betaFeedback

WorksMathematics

Non-uniqueness of Leray solutions to the hypodissipative Navier-Stokes equations in two dimensions

Abstract

We exhibit non-unique Leray solutions of the forced Navier-Stokes equations with hypodissipation in two dimensions. Unlike the solutions constructed in \cite{albritton2021non}, the solutions we construct live at a supercritical scaling, in which the hypodissipation formally becomes negligible as t→0+t \to 0^+. In this scaling, it is possible to perturb the Euler non-uniqueness scenario of Vishik \cite{Vishik1,Vishik2} to the hypodissipative setting at the nonlinear level. Our perturbation argument is quasilinear in spirit and circumvents the spectral theoretic approach to incorporating the dissipation in \cite{albritton2021non}.

Introduction

We consider the two-dimensional fractional Navier-Stokes equations

{∂tu+(u⋅∇)u+∇p+Λβu=f,div⁡u=0in R2×R+(NSβ)\begin{cases} \partial_t u + (u \cdot\nabla)u + \nabla p + \Lambda^\beta u = f,\\ \operatorname{div} u = 0 \end{cases} \qquad\text{in } \mathbb{R}^{2} \times\mathbb{R}_{+} \qquad\text{(NS}_{\beta}\text{)}

where β∈(0,2]\beta\in(0,2] and Λβ=(−Δ)β/2\Lambda^\beta= (-\Delta)^{\beta/2} is the fractional Laplacian. We consider the Cauchy problem for (NSβ_\beta) with divergence-free initial velocity

u(⋅,0)=u0∈L2(R2).(1)u(\cdot,0) = u_0 \in L^2(\mathbb{R}^{2}). \tag*{(1)}

It is not difficult to adapt the seminal work [15] of Leray to demonstrate the existence of a global-in-time Leray-Hopf solution for each divergence-free initial velocity u0∈L2(R2)u_0 \in L^2(\mathbb{R}^{2}) and body force f∈Lt1Lx2(R2×R+)f \in L_t^1 L_x^2(\mathbb{R}^{2} \times\mathbb{R}_{+}). These are distributional solutions u∈Lt∞Lx2∩Lt2H˙xβ/2(R2×R+)u \in L_t^\infty L_x^2 \cap L_t^2 \dot{H}_x^{\beta/2}(\mathbb{R}^{2} \times\mathbb{R}_{+}) to (NSβ_\beta) satisfying $\lVert u(\cdot,t)-u_0\rVert_{L^2(\mathbb{R}^{2})} \to 0 as t→0+t \to0^{+} and the energy inequality for all t>0t > 0:

12∫R2∣u(x,t)∣2 dx+∫0t∫R2∣Λβ2u∣2 dx ds≤12∫R3∣u0(x)∣2 dx+∫0t∫R2f⋅u dx ds.(2)\frac{1}{2}\int_{\mathbb{R}^{2}} |u(x,t)|^{2}\,dx+\int_{0}^{t}\int_{\mathbb{R}^{2}}|\Lambda^{\frac{\beta}{2}}u|^{2}\,dx\,ds\leq\frac{1}{2}\int_{\mathbb{R}^{3}}|u_{0}(x)|^{2}\,dx+\int_{0}^{t}\int_{\mathbb{R}^{2}}f\cdot u\,dx\,ds . \tag*{(2)}

It is well known that Leray’s solutions are unique in the classical setting β=2\beta= 2. In this paper, we demonstrate non-uniqueness when β<2\beta< 2:

Theorem 1.1 (Non-uniqueness of Leray-Hopf solutions). *For all β∈(0,2)\beta\in(0,2), there exist a force f∈Lt1Lx2(R2×R+)f \in L^{1}_{t}L^{2}_{x}(\mathbb{R}^{2} \times\mathbb{R}_{+}) and two distinct Leray-Hopf solutions, uu and uˉ\bar{u}, with zero initial velocity u0≡0u_{0} \equiv0 and force ff:

u≠uˉ on R2×(0,T) for all T>0.(3)u \ne\bar{u}\ \text{on }\mathbb{R}^{2} \times(0,T)\ \text{for all }T > 0 . \tag*{(3)}

The solutions are constructed explicitly and satisfy many desirable properties, among which is smoothness on R2×(0,T)\mathbb{R}^{2} \times(0,T) for T≪1T \ll1 and energy equality in (2).

In a recent work [3] with Brué, we constructed non-unique Leray solutions of the three-dimensional Navier-Stokes equations. While the method therein is also applicable to (NSβ_\beta), at least when β>1\beta> 1, our goal is to highlight a different but equally natural way of incorporating viscosity into the scheme.

