Navier-Stokes Existence and Smoothness: Fefferman A

Navier-Stokes Existence and Smoothness: Fefferman A

Definitions and mathematical statement

A vector field $u_0\in\mathcal S({\mathbb{R}}^3;{\mathbb{R}}^3)$ is Schwartz if $\sup_x(1+|x|)^N|\partial^\alpha u_0(x)|<\infty$ for every multi-index $\alpha$ and integer $N\ge0$. For every such field with $\nabla\cdot u_0=0$ and every $\nu>0$, seek real smooth $u,p$ on ${\mathbb{R}}^3\times[0,\infty)$ satisfying \[ \partial_tu+(u\cdot\nabla)u=\nu\Delta u-\nabla p,\qquad \nabla\cdot u=0,\qquad u(\cdot,0)=u_0, \] \[ \sup_{t\ge0}\int_{{\mathbb{R}}^3}|u(x,t)|^2\,dx<\infty. \] Smoothness is required at time zero as well. No forcing, periodic domain, or smallness hypothesis is allowed in this selected statement, Fefferman (A).

Formulation and concept references: [S1].

Short English statement

Must every smooth, rapidly decreasing initial flow in unforced three-dimensional viscous fluid motion remain smooth forever with bounded total energy?

Research attempt

Scope and current literature check (27 September 2026).

The quantifiers are: for every real divergence-free Schwartz datum on $\mathbb R^3$ and every fixed $\nu>0$, construct a smooth global solution of the unforced equation with bounded kinetic energy. The selected statement is Fefferman (A), not the assertion that some allowed alternative in the Millennium formulation has been established.

This distinction is essential for the supplied recent sources. Theorem 1.1 on page 1 of the analytical paper [S4] states a construction with zero initial datum and a smooth compactly supported external force, leading to blowup and alternatives (C) and (D). The announcement [S2] describes that forced result. Clay's 11 September 2026 announcement [S3] discusses the announced advance and its evaluation process. Reading these statements does not verify the complete proof, and their forced hypotheses do not supply a counterexample to the unforced target here. No associated Lean files were executed or treated as independent certification.

For the present argument, local strong-solution theory is the input supplied by Kato [E1]. The endpoint theorem of Escauriaza--Seregin--Šverák [E2] says, in particular, that a suitable finite-energy Cauchy solution with bounded $L^\infty_tL^3_x$ norm cannot have a finite-time singularity. This is a conditional theorem, not an available bound for arbitrary data. Tao's averaged-equation construction [E3] warns that energy cancellation alone does not distinguish the actual Navier--Stokes nonlinearity from every blowup model. The calculations below reproduce classical estimates and test their exact failure point; no novelty is claimed for the small-data or continuation criteria.

Attack map and conventions.

The proof route is to control vortex stretching using energy and enstrophy, then test whether a scale-invariant small-data argument extends to large data. The counterexample route is to examine whether a forced singular trajectory can be converted into an unforced one by removing its force. A third, conditional route isolates the accumulated positive strain that would have to be controlled by a sharper geometric argument.

Let $[0,T_*)$ be the maximal smooth strong-solution interval issued from $u_0$. All identities below are used before $T_*$, where Sobolev regularity and spatial cutoffs justify integration by parts. Write \[ \omega=\nabla\times u,\quad S=\tfrac12(\nabla u+\nabla u^{\mathsf T}),\quad E=\|u\|_2^2,\quad Y=\|\omega\|_2^2,\quad Z=\|\nabla\omega\|_2^2,\quad N=\int_{\mathbb R^3}\omega\cdot S\omega\,dx. \] For divergence-free fields, Plancherel gives $Y=\|\nabla u\|_2^2$ and $Z=\|\Delta u\|_2^2$. The pressure is recovered, up to a function of time, from $-\Delta p=\partial_i\partial_j(u_i u_j)$, with repeated indices summed. Pressure is not assumed to retain Schwartz decay; the target only requires smoothness and the velocity energy bound.

Exact identities and the first obstruction.

