This paper studies how entropy is produced when a very rarefied gas—so rare that molecules hardly bump into each other—moves inside a container split into compartments. In that Knudsen regime, the motion of one molecule between collisions with the walls largely controls transport. The authors model the molecule as a “random billiard” that reflects or passes through semi‑permeable walls according to probabilistic scattering rules, and they use that model to quantify irreversibility and heat flow in steady state.

At a high level the main theoretical quantity is the stationary entropy production rate, which measures how different the usual forward motion of the particle is from its time‑reversed motion. The authors first express this rate as a relative entropy (an information‑theory notion of distance) between forward and backward path probabilities. Under a reciprocity condition — a time‑reversibility requirement tied to a surface Maxwellian distribution — this abstract definition simplifies to a familiar thermodynamic form: the mean energy delivered to the walls divided by the walls’ local temperatures. The paper calls this a stochastic Clausius relation.

To make computations practical the authors develop a modular, compartment‑by‑compartment analysis. Each open compartment is described by a scattering operator (which encodes how a particle leaves the compartment given how it entered) and by sojourn statistics such as expected time and expected number of collisions. The sequence of entrance states forms a Markov chain. Using the chain’s stationary distribution and a renewal‑reward theorem (a standard averaging method for repeated visits), the local, per‑compartment entropy contributions are assembled into a global entropy production rate.

The framework is illustrated with a family of explicit scattering models that generalize the classical Maxwell‑Smoluchowski law. These models depend on a few physical parameters: temperature T, a porosity p for wall gaps, a potential barrier height C for those gaps, and an accommodation coefficient α that gives the chance of full thermalization versus specular reflection. The authors work through examples of increasing complexity and give a central example of a cyclic three‑compartment device with two walls at different temperatures and a potential barrier. For the case of full thermal accommodation they obtain closed‑form formulas for the entropy production and for a net probability circulation around the cycle, a thermal‑ratchet effect analogous to thermal transpiration.