← All problems

Navier–Stokes existence and smoothness

91 years old · documented by

Fix a viscosity ν>0\nu>0. The incompressible Navier–Stokes equations in R3\mathbb R^3 for a velocity field u(x,t)∈R3u(x,t)\in\mathbb R^3 and a pressure p(x,t)∈Rp(x,t)\in\mathbb R, with x∈R3x\in\mathbb R^3 and t≥0t\ge0, are

∂tu+(u⋅∇)u=ν Δu−∇p+f,∇⋅u=0,u(x,0)=u0(x).\partial_t u+(u\cdot\nabla)u=\nu\,\Delta u-\nabla p+f,\qquad \nabla\cdot u=0,\qquad u(x,0)=u_0(x).

Here u0u_0 is a smooth divergence-free vector field, f(x,t)f(x,t) is a given external force, and Δ=∑j=13∂2/∂xj2\Delta=\sum_{j=1}^3\partial^2/\partial x_j^2. Following Fefferman's official Clay description [1], data and solutions come in two settings.

Whole space. The data decay rapidly: for all multi-indices α\alpha, all m≥0m\ge0 and all K≥0K\ge0,

∣∂xαu0(x)∣≤CαK(1+∣x∣)−K,∣∂xα∂tmf(x,t)∣≤CαmK(1+∣x∣+t)−Kon R3×[0,∞).|\partial_x^\alpha u_0(x)|\le C_{\alpha K}(1+|x|)^{-K},\qquad |\partial_x^\alpha\partial_t^m f(x,t)|\le C_{\alpha mK}(1+|x|+t)^{-K}\quad\text{on }\mathbb R^3\times[0,\infty).

A solution is admissible if u,p∈C∞(R3×[0,∞))u,p\in C^\infty(\mathbb R^3\times[0,\infty)) and the energy is bounded:

sup⁡t≥0∫R3∣u(x,t)∣2 dx<∞.\sup_{t\ge0}\int_{\mathbb R^3}|u(x,t)|^2\,dx<\infty.

Periodic. With e1,e2,e3e_1,e_2,e_3 the unit vectors, u0(x+ej)=u0(x)u_0(x+e_j)=u_0(x) and f(x+ej,t)=f(x,t)f(x+e_j,t)=f(x,t), u0u_0 is smooth, and for all α,m,K\alpha,m,K

∣∂xα∂tmf(x,t)∣≤CαmK(1+t)−Kon R3×[0,∞).|\partial_x^\alpha\partial_t^m f(x,t)|\le C_{\alpha mK}(1+t)^{-K}\quad\text{on }\mathbb R^3\times[0,\infty).

A solution is admissible if u,p∈C∞(R3×[0,∞))u,p\in C^\infty(\mathbb R^3\times[0,\infty)) and u(x+ej,t)=u(x,t)u(x+e_j,t)=u(x,t) and p(x+ej,t)=p(x,t)p(x+e_j,t)=p(x,t) for j=1,2,3j=1,2,3. Periodicity of pp is required by the errata to [1]. This is the problem on the torus T3=R3/Z3\mathbb T^3=\mathbb R^3/\mathbb Z^3.

Problem. Prove at least one of the following four statements.

  • (A) Existence and smoothness on R3\mathbb R^3. For every ν>0\nu>0 and every smooth divergence-free u0u_0 satisfying the whole-space decay bound, with f≡0f\equiv0, there is an admissible whole-space solution.
  • (B) Existence and smoothness on R3/Z3\mathbb R^3/\mathbb Z^3. For every ν>0\nu>0 and every smooth, divergence-free, periodic u0u_0, with f≡0f\equiv0, there is an admissible periodic solution.
  • (C) Breakdown on R3\mathbb R^3. For every ν>0\nu>0 there are a smooth divergence-free u0u_0 and a smooth ff, both satisfying the whole-space decay bounds, for which no admissible whole-space solution exists.
  • (D) Breakdown on R3/Z3\mathbb R^3/\mathbb Z^3. For every ν>0\nu>0 there are a smooth, divergence-free, periodic u0u_0 and a smooth periodic ff with the stated time decay for which no admissible periodic solution exists.

