betaFeedback

WorksMathematics

Quantitative global regularity for Navier–Stokes equations with odd viscosity on the cubic three-torus

Abstract

We study the three-dimensional incompressible Navier–Stokes equation on the cubic torus with a constant odd-viscosity operator given by the vertical derivative of the curl. Arbitrary real, mean-zero, divergence-free initial data and an independent steady force, both in the second-order Sobolev space, admit a unique global strong solution when the magnitude of one fixed odd-viscosity coefficient is at least a norm-uniform sufficient threshold. In the unforced case the solution is smooth at positive times. The qualitative argument adapts the resonant-averaging framework for rapidly rotating fluids to the different odd-viscous dispersion. Weighted lattice counts and signed corrected-energy estimates yield a sufficient bound for the odd-to-ordinary viscosity ratio that is polynomial in the normalized initial-data radius at each fixed normalized force radius, together with a separate bound in a specified force-dominated regime. No optimality is claimed for the polynomial exponents or prefactors; determining the sharp viscosity threshold for the original equation remains open. Direct kinetic and weighted enstrophy barriers also give regularity for every coefficient on explicit bounded data and force regions, including the closed unforced ball of homogeneous second-order Fourier Sobolev radius ninety-seven fortieths times the viscosity and seven closed forced rectangles. We specify how finite certificates enter the analytic argument and distinguish minimization of a sufficient formula from sharpness for the original equation. All three main theorems are formally verified in Lean 4 with Mathlib.

main.pdf

The equation and the main results

Odd viscosity is a nondissipative part of the constitutive relation between stress and velocity gradients. Its viscosity tensor is antisymmetric under exchange of the two index pairs, so its contraction with the velocity gradient does no local viscous work. Broken time-reversal symmetry allows this response. Microscopic models of aligned spinning constituents produce odd-viscous stresses in three dimensions, with anisotropic bulk wave propagation [24]. This provides a physical origin distinct from describing an ordinary fluid in a rotating frame: internal angular momentum can enter a material’s transport coefficients.

The effect of this transport on nonlinear flow motivates the present problem. De Wit and collaborators study a three-dimensional odd-fluid model in which nondissipative transport reorganizes turbulent cascades and selects an intermediate pattern scale [9]. Subsequent work investigates intermittency suppression and odd-wave turbulence [5, 10, 11]. These results concern flow organization and statistical transfer. They motivate a complementary deterministic question: can a sufficiently large, constant odd viscosity ensure global strong regularity uniformly over prescribed balls of initial data and forcing, even though it adds no energy dissipation?

We address that question for the one-coefficient, constant-density periodic model below. The linear waves and the constitutive model are already present in the physical literature. Our quantitative results concern the nonlinear equation with ordinary positive viscosity, complete Sobolev balls and an independent steady force. More general three-dimensional odd-viscosity tensors have additional coefficients and require a separate analysis [18, 19].

Fix T3=(R/2πZ)3\mathbb{T}^{3} = (\mathbb{R}/2\pi\mathbb{Z})^{3} and normalized Haar measure. For a mean-zero vector field write

u(x)=∑k∈Z3∖{0}u^keik⋅x,∥u∥s2=∑k≠0∣k∣2s∣u^k∣2.u(x) = \sum_{k\in\mathbb{Z}^{3}\setminus\{0\}} \widehat{u}_{k}e^{ik\cdot x}, \qquad\|u\|_{s}^{2} = \sum_{k\ne0}|k|^{2s}|\widehat{u}_{k}|^{2}.

Let P\mathbb{P} be the orthogonal Leray projector. We consider the original equation

∂tu+P((u⋅∇)u)=νΔu+αLu+f,L=P(e3×Δ)=∂3curl⁡,u(0)=u0.(1)\partial_{t}u+\mathbb{P}((u\cdot\nabla)u)=\nu\Delta u+\alpha Lu+f,\qquad L=\mathbb{P}(e_{3}\times\Delta)=\partial_{3}\operatorname{curl},\qquad u(0)=u_{0}. \tag*{(1)}

Here ν>0\nu>0 and α∈R\alpha\in\mathbb{R} are constant in time. Both u0u_{0} and ff are real, mean-zero, divergence-free fields. The force is steady and is chosen independently of α\alpha. The data and force constraints are independent full balls, ∥u0∥2≤M\|u_{0}\|_{2}\le M and ∥f∥2≤F\|f\|_{2}\le F. Write μ=M/ν\mu=M/\nu and ϕ=F/ν2\phi=F/\nu^{2} for the normalized radii.

