ΠS)^4+4Id divided by γ⋆^4 (Estimate (31)).
  • The simple preconditioner (Id−iΠ)/2 gives the cleaner bound ∏_{j=1}^n σ_j(S−iId)/γ⋆, removing the fourth power on γ⋆ (Lemma 6.1).
  • With a rank-r coarse space from the truncated SVD of S−iId, the GMRES residual bound shifts to the tail singular values σ_{r+j}(S−iId), so enlarging r can compensate a small γ⋆ (Proposition 6.4).
  • When A−A+ and T−T+ lie in a Schatten p-class, the product bounds become explicit rates involving (∥S−S_r∥_{C_p}/(γ⋆ n^{1/p}))^n (Corollaries 5.7, 6.2, 6.5).
  • The coarse-space solve is practical because the Woodbury formula reduces it to inverting an r×r matrix (Equation (45)).

Reading between the lines

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

  • This is effectively a parametrix construction: the coarse space is a finite-dimensional inverse of the compact tail, so the same template could apply to other wave domain-decomposition formulations whose scattering maps are compact perturbations of constants.
  • The continuous analysis suggests a discrete-hindsight prediction: on fine meshes, the convergence rate of the preconditioned iteration should track the decay of the discrete singular values of S−iId, so adaptively choosing r to capture all singular values above γ⋆ should make the method insensitive to frequency and to the number of subdomains.
  • The paper leaves the choice of T open; if a concrete non-local impedance can be constructed for which the Schatten p-norm ∥T−T+∥_{C_p} is small, the convergence bounds would become quantitative rather than existential.
  • Because the analysis is at the continuous level, a cheap numerical check of the singular-value tail on a family of subdomain partitions would test whether the predicted superlinear phase appears in practice.
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.

Referee Report

2 major / 4 minor

Summary. The paper analyzes a continuous-level skeleton formulation of the Generalized Optimized Schwarz Method (GOSM) for the Helmholtz problem (1) in a non-overlapping domain decomposition, written as (Id + ΠS)p = f. Under the standing assumption that the impedance operator T differs from the reference DtN operator T+ by a compact (eventually Schatten-p) perturbation, the paper proves that S − iId is compact, derives superlinear GMRES convergence bounds via Moret's theorem, and proposes two preconditioners: a first-order one based on (Id − iΠ)/2 and a refined coarse-space preconditioner built from a rank-r truncated SVD Sr = iId + Kr of S, inverted by the Woodbury formula. The main results are Proposition 6.4 and Corollary 6.5, which bound the GMRES residual of the preconditioned equation (46) by products of tail singular values σ_{r+j}(S − iId), yielding superlinear convergence controlled by the singular-value decay.

Significance. The paper is a credible and largely correct continuous-level analysis of a parametrix for GOSM. The algebraic chain from Lemma 5.1 to Corollary 5.3 is clean, the use of Moret's theorem is appropriate, and the Woodbury-based coarse-space inversion in Section 6.2 is a concrete algorithmic proposal. If the standing compactness assumption on T is satisfied, the proposed coarse space removes the subdomain-count and inf-sup dependence from the asymptotic convergence rate, and the explicit Schatten-class estimates give quantitative rates. The main weakness is that the key hypothesis is not instantiated: apart from the passing mention of T = T+, the paper gives no concrete nonlocal T satisfying the compactness/Schatten assumption, and it does not analyze the cost of applying such a T. This limits the applicability of the claimed construction as it stands, but the issue is local and fixable within the manuscript's scope.