By the scaling u↦λu(x,λt)u\mapsto\lambda u(x,\lambda t), each statement for one value of ν\nu is equivalent to the same statement for every ν>0\nu>0. Statements (A) and (B) concern the unforced equations. Statements (C) and (D) allow a force.

Background

The equations are Newton's law for a viscous incompressible fluid. Viscosity damps velocity gradients, while the nonlinear term (u⋅∇)u(u\cdot\nabla)u can move energy to ever smaller scales. The energy identity bounds ∫∣u∣2\int|u|^2 for all time. In three dimensions this bound is supercritical under the natural scaling and does not control the derivatives needed for smoothness. [1]

Leray constructed global weak solutions with finite energy in 1934 [2], and Hopf extended this to bounded domains [3]. Whether weak solutions from smooth data stay smooth, and whether they are unique, was left open. For any smooth data there is a smooth solution on some interval [0,T)[0,T). If the maximal TT is finite, then sup⁡x∣u(x,t)∣→∞\sup_x|u(x,t)|\to\infty as t↑Tt\uparrow T. So (A) and (B) are equivalent to ruling out finite-time blowup. In two space dimensions global regularity has long been known. [1]

References
  1. C. L. Fefferman, "Existence and smoothness of the Navier–Stokes equation", in J. Carlson, A. Jaffe and A. Wiles (eds.), The Millennium Prize Problems, Clay Math. Inst. and AMS (2006), with errata. Clay PDF
  2. J. Leray, "Sur le mouvement d'un liquide visqueux emplissant l'espace", Acta Math. 63 (1934), 193–248. doi:10.1007/BF02547354
  3. E. Hopf, "Über die Anfangswertaufgabe für die hydrodynamischen Grundgleichungen", Math. Nachr. 4 (1950), 213–231. doi:10.1002/mana.3210040121
  4. H. Fujita and T. Kato, "On the Navier–Stokes initial value problem. I", Arch. Rational Mech. Anal. 16 (1964), 269–315. doi:10.1007/BF00276188
  5. H. Koch and D. Tataru, "Well-posedness for the Navier–Stokes equations", Adv. Math. 157 (2001), 22–35. doi:10.1006/aima.2000.1937
  6. L. Caffarelli, R. Kohn and L. Nirenberg, "Partial regularity of suitable weak solutions of the Navier–Stokes equations", Comm. Pure Appl. Math. 35 (1982), 771–831. doi:10.1002/cpa.3160350604
  7. L. Escauriaza, G. A. Seregin and V. Šverák, "L3,∞L_{3,\infty}-solutions of the Navier–Stokes equations and backward uniqueness", Russian Math. Surveys 58 (2003), 211–250. doi:10.1070/RM2003v058n02ABEH000609
  8. T. Tao, "Finite time blowup for an averaged three-dimensional Navier–Stokes equation", J. Amer. Math. Soc. 29 (2016), 601–674. arXiv:1402.0290
  9. T. Buckmaster and V. Vicol, "Nonuniqueness of weak solutions to the Navier–Stokes equation", Ann. of Math. 189 (2019), 101–144. arXiv:1709.10033
  10. D. Albritton, E. Brué and M. Colombo, "Non-uniqueness of Leray solutions of the forced Navier–Stokes equations", Ann. of Math. 196 (2022), 415–455. arXiv:2112.03116
  11. OpenAI, "Finite time blowup for Navier–Stokes", preprint (September 2026). PDF; Lean formalization at github.com/openai/NavierStokesAndEuler
  12. Wikipedia, Navier–Stokes existence and smoothness.

Source: Clay Mathematics Institute

Progress summary

Refreshed
Open

The problem remains open: a recent claimed construction has not been independently validated, and global smoothness in three dimensions is unresolved.

Fefferman’s formulation asks for either global smooth solutions or an admissible finite-time breakdown, in whole space or on the three-dimensional torus. The forced-breakdown announcement would address two alternatives, but no independent expert assessment has validated it.

