Identifying the natural frequencies of a vibrating structure from measured data requires fitting a mathematical model to the frequency response. The model is a ratio of polynomials. Higher-order polynomials fit the data better but introduce spurious poles — mathematical artifacts that look like resonances but correspond to nothing physical. Distinguishing real modes from artifacts is the central problem of operational modal analysis.
The conventional approach builds a stability diagram: fit models of increasing order and plot which poles persist. Stable poles appear across many model orders; spurious poles appear and vanish erratically. The method works but produces cluttered diagrams that require expert interpretation.
Note (2026-11-12; read against the paper's full text): the paper's claim is that spurious roots can be eliminated from the stability diagrams without compromising accuracy for the cases considered (a numerical plate model, plates with low and high damping, and an electric-vehicle motor stator); it says the conventional method produced stable spurious poles, especially at high frequency, which made it hard to tell them from physical modes. It does not report that the sparse representation separated physical modes that were nearly indistinguishable in the full polynomial model: for the stator it notes larger differences in the mode shapes than in the first plate case, which it attributes to noise-induced perturbations of the polynomial coefficients and to the many closely spaced modes. The essay's closing remarks (controlled hallucination followed by selective destruction) are its own.
Strategic destabilization offers a different approach (arXiv:2603.20614). Orthogonal matching pursuit sparsifies the polynomial coefficients, removing terms that contribute only to spurious poles. The surviving coefficients produce clean stability diagrams with the real modes clearly visible and the artifacts eliminated.
The method was validated on an electric vehicle motor stator core — a complex vibrating structure with closely spaced modes. The sparse representation separated physical modes that were nearly indistinguishable in the full polynomial model.
The structural insight: finding what's real requires systematically generating and then eliminating what's false. The full polynomial model must first contain all the spurious poles so that the sparsification step can identify and remove them. You cannot find the true modes by restricting the model — the model must be rich enough to contain both real and spurious solutions, then pruned. Discovery proceeds by controlled hallucination followed by selective destruction. The real modes are what survives the culling, not what the model produces first.