The direct trace-formula scheme of Chen, Duan and Rokhlin (J. Comput. Phys. 228:3209–3231, 2009) reconstructs the density ρ\rho, potential qq, and attenuation γ\gamma of a laterally invariant (layered) acoustic medium from the boundary impedance measured over a frequency band for at least two incidence angles, with reconstruction error decaying as 1/am−11/a^{m-1} for profiles with mm continuous derivatives and band limit aa. Its convergence analysis, however, is proved only for non-attenuating media (γ=0\gamma=0), because the boundedness estimate underlying the Neumann-series expansion requires the Liouville change of variable to be real; numerically the scheme still works for small attenuation, and it destabilizes or fails for nonsmooth and discontinuous profiles. Establishing a rigorous theoretical foundation — convergence and stability guarantees, or a rigorously analyzed modification — for nonzero attenuation, and for nonsmooth or discontinuous coefficient profiles (with quantitative noise handling via overdetermined multi-incidence data), remains open.

Background

Inverse scattering in a layered acoustic medium asks for the medium's depth profiles — sound speed c(z)c(z), density ρ(z)\rho(z), and attenuation γ(z)\gamma(z), all assumed to vary only with depth zz — to be reconstructed from measurements of incident and scattered monochromatic fields. After separating a time-harmonic field with incidence angle θ\theta to the layer normal, the depth-dependent part satisfies the one-dimensional Helmholtz equation ϕ′′−(ρ′/ρ)ϕ′+k2(1+q+iγ−α2)ϕ=0\phi''-(\rho'/\rho)\phi'+k^2(1+q+i\gamma-\alpha^2)\phi=0, where α=sin⁡θ\alpha=\sin\theta is the inclination parameter, and the potential qq and attenuation γ\gamma encode the sound speed through c02/c2(z)=1+q(z)+iγ(z)c_0^2/c^2(z)=1+q(z)+i\gamma(z) with c0c_0 the background speed. Chen, Duan and Rokhlin (2009), extending the trace-formula method of Chen and Rokhlin (1992) and Chen (1997) which allowed only the speed to vary, reconstruct ρ\rho, qq, γ\gamma on [0,1][0,1] from the impedance function p+(0,k)p_+(0,k) — a logarithmic derivative of the outgoing solution obtainable from reflection data — sampled over a symmetric frequency band [−a,a][-a,a] for at least two values of α\alpha. Their algorithm is direct (non-iterative), costs O(Nk⋅Nx)O(N_k\cdot N_x), and is built on trace formulas: identities expressing the coefficients of the large-kk asymptotic expansion of the impedance in terms of ρ\rho, qq, γ\gamma and their derivatives. For media whose profiles have mm continuous derivatives, the reconstruction error decays as 1/am−11/a^{m-1} as aa grows — Fourier-transform-like accuracy. The analytical core converts the Helmholtz equation to Schrödinger form via a Liouville-type change of independent variable t(x)=∫1+q−α2+iγ dτt(x)=\int\sqrt{1+q-\alpha^2+i\gamma}\,d\tau; the reconstruction is then reduced to a Volterra integral equation solved by a Neumann series (an iterative series expansion), whose key boundedness estimate ∣Fk∣≤1|F_k|\le 1 requires tt to be real — which holds if and only if γ=0\gamma=0. Consequently the convergence theory covers only non-attenuating media, while the numerical experiments show the scheme still working when ∣γ(x)∣≪∣1+q(x)−α2∣|\gamma(x)|\ll|1+q(x)-\alpha^2| and failing for larger attenuation. Two further regularity gaps are documented in the same paper. First, for profiles that are continuous but whose derivatives jump, in the presence of attenuation, the trace-formula integrand p+(x,k)−p−(x,k)p_+(x,k)-p_-(x,k) becomes extremely oscillatory and blows up as kk increases near the right edge of the inhomogeneity, and the algorithm does not converge. Second, for discontinuous profiles the reconstruction exhibits Gibbs-type oscillatory artifacts (the classical overshoot and ringing of Fourier-type approximation at a jump discontinuity) and in general does not converge, with two special categories — qq discontinuous with ρ\rho constant and γ\gamma absent; γ\gamma discontinuous with ρ,q\rho,q constant — still converging at order O(1/a)O(1/\sqrt a). Finally, because two unknowns (the complex combination q+iγq+i\gamma and the density ρ\rho) must be recovered, at least two inclination parameters are needed, and the authors propose — without developing — using more than two α\alpha's to form an overdetermined system (more equations than unknowns) as a noise-control mechanism; the observed reconstruction error is proportional to perturbations of the impedance data with proportionality constant 1. Adjacent methodology exists but covers different ground: Gel'fand–Levitan-type integral-equation methods (recently extended numerically to simultaneous velocity–density inversion with jumps), iterative optimization with frequency continuation, boundary-control-based non-iterative reconstruction of the wave speed, and a recent direct method for piecewise-constant wave speed based on orthogonal polynomials on the unit circle. None provides the missing convergence theory for the attenuating, multi-parameter, or low-regularity regimes of the trace-formula scheme.

