pith. sign in

arxiv: 2606.30495 · v1 · pith:LT7Q3J3Pnew · submitted 2026-06-29 · 🧮 math.NA · cs.AI· cs.NA· math-ph· math.MP

McMg: A Learned Phase-Space Multi-channel Multigrid Preconditioner for Helmholtz Equation

Jiwei Jia , Xinliang Liu , Juntao Wang , Jinchao Xu This is my paper

Pith reviewed 2026-06-30 04:39 UTC · model grok-4.3

classification 🧮 math.NA cs.AIcs.NAmath-phmath.MP
keywords Helmholtz equationmultigrid preconditionerlearned preconditionersphase-space methodshigh-frequency wavesnumerical PDEheterogeneous media
0
0 comments X

The pith

McMg retains unresolved wave phase and direction in extra channels during coarsening to precondition high-wavenumber Helmholtz equations with fewer iterations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes McMg, a learned multigrid preconditioner that maps residuals to corrections for heterogeneous Helmholtz equations at high wavenumbers. It coarsens physical space while storing local amplitude, phase, direction, and scattering information in additional channel dimensions at each coarse node. This design combines multi-channel transfer operators, adaptive stencils, and medium-dependent smoothers whose coefficients are generated from the wave speed. For a fixed medium the online V-cycle reduces to fixed-coefficient convolutions after a one-time setup. Models trained on small domains extend to larger scales and higher wavenumbers through Layer-by-Layer Progressive Finetuning, and experiments on three-dimensional high-frequency and high-contrast problems report fewer iterations and lower wall-clock time than classical and existing neural baselines.

Core claim

McMg coarsens physical space while retaining unresolved local wave information in the channel dimension so that each coarse node carries a learned packet of amplitude, phase, direction, and scattering coefficients rather than a single scalar. The architecture uses linear multi-channel transfer operators together with locally adaptive stencils and medium-dependent smoothers; nonlinear physical features are cached once, after which each iteration is linear in the residual. Layer-by-Layer Progressive Finetuning extends the learned operator to larger domains and higher effective wavenumbers by adding and tuning only new coarse levels.

What carries the argument

Multi-channel transfer operators that coarsen space while preserving phase-space wave packets in the channel dimension.

If this is right

  • McMg requires substantially fewer iterations than classical multigrid on high-frequency heterogeneous problems.
  • It uses less wall-clock time than existing neural preconditioners on large-scale three-dimensional cases.
  • For any fixed medium the V-cycle remains linear after a single setup phase that caches medium-dependent coefficients.
  • Models trained on small domains transfer directly or via LLPF to larger domains and higher effective wavenumbers.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same channel-based retention of local propagation information could be tested on time-harmonic Maxwell or acoustic wave equations.
  • Combining the learned coarse-grid packets with adaptive mesh refinement around high-contrast interfaces might further reduce iteration counts.
  • The linearity of the online phase suggests the method could serve as a drop-in component inside existing iterative frameworks without changing outer solvers.

Load-bearing premise

Learned multi-channel operators and smoothers trained on small domains can be extended to larger domains and higher wavenumbers by adding and finetuning only new coarse levels without retraining the full model.

What would settle it

On a three-dimensional high-contrast Helmholtz problem whose domain size and wavenumber exceed the training range by a factor of two, McMg would fail to show fewer iterations than a strong classical multigrid baseline.

Figures

Figures reproduced from arXiv: 2606.30495 by Jinchao Xu, Jiwei Jia, Juntao Wang, Xinliang Liu.

