pith. sign in

arxiv: 2605.21830 · v1 · pith:Q4A4BJUXnew · submitted 2026-05-20 · ⚛️ physics.comp-ph

Solving forward and inverse wave scattering via boundary integral equations and deep learning. Applications to cloaking design

Camille Carvalho , Elsie Cortes , Chrysoula Tsogka , Symeon Papadimitropoulos This is my paper

Pith reviewed 2026-05-22 07:11 UTC · model grok-4.3

classification ⚛️ physics.comp-ph
keywords wave cloakingdeep learningboundary integral equationsHelmholtz equationscattering reductionlayered mediainverse designcomputational physics
0
0 comments X

The pith

Object-fitted layered cloaks reduce wave scattering more effectively than circular designs for irregular objects

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

The paper introduces a deep learning framework using an encoder-decoder architecture to design and evaluate cloaking devices made of concentric layered media for two-dimensional waves governed by the Helmholtz equation. Training relies on data generated by a boundary element method for varying layer thicknesses, applied separately to circular, star-shaped, and kite-shaped objects. Results indicate that object-fitted layer configurations achieve greater scattering reduction than simpler circular-layer designs. A sympathetic reader would care because this supplies a practical data-driven route to compare cloaking strategies and optimize performance based on object geometry without repeated analytic or full-field solves.

Core claim

The authors demonstrate through encoder-decoder networks trained on boundary element solutions that object-fitted concentric layered cloaks, parameterized uniformly by layer thicknesses, consistently outperform circular-layer designs in scattering reduction for circular, star-shaped, and kite-shaped objects under Helmholtz wave propagation.

What carries the argument

Encoder-decoder neural network trained on geometry-specific datasets of layer thickness parameters and corresponding scattering responses obtained from boundary element method solutions.

Load-bearing premise

The boundary element method solutions used to generate training data accurately represent the true scattering behavior for the chosen geometries and material parameters without significant discretization or modeling errors.

What would settle it

An independent numerical computation of the scattering cross-section for a star-shaped object with both circular and object-fitted layers using a different solver such as finite elements, showing no consistent advantage for the fitted configuration.

Figures

Figures reproduced from arXiv: 2605.21830 by Camille Carvalho, Chrysoula Tsogka, Elsie Cortes, Symeon Papadimitropoulos.

Figure 1
Figure 1. Figure 1: Example of a 5-layer configuration with star-shaped layers. The wave numbers kj , permittivities εj , and radii rj are labeled, including an innermost object defined by rob, kob, and εob in region Ω5 = Ωob. An incident plane wave u in generates a scattered field u sc bg in the background medium in Ω0 = Ωbg, defined by kbg and εbg. 4 [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The real part of the wave field solution is shown. (Left) Incoming wave field [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Decoder architecture mapping layer thicknesses [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Combined encoder-decoder architecture trained to act as the identity. The [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Log plot of the absolute error of the boundary data [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Sketch of the sample distribution and setting to generate dataset. The 200-point [PITH_FULL_IMAGE:figures/full_fig_p020_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Training and validation MSE loss curves for the simple circular-object case with [PITH_FULL_IMAGE:figures/full_fig_p020_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Evaluation of the decoder’s output for one test sample. The top row corresponds [PITH_FULL_IMAGE:figures/full_fig_p021_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Evaluation of the encoder-decoder’s output for the test sample in Figure 8. [PITH_FULL_IMAGE:figures/full_fig_p022_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Visualization of the cloaking performance of the neural network. Only the real [PITH_FULL_IMAGE:figures/full_fig_p024_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Cloaking parameters produced by the neural network for the simple circular [PITH_FULL_IMAGE:figures/full_fig_p025_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Cloaking parameters produced by the neural network for the star-object case [PITH_FULL_IMAGE:figures/full_fig_p025_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Cloaking parameters produced by the neural network for the star-object case [PITH_FULL_IMAGE:figures/full_fig_p026_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Cloaking parameters produced by the neural network for the [PITH_FULL_IMAGE:figures/full_fig_p027_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Cloaking parameters produced by the neural network for the kite-object case [PITH_FULL_IMAGE:figures/full_fig_p028_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Cloaking parameters produced by the neural network for the [PITH_FULL_IMAGE:figures/full_fig_p029_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Cloaking parameters produced by the neural network for the kite-object case [PITH_FULL_IMAGE:figures/full_fig_p029_17.png] view at source ↗
read the original abstract