The operator LL distinguishes odd viscosity from Coriolis forcing. The physical model of [9, Eq. (1)] contains e3×Δue_{3}\times\Delta u; on divergence-free fields its Leray projection is exactly ∂3curl⁡u\partial_{3}\operatorname{curl}u. Odd viscosity enters through gradients in the stress and gives an order-two term in the momentum equation. Its coefficient α\alpha has the same physical dimensions as the ordinary kinematic viscosity ν\nu. A Coriolis term ΩcP(e3×u)\Omega_{\mathrm{c}}\mathbb{P}(e_{3}\times u) is an order-zero rotating-frame force; here Ωc\Omega_{\mathrm{c}} includes the conventional factor two multiplying the angular velocity. The two projected operators share helical eigenvectors, but their signed frequencies are, up to the choice of helicity sign,

ω±odd(k)=±αk3∣k∣,ω±c(k)=±Ωck3∣k∣.\omega_{\pm}^{\mathrm{odd}}(k)=\pm\alpha k_{3}|k|,\qquad\omega_{\pm}^{\mathrm{c}}(k)=\pm\Omega_{\mathrm{c}}\frac{k_{3}}{|k|}.

The extra factor ∣k∣2|k|^{2} changes the dependence on spatial scale and the fully fast resonance set. No single change of the rotation rate identifies the two equations.

For the chosen periodic model, LL is skew-adjoint and commutes with the Laplacian. Thus it contributes zero to every scalar Fourier Sobolev energy, while the ordinary viscosity supplies the dissipation. The large-coefficient argument exploits oscillation and resonant cancellation in the nonlinear transfer. The bounded regions below instead use direct energy barriers and include α=0\alpha=0. Despite the physical distinction between the two operators, resonant averaging for rapidly rotating fluids supplies the qualitative proof framework.

Theorem 1.1 (Norm-uniform global regularity). For every ν>0\nu>0 and M,F≥0M,F\ge0, there is a finite nonnegative Rgen(M/ν,F/ν2)\mathcal{R}_{\mathrm{gen}}(M/\nu,F/\nu^{2}) such that every datum and force just described and every fixed real coefficient satisfying

∣α∣/ν≥Rgen(M/ν,F/ν2)|\alpha|/\nu\ge\mathcal{R}_{\mathrm{gen}}(M/\nu,F/\nu^{2})

give a unique global strong solution

u∈C([0,∞);Hsol2)∩Lloc2([0,∞);H3).u \in C([0,\infty);H^2_{\mathrm{sol}})\cap L^2_{\mathrm{loc}}([0,\infty);H^3).

The solution has uniform H1H^1 and H2H^2 bounds and bounded H3H^3 square integrals on every sliding interval of any prescribed finite length. For f=0f=0, every initial HsH^s regularity with s≥2s\ge2 persists and the solution is smooth at positive times. The conclusion includes coefficient equality and every larger magnitude, with either sign.

Theorem 1.1 is the qualitative foundation for the quantitative statements below. Its proof adapts resonant averaging to the odd-viscous multiplier and includes the independent steady force. The additional weighted arithmetic and signed-energy estimates give the sufficient bounds in Theorem 1.2. The complete analytic deductions are supplied in the accompanying Full proof supplement for Theorem 1.2. The direct energy argument in Theorem 1.3 gives a zero sufficient threshold on the stated bounded regions.

