Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Optimal Spectral Approximation in the Overlaps for Generalized Finite Element Methods

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read This paper proves that solving local spectral problems only on rings around subdomain boundaries keeps the nearly exponential error decay of MS-GFEM while cutting the cost of the eigencomputations.

desk verdict Genuinely useful ring-based MS-GFEM with credible but incomplete analysis: the headline near-exponential decay is proved in infinite dimensions, while the implemented discrete method rests on an unproved assertion. read the letter →

arxiv 2507.12226 v1 pith:KWPZ2EKU submitted 2025-07-16 math.NA cs.NA

classification math.NAcs.NA MSC 65F1065N1565N3065N55
keywords generalizedfiniteelementmethodmultiscalespectralKolmogorovn-widthring-localizedeigenproblemsapriorierrorestimaterestrictedadditiveSchwarzpreconditionerhigh-contrastcoefficientsoperator-harmonicextension
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 proves that the Multiscale Spectral Generalized Finite Element Method does not need to solve its local eigenvalue problems on entire subdomains: solving them on thin rings around the subdomain boundaries is enough to retain nearly exponential accuracy. Specifically, with $n$ local basis functions per subdomain it proves the a priori error bound $\|u-u_G\|_a \le C_d \exp(c_d) \sqrt{\kappa\kappa^*} \exp(-c_d n^{1/d})$, with constants depending only on the coefficient contrast and the geometry of the subdomains and rings. This matters because ring-localized eigenproblems are substantially smaller and have sparsity patterns closer to a problem of one dimension lower, so the spectral computations run much faster. The same analysis transfers to a two-level restricted additive Schwarz preconditioner, giving a convergence rate equal to the multiscale approximation error. Numerical experiments confirm the near-exponential decay and show that roughly twice as many modes are needed on rings as on whole subdomains, while the runtime for the eigenproblems drops by large factors.

What carries the argument

The central object is the ring-localized restriction operator $P_i^R : H_{a,0}(R_i^*) \to H^1_0(R_i)$ with $v \mapsto \chi_i^R v$, where $R_i$ is a ring around the boundary of the subdomain $\omega_i$, $R_i^*$ is an oversampling ring, and $\chi_i^R$ is a cut-off function that equals the partition-of-unity function on the ring and vanishes toward the interior. Its Kolmogorov $n$-width $d_n(R_i,R_i^*)$ is characterized by the eigenproblem on the ring, and the eigenfunctions are operator-harmonically extended into the subdomain interior to build the local basis. The near-exponential decay of the $n$-width is proved through a multiplicative decomposition $P^R = X_aQ_2Q_1$ over nested concentric rings, using a Caccioppoli inequality and a weak approximation property. The key identity in the local error bound is energy minimality: on the interior region the operator-harmonic extension has no larger energy than any function with the same trace, which is what lets an estimate on the ring control the error on the whole subdomain.

What would settle it

Run the fully discrete version of the ring-based method on a fixed high-contrast coefficient with rings of fixed physical width, refine the fine mesh, and measure the relative energy error as a function of the number of local eigenfunctions $n$. If the error at fixed $n$ grows as the mesh is refined, or if the fitted exponent $c$ in $\exp(-c n^{1/d})$ degrades with mesh refinement, then the asserted discrete analogue of the a priori bound does not hold for the implemented method.

Watch

Extended reading notes

Core claim

In the paper's own terms, the discovery is that the overlaps, the regions where neighboring subdomains meet, carry the spectral information needed for optimal local approximation. Replacing the whole-subdomain eigenproblem by the ring eigenproblem on $R_i^*$, followed by an operator-harmonic extension from the inner boundary to the full oversampling domain, preserves the nearly exponential decay of the relevant Kolmogorov $n$-width: $d_n(R_i,R_i^*) \le C_{d,i} \exp(-c_{d,i} n^{1/d})$. Combined with the general GFEM error estimate, this yields an a priori bound of the form $\|u-u_G\|_a \le C_d \exp(c_d) \sqrt{\kappa\kappa^*} \exp(-c_d n^{1/d})$ for the global approximation. The paper also proves that the resulting method works as a two-level restricted additive Schwarz preconditioner whose contraction factor is exactly the approximation error constant. The main price is that more local eigenfunctions are needed per subdomain than in the whole-subdomain MS-GFEM, because the ring does not see the coefficient inside the subdomain interior.

