Completing the rigorous high-frequency bridge from inverse Helmholtz scattering with beam (Husimi) data to inverse Liouville scattering for general refractive indices

Chen, Ding, Li and Zepeda-Núñez (SIAM J. Imaging Sci. 2023) showed that probing an unknown refractive index with tightly concentrated monochromatic beams and measuring only intensity data through the Husimi transform makes the inverse Helmholtz scattering problem converge, in the high-frequency limit, to the inverse Liouville (ray-tracing) problem, thereby inheriting its stability in contrast to unstable plane-wave inversion. The bridge is rigorous only under restrictive hypotheses: the semiclassical convergence is proved for radially symmetric indices, the stability theorem holds only for near-constant, non-trapping media, and translating Husimi intensity data into the phase data used by the classical problem is an unsettled phase-retrieval question. This problem asks to complete (or rigorously refute) the bridge for general refractive indices: establish the convergence of the Wigner/Husimi data to the Liouville exit data on the non-trapped phase-space set for non-radial, less regular, discontinuous, and trapping media; characterize stability beyond the near-constant non-trapping regime; and settle uniqueness, stability, constructibility, or impossibility of the Husimi-to-phase data translation. All three components remain open in the literature to date.

Background

The inverse scattering problem for time-harmonic waves asks for the spatially varying refractive index $n(x)$ inside a bounded domain from measurements of scattered waves outside it. The forward model is the Helmholtz equation $\Delta u^{k} + k^{2} n(x) u^{k} = -S^{k}_{H}$ with an outgoing radiation condition at infinity, where $k$ is the wavenumber. With the classical illumination - plane waves, which are direction-localized but spatially delocalized - the fixed-frequency inverse problem is severely ill-conditioned and unstable; Hähner and Hohage (2001) quantify this instability, and in geophysical full-waveform inversion (FWI, PDE-constrained fitting of the measured waveform) it manifests as cycle-skipping: convergence of the nonconvex data misfit to spurious local minima. Chen, Ding, Li and Zepeda-Núñez (2023) proposed a reformulation grounded in the wave-particle duality: probe with tight beams and measure in phase space. Their sources are Gaussian wave packets of width $1/k$, concentrated both in position $x_s$ and direction $v_s$ as $k\to\infty$, and the data are taken through the Husimi transform $H^{k}u(x,v) = (k/2\pi)^{d}\,|(u(\cdot), \varphi^{k}_{v}(x-\cdot))_{L^{2}}|^{2}$, the squared modulus of the field tested against oscillatory wave packets localized in position and direction - a spectrogram-type, intensity-only (phase-blind) measurement. The analytic bridge between the wave and particle descriptions is the Wigner transform $W^{k}[u](x,v)$, a phase-space quasi-probability distribution of the wave field: in the high-frequency (semiclassical) limit it satisfies, formally and under hypotheses rigorously, the Liouville equation $v\cdot\nabla_{x} f + \tfrac{1}{2}\nabla_{x} n(x)\cdot\nabla_{v} f = S_{L}$, the kinetic description of point particles (rays) governed by the Hamiltonian $\tfrac{1}{2}|v|^{2} - \tfrac{1}{2}n(x)$, whose characteristics are $\dot{x}=v$ and $\dot{v}=\tfrac{1}{2}\nabla_{x} n(x)$. Convergence holds in the weak-* sense - testing the phase-space densities against smooth observables rather than pointwise evaluation. The rigorous Wigner-limit theory for the Helmholtz equation with source terms was built by Benamou, Castella, Katsaounis and Perthame (2002) and Castella, Perthame and Runborg (2002), with sharp interfaces and discontinuous indices treated in the forward setting by Fouassier (2007, 2008); Chen and Li (2021) established the analogous semiclassical bridge for the time-dependent Schrödinger inverse problem, where the Wigner-limit theory for general potentials is classical. The central results of Chen, Ding, Li and Zepeda-Núñez (2023) are two theorems. First, an asymptotic equivalence: as $k\to\infty$, the Wigner transform of the beam-generated Helmholtz solution converges weak-* to the Liouville solution driven by a source concentrated at $(x_s, v_s)$, and the Husimi data converge to the classical exit data - the in-out map that pairs each incoming phase-space point with the point where its ray exits - so the inverse Helmholtz problem with beam data inherits the well-posedness of the inverse Liouville problem, an X-ray-transform-type inversion (recovering a function from its integrals along the ray family) analyzed near constant indices by Novikov (1999). Second, a stability theorem: if $n$ is close to a constant - both $\nabla n$ and the Hessian $Hn$ uniformly bounded by a small $\Delta$ - and non-trapping, meaning no ray of the Hamiltonian flow stays confined to the domain forever, then the Liouville data determine $n$ approximately via the X-ray transform with error of order $\Delta^{2}$. Each pillar rests on restrictions. The convergence theorem is rigorously justified only under additional conditions on $n$: the authors note that a full proof for general $n$ is complicated and certify radial symmetry as the rigorous case, leaning on the Wigner-limit literature above. The stability theorem leaves strongly varying, discontinuous, and trapping media (sharp interfaces, resonant cavities) uncharacterized. And while in the strict double limit $k\to\infty$ with vanishing source width the Husimi spectrogram concentrates on the exit ray and thereby bypasses phase information, at finite $k$ the translation from Husimi intensity data to the complex-valued phase data is an instance of phase retrieval - recovering a signal from windowed Fourier magnitudes - a problem class known to be unstable in general infinite-dimensional settings (Alaifari, Daubechies, Grohs and Yin 2019; Alaifari and Grohs 2021), though stable for structured signal classes (Grohs and Rathmair 2019; see the survey by Grohs, Koppensteiner and Rathmair 2020). These three gaps delimit the open problem.

