The Proof That Changed Math but Not Your Simulation

In 2025, Deng Yu and his collaborators released a lengthy proof on the preprint platform, deriving the incompressible Navier–Stokes equations rigorously from hard‑sphere gas dynamics. This work completed the Boltzmann‑to‑continuum derivation that had been a long‑standing problem in mathematical physics. It drew interest among specialists in partial differential equations and statistical mechanics.

At the same time, neither the Ansys Fluent user forums nor the technical discussion groups of CFD engineers have shown any mention of this proof. Daily conversations remain focused on: the setting of interface compression coefficients in multiphase models, the convergence of pressure‑velocity coupling on unstructured grids, and the influence of wall Y⁺ values on drag coefficients. This gap is not a matter of differing attitudes but reflects the distinct objects of study in mathematics and engineering.

I. A qualitative proof does not modify quantitative numerical schemes

The core solvers of commercial software such as Fluent and Star‑CCM+ use the finite volume method (FVM). They integrate the N‑S equations over control volumes, discretise the partial differential terms into algebraic equation systems, and solve them iteratively (e.g., SIMPLE, PISO algorithms) to obtain numerical solutions. The accuracy of this process is governed by two error sources: discretisation errors (depending on mesh density and interpolation order) and modelling errors (depending on the turbulence model's closure assumptions for unresolved scales).

Deng's proof belongs to the qualitative theory of partial differential equations – it establishes existence, uniqueness, and the limiting convergence from microscopic statistical mechanics to macroscopic equations. It does not provide a new numerical flux scheme, modify the under‑relaxation factors in pressure‑correction equations, or alter the logarithmic law coefficients in wall functions. Consequently, no part of the solver kernel requires an update because of this proof. Commercial vendors like ANSYS and Siemens will not add algorithmic iterations based on this mathematical result.

The continuum assumption (Kn < 0.01) has been validated across decades of wind‑tunnel tests and field measurements in aerospace, energy, and chemical engineering. The mathematical proof elevates that empirical validation to logical necessity, but it does not change any operational step engineers take when setting boundary conditions in Workbench.

II. Turbulence modelling and rarefied gases: two unaffected branches

The core difficulty in turbulence models (RANS, LES) is the closure problem – the Reynolds stress tensor must be modelled in terms of mean or filtered quantities. Existing approaches include the Boussinesq eddy‑viscosity approximation (k‑ε family), second‑moment closure (RSM), and machine‑learning‑assisted models initialised from analytical bases.

Deng's proof assumes a key condition: the molecular velocity distribution function is a small perturbation about the local Maxwellian, i.e., the fluid is in a thermodynamic near‑equilibrium state. In wall‑bounded turbulence, however, fluctuating velocity amplitudes of coherent structures typically reach 15‑20% of the mean flow, and energy transfers across scales spanning several orders of magnitude in the inertial subrange. Such flow regimes do not satisfy the near‑equilibrium small‑perturbation condition. Therefore, the proof provides no mathematical constraint on any Reynolds‑stress modelling scheme. The next generation of turbulence models will still rely on feature extraction from DNS databases and training of physics‑constrained neural networks – entirely unrelated to the hard‑sphere collision integral.

In the rarefied‑gas domain, the existing engineering tool DSMC (Direct Simulation Monte Carlo) is built on statistical sampling of hard‑sphere collisions. Deng's proof confirms at an academic level that as the number of simulated molecules tends to infinity and the time step tends to zero, DSMC's statistical average converges to the N‑S solution. In engineering practice, however, the particle count is limited by computational resources (typically 10⁷–10⁸), leaving a gap of more than ten orders of magnitude from the Avogadro number. This convergence proof does not change the time‑step selection criteria for DSMC or reduce its statistical noise.

The SPH (Smoothed Particle Hydrodynamics) method, used in film special effects, blast simulations, and dam‑break modelling, describes particle interactions via kernel interpolations. Its force terms come from artificially introduced equations of state and artificial viscosity coefficients, which are formally inconsistent with the binary hard‑sphere collision term in the Boltzmann collision integral. Deng's work has no connection with SPH algorithms.

III. The specific items in the engineer’s manual that remain unchanged

For practitioners in engineering simulation, the scope of influence of this proof can be precisely itemised:

• Geometry cleaning: unchanged. Defective surfaces and fine features imported from CAD still need repair.

• Meshing strategy: unchanged. The number of prism layers, growth ratios, and target Y⁺ values are still dictated by the turbulence model.

• Solver settings: unchanged. The CFL limit, pressure‑velocity coupling scheme, and discretisation order of the momentum equation remain unaffected.

• Convergence criteria: unchanged. The standards of reducing residuals by three orders of magnitude or monitoring physical quantities until steady are still applied.

• Post‑processing verification: unchanged. The quantitative accuracy of lift and drag coefficients, wall shear stress, and the separation point location is still assessed through mesh‑independence studies.

The only new citation scenario is this: when the Knudsen number approaches the transitional range of 0.01‑0.1, and the multidisciplinary team questions the continuum assumption, engineers may cite this proof as a mathematical argument that the N‑S equations still possess microscopic convergence in that range. However, this citation provides no operational guidance on controlling discretisation errors.

IV. A verifiable observational coordinate

Mathematicians produce proofs in the abstract spaces of functional analysis and ergodic theory; their products are existence theorems and convergence thresholds. Engineers produce simulations on discrete grids in Euclidean space; their products are numerical matrices of pressure and velocity fields. When both describe the same physical fact – for example, the wall friction coefficient of a turbulent boundary layer over a flat plate – the mathematical proof gives the logical assertion that “this coefficient exists and is monotonically bounded with Reynolds number,” while the numerical iteration gives the engineering estimate that “this coefficient is 0.00325 on a particular mesh.”

Both share experimental databases as calibration benchmarks, but their methodologies do not intersect. Deng's work verifies the logical self‑consistency of the N‑S equations at the microscopic level – a verification that does not affect the CFD software codebase. This is not conservatism or underestimation; it is the natural boundary of disciplinary division.