Figure 1
Figure 1. Figure 1: McMg architecture. The setup phase precomputes and caches per-level multi-channel physical features, while [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Visualization of the learned coarse-space basis responses. We inject a unit impulse at different coarse levels [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Diversity of the multi-channel coarse space. (a) Fine-grid responses obtained by activating channels [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Schematic of the Layer-by-Layer Progressive Finetuning (LLPF) strategy. The process begins with a base [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Representative velocity models for the two 2D benchmarks: (a) the high-frequency OpenBreastUS breast [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Inference performance of Born+McMg on the OpenBreastUS breast dataset, showing predicted wavefields [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Inference performance of Born+McMg on the OpenFWI CurveFault-B dataset, showing predicted wavefields [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Inference performance of Born+McMg on the Kimberlina 1.2 CCUS dataset, showing predicted wavefields [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Velocity models used for generalization analysis: (a) a representative [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Convergence performance of McMg on a randomly cropped Marmousi patch. [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Convergence performance of McMg on the full Marmousi model. [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Effective support of the learned Green’s function on a [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Spectral convergence analysis. (a) Sound speed map. (d) Wavefield real part. (b, e) Spectral error at [PITH_FULL_IMAGE:figures/full_fig_p017_13.png] view at source ↗
read the original abstract

Solving heterogeneous Helmholtz equations at high wavenumbers remains challenging because the discretized operator is indefinite, pollution degrades phase accuracy, and scalar coarse-grid correction can discard the local phase and propagation-direction information carried by oscillatory errors. We propose Multi-channel Multigrid (McMg), a learned phase-space multigrid preconditioner for heterogeneous Helmholtz equations. Rather than predicting the solution directly, McMg maps residuals to corrections within an iterative framework. Its central idea is to coarsen physical space while retaining unresolved local wave information in the channel dimension: each coarse node carries a learned packet of amplitude, phase, direction, and scattering coefficients rather than a single scalar unknown. The architecture combines linear multi-channel transfer operators with locally adaptive stencils, neural PDE operators, and medium-dependent smoothers whose coefficients are generated from the wave speed. For a fixed medium, the V-cycle is linear in the residual; nonlinear physical features are computed once in a setup phase and cached, so each online iteration reduces to convolutions with fixed coefficients. We further study generalization across scales. Models trained on small domains transfer directly to larger domains and higher effective wavenumbers, and a Layer-by-Layer Progressive Finetuning (LLPF) strategy extends the support of the learned Green's operator by adding and finetuning only new coarse levels. Numerical experiments on high-frequency, high-contrast, and large-scale three-dimensional problems demonstrate that McMg requires substantially fewer iterations and less wall-clock time than strong classical baselines, while consistently outperforming existing neural preconditioners.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes McMg, a learned phase-space multi-channel multigrid preconditioner for heterogeneous Helmholtz equations. It coarsens physical space while retaining local wave information (amplitude, phase, direction, scattering) in channel dimensions via learned multi-channel transfer operators, locally adaptive stencils, neural PDE operators, and medium-dependent smoothers (cached after setup). The V-cycle is linear for fixed media; generalization is achieved by direct transfer from small-domain training plus Layer-by-Layer Progressive Finetuning (LLPF) to add/finetune coarse levels for larger domains and higher wavenumbers. Numerical experiments claim substantially fewer iterations and lower wall-clock time than classical baselines and existing neural preconditioners on high-frequency, high-contrast, large-scale 3D problems.

Significance. If the empirical claims hold with reproducible setups, McMg would represent a meaningful advance in scalable preconditioning for indefinite high-wavenumber Helmholtz problems, by embedding phase-space information into multigrid and demonstrating practical generalization via LLPF without full retraining. This could impact applications requiring repeated solves at scale, provided the learned components transfer reliably.

major comments (2)
  1. [Numerical experiments / abstract] Numerical experiments (abstract and corresponding section): the central performance claim of substantially fewer iterations and less wall-clock time versus strong classical and neural baselines is asserted without concrete setup details (domain sizes, wavenumber ranges, contrast levels, grid resolutions), error metrics, baseline implementations, or statistical controls. This prevents verification of the headline advantage and makes the generalization claim via LLPF untestable from the reported data.
  2. [Generalization / LLPF description] Generalization via LLPF (abstract and method description): the claim that models trained on small domains transfer directly and that LLPF extends the learned Green's operator to larger domains/higher wavenumbers while preserving iteration counts relies on unquantified assertions. No data is supplied on channel-wise packet degradation under wavenumber increase, epochs needed versus full retraining, or effectiveness of cached medium-dependent smoothers when coarse-grid spacing changes; if this step fails, the reported 3D high-frequency gains do not follow.
minor comments (1)
  1. [Introduction / method overview] The abstract and method sections introduce several new entities (learned phase-space packets, multi-channel transfers) without a compact notation table or diagram clarifying the channel dimension versus physical coarsening.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on reproducibility and the LLPF generalization strategy. We address both major points below by agreeing to enhance the manuscript with additional concrete details and quantitative data.