Taking the $L^2$ product with $u$ eliminates pressure and transport because $\nabla\cdot u=0$. Curling the equation gives \[ \partial_t\omega+(u\cdot\nabla)\omega =\nu\Delta\omega+(\omega\cdot\nabla)u. \] The antisymmetric part of $\nabla u$ contributes zero to its quadratic form on $\omega$. Hence \begin{equation}\label{wsref-0-6e-73-3a-62-61-6c-61-6e-63-65-73} E'=-2\nu Y,\qquad \tfrac12Y'+\nu Z=N. \end{equation} In particular, $\int_0^tY(s)\,ds\leq E(0)/(2\nu)$, but stretching has no analogous sign. Tracelessness of $S$ permits a positive eigenvalue; it does not imply $\omega\cdot S\omega\leq0$.

The Biot--Savart singular-integral estimate gives $\|\nabla u\|_3\leq C\|\omega\|_3$. Interpolation between $L^2$ and the three-dimensional Sobolev bound $\|\omega\|_6\leq C\|\nabla\omega\|_2$ then gives, with one fixed universal constant $C_0$, \begin{equation}\label{wsref-0-6e-73-3a-73-74-72-65-74-63-68} |N|\leq\|\nabla u\|_3\|\omega\|_3^2 \leq C_0Y^{3/4}Z^{3/4} \leq\tfrac\nu2 Z+C_1\nu^{-3}Y^3. \end{equation} The last step is Young's inequality with exponents $4/3$ and $4$. Thus $Y'\leq K\nu^{-3}Y^3$ for a universal $K>0$. This estimate gives a positive lifespan, not a bound valid for all times.

If $T_*<\infty$, bounded $Y$ would imply bounded $L^6$ velocity and allow continuation, as justified below. There is therefore a sequence $s\uparrow T_*$ with $Y(s)\to\infty$. On an interval where $Y>0$, \[ \frac d{dt}Y^{-2}\geq-2K\nu^{-3}. \] Integration from $t$ to those $s$ yields the necessary lower bound \begin{equation}\label{wsref-0-6e-73-3a-72-61-74-65} Y(t)\geq\frac{\nu^{3/2}}{\sqrt{2K(T_*-t)}}. \end{equation} This is compatible with $\int_0^{T_*}Y<\infty$. There is no contradiction between the known energy budget and the derived blowup-rate condition.

To stress-test that conclusion explicitly, put $Y_*(t)=A(T-t)^{-1/2}$ and \[ E_*(t)=E_0-4\nu A\left(\sqrt T-\sqrt{T-t}\right). \] If $A^2\geq\nu^3/(2K)$ and $E_0>4\nu A\sqrt T$, these positive scalar functions satisfy $E_*'=-2\nu Y_*$ and $Y_*'\leq K\nu^{-3}Y_*^3$, yet $Y_*$ diverges at $T$. These are not velocity fields or solutions. They disprove only the proposed inference from these two scalar inequalities to global regularity.

A proved small-data subcase with the correct scaling.

Fourier-space Cauchy--Schwarz yields \[ Y=\|\nabla u\|_2^2\leq\|u\|_2\|\Delta u\|_2=E^{1/2}Z^{1/2}. \] For $Z>0$ this implies $Y^3/Z\leq EY$. Returning to the estimate before Young's inequality gives the stronger, scale-sensitive statement \begin{equation}\label{wsref-0-6e-73-3a-69-6e-76-61-72-69-61-6e-74} |N|\leq C_0(EY)^{1/4}Z, \qquad \tfrac12Y'+\left(\nu-C_0(EY)^{1/4}\right)Z\leq0. \end{equation} If $Z=0$, finite energy makes $u$ spatially constant and hence zero, so there is no exceptional nonzero case.

Proposition 5.1. If $C_0(E(0)Y(0))^{1/4}<\nu$, then the datum in the selected target has a global smooth solution, $E(t)\leq E(0)$, and $Y(t)\leq Y(0)$.

Proof. The zero datum gives $u=0$. Otherwise put $\theta_0=C_0(E(0)Y(0))^{1/4}/\nu<1$ and choose $\theta_1\in(\theta_0,1)$. Before the first time the coefficient ratio $C_0(EY)^{1/4}/\nu$ reaches $\theta_1$, equations (ns:balances) and (ns:invariant) make both $E$ and $Y$ nonincreasing. Their product cannot reach that larger threshold. By continuity the crossing never occurs. In fact \[ Y(t)+2(1-\theta_0)\nu\int_0^t Z(s)\,ds\leq Y(0) \quad(t<T_*). \] The homogeneous Sobolev inequality now bounds $\|u(t)\|_6\leq C Y(0)^{1/2}$ uniformly.

