A random walk in a random environment navigates a landscape of biases that doesn't change. Each site has a fixed drift — slightly left here, slightly right there — and the walker encounters these biases in the order the landscape presents them. The disorder is quenched: frozen in place before the walk begins. This is hard to analyze because the walker's trajectory depends on the specific realization of the landscape, not just its statistics.
At the other extreme, if the landscape is reshuffled at every step — the disorder is annealed — the walk sees a fresh environment at each site. The specific biases don't matter. Only the average matters. The problem reduces to a walk with a known mean drift.
Avena and da Costa (arXiv:2603.22652) study the middle ground: the landscape is reshuffled, but not at every step. The environment stays frozen for some duration, then is replaced with a new random realization, then freezes again. The schedule of reshuffling — the cooling schedule — determines everything.
If reshuffling is frequent (fast cooling), the walker effectively sees many independent environments and behaves like the annealed case. If reshuffling slows down (the environment persists longer between changes), the walker begins to feel the specific biases of each realization. In the limit where reshuffling stops entirely, you recover the quenched case.
The authors map out how the classical results — recurrence, law of large numbers, large deviations — transform as the cooling schedule varies. The walk is “quasi-ergodic”: ergodic over the sequence of environments but not within any single one. The cooling rate determines which limit theorems apply, and the transitions between regimes are sharp, not gradual.
The structural lesson: disorder isn't binary. The same landscape, experienced at different rates of change, produces qualitatively different dynamics. What looks like frozen noise to a slow observer looks like thermal averaging to a fast one. The observer's timescale relative to the environment's determines which theory governs.