it takes navier-stokes to move the world

The Navier-Stokes (N-S) discussion online is confusing. I attempted to explain it to myself with AI assistance (several). Why? N-S equations are used in the real world e.g. surfboard design and arterial blood flow sims, it seems important to know what the excitement is about.

Intuition: fill a 3D space with fluid (e.g. water or air). Set it moving in a way that is smooth, then take your hands away. Now drop a speck of dust into it and watch it spin. Can the fluid, and the flow around it, make that speck spin infinitely fast in finite time?

N-S contains a 'runaway' mechanism in 3D. Notice first, stretch a spinning thread of fluid and it thins and spins faster. Imagine an ice skater spinning in place, then pulling her arms in. Faster spin can drive more flow around the thread of fluid that causes further stretch, which drives faster spin again. Against that, the fluid's own friction (viscosity) bleeds energy away. Does the resulting spin outrun this friction in finite time? Nobody has proved it either way. The Clay Mathematics Institute put a $1M bounty on it.

Clay asked Princeton University mathematician Charles Fefferman to formalise what needs to be proved to win the prize [1]. Fefferman gives four precise statements (grouped in two directions); proving any one of them would be grounds to claim the prize.

Historically, mathematicians Córdoba & Martínez-Zoroa pursued the 'forced case' direction that uses an external force on a fluid to show 'blow-up' (runaway) in a simpler than N-S equation, but the external force used was 'rough' (not infinitely differentiable).

Recently, Buckmaster-Alpöge, with AI assistance. extend this approach to a set of simpler than N-S equations (IPM, Boussinesq, Euler 3D) using a 'smooth' external force and formalize it in Lean for automated verification [2]. Notice, Euler 3D is N-S with the viscosity term deleted. This alone is insufficient to claim the prize. But, famed mathematician Terence Tao suggests that this approach is likely to be extensible to Navier-Stokes [3], if I've understood him well. Hence the excitement among mathematicians + physicists + enthusiasts.

Separately, OpenAI claim it has solved the 'forced case' for Navier-Stokes (not 3 simpler equations), and formalised it in Lean [4], but will not claim the prize. I am not a specialist who can carefully compare OAI's approach v. the Buckmaster-Alpöge approach (nb. Alpöge is Anthropic-employed but worked on this in his personal time). That's the discussion online I'm not interested in, and somewhat gets in the way of understanding this result.

Upshot: it seems spin can run away from friction in the forced case; independent verification awaits. What a time to be alive.

[1] https://www.claymath.org/wp-content/uploads/2022/06/navierstokes.pdf
[2] https://cims.nyu.edu/~tristanb/statement.pdf
[3] https://terrytao.wordpress.com/2026/09/07/finite-time-blowup-with-smooth-forcing-term-for-the-incompressible-porous-medium-boussinesq-and-incompressible-euler-equations/
[4] https://cdn.openai.com/pdf/32d9f210-8b73-45e0-91bc-82a30aef8a9a/navier-stokes.pdf

1. (1) Navier-Stokes: ∂ₜu + (u·∇)u = −∇p + νΔu + f

   (2) Euler: ∂ₜu + (u·∇)u = −∇p + f

   Both (1)-(2) with ∇·u = 0 (incompressible). Derived from conservation laws, except for νΔu (Newtonian assumption: double the shear rate, double the stress). Note, comparing (1) and (2), Euler has no viscosity (friction) term (νΔu).

   Notation:

   u - velocity vector
   p - pressure
   ν - kinematic viscosity
   f - external force
   ∇ - gradient
   Δ - laplacian (2nd derivative)
   ∂ₜu - change in velocity wrt time
   ∇p - pressure gradient
   νΔu - diffusion (friction via Laplacian)
   (u·∇)u - advection: fluid carrying itself along. Non-linear term.
   ∇·u = 0 - volume is conserved ('incompressibility')

2. Most real-world use of N-S is an approximation. Why? The root problem with N-S seems to be the non-linear term (u·∇)u. In F = ma terms, acceleration follows ∂ₜu + (u·∇)u, where the second term is quadratic and represents the fluid transporting itself. Because it's quadratic, averaging is lossy: the mean of the true evolution isn't the evolution of the mean.

   Book-keeping splits velocity into mean plus fluctuation (u = ū + u′) to get four terms; two average away, but the one quadratic in u′ doesn't. The surviving term, ⟨u′u′⟩, is called 'Reynolds stress'. Writing an equation for it needs the third moment, which needs the fourth, with no last link (closure problem). So it seems most approximations model the Reynolds stress as a function of the mean flow (ū) instead, importing ideas from molecular viscosity to do so (physical intuition).

   Ultimately a guess with fitted constants. Interesting.