This paper studies what happens when a tiny random change is added to the usual elastic reflection rule in billiards. A billiard here means a particle that moves in straight lines inside a fixed shape and bounces off the boundary. The authors show that, depending on whether the original (deterministic) billiard is chaotic or integrable, the long‑time statistical behaviour falls into two different universality classes.

The perturbation they introduce is simple: instead of reflecting exactly as in the elastic rule, the outgoing angle is randomized by a small amount. For very small noise the system becomes ergodic (everywhere accessible in the long run), but the invariant or steady distributions differ. For chaotic billiards the steady state is very close to what arises from the classical Knudsen model of diffuse scattering, where outgoing angles follow a cosine law (probability proportional to cos(angle)). For integrable tables — like circles and ellipses — the steady state instead resembles the Evans model, where outgoing angles are uniformly random. That uniform angular law produces a spatial stationary measure that is typically not uniform along the boundary.

The authors support these claims with analytic arguments and numerical experiments. They study examples such as a dispersing “diamond” billiard and variants of a lemon billiard to show the chaotic/ergodic case returns a Knudsen‑like angular distribution and an almost uniform spatial density. For elliptic (integrable) tables they compare numerics to the known Evans solution and find very close agreement: angles are nearly uniformly distributed and the particle density along the boundary develops characteristic non‑uniform peaks. The paper reports large simulations (figures are based on about 10^9 collisions) and also derives an explicit invariant spatial density for the Evans ellipse that matches the numerics.