major comments (2)
  1. [Section 4(a), Lemma 5.2, Corollaries 5.7, 6.2, 6.5] The entire low-rank compressibility statement of the paper rests on the standing assumption in Section 4(a) that T − T+ is compact (or Schatten-p). The manuscript does not exhibit any concrete T for problem (1) other than T = T+, mentioned in passing in Section 5.2. For the standard local impedance T = ηId used in optimized Schwarz methods, T − T+ is not compact: T+ is an order-one isomorphism H^{1/2}(∂Ωj) → H^{-1/2}(∂Ωj), whereas ηId is compact in that scale. The construction therefore fails for the usual Robin-type transmission condition. Please either (i) prove that a class of admissible nonlocal impedance operators satisfies the compactness/Schatten assumption and discuss the cost of applying them, or (ii) state explicitly and prominently that the method is restricted to such T, with T = T+ as the only known example. As written, the reader cannot tell whether the proposed coarse space is applicable to the GOSM instances that motivated the paper.
  2. [Lemma 5.2 proof and Lemma 5.6] The factorization displayed in the proof of Lemma 5.2 is incomplete. Setting R = (A − iB*TB)^{-1} and R+ = (A+ − iB*T+B)^{-1}, a correct derivation gives S − iId = 2i[(T − T+)B R+ B* + T B R(A+ − A − iB*(T+ − T)B)R+ B*], whereas the proof drops the term 2i(T − T+)B R+ B*. The compactness conclusion survives because this extra term is compact, but the Schatten bound in Lemma 5.6 and the subsequent quantitative estimates in Corollaries 5.7, 6.2, and 6.5 need an additional contribution involving ∥T − T+∥_{C_p} with a constant factor, and they are not supported by the proof as stated. Please correct the factorization and re-derive the constants.
minor comments (4)
  1. [Throughout] The abbreviation 'GMRes' is used consistently but is nonstandard; the conventional spelling is 'GMRES' (see [30,31]). Please make the notation consistent.
  2. [Section 5.2] The sentence referring to 'the explicit expression (2)' should refer to formula (22) for the scattering operator S, not to the variational form (2).
  3. [Introduction and Section 6.2] Typos: 'the later' should be 'the latter', 'developpement' should be 'development', 'reasonnable' should be 'reasonable', and 'equiped' should be 'equipped'. Also, 'Stecklov-Poincaré' should be 'Steklov-Poincaré'.
  4. [Section 5 (after Eq. (31))] The inequality '⌊n/4⌋ ≤ n/4 ≤ ⌊n/4⌋ + 1' should have a strict inequality '⌊n/4⌋ ≤ n/4 < ⌊n/4⌋ + 1'; the subsequent argument is unaffected.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the superlinear convergence claims follow from an explicit compactness hypothesis and standard external theorems, not from a fitted or self-referential construction.

full rationale

The derivation is a conditional analysis, not a self-fulfilling prediction. Section 4(a) explicitly assumes that the impedance operator T differs from the reference DtN operator T+ by a compact (later Schatten-p) perturbation. Lemma 5.2 derives compactness of S−iId from that assumption together with the compact embedding of H^1 into L^2, via a resolvent factorization; the conclusion is not the same operator as the hypothesis, and no parameter is fitted. The coarse space is then defined as the truncated SVD of S−iId, and the convergence bounds follow from Moret's theorem, Woodbury's formula, and Schatten-class estimates. These are external, parameter-free results. Self-citations to [3,4,6] supply the GOSM formalism and the coercivity bound (25); this is ordinary reliance on prior published theorems, and the superlinear convergence itself rests on compactness rather than on the numerical value of gamma-star. The lack of a concrete T satisfying hypothesis 4(a) and the apparent algebraic gap in the factorization displayed in Lemma 5.2 are correctness and scope concerns, not circularity.

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

The analysis is parameter-free: no numbers are fitted to data, and the coarse-space rank r is a free design parameter whose adequacy condition sigma_{r+1}(S - iId) < gamma* is stated rather than fitted. The load-bearing hypotheses are: well-posedness of the Helmholtz problem (alpha > 0 in (3)); the impedance operator T is positive definite self-adjoint with T - T+ compact (Section 4a), strengthened to the Schatten p-class in Lemma 5.6; and the whole GOSM framework including the coercivity bound (25) imported from the author's prior papers [3,4,6]. Standard functional analysis tools (Moret's theorem, Woodbury's formula, Schatten theory, Weyl's laws) are used as background.