Load-bearing premise

The a priori analysis is carried out in infinite dimensions, and the paper asserts, without proof, that the discrete analogue holds with the same ring geometry and the same constants; if the discrete versions of energy minimality, Caccioppoli inequalities, and the weak approximation property do not hold as claimed, the practical error bound for the implemented method fails.

Editorial extensions

If this is right

  • If the a priori bound is correct, local eigenvalue problems in MS-GFEM can be restricted to rings around subdomain boundaries without sacrificing the near-exponential accuracy guarantee, at the price of roughly twice as many modes per subdomain.
  • The same approximation error controls the contraction factor of the preconditioned Richardson iteration and the GMRES residual bound, so the ring method gives a coefficient-robust two-level preconditioner.
  • The error constant scales with the square root of the local contrast and with the ratio of oversampling diameter to ring width, but not with the mesh size, so the method is robust under minimal structural assumptions on the coefficient.
  • The number of local eigenfunctions $n$ can be chosen according to a target accuracy, and the computational savings from smaller, sparser eigenproblems grow with the spatial dimension of the problem.

Reading between the lines

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

  • The paper does not state this, but the same ring-localization mechanism should transfer to other elliptic problems, such as Stokes or Maxwell systems, wherever energy minimality, a Caccioppoli inequality, and a weak approximation property hold, giving the same cost reduction for spectral coarse spaces.
  • The analysis suggests a quantitative trade-off: making rings thinner improves sparsity and factorization cost but shrinks the ring width that enters the constants, so an optimal ring width could be chosen by balancing fill-in against the error constant.
  • A testable consequence implied by the proof structure but not tested in the paper is that, for a fixed number of modes and a mesh resolving the coefficient, the energy error should remain roughly independent of mesh refinement, while preconditioned GMRES iteration counts should stay bounded as the mesh is refined.
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

3 major / 5 minor

Summary. The paper proposes a variant of the multiscale spectral generalized finite element method (MS-GFEM) in which the local spectral approximation spaces are computed on rings around subdomain boundaries and then operator-harmonically extended to the whole subdomain. The main theoretical results are a local approximation bound (Theorem 5.1) expressed in terms of a ring-based Kolmogorov n-width, a nearly exponential decay estimate for that n-width (Theorem 5.3), and a global a priori error estimate (Theorem 5.4, Corollary 5.5). The method is also proposed as a two-level restricted additive Schwarz preconditioner, with convergence stated via the same error bound. Numerical experiments on high-contrast coefficients compare the method with the standard MS-GFEM in terms of accuracy, iteration counts, and computational cost.

Significance. If the main claims hold, localizing the spectral computations to rings reduces the size and bandwidth of the local eigenproblems while retaining the nearly exponential convergence of MS-GFEM. The paper's local error estimate, Theorem 5.1, is clean and self-contained, and the numerical study is thorough and reproducible, with code provided. The a priori bound is parameter-free in the sense that no constants are fitted to data, and the claimed decay in n is a falsifiable prediction. However, the central decay theorem is only sketched, and the fully discrete version of the analysis is asserted rather than proved, so the paper's strongest claim is currently established only in the infinite-dimensional setting.