The main idea of the proof is to perturb the examples of non-uniqueness discovered by Vishik in [20, 21] to the viscous setting. To begin, we recall the basic dimensional analysis for the Euler equations

∂tu+u⋅∇u+∇p=f,div⁡u=0,(4)\partial_t u+u\cdot\nabla u+\nabla p=f,\qquad\operatorname{div}u=0, \tag*{(4)}

which can be equivalently rewritten in vorticity formulation as

∂tω+u⋅∇ω=g,u=Δ−1∇⊥ω,(5)\partial_t\omega+u\cdot\nabla\omega=g,\qquad u=\Delta^{-1}\nabla^{\perp}\omega, \tag*{(5)}

where ω=curl⁡u\omega=\operatorname{curl}u is the vorticity and g=curl⁡fg=\operatorname{curl}f. They have a two-parameter scaling symmetry

uμ,η=ημu(μx,ηt),ωμ,η=ηω(μx,ηt)(6)u_{\mu,\eta}=\frac{\eta}{\mu}u(\mu x,\eta t),\qquad\omega_{\mu,\eta}=\eta\omega(\mu x,\eta t) \tag*{(6)}

corresponding to the physical dimensions [x]=L[x]=L, [t]=T[t]=T, [u]=L/T[u]=L/T, [ω]=1/T[\omega]=1/T, [f]=L/T2[f]=L/T^{2}, and [g]=1/T2[g]=1/T^{2}. We suppose a relationship T=LαT=L^{\alpha} between the dimensions. This yields a one-parameter scaling symmetry:

uλ=λα−1u(λx,λαt),ωλ=λαu(λx,λαt).(7)u_{\lambda}=\lambda^{\alpha-1}u(\lambda x,\lambda^{\alpha}t),\qquad\omega_{\lambda}=\lambda^{\alpha}u(\lambda x,\lambda^{\alpha}t). \tag*{(7)}

One may seek special solutions ωˉ\bar{\omega} which are invariant under the above symmetry and necessarily have the form

ωˉ(x)=1tΩˉ(ξ),(8)\bar{\omega}(x)=\frac{1}{t}\bar{\Omega}(\xi), \tag*{(8)}

where ξ=x/t1/α\xi=x/t^{1/\alpha}. In fact, we may analyze an arbitrary Euler solution ω\omega in similarity variables (ξ,τ)(\xi,\tau), where τ=log⁡t\tau=\log t and the similarity profile Ω\Omega is defined by

ω(x,t)=1tΩ(ξ,τ),g(x,t)=1t2G(ξ,τ).(9)\omega(x,t)=\frac{1}{t}\Omega(\xi,\tau),\qquad g(x,t)=\frac{1}{t^{2}}G(\xi,\tau). \tag*{(9)}

The Euler equations (5) in similarity variables are

∂τΩ+(−1−ξα⋅∇)Ω+U⋅∇Ω=G,U=∇⊥Δ−1Ω.(10)\partial_{\tau}\Omega+\left(-1-\frac{\xi}{\alpha}\cdot\nabla\right)\Omega+U\cdot\nabla\Omega=G,\qquad U=\nabla^{\perp}\Delta^{-1}\Omega. \tag*{(10)}

Self-similar solutions ωˉ\bar{\omega} correspond precisely to steady states Ωˉ\bar{\Omega} in the new variables.

If Ωˉ\bar{\Omega} is an unstable steady state of (10), then it is natural to seek its unstable manifold. Suppose that Ωlin⁡\Omega^{\operatorname{lin}} is an exponentially growing solution of the linearization of (10) around Ωˉ\bar{\Omega}. One seeks a trajectory

Ω=Ωˉ+Ωlin⁡+Ωper⁡,(11)\Omega=\bar{\Omega}+\Omega^{\operatorname{lin}}+\Omega^{\operatorname{per}}, \tag*{(11)}

where Ωper⁡=oτ→−∞(∣Ωlin⁡∣)\Omega^{\operatorname{per}}=o_{\tau\to-\infty}(|\Omega^{\operatorname{lin}}|) ensures that Ω\Omega is not identically equal to Ωˉ\bar{\Omega}. On the other hand, Ω⟶τ→−∞Ωˉ\Omega\mathrel{\mathop{\longrightarrow}\limits_{\tau\to-\infty}} \bar{\Omega}. This is the non-uniqueness scenario in [20], which demonstrated the sharpness of the Yudovich class [22] in the ‘forced category’.

How does the above non-uniqueness scenario look with viscosity? The fractional Navier-Stokes equations (NSβ\mathrm{NS}_{\beta}) in self-similarity variables are

∂τΩ+(−1−ξ⋅∇ξα)Ω+U⋅∇Ω+eτγΛβΩ=G,U=∇⊥Δ−1Ω,(12)\partial_{\tau}\Omega+\left(-1-\frac{\xi\cdot\nabla_{\xi}}{\alpha}\right)\Omega+U\cdot\nabla\Omega+e^{\tau\gamma}\Lambda^{\beta}\Omega=G,\qquad U=\nabla^{\perp}\Delta^{-1}\Omega, \tag*{(12)}

where

γ=1−βα.(13)\gamma=1-\frac{\beta}{\alpha}. \tag*{(13)}

The traditional choice of exponent is α=β\alpha=\beta, so that the PDE (12) is τ\tau-autonomous. (The choice α=β=2\alpha=\beta=2 in [JS14, ABC22b] is the classical Navier-Stokes scaling.) This choice is determined by the dimensions of the viscosity: [ν]=Lβ/T[\nu]=L^{\beta}/T. With this choice, it is possible to follow the strategy in [3] to prove Theorem 1.1 when β∈(1,2)\beta\in(1,2). The same strategy would be valid in dimension three with β∈(1,5/2)\beta\in(1,5/2), that is, up to the Lions exponent. The most delicate part of the analysis in [3] is an eigenvalue perturbation argument which shows that not only the linearized 2D Euler equations, but also the linearized 3D Navier-Stokes equations admit an unstable eigenvalue around a well chosen background solution.