Known results

  • Leray (1934) and Hopf (1950) constructed global finite-energy weak solutions, without establishing smoothness or uniqueness.
  • Fujita–Kato (1964) and Koch–Tataru (2001) proved global smooth existence for sufficiently small data in critical spaces such as H˙1/2\dot H^{1/2} and BMO−1\mathrm{BMO}^{-1}.
  • Caffarelli–Kohn–Nirenberg (1982) proved partial regularity; Escauriaza–Seregin–Šverák (2003) ruled out blowup under an Lt∞Lx3L^\infty_tL^3_x bound.
  • Buckmaster–Vicol (2019) and Albritton–Brué–Colombo (2022) proved nonuniqueness for related weak or forced problems, not the Clay alternatives.

Community submission (unverified), November 4, 2026

A submitted argument claims that a nonstandard-analysis framework closes a fluctuation hierarchy and yields global smooth solutions for smooth divergence-free data. It provides no independent verification and does not establish the result.

Current status (as of November 2026): All four alternatives remain unsettled; the claimed forced-breakdown result lacks independent corroboration, and the community submission is unverified.

Sources

Solutions 6

Quadrupolar Structure, Directional Decomposition, and Critical Cancellation in the Three-Dimensional Navier–Stokes EquationsSee full solutionHide full solution

We develop an exact structural formulation of the vorticity dynamics for the three-dimensional incompressible Navier–Stokes equations. Writing ω=ρξ with |ξ|=1, we derive the amplitude–direction decomposition, the exact viscous splitting, the directional stretching identity, and the trace-free quadrupole tensor formulation. We then identify the signed angular structure of the Biot–Savart strain and derive the corrected mixed amplitude–direction pairing. The angular mean-zero and L² identities isolate a finite-dimensional, quadrupolar interaction channel. A dyadic formulation is given using Littlewood–Paley projectors. The results are structural identities and proved estimates; no global-regularity conclusion is asserted.

  • Article_I_Quadrupolar_Structure_DeWatson-Ledetsambali.pdf286,450 bytesOpen
  • Article_II_Critical_Quadrupole_Endpoint_DeWatson-Ledetsambali.pdf275,753 bytesOpen
  • navier-1-compressed.pdf6,679,767 bytesOpen
  • Article_III_Conditional_Global_Regularity_DeWatson-Ledetsambali.pdf268,163 bytesOpen
We recast the 3D incompressible Navier–Stokes equations with monads of scale ε, where incompressibility becomes T_x+T_y+T_z=3. Under this constraint convection and viscosity vanish inside a monad, and the nonlinearity passes into the pressure source ∇u·∇u. With ℓ_f≫ε the analytic radius stays β_c>0, so BKM never triggers and blow-up is excluded.DOI:[10.5281/zenodo.23233956]See full solutionHide full solution

Solution*. We reformulate the three-dimensional incompressible Navier–Stokes equations using nonstandard analysis together with a difference-differential algebra, and show that in this formulation the obstruction attributed to the convective term is relocated rather than approximated.

The basic object is not a point but a monad: the infinitesimal neighbourhood of a real point, of linear scale ε. Inside a monad the velocity takes a single value, so the derivative along any direction is replaced by a difference quotient between adjacent cells. Incompressibility, written at the level of the three shift operators T_x, T_y, T_z, becomes the algebraic constraint T_x+T_y+T_z=3, equivalently the difference-differential constraint K_m=0. Under this constraint the convective term and the Laplacian vanish identically inside every monad, leaving only the time derivative on the left and the pressure gradient together with the external force on the right.

The nonlinearity does not disappear. Taking the divergence of the in-monad equation returns it as the source ∇u·∇u of the pressure Poisson equation, i.e. as the nonlinear part of Zharinov's hidden integrability condition Δp+∇u·∇u=0, whose addition is what makes the system formally integrable. Because the fluctuation inside a monad is a linear functional of the mean velocity gradient, the Reynolds stress acquires an explicit expression in terms of that gradient; the closure problem is dissolved rather than modelled, and the moment hierarchy closes at every order with the single ratio εG.

With the hierarchy closing geometrically, the analytic radius β_c is bounded below by a positive quantity for as long as the construction is valid, so the BKM blow-up criterion—divergence of the time integral of the vorticity sup-norm—is never met, and global smooth solutions exist for smooth divergence-free data.