For completeness, the continuation step is quantitative. In the mild equation, the heat--Leray estimate \[ \|e^{\nu\tau\Delta}\mathbb P\nabla\cdot F\|_6 \leq C(\nu\tau)^{-3/4}\|F\|_3 \] comes from heat-kernel differentiation, $L^3$--$L^6$ smoothing, and boundedness of the Leray projector on $L^3$. Since $\|u\otimes v\|_3\leq\|u\|_6\|v\|_6$, the mild bilinear term on a time interval of length $\delta$ has norm at most $C\nu^{-3/4}\delta^{1/4}\|u\|_{C_tL^6_x}\|v\|_{C_tL^6_x}$. The contraction argument therefore gives a lifespan bounded below by $c\nu^3M^{-4}$ whenever the initial $L^6$ norm is at most $M>0$. This is the local theory used in [E1]. Restarting near any proposed finite $T_*$ with the uniform bound just obtained extends the solution past it. Uniqueness of the strong solution makes the overlapping constructions agree. Smoothing for positive time, together with the initial Schwartz regularity, supplies the requested smoothness, including time zero. The energy identity supplies the global energy bound.$\square$

This subcase has the right invariance but is not the full target. Under \[ u_\lambda(x,t)=\lambda u(\lambda x,\lambda^2t),\qquad p_\lambda(x,t)=\lambda^2p(\lambda x,\lambda^2t), \] viscosity is unchanged and $E_\lambda=\lambda^{-1}E$, $Y_\lambda=\lambda Y$ at corresponding times. Thus $EY/\nu^4$ is invariant. A rescaling can make energy arbitrarily small while making enstrophy large; it cannot move arbitrary initial data into Proposition 5.1. Decreasing amplitude alone changes the nonlinear balance and creates different initial data.

A concrete conditional continuation target.

Let $s_+(x,t)=\max\{0,\lambda_{\max}(S(x,t))\}$ and $b(t)=\|s_+(\cdot,t)\|_\infty$. Then $N\leq b(t)Y$, and \[ Y(t)\leq Y(0)\exp\left(2\int_0^t b(s)\,ds\right). \] Consequently finite accumulated positive strain up to a finite endpoint rules out blowup there by the same $L^6$ continuation argument. This criterion is proved here, but the integral is not controlled by energy.

A more selective estimate would suffice. Given $0<\theta<1$, suppose one can choose a nonnegative $k\in L^1(0,T_*)$ such that \begin{equation}\label{wsref-0-6e-73-3a-74-61-69-6c} \int_{\{s_+>k(t)\}}s_+(x,t)|\omega(x,t)|^2\,dx \leq\theta\nu Z(t) \quad\text{for almost every }t<T_*. \end{equation} The complementary region contributes at most $k(t)Y(t)$, giving $\tfrac12Y'+(1-\theta)\nu Z\leq kY$ and hence continuation. Unlike a global $L^\infty$ strain estimate, (ns:tail) allows large strain where its vorticity-weighted contribution can be absorbed by diffusion. Proving this for arbitrary initial data is an explicit missing geometric estimate; it is an additional assumption, not a consequence of incompressibility or a renamed energy bound.

Counterexample route: can the force be removed?

