Time-domain direct sampling for inverse electromagnetic scattering beyond the point-like quasi-static regime: extended scatterers, shape reconstruction, and reliable parameter selection

The time-domain direct sampling method of Geng, Song, Wang and Wang (SIAM J. Imaging Sci. 2025) locates dielectric scatterers in electromagnetic inverse medium scattering from time-dependent boundary data generated by a single incident source, with no restriction on the pulse waveform. Its justification and numerics, however, are confined to well-separated point-like scatterers in the quasi-static scaling regime where the scatterer size times the frequency is small: the localization estimate carries an error term forcing the scatterer scale to zero, the truncation frequency and damping parameter are chosen heuristically, and shape recovery fails for a single source. The open problem is to extend this framework, with rigorous justification, to extended scatterers outside the quasi-static regime, to shape reconstruction of electromagnetic scatterers, and to principled selection of the truncation frequency and damping parameters. No published resolution of any of these components for Maxwell's equations is known to date: sibling extensions exist only for acoustics, where shape recovery still requires multiple sources numerically and the damping choice is explicitly left open, and, more recently, for elasticity, still in the point-like regime.

Background

Inverse electromagnetic medium scattering asks what can be learned about an unknown object from the electromagnetic waves it scatters. In the time-domain setting considered here, a dielectric inhomogeneity — characterized by an electric permittivity $\epsilon(x)$ that differs from the vacuum value $\epsilon_0$ on a compact region, with magnetic permeability fixed at the vacuum value $\mu_0$ and zero electric conductivity — is illuminated by a causal pulse, and the scattered electric field $\mathcal{E}^s(x,t)$ is recorded over time on a measurement surface $\Gamma$ away from the object. Pulsed broadband data of this kind is what ultra-wideband radar and pulsed nondestructive-evaluation instruments produce, so methods that exploit it directly are of practical as well as theoretical interest.

Because iterative reconstruction is expensive and needs prior information, a large literature develops qualitative (sampling-type) methods, which build an indicator function on a grid of sampling points: large indicator values mark where scatterers are, without solving a nonlinear optimization. The direct sampling method (DSM), introduced for frequency-domain electromagnetic medium scattering by Ito, Jin and Zou, evaluates the indicator by correlating measured data with analytically known test fields. The linear sampling method (LSM) instead characterizes the scatterer through the solvability of a linear integral equation (the near-field equation) driven by the data. Time-domain versions of both families go back to Chen, Haddar, Lechleiter and Monk and to Guo, Monk and Colton. For the LSM, Cakoni, Monk and Selgas rigorously proved solvability of the time-domain near-field equation for acoustic waves, while Lähivaara, Monk and Selgas developed the time-domain LSM for electromagnetic waves numerically, its solvability for Maxwell's equations remaining open.

The source, by Geng, Song, Wang and Wang (arXiv 2024; published in SIAM J. Imaging Sci. 2025), constructs a time-domain DSM for electromagnetic medium scattering that needs only a single incident source. Its indicator $\mathcal{I}(z)$ integrates in time the squared magnitude of a boundary integral that time-shifts the measured scattered field by the travel time $c_{0}^{-1}|x-z|$ (with $c_{0}$ the speed of light in vacuum) from each receiver to the sampling point $z$ and weights it by a damped free-space factor $\mathrm{e}^{-\sigma t}/(4\pi|x-z|)$; the time shift makes $\mathcal{I}(z)$ peak where a scatterer would re-emit the recorded signal. The proof works through the Fourier–Laplace transform (the Laplace transform evaluated along a vertical line $\omega+\mathrm{i}\sigma$ in the complex frequency plane, which converts causal time-domain fields into frequency-domain ones while retaining a damping parameter $\sigma>0$), showing that the time-domain indicator is a superposition of frequency-domain indicators whose localization mechanism is governed by modified spherical Bessel functions $i_n$ — the regular radial solutions arising when the transformed field is expanded on the sphere, whose exponential growth pins the indicator near the scatterers.

The proved localization is, however, restricted. The scatterers must form a finite union of well-separated point-like components $\Omega_j=y_j+\rho B_j$ (centers $y_j$, common small scale $\rho$), and the estimate carries error terms $\mathcal{O}(\rho|\omega_0|)$ from the quasi-static approximation — the low-frequency regime in which the scatterer is small compared with the wavelength so the internal field is nearly uniform — forcing $\rho|\omega_0|\to 0$; $\mathcal{O}(|\omega_0|^{-2s})$ from truncating frequencies above $\omega_0$; and $N_j\mathcal{O}(1/L)^{2}$ from multiple scattering between components separated by $L$. Numerically the method locates small cube-shaped scatterers well but, with one source, does not recover their shape. The companion acoustic paper of Guo, Li and Wang, on which the electromagnetic analysis is modeled, shows the same structure: there, shape recovery for extended targets requires multiple incident sources and remains heuristic, and the choice of the damping parameter $\sigma$ is explicitly left as future work.