Problem Statement

The problem is to complete - or rigorously refute - the high-frequency bridge between the wave and ray descriptions of inverse scattering with beam (Husimi) data, by removing the restrictions under which it is currently proved. In the framework of Chen, Ding, Li and Zepeda-Núñez (Helmholtz equation $\Delta u^{k} + k^{2} n(x) u^{k} = -S^{k}_{H}$ with outgoing radiation condition, Gaussian-beam sources of width $1/k$ concentrated at $(x_s, v_s)$, Husimi intensity data measured on the boundary, Liouville equation with Hamiltonian $\tfrac{1}{2}|v|^{2} - \tfrac{1}{2}n(x)$, and the in-out map $(x_s, v_s) \mapsto (x_r, v_r)$ pairing incoming and exit phase-space points), establish: (1) Convergence for general media: prove or disprove that, as $k\to\infty$, the Wigner transforms of the beam-generated solutions converge weak-* to the Liouville solution and the Husimi data converge to the classical exit data for refractive indices beyond the radially symmetric case - including non-radial smooth, less regular, discontinuous, and trapping $n$ - where the exit-data convergence is asserted on the non-trapped phase-space set, the rays of a trapping medium that stay confined forever contributing only to the interior Liouville limit, with rigorous treatment of the radiation condition at infinity and of the concentrated-source scaling, and with the media class explicitly stated. (2) Stability beyond near-constant media: characterize the stability of the beam-data inverse problem - quantitative data-to-medium bounds, or provable instability - for strongly varying, discontinuous, or trapping refractive indices (a medium is trapping if some ray of the Hamiltonian flow remains confined forever; for such media the data map used where the in-out map fails to be defined must be specified), replacing the current hypotheses that $\|\nabla n\|_{L^{\infty}}$ and the uniform Frobenius norm of the Hessian $Hn$ are bounded by a small $\Delta$ and that $n$ is non-trapping. (3) Husimi-to-phase translation: determine whether the Husimi intensity data determine the complex-valued boundary field - equivalently the phase data the classical inversion needs - outside the single-beam regime where the limiting spectrogram concentrates on one ray: uniqueness, stability, a constructive reconstruction algorithm, or an impossibility/instability result. These three components jointly serve one scientific objective: a mathematically justified (or rigorously refuted) statement that probing with tightly concentrated monochromatic beams yields asymptotically stable reconstruction of the medium for general refractive indices, with ray tomography as the high-frequency limit. An answer that settles one of (1)-(3) for an explicitly stated class of media is admissible as an answer to that component; a claim to have resolved the problem as a whole must address all three. The source work additionally flags two companion practical directions that are recorded but not part of this mathematical completion boundary: numerical schemes quantifying discretization-versus-asymptotic errors at very high $k$, and globally convergent algorithms for the nonconvex PDE-constrained inversion.