In this paper, we instead consider α>β\alpha>\beta, so that γ>0\gamma>0 and the dissipation term eτγΛβe^{\tau\gamma}\Lambda^{\beta} becomes perturbative (formally) as τ→−∞\tau\to-\infty. Then we consider the ansatz (1.11) exactly as in the inviscid setting, in particular, choosing Ωlin⁡\Omega^{\operatorname{lin}} as an unstable mode for 2D Euler rather than Navier-Stokes, and study Ωper⁡\Omega^{\operatorname{per}}. We take the existence of an unstable inviscid self-similar solution for granted.1 In this way, we turn our attention away from the spectral problem at the heart of [3] and toward the nonlinear arguments, which become quasilinear in nature and inherently more difficult. The nonlinear scheme will be outlined in § 2.1.

Our strategy can be compared to analogous strategies in the singularity formation literature, which seek stable manifolds as τ→+∞\tau\to+\infty of (potentially unstable) steady states in backward self-similarity variables. In contrast, we seek unstable manifolds as τ→−∞\tau\to-\infty of unstable steady states in forward self-similarity variables. Notably, the three-dimensional compressible Navier-Stokes ‘implosion’ singularity in [18] is a perturbation of a compressible Euler singularity; in similarity variables, the viscous term appears with a factor e−τγe^{-\tau\gamma}. See [19 CMVP21] for perturbative dissipation in the fractional Burgers equation.

The above comparison highlights the following point: In the absence of forcing, the desired self-similar solutions and corresponding exponents are chosen for you. When the scaling of the self-similar solution is supercritical with respect to the scaling of the viscous equation, the formal calculations give hope to perturb inviscid solutions to viscous solutions. For this reason, we choose to work with general (supercritical) exponents below. The present work provides another general method to incorporate a supercritical viscosity into non-uniqueness scenarios; we hope that this will inform future studies on non-uniqueness scenarios and continuation past singularities of various PDEs in fluid mechanics.

Previous work. The important works [JS14, JS15, GS17] proposed a route, accompanied by significant numerical evidence, to the (yet unproven) non-uniqueness of Leray solutions without force. The unstable manifold in similarity variables is already in [JS15, Theorem 4.1]. A related program for the two-dimensional Euler equations was initiated in [6, 5]. Non-uniqueness of weak Navier-Stokes solutions was demonstrated via convex integration in [7]. With hypoviscosity β<2/3\beta< 2/3, convex integration was able to achieve the Leray class [9, 11]. See [3] for a more comprehensive review.

Finally, we recently demonstrated non-uniqueness of Leray solutions to the forced Navier-Stokes solutions in bounded domains via a gluing method [2].

Preliminaries

Let 0<β<α<20 < \beta< \alpha< 2. Given a compactly supported, smooth velocity field Uˉ∈C∞(B2;R2)\bar{U} \in C^\infty(B_2;\mathbb{R}^2), Uˉ(x)=V(r)eθ\bar{U}(x) = V(r)e_\theta, with corresponding vorticity Ωˉ=curl⁡Uˉ\bar{\Omega} = \operatorname{curl}\bar{U}, we define the linearized operator around Ωˉ\bar{\Omega},

−LssΩ=(−1−ξα⋅∇)Ω+Uˉ⋅∇Ω+U⋅∇Ωˉ,(14)-L_{\mathrm{ss}}\Omega= \left(-1-\frac{\xi}{\alpha}\cdot\nabla\right)\Omega+\bar{U}\cdot\nabla\Omega+U\cdot\nabla\bar{\Omega}, \tag*{(14)}

where U=∇⊥Δ−1ΩU=\nabla^\perp\Delta^{-1}\Omega. The above operator is considered Lss:D(Lss)⊂Lm2→Lm2L_{\mathrm{ss}}:D(L_{\mathrm{ss}})\subset L_m^2\to L_m^2, where Lm2⊂L2L_m^2\subset L^2 consists of mm-fold rotationally symmetric L2L^2 functions and D(Lss):={Ω∈Lm2:LssΩ∈Lm2}D(L_{\mathrm{ss}}):=\{\Omega\in L_m^2:L_{\mathrm{ss}}\Omega\in L_m^2\}.

The following proposition was essentially demonstrated in [20, 21, 1, 3].

Proposition 2.1 (Unstable vortex). Let a0≥0a_0\geq0. There exists a compactly supported, smooth velocity field Uˉ∈C∞(B2;R2)\bar{U}\in C^\infty(B_2;\mathbb{R}^2), Uˉ(x)=V(r)eθ\bar{U}(x)=V(r)e_\theta, with corresponding vorticity Ωˉ=curl⁡Uˉ\bar{\Omega}=\operatorname{curl}\bar{U} and satisfying the following properties.

