# PTT-042: Navier–Stokes — The Proof You Can Enter

**Prospective Proof-to-Theory benchmark · frozen before later commentary**

**Frozen candidate SHA-256:** `9a5d4220f22ad535f5e1528bd01cef89f8bf446549191f517d20ef724073a203`

The source is OpenAI's 166-page *Finite time blowup for Navier–Stokes*, the exact pinned `openai/NavierStokesAndEuler` repository at commit `f9e8bc5b38b6e212696e8a30e3e91517af887bbd`, and the Clay Mathematics Institute problem statement.

PTT-042 was run prospectively. No later explanatory source was supplied as an answer key.

## 1. The claim

The source does **not** claim a zero-force Navier–Stokes blowup.

For every viscosity `ν > 0`, it claims there is a smooth force `f`, compactly supported in space and in positive time, and a smooth solution `u,p` on `R^3 × [0,1)` with zero initial velocity such that:

- `u` solves the exact forced `ν`-Navier–Stokes equation;
- `div u = 0`;
- `u` and `p` remain inside one fixed compact spatial set before time one;
- the kinetic energy remains uniformly bounded;
- `limsup_{t→1} ||u(t)||_∞ = ∞`;
- there is no global smooth solution with the same force, same zero datum, and uniformly bounded whole-space kinetic energy.

The force remains smooth through the singular time. That is one of the central burdens of the proof.

The source also gives a periodic `T^3` corollary and maps the whole-space and periodic results to the forced Clay alternatives (C) and (D).

It does **not** establish zero-force blowup, blowup for every datum, one force working for all `ν`, an inviscid conclusion, or failure of global weak solutions.

## 2. The problem the proof has to solve

A concentrating flow is not enough.

The construction has to produce a flow that becomes unbounded while all of the following remain true at the same time:

1. the equation is exact;
2. the force stays smooth;
3. incompressibility survives every correction and cutoff;
4. energy stays bounded;
5. the singular growth path survives all later repairs;
6. the whole-space uniqueness argument applies to an arbitrary same-force finite-energy competitor;
7. viscosity scaling and periodization do not change the theorem being proved.

That makes the proof easier to carry as four coupled obligations.

### [G] Growth

The shrinking carrier has to concentrate strongly enough to create the singular path, while occupying little enough volume to remain compatible with bounded energy.

### [S] Stress

The leading tangential residual of the carrier has to be rewritten as annular stress inside a **positive covariance cone** realizable by two wave families. Positivity is load-bearing because the wave amplitudes enter through squares.

### [F] Force

The waves only cancel the leading stress after averaging. Their exact physical realization creates new oscillatory, nonlinear, pressure, evaluation and cutoff errors. The construction iterates corrections until the **full exact residual** is flat in every required Cartesian derivative. Only then can that residual extend smoothly through time one and serve as the forcing term.

### [T] Terminal theorem

The corrected object still has to satisfy the full energy estimate, the arbitrary-competitor uniqueness theorem, viscosity scaling, disjoint nonlinear periodization, and the exact Clay comparator class.

The useful mnemonic is:

`Growth → Stress → Force → Terminal theorem`

This is a reading interface, not a replacement proof.

## 3. The source-bound proof spine

PTT-042 reorganized the source into the following dependency chain:

`N0 target theorem`
→ `N1 growth carrier + N2 profile/stress compatibility`
→ `N3 background correction`
→ `N4 oscillatory stress realization`
→ `N5 full residual correction`
→ `N6 residual becomes smooth force`
→ `N7 energy + N8 whole-space comparison`
→ `N9 viscosity scaling`
→ `N10 periodic bridge`
→ `N11 Clay comparator bridge`

Two gates are especially easy to lose in a summary.

**N6:** the residual has to become a legal smooth force, not just a small defect.

**N8:** the whole-space comparison must work against any smooth same-force uniformly finite-energy competitor, not just one confined to the construction's compact region.