read point-by-point responses

  1. Referee: [Numerical experiments / abstract] Numerical experiments (abstract and corresponding section): the central performance claim of substantially fewer iterations and less wall-clock time versus strong classical and neural baselines is asserted without concrete setup details (domain sizes, wavenumber ranges, contrast levels, grid resolutions), error metrics, baseline implementations, or statistical controls. This prevents verification of the headline advantage and makes the generalization claim via LLPF untestable from the reported data.

    Authors: We agree that the abstract is high-level and that Section 4 would benefit from a consolidated summary of parameters for easier verification. In the revision we will insert a table listing all domain sizes, wavenumber ranges, contrast levels, grid resolutions, error metrics (relative residual norms), baseline solver implementations with tolerances, and any multi-run statistics. This directly addresses the verifiability concern without altering the existing experimental results. revision: yes

  2. Referee: [Generalization / LLPF description] Generalization via LLPF (abstract and method description): the claim that models trained on small domains transfer directly and that LLPF extends the learned Green's operator to larger domains/higher wavenumbers while preserving iteration counts relies on unquantified assertions. No data is supplied on channel-wise packet degradation under wavenumber increase, epochs needed versus full retraining, or effectiveness of cached medium-dependent smoothers when coarse-grid spacing changes; if this step fails, the reported 3D high-frequency gains do not follow.

    Authors: The current manuscript demonstrates successful transfer and LLPF through iteration-count preservation on the reported 3D cases, but we acknowledge the absence of explicit quantification on packet degradation, finetuning epoch counts, and smoother cache behavior under grid changes. We will add a new subsection with supporting tables or plots quantifying these quantities (channel retention metrics, epoch comparisons, and smoother validation across coarse spacings) to make the generalization claims fully substantiated. revision: yes

Circularity Check

0 steps flagged

No significant circularity; performance claims rest on independent numerical experiments

full rationale

The paper defines McMg as a learned preconditioner whose multi-channel operators and medium-dependent smoothers are trained on small domains, then applies the resulting fixed V-cycle to larger problems. The headline claim of substantially fewer iterations and lower wall-clock time versus classical and neural baselines is presented as the outcome of explicit numerical experiments on high-frequency, high-contrast 3D instances; these results are not shown to reduce by construction to the training loss or to any self-citation. The LLPF generalization step is an empirical assertion supported by the same experiments rather than a definitional identity. No self-definitional, fitted-input-renamed-as-prediction, or load-bearing self-citation steps appear in the supplied text.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the existence of trainable multi-channel operators that preserve wave information and on the assumption that a one-time setup phase can cache all nonlinear medium features. No explicit free parameters or invented physical entities are named; the learned packets constitute a new representational entity whose independent evidence is the reported numerical improvement.

axioms (1)
  • domain assumption Standard assumptions of multigrid convergence theory for indefinite operators hold when the coarse-grid correction is replaced by a learned multi-channel operator.
    Invoked implicitly when claiming that the V-cycle remains an effective preconditioner.
invented entities (1)
  • Learned phase-space packet (amplitude, phase, direction, scattering coefficients per coarse node) no independent evidence
    purpose: To retain unresolved local wave information that scalar coarse-grid values discard.
    Central representational innovation described in the abstract; no external falsifiable prediction supplied.

pith-pipeline@v0.9.1-grok · 5821 in / 1396 out tokens · 49546 ms · 2026-06-30T04:39:30.938510+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

89 extracted references · 16 canonical work pages · 5 internal anchors

  1. [1]