Problem Statement

Consider time-dependent electromagnetic scattering in $\mathbb{R}^{3}$ governed by Maxwell's equations: a dielectric inhomogeneity with permittivity $\epsilon(x)\in L^{\infty}(\mathbb{R}^{3})$ (contrast $\epsilon(x)-\epsilon_{0}$ compactly supported in $\Omega$, permeability $\mu\equiv\mu_{0}$, zero conductivity) is illuminated by a causal incident pulse $\mathcal{E}^{i}$, and the scattered electric field $\mathcal{E}^{s}(x,t)$ is measured for all $t\geq 0$ on a measurement surface $\Gamma$ away from the scatterers — in the analysis, a sphere $\{x\in\mathbb{R}^{3}:|x|=r\}$ of radius $r$ enclosing them. Geng–Song–Wang–Wang introduced, for sampling points $z$ in a search domain containing $\Omega$, the indicator

$$\mathcal{I}(z)=\int_{-\infty}^{+\infty}\left|\int_{\Gamma}\mathcal{E}^{s}\bigl(x,t+c_{0}^{-1}|x-z|\bigr)\frac{\mathrm{e}^{-\sigma(t+c_{0}^{-1}|x-z|)}}{4\pi|x-z|}\,\mathrm{d}s(x)\right|^{2}\mathrm{d}t$$

where $c_{0}=(\epsilon_{0}\mu_{0})^{-1/2}$ is the speed of light in vacuum and $\sigma>0$ is the damping parameter of the Fourier–Laplace transform. They proved, via the Fourier–Laplace transform and an expansion in modified spherical Bessel functions, that near each scatterer component

$$\mathcal{I}(z)\leq\rho^{6}\mathrm{e}^{-2\sigma c_{0}^{-1}r}\left\{M_{j}\left(1+\mathcal{O}\Big(\frac{|\omega_{0}|}{r}\Big)+\mathcal{O}(\rho|\omega_{0}|)\right)^{2}+\mathcal{O}\big(|\omega_{0}|^{-2s}\big)+N_{j}\mathcal{O}\Big(\frac{1}{L}\Big)^{2}\right\}$$

with equality if and only if $z=y_{j}$ — but only for well-separated point-like scatterers $\Omega_{j}=y_{j}+\rho B_{j}$ (centers $y_{j}$, common small scale $\rho$, inter-component separation $L$), with a truncation frequency $\omega_{0}$ satisfying $1<|\omega_{0}|\ll 1/\rho$ and a decay order $s\gg 1$, in the asymptotic regime $r\to\infty$, $\rho|\omega_{0}|\to 0$, $L\to\infty$ and $|\omega_{0}|^{2s}\to\infty$; here $M_{j}$ and $N_{j}$ are positive constants determined by the incident field and the permittivity contrast on each component. The error terms thus force small scatterers at low frequency (the quasi-static term $\mathcal{O}(\rho|\omega_{0}|)$), a bounded truncation frequency, and large inter-scatterer separation.

The open problem is to extend this time-domain direct sampling framework for Maxwell's equations, with rigorous justification, along three jointly linked fronts that together serve one objective — turning the single-source time-domain direct sampling method from a locator of small inclusions, valid in an asymptotic regime, into a rigorously justified and parameter-robust imaging method for general electromagnetic scatterers:

1. Justify the indicator's localization (and ideally its quantitative decay away from the scatterers and its stability under noisy data) for extended scatterers of finite size, dropping the quasi-static smallness $\rho|\omega_{0}|\to 0$ and the point-like decomposition of the unknown object.
2. Establish, with rigorous justification, shape reconstruction of electromagnetic scatterers, going beyond the current state in which a single source yields locations but not shapes.
3. Establish principled rules for selecting the truncation frequency $\omega_{0}$ and the damping parameter $\sigma$, so that the leading localization term provably dominates the truncation remainder while retaining useful frequency coverage, in place of the current heuristics (maximum or center frequency of the incident field).

A related but separable theoretical question, flagged by the same source and not a prerequisite for the objective above, is the solvability of the near-field equation underlying the time-domain linear sampling method for electromagnetic waves, proved for acoustic waves by Cakoni–Monk–Selgas (2021) but open for Maxwell's equations. This question is optional context rather than a required deliverable: the task is resolved by the three fronts above.