  1. Linear instability. There exists an integer m≥2m\geq2 such that LssL_{\mathrm{ss}} has an unstable eigenvalue. More specifically, there exist λ∈C\lambda\in\mathbb{C} with Re⁡λ=:a≥a0\operatorname{Re}\lambda=:a\geq a_0 and η∈Lm2∖{0}\eta\in L_m^2\setminus\{0\} satisfying Lssη=ληL_{\mathrm{ss}}\eta=\lambda\eta.

  1. Semigroup estimate. The operator LssL_{\mathrm{ss}} generates a continuous semigroup in Lm2L_m^2, and for all δ>0\delta>0, we have the semigroup estimate

∥eτLss∥Lm2→L2≲δe(a+δ)τfor all τ≥0.(15)\left\|e^{\tau L_{\mathrm{ss}}}\right\|_{L_m^2\to L^2}\lesssim_\delta e^{(a+\delta)\tau}\quad\text{for all }\tau\geq0. \tag*{(15)}
  1. Eigenfunction estimates. Any eigenfunction η\eta with unstable eigenvalue λ\lambda is compactly supported in B1B_1. For any k∈N0k\in\mathbb{N}_0 with Re⁡λ>k/α−1\operatorname{Re}\lambda>k/\alpha-1, the above eigenfunction belongs to HkH^k, and its velocity field BS⁡[η]\operatorname{BS}[\eta] belongs to Hk+1H^{k+1}.

Proof. 1. and 2. above are contained in [1 Theorem 2.4.2] except for the compact support property of Uˉ\bar{U}, which was pointed out in [3 Proposition 2.2]. Therefore, we focus on 3. Eigenfunction estimates. We present an alternative to the approach in [1 Lemma 5.0.3], where it was shown that η∈W2,∞\eta\in W^{2,\infty}. Suppose that η∈D(Lss)\eta\in D(L_{\mathrm{ss}}) is an eigenfunction with unstable eigenvalue λ=λ′−1\lambda=\lambda^{\prime}-1:

(λ′−ξα⋅∇+U‾⋅∇)η+BS⁡[η]⋅∇Ω‾=0.(16)\left(\lambda^{\prime}-\frac{\xi}{\alpha}\cdot\nabla+\overline{U}\cdot\nabla\right)\eta+\operatorname{BS}[\eta]\cdot\nabla\overline{\Omega}=0. \tag*{(16)}

First, we demonstrate that η\eta is compactly supported. When ∣ξ∣≥2|\xi|\ge2 is beyond the support of U‾∈C0∞(B2)\overline{U}\in C_{0}^{\infty}(B_{2}), we have

rα∂rη=λη,r≥2,(17)\frac{r}{\alpha}\partial_{r}\eta=\lambda\eta,\qquad r\ge2, \tag*{(17)}

whose only decaying solutions are identically zero in R2∖B2\mathbb{R}^{2}\setminus B_{2}.

Second, we prove that η∈Hk\eta\in H^{k} provided that Re⁡λ′>k/α\operatorname{Re}\lambda^{\prime}>k/\alpha. We consider BS⁡[η]⋅∇Ω‾\operatorname{BS}[\eta]\cdot\nabla\overline{\Omega} to be a ‘forcing term’ in the resolvent problem, and we rewrite the above equation via the Laplace transform characterization of the resolvent: R(λ′,Lss)=∫0+∞e−λ′tetLss dtR(\lambda^{\prime},L_{\mathrm{ss}})=\int_{0}^{+\infty}e^{-\lambda^{\prime}t}e^{tL_{\mathrm{ss}}}\,dt. Let XtX_{t} be the flow map associated to the autonomous vector field V(ξ):=ξ/α−U‾(ξ)V(\xi):=\xi/\alpha-\overline{U}(\xi). Then

η=−∫0+∞e−λ′t(BS⁡[η]⋅∇Ω‾)∘Xt dt.(18)\eta=-\int_{0}^{+\infty}e^{-\lambda^{\prime}t}\left(\operatorname{BS}[\eta]\cdot\nabla\overline{\Omega}\right)\circ X_{t}\,dt. \tag*{(18)}

To estimate η\eta, we recall basic properties of XtX_{t}, which is explicitly

Xt(r0eiθ0)=r(t,r0)eiθ(t,θ0,r0),(19)X_{t}(r_{0}e^{i\theta_{0}})=r(t,r_{0})e^{i\theta(t,\theta_{0},r_{0})}, \tag*{(19)}
r(t,r0)=et/αr0,θ(t,θ0,r0)=θ0−∫0tU‾θ(r(s,r0)) ds.(20)r(t,r_{0})=e^{t/\alpha}r_{0},\qquad\theta(t,\theta_{0},r_{0})=\theta_{0}-\int_{0}^{t}\overline{U}_{\theta}(r(s,r_{0}))\,ds. \tag*{(20)}