    International conference on machine learning , pages=

    On enhancing expressive power via compositions of single fixed-size relu network , author=. International conference on machine learning , pages=. 2023 , organization=

    2023

  2. [2]

    International Conference on Learning Representations , volume=

    MgNO: Efficient parameterization of linear operators via multigrid , author=. International Conference on Learning Representations , volume=

  3. [3]

    Journal of Machine Learning Research , volume=

    Neural operator: Learning maps between function spaces with applications to pdes , author=. Journal of Machine Learning Research , volume=

  4. [4]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Learning across scales---multiscale methods for convolution neural networks , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  5. [5]

    International conference on machine learning , pages=

    Beyond finite layer neural networks: Bridging deep architectures and numerical differential equations , author=. International conference on machine learning , pages=. 2018 , organization=

    2018

  6. [6]

    Advances in neural information processing systems , volume=

    Neural ordinary differential equations , author=. Advances in neural information processing systems , volume=

  7. [7]

    Advances in neural information processing systems , volume=

    Deep equilibrium models , author=. Advances in neural information processing systems , volume=

  8. [8]

    Advances in neural information processing systems , volume=

    On training implicit models , author=. Advances in neural information processing systems , volume=

  9. [9]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Jfb: Jacobian-free backpropagation for implicit networks , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  10. [10]

    Advances in Neural Information Processing Systems , volume=

    Deep equilibrium based neural operators for steady-state pdes , author=. Advances in Neural Information Processing Systems , volume=

  11. [11]

    National Science Open , volume=

    Learning neural operators on riemannian manifolds , author=. National Science Open , volume=. 2024 , publisher=

    2024

  12. [12]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Film: Visual reasoning with a general conditioning layer , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  13. [13]

    Advances in neural information processing systems , volume=

    Dynamic filter networks , author=. Advances in neural information processing systems , volume=

  14. [14]

    Advances in neural information processing systems , volume=

    Condconv: Conditionally parameterized convolutions for efficient inference , author=. Advances in neural information processing systems , volume=

  15. [15]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Dynamic convolution: Attention over convolution kernels , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  16. [16]

    HyperNetworks

    Hypernetworks , author=. arXiv preprint arXiv:1609.09106 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  17. [17]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Pixel-adaptive convolutional neural networks , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  18. [18]

    Links , volume=

    A proposal on machine learning via dynamical systems , author=. Links , volume=

  19. [19]

    Proceedings of the 27th international conference on international conference on machine learning , pages=

    Learning fast approximations of sparse coding , author=. Proceedings of the 27th international conference on international conference on machine learning , pages=

  20. [20]

    Journal of computational physics , volume=

    A convergent Born series for solving the inhomogeneous Helmholtz equation in arbitrarily large media , author=. Journal of computational physics , volume=. 2016 , publisher=

    2016

  21. [21]

    Applied Numerical Mathematics , volume=

    On a class of preconditioners for solving the Helmholtz equation , author=. Applied Numerical Mathematics , volume=. 2004 , publisher=

    2004

  22. [22]

    SIAM Journal on Numerical Analysis , volume=

    A source transfer domain decomposition method for Helmholtz equations in unbounded domain , author=. SIAM Journal on Numerical Analysis , volume=. 2013 , publisher=

    2013

  23. [23]

    Journal of Computational Physics , volume=

    Trace transfer-based diagonal sweeping domain decomposition method for the Helmholtz equation: Algorithms and convergence analysis , author=. Journal of Computational Physics , volume=. 2022 , publisher=

    2022

  24. [24]

    Numerical analysis of multiscale problems , pages=

    Why it is difficult to solve Helmholtz problems with classical iterative methods , author=. Numerical analysis of multiscale problems , pages=. 2011 , publisher=

    2011

  25. [25]

    Nature Reviews Physics , volume=

    Neural operators for accelerating scientific simulations and design , author=. Nature Reviews Physics , volume=. 2024 , publisher=

    2024

  26. [26]