References

1. Chen Geng, Minghui Song, Xianchao Wang, Yuliang Wang. Time-Domain Direct Sampling Method for Inverse Electromagnetic Scattering with a Single Incident Source. SIAM Journal on Imaging Sciences 18(2):1208–1234, 2025. DOI: 10.1137/24m1701071, https://doi.org/10.1137/24m1701071 (preprint: arXiv:2410.07703, https://arxiv.org/abs/2410.07703)
2. Yukun Guo, Hongjie Li, Xianchao Wang. A Novel Time-Domain Direct Sampling Approach for Inverse Scattering Problems in Acoustics. SIAM Journal on Applied Mathematics 84(5):2152–2174, 2024. DOI: 10.1137/23m1622854, https://doi.org/10.1137/23m1622854 (preprint: arXiv:2312.04241, https://arxiv.org/abs/2312.04241)
3. Lefu Cai, Hongjie Li, Xianchao Wang. A Non-Decoupled Time-Domain Direct Sampling Method for Inverse Elastic Medium Scattering. arXiv:2607.08067, 2026. https://arxiv.org/abs/2607.08067
4. Timo Lähivaara, Peter Monk, Virginia Selgas. The Time Domain Linear Sampling Method for Determining the Shape of Multiple Scatterers Using Electromagnetic Waves. Computational Methods in Applied Mathematics 22(4):889–913, 2022. DOI: 10.1515/cmam-2021-0190, https://doi.org/10.1515/cmam-2021-0190
5. Fioralba Cakoni, Peter Monk, Virginia Selgas. Analysis of the linear sampling method for imaging penetrable obstacles in the time domain. Analysis & PDE 14(3):667–688, 2021. DOI: 10.2140/apde.2021.14.667, https://doi.org/10.2140/apde.2021.14.667
6. Q. Chen, H. Haddar, A. Lechleiter, P. Monk. A sampling method for inverse scattering in the time domain. Inverse Problems 26(8):085001, 2010. DOI: 10.1088/0266-5611/26/8/085001, https://doi.org/10.1088/0266-5611/26/8/085001
7. Kazufumi Ito, Bangti Jin, Jun Zou. A direct sampling method for inverse electromagnetic medium scattering. Inverse Problems 29(9):095018, 2013. DOI: 10.1088/0266-5611/29/9/095018, https://doi.org/10.1088/0266-5611/29/9/095018
8. Y. Guo, P. Monk, D. Colton. Toward a time domain approach to the linear sampling method. Inverse Problems 29(9):095016, 2013. DOI: 10.1088/0266-5611/29/9/095016, https://doi.org/10.1088/0266-5611/29/9/095016
9. Juan Liu, Fuquan Yang, Xiaoyi Yan, Jiguang Sun. Extended sampling method for inverse electromagnetic scattering problems using a single incident wave. Inverse Problems 42(1):015010, 2026. DOI: 10.1088/1361-6420/ae3799, https://doi.org/10.1088/1361-6420/ae3799
10. Yu Sun, Bo Chen, Peng Gao, Qiuyi Li, Yao Sun. Inverse obstacle scattering with a single moving emitter. arXiv:2510.09385, 2025. https://arxiv.org/abs/2510.09385
11. Fenglin Sun, Hongxia Guo. A direct sampling method for inverse time-dependent electromagnetic source problems: reconstruction of the radiating time and spatial support. arXiv:2603.09143, 2026. https://arxiv.org/abs/2603.09143
12. Thu Le, Dinh-Liem Nguyen. On orthogonality sampling method for Maxwell's equations and its applications to experimental data. Philosophical Transactions of the Royal Society A 383(2305):20240051, 2025. DOI: 10.1098/rsta.2024.0051, https://doi.org/10.1098/rsta.2024.0051
13. Lu Zhao, Heping Dong, Zhiyong Cheng. A novel time-domain iterative method for a three-dimensional inverse acoustic obstacle scattering problem. arXiv:2607.02286, 2026. https://arxiv.org/abs/2607.02286
14. Xiaodong Liu, Xianchao Wang. Quantitative Direct Sampling for Initial Acoustic Sources. arXiv:2604.20187, 2026. https://arxiv.org/abs/2604.20187
15. Bangti Jin, Fengru Wang, Jun Zou. An Iterative Direct Sampling Method for Reconstructing Moving Inhomogeneities in Parabolic Problems. arXiv:2511.08197, 2025. https://arxiv.org/abs/2511.08197