Problem Statement

Establish a rigorous theoretical foundation — convergence and stability guarantees, for the original scheme or through a rigorously analyzed modification — for trace-formula direct reconstruction of the layered acoustic medium parameters (ρ,q,γ)(\rho,q,\gamma) from impedance data p+(0,k)p_+(0,k) on a frequency band [−a,a][-a,a] for two or more inclination parameters α\alpha, in the regimes excluded by the analysis of Chen, Duan and Rokhlin (2009). The facets, which jointly define the regime coverage the method's own numerics claim but its theory does not, and which must all be addressed for a full resolution (a result covering only some of them is partial): (i) nonzero attenuation — prove convergence with explicit error bounds for γ≠0\gamma\ne 0 under explicit small-attenuation conditions of the type ∣γ(x)∣≪∣1+q(x)−α2∣|\gamma(x)|\ll|1+q(x)-\alpha^2| (larger attenuation requiring a rigorously analyzed modification), replacing or repairing the Neumann-series boundedness estimate that requires the Liouville variable t(x)=∫1+q−α2+iγ dτt(x)=\int\sqrt{1+q-\alpha^2+i\gamma}\,d\tau to be real; (ii) nonsmooth profiles — for continuous ρ,q\rho,q with discontinuous derivatives in the presence of attenuation, explain and control the oscillatory blow-up of the trace-formula integrand p+−p−p_+-p_-; (iii) discontinuous profiles — give rigorous convergence guarantees (or a rigorously analyzed modification with proven rates) for discontinuous ρ,q,γ\rho,q,\gamma beyond the two special categories currently known to converge at O(1/a)O(1/\sqrt a), addressing the Gibbs-type artifacts. As an auxiliary facet of the same objective, develop the proposed overdetermined formulation with more than two α\alpha's into a quantified noise-robust reconstruction with proven error propagation.

References

  1. Yu Chen, Ran Duan, Vladimir Rokhlin, 'On the inverse scattering problem in the acoustic environment', Journal of Computational Physics 228(9):3209–3231, 2009. DOI: 10.1016/j.jcp.2008.12.034, https://doi.org/10.1016/j.jcp.2008.12.034
  2. Ran Duan, Vladimir Rokhlin, 'On the inverse scattering problem in the acoustic environment', Technical Report YALEU/DCS/TR-1395, Yale University, March 2008. DOI: 10.21236/ada487942, http://www.cs.yale.edu/publications/techreports/tr1395.pdf
  3. Yu Chen, Vladimir Rokhlin, 'On the inverse scattering problem for the Helmholtz equation in one dimension', Inverse Problems 8(3):365–391, 1992. DOI: 10.1088/0266-5611/8/3/002, https://doi.org/10.1088/0266-5611/8/3/002
  4. Yu Chen, 'Inverse scattering via Heisenberg's uncertainty principle', Inverse Problems 13(2):253–282, 1997. DOI: 10.1088/0266-5611/13/2/005, https://doi.org/10.1088/0266-5611/13/2/005
  5. Sophia Bugarija, Peter C. Gibson, Guanghui Hu, Peijun Li, Yue Zhao, 'Inverse scattering for the one-dimensional Helmholtz equation with piecewise constant wave speed', Inverse Problems 36(7):075008, 2020. DOI: 10.1088/1361-6420/ab89c4, arXiv:1912.05360, https://arxiv.org/abs/1912.05360
  6. Gang Bao, Faouzi Triki, 'Stability for the multifrequency inverse medium problem', Journal of Differential Equations 269(9):7106–7128, 2020. DOI: 10.1016/j.jde.2020.05.021, arXiv:2002.09242, https://arxiv.org/abs/2002.09242
  7. Kiril Datchev, Maarten V. de Hoop, 'Iterative reconstruction of the wave speed for the wave equation with bounded frequency boundary data', Inverse Problems 32(2):025008, 2016. DOI: 10.1088/0266-5611/32/2/025008, arXiv:1506.09014, https://arxiv.org/abs/1506.09014
  8. Elena Beretta, Maarten V. de Hoop, Lingyun Qiu, Otmar Scherzer, 'Inverse boundary value problem for the Helmholtz equation: multi-level approach and iterative reconstruction', arXiv:1406.2391, 2014. https://arxiv.org/abs/1406.2391
  9. Peter C. Gibson, 'Acoustic imaging of layered media', Journal of Computational Physics 372:524–545, 2018. DOI: 10.1016/j.jcp.2018.06.053, https://doi.org/10.1016/j.jcp.2018.06.053
  10. Ru-Shan Wu, Huijing He, 'Simultaneous density-velocity inversion by extended Gel'fand-Levitan-Marchenko solution in layered acoustic media for oblique incidence', arXiv:2007.02391, 2020. https://arxiv.org/abs/2007.02391
  11. Tianyu Yang, Yang Yang, 'A stable non-iterative reconstruction algorithm for the acoustic inverse boundary value problem', Inverse Problems & Imaging 16(1), 2022. DOI: 10.3934/ipi.2021038, https://doi.org/10.3934/ipi.2021038
  12. Rong Sun, Ganghua Yuan, Yue Zhao, 'Stability for inverse potential scattering with attenuation', Journal of Differential Equations 431:113259, 2025. DOI: 10.1016/j.jde.2025.113259, https://doi.org/10.1016/j.jde.2025.113259
  13. K. V. Koshel, E. A. Ryzhov, V. N. Zyryanov, 'A modification of the invariant imbedding method for a singular boundary value problem', Communications in Nonlinear Science and Numerical Simulation 19(3):459–470, 2014. DOI: 10.1016/j.cnsns.2013.07.001, https://doi.org/10.1016/j.cnsns.2013.07.001