major comments (3)
  1. [Section 5.2, after Corollary 5.5] The transition from the infinite-dimensional theory to the implemented discrete method is asserted rather than proved. The abstract and Corollary 5.5 claim the a priori error bound for the proposed method, but Theorem 5.3 and Corollary 5.5 concern the continuous problems (4.1), (4.13), and (4.17), while the experiments and preconditioner use Q1 finite elements on a fine mesh. The statement that discrete analogues of energy-minimality, the Caccioppoli inequality, and the weak approximation property 'can be adapted easily' is insufficient: the discrete harmonic extension in (4.12)-(4.13) must satisfy an energy-minimality comparison with constants independent of h, and the discrete Caccioppoli and weak approximation inequalities must be uniform in h while retaining the H*/delta factors. In the implementation of Section 7, eta_i transitions over one fine-element layer, so the gradient bound (4.10) involves h^{-1} unless an explicit argument is supplied. Please provide a complete proof of the fully discrete analogue or, alternatively, state precisely which mesh and quadrature hypotheses are needed and revise the abstract accordingly.
  2. [Theorem 5.3] The proof of the main decay estimate is only a sketch. It states that the result 'can be proved by slightly modifying' [Ma24, Thm. 3.8 (i)] and then lists ingredients, but the ring geometry introduces substantive changes: an intermediate subdomain R_{N+1}, the multiplicative n-width inequality (5.7), a modified constant Theta, and the treatment of boundary subdomains. Because this theorem is the sole source of the exp(-c n^{1/d}) factor in Corollary 5.5, the paper should provide a complete proof or a detailed appendix. This is especially important because [Ma24] is an unpublished preprint and the constants C_{d,i} and c_{d,i} in (5.6) are not verified for the ring geometry in the present text. In addition, the multiplicative estimate (5.7) is asserted without proof and is not a standard immediate property of Kolmogorov n-widths; it needs a justification or a precise reference.
  3. [Section 5.2, Theorem 5.4] In the boundary-subdomain case of the proof of Theorem 5.4, the displayed estimate uses d_n(R,R*) but the accompanying sentence says 'using d_{n-1}(R,R*) instead of d_n(R,R*) in (5.12).' Since boundary subdomains have no constant component and use n harmonic extensions, the index should be n, not n-1. The contradiction between the display and the parenthetical should be corrected, and the indexing should be clarified for both interior and boundary subdomains.
minor comments (5)
  1. [Section 5.2, after Corollary 5.5] The phrase 'All of these properties have discrete analogues and hence the modification ... does not affect the analysis' is presented as a limitation statement but is not supported by any details. At minimum, the paper should cite a reference where the discrete analogues of energy-minimality, Caccioppoli, and the weak approximation property are proved in the present ring geometry, or state that this is part of future work.
  2. [Section 4.2 and Theorem 5.3] The nesting of rings in the proof of Theorem 5.3 is typeset unclearly: the sequence 'R* = R_1 > ... > R_{N+1} = R_{N+1} > R' contains an apparent notational collision. Please rewrite with consistently indexed rings so that the intermediate subdomain is unambiguous.
  3. [Section 7] The numerical setup uses brick-like subdomains and rings defined by extending supp(chi_i) by a fixed number of fine-element layers, whereas Theorem 5.3 is stated for concentric cubic rings. The text says the generalization follows from [Ma24, Rem. 3.15], but it should explicitly state whether the experimental geometry satisfies the theorem's hypotheses, or how the experimental results should be interpreted if it does not.
  4. [Section 6.2] The matrix form of the preconditioner overloads the symbol chi_i: it is used both for the partition-of-unity function and for the matrix representation of the operator v^h -> I_h(chi_i v^h). Using a distinct symbol such as [chi_i] or C_i for the discrete operator would improve readability.
  5. [Abstract] The abstract states that the paper proves a nearly exponential a priori decay result for the proposed method. In light of the missing fully discrete analysis, this claim should be qualified to the infinite-dimensional problems, or the discrete proof should be completed so that the claim covers the implemented method.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the ring-based n-width estimate is an adaptation of external machinery, not an assumption of the target result, and no fitted parameter is relabeled as a prediction.

