A chaotic system amplifies uncertainty exponentially — the Lyapunov exponent quantifies how fast initially close trajectories diverge. Measuring the system periodically should reduce this uncertainty: each measurement collapses the cloud of possible states back toward a point. The question is whether measurement can keep up with chaos.
Gerbino, Giachetti, Le Doussal, and De Luca (arXiv:2501.00547) show this competition undergoes a sharp phase transition. Map the state estimation problem onto a directed polymer on a Cayley tree, and you find two phases: a chaotic phase where the Lyapunov exponent is reduced but positive (uncertainty still grows), and a strong-measurement phase where uncertainty remains bounded forever. There is a critical measurement rate separating them.
The structural surprise is that the measurement-induced phase transition coincides in location with the freezing transition of the directed polymer — but the critical properties differ. Same threshold, different universality. The point where measurements tame chaos and the point where the polymer freezes into a single optimal path happen at the same coupling strength, yet the fluctuations near each transition scale differently. The physical systems are mapped to each other by an exact correspondence, but the mapping distorts the critical exponents.
This matters beyond chaos theory because it demonstrates that equivalence of models need not imply equivalence of their phase transitions. Two descriptions of the same system can agree on when things change while disagreeing on how the change unfolds.