assumptions (5)
  • domain assumption The Helmholtz boundary value problem is uniquely solvable: the inf-sup constant alpha in (3) is strictly positive.
    Standard well-posedness assumption for the continuous problem; it feeds the coercivity bound (25) on which every GMRES estimate in Sections 5 and 6 depends.
  • domain assumption The impedance operator T is subdomainwise block-diagonal, positive definite, self-adjoint, and T - T+ is compact (Section 4, item a).
    Load-bearing for Lemma 5.2: it is what makes S - iId compact and hence low-rank compressible. The paper assumes this for T rather than deriving it for a concrete choice attached to problem (1); only the trivial case T = T+ is explicitly mentioned (end of Section 5.2).
  • domain assumption A - A+ and T - T+ belong to the Schatten p-class for some p >= 1 (Lemmas 5.6, Corollaries 5.7, 6.2, 6.5).
    Strengthens compactness into quantitative singular-value decay, needed for the explicit convergence rates. For A - A+ the paper sketches a Weyl's law argument (end of Section 5.2); for T - T+ it remains an assumption on the transmission condition.
  • domain assumption The GOSM framework is imported from prior work: forms of Pi (20) and S (22), equivalence Lemma 4.1, isometry (21), contractivity (23), and the coercivity bound (25) from [3,4,6].
    These are the author's own published results; importing them is standard, but it means the quantitative backbone of the paper is not re-derived here and carries the correction noted in the red flags.
  • standard math Background functional analysis: compact operators form a two-sided ideal; singular-value characterization of compactness and Schatten classes; Moret's theorem (Thm 5.4); Woodbury formula; Weyl's laws for Neumann Laplacian and Poincare-Steklov eigenvalues.
    Standard cited machinery (Birman-Solomyak [1], Ringrose [26], Moret [21], Woodbury [36], Netrusov-Safarov [23], Rozenblum [27]); assumed correct.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Coarse space preconditioning for Generalized Optimized Schwarz Methods. Part I: continuous case." pith.science (2026). https://pith.science/paper/DOTM34WN

@misc{pith2026260808505,
  author       = {Pith},
  title        = {Pith review of: Coarse space preconditioning for Generalized Optimized Schwarz Methods. Part I: continuous case},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DOTM34WN}},
  note         = {Machine review of arXiv:2608.08505}
}
read the original abstract

The Generalized Optimized Schwarz Method (GOSM) originally proposed in [Claeys, 2021] is a variant of Depr\'es algorithm, a domain decomposition strategy for the solution of harmonic wave propagation problems. It imposes transmission conditions through interfaces by means of a non-local exchange operator. Conducting our analysis at the continuous level, in an infinite dimensional setting, we propose a coarse space construction for the preconditioning of the GOSM formulation, and provide estimates for the convergence of GMRes applied to the preconditioned equation.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 33 canonical work pages

  1. [1]

    M. S. Birman and M. Z. Solomyak.Spectral theory of self-adjoint operators in Hilbert space, volume 5 ofMath. Appl., Sov. Ser.Dordrecht etc.: Kluwer Academic Publishers, 1987

    work page 1987

  2. [2]

    Spectral coarse spaces based on indefinite operators: the $H_k$-GenEO method

    T. Chaumont-Frelet, V. Dolean, M. Fry, I.G. Graham, and M. Langer. Spectral coarse spaces based on indefinite operators: theHk-GenEO method. Preprint, arXiv:2605.31552 [math.NA] (2026), 2026

    work page Pith review arXiv 2026

  3. [3]

    X. Claeys. Non-local variant of the optimised Schwarz method for arbitrary non- overlapping subdomain partitions.ESAIM, Math. Model. Numer. Anal., 55(2):429–448, 2021

    work page 2021

  4. [4]

    X. Claeys. Nonlocal optimized Schwarz method for the Helmholtz equation with physical boundaries.SIAM J. Math. Anal., 55(6):7490–7512, 2023

    work page 2023

  5. [5]