    Journal of Computational physics , volume=

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations , author=. Journal of Computational physics , volume=. 2019 , publisher=

    2019

  27. [27]

    Nature machine intelligence , volume=

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators , author=. Nature machine intelligence , volume=. 2021 , publisher=

    2021

  28. [28]

    Fourier Neural Operator for Parametric Partial Differential Equations

    Fourier neural operator for parametric partial differential equations , author=. arXiv preprint arXiv:2010.08895 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  29. [29]

    Journal of Machine Learning Research , volume=

    Fourier neural operator with learned deformations for pdes on general geometries , author=. Journal of Machine Learning Research , volume=

  30. [30]

    arXiv preprint arXiv:2204.11127 (2022)

    U-no: U-shaped neural operators , author=. arXiv preprint arXiv:2204.11127 , year=

    work page arXiv

  31. [31]

    Advances in neural information processing systems , volume=

    Choose a transformer: Fourier or galerkin , author=. Advances in neural information processing systems , volume=

  32. [32]

    Journal of Computational Physics , volume=

    Mitigating spectral bias for the multiscale operator learning , author=. Journal of Computational Physics , volume=. 2024 , publisher=

    2024

  33. [33]

    International conference on machine learning , pages=

    On the spectral bias of neural networks , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  34. [34]

    The Thirteenth International Conference on Learning Representations , year=

    MGCFNN: A Neural MultiGrid Solver with Novel Fourier Neural Network for High Wave Number Helmholtz Equations , author=. The Thirteenth International Conference on Learning Representations , year=

  35. [35]

    SIAM Journal on Scientific Computing , volume=

    A neural multigrid solver for helmholtz equations with high wavenumber and heterogeneous media , author=. SIAM Journal on Scientific Computing , volume=. 2025 , publisher=

    2025

  36. [36]

    Nature Machine Intelligence , volume=

    Blending neural operators and relaxation methods in PDE numerical solvers , author=. Nature Machine Intelligence , volume=. 2024 , publisher=

    2024

  37. [37]

    arXiv preprint arXiv:2402.05598 , year=

    Neural operators meet conjugate gradients: The FCG-NO method for efficient PDE solving , author=. arXiv preprint arXiv:2402.05598 , year=

    work page arXiv

  38. [38]

    arXiv preprint arXiv:2312.11093 , year=

    MGCNN: a learnable multigrid solver for sparse linear systems from PDEs on structured grids , author=. arXiv preprint arXiv:2312.11093 , year=

    work page arXiv

  39. [39]

    SIAM Journal on Scientific Computing , volume=

    Multigrid-augmented deep learning preconditioners for the Helmholtz equation using compact implicit layers , author=. SIAM Journal on Scientific Computing , volume=. 2024 , publisher=

    2024

  40. [40]

    Communications on pure and applied mathematics , volume=

    Radiation boundary conditions for acoustic and elastic wave calculations , author=. Communications on pure and applied mathematics , volume=

  41. [41]

    Journal of computational physics , volume=

    A perfectly matched layer for the absorption of electromagnetic waves , author=. Journal of computational physics , volume=. 1994 , publisher=

    1994

  42. [42]

    2001 , publisher=

    Chebyshev and Fourier spectral methods , author=. 2001 , publisher=

    2001

  43. [43]

    Journal of computational physics , volume=

    Approximation of radiation boundary conditions , author=. Journal of computational physics , volume=. 1981 , publisher=

    1981

  44. [44]

    Wavelets, multilevel methods and elliptic PDEs , pages=

    An introduction to multilevel methods , author=. Wavelets, multilevel methods and elliptic PDEs , pages=. 1997 , publisher=

    1997

  45. [45]

    SIAM Journal on Scientific Computing , volume=

    Accuracy properties of the wave-ray multigrid algorithm for Helmholtz equations , author=. SIAM Journal on Scientific Computing , volume=. 2006 , publisher=

    2006

  46. [46]

    SIAM Journal on numerical analysis , volume=

    Is the pollution effect of the FEM avoidable for the Helmholtz equation considering high wave numbers? , author=. SIAM Journal on numerical analysis , volume=. 1997 , publisher=

