A mathematical preprint has tested whether the shape of a three-dimensional acoustic obstacle can be reconstructed from a small set of time-dependent sound-wave backscattering measurements. In synthetic numerical experiments, the method produced good qualitative reconstructions for a tilted peanut shape with 8 or 14 directions, while a sea-star shape was judged satisfactory only with 14.
The study is an arXiv version 1 preprint dated 20 August 2026. Its evidence combines mathematical analysis with numerical experiments on two modeled shapes, so the reported performance applies to the stated model and cases rather than establishing a general rule for arbitrary three-dimensional obstacles.
Turning returning waves into a shape
The task is an inverse-scattering problem: infer an obstacle’s boundary from the far-field pattern produced when time-dependent plane waves strike it. The paper focuses on whether a few such backscattering directions can provide enough information for reconstruction.
At the heart of the approach is a mathematical sensitivity calculation. The preprint shows that the far-field measurement is Fréchet differentiable with respect to changes in the obstacle boundary—a formal way of saying that the measurement has a well-defined first-order response to a small shape perturbation. It also characterizes the corresponding time-domain derivative through a transmission problem for the perturbed wave field.
That derivative is used in a regularized Gauß–Newton procedure, an iterative fitting method that updates the estimated shape while controlling unstable changes. The calculations combine convolution quadrature for the time-dependent part with a boundary element method for the obstacle surface. The reconstruction is limited to star-shaped boundaries and represents the boundary with spherical harmonics up to degree 5.
More directions helped, but the shapes differed
For the tilted peanut, reconstructions using 8 and 14 backscattering directions were judged qualitatively good, while 6 directions appeared insufficient. The final relative residuals—the remaining mismatch between calculated and target data—were 0.17 with 6 directions and 0.01 with both 8 and 14 directions.
The sea-star test was more demanding. Only the reconstruction using 14 directions was judged satisfactory; the 6- and 8-direction cases were judged insufficient. Their final relative residuals were 0.02, 0.05 and 0.02, respectively, for 6, 8 and 14 directions.
The authors also report an analytical time-discretization error rate proportional to τ^{2m−1}, under the stated regularity assumptions. The numerical examples used a final observation time of 20 and 300 equally spaced time steps.
A limited test of noisy data
The 14-direction sea-star experiment was also run with added noise. At 5% noise, the final relative residual was 0.06 after 11 Gauß–Newton steps; at 15% noise, it was 0.16 after 9 steps; and at 30% noise, it was 0.29 after 9 steps. The paper describes the reconstructions at 15% and 30% noise as satisfactory.
These figures are examples rather than a formal statistical robustness estimate: the analysis reports no repeated-noise summaries or confidence intervals. With only two modeled shapes and selected direction sets and noise levels, the study does not establish how the method will perform across broader obstacle classes or physical measurement conditions.
Taken together, the preprint supports the reconstruction method for the stated acoustic model and the two synthetic examples. It leaves open how the approach will perform for broader geometries, heterogeneous materials, measurement-model errors, and comparisons with alternative algorithms.
The research was funded by the Deutsche Forschungsgemeinschaft through Project-ID 258734477 and SFB 1173. One author also received Research Council of Finland support through grant 359182.
Paper data and sources
Original title: Domain derivative and shape reconstruction for an inverse backscattering problem for the wave equation
Authors: Roland Griesmaier, Marvin Knöller, Eliane Kummer
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text