We propose a deep learning framework based on an encoder-decoder architecture for the design and evaluation of cloaking devices, demonstrated in this work for two-dimensional wave propagation governed by the Helmholtz equation. The cloaks under consideration are concentric layered media surrounding the object, whose geometry and material parameters determine the scattering response. We consider circular and object-fitted layer configurations and parameterize all designs by the layer thicknesses, enabling a unified representation for direct comparison of different cloaks for the same object. Training data are generated using a boundary element formulation suitable for geometries where analytic solutions are not available, and neural networks are trained with standard hyperparameters on geometry-specific datasets. The proposed approach is applied to circular, star-shaped, and kite-shaped objects. Results show that object-fitted configurations consistently outperform simpler circular-layer designs in scattering reduction, highlighting the importance of geometry in cloaking performance. Overall, we present a flexible, data-driven approach for systematic comparison of cloaking strategies, with potential extension to more complex geometries and broadband settings.

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 / 2 minor

Summary. The manuscript proposes a deep learning framework based on an encoder-decoder architecture to design and evaluate cloaking devices for two-dimensional Helmholtz wave scattering. Training data are generated via a boundary element method (BEM) formulation for circular, star-shaped, and kite-shaped objects; neural networks are trained on geometry-specific datasets to predict scattering responses. All cloaks are parameterized uniformly by layer thicknesses, enabling direct comparison between concentric circular-layer designs and object-fitted layer configurations. The central result is that object-fitted configurations consistently outperform circular-layer designs in scattering reduction for the tested objects.

Significance. If the central comparison holds after addressing validation concerns, the work provides a flexible, data-driven methodology for systematic evaluation of cloaking strategies across irregular geometries where analytic solutions are unavailable. The unified thickness-based parameterization and geometry-specific training are strengths that facilitate reproducible comparisons; the approach has clear potential for extension to broadband or three-dimensional settings.

major comments (2)
  1. [Section 3] Section 3 (BEM data generation): No validation of the BEM solver against analytic Mie-series solutions is reported for the circular-object case, even though such references exist and the abstract notes BEM is used 'where analytic solutions are not available.' Without this cross-check, it remains possible that discretization or quadrature errors differ systematically between smooth circular boundaries and the non-smooth star/kite geometries, which would undermine the claim that object-fitted layers outperform circular ones.
  2. [Section 4.2 and Table 2] Section 4.2 and Table 2: The reported scattering-reduction metrics for object-fitted vs. circular configurations lack error bars or convergence studies with respect to BEM mesh density or NN training-set size. This makes it difficult to assess whether the observed outperformance is statistically robust or sensitive to the particular discretization choices.
minor comments (2)
  1. [Figure 3] Figure 3: The color scale for the scattered-field plots is not labeled with units or normalized range, making quantitative comparison between panels difficult.
  2. Notation: The symbol for layer thickness is introduced inconsistently (sometimes t_i, sometimes d_i) across the methods and results sections; a single consistent symbol would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback on our manuscript. We address each major comment below and have incorporated revisions to strengthen the validation and robustness analysis in the next version of the paper.

read point-by-point responses

  1. Referee: [Section 3] Section 3 (BEM data generation): No validation of the BEM solver against analytic Mie-series solutions is reported for the circular-object case, even though such references exist and the abstract notes BEM is used 'where analytic solutions are not available.' Without this cross-check, it remains possible that discretization or quadrature errors differ systematically between smooth circular boundaries and the non-smooth star/kite geometries, which would undermine the claim that object-fitted layers outperform circular ones.

    Authors: We agree that explicit validation of the BEM solver on the circular geometry would increase confidence in the results for non-circular objects. Although the abstract highlights BEM for cases without analytic solutions, we recognize the value of this cross-check. In the revised manuscript we have added a new paragraph and accompanying figure in Section 3 that directly compares BEM-computed scattering cross-sections with the analytic Mie-series solution for the circular object under several layer-thickness configurations. The relative difference remains below 0.5 % for the mesh parameters used throughout the study, confirming that discretization errors do not introduce systematic bias that would favor one cloak geometry over the other. revision: yes

  2. Referee: [Section 4.2 and Table 2] Section 4.2 and Table 2: The reported scattering-reduction metrics for object-fitted vs. circular configurations lack error bars or convergence studies with respect to BEM mesh density or NN training-set size. This makes it difficult to assess whether the observed outperformance is statistically robust or sensitive to the particular discretization choices.

