Kinetic theory and the Boltzmann equation

Let $f(\mathbf x,\mathbf p,t)$ be the number density in one-particle phase space, normalized so that integrating over position and momentum gives the particle number. For nonrelativistic particles, streaming transports this distribution with velocity $\mathbf p/m$, while an external force changes momentum. The collision term accounts for particles entering and leaving a given momentum region through interactions. The equation evolves an ensemble density; a feature moving in its plot is not the trajectory of one labeled molecule.

A Boltzmann collision integral for a classical dilute gas assumes short, predominantly binary collisions and negligible pre-collision correlations of the relevant incoming particles. This molecular-chaos closure is not the statement that every pair correlation vanishes forever: collisions generate correlations, which the reduced model does not retain in full. Dense fluids, long collision durations, strong long-range collective forces, and quantum-degenerate occupations require modified kinetic descriptions. Boundaries and external forcing must be supplied as part of a concrete initial-boundary-value problem.

Elastic binary collisions conserve particle count, total momentum, and total kinetic energy. Accordingly, integrating the collision term against $1$, $\mathbf p$, or $\mathbf p^2/(2m)$ gives zero, assuming the collision kernel respects those conservation laws. Other moments can relax: an anisotropic momentum distribution can become more isotropic without changing its total kinetic energy. The local Maxwell distribution makes the gain and loss terms balance when its parameters are compatible with local collision equilibrium.

The classical H-theorem states that a suitably integrated $H=\int f\ln(f/f_*)$ does not increase for the closed Boltzmann description, with a fixed reference density making the logarithm dimensionless. Boundary entropy flux and forcing must be included when the system is open. This reduced irreversible result does not prove that the exact reversible many-particle dynamics raises entropy on every microscopic history. The closure and special assumptions about incoming correlations select the direction in which the kinetic relaxation argument applies.

For a dilute identical hard-sphere gas, an equilibrium mean-free-path estimate is $\ell=1/(\sqrt2 n\sigma)$, where $\sigma$ is the total collision cross-section. Comparing that microscopic length with a macroscopic variation scale $L$ defines the Knudsen number. When $\ell/L\ll1$ and observation times are long compared with collision times, a near-local-equilibrium fluid expansion may be useful. Small Knudsen number alone does not guarantee weak gradients everywhere, especially near boundaries or sharp shocks.

When the mean free path is comparable to the apparatus, particles can travel ballistically between walls and the local Navier–Stokes approximation may fail even if the dilute Boltzmann equation remains appropriate. Taking moments produces conservation equations, but those equations contain stress and heat flux that require closure. Hydrodynamics therefore does not follow merely by integrating once and renaming the moments. The hierarchy between collision length, gradient scale, collision time, and driving time determines which reduced model is defensible.