The framework has an explicit domain of validity. It requires the external-force scale to be large compared with the monad, ℓ_f≫ε, i.e. λ_f=ε/ℓ_f≪1; if the force varies at the monad scale the velocity inside a monad is no longer uniform and both the vanishing and the transfer above fail. The argument is carried out monad by monad, and its conclusions concern the monad-wise linearised system; passing from that system to a global solution is a separate step, and the hypotheses under which the analytic radius stays positive are stated explicitly rather than assumed.DOI:[10.5281/zenodo.23233956]

  • 阅读版-中文-含结论.pdf1,918,864 bytesOpen
  • Monad-Elimination-and-Difference-Differential-Constraints-EN.pdf1,332,030 bytesOpen
Exposition (not a new solution) for the Exposition category, JSP-000005. A 15-page guided essay on OpenAI's forced finite-time blowup construction (alternatives C and D). It explains the residual reformulation, the collapsing anisotropic vortex, how amplified pulses supply the missing annular stress, the infinite-order correction cycle, uniqueness, the torus case, and what the Lean proof certifies. Claims are tagged by provenance; (A)/(B) remain open.See full solutionHide full solution

Exposition, not a new solution. Submitted for the Exposition category of the MathDB × Justin Sun Prize pilot (JSP-000005).

This 15-page essay explains OpenAI's Finite time blowup for Navier–Stokes (September 2026) and its Lean formalization (openai/NavierStokesAndEuler, commit f9e8bc5b). For every ν>0\nu>0 the paper constructs a smooth, compactly supported force and a flow that starts from rest, keeps bounded energy, and blows up at t=1t=1. This addresses alternatives (C) and (D).

What the essay covers

  • The reduction to a cancellation problem: the residual R(u,p)=∂tu+(u⋅∇)u−νΔu+∇p\mathcal R(u,p)=\partial_t u+(u\cdot\nabla)u-\nu\Delta u+\nabla p must extend smoothly through the singular time.
  • The anisotropic collapsing vortex: ℓr≍τ1/2\ell_r\asymp\tau^{1/2}, ℓz≍τ1/2−h\ell_z\asymp\tau^{1/2-h}, speeds ≍τ−1/2−h\asymp\tau^{-1/2-h}. Energy →0\to0 and dissipation is integrable since h<1/6h<1/6.
  • The missing momentum as an annular stress, supplied by two families of amplified oscillatory pulses through their Reynolds stress (the admissible stress cone).
  • The infinite-order correction cycle (σj+1=σj+110\sigma_{j+1}=\sigma_j+\tfrac1{10}) and its summation with shrinking cutoffs.
  • Localization, the force extension past t=1t=1, the energy bound, uniqueness, and the torus case.
  • What the Lean/Comparator development certifies, and what it does not.
  • A reading map of the 166-page paper and a glossary.

Every statement is tagged [Paper], [Check] (verified independently), [Heuristic] or [Context]. The essay states that independent review is pending and that the unforced alternatives (A) and (B) remain open. All mathematics is due to the cited authors; this is exposition only.

Fixed version (v1.0, 29 Oct 2026): https://github.com/Thanh-Van-2001/navier-stokes-blowup-exposition/releases/tag/v1.0 PDF: https://github.com/Thanh-Van-2001/navier-stokes-blowup-exposition/releases/download/v1.0/ns_exposition.pdf

Prepared with AI assistance (Claude); all statements were checked against the source paper and repository. A narrated video companion will be added.

  • ns_exposition_video.mp421,447,216 bytesOpen
  • ns_exposition.pdf507,556 bytesOpen
Exposition (not a new solution) for the Exposition category, JSP-000005. A 10-page essay and a 5:52 narrated video on OpenAI's forced finite-time blowup construction (alternatives C and D). Claims are tagged by origin, and it reports our own rerun of the Lean proof with Comparator and the limits of that check. (A) and (B) remain open.See full solutionHide full solution

Exposition, not a new solution. Submitted for the Exposition category of the MathDB × Justin Sun Prize pilot (JSP-000005).

Author: @jhonsu777 (2xrqx82e2@mozmail.com). Prepared with AI assistance (Claude, Anthropic). All the mathematics is due to the cited authors.