    Authors: We acknowledge that the lack of quantitative uncertainty measures limits the ability to judge statistical robustness. In the revised manuscript we have updated Table 2 to include error bars given by the standard deviation across five independent neural-network training runs that differ only in random seed. We have also added a short convergence subsection in Section 4.2 that reports the change in scattering-reduction values when the BEM mesh density is doubled for a representative circular-object case; the metrics vary by less than 2 %, indicating that the reported outperformance of object-fitted layers is stable with respect to the discretization choices employed. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on independent BEM data generation

full rationale

The paper generates training data via boundary element method for circular, star-shaped, and kite-shaped objects, then trains encoder-decoder networks on geometry-specific datasets to predict scattering responses from layer thicknesses. The central comparison—that object-fitted configurations outperform circular-layer designs—is obtained by evaluating the trained networks on both parameterization types using the same externally generated BEM data. No step reduces a claimed prediction to a fitted parameter or self-citation by construction; the BEM solver is treated as an independent forward model, and the neural network approximates rather than tautologically reproduces its inputs. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework rests on standard assumptions of the Helmholtz equation for time-harmonic waves and the accuracy of the boundary element method for generating labels. No new physical entities are introduced. Hyperparameters are described as standard, implying limited free parameters beyond those implicit in the NN architecture.

axioms (2)
  • domain assumption The 2D Helmholtz equation governs the wave propagation in the layered media.
    Invoked in the abstract as the governing equation for the scattering problem.
  • domain assumption Boundary element method provides sufficiently accurate forward solutions for training data generation.
    Used to create datasets for geometries without analytic solutions.

pith-pipeline@v0.9.0 · 5717 in / 1394 out tokens · 27604 ms · 2026-05-22T07:11:34.994801+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages

  1. [1]

    1063/5.0048846

    K.-T.Lee, C.Ji, H.Iizuka, D.Banerjee, Opticalcloakingandinvisibility: From fiction toward a technological reality 129 (23) 231101.doi:10. 1063/5.0048846

    work page

  2. [2]

    Cassier, T

    M. Cassier, T. DeGiovanni, S. Guenneau, F. G. Vasquez, Active exterior cloaking for the 2d helmholtz equation with complex wavenumbers and application to thermal cloaking (2022).arXiv:2203.02075. URLhttps://arxiv.org/abs/2203.02075

    work page arXiv 2022

  3. [3]

    J. B. Pendry, D. Schurig, D. R. Smith, Controlling Electromagnetic Fields 312 (5781) 1780–1782.doi:10.1126/science.1125907

    work page doi:10.1126/science.1125907

  4. [4]

    A. Alù, N. Engheta, Achieving transparency with plasmonic and meta- material coatings, Physical Review E 72 (1) (2005) 016623.doi: 10.1103/PhysRevE.72.016623

    work page doi:10.1103/physreve.72.016623 2005

  5. [5]

    W. Cai, U. K. Chettiar, A. V. Kildishev, V. M. Shalaev, Optical cloaking with metamaterials 1 (4) 224–227.doi:10.1038/nphoton.2007.28

    work page doi:10.1038/nphoton.2007.28 2007

  6. [6]

    Schweiger, S

    M. Schweiger, S. R. Arridge, The finite-element method for the prop- agation of light in scattering media: Frequency domain case, Medical Physics 24 (6) (1997) 895–902.doi:10.1118/1.598008

    work page doi:10.1118/1.598008 1997

  7. [7]

    J. L. Volakis, A. Chatterjee, L. C. Kempel, Review of the finite-element method for three-dimensional electromagnetic scattering, J. Opt. Soc. Am. A 11 (4) (1994) 1422–1433.doi:10.1364/JOSAA.11.001422. 31

    work page doi:10.1364/josaa.11.001422 1994

  8. [8]

    Frehner, S

    M. Frehner, S. M. Schmalholz, E. H. Saenger, H. Steeb, Compari- son of finite difference and finite element methods for simulating two- dimensional scattering of elastic waves, Physics of the Earth and Plane- tary Interiors 171 (1) (2008) 112–121.doi:10.1016/j.pepi.2008.07. 003

    work page doi:10.1016/j.pepi.2008.07 2008

  9. [9]