full rationale

The derivation chain runs from the local eigenproblem (4.17) to the n-width dn(R_i,R*_i) via the standard characterization, then to the local estimate (5.3) and global bound (5.10). Theorem 5.1 proves ||P u-v||_{a,omega} <= dn(R,R*)||u||_{a,R*} by an energy-minimality argument (Eqs. (5.4)-(5.5)) that transfers the comparison from the subdomain to the ring; it does not assume the desired approximation. Theorem 5.3 obtains the nearly exponential decay of the ring n-width by modifying the proof of [Ma24, Thm. 3.8(i)], whose stated assumptions are cubes rather than rings; the paper explicitly replaces nested cubes by nested rings and derives new factors (5.8)-(5.9). This is an adaptation, not an import of the target result. Corollary 5.5 is the composition of Theorem 5.4 with Theorem 5.3; the constants are explicit functions of contrast and geometry, and no parameter is fitted to error data. The preconditioner results (Prop. 6.2, Thm. 6.3) are quoted from [SMS24] with references to its proofs, which is independent support rather than a circular reduction. The paper does rely on co-authored citations ([MSD22] for the GFEM error estimate, [SMS24] for the preconditioner analysis), but these are parameter-free results whose assumptions do not include the ring method, so they do not constitute self-citation load-bearing circularity. The only notable weakness is the unproved assertion after Corollary 5.5 that the discrete analogue 'can be adapted easily' because energy minimality, Caccioppoli inequalities, and the weak approximation property have discrete analogues; that is a completeness gap in the manuscript, not a circularity, and therefore does not raise the circularity score.

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

No data fitting appears; the constants in the a priori bounds are existence constants depending on contrast and geometry. The main external burden is the n-width decay machinery of [Ma24], plus the unproved discrete adaptation. No new physical entities are introduced.

assumptions (5)
  • domain assumption The coefficient A is uniformly elliptic with 0 < alpha_min <= alpha_max < infinity (2.2).
    The whole analysis is for this model problem; extension to other coercive problems is claimed but not proved.
  • domain assumption Partition of unity {chi_i} satisfies (3.1) with |grad chi_i| <= C_chi / delta_i.
    This is the standard GFEM assumption used in Theorem 3.1 and throughout the local error analysis.
  • domain assumption For each subdomain there exist rings R_i, R_i^* and a cut-off eta_i with the properties in Section 4.2, including dist(diff tilde{omega}_i, diff R_i) ≳ delta_i.
    This geometric cut-off construction is the backbone of the ring method and is assumed to hold for the chosen domain decompositions.
  • domain assumption Theorem 5.3 assumes R_i and R_i^* are concentric (truncated) cubes with delta* = dist_inf(R, diff R* \ diff Omega).
    The near-exponential decay is proved only for this restricted geometry; the paper refers to [Ma24, Rem. 3.15] for generalization.
  • ad hoc to paper Discrete analogues of energy minimality, the Caccioppoli inequality, and the weak approximation property transfer to the fully discrete setting (Section 5.2).
    The paper asserts that these properties have discrete analogues but does not prove them for rings; the numerical method relies on this assertion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Optimal Spectral Approximation in the Overlaps for Generalized Finite Element Methods." pith.science (2026). https://pith.science/paper/KWPZ2EKU

@misc{pith2026250712226,
  author       = {Pith},
  title        = {Pith review of: Optimal Spectral Approximation in the Overlaps for Generalized Finite Element Methods},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KWPZ2EKU}},
  note         = {Machine review of arXiv:2507.12226}
}
read the original abstract

In this paper, we study a generalized finite element method for solving second-order elliptic partial differential equations with rough coefficients. The method uses local approximation spaces computed by solving eigenvalue problems on rings around the boundary of local subdomains. Compared to the corresponding method that solves eigenvalue problems on the whole subdomains, the problem size and the bandwidth of the resulting system matrices are substantially reduced, resulting in faster spectral computations. We prove a nearly exponential a priori decay result for the local approximation errors of the proposed method, which implies the nearly exponential decay of the overall approximation error of the method. The proposed method can also be used as a preconditioner, and only a slight adaptation of our theory is necessary to prove the optimal convergence of the preconditioned iteration. Numerical experiments are presented to support the effectiveness of the proposed method and to investigate its coefficient robustness.

Figures

Figures reproduced from arXiv: 2507.12226 by the authors.