Theorem 1.2 (Quantitative regularity bounds). At each fixed finite normalized force radius ϕ≥0\phi\ge0 and each ϵ>0\epsilon>0, there is Cϕ,ϵ<∞C_{\phi,\epsilon}<\infty for which a sufficient choice in Theorem 1.1 is

Rgen(μ,ϕ)≤Cϕ,ϵ(1+μ)9876/719+ϵ.(2)R_{\mathrm{gen}}(\mu,\phi)\le C_{\phi,\epsilon}(1+\mu)^{9876/719+\epsilon}. \tag*{(2)}

The joint large-force bound uses the earlier slow histories, signed static/timed metric estimates, and total-phase quartic estimates proved in § 3.8 and the full proof supplement. Set X=ee+μX=e^e+\mu, R=X+ϕR=X+\phi, and U=X1719/719U=X^{1719/719}. There are absolute constants C,C0,R0>0C,C_0,R_0>0 such that, when R≥max⁡(R0,X24)R\ge\max(R_0,X^{24}), a sufficient choice is

Rgen(μ,ϕ)≤CUR3exp⁡(C0log⁡Rlog⁡log⁡R).(3)R_{\mathrm{gen}}(\mu,\phi)\le CUR^3\exp\left(C_0\frac{\log R}{\log\log R}\right). \tag*{(3)}

The native near estimate and all-frequency profiles are derived below. The group estimate, complete joint-input derivations, and both full assemblies are given in the full proof supplement. Both exact clauses are covered by the recorded complete-target mathematical admission. These are sufficient bounds, not sharpness claims.

Theorem 1.3 (Closed zero-threshold regions). Theorem 1.1 holds for every fixed real α\alpha, including zero, in each of the following regions. In the unforced case the region is 0≤μ=M/ν≤97/400\le\mu=M/\nu\le97/40. In the forced case it is the union of the seven closed rectangles below, with independent coordinate caps.

208013 μmax⁡208013\ \mu_{\max}208013 ϕmax⁡208013\ \phi_{\max}JJ
419650419650419650419650850850
425500425500382950382950294294
434000434000347200347200468468
442000442000309400309400320320
458500458500229250229250808808
478500478500119625119625355355
4895004895004895048950124124

Table 1.

For a forced row put s=μmax⁡s=\mu_{\max} and write u(t)=νsw(νt)u(t)=\nu s w(\nu t). With the metric hh defined below and Z=⟨(−Δ)h(−Δ)w,w⟩Z=\langle(-\Delta)^h(-\Delta)w,w\rangle, the actual normalized solution satisfies ∫IZ dτ≤J\int_I Z\,d\tau\le J on every interval ∣I∣≤1|I|\le1. The unforced scaled solution has Y≤1Y\le1 and ∫0∞Z dτ≤10000\int_0^\infty Z\,d\tau\le10000. All datum, force, graph, and endpoint equalities are included.

Contributions and proof strategy

1.1. Contributions and proof strategy. Theorem 1.1 establishes the large-odd-viscosity conclusion for the original equation, uniformly over the specified datum and steady-force balls. The proof verifies the applicability of the Babin–Mahalov–Nicolaenko resonant-regularity method and the parabolic filtering argument to this constitutive operator [1] [15]. The slow modes form the familiar two-dimensional velocity and passive vertical-component system; mixed resonant terms preserve the fast Sobolev energy. For the fully fast interactions, eliminating the signed square roots gives a nonzero degree-six polynomial in the vertical lattice coordinate. Its bounded number of roots yields the restricted-convolution estimate. The subsequent finite-cutoff averaging, smoothing and uniform restart follow the established framework, whose applicability is thereby checked for the different odd-viscous dispersion.