What it covers

  • Why the question reduces to a cancellation problem: any flow solves the equation with some force, so the force must stay smooth through the singular time.
  • The collapsing anisotropic vortex and its scales, recomputed symbolically.
  • How oscillating pulses supply the missing stress in the annulus, and the correction cycle.
  • Localization, the uniqueness lemma, and the torus case.
  • Heuristic consistency checks against the L3L^3 and L5L^5 regularity criteria. These are our own observations, not proofs.
  • Three separate questions about the Lean proof: whether it passes, whether its statement matches Clay's, and whether the written paper is right line by line.

What we ran. On commit f9e8bc5b of openai/NavierStokesAndEuler, with Lean v4.34.0-rc2: the build completed, #print axioms shows only propext, Classical.choice and Quot.sound, leanchecker --fresh exited with code 0, and Comparator accepted the solution with both the Lean kernel and nanoda on a clean volume.

Limits. We did not verify the written proof, and this is not an independent review. Our reading of the statement's fidelity is not an expert certification. Our Comparator run lacked the recommended systemd-run restriction. We confirmed one Lean-to-paper mismatch reported by Bastounis, Circelli and Hansen (arXiv:2610.08144): equation (8.19) has Cm+4C^{m+4}, while the Lean lemma assumes bounds at w.length + 5. Alternatives (A) and (B) remain open.

Materials

Exposition - An 11-page essay and two short videos that explain OpenAI's forced finite-time blowup for 3D Navier–Stokes to readers who know calculus but not fluid PDESee full solutionHide full solution

Exposition for JSP-000005: Navier–Stokes existence and smoothness.

Author: Zinan Xing. Date: November 4, 2026.

Exposition essay: https://github.com/neoskyz/navier-stokes-exposition/blob/v1.0/navier-stokes-exposition.pdf

Videos: V1, narrated (author's voice):

Open on YouTube

V2, animation flow + music, no narration:

Open on YouTube

Sources:

Contributions: A reader-level narrative of the construction, a clause-by-clause check of the Lean statements against Fefferman's text, and two companion videos.

(The essay, figures and videos were prepared with the assistance of Claude)

An article and 9:43 narrated film explaining the announced smooth-force Navier–Stokes breakdown: the mechanism, four official alternatives, Lean statements and execution evidence. Includes a short reading route, claim and source ledgers, and exact elementary checks. Forced C and D are claimed and formalized; unforced A and B remain open. Fixed version v1.2.0.See full solutionHide full solution

Exposition by Yao Xu: an article and a narrated film on how a smooth force is claimed to drive a fluid from rest into a finite-time velocity singularity while energy stays bounded. The mechanism starts with a designed flow, makes its residual the force, and uses oscillatory momentum transfer and corrections at every derivative order to keep that force smooth through blowup.

Read the article, watch the 4K film, or inspect the evidence. The PDF and a 1080p film copy are attached here. The repository includes timed subtitles and chapter navigation; the article offers a ten-minute reading route and a compact notation key.

Claimed: OpenAI's construction excludes global smooth solutions with bounded energy on the whole space (C), and global smooth periodic solutions on the torus (D), for its chosen data and force.
Lean: our 9 November 2026 run built the pinned development, printed only the three standard foundational axioms for both endpoints, and recorded acceptance by both Comparator kernels. Mathlib's compiled files came from its binary cache; Comparator ran after the full project build, as configured.
Open: unforced alternatives A and B. The exposition has not been peer reviewed.

Original contributions:

  • A guided explanation and film with explicit claim statuses.
  • Clause-by-clause comparison of the formal and official statements.
  • Exact elementary checks, negative controls and an import-closure source audit.
  • A dated evidence account, including two manuscript–Lean estimate differences.
  • Our source-build, axiom and two-kernel execution records.

The mathematics belongs to the cited authors. Fixed version: v1.2.0, commit 12fe29c6a06ca339add2e7118cbf8ea4e07472f2, published 9 November 2026, 23:02 UTC (Software Heritage save request 2634605). First public version: v1.0.0, 9 November 2026, 00:36 UTC; Software Heritage snapshot, captured at 00:36:28 UTC.

  • navier-stokes-exposition-film-1080p-v1.2.0.mp491,912,438 bytesOpen
  • navier-stokes-exposition-article-v1.2.0.pdf1,224,303 bytesOpen