    1997

  47. [47]

    Journal of computational Physics , volume=

    A dispersion minimizing scheme for the 3-D Helmholtz equation based on ray theory , author=. Journal of computational Physics , volume=. 2016 , publisher=

    2016

  48. [48]

    Advances in Applied Mathematics and Mechanics , volume=

    Local Interaction Simulation Approach for the Acoustic Wave Equation with Perfectly Matched Layer , author=. Advances in Applied Mathematics and Mechanics , volume=

  49. [49]

    Bulletin of the Seismological Society of America , volume=

    3D heterogeneous staggered-grid finite-difference modeling of seismic motion with volume harmonic and arithmetic averaging of elastic moduli and densities , author=. Bulletin of the Seismological Society of America , volume=. 2002 , publisher=

    2002

  50. [50]

    Journal of computational physics , volume=

    A Helmholtz equation solver using unsupervised learning: Application to transcranial ultrasound , author=. Journal of computational physics , volume=. 2021 , publisher=

    2021

  51. [51]

    Gaussian Error Linear Units (GELUs)

    Gaussian error linear units (gelus) , author=. arXiv preprint arXiv:1606.08415 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  52. [52]

    Advances in Neural Information Processing Systems , volume=

    OpenFWI: Large-scale multi-structural benchmark datasets for full waveform inversion , author=. Advances in Neural Information Processing Systems , volume=

  53. [53]

    Communications on Pure and Applied Mathematics , volume=

    Approximate separability of the Green's function of the Helmholtz equation in the high frequency limit , author=. Communications on Pure and Applied Mathematics , volume=. 2018 , publisher=

    2018

  54. [54]

    Frontiers in Physics , volume=

    The Green-function transform and wave propagation , author=. Frontiers in Physics , volume=. 2014 , publisher=

    2014

  55. [55]

    Journal of Computational and Applied Mathematics , volume=

    Pollution and accuracy of solutions of the Helmholtz equation: A novel perspective from the eigenvalues , author=. Journal of Computational and Applied Mathematics , volume=. 2021 , publisher=

    2021

  56. [56]

    1999 , publisher=

    Waves and fields in inhomogenous media , author=. 1999 , publisher=

    1999

  57. [57]

    SIAM Journal on Numerical Analysis , volume=

    Wavenumber explicit convergence of a multiscale generalized finite element method for heterogeneous Helmholtz problems , author=. SIAM Journal on Numerical Analysis , volume=. 2023 , publisher=

    2023

  58. [58]

    Multiscale Modeling & Simulation , volume=

    Multi-resolution localized orthogonal decomposition for Helmholtz problems , author=. Multiscale Modeling & Simulation , volume=. 2022 , publisher=

    2022

  59. [59]

    SIAM Journal on Scientific Computing , volume=

    Super-localized orthogonal decomposition for high-frequency Helmholtz problems , author=. SIAM Journal on Scientific Computing , volume=. 2024 , publisher=

    2024

  60. [60]

    arXiv preprint arXiv:2407.04364 , year=

    Robust Multiscale Methods for Helmholtz equations in high contrast heterogeneous media , author=. arXiv preprint arXiv:2407.04364 , year=

    work page arXiv

  61. [61]

    I , author=

    Variational principles for scattering processes. I , author=. Physical Review , volume=. 1950 , publisher=

    1950

  62. [62]

    SIAM Journal on Numerical Analysis , volume=

    Two-level hybrid Schwarz preconditioners for the Helmholtz equation with high wave number , author=. SIAM Journal on Numerical Analysis , volume=. 2025 , publisher=

    2025

  63. [63]

    Computer Methods in Applied Mechanics and Engineering , volume=

    Stable multiscale Petrov--Galerkin finite element method for high frequency acoustic scattering , author=. Computer Methods in Applied Mechanics and Engineering , volume=. 2015 , publisher=

    2015

  64. [64]

