Configuration-Space Curvature and the Navier–Stokes Singular Set

Two programmes in mathematics have developed alongside one another for sixty years without quite meeting. The first, initiated by Arnold in 1966, treats an ideal incompressible fluid as a geodesic on the infinite-dimensional Riemannian manifold of volume-preserving diffeomorphisms and reads the instability of the flow from the curvature of that manifold. The second, developed by Scheffer and by Caffarelli, Kohn and Nirenberg through the 1970s and 1980s, bounds the singular set of a viscous fluid by asking when the energy of the flow concentrates in small parabolic regions. The preprint described in this article shows that the two programmes meet through a single operator identity: the curvature of the configuration group is the pressure Hessian, the nonlocal transform that drives the dynamics of the velocity gradient. Read against the partial-regularity theory, this identity places the Navier–Stokes singular set where the curvature concentrates or the pressure depletes.

1. The configuration group of an ideal fluid

The configuration space of an ideal incompressible fluid on a closed Riemannian manifold is the group

of smooth diffeomorphisms that preserve the volume form . Demanding that the diffeomorphism preserve volume is the same as demanding that the fluid be incompressible: no region is compressed or expanded. A configuration of the fluid is a placement of every fluid particle, so a path in is a motion of the fluid.

The group carries a natural Riemannian metric. At the identity its tangent space is the Lie algebra of smooth divergence-free vector fields on , and the natural inner product there is the pairing

Extending this by right-invariance produces the right-invariant metric on . This is not an arbitrary choice: the right-invariance says that the kinetic energy of the fluid is invariant under relabelling the particles, a fundamental symmetry of an ideal fluid. Arnold's observation^[1] is that the geodesics of this metric are exactly the solutions of the Euler equations of ideal fluid mechanics.

The pressure appears here as a Lagrange multiplier enforcing the incompressibility constraint, exactly as in classical mechanics: it is the normal reaction force that keeps the motion on the submanifold inside the flat ambient group of all diffeomorphisms.

For the incompressible equations on the flat torus , we work at the Lie algebra level. The Levi-Civita connection of the metric is the ordinary directional derivative followed by the Leray projection onto the divergence-free part:

The complementary projection onto gradients is what the incompressibility constraint removes from each infinitesimal step. The pressure is what produces.

2. Sectional curvature and the role of pressure

The curvature tensor of the Levi-Civita connection is

Arnold^[1] computed the sectional curvature for smooth divergence-free pairs and found it predominantly negative. For Fourier mode pairs with wavevectors and , the sectional curvature is

(the precise formula involves the projection), and it is negative unless , in which case it vanishes; Arnold called this the same-shell condition. The sign is carried entirely by the pressure: the Fourier-mode formula resolves into the negative energy of a pressure gradient associated to the pair.

Negative sectional curvature has a direct dynamical interpretation. By the Jacobi equation, it forces exponential divergence of neighbouring geodesics: two nearby Euler flows starting close together in configuration space separate exponentially fast. This is the Lagrangian instability of the ideal flow. Rouchon (1992)^[3] and Misiołek (1993)^[4] established that this geometric instability is equivalent to the Eulerian instability of the velocity field; Preston (2004)^[5] placed the equivalence on a rigorous footing for all incompressible configurations.

The curvature is thus a measure of how violently the incompressibility constraint couples nearby fluid motions. Its sign and size are carried by the pressure.

3. The pressure Hessian in viscous dynamics

For the viscous problem on , the velocity gradient evolves by

obtained by differentiating the Navier–Stokes momentum equation and using incompressibility to identify the pressure term. The three contributions are the self-stretching , the pressure Hessian , and the viscous term.

The self-stretching is local: it depends only on the velocity gradient at the same point. If the pressure Hessian were absent, this equation would be the Vieillefosse–Cantwell restricted Euler system,^[6][7] which blows up in finite time from generic smooth initial data. The pressure Hessian is what prevents that blowup, but it does so nonlocally.

The pressure is determined from the flow by the Poisson equation

where is the strain tensor and is the vorticity. The quantity is the strain–enstrophy imbalance. Inverting the Laplacian and differentiating twice gives

where is the matrix Riesz transform with Fourier symbol . The pressure Hessian is a nonlocal zeroth-order Fourier multiplier applied to : its value at a point depends on the distribution of over the entire torus.

This nonlocality is why local models of the velocity-gradient dynamics are inadequate. Every local closure, ignoring the pressure Hessian, discards the information the configuration-space curvature carries, which is precisely what the next section makes explicit.

