Search

Universal computation in forced Navier–Stokes flows

Proves a pushed fluid can run any program: a marked particle enters a fixed box exactly when the program halts.

9 papers September 27, 2026 Explained by AI

A Turing machine is a simple model of any computer program. These nine papers turn any such machine and its input into a recipe for pushing a thick fluid, like honey in a box. The fluid starts at rest. One marked drop of fluid reaches a fixed target region exactly when the program stops. Since no method can decide in general whether a program stops, no method can decide in general where that drop goes.

First step
—
Significance
Solid
It moves fluid computation from specially chosen geometries to a fixed flat domain, with viscosity and a fluid at rest. Lean covers several versions.
Trust
Formal proof listed
Field
Partial differential equations
9 papers

The explanation

Start with a simple flow on a cube whose opposite faces are glued, called a flat torus. Let the fluid move only in the X direction, with speed A(t)·sin(2πY). Here Y is height and A(t) is a strength you choose. This flow is incompressible: it never squeezes the fluid. For this flow the fluid needs no pressure. The force only has to speed it up and beat friction. With viscosity ν = 1 and A(t) = t, the needed force is (1 + 4π²t)·sin(2πY). At height Y = 1/4 and time 1 this is about 40.48. A drop at that height travels t²/2, so 0.5 units by time 1.

This is the basic trick. The Navier–Stokes equations describe a viscous, incompressible fluid. Viscosity is its internal friction. If you choose the motion first, the force that produces it is just acceleration minus the friction term. So the hard part is not the force. It is to choose a motion that runs a program, written from the program's table of rules without running it first.

A Turing machine is a simple model of a computer: a tape of symbols, a head, and a finite table of rules. It halts if it reaches a stop state. One step of the machine can erase information, but a smooth flow can always be run backward. So the papers keep a record of past steps, an idea going back to Bennett. Each step then becomes a reversible map on small rectangles or boxes. Shear flows like the one above carry out these maps.

Earlier fluid computers used steady flows on specially shaped spaces. Cardona, Miranda, Peralta-Salas and Presas built one on a reshaped three-sphere. Here the space is flat and fixed, and the fluid starts at rest.

Take any machine and input. The papers claim a smooth force where a fixed particle enters a fixed region exactly when the machine halts. So deciding where the particle goes is undecidable: no general method can answer it. Variants make the force constant after time one, or let force and velocity fade faster than any power of time. In another, the velocity at a fixed spot reveals halting. The last paper puts the whole flow in one bounded region of ordinary space.

How sure is this? Lean, a computer proof checker, has formal statements of several versions. They include the sheet maps, the fading-force memory, the velocity tests and the bounded-region computer in ordinary space. So those are proved, as far as the Lean statements match the papers. Other variants, such as the eventually constant force, are claims only. No specialist review is published.

What it doesn't show. This does not say that natural, unforced fluid flows compute, and it says nothing about whether Navier–Stokes solutions can blow up. The force is designed by hand for each program and input.

Where it came from

This result comes from an unreleased OpenAI model, published in the openai/math release on 6 October 2026.

Hype check

Lean statements cover several of the main constructions. The forces are hand-built for each program; this says nothing about natural flows or about Navier–Stokes blowup. No specialist review is published.

Words used

Navier–Stokes equations
The equations of motion for a viscous fluid, like water or honey.
Incompressible
The fluid is never squeezed or stretched in volume; its velocity has zero divergence.
Viscosity
The internal friction of a fluid, written ν.
Flat torus
A cube whose opposite faces are glued, so fluid leaving one side comes back on the other.
Turing machine
A simple model of a computer: a tape of symbols, a moving head, and a finite table of rules.
Halting
A Turing machine reaching its stop state after finitely many steps.
Undecidable
No single method can give the right yes-or-no answer for every case.
Shear flow
A flow where layers slide past each other at different speeds.

The original papers (9)

Lean notes

All quoted lines (13)
  1. The force identity itself is elementary: for a prescribed incompressible velocity, its material acceleration minus its viscous term is a body force.
    Universal Computation with Eventually Stationary Navier–Stokes Forcing Back to text
  2. The substance of the argument is to prescribe the velocity from the finite machine table without running that machine.
    Universal Computation with Eventually Stationary Navier–Stokes Forcing Back to text
  3. A difficulty is that a tape instruction may erase information whereas a smooth flow has invertible finite-time maps.
    Universal Computation with Eventually Stationary Navier–Stokes Forcing Back to text
  4. Landauer discusses retaining intermediate information to avoid this loss; Bennett implements reversible simulation by recording instruction indices.
    Universal Computation with Eventually Stationary Navier–Stokes Forcing Back to text
  5. We realize finite reciprocal affine instruction maps by smooth incompressible shear flows on a flat three-torus.
    Finite Instructions and Solenoidal Shear Flows Back to text
  6. Cardona, Miranda, Peralta-Salas and Presas obtain stationary Euler flows on an adapted Riemannian three-sphere, using an area-preserving realization of generalized shifts and a suspension.
    Universal Computation with Eventually Stationary Navier–Stokes Forcing Back to text
  7. Here the flat metric is fixed and the fluid starts at rest.
    Universal Computation with Eventually Stationary Navier–Stokes Forcing Back to text
  8. We construct smooth external forces for incompressible Navier–Stokes flow on a fixed flat three-torus at any fixed positive computable viscosity, from zero initial velocity, such that a fixed particle enters a fixed open set exactly when a prescribed Turing machine halts.
    A Fixed Particle Test for Computation in a Forced Viscous Flow Back to text
  9. The resulting reachability problem is undecidable.
    A Fixed Particle Test for Computation in a Forced Viscous Flow Back to text
  10. At every fixed positive computable viscosity, we construct smooth mean-zero forces on the flat three-torus that become stationary after time one and make a fixed particle, initially in a fluid at rest, enter a fixed open set exactly when a prescribed Turing machine halts.
    Universal Computation with Eventually Stationary Navier–Stokes Forcing Back to text
  11. At any fixed positive computable viscosity, smooth forces can make a fixed fluid particle detect the halting of an arbitrary machine, starting from rest, while every mixed derivative of the force and velocity decreases faster than every inverse power of time.
    Computation under Rapidly Vanishing Navier–Stokes Forcing Back to text
  12. We construct smooth forces for three-dimensional incompressible Navier–Stokes flow, from rest at any fixed positive computable viscosity, whose velocity field detects whether a prescribed machine halts.
    Velocity-Field Detection of Computation in Forced Navier–Stokes Flows Back to text
  13. A balanced three-stack recorder then assigns every machine and finite input a smooth Navier–Stokes force with one compact spatial support, periodic after a loading interval, at any fixed positive computable viscosity.
    Incompressible Box Transport and Finite Computation Back to text

Other sources

  1. openai/math, Lean scope note for family 376 The formalization constructs smooth forced three-dimensional Navier–Stokes flows from rest whose velocity fields detect whether a prescribed machine halts, at any positive computable viscosity.
  2. openai/math release notes (2026): model provenance and review status manuscripts and supporting proof artifacts produced by an internal OpenAI model

Browse Partial differential equations Lean-checked resultsSolid results