    X. Claeys. Nonselfadjoint impedance in generalized optimized Schwarz methods.IMA J. Numer. Anal., 43(5):3026–3054, 2023

    work page 2023

  6. [6]

    Robusttreatmentofcross-pointsinoptimizedSchwarzmethods

    X.ClaeysandE.Parolin. Robusttreatmentofcross-pointsinoptimizedSchwarzmethods. Numer. Math., 151(2):405–442, 2022. 15

    work page 2022

  7. [7]

    B. Després. Domain decomposition method and the Helmholtz problem.Mathematical and Numerical aspects of wave propagation phenomena, pages 44–52, 1991

    work page 1991

  8. [8]

    Dolean, P

    V. Dolean, P. Jolivet, and F. Nataf.An introduction to domain decomposition methods. Algorithms, theory, and parallel implementation, volume 144 ofOther Titles Appl. Math. Philadelphia, PA: Society for Industrial and Applied Mathematics (SIAM), 2015

    work page 2015

Show all 36 references
  1. [9]

    Eiermann and O.G

    M. Eiermann and O.G. Ernst. Geometric aspects of the theory of Krylov subspace meth- ods.Acta Numerica, 10:251–312, 2001

    2001

  2. [10]

    S. C. Eisenstat, H. C. Elman, and M. H. Schultz. Variational iterative methods for nonsymmetric systems of linear equations.SIAM J. Numer. Anal., 20:345–357, 1983

    1983

  3. [11]

    H. C. Elman.Iterative Methods for Sparse Nonsymmetric Systems of Linear Equations. PhD thesis, Yale University, Department of Computer Science, New Haven, CT, 1982

    1982

  4. [12]

    Farhat, P

    C. Farhat, P. Avery, R. Tezaur, and J. Li. FETI-DPH: a dual-primal domain decompo- sition method for acoustic scattering.J. Comput. Acoust., 13(3):499–524, 2005

    2005

  5. [13]

    Farhat, A

    C. Farhat, A. Macedo, and M. Lesoinne. A two-level domain decomposition method for the iterative solution of high frequency exterior Helmholtz problems.Numer. Math., 85(2):283–308, 2000

    2000

  6. [14]

    Farhat, A

    C. Farhat, A. Macedo, M. Lesoinne, F.-X. Roux, F. Magoulès, and A. de la Bourdonnaye. Two-level domain decomposition methods with Lagrange multipliers for the fast iterative solution of acoustic scattering problems.Comput. Methods Appl. Mech. Eng., 184(2- 4):213–239, 2000

    2000

  7. [15]

    Golub and C.F

    G.H. Golub and C.F. Van Loan.Matrix computations.Baltimore, MD: The Johns Hopkins University Press, 4th ed. edition, 2013

    2013

  8. [16]

    W.W. Hager. Updating the inverse of a matrix.SIAM Rev., 31(2):221–239, 1989

    1989

  9. [17]

    Liesen and Z

    J. Liesen and Z. Strakoš.Krylov subspace methods. Principles and analysis. Numer. Math.Sci.Comput.Oxford: OxfordUniversityPress, reprintofthe2013hardbackedition edition, 2015

    2015

  10. [18]

    C. Ma. A unified framework for multiscale spectral generalized FEMs and low-rank approximations to multiscale PDEs.Foundations of Computational Mathematics, Apr 2025

    2025

  11. [19]

    C. Ma, C. Alber, R. Scheichl, and Y. Zhang. Two-level Restricted Additive Schwarz pre- conditioner based on Multiscale Spectral Generalized FEM for Heterogeneous Helmholtz Problems. Preprint, arXiv:2409.06533 (2024), 2024

    2024 arXiv

  12. [20]

    Ma and Y

    C. Ma and Y. Zhang. Multiscale model reduction and two-level Schwarz preconditioner for H(curl) elliptic problems, 2025

    2025

  13. [21]

