A directional flow in the model
A mathematical model of a dilute gas reports that more molecules cross from a warmer cell into a cooler neighboring cell during the modeled free-streaming stage. The same calculation gives a non-zero net momentum flux in that direction. The paper treats the resulting narrowing of the temperature gap as a microscopic picture of heat moving from hot to cold, and identifies it with the direction described by Clausius's Second Law. The finding is conditional on the model's assumptions; it is not a measurement of heat flow.
The setup is deliberately spare. It imagines two adjacent cells with the same molecule count but different volumes, temperatures and densities, while assuming uniformity in the other spatial directions. The analysis begins with Boltzmann-Grad scaling, a dilute-gas framework in which molecule count and molecular size are kept in a fixed relationship as the model considers more and more molecules. It then splits each local update into two stages: free streaming, as molecules cross the interface, followed by collisions that move the combined distribution toward a new Gaussian, or bell-shaped, equilibrium.
What crosses the boundary
During free streaming, the paper analyzes one-dimensional velocity components using a Gaussian distribution. With the cooler cell next to the warmer one, its crossing calculation finds an imbalance favoring warmer-to-cooler crossings. When the momentum carried by those crossings is counted, the hotter-origin molecules contribute greater average momentum, leaving a modeled net momentum flux toward the cooler cell. Equal cell properties serve as the equilibrium baseline. The paper reports no numerical size or uncertainty interval for either flux.
That first stage does not end the modeled process. The merged velocity distribution is then allowed to relax through collisions. To estimate how many collisions are needed, the paper invokes the Central Limit Theorem, which describes how repeated contributions can approach a bell-shaped average. It states that the required count grows roughly as the inverse square of the chosen tolerance for closeness to Gaussianity. For modest temperature contrasts, the generic scale is on the order of one to ten collisions, but the paper gives no fitted or simulated value.
The timing estimate is similarly approximate. The free-streaming interval is put at about one mean collision time, meaning the average interval between collisions. Including the adjustment stage, the full modeled episode is estimated at about one plus the required collision count times that interval. The framework therefore supplies a microscopic timescale, but it is an order-of-magnitude estimate rather than a sharp bound.
A proposed answer to an old paradox
The broader claim concerns Loschmidt's paradox: if microscopic mechanics can be run backward, why should the macroscopic world favor one direction of time? The paper's answer is that the relaxation step uses coarse-graining, a statistical simplification that discards microscopic information. Reversing the recorded, simplified description would therefore require recovering information that the description no longer retains; it is not simply the ordinary reversal of every microscopic trajectory. The paper presents this as a conceptual resolution within its framework, not as an independent test.
The paper extends the argument into information theory by invoking Landauer's principle. It states that erasing one bit at a given temperature dissipates at least the Boltzmann constant multiplied by that temperature and the natural logarithm of 2, then applies that minimum to information needed for a reversal. This is a theoretical accounting inside the proposed picture. The paper does not report a system-specific measurement of the cost.
A model with clear boundaries
The paper acknowledges that its collision model is idealized. The linear-mixing ansatz represents an outgoing velocity component as a combination of two incoming components, but it omits the exact three-dimensional dependence on relative speed and scattering direction. The collision count and adjustment time are order-of-magnitude or mean-field estimates rather than sharp convergence bounds. The wider setup remains the stated adjacent-cell, dilute-gas construction, with local quasi-equilibrium and a one-dimensional Gaussian velocity analysis.
The paper closes by suggesting a connection between its encoded irreversibility and the existence of smooth solutions to the Navier-Stokes equations. That implication is explicitly speculative and is not established mathematically.
Paper data and sources
Original title: The Hidden Second Law of Thermodynamics behind the Boltzmann-Grad Limit
Authors: Zhaohua Wu
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-25
DOI: Not available
Original paper · Full text