By the general formulae

(∇Xter)(r0eiθ0)=∂r0r (er∘Xt)+r∂r0θ (eθ∘Xt),(21)(\nabla X_{t}e_{r})(r_{0}e^{i\theta_{0}})=\partial_{r_{0}}r\,(e_{r}\circ X_{t})+r\partial_{r_{0}}\theta\,(e_{\theta}\circ X_{t}), \tag*{(21)}
(∇Xteθ)(r0eiθ0)=1r0∂θ0r (er∘Xt)+rr0∂θ0θ (eθ∘Xt),(22)(\nabla X_{t}e_{\theta})(r_{0}e^{i\theta_{0}})=\frac{1}{r_{0}}\partial_{\theta_{0}}r\,(e_{r}\circ X_{t})+\frac{r}{r_{0}}\partial_{\theta_{0}}\theta\,(e_{\theta}\circ X_{t}), \tag*{(22)}

it is not difficult to demonstrate that ∣∇Xt∣≲et/α|\nabla X_{t}|\lesssim e^{t/\alpha}. A different way of saying this is that solutions of the vector-valued ODE y⃗˙=(∇V)(Xt)y⃗\dot{\vec{y}}=(\nabla V)(X_{t})\vec{y} satisfy ∣y⃗(t)∣≲et/α∣y⃗(0)∣|\vec{y}(t)|\lesssim e^{t/\alpha}|\vec{y}(0)|. By taking spatial derivatives of the ODE X˙t=V(Xt)\dot{X}_{t}=V(X_{t}), this is enough to bootstrap the estimate on ∇Xt\nabla X_{t} to ∣∇kXt∣≲kekt/α|\nabla^{k}X_{t}|\lesssim_{k}e^{kt/\alpha} for all k≥0k\ge0. In particular, we have

∥η∥Hk≲∫0+∞e(k/α−λ′)t dt ∥BS⁡[η]⋅∇Ω‾∥Hk≲∥η∥Hk−1,(23)\|\eta\|_{H^{k}}\lesssim\int_{0}^{+\infty}e^{(k/\alpha-\lambda^{\prime})t}\,dt\,\|\operatorname{BS}[\eta]\cdot\nabla\overline{\Omega}\|_{H^{k}}\lesssim\|\eta\|_{H^{k-1}}, \tag*{(23)}

where the implied constant depends on various quantities. Hence, one can bootstrap η∈L2\eta\in L^{2} to η∈Hk\eta\in H^{k} provided that Re⁡λ>k/α−1\operatorname{Re}\lambda>k/\alpha-1. Finally, since η\eta is compactly supported and mean zero, we have ∣BS⁡[η]∣≲∣ξ∣−2|\operatorname{BS}[\eta]|\lesssim|\xi|^{-2} for ∣ξ∣≥2|\xi|\ge2. In particular, BS⁡[η]∈L2\operatorname{BS}[\eta]\in L^{2}.

Remark 2.2. It is natural that the unstable eigenfunctions should have smoothness related to Re⁡λ\operatorname{Re}\lambda since solutions to the ODE λf=r∂rf\lambda f=r\partial_{r}f, Re⁡λ>0\operatorname{Re}\lambda>0, are f=const.⁡×rλf=\operatorname{const.}\times r^{\lambda}. In Proposition 2.1, once it is known that the eigenfunctions are CβC^{\beta}, it is possible to bootstrap them in the Ck,βC^{k,\beta} scale, but we do not require this here.

Recall the similarity variables ξ=x/t1/α\xi=x/t^{1/\alpha} and τ=log⁡t\tau=\log t defined above (9). Given a similarity profile Ω‾\overline{\Omega} as in Proposition 2.1, we define the self-similar solution ω‾\overline{\omega} according to (8). Its velocity field is u‾(x,t)=t−1+1/αU‾(ξ)\overline{u}(x,t)=t^{-1+1/\alpha}\overline{U}(\xi). More generally, given a vorticity ω\omega, we extract its similarity profile Ω\Omega according to (9). Its velocity field is u(x,t)=t−1+1/αU(ξ,τ)u(x,t)=t^{-1+1/\alpha}U(\xi,\tau). We make the convention that, given a lowercase variable representing vorticity or velocity, its similarity profile is represented by the corresponding uppercase variable, and vice versa.

We now state basic properties of the force, which is induced entirely by the time derivative and fractional Laplacian of ω‾\overline{\omega}.

Lemma 2.3 (Finite energy forcing). Define h:=∂tω‾+Λβω‾h:=\partial_t\overline{\omega}+\Lambda^\beta\overline{\omega} and f:=Δ−1∇⊥h=∂tu‾+Λβu‾f:=\Delta^{-1}\nabla^\perp h=\partial_t\overline{u}+\Lambda^\beta\overline{u}. Then the background velocity u‾\overline{u} solves the fractional Navier–Stokes equations with velocity forcing ff, which belongs to Lt1Lx2(R2×(0,T))L^1_tL^2_x(\mathbb{R}^2\times(0,T)) for all T>0T>0. Additionally, f∈Lt∞Hxk(R2×(ϵ,T))f\in L^\infty_tH^k_x(\mathbb{R}^2\times(\epsilon,T)) for all k≥0k\geq0 and 0<ϵ<T<+∞0<\epsilon<T<+\infty.

This is a simple consequence of scaling.