Figure 4.1
Figure 4.1. Illustration of the subdomains ωi and ω ∗ i (left) and of the rings Ri and R∗ i (right) in the case of an interior subdomain. conditions are imposed. The global particular function is then given by u p := X M i=1 χiu p i , which satisfies u p ∈ H1 g (Ω) by the partition of unity property, cf. (3.1). The GFEM solution is then defined as the Galerkin approximation in the affine space u p + Sn(Ω), i.e., we seek u G = u… view at source ↗
Figure 4.2
Figure 4.2. One dimensional illustration of the cut-off func￾tion χ R i (left) and the corresponding ring Ri (right). approximation space we consider the first ni eigenpairs (uk, λk) ∈ Ha(R∗ i ) × R of (4.11) aR∗ i (uk, φ) = λk aRi (χ R i uk, χR i φ) ∀φ ∈ Ha(R ∗ i ). To obtain functions defined on the full oversampling domain ω ∗ i , we perform an operator-harmonic extension of the eigenfunctions. This extension is performed by… view at source ↗
Figure 7.1
Figure 7.1. Coefficients A used in the numerical experiments. discretization. We always choose the fine mesh size small enough to resolve the co￾efficient. The computational domain is divided into M non-overlapping brick-like subdomains. Each subdomain is extended by two layers of fine elements in each space direction to create the overlapping subdomains {ωi} m i=1 of Ω. Additionally, each subdomain ωi is extended by ℓ layers o… view at source ↗
Figures from the paper (4 more)
Figure 7
Figure 7. Figure 7: (left) illustrates the (nearly-)exponential convergence of MS-GFEM [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 7.2
Figure 7.2. Figure 7.2: Errors with MS-GFEM and MS-GFEM-R for the skyscraper coefficient in [PITH_FULL_IMAGE:figures/full_fig_p016_7_2.png]
Figure 7.3
Figure 7.3. Figure 7.3: Eigenvalues (top row) of MS-GFEM and MS-GFEM-R for the channelized coefficient in [PITH_FULL_IMAGE:figures/full_fig_p016_7_3.png]
Figure 7.4
Figure 7.4. Figure 7.4: Computational times for eigenproblems (left) and av￾erage number of nonzero entries of LU-factors per row (right) for the MS-GFEM and the MS-GFEM-R as a function of the num￾ber m of fine-scale elements of the subdomains in each space di￾rection. The curves of the MS-…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. An Order-One Lower Bound on the Error of Scalable Generalized Multiscale Finite Element Space Constructions

    math.NA 2026-07 accept novelty 8.0 of 10

    For every deterministic fixed-visibility multiscale finite-element construction with bounded support, bounded coefficient-information radius, and bounded local multiplicity, a smooth periodic coefficient and smooth ri...

  2. Robust spectral preconditioning for high-P\'{e}clet number convection-diffusion

    math.NA 2025-09 conditional novelty 6.0 of 10

    A spectral coarse-space preconditioner based on MS-GFEM makes GMRES robust to Péclet number and subdomain count for convection-diffusion, demonstrated in experiments up to 100,000 subdomains.

Reference graph

Works this paper leans on

43 extracted references · 41 canonical work pages · cited by 2 Pith papers

  1. [1]

    Al Daas, P

    H. Al Daas, P. Jolivet, and T. Rees. Efficient algebraic two-level S chwarz preconditioner for sparse matrices. SIAM J. Sci. Comput. , 45(3):A1199--A1213, 2023

  2. [2]

    Altmann, P

    R. Altmann, P. Henning, and D. Peterseim. Numerical homogenization beyond scale separation. Acta Numer. , 30:1--86, 2021

  3. [3]

    Babuska, X

    I. Babuska, X. Huang, and R. Lipton. Machine computation using the exponentially convergent multiscale spectral generalized finite element method. ESAIM: Math. Model. Numer. Anal. , 48(2):493--515, 2014

  4. [4]

    P. E. Bjørstad, J. Koster, and P. Krzyzanowski. Domain Decomposition Solvers for Large Scale Industrial Finite Element Problems , page 373–383. Springer Berlin Heidelberg, 2001

  5. [5]

    Babuska and R

    I. Babuska and R. Lipton. Optimal local approximation spaces for generalized finite element methods with application to multiscale problems. Multiscale Model. Simul. , 9(1):373--406, 2011