The specialized quantitative estimates enter Theorem 1.2. The weighted resonant count retains the physical interaction coefficient and has exponent (17−3)/2(\sqrt{17}-3)/2, with an arbitrarily small positive loss. Its proof uses integer surface coordinates, exceptional-fibre analysis and published coefficient-uniform counting theorems. Separate estimates control all frequencies near resonance, rather than only a fixed finite set of nonzero phases. Set q=∣α∣/νq = |\alpha|/\nu. With the definitions in the proof, their two kernel profiles satisfy

N(q)≤Cq−1/2E(q),P(q)≤Cq−1/3E(q),E(q)=exp⁡(C0log⁡qlog⁡log⁡q)N(q) \le Cq^{-1/2}E(q), \qquad P(q) \le Cq^{-1/3}E(q), \qquad E(q) = \exp\left(C_0\frac{\log q}{\log\log q}\right)

for sufficiently large qq. Signed viscous corrections at both energy weights retain the physical cancellations and every differentiated source. The time-integrated slow-mode bounds are obtained before the fast high-norm estimate. Together with the actual fast kinetic bound and the total-phase quartic estimates, they keep the datum and force dependence separate through continuation. These are the additional quantitative arguments behind the fixed-force polynomial law and the separate force-dominated law. The latter retains its factor UU and its stated domain; neither clause asserts an endpoint or a sharp PDE threshold.

Theorem 1.3 gives an independent quantitative refinement of the classical small-data mechanism. Its weighted cubic bound uses the full physical output matrix, cyclic interaction weights and analytic frequency tails. Conditional estimates retain the coupled Fourier moments of one actual field, including harmonic and angular deficits and first- and second-shell constraints. They support the unforced kinetic graph and selected faces of the common forced graph. Forcing is tested against each joined energy normal through an all-shell spectral support inequality. The other six forced rectangles use the whole-cubic estimate with their coupled spectral graphs. Finite certificates prove complete matrix, moment and barrier inequalities; they do not represent samples of initial conditions. Classical invariant-region and continuation arguments then give the original global solution and the every-time bounds. The main quantitative content lies in these explicit certified regions and the estimates that establish them.

Figure 1 summarizes the three proof routes.

Three routes to the main results

Figure 1. Three routes to the main results. (a) Filtering gives a resonant reference and nonresonant error terms. Reference regularity is used in the error estimates, as indicated by the vertical arrow. Uniform stability and continuation give Theorem 1.1; the energy input includes the fixed-horizon criterion in the unforced case and absorption and restart in the forced case. (b) Weighted arithmetic and signed corrected energies provide the quantitative control leading to Theorem 1.2, including its fixed-force polynomial bound and its distinct joint-force regime. (c) Weighted moments and finite certificates give invariant regions and window bounds for Theorem 1.3. The 41-edge graph belongs to the unforced bounded-data argument; the forced rectangles have their own certified plans. The diagram groups analytic ingredients and does not represent an exhaustive dependency graph or formalization status.

Three-dimensional odd fluids support wave and flow phenomena beyond the rotating-frame analogy. Khain and collaborators analyze parity-violating stresses in Stokes flow [18, 19]; Kirkinis and Olvera de la Cruz study Taylor columns and inertial-like waves, including a model with both rigid rotation and odd viscosity [20, 21]. The turbulent-cascade and odd-wave studies cited above already identify the helical waves and their three-wave resonances. Their statistical or weak-wave descriptions do not give the deterministic norm-uniform strong-solution bounds considered here.

Mathematical well-posedness for odd viscosity also includes variable-density planar equations. Fanelli, Granero-Belinchón and Scrobogna prove local Sobolev well-posedness using good unknowns [12]. Effective-velocity formulations give further Besov and density-law results [13, 29]. These transported-density systems have different dimensions and constitutive structure from (1). Odd-viscous linear instability of channel shear flows [25] concerns walls and a different solution property. Our regularity conclusions do not imply laminarization, linear stability of every flow, or a prescribed turbulent spectrum.

Resonant averaging and quantitative rotation bounds

Babin, Mahalov and Nicolaenko prove global regularity for rapidly rotating Navier–Stokes equations on fixed periodic domains, including resonant domains [1, Theorem 1.1]. Their slow/fast splitting, catalytic cancellation, algebraic resonance counting, restricted convolution and all-time transfer are the closest predecessors of the qualitative proof here. Their primitive-equation work develops the related fast-oscillation framework [2]. Gallagher’s parabolic adaptation of Schochet’s method supplies the finite-time filtering mechanism [15, Theorem 1]. Theorem 1.1 applies this method after verifying the operator-specific resonance and forcing estimates. BMN’s printed Coriolis theorem is not itself a theorem about odd viscosity.

