Classification of limiting dynamics in the Boltzmann-Grad limit of hard-sphere gases from the microscopic correlation structure of the initial data

For a dilute gas of hard spheres in the Boltzmann-Grad limit (particle diameter tending to zero with the number of particles tending to infinity at fixed mean free path), Lanford's theorem derives the Boltzmann equation from Newtonian dynamics, but only for initial data whose n-particle marginals factorize (chaotic data) and on a short time interval. Bodineau, Gallagher, Saint-Raymond and Simonella (2018) showed that convergence of the marginals can be lost on singular 'bad sets' and constructed an explicit grand-canonical initial state whose Boltzmann-Grad limit is free transport rather than the Boltzmann equation, proving that the limiting dynamics is not determined by the one-particle density alone. The open problem is the classification question they pose: characterize quantitatively which microscopic correlation structures of the initial n-particle distribution lead to which limiting dynamics (the Boltzmann equation, Enskog-type hierarchies retaining correlations, or free transport), and describe the correlations that persist at positive times and encode the memory of the Hamiltonian dynamics. A sibling issue raised by the same paper — extending Lanford's short validity time — was resolved for chaotic data by Deng, Hani and Ma (2024), who proved validity for arbitrarily long times, so the classification of correlated data is the surviving open core.

Background

A gas of $N$ identical hard spheres of diameter $\varepsilon$ on a periodic box evolves by elastic binary collisions, a deterministic, time-reversible Hamiltonian flow. In the Boltzmann-Grad scaling $N\varepsilon^{d-1}=O(1)$ (fixed inverse mean free path, in dimension $d\geq 2$), the natural macroscopic description is the Boltzmann equation, a nonlinear kinetic equation for the one-particle density $f(t,x,v)$ whose collision term is quadratic and involves only pre-collisional velocities. Lanford's theorem (1975) rigorously derives the Boltzmann equation in this limit, but only on a time interval $[0,t^{*}]$, a small fraction of the mean free time, and for chaotic initial data: the n-particle marginals $f_N^{(n)}$ must converge to products $f_0^{\otimes n}$, an assumption expressing the statistical independence of any bounded group of particles. The proof passes through the BBGKY hierarchy, the chain of equations for all marginals coupled by the next marginal, whose formal limit is the Boltzmann hierarchy; chaos decouples this hierarchy into copies of the Boltzmann equation. Two difficulties surfaced in the refinement by Bodineau, Gallagher, Saint-Raymond and Simonella. First, the convergence of $f_N^{(n)}(t)$ to $f(t)^{\otimes n}$ provably fails on certain singular sets $\mathcal{B}_{\varepsilon_0}^{n+}$ ('bad sets') of configurations whose backward free flow exhibits recollisions within a time $T$; convergence holds only outside them, a one-sided statement, and the choice of the bad sets encodes the arrow of time — this is where irreversibility enters the limit. Second, chaos is not preserved as an assumption of the same strength at positive times, because the dynamics creates correlations on these sets. The same authors constructed an explicit counterexample: a grand-canonical ensemble (a probability distribution over the number of particles, with rescaled correlation functions describing each fixed-n sector) obtained by modifying a local-equilibrium state precisely on the forward bad sets; in the Boltzmann-Grad limit its correlation functions follow free transport (uncorrelated straight-line motion) rather than the Boltzmann equation on a small time interval, showing that non-chaotic data can change the nature of the dissipative limit. The canonical ensemble (exactly $N$ particles) version of this construction requires cluster-expansion techniques at fixed particle number and is not available. Adjacent descriptions of correlated limits exist: the Enskog hierarchy (the version of the Boltzmann equation in which the collision term retains the exact geometric collision-impact parameter, i.e. the finite particle diameter) admits rigorous solutions describing the exact empirical measure of individual finite hard-sphere configurations ('microscopic solutions'); the Kyiv school (Gerasimenko, Gapyak and coauthors) derives generalized non-Markovian Enskog equations — equations whose collision term retains memory of the particles' past trajectories — as the Boltzmann-Grad asymptotics of the evolution of observables for general data; and global-in-time well-posedness of the Boltzmann hierarchy itself has recently been established for polynomially weighted $L^{\infty}$ data. On the time restriction, Deng, Hani and Ma (2024) proved, for factorized grand-canonical data with bounded $\nabla_x f_0$ and a bounded Boltzmann solution, that the marginals converge to products uniformly up to any finite time $t_{\mathrm{fin}}$ — an arbitrarily long validity — which removes the short-time limitation of Lanford's theorem for chaotic data. What no result addresses is the dependence of the limit dynamics on the correlation structure of the initial state: between the two extreme cases now rigorous (chaotic data giving the Boltzmann equation, and bad-set-correlated data giving free transport) there is no systematic picture.