  6. [6]

    Babuska, R

    I. Babuska, R. Lipton, P. Sinz, and M. Stuebner. Multiscale-spectral GFEM and optimal oversampling. Comput. Methods Appl. Mech. Eng. , 364:112960, 2020

  7. [7]

    Babuska and J

    I. Babuska and J. M Melenk. The partition of unity method. Int. J. Numer. Methods Eng. , 40(4):727--758, 1997

  8. [8]

    Babuska and J

    I. Babuska and J. E. Osborn. Can a finite element method perform arbitrarily badly? Math. Comp. , 69(230):443–463, 1999

Show all 43 references
  1. [9]

    Buhr and K

    A. Buhr and K. Smetana. Randomized local model order reduction. SIAM J. Sci. Comput. , 40(4):A2120–A2151, 2018

  2. [10]

    Bastian, R

    P. Bastian, R. Scheichl, L. Seelinger, and A. Strehlow. Multilevel spectral domain decomposition. SIAM J. Sci. Comput. , 45(3):S1--S26, 2022

  3. [11]

    E. T. Chung, Y. Efendiev, and W. T. Leung. Constraint energy minimizing generalized multiscale finite element method. Comput. Methods Appl. Mech. Engrg. , 339:298--319, 2018

  4. [12]

    Cai and M

    X.-C. Cai and M. Sarkis. A restricted additive S chwarz preconditioner for general sparse linear systems. SIAM J. Sci. Comput. , 21(2):792–797, 1999

  5. [13]

    J. W. Demmel, J. R. Gilbert, and X. S. Li. Super LU users' guide. 1999

  6. [14]

    C. R. Dohrmann and O. B. Widlund. A BDDC algorithm with deluxe scaling for three‐dimensional H(curl) problems. Commun. Pure Appl. Math. , 69(4):745–770, 2015

  7. [15]

    W. E and B. Engquist. The heterogeneous multiscale methods. Commun. Math. Sci. , 1(1):87--132, 2003

  8. [16]

    Ern and J.-L

    A. Ern and J.-L. Guermond. Theory and Practice of Finite Elements , volume 159 of Applied Mathematical Sciences . Springer New York, 2004

  9. [17]

    Efendiev, J

    Y. Efendiev, J. Galvis, and T. Y. Hou. Generalized multiscale finite element methods ( GM s FEM ). J. Comput. Phys. , 251:116--135, 2013

  10. [18]

    Efendiev, J

    Y. Efendiev, J. Galvis, R. Lazarov, and J. Willems. Robust domain decomposition preconditioners for abstract symmetric positive definite bilinear forms. ESAIM: Math. Model. Numer. Anal. , 46(5):1175--1199, 2012

  11. [19]

    Freese, M

    P. Freese, M. Hauck, T. Keil, and D. Peterseim. A super-localized generalized finite element method. Numer.Math. , 156(1):205–235, 2023

  12. [20]

    Galvis and Y

    J. Galvis and Y. Efendiev. Domain decomposition preconditioners for multiscale flows in high contrast media: reduced dimension coarse spaces. Multiscale Model. Simul. , 8(5):1621--1644, 2010

  13. [21]

    Grasedyck, I

    L. Grasedyck, I. Greff, and S. Sauter. The AL basis for the solution of elliptic problems in heterogeneous media. Multiscale Model. Simul. , 10(1):245--258, 2012

  14. [22]

    M. J. Gander, A. Loneland, and T. Rahman. Analysis of a new harmonically enriched multiscale coarse space for domain decomposition methods. arXiv e-print 1512.05285 , 2015

  15. [23]

    Heinlein, A

    A. Heinlein, A. Klawonn, J. Knepper, and O. Rheinbach. Adaptive GDSW coarse spaces for overlapping S chwarz methods in three dimensions. SIAM J. Sci. Comput. , 41(5):A3045–A3072, 2019

  16. [24]

    Henning and D

    P. Henning and D. Peterseim. Oversampling for the multiscale finite element method. Multiscale Model. Simul. , 11(4):1149--1175, 2013