    I. Moret. A note on the superlinear convergence of GMRES.SIAM J. Numer. Anal., 34(2):513–516, 1997. 16

    1997

  14. [22]

    Nataf and E

    F. Nataf and E. Parolin. Coarse spaces for non-symmetric two-level preconditioners based on local extended generalized eigenproblems. Preprint, arXiv:2404.02758 [math.NA] (2025), 2025

    2025 arXiv

  15. [23]

    Netrusov and Y

    Y. Netrusov and Y. Safarov. Weyl asymptotic formula for the Laplacian on domains with rough boundaries.Commun. Math. Phys., 253(2):481–509, 2005

    2005

  16. [24]

    Nevanlinna.Convergence of iterations for linear equations

    O. Nevanlinna.Convergence of iterations for linear equations. Basel: Birkhäuser, 1993

    1993

  17. [25]

    Pechstein.Finite and boundary element tearing and interconnecting solvers for mul- tiscale problems, volume 90 ofLecture Notes in Computational Science and Engineering

    C. Pechstein.Finite and boundary element tearing and interconnecting solvers for mul- tiscale problems, volume 90 ofLecture Notes in Computational Science and Engineering. Springer, Heidelberg, 2013

    2013

  18. [26]

    Compactnon-self-adjointoperators

    J.R.Ringrose. Compactnon-self-adjointoperators. VanNostrandReinholdMathematical Studies. 35. London etc.: Van Nostrand Reinhold Company. VI, 238 p, 1971

    1971

  19. [27]

    Rozenblum

    G. Rozenblum. Weyl asymptotics for Poincaré-Steklov eigenvalues in a domain with Lipschitz boundary.J. Spectr. Theory, 13(3):755–803, 2023

    2023

  20. [28]

    Rudin.Real and complex analysis.New York, NY: McGraw-Hill, 3rd ed

    W. Rudin.Real and complex analysis.New York, NY: McGraw-Hill, 3rd ed. edition, 1987

    1987

  21. [29]

    Rudin.Functional analysis.New York, NY: McGraw-Hill, 2nd ed

    W. Rudin.Functional analysis.New York, NY: McGraw-Hill, 2nd ed. edition, 1991

    1991

  22. [30]

    Saad.Iterative Methods for Sparse Linear Systems

    Y. Saad.Iterative Methods for Sparse Linear Systems. Society for Industrial and Applied Mathematics, jan 2003

    2003

  23. [31]

    Saad and M.H

    Y. Saad and M.H. Schultz. GMRES: A generalized minimal residual algorithm for solving nonsymmetric linear systems.SIAM J. Sci. Stat. Comput., 7:856–869, 1986

    1986

  24. [32]

    Spillane, V

    N. Spillane, V. Dolean, P. Hauret, F. Nataf, C. Pechstein, and R. Scheichl. Abstract robust coarse spaces for systems of PDEs via generalized eigenproblems in the overlaps. Numer. Math., 126(4):741–770, 2014

    2014

  25. [33]

    Fast-convergenttwo-levelrestrictedadditiveSchwarz methods based on optimal local approximation spaces.SIAM J

    A.Strehlow, C.Ma, andR.Scheichl. Fast-convergenttwo-levelrestrictedadditiveSchwarz methods based on optimal local approximation spaces.SIAM J. Sci. Comput., 48(1):31– 58, 2026

    2026

  26. [34]

    Toselli and O

    A. Toselli and O. Widlund.Domain decomposition methods—algorithms and theory, volume 34 ofSpringer Series in Computational Mathematics. Springer-Verlag, Berlin, 2005

    2005

  27. [35]

    R. Winther. Some superlinear convergent results for the conjugate gradient method. SIAM J. Numer. Anal., 17:14–17, 1980

    1980

  28. [36]

    Woodbury.Inverting modified matrices

    M.A. Woodbury.Inverting modified matrices. Princeton University, Princeton, NJ, 1950. Statistical Research Group, Memo. Rep. no. 42,. 17

    1950

Pith tools

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