Take a smooth forced reference flow $U$ and an unforced solution $v$ on a common regular interval, with equal initial data. Put $w=v-U$ and let $f$ be the force of $U$. Subtraction gives \[ \partial_tw+(U\cdot\nabla)w+(w\cdot\nabla)U+(w\cdot\nabla)w =\nu\Delta w-\nabla q-f. \] The two transport terms vanish in the $L^2$ product with $w$, so \[ \tfrac12\frac d{dt}\|w\|_2^2+\nu\|\nabla w\|_2^2 \leq\|\nabla U\|_\infty\|w\|_2^2+\|f\|_2\|w\|_2. \] Regularizing the norm at zero and applying Gronwall proves \begin{equation}\label{wsref-0-6e-73-3a-66-6f-72-63-65} \|w(t)\|_2\leq \int_0^t\|f(s)\|_2 \exp\left(\int_s^t\|\nabla U(r)\|_\infty\,dr\right)ds. \end{equation} The estimate gives no uniform control near a singular reference trajectory when the accumulated gradient diverges. Even a small $L^2$ difference would not preserve unbounded pointwise velocity, because concentrations can have arbitrarily small $L^2$ norm. Also, setting $f$ equal to a residual of a chosen ansatz must establish the prescribed force regularity; it does not make that residual zero. Thus deleting the force in [S4] is not a justified construction of a counterexample to (A).

Verification, remaining work, and outcome.

The signs in the curl identity, the powers of $\nu$ in Young's inequality and the local lifespan, and the scaling of $E,Y$ were checked directly. The zero-datum and $Z=0$ cases were separated before division. The scalar blowup example satisfies only the inequalities claimed for it. No finite numerical simulation or check of a few Fourier modes could establish the all-data, all-time conclusion, and none is represented as doing so.

The strongest proved conclusion is the stated scale-invariant small-data subcase, together with the positive-strain and strain-tail continuation implications. For general Schwartz data the first missing step in this proof route is an a priori bound such as (ns:tail), or another estimate controlling a critical continuation norm before the maximal time. The energy identity and (ns:stretch) cannot supply it by themselves.

Three concrete next targets are a vorticity-alignment estimate implying (ns:tail); a bound on $L^\infty_tL^3_x$ permitting use of [E2]; and a stability estimate for a singular construction in a norm strong enough to transfer blowup while removing the force. Any candidate counterexample must satisfy the actual unforced equation, have Schwartz initial data and finite energy, and evade the continuation criteria above.

UNRESOLVED. No all-data regularity theorem or unforced singular solution is obtained. Written by GPT-6 Astra Ultra from the exact catalogue target, with primary-source scope checks and direct derivations; the argument is an independently written research attempt, not a verification of the announced forced proof.

Sources

• S1 Charles L. Fefferman, Existence and Smoothness of the Navier-Stokes Equation, official Clay problem description. https://www.claymath.org/wp-content/uploads/2022/06/navierstokes.pdf. Formulation source. Consulted 24 September 2026: Statement (A), page 2.
• S2 On the Navier–Stokes Millennium Prize Problem. https://openai.com/index/navier-stokes-solution/. Further reference; not a proof endorsement.
• S3 Navier-Stokes Announcement — Clay Mathematics Institute. https://www.claymath.org/news/navier-stokes-announcement/. Further reference; not a proof endorsement.
• S4 On the Navier–Stokes Millennium Prize Problem — analytical paper. https://cdn.openai.com/pdf/32d9f210-8b73-45e0-91bc-82a30aef8a9a/navier-stokes.pdf. Further reference; not a proof endorsement.
• E1 Tosio Kato, Strong $L^p$-solutions of the Navier--Stokes equation in $\mathbb R^m$, with applications to weak solutions, Mathematische Zeitschrift 187 $1984$, 471--480. https://doi.org/10.1007/BF01174182. Local strong-solution framework; publisher record checked 27 September 2026. The $L^6$ continuation estimate used here is displayed in the argument.
• E2 L. Escauriaza, G. A. Seregin, and V. Šverák, $L_{3,\infty}$-solutions of the Navier--Stokes equations and backward uniqueness, Russian Mathematical Surveys 58 $2003$, 211--250. https://www.mathnet.ru/eng/rm609. Endpoint regularity theorem; journal record and abstract checked 27 September 2026. Here $L_{3,\infty}$ denotes $L^\infty$ in time with values in spatial $L^3$, not weak spatial $L^3$.
• E3 Terence Tao, Finite time blowup for an averaged three-dimensional Navier--Stokes equation, Journal of the American Mathematical Society 29 $2016$, 601--674. https://arxiv.org/abs/1402.0290. Primary source for the averaged-nonlinearity obstruction; consulted 27 September 2026. It concerns a modified equation and is not an unforced Navier--Stokes counterexample.