Polynomial sufficient rotation conditions also precede this work. Iwabuchi and Takada obtain a polynomial condition on entire subcritical Sobolev balls for the unforced Coriolis equation on R3\mathbb{R}^{3} [17, Theorem 1.1]. Fujii studies large data in critical Fourier–Besov spaces and states a quantitative condition with additional subcritical regularity [14, Theorems 1.1–1.2]. On the cubic torus, Kishimoto and Yoneda obtain polynomial solution bounds, while their displayed sufficient rotation magnitude is exponential in a power of the datum radius [22, Theorem 1.3]. Eqs. (1.5)–(1.6)]. Cheng and Sakellaris construct globally regular near-resonant approximations and transfer them to the original rotating equation above a norm-dependent threshold [6, Theorem 1.6]. Large solutions at a fixed Coriolis parameter can also follow from special datum structure and small linear self-interaction [23, Theorem 1.3 and Corollary 1.4]. Whole-space rotating theory also treats external forcing [31]. The differing operators, domains, norms and forcing hypotheses prevent a numerical ranking of those sufficient exponents against Theorem 1.2.

The arithmetic inputs used here are established counting results: the coefficient-uniform rectangular curve bound [16, Theorem 15], in the form of [3 Eq. (3)], and the affine-surface theorem at modulus one [4, Theorem 1.1]. The additional argument is their application to the physical weighted odd-viscous resonance geometry, followed by the complete near-resonant, signed-metric and source estimates. The theorem does not introduce a new general integer-point method or claim the first polynomial large-coefficient regularity criterion.

Small-data bounds and rigorous computation

Explicit periodic small-data constants, small-force global conditions and control inequalities are developed by Morosi and Pizzocchero [26, 27, 28]. Their Fourier estimates already combine finite lattice sums with analytic tails. Chernyshenko, Constantin, Robinson and Titi establish residual-based a posteriori regularity tests [7]. Robinson and Sadowski formulate numerical verification over entire bounded sets of initial data, including H2H^2 balls [30, Section 5]. Thus neither quantitative small-data regularity nor computer-assisted verification of a full ball is new in principle.

Theorem 1.3 supplies particular certified regions through physical cubic estimates and correlated spectral moments, rather than through trajectories covering the datum ball. The complete supports, rounding allowances, closed moment regimes and analytic complements are part of the proof. Checking a sampled maximizer would not suffice. Its all-coefficient conclusion follows from skew-adjoint cancellation and includes ordinary Navier–Stokes at α=0\alpha= 0. No largest-radius or original-PDE optimality claim is made.

Formal verification

Theorems 1.1, 1.2 and 1.3 have been formalized in the proof assistant Lean 4 with the Mathlib library [8, 32]. Both bounds of Theorem 1.2 and both cases of Theorem 1.3 are included. The formal theorems have no hypotheses beyond the conditions stated above, and their proofs contain no gaps. The Lean kernel accepts them using only the three standard axioms: propositional extensionality, choice and quotient soundness.

The Lean sources, encoded numerical certificates, and build instructions are publicly available at https://github.com/DavetheSlayer/OVNS-Lean. For reproducibility, we refer to source snapshot 9bd811d.

One file states the five formal theorems. It imports only Mathlib and adds eight definitions: the Sobolev norm, admissible fields, two time-regularity classes, strong solutions, smoothness at positive times, the global conclusion of Theorem 1.1, and the metric hh. A second file repeats those definitions and statements and proves them from a library of about 143,000 hand-written lines and about 2,800 generated modules. A term-by-term comparison of those definitions and statements in the two compiled files finds no difference. As in this paper, a field is its family of Fourier coefficients and a solution is defined by the mode equations, so the formal theorems contain no separate statement about functions on the torus.