Problem Statement

For the hard-sphere gas in the Boltzmann-Grad limit, establish a systematic classification of the possible limiting dynamics in terms of the microscopic correlation structure of the initial n-particle distribution. Concretely: (a) identify quantitative properties of the initial marginals $f_{N,0}^{(n)}$ — expressed through their correlation (cumulant) structure, including modifications concentrated on the singular sets of recolliding configurations — that determine the low-density limit, and characterize classes of initial states for which the limit is governed, respectively, by (i) the Boltzmann equation (decoupled hierarchy), (ii) an Enskog-type or otherwise correlated hierarchy dynamics, or (iii) free transport; the classification must place the two known boundary cases — chaotic data yielding the Boltzmann equation (Lanford; Deng-Hani-Ma for long times) and the bad-set-modified grand-canonical data of Bodineau-Gallagher-Saint-Raymond-Simonella (2018) yielding free transport — inside a single coherent framework, and must decide two independent questions about the list (i)-(iii), each resolved either way — by exhibiting the relevant class of initial data or by proving that no such class exists: whether the correlated class (ii) is nonempty (for example, data whose limit retains dynamical correlations described by an Enskog-type equation) and whether the list (i)-(iii) is complete (no class of initial data has a limit governed by none of (i)-(iii)). (b) As an integral part of the same objective, describe the structure of the correlations of the marginals $f_N^{(n)}(t)$ at positive times — on and off the bad sets — and how they encode the memory of the Hamiltonian dynamics that the Boltzmann description discards. The classification comprises the following subgoals: (a-1) a quantitative criterion on the initial cumulant structure that selects the low-density limit; (a-2) the characterization of each of the classes (i)-(iii) that is nonempty, with both known boundary cases placed in one framework under that criterion; (a-3) a decision, resolved either way, on whether the correlated class (ii) is nonempty and on whether the list (i)-(iii) is complete; and (b) the positive-time correlation structure of the marginals. Part (b) is posed jointly with part (a) and does not stand alone.

References

1. T. Bodineau, I. Gallagher, L. Saint-Raymond, S. Simonella, 'One-sided convergence in the Boltzmann-Grad limit', Annales de la Faculté des sciences de Toulouse: Mathématiques (6) 27(5) (2018) 985-1022, DOI 10.5802/afst.1589, https://doi.org/10.5802/afst.1589; preprint arXiv:1612.03722, https://arxiv.org/abs/1612.03722
2. Y. Deng, Z. Hani, X. Ma, 'Long time derivation of the Boltzmann equation from hard sphere dynamics', preprint arXiv:2408.07818 (2024), https://arxiv.org/abs/2408.07818
3. Y. Deng, Z. Hani, X. Ma, 'Hilbert's sixth problem: derivation of fluid equations via Boltzmann's kinetic theory', preprint arXiv:2503.01800 (2025), https://arxiv.org/abs/2503.01800
4. T. Bodineau, I. Gallagher, L. Saint-Raymond, 'The Brownian motion as the limit of a deterministic system of hard-spheres', Inventiones mathematicae 203(2) (2016) 493-553, DOI 10.1007/s00222-015-0593-9, https://doi.org/10.1007/s00222-015-0593-9
5. T. Bodineau, I. Gallagher, L. Saint-Raymond, 'From hard sphere dynamics to the Stokes-Fourier equations: an L2 analysis of the Boltzmann-Grad limit', Annals of PDE 3(1) (2017), Art. 2, DOI 10.1007/s40818-016-0018-0, https://doi.org/10.1007/s40818-016-0018-0
6. R. Denlinger, 'Observables and Strong One-Sided Chaos in the Boltzmann-Grad Limit', preprint arXiv:1706.09961 (2017), https://arxiv.org/abs/1706.09961
7. R. Denlinger, 'Structure of correlations for the Boltzmann-Grad limit of hard spheres', Journal of Mathematical Physics 58(12) (2017) 121501, DOI 10.1063/1.4998194, https://doi.org/10.1063/1.4998194; preprint arXiv:1707.02274, https://arxiv.org/abs/1707.02274
8. M. Pulvirenti, S. Simonella, A. Trushechkin, 'Microscopic solutions of the Boltzmann-Enskog equation in the series representation', Kinetic & Related Models 11(4) (2018) 911-931, DOI 10.3934/krm.2018036, https://doi.org/10.3934/krm.2018036; preprint arXiv:1802.05926, https://arxiv.org/abs/1802.05926
9. I. Ampatzoglou, J. K. Miller, N. Pavlović, M. Tasković, 'On the global in time existence and uniqueness of solutions to the Boltzmann hierarchy', Journal of Functional Analysis 289 (2025) 111079, DOI 10.1016/j.jfa.2025.111079, https://doi.org/10.1016/j.jfa.2025.111079; preprint arXiv:2402.00765, https://arxiv.org/abs/2402.00765
10. A. Di Stefano, S. Simonella, R. Winter, 'Pointwise decay of truncated correlations in chaotic states at low density', preprint arXiv:2012.01246 (2020), https://arxiv.org/abs/2012.01246
11. V. I. Gerasimenko, I. V. Gapyak, 'The Boltzmann-Grad asymptotic behavior of collisional dynamics', Reviews in Mathematical Physics 33(2) (2021) 2130001, DOI 10.1142/S0129055X21300016, https://doi.org/10.1142/S0129055X21300016; preprint arXiv:2001.06786 (with subtitle 'a brief survey'), https://arxiv.org/abs/2001.06786
12. T. Bodineau, I. Gallagher, L. Saint-Raymond, S. Simonella, 'Long-time derivation at equilibrium of the fluctuating Boltzmann equation', The Annals of Probability 52(1) (2024) 217-295, DOI 10.1214/23-AOP1656, https://doi.org/10.1214/23-AOP1656; preprint arXiv:2201.04514, https://arxiv.org/abs/2201.04514
13. M. Pulvirenti, S. Simonella, 'On the cardinality of collisional clusters for hard spheres at low density', Discrete and Continuous Dynamical Systems 41(8) (2021) 3903-3914, DOI 10.3934/dcds.2021021, https://doi.org/10.3934/dcds.2021021; preprint arXiv:2005.09962, https://arxiv.org/abs/2005.09962