The Huxley-Zel'dovich front — a reaction-diffusion system with reversible reactions and discrete particles — propagates at an average speed set by the reaction kinetics and diffusion rate. In the limit of large particle number N, the front speed is deterministic. At finite N, shot noise from the discrete particles introduces fluctuations.
The expected scaling: both the systematic speed shift δc and the front diffusion coefficient D_f scale as 1/N. More particles, less noise, sharper front. This is confirmed by perturbation theory and Monte Carlo simulation.
The large deviations are the surprise. Rare events — extreme speed fluctuations far from the mean — are driven by optimal histories in which the front propagates at a speed very different from the average. The system can achieve this by coordinating particle births and deaths along the front, creating transient configurations that push the front faster or slower than its typical rate. In the most extreme case, the front travels in the wrong direction — retreating instead of advancing — and does so through a specific reaction pathway that the typical dynamics would never explore.
The optimal wrong-direction pathway is not the typical dynamics run backward. It's a distinct trajectory through configuration space, one that concentrates the available randomness into a coordinated push. The front's rare retreat is not disorder — it's organization, a fluctuation so improbable that the only way to achieve it is through a precise, atypical sequence of events. Large deviations reveal the system's hidden capacity for order within its noise.