Derivation of the Boltzmann equation from hard-sphere dynamics (after Y. Deng, Z. Hani, and X. Ma)

Consider a microscopic system of $N$ hard spheres that are initially independent (modulo the exclusion condition on particle positions) and identically distributed in $\mathbb{R}^3$. When the number $N$ of particles goes to infinity and the diameter $\varepsilon$ of the particles goes to zero, and under the weak density assumption $N\varepsilon^2=1$, it has been known since the work of Lanford that the empirical measure for the particles converges to the solution of the Boltzmann equation in a short time interval. In particular, the particles remain dynamically independent, in this limit and in the short time interval where the correlations induced by their collisions remain under control. In a recent work, Y. Deng, Z. Hani and X. Ma successfully obtained the same convergence result in arbitrary large time; more precisely, the convergence result holds for any time interval on which the Boltzmann equation has a regular solution. In this note, we explain a few elements of their proof.

💡 Research Summary

The paper under review provides a concise exposition of the recent breakthrough by Y. Deng, Z. Hani, and X. Ma (2024) concerning the derivation of the Boltzmann equation from a deterministic hard‑sphere system for arbitrarily long times, as long as the Boltzmann equation admits a smooth solution on the interval considered. The classical result, due to Lanford (1975), establishes that in the low‑density Boltzmann–Grad scaling (N particles, diameter ε, with Nε² = 1) the empirical distribution of the particles converges to the solution of the Boltzmann equation, but only for a short time interval of order a small fraction of the mean free time. The limitation stems from the fact that Lanford’s proof relies on a perturbative expansion of the BBGKY hierarchy into a series of “collision trees”. This series converges only while the number of collisions per particle remains small; beyond that, the combinatorial growth of terms overwhelms the smallness provided by the scaling.

Deng, Hani, and Ma overcome this obstacle by introducing three intertwined innovations: (1) a cluster expansion that groups together simultaneous multi‑particle collisions into single graphical objects called clusters; (2) a weighted estimate that assigns a carefully chosen weight to each particle pair based on their spatial separation and relative velocity, thereby controlling the geometric constraints imposed by the hard‑sphere exclusion; and (3) a recursive collision‑tree structure that records how clusters themselves interact over time. The cluster expansion replaces the simple binary‑collision trees of Lanford with a hierarchy of graphs that can accommodate arbitrarily many particles colliding in a short time window. The weighted estimate ensures that each additional collision contributes a factor that is uniformly bounded (or even decays) under the Boltzmann‑Grad scaling, preventing the exponential blow‑up of the series. The recursive tree then allows one to iterate the cluster expansion over successive time intervals, keeping track of the “memory” of past collisions while still preserving a uniform bound on the total contribution.

Technically, the authors start from the Liouville equation for the N‑particle distribution and write the BBGKY hierarchy. They then perform a Duhamel expansion, but instead of stopping after a fixed number of iterates, they reorganize the terms into clusters. Each cluster is associated with a “collision graph” that encodes which particles interact and when. By integrating over the admissible configurations (respecting the hard‑sphere exclusion) and using the weighted bound, they prove that the contribution of clusters with k collisions decays like Cⁿ/k! for some constant C independent of N and ε. This factorial decay is sufficient to sum the series for any finite time T, provided the solution of the Boltzmann equation remains smooth on