4. The main identity

The connection between the curvature of and the pressure Hessian is established through the Gauss equation for the embedding

The ambient group of all smooth diffeomorphisms carries a flat metric: its Riemannian curvature vanishes. The curvature of the submanifold is therefore entirely accounted for by its shape inside the flat ambient space, by the second fundamental form of the embedding.

For the embedding of inside , the second fundamental form is the component of the ambient acceleration orthogonal to , that is, the gradient part produced by the incompressibility projection:

The second fundamental form of the incompressibility constraint is the pressure gradient. The incompressibility constraint curves inside the flat ambient group, and the shape of that curving is carried by the pressure.

The Gauss equation then gives the curvature of from the shape operator.

The first term, , is the Leray-projected pressure Hessian applied to . This is the same operator that drives the velocity-gradient evolution equation: the configuration-space curvature and the pressure Hessian of the vortex-stretching dynamics are one object.

The second term involves the mixed pressure , which depends on the pair antisymmetrically. On computing the sectional curvature , the mixed-pressure term integrates to zero by incompressibility and the identity vanishes, recovering Arnold's sectional formula. The operator identity carries more information than the sectional curvature: it identifies the curvature as a specific operator on , not just a scalar assigned to two-planes.

The identity is established by a direct computation through the Gauss equation, working in Fourier space to handle the Leray projection. The identities used are verified independently by symbolic and spectral computation, available in the verify/ directory of the companion repository.

5. Measuring the curvature: the strain-enstrophy imbalance

The main identity identifies the curvature operator qualitatively. The next step is to measure its size.

From the Poisson equation , the pointwise trace of the pressure Hessian is

The local trace of the curvature is therefore the strain–enstrophy imbalance . The curvature concentrates where strain and enstrophy are out of balance and vanishes on the set of pressureless flows.

The global size follows from the Riesz isometry.

The proof is immediate: because Riesz transforms are isometric on . The physical content is that the norm of the nonlocal pressure Hessian (the size of the operator driving the velocity-gradient dynamics) equals the norm of the local pointwise imbalance between strain energy and enstrophy.

Arnold's same-shell degeneracy is a special case. For a Fourier mode pair on the same shell , the imbalance associated to that pair is zero, so the curvature vanishes, recovering Arnold's result from the more general identity.

The total imbalance over the torus vanishes at every time,

because the identity holds for every divergence-free field on a closed manifold (by integration by parts and incompressibility). The imbalance is produced locally and cancelled locally; it is redistributed by the dynamics but has no global source.

6. The Caffarelli–Kohn–Nirenberg partial-regularity theory

We now turn to the regularity theory of the viscous problem, against which the identity will be read.

A Leray–Hopf weak solution of the Navier–Stokes equations on is a divergence-free field satisfying the equations in the sense of distributions and the global energy inequality. Leray (1934)^[8] and Hopf (1951)^[9] showed such solutions exist for all time from any initial datum. Whether every smooth initial datum produces a smooth solution for all time is the Clay Millennium Problem.

Where global smoothness is open, the partial-regularity theory bounds the set of potential singularities. A point in spacetime is regular if the solution is locally bounded near it; the singular set is the complement. Scheffer (1976)^[10] and Caffarelli, Kohn and Nirenberg (1982)^[11] proved that has parabolic Hausdorff dimension at most one: it cannot contain a curve in space at a fixed time, and its one-dimensional parabolic measure is zero.

The central tool is the following criterion.

A suitable weak solution is one satisfying a local energy inequality, a mild additional condition on the Leray–Hopf class that is satisfied, for example, by solutions constructed by appropriate mollification. The -regularity criterion says: a point is regular if and only if the scaled dissipation (the average of over a backwards parabolic ball, normalised by the ball's radius) can be made small at all sufficiently small scales.

7. Curvature concentration and the singular set

The pointwise identity

(since and ) inserts the curvature into the -regularity criterion. Whenever the curvature is small, the full gradient energy is controlled by the enstrophy alone.

The proof is a direct consequence of the -regularity criterion. Suppose the curvature does not concentrate, so . From the gradient split, , and since the second term vanishes, . Since , the scaled strain energy and half-enstrophy differ by , giving the claimed balance.

The companion coercive statement goes the other direction: if the flow remains imbalanced near the singular point (meaning for all small and some fixed ), then non-concentrating curvature forces regularity, because the curvature controls the full gradient energy up to the factor .

Together: away from balance, the curvature governs regularity. At a singular point with bounded curvature, the flow must enter strain–enstrophy balance; equivalently, the pressure depletes locally, with in scaled average.

