Exceptional points in non-Hermitian systems produce enhanced sensitivity: near the point where eigenvalues coalesce, a small perturbation produces a disproportionately large eigenvalue splitting. For an Nth-order exceptional point, the splitting scales as the Nth root of the perturbation. This sublinear amplification looks like a sensor advantage — measure the splitting, infer a perturbation smaller than any conventional method could detect.
The problem is noise. At the exceptional point, the eigenbasis collapses — the eigenvectors coalesce along with the eigenvalues — and this collapse amplifies noise as much as it amplifies signal. The signal-to-noise ratio does not improve. The enhanced sensitivity and the enhanced noise cancel, and the net measurement precision stays the same as a conventional system. The exceptional point gives back with one hand what it takes with the other.
Note (2026-11-11): checked against the paper's full text. The system (two yttrium iron garnet spheres coupled to a cavity mode), the third-order exceptional point of coherent perfect absorption, the separation of the absorption exceptional point from the resonance one and the avoidance of the noise divergence from eigenbasis collapse are as in the paper. The figures are mixed up. The paper reports two results: a twelve-fold enhancement of the signal-to-noise ratio for frequency-based magnetic-field sensing, and, by using the sensitivity of the minimum output intensity near coherent perfect absorption, a seventy-fold improvement of the signal-to-noise ratio with a 400-fold increase in responsivity, compared with a non-coherent-perfect-absorption system. The essay gives the twelve-fold figure and then 'a 400-fold increase in responsivity with a sevenfold SNR gain', which does not match either: the intensity-readout gain is seventy-fold, not sevenfold, and it belongs to a different readout from the twelve-fold result. The essay's statement that at a conventional exceptional point the sensitivity and noise gains cancel so that the signal-to-noise ratio does not improve is its simplification of the paper's statement that noise diverges because the eigenvectors are nonorthogonal.
Wang et al. (arXiv:2603.11604, 2026) separate the two effects. In a passive cavity magnonic system with two yttrium iron garnet spheres, they engineer a third-order exceptional point of coherent perfect absorption — a point where three eigenvalues coalesce in the absorption spectrum rather than the resonance spectrum. The absorption exceptional point provides the enhanced sensitivity. But because the resonance point is elsewhere, the eigenbasis collapse that amplifies noise is not at the same location. The noise divergence occurs at a different operating point than the sensitivity enhancement.
The result is a twelve-fold improvement in signal-to-noise ratio: the sensitivity advantage is real and the noise penalty is displaced. A 400-fold increase in responsivity with a sevenfold SNR gain.
The structural point: the sensitivity and the noise were never intrinsically coupled. They appeared coupled because previous systems placed them at the same exceptional point. Separating absorption from resonance separates the useful degeneracy from the destructive one. The limitation was not in the physics of exceptional points but in the engineering choice to collapse all degeneracies at the same location.