  17. [25]

    Hauck and D

    M. Hauck and D. Peterseim. Super-localization of elliptic multiscale problems. Math. Comp. , 92(341):981--1003, 2022

  18. [26]

    Heinlein and K

    A. Heinlein and K. Smetana. A fully algebraic and robust two-level schwarz method based on optimal local approximation spaces. arXiv e-print 2207.05559 , 2022

  19. [27]

    T. Y. Hou and X.-H. Wu. A multiscale finite element method for elliptic problems in composite materials and porous media. J. Comput. Phys. , 134(1):169--189, 1997

  20. [28]

    H. H. Kim, E. Chung, and J. Wang. BDDC and FETI-DP preconditioners with adaptive coarse spaces for three-dimensional elliptic problems with oscillatory and high contrast coefficients. J. Comput. Phys. , 349:191–214, 2017

  21. [29]

    Klawonn, M

    A. Klawonn, M. K\" u hn, and O. Rheinbach. Adaptive coarse spaces for FETI-DP in three dimensions. SIAM J. Sci. Comput. , 38(5):A2880–A2911, 2016

  22. [30]

    A unified framework for multiscale spectral generalized FEMs and low-rank approximations to multiscale PDEs

    C Ma. A unified framework for multiscale spectral generalized FEMs and low-rank approximations to multiscale PDEs . arXiv e-print 2311.08761 , 2024

  23. [31]

    J. M. Melenk. On generalized finite element methods. Ph.D. thesis, Department of Mathematics, University of Maryland, 1995

  24. [32]

    M lqvist and D

    A. M lqvist and D. Peterseim. Localization of elliptic multiscale problems. Math. Comp. , 83(290):2583--2603, 2014

  25. [33]

    Matache and C

    A.-M. Matache and C. Schwab. Two-scale FEM for homogenization problems. ESAIM: Math. Model. Numer. Anal. , 36(4):537--572, 2002

  26. [34]

    Mandel and B

    J. Mandel and B. Sousedik. Adaptive selection of face coarse degrees of freedom in the BDDC and the FETI-DP iterative substructuring methods. Comput. Methods Appl. Mech. Eng. , 196(8):1389–1399, 2007

  27. [35]

    Ma and R

    C. Ma and R. Scheichl. Error estimates for discrete generalized fems with locally optimal spectral approximations. Math. Comp. , 91(338):2539--2569, 2022

  28. [36]

    C. Ma, R. Scheichl, and T. Dodwell. Novel design and analysis of generalized finite element methods based on locally optimal spectral approximations. SIAM J. Numer. Anal. , 60(1):244--273, 2022

  29. [37]

    Mandel, B

    J. Mandel, B. Sousedik, and J. Sistek. Adaptive BDDC in three dimensions. Math. Comput. Simul. , 82(10):1812–1831, 2012

  30. [38]

    Nataf, H

    F. Nataf, H. Xiang, V. Dolean, and N. Spillane. A coarse space construction based on local dirichlet-to-neumann maps. SIAM J. Sci. Comput. , 33(4):1623–1642, 2011

  31. [39]

    H. Owhadi. Multigrid with rough coefficients and multiresolution operator decomposition from hierarchical information games. SIAM Rev. , 59(1):99--149, 2017

  32. [40]

    Pechstein and C

    C. Pechstein and C. Dohrmann. A unified framework for adaptive BDDC . Electron. Trans. Numer. Anal. , 46:337--358, 08 2017

  33. [41]

    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. , 126:741--770, 2014

  34. [42]

    Strehlow, C

    A. Strehlow, C. Ma, and R. Scheichl. Fast-convergent two-level restricted additive schwarz methods based on optimal local approximation spaces. arXiv e-print 2408.16282 , 2024

  35. [43]

    Spillane and D.J

    N. Spillane and D.J. Rixen. Automatic spectral coarse spaces for robust finite element tearing and interconnecting and balanced domain decomposition algorithms. Int. J. Numer. Methods Eng. , 95(11):953–990, 2013

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.