The transport law. The imbalance is not a passive scalar. Differentiating the Navier–Stokes equation to get the evolution of and tracing gives

The three source terms are: the inertial production (the determinant of the velocity gradient, a null Lagrangian that integrates to zero over ), the curvature work (which also integrates to zero by incompressibility and the divergence theorem), and the viscous redistribution. Each term carries zero net integral, so the total imbalance is conserved and each term acts locally with no global source. A singularity approached through balance requires these three terms to cancel pointwise up to the singular time; the next section makes this cancellation rigid.

8. Excluding balanced singularities in the scale-critical class

The dichotomy leaves one regime the curvature does not detect: the balanced regime where in scaled average and the pressure depletes. We show this regime is rigid in the scale-critical class.

Blow-up limits. The parabolic rescaling at is

This preserves the Navier–Stokes equations and rescales the scaled dissipation as , and similarly for .

If the solution obeys a Type-I bound at , meaning on a backwards neighbourhood, then uniformly in . The -regularity theory propagates this to uniform local bounds on all derivatives, and the blow-up procedure of Seregin and Šverák^[12] extracts a subsequence converging in to a bounded smooth ancient solution on .

The pressureless condition follows directly from the balance hypothesis: scaling shows for every , so pointwise. Then , and since the pressure has no spatially harmonic part (it is recovered from via the Poisson equation), and the momentum equation reduces to a heat equation with divergence-free drift.

Type-I exclusion. The pressureless system is a uniformly parabolic equation with divergence-free drift and no zeroth-order term. The maximum principle applies to each component.

For each component, the maximum principle on strips shows is non-increasing and is non-decreasing. The Type-I decay forces both to zero as , so the supremum is non-positive and the infimum is non-negative at every time, giving . Since contradicts , the balanced alternative cannot occur at a Type-I singular point.

Scale-critical extension. The Type-I argument uses the pointwise decay only to force as . The same conclusion holds without pointwise decay whenever the limit lies in the scale-critical class , through the Liouville theorem for divergence-free drifts of Seregin, Silvestre, Šverák and Zlatoš (2012).^[14]

The proof rewrites the drift term in divergence form: where is the skew-symmetric tensor potential of . The endpoint Riesz potential estimate gives , and the Liouville theorem of Seregin–Silvestre–Šverák–Zlatoš^[14] forces every bounded ancient suitable weak solution to be constant.

At a singular point satisfying the local bound

the rescaled sequence is bounded in on every fixed cylinder. The -regularity theory in the critical class upgrades this to uniform local bounds and convergence to a bounded smooth ancient solution with finite norm. The critical Liouville lemma forces this limit to be constant, contradicting the singularity lower bound.

The Escauriaza–Seregin–Šverák theorem (2003)^[13] shows that a global bound prevents any singularity. The local condition used here is strictly weaker: it requires only that the norm near the potential singularity is locally bounded. A solution could fail the global condition while satisfying the local one at individual spacetime points.

9. Beyond the critical class

The results of Section 8 close the balanced alternative in the scale-critical class. Beyond that class, the blow-up procedure still produces a bounded ancient pressureless flow with and , but the drift is now only bounded, carrying no spatial decay, and the Liouville theorem for divergence-free drifts does not apply.

The pressureless structure forces rigid algebraic constraints on the velocity gradient . Incompressibility gives , and the balance eliminates the quadratic term from the characteristic polynomial. The Cayley–Hamilton theorem then gives

The velocity gradient is either nilpotent (where , so there is no inertial production) or its eigenvalues are the three cube roots of , a fixed pattern determined entirely by a single scalar. The gradient is as constrained algebraically as a trace-free matrix with zero Frobenius norm can be.

The maximum principle supplies unconditional structure: for each component , the map is non-increasing and is non-decreasing. The spatial oscillation

is non-increasing in time, and the limit is constant if and only if its past oscillation vanishes: for each component. The Liouville question reduces to whether the past oscillation is forced to zero by the pressureless structure, and that is one number per component.

The Tao (2016)^[15] averaged-equation barrier explains why this cannot be settled by the energy identity alone. Tao constructed a bilinear operator that preserves the energy cancellation , shares the Euler operator's scaling and upper-bound estimates, and yet admits smooth finite-time blowup. Any argument that uses only the energy identity, scaling, and upper bounds for the nonlinearity applies equally to , so it cannot decide regularity for the true Euler operator . The averaging that produces discards both the Riesz identity and the self-stretching determinant : these are exactly the data the curvature carries.

