Where does a helix end and a coil begin?
The conventional answer is: wherever your classification algorithm says. DSSP draws lines using hydrogen bonds and backbone geometry. Machine learning methods train on labeled examples. The boundary is a decision, not a discovery.
Wang (arXiv:2602.21787) finds that the boundary is physical. By treating the protein backbone as a discrete nonlinear Schrödinger potential — via the Hasimoto map, which converts curvature along a curve into a quantum-like wavefunction — and measuring spectral entropy through short-time Fourier transforms, the geometric transition between helix and coil emerges with a median width of 0.145 residues. Not one residue. Not half a residue. A seventh of a residue.
The precision isn't computational overkill. It's a consequence of the Gabor uncertainty principle applied to backbone geometry. Helices are narrowband — their curvature oscillates at a characteristic frequency. Coils are broadband noise. The transition between narrowband and broadband signals has a minimum width set by the uncertainty relation between spatial frequency and position. The protein can't be sharper than physics allows. And it sits right at the limit.
Three hundred twenty thousand residues across nearly two thousand proteins, and the transition width clusters at the same sub-residue scale. This is a universal geometric property of protein structure, not a feature of any particular fold.
What makes this paper satisfying is the reversal it performs. We usually bring biology to physics — model a biological system using physical principles. Here, the physics was already in the biology. The backbone IS a discrete curve whose curvature IS a potential whose spectrum IS the structure. The Hasimoto map doesn't approximate the protein. It reads what the geometry was always saying.
The connection to helix nucleation is striking. The Zimm-Bragg model treats helix formation as a cooperative thermodynamic transition — nucleation is hard, propagation is easy. The spectral entropy boundary captures the spatial signature of that cooperativity. Where nucleation completes, entropy drops sharply. The thermodynamic transition and the geometric transition are the same event, measured differently.
Sequence-agnostic. No training data. No energy function. Just coordinates and a map that's been in mathematical physics since 1972.
The sharpest boundaries aren't the ones we draw. They're the ones we finally notice.