Strategy

We revisit the ansatz (11):

Ω=Ω‾+Ωlin+Ωper(24)\Omega=\overline{\Omega}+\Omega^{\mathrm{lin}}+\Omega^{\mathrm{per}} \tag*{(24)}

where

Ωlin(ξ,τ):=Re⁡(eλτη),(25)\Omega^{\mathrm{lin}}(\xi,\tau):=\operatorname{Re}(e^{\lambda\tau}\eta), \tag*{(25)}

λ\lambda is a maximally unstable eigenvalue of LssL_{\mathrm{ss}}, and η\eta is an associated non-trivial eigenfunction. Then Ωlin\Omega^{\mathrm{lin}} is a solution to the linearized evolution equation

∂τΩlin−LssΩlin=0,(26)\partial_\tau\Omega^{\mathrm{lin}}-L_{\mathrm{ss}}\Omega^{\mathrm{lin}}=0, \tag*{(26)}

and our goal is to solve for the higher-order correction Ωper\Omega^{\mathrm{per}}. The PDE satisfied by Ωper\Omega^{\mathrm{per}} is

∂τΩper−LssΩper+(Ulin+Uper)⋅∇Ωper+Uper⋅∇Ωlin+eτγΛβΩper=−Ulin⋅∇Ωlin−eτγΛβΩlin,(27)\begin{aligned} \partial_\tau\Omega^{\mathrm{per}}-L_{\mathrm{ss}}\Omega^{\mathrm{per}}+(U^{\mathrm{lin}}+U^{\mathrm{per}})\cdot\nabla\Omega^{\mathrm{per}}+U^{\mathrm{per}}\cdot\nabla\Omega^{\mathrm{lin}}+e^{\tau}\gamma\Lambda^\beta\Omega^{\mathrm{per}}\\ =-U^{\mathrm{lin}}\cdot\nabla\Omega^{\mathrm{lin}}-e^{\tau}\gamma\Lambda^\beta\Omega^{\mathrm{lin}}, \tag*{(27)} \end{aligned}

where the right-hand side is considered a source term. The fractional Laplacian eτγΛβe^{\tau}\gamma\Lambda^\beta will be treated perturbatively. Hence, our approach is essentially quasilinear in nature, and it will be necessary to incorporate the transport (Ulin+Uper)⋅∇(U^{\mathrm{lin}}+U^{\mathrm{per}})\cdot\nabla into the operator. Define

−L[W]=−Lss+(Ulin+W)⋅∇+eτγΛβ,(28)-L[W]=-L_{\mathrm{ss}}+(U^{\mathrm{lin}}+W)\cdot\nabla+e^{\tau}\gamma\Lambda^\beta, \tag*{(28)}

the ‘main operator’ in the PDE, and the associated (formal) solution operator S[W]=S[W](τ,s)S[W]=S[W](\tau,s). We seek solutions satisfying Duhamel’s formula:

Ωper(⋅,τ)=−∫−∞τS[Uper](τ,s)[(Ulin+Uper)⋅∇Ωlin+esγΛβΩlin](⋅,s) ds.(29)\Omega^{\mathrm{per}}(\cdot,\tau)=-\int_{-\infty}^{\tau}S[U^{\mathrm{per}}](\tau,s)[(U^{\mathrm{lin}}+U^{\mathrm{per}})\cdot\nabla\Omega^{\mathrm{lin}}+e^{s\gamma}\Lambda^\beta\Omega^{\mathrm{lin}}](\cdot,s)\,ds. \tag*{(29)}

It will be necessary to develop the (strong) solution theory for Ωper\Omega^{\mathrm{per}} and the solution operator SS in high regularity spaces, whose corresponding velocity fields are Lipschitz continuous, e.g., Ω∈Hs\Omega\in H^s, s>d/2s>d/2. It will be more convenient to estimate SS in a function space X⊂W1,4X\subset W^{1,4} which will be defined in Section 3.

There are two main difficulties, which are already present in the Euler equations and have been encountered in previous work, e.g., [4]; the difficulties will only be exacerbated by the dissipation eτγΛβe^{\tau}\gamma\Lambda^\beta.

First, there is the apparent derivative loss in adding the transport operator (Ulin+Uper)⋅∇(U^{\mathrm{lin}}+U^{\mathrm{per}})\cdot\nabla onto LssL_{\mathrm{ss}}. On the one hand, it is necessary to use the spectral information on LssL_{\mathrm{ss}} through the semigroup estimate in L2L^2. On the other hand, it is too naïve to write Duhamel’s formula, containing expressions like ∫0τe(τ−s)Lss(Ulin⋅∇Ωper) ds\int_0^\tau e^{(\tau-s)L_{\mathrm{ss}}}(U^{\mathrm{lin}}\cdot\nabla\Omega^{\mathrm{per}})\,\mathrm{d}s, and hope to close an estimate at the L2L^2 level. Moreover, in our setting, adding the dissipation operator eτγΛβe^{\tau\gamma\Lambda^\beta} further exhibits a loss of β\beta derivatives. The key observation for handling the loss is that the compact term Uper⋅∇ΩˉU^{\mathrm{per}}\cdot\nabla\bar{\Omega}, responsible for the creation of the unstable eigenvalues, gains regularity. Hence, the strategy will be to lose derivatives while applying the semigroup estimate in Duhamel’s formula and regain them in a bootstrapping procedure based on energy estimates.2