    Bonnet, Boundary Integral Equation Methods for Solids and Fluids 34 (4) 301–302.doi:10.1023/A:1004795120236

    M. Bonnet, Boundary Integral Equation Methods for Solids and Fluids 34 (4) 301–302.doi:10.1023/A:1004795120236

    work page doi:10.1023/a:1004795120236

  10. [10]

    Masset, R

    Y. Liu, On the BEM for acoustic wave problems, Engineering Anal- ysis with Boundary Elements 107 (2019) 53–62.doi:10.1016/j. enganabound.2019.07.002

    work page doi:10.1016/j 2019

  11. [11]

    A. H. Barnett, Evaluation of Layer Potentials Close to the Boundary for Laplace and Helmholtz Problems on Analytic Planar Domains, SIAM J. Sci. Comput. 36 (2) (2014) A427–A451, publisher: Society for Industrial and Applied Mathematics. URLhttps://doi.org/10.1137/120900253

    work page doi:10.1137/120900253 2014

  12. [12]

    Colton, R

    D. Colton, R. Kress, Inverse scattering, in: Handbook of Mathematical Methods in Imaging: Volume 1, Second Edition, 2015, pp. 649–700

    work page 2015

  13. [13]

    Z. Liu, D. Zhu, S. P. Rodrigues, K.-T. Lee, W. Cai, Generative Model for the Inverse Design of Metasurfaces 18 (10) 6570–6576.doi:10.1021/ acs.nanolett.8b03171

    work page

  14. [14]

    W. Ma, F. Cheng, Y. Liu, Deep-Learning-Enabled On-Demand De- sign of Chiral Metamaterials 12 (6) 6326–6334.doi:10.1021/acsnano. 8b03569

    work page doi:10.1021/acsnano

  15. [15]

    W. Ma, F. Cheng, Y. Xu, Q. Wen, Y. Liu, Probabilistic Representation and Inverse Design of Metamaterials Based on a Deep Generative Model withSemi-SupervisedLearningStrategy31(35)1901111.doi:10.1002/ adma.201901111

    work page

  16. [16]

    Cerniauskas, H

    G. Cerniauskas, H. Sadia, P. Alam, Machine intelligence in metama- terials design: A review, Oxford Open Materials Science 4 (1) (2024) itae001.arXiv:https://academic.oup.com/ooms/article-pdf/4/1/ itae001/56750204/itae001.pdf,doi:10.1093/oxfmat/itae001. 32

    work page doi:10.1093/oxfmat/itae001 2024

  17. [17]

    Zhang, L

    H. Zhang, L. Kang, S. D. Campbell, J. T. Young, D. H. Werner, Data driven approaches in nanophotonics: A review of AI-enabled metadevices, Nanoscale 17 (41) (2025) 23788–23803.doi:10.1039/ D5NR02043C

    work page 2025

  18. [18]

    D. Liu, Y. Tan, E. Khoram, Z. Yu, Training Deep Neural Networks for the Inverse Design of Nanophotonic Structures 5 (4) 1365–1369.doi: 10.1021/acsphotonics.7b01377

    work page doi:10.1021/acsphotonics.7b01377

  19. [19]

    Kiarashinejad, S

    Y. Kiarashinejad, S. Abdollahramezani, A. Adibi, Deep learning ap- proach based on dimensionality reduction for designing electromagnetic nanostructures 6 (1) 12.doi:10.1038/s41524-020-0276-y

    work page doi:10.1038/s41524-020-0276-y

  20. [20]

    W. W. Ahmed, M. Farhat, X. Zhang, Y. Wu, Deterministic and probabilistic deep learning models for inverse design of broadband acoustic cloak, Phys. Rev. Res. 3 (2021) 013142.doi:10.1103/ PhysRevResearch.3.013142

    work page 2021

  21. [21]

    Ra, Problems in inverse scattering-illposedness, resolution, local minima, and uniquenesse, Communications of the Korean Mathematical Society 16 (3) (2001) 445–458

    J.-W. Ra, Problems in inverse scattering-illposedness, resolution, local minima, and uniquenesse, Communications of the Korean Mathematical Society 16 (3) (2001) 445–458

    work page 2001

  22. [22]

    R.-T. Wu, M. Jokar, M. R. Jahanshahi, F. Semperlotti, A physics- constrained deep learning based approach for acoustic inverse scattering problems 164 108190.doi:10.1016/j.ymssp.2021.108190

    work page doi:10.1016/j.ymssp.2021.108190 2021

  23. [23]