The supercritical balanced regime therefore reduces to a single Liouville question: whether a bounded ancient pressureless flow is constant. The Cayley–Hamilton rigidity, the monotone oscillation structure, and the energy supersolution property (the kinetic energy satisfies

when ) are the structural data that the averaging discards. They are what a proof would need to read. The two-dimensional and axisymmetric cases are settled; the three-dimensional case is open.

10. Summary

The Gauss equation for the embedding gives the curvature of the configuration group from the shape of incompressibility. The ambient group is flat; the whole curvature is carried by the second fundamental form of the constraint, which is the pressure gradient. The curvature operator is the Leray-projected pressure Hessian. Its size is the strain–enstrophy imbalance , vanishing precisely on pressureless flows and recovering Arnold's same-shell degeneracy.

Read against the Caffarelli--Kohn--Nirenberg theory: at a singular point the curvature concentrates or the flow enters strain–enstrophy balance. Away from balance the curvature controls the local energy coercively and non-concentrating curvature forces regularity. In the scale-critical class the balanced alternative produces a bounded ancient pressureless flow whose drift is critical, and the Liouville theorem for divergence-free drifts excludes it: every scale-critical singular point, including all Type-I points, is curvature-concentrating.

Beyond the critical class the balanced regime is supercritical. The energy identity is blind there, as the Tao barrier shows. The curvature is the structure of the Euler nonlinearity that the barrier leaves visible: the Cayley–Hamilton rigidity, the monotone oscillation, and the energy supersolution are data the averaging discards. They reduce the supercritical balanced regime to whether a bounded ancient pressureless flow in three dimensions is constant; one number per component past the reach of the energy identity.

References

1. Arnold, V. I. (1966). Sur la géométrie différentielle des groupes de Lie de dimension infinie et ses applications à l'hydrodynamique des fluides parfaits. Annales de l'Institut Fourier, 16(1), 319–361. DOI
2. Ebin, D. G. and Marsden, J. E. (1970). Groups of diffeomorphisms and the motion of an incompressible fluid. Annals of Mathematics, 92(1), 102–163. DOI
3. Rouchon, P. (1992). The Jacobi equation, Riemannian curvature and the motion of a perfect incompressible fluid. European Journal of Mechanics B/Fluids, 11(3), 317–336.
4. Misiołek, G. (1993). Stability of flows of ideal fluids and the geometry of the group of diffeomorphisms. Indiana University Mathematics Journal, 42(1), 215–235. DOI
5. Preston, S. C. (2004). For ideal fluids, Eulerian and Lagrangian instabilities are equivalent. Geometric and Functional Analysis, 14(5), 1044–1064.
6. Vieillefosse, P. (1982). Local interaction between vorticity and shear in a perfect incompressible fluid. Journal de Physique, 43(6), 837–842. DOI
7. Cantwell, B. J. (1992). Exact solution of a restricted Euler equation for the velocity gradient tensor. Physics of Fluids A, 4(4), 782–793. DOI
8. Leray, J. (1934). Sur le mouvement d'un liquide visqueux emplissant l'espace. Acta Mathematica, 63, 193–248. DOI
9. Hopf, E. (1951). Über die Anfangswertaufgabe für die hydrodynamischen Grundgleichungen. Mathematische Nachrichten, 4(1–6), 213–231. DOI
10. Scheffer, V. (1976). Partial regularity of solutions to the Navier–Stokes equations. Pacific Journal of Mathematics, 66(2), 535–552. DOI
11. Caffarelli, L., Kohn, R., and Nirenberg, L. (1982). Partial regularity of suitable weak solutions of the Navier–Stokes equations. Communications on Pure and Applied Mathematics, 35(6), 771–831. DOI
12. Seregin, G. and Šverák, V. (2009). On type I singularities of the local axi-symmetric solutions of the Navier–Stokes equations. Communications in Partial Differential Equations, 34(2), 171–201. DOI
13. Escauriaza, L., Seregin, G. A., and Šverák, V. (2003). -solutions of the Navier–Stokes equations and backward uniqueness. Russian Mathematical Surveys, 58(2), 211–250. DOI
14. Seregin, G., Silvestre, L., Šverák, V., and Zlatoš, A. (2012). On divergence-free drifts. Journal of Differential Equations, 252(1), 505–540. DOI
15. Tao, T. (2016). Finite time blowup for an averaged three-dimensional Navier–Stokes equation. Journal of the American Mathematical Society, 29(3), 601–674. DOI