Hilbert's sixth problem, posed in 1900, asked for an axiomatic treatment of physics — starting from the most fundamental level and deriving everything above it through rigorous mathematics. The hardest piece: deriving the equations of fluid mechanics from Newton's laws of particle motion. Water is made of molecules. Molecules obey Newton's laws. The Navier-Stokes equations describe how water flows. The gap between these descriptions has been open for 125 years.
In 2025, Yu Deng, Zaher Hani, and Xiao Ma closed it. Their proof requires two limits taken in sequence, not one.
The first limit takes the number of particles to infinity while shrinking their size to zero. This produces the Boltzmann equation — a statistical description of how populations of particles distribute themselves in velocity space, colliding and scattering. The individual particle disappears. What remains is a probability distribution.
The second limit takes the collision rate to infinity — equivalently, the mean free path to zero. Particles collide so frequently that the velocity distribution relaxes to local equilibrium everywhere, and what emerges are the macroscopic equations: compressible Euler for inviscid flow, incompressible Navier-Stokes-Fourier for viscous flow with heat conduction.
The critical structural point: you cannot take both limits simultaneously. The derivation fails if you try to go directly from particles to fluids. The Boltzmann equation is not merely a convenient waystation — it is mathematically necessary. The information destruction that produces fluid behavior happens in two distinct stages, and collapsing them into one produces nothing rigorous.
This means the kinetic scale — the scale at which you track distributions rather than individuals, but haven't yet reached continuous flow — is not an approximation to be refined away. It is the mandatory bridge. The macroscopic description is not a simplification of the microscopic one. It is a simplification of the intermediate one, which is itself a simplification of the microscopic one. Two lossy compressions in sequence, each discarding different information, neither replaceable by the other.