    Lee, GELU Activation Function in Deep Learning: A Comprehensive Mathematical Analysis and Performance.arXiv:2305.12073,doi:10

    M. Lee, GELU Activation Function in Deep Learning: A Comprehensive Mathematical Analysis and Performance.arXiv:2305.12073,doi:10. 48550/arXiv.2305.12073

    work page arXiv

  24. [24]

    T. Song, Comparative Analysis of Activation Functions in Simple Con- volutional Neural Networks, in: 2024 IEEE 6th International Conference on Civil Aviation Safety and Information Technology (ICCASIT), pp. 1031–1036.doi:10.1109/ICCASIT62299.2024.10827900

    work page doi:10.1109/iccasit62299.2024.10827900 2024

  25. [25]

    S. R. Dubey, S. K. Singh, B. B. Chaudhuri, Activation Functions in Deep Learning: A Comprehensive Survey andBenchmark.arXiv:2109. 14545,doi:10.48550/arXiv.2109.14545. 33

    work page doi:10.48550/arxiv.2109.14545

  26. [26]

    Maurya, D

    R. Maurya, D. Aggarwal, T. Gopalakrishnan, N. N. Pandey, Enhancing Deep Neural Network Convergence and Performance: A Hybrid Activa- tion Function Approach by Combining ReLU and ELU Activation Func- tion, in: 2023 Second International Conference on Informatics (ICI), pp. 1–5.doi:10.1109/ICI60088.2023.10421353

    work page doi:10.1109/ici60088.2023.10421353 2023

  27. [27]

    Kress, Boundary integral equations in time-harmonic acoustic scat- tering, Mathematical and Computer Modelling 15 (3) (1991) 229–243

    R. Kress, Boundary integral equations in time-harmonic acoustic scat- tering, Mathematical and Computer Modelling 15 (3) (1991) 229–243. URLhttps://doi.org/10.1016/0895-7177(91)90068-I

    work page doi:10.1016/0895-7177(91)90068-i 1991

  28. [28]

    Carvalho, S

    C. Carvalho, S. Chaillat, E. Cortes, C. Tsogka, Discretization in Mul- tilayered Media with High Contrasts: Is It All About the Boundaries? (Feb. 2026).arXiv:2602.18629

    work page arXiv 2026

  29. [29]

    Raissi, P

    A. Klöckner, A. Barnett, L. Greengard, M. O’Neil, Quadrature by ex- pansion: A new method for the evaluation of layer potentials, Journal of Computational Physics 252 (2013) 332–349.doi:10.1016/j.jcp. 2013.06.027

    work page doi:10.1016/j.jcp 2013

  30. [30]

    Carvalho, S

    C. Carvalho, S. Khatri, A. D. Kim, Asymptotic approximations for the close evaluation of double-layer potentials, SIAM Journal on Scientific Computing 42 (1) (2020) A504–A533.doi:10.1137/18M1218698

    work page doi:10.1137/18m1218698 2020

  31. [31]

    Pérez-Arancibia, C

    C. Pérez-Arancibia, C. Turc, L. Faria, Planewave density interpolation methods for 3d helmholtz boundary integral equations, SIAM Journal on Scientific Computing 41 (4) (2019) A2088–A2116.doi:10.1137/ 19M1239866

    work page 2019

  32. [32]

    E. A. Cortes, C. Carvalho, S. Chaillat, C. Tsogka, Fast and accurate boundary integral equation methods for the multi-layer transmission problem, in: WAVES2024-The16thInternationalConferenceonMath- ematical and Numerical Aspects of Wave Propagation, Edmond, Berlin, Germany, 2024.doi:10.17617/3.MBE4AA

    work page doi:10.17617/3.mbe4aa 2024

  33. [33]

    Q. Sun, E. Klaseboer, B.-C. Khoo, D. Y. C. Chan, Boundary regularized integral equation formulation of the Helmholtz equation in acoustics, Royal Society Open Science 2 (1) (2015) 140520

    work page 2015

  34. [34]

    M. Yan, W. Yan, M. Qiu, Invisibility Cloaking by Coordinate Trans- formation, in: Progress in Optics, Vol. 52, Elsevier, 2009, pp. 261–304. doi:10.1016/S0079-6638(08)00006-1. 34

    work page doi:10.1016/s0079-6638(08)00006-1 2009