OpenAI Shows Forced Navier-Stokes Flows Can Carry Universal Computation

OpenAI published the paper 'Universal computation in forced Navier–Stokes flows', accompanied by a Lean formalized proof (in the openai/math repository), claiming that forced Navier-Stokes flows can carry universal computation. Oxford scholar Anders Sandberg called it his favorite of OpenAI's mathematical results and posted a series of technical takes on its implications and limitations.

Confirmed

• The paper is 'Universal computation in forced Navier–Stokes flows', with Lean formalization; the code is in the openai/math repository
• The intuitive picture of the construction: take a universal Turing machine and a program, generate a time-varying force field that drives fluid on a 3D torus, and let particles transported by advection fall into boxes representing 0 or 1 according to the computation's output
• Sandberg noted the result is 'less stunning than it sounds': nearly all of the complexity is shifted into the force field driving the fluid
• Still, the construction has substantive constraints: nonzero viscosity, smooth force field, zero initial velocity, and the force is not an oracle—so it is not a trivial construction
• Sandberg analyzed the error budget: spatial error is manageable, but error tightens at a quadratic rate as computation time grows
• He also clarified a conceptual point: the essence is not 'fluid can compute' but 'fluid can contribute the nonlinear steps of a computation', which is a decent starting point

Why it matters

• The result links computability with fluid dynamics (the Navier-Stokes equations) and carries Lean formalization, making it a rare crossover between formalized mathematics and physics
• Sandberg's reading offers a crucial complexity-theoretic perspective: outsiders should avoid misreading it as 'fluids are intelligent' or having independent computational ability—the program's information is really carried by a carefully engineered force field

2026-10-16 ~ 2026-10-16 · 5 related posts

• [source] OpenAI Lean-proves universal computation in forced Navier–Stokes flows — anderssandberg · 2026-10-16
• How the fluid computation works: force field advects particle to 0/1 boxes — anderssandberg · 2026-10-16
• [source] Sandberg: all the complexity in fluid computation hides in the force field — anderssandberg · 2026-10-16
• [source] Anders Sandberg on fluid computation: error budgets tighten quadratically over time — anderssandberg · 2026-10-16
• Fluids Don't Compute—They Contribute the Nonlinear Step, Sandberg Argues — anderssandberg · 2026-10-16