Removing either changes the theorem.

## 4. Where the proof is actually hard

The source already contains a substantial physical explanation. PTT-042 did not discover a hidden short proof.

The difficult bridges that remain are concrete:

- why the asymmetric carrier has the required positive stress cone throughout the full annulus;
- why exactly the displayed radial moment conditions are the right obstructions;
- how shear amplification and viscous damping coexist for both wave families;
- why every correction cycle pushes the complete nonlinear residual upward in regularity despite derivative losses;
- why the locally finite corrected sum is flat in **all** Cartesian jets needed for smooth force extension;
- why the full corrected localized field has uniformly bounded energy;
- why whole-space uniqueness closes against an unbounded-support finite-energy competitor;
- whether an independent build of the pinned Lean repository confirms the reported declaration dependencies and axiom sets.

That is why the frozen mechanism and representation rulings are both `PARTIAL`.

## 5. What breaks when you remove a load-bearing part

PTT-042 predicts the following failures for the **sourced architecture**:

1. Remove one wave family → generic two-component stress realization fails.
2. Impose exact midplane reflection symmetry → the source's two shear mechanisms degenerate together.
3. Delete the zero radial moments → the anti-derivative stress leaks outside the annulus and exterior/pressure matching fails.
4. Stop after finitely many correction cycles → sufficiently high derivative order is uncontrolled, so the force is not certified `C^∞` through time one.
5. Keep only averaged wave cancellation → the exact physical PDE still has oscillatory, nonlinear, pressure and chain-rule errors.
6. Change the pulse schedule so viscosity does not damp before cutoff → the source's flatness estimates fail.
7. Multiply velocity directly by a cutoff → incompressibility generally breaks.
8. Periodize overlapping supports → nonlinear cross-interactions appear between translated copies.
9. Drop the finite-energy competitor condition → the current whole-space uniqueness theorem no longer closes; the result is `UNKNOWN` in the enlarged class.

These are failure predictions for this proof architecture, not impossibility theorems for all possible methods.

## 6. What the frozen run earned

| Dimension | Ruling |
|---|---|
| Theorem / scope | `PASS` |
| Dependency reconstruction | `PARTIAL` |
| Human mechanism | `PARTIAL` |
| Representation compression | `PARTIAL` |
| Source / formal alignment | `PARTIAL` |
| Intervention quality | `PARTIAL` |

The strongest defensible conclusion is narrower than “AI solved Navier–Stokes.”

The locally frozen paper claims, and the pinned Lean source declares proofs of, a smooth-forced finite-time singular flow for every positive viscosity with uniformly bounded energy, together with the periodic corollary and Clay comparator bridges. PTT-042 reconstructed the theorem-facing architecture. It did not independently verify the 166-page proof or produce a substantially simpler human proof.

## 7. Freeze and provenance

Candidate SHA-256:

`9a5d4220f22ad535f5e1528bd01cef89f8bf446549191f517d20ef724073a203`

Frozen size: 36,846 bytes.

Model: `gpt-6-sol`.

The Console version used for the run wrote the exact PTT-042 candidate bytes under the stale filename `PTT041_CANDIDATE.md`. The bytes and hash were preserved; the incident was recorded; the benchmark was not rerun.

## Sources

1. OpenAI, *Finite time blowup for Navier–Stokes*  
   https://cdn.openai.com/pdf/32d9f210-8b73-45e0-91bc-82a30aef8a9a/navier-stokes.pdf
2. OpenAI, `NavierStokesAndEuler`, pinned commit `f9e8bc5b38b6e212696e8a30e3e91517af887bbd`  
   https://github.com/openai/NavierStokesAndEuler
3. Charles L. Fefferman, *Existence and Smoothness of the Navier–Stokes Equation*, Clay Mathematics Institute  
   https://www.claymath.org/wp-content/uploads/2022/06/navierstokes.pdf