References

1. Shi Chen, Zhiyan Ding, Qin Li, Leonardo Zepeda-Núñez, High-Frequency Limit of the Inverse Scattering Problem: Asymptotic Convergence from Inverse Helmholtz to Inverse Liouville, SIAM Journal on Imaging Sciences 16(1):111-143, 2023. DOI: 10.1137/22M147075X. https://doi.org/10.1137/22M147075X (preprint arXiv:2201.03494, https://arxiv.org/abs/2201.03494)
2. Shi Chen, Qin Li, Semiclassical limit of an inverse problem for the Schrödinger equation, Research in the Mathematical Sciences 8, Article 39, 2021. DOI: 10.1007/s40687-021-00279-y. https://doi.org/10.1007/s40687-021-00279-y
3. J.-D. Benamou, F. Castella, T. Katsaounis, B. Perthame, High frequency limit of the Helmholtz equations, Revista Matemática Iberoamericana 18(1):187-209, 2002. DOI: 10.4171/RMI/315. https://doi.org/10.4171/RMI/315
4. F. Castella, B. Perthame, O. Runborg, High frequency limit of the Helmholtz equation. II. Source on a general smooth manifold, Communications in Partial Differential Equations 27(3-4):607-651, 2002. DOI: 10.1081/PDE-120002867. https://doi.org/10.1081/PDE-120002867
5. Elise Fouassier, High frequency limit of Helmholtz equations: refraction by sharp interfaces, Journal de Mathématiques Pures et Appliquées 87(2):144-192, 2007. DOI: 10.1016/j.matpur.2006.11.002. https://doi.org/10.1016/j.matpur.2006.11.002
6. Elise Fouassier, High frequency limit of Helmholtz equations: the case of a discontinuous index, Journées Équations aux Dérivées Partielles, 2008. DOI: 10.5802/jedp.31. https://doi.org/10.5802/jedp.31
7. Rima Alaifari, Ingrid Daubechies, Philipp Grohs, Rujie Yin, Stable Phase Retrieval in Infinite Dimensions, Foundations of Computational Mathematics 19(4):869-900, 2019. DOI: 10.1007/s10208-018-9399-7. https://doi.org/10.1007/s10208-018-9399-7
8. Rima Alaifari, Philipp Grohs, Gabor phase retrieval is severely ill-posed, Applied and Computational Harmonic Analysis 50:401-419, 2021. DOI: 10.1016/j.acha.2019.09.003. https://doi.org/10.1016/j.acha.2019.09.003
9. Philipp Grohs, Sarah Koppensteiner, Martin Rathmair, Phase Retrieval: Uniqueness and Stability, SIAM Review 62(2):301-350, 2020. DOI: 10.1137/19M1256865. https://doi.org/10.1137/19M1256865
10. Philipp Grohs, Martin Rathmair, Stable Gabor Phase Retrieval and Spectral Clustering, Communications on Pure and Applied Mathematics 72(5):981-1043, 2019. DOI: 10.1002/cpa.21799. https://doi.org/10.1002/cpa.21799
11. Roman G. Novikov, Small angle scattering and X-ray transform in classical mechanics, Arkiv för Matematik 37(1-2):141-169, 1999. DOI: 10.1007/BF02384831. https://doi.org/10.1007/BF02384831
12. Peter Hähner, Thorsten Hohage, New Stability Estimates for the Inverse Acoustic Inhomogeneous Medium Problem and Applications, SIAM Journal on Mathematical Analysis 33(3):670-685, 2001. DOI: 10.1137/S0036141001383564. https://doi.org/10.1137/S0036141001383564
13. Borong Zhang, Leonardo Zepeda-Núñez, Qin Li, Solving the wide-band inverse scattering problem via equivariant neural networks, Journal of Computational and Applied Mathematics 451:116050, 2024. DOI: 10.1016/j.cam.2024.116050. https://doi.org/10.1016/j.cam.2024.116050
14. Borong Zhang, Martín Guerra, Qin Li, Leonardo Zepeda-Núñez, Back-Projection Diffusion: Solving the wideband inverse scattering problem with diffusion models, Computer Methods in Applied Mechanics and Engineering 443:118036, 2025. DOI: 10.1016/j.cma.2025.118036. https://doi.org/10.1016/j.cma.2025.118036