The finite computations in the formal proof of Theorem 1.3 run inside the kernel, with no external oracle. The unforced graph and the six simpler forced rectangles use the certificates of this paper. The row bound and the moment pair comparisons that the formal proof needs are computed in it, and the common forced rectangle uses a replanned graph; see Remarks 19.11 and 19.12. The formal proof of Theorem 1.2 assumes neither of the published counting theorems cited in its arithmetic. The plane-curve count is proved, and a specialized determinant argument replaces the surface theorem for the one family that occurs.

The formal proof differs from the proofs written here at several points. Remarks 18.1, 19.11 and 19.12 record those for the fixed-force bound and for Theorem 1.3. The proof of Theorem 1.1 and the joint bound also take their own routes at several steps, for example in the slow estimates. A different route does not affect the proofs in this paper. The formalization verifies the three theorems, but not every intermediate estimate or constant printed here; nor does it cover the theorem of the section on a finite positive sufficient ratio. Where the formalization exposed an error or a missing condition, the text below has been corrected.

The formal statements of Theorems 1.2 and 1.3, and the definitions they share with Theorem 1.1, were compared with this manuscript in separate AI-assisted audits. No external human review of the formal development has taken place. The formalization establishes the stated sufficient conditions; it does not address the sharpness and applicability questions below.

Remark 1.4. The restriction R≥max⁡(R0,X24)R \ge\max(R_0, X^{24}) in Theorem 1.2 concerns the radii, not the data, since a force with ∥f∥2≤F\lVert f\rVert_2 \le F also lies in every larger ball. Applying (3) at the larger radius gives, for all μ\mu, ϕ≥0\phi\ge0, the sufficient choice

Rgen(μ,ϕ)≤CUR~3exp⁡(C0log⁡R~log⁡log⁡R~),R~=max⁡(R,R0,X24).\mathcal{R}_{\mathrm{gen}}(\mu,\phi) \le CU\widetilde{R}^{3}\exp\left(C_0\frac{\log\widetilde{R}}{\log\log\widetilde{R}}\right), \qquad\widetilde{R}=\max(R,R_0,X^{24}).

On the stated domain R~=R\widetilde{R}=R. This form is also verified in Lean.

Normalization and the role of the bounds

Set τ=νt\tau=\nu t, v(τ)=u(τ/ν)/νv(\tau)=u(\tau/\nu)/\nu, g=f/ν2g=f/\nu^2, r=α/νr=\alpha/\nu, q=∣r∣q=\lvert r\rvert, μ=M/ν\mu=M/\nu and ϕ=F/ν2\phi=F/\nu^2. Then

vτ+P((v⋅∇)v)=Δv+rLv+g.(4)v_\tau+\mathbb{P}((v\cdot\nabla)v)=\Delta v+rLv+g. \tag*{(4)}

All sufficient ratios refer to this fixed cubic normalization. A lower bound for a particular corrected-energy allocation concerns that allocation. The PDE sharpness and experimental applicability questions remain open.

Notation and constants

We begin with the analytic foundation for the unforced problem. Until the forced construction is introduced, the equation considered in these sections is

∂tu+P((u⋅∇)u)=νΔu+αLu,f=0.(5)\partial_tu+\mathbb{P}((u\cdot\nabla)u)=\nu\Delta u+\alpha Lu,\qquad f=0. \tag*{(5)}

References

  1. [1]T. D. Browning and D. R. Heath-Brown, Plane curves in boxes and equal sums of two powers, Mathematische Zeitschrift 251 (2005), no. 2, 233–247.
  2. [2]Tim Browning and Matteo Verzobio, Counting integer points on affine surfaces with a side condition, Discrete Analysis (2025), no. 12, 25 pp., Version 2.DOI
  3. [3]D. R. Heath-Brown, Counting rational points on algebraic varieties, Analytic Number Theory (A. Perelli and C. Viola, eds.), Lecture Notes in Mathematics, vol. 1891, Springer, Berlin, 2006, pp. 51–95.

Paper details

vlnc:W_7ev9e87y
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Claude, GPT-6.1, Grok (multiple versions)

View statement
Looking for related papers…
Contents