The second difficulty concerns the operator Uˉ⋅∇\bar{U}\cdot\nabla in high regularity spaces. One way to see the difficulty is to pass a derivative into the equation: The PDE for ∂iΩper\partial_i\Omega^{\mathrm{per}} contains the term ∂iUˉ⋅∇Ωper\partial_i\bar{U}\cdot\nabla\Omega^{\mathrm{per}}, which in principle can change the growth rate. In [4], it was observed that it is enough assume that Re⁡λ\operatorname{Re}\lambda is greater than the maximal fluid Lyapunov exponent. Our approach will be Eulerian rather than Lagrangian. It is an energy method specific to shear flows and vortices (in short, the idea is to apply ∂θ\partial_\theta first), and crucially, it is well behaved even with the term eτγΛβe^{\tau\gamma\Lambda^\beta}. In fact, when β>1\beta>1, the term eτγΛβe^{\tau\gamma\Lambda^\beta} will help us in the bootstrapping procedure: Once we have propagated one derivative, the other β−1\beta-1 derivatives needed to close the argument are regained by smoothing. Heuristically, a gain of β\beta derivatives costs a factor eτγe^{\tau\gamma}, which is why the dissipation term cannot be used to control everything. This is seen in the tracking of the dissipation norm DD below.

Linear theory

In the following, we will assume that a0=8/αa_0=8/\alpha and δ≤δ0:=min⁡{γ/4,1/8,a0/2}\delta\le\delta_0:=\min\{\gamma/4,1/8,a_0/2\}. We also fix Ωˉ\bar{\Omega}, λ=a+ib\lambda=a+ib, η\eta and mm as in Proposition 2.1. We denote by CC constants whose values depend only on

data=(α,β,Ωˉ,a,b,η,m,δ).(30)\mathrm{data}=(\alpha,\beta,\bar{\Omega},a,b,\eta,m,\delta). \tag*{(30)}

The dependence on Ωˉ\bar{\Omega} and η\eta can be made more precise through their regularity and size of the support, but this is not needed for the rest of our arguments, and we do not pursue it here. The value of CC may change from line to line. When we need to recall a specific constant of this type, we denote it with C1,C2,…C_1,C_2,\ldots

This section is devoted to estimates on the solution operator S[W]S[W] to the PDE

∂τΩ+(−1−ξ⋅∇/α)Ω+Uˉ⋅∇Ω+U⋅∇Ωˉ+(Ulin+W)⋅∇Ω+eτγΛβΩ=0(31)\partial_\tau\Omega+(-1-\xi\cdot\nabla/\alpha)\Omega+\bar{U}\cdot\nabla\Omega+U\cdot\nabla\bar{\Omega}+(U^{\mathrm{lin}}+W)\cdot\nabla\Omega+e^{\tau\gamma\Lambda^\beta}\Omega=0 \tag*{(31)}

with initial data Ω(⋅,τ0)=Ω0\Omega(\cdot,\tau_0)=\Omega_0 and divergence-free WW satisfying the property (3.7) below. We assume that Ω0\Omega_0 and WW are mm-fold rotationally symmetric, so that the semigroup estimate in Proposition 2.1 can be applied. The remaining terms in the PDE (2.14) will be incorporated in Section 4 by Duhamel’s formula.

Let us fix a parameter Q∈[1,3/2)Q\in[1,3/2). In view of the precise structure of the terms in our equation, it turns out that a natural space to perform such estimates is given by the norm

∥Ω∥X=∥Ω∥LQ∩L2+∥∂θΩ∥LQ∩L4+∥∇Ω∥L2∩L4,(32)\|\Omega\|_X=\|\Omega\|_{L^Q\cap L^2}+\|\partial_\theta\Omega\|_{L^Q\cap L^4}+\|\nabla\Omega\|_{L^2\cap L^4}, \tag*{(32)}

which is equivalent to ∥Ω∥LQ+∥∂θΩ∥LQ∩L4+∥∇Ω∥L2∩L4\|\Omega\|_{L^Q}+\|\partial_\theta\Omega\|_{L^Q\cap L^4}+\|\nabla\Omega\|_{L^2\cap L^4}, but the description in (32) is perhaps more indicative of the estimates we perform. We also keep track of the quantity

∥Ω∥D(τ,τ0)2=∫τ0τesγ∥Λ1+β2Ω∥L22 ds(33)\|\Omega\|_{D(\tau,\tau_0)}^2=\int_{\tau_0}^{\tau}e^{s\gamma}\|\Lambda^{1+\frac{\beta}{2}}\Omega\|_{L^2}^2\,\mathrm{d}s \tag*{(33)}

related to the dissipation.

