The study reports that its model predicts different resonance behavior for layered spheres depending on whether the outer shell is finite or nearly vanishing. In the parameter scans, finite-shell configurations occupied resonant regions that were non-resonant in the nearly vanishing-shell limit. The result is a map from a modeled acoustic system, with its reach set by the assumptions of the formulation.
A map built from acoustic boundaries
The work asks under which conditions different resonant regimes appear and how they relate to core–shell parameters. In plain terms, the model examines a sphere with an inner core and an outer layer, comparing each region’s acoustic impedance—a measure used to describe its response to sound—with the surrounding host. The calculations vary shell thickness and consider contrasts driven by sound speed, density, or a mixture of the two.
To calculate scattering, the authors impose continuity of acoustic pressure and normal particle velocity at both fluid interfaces. For each multipole order, they solve a 4 × 4 system. The core’s internal acoustic energy is computed through an analytical route that avoids the radial and angular derivative expansion and numerical integrations required in the standard formulation.
Resonance is assessed from internal acoustic-energy spectra. For the compact classification step, logistic regression uses α, β and δ to predict whether a resonance exists; a configuration is labeled resonant when P_res reaches at least 0.5, equivalent to a score g of at least 0.
Shell thickness changes the resonance landscape
One of the clearest patterns appears in the mixed and sound-speed-driven regimes. A region in which both the core and shell impedances exceed the host impedance remains strictly non-resonant across shell thicknesses. This is a result for the modeled regime and does not establish what would happen once losses or solid-mechanical effects are included.
The scans therefore suggest that shell geometry is part of the resonance condition in this model. A conclusion drawn from the nearly vanishing-shell limit does not cover every finite-shell configuration represented in the scans, because some finite-shell combinations fall inside resonant regions that disappear in that limit.
A classifier separates resonant from non-resonant cases
The classifier’s test results were strongest in the sound-speed-driven regime: accuracy was 99.04%, precision 99.64%, recall 98.08%, and ROC-AUC 0.9997. In the mixed regime, the corresponding figures were 97.57%, 96.95%, 96.36%, and 0.9982. In the density-driven regime, they were 95.37%, 90.96%, 78.80%, and 0.9837.
The lower recall in the density-driven regime is the main weakness in the comparison: recall measures the share of resonant cases that the classifier captures. The reported positive rate was roughly 16% in that regime, compared with roughly 36% to 42% in the other two. Confidence intervals were not reported, and the figures describe performance in the modeled test sets rather than a guarantee for physical systems outside the formulation.
For the positions of successive resonance peaks, the study uses a different relation: a recurrence rule whose coefficients are second-order polynomials in impedance asymmetry η and shell thickness δ. The fit used 6,800 anchored configurations in each of the sound-speed-driven and mixed regimes. Those sets contained 2,826 and 2,486 resonant curves, respectively.
On test data, the recurrence fit performed better in the sound-speed-driven regime, with R² = 0.9989, RMSE = 0.0094, MAE = 0.0034, and MAPE = 1.32%. In the mixed regime, the values were R² = 0.9812, RMSE = 0.0356, MAE = 0.0238, and MAPE = 6.51%.
That recurrence analysis did not include the density-driven regime. The authors excluded it because its resonant configurations were fewer and less regular, so the study reports only the logistic classifier for that regime.
Checks against published examples
The authors also compared the method with a reference configuration associated with Liu. The configuration was classified as resonant with P_res = 0.9999. The comparison uses an effective-fluid approximation for a system containing elastic solids, so it is not a full test of solid mechanics.
For a Marston fluid-sphere configuration, the classifier gave P_res = 0.9984, and the resolved peak was reported to match the literature value. This validation is a comparison with a cited literature result rather than a new experiment conducted in the study.
What the calculations leave out
The formulation is deliberately limited in physical scope. It omits viscosity, thermal conduction, deformation, viscoelasticity, thermal and mechanical expansion, and conversion of longitudinal waves into transverse waves. Those omissions mean the phase boundaries, classifier scores and recurrence fits apply to the idealized formulation and do not establish accuracy when those effects matter.
The Liu comparison carries an additional qualification because the solid–solid system was represented through an effective-fluid approximation. The results therefore provide a basis for screening modeled core–shell combinations, while leaving open how well the same relations would generalize to full elastic-solid calculations and experiments.
Publication note
The supplied document is an arXiv preprint, version 1, dated 20 Aug 2026. The authors report financial support from FAPESP and CNPq.
Paper data and sources
Original title: Acoustic Resonance Distribution for Core-Shell Scatterers
Authors: Naruna E. Rodrigues, Gilberto Nakamura, Odemir M. Bruno, Alexandre S. Martinez
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text