    Mathematics of Computation , volume=

    Eliminating the pollution effect in Helmholtz problems by local subscale correction , author=. Mathematics of Computation , volume=

  65. [65]

    arXiv preprint arXiv:2511.16808 , year=

    Vanka-smoothed shifted Laplacian multigrid preconditioners for the Helmholtz equations , author=. arXiv preprint arXiv:2511.16808 , year=

    work page arXiv

  66. [66]

    Journal of Computational Physics , volume=

    Block-implicit multigrid solution of Navier-Stokes equations in primitive variables , author=. Journal of Computational Physics , volume=. 1986 , publisher=

    1986

  67. [67]

    Electron

    Wave-ray multigrid method for standing wave equations , author=. Electron. Trans. Numer. Anal , volume=

  68. [68]

    arXiv preprint arXiv:2507.15035 , year=

    Openbreastus: Benchmarking neural operators for wave imaging using breast ultrasound computed tomography , author=. arXiv preprint arXiv:2507.15035 , year=

    work page arXiv

  69. [69]

    Communications on pure and applied mathematics , volume=

    Sweeping preconditioner for the Helmholtz equation: hierarchical matrix representation , author=. Communications on pure and applied mathematics , volume=. 2011 , publisher=

    2011

  70. [70]

    Journal of Computational Physics , volume=

    The method of polarized traces for the 2D Helmholtz equation , author=. Journal of Computational Physics , volume=. 2016 , publisher=

    2016

  71. [71]

    SIAM Journal on Scientific Computing , volume=

    Optimized Schwarz methods without overlap for the Helmholtz equation , author=. SIAM Journal on Scientific Computing , volume=. 2002 , publisher=

    2002

  72. [72]

    arXiv preprint arXiv:2508.20650 , year=

    Self-Composing Neural Operators with Depth and Accuracy Scaling via Adaptive Train-and-Unroll Approach , author=. arXiv preprint arXiv:2508.20650 , year=

    work page arXiv

  73. [73]

    Diff-ANO: Towards Fast High-Resolution Ultrasound Computed Tomography via Conditional Consistency Models and Adjoint Neural Operators

    Diff-ano: Towards fast high-resolution ultrasound computed tomography via conditional consistency models and adjoint neural operators , author=. arXiv preprint arXiv:2507.16344 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  74. [74]

    arXiv preprint arXiv:2404.02156 , year=

    Convergence of overlapping domain decomposition methods with PML transmission conditions applied to nontrapping Helmholtz problems , author=. arXiv preprint arXiv:2404.02156 , year=

    work page arXiv

  75. [75]

    2000 , publisher=

    Spectral methods in MATLAB , author=. 2000 , publisher=

    2000

  76. [76]

    SIAM Journal on Scientific Computing , volume=

    Generalized hybrid iterative methods for large-scale Bayesian inverse problems , author=. SIAM Journal on Scientific Computing , volume=. 2017 , publisher=

    2017

  77. [77]

    Multiscale Modeling & Simulation , volume=

    Generalized multiscale finite element methods for wave propagation in heterogeneous media , author=. Multiscale Modeling & Simulation , volume=. 2014 , publisher=

    2014

  78. [78]

    Neural Preconditioned Born Series: A Metric-Matched Framework for Learning-based Preconditioners

    Neural Preconditioned Born Series: A Metric-Matched Framework for Learning-based Preconditioners , author=. arXiv preprint arXiv:2603.18527 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  79. [79]

    2022 , month =

    Gasperikova, Erika and Alumbaugh, David and Crandall, Dustin and Commer, Michael and Feng, Shihang and Harbert, William and Li, Yaoguo and Lin, Youzuo and Samarasinghe, Savini Manthila , title =. 2022 , month =. doi:10.18141/1887287 , url =

    work page doi:10.18141/1887287 2022

  80. [80]

    Machine Learning for Computational Science and Engineering , volume=

    Leveraging operator learning to accelerate convergence of the preconditioned conjugate gradient method , author=. Machine Learning for Computational Science and Engineering , volume=. 2025 , publisher=

    2025

Showing first 80 references.