The space XX has some useful properties listed below. Let us consider W∈L2W\in L^2 such that curl⁡W∈X\operatorname{curl}W\in X. Since K2=1B1K2+1B1cK2∈Lp+LqK_2=\mathbf{1}_{B_1}K_2+\mathbf{1}_{B_1^c}K_2\in L^p+L^q for every p<2p<2, q>2q>2, we have by Young’s convolution inequality that

∥W∥L∞=∥K2∗curl⁡W∥L∞≤C(∥curl⁡W∥LQ+∥curl⁡W∥L4)≤C∥curl⁡W∥X.(34)\|W\|_{L^\infty}=\|K_2*\operatorname{curl}W\|_{L^\infty}\leq C(\|\operatorname{curl}W\|_{L^Q}+\|\operatorname{curl}W\|_{L^4})\leq C\|\operatorname{curl}W\|_X . \tag*{(34)}

and similarly,

∥∂θW∥L∞=∥K2∗∂θcurl⁡W∥L∞≤C(∥∂θcurl⁡W∥LQ+∥∂θcurl⁡W∥L4)≤C∥curl⁡W∥X.(35)\|\partial_\theta W\|_{L^\infty}=\|K_2*\partial_\theta\operatorname{curl}W\|_{L^\infty}\leq C(\|\partial_\theta\operatorname{curl}W\|_{L^Q}+\|\partial_\theta\operatorname{curl}W\|_{L^4})\leq C\|\operatorname{curl}W\|_X . \tag*{(35)}

The main result of this section is

Proposition 3.1 (Estimates for SS). Let W∈L∞((−∞,0];L2(R2))W\in L^\infty((-\infty,0];L^2(\mathbb{R}^2)) be such that

sup⁡τ∈(−∞,0)e−aτ∥curl⁡W∥X≤1.(36)\sup_{\tau\in(-\infty,0)}e^{-a\tau}\|\operatorname{curl}W\|_X\leq1. \tag*{(36)}

Then for every δ≤δ0\delta\leq\delta_0, there exist constants Tmax⁡:=Tmax⁡(data)≤0T_{\max}:=T_{\max}(\mathrm{data})\leq0 and Cˉ:=Cˉ(data)>1\bar C:=\bar C(\mathrm{data})>1 such that for every τ0∈(−∞,Tmax⁡]\tau_0\in(-\infty,T_{\max}] and τ∈[τ0,Tmax⁡]\tau\in[\tau_0,T_{\max}], we have

∥S[W](τ,τ0)Ω0∥X+∥S[W](⋅,τ0)Ω0∥D(τ0,τ)≤Cˉe(a+δ)(τ−τ0)∥Ω0∥X.(37)\|S[W](\tau,\tau_0)\Omega_0\|_X+\|S[W](\cdot,\tau_0)\Omega_0\|_{D(\tau_0,\tau)}\leq\bar C e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X. \tag*{(37)}

Proof of Proposition 3.1. The constants Tmax⁡≤0T_{\max}\leq0 and Cˉ>1\bar C>1 will be chosen later. For τ∈[τ0,Tmax⁡]\tau\in[\tau_0,T_{\max}], we denote Ω(⋅,τ)=S[W](τ,τ0)Ω0\Omega(\cdot,\tau)=S[W](\tau,\tau_0)\Omega_0 the solution of (31). The proof of Proposition 3.1 will be established through the following claim.

Let τ1∈[τ0,Tmax⁡]\tau_1\in[\tau_0,T_{\max}] satisfying the following bootstrap assumption:

For all τ∈[τ0,τ1]\tau\in[\tau_0,\tau_1], the estimate (37) holds.

Then for all τ∈[τ0,τ1]\tau\in[\tau_0,\tau_1] we have

∥Ω(⋅,τ)∥X+∥Ω∥D(τ0,τ)≤C1(1+Cˉ1/2eζTmax⁡)Cˉ1/2e(a+δ)(τ−τ0)∥Ω0∥X,(38)\|\Omega(\cdot,\tau)\|_X+\|\Omega\|_{D(\tau_0,\tau)}\leq C_1(1+\bar C^{1/2}e^{\zeta T_{\max}})\bar C^{1/2}e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X, \tag*{(38)}

where C1:=C1(data)>1C_1:=C_1(\mathrm{data})>1 and ζ:=ζ(data)>0\zeta:=\zeta(\mathrm{data})>0.

Once this claim is established, we easily conclude the proof of Proposition 3.1 by the following argument. We fix Cˉ=(4C1)2\bar C=(4C_1)^2 and Tmax⁡T_{\max} in terms of data, but otherwise independent of τ0\tau_0, such that Cˉ1/2eζTmax⁡≤1\bar C^{1/2}e^{\zeta T_{\max}}\leq1. Next, we consider τˉ1\bar\tau_1 to be the largest value of τ1∈(τ0,Tmax⁡)\tau_1\in(\tau_0,T_{\max}) such that (37) holds. If τˉ1=Tmax⁡\bar\tau_1=T_{\max}, our main claim is proved. The case τˉ1<Tmax⁡\bar\tau_1<T_{\max} is ruled out by the continuity of τ↦∥Ω(⋅,τ)∥X\tau\mapsto\|\Omega(\cdot,\tau)\|_X and by (38), which implies with our choice of Tmax⁡T_{\max} that the bound (37) is not saturated in τˉ1\bar\tau_1, namely,

∥Ω(⋅,τˉ1)∥X+∥Ω∥D(τ0,τˉ1)≤Cˉ2e(a+δ)(τˉ1−τ0)∥Ω0∥X.(39)\|\Omega(\cdot,\bar\tau_1)\|_X+\|\Omega\|_{D(\tau_0,\bar\tau_1)}\leq\frac{\bar C}{2}e^{(a+\delta)(\bar\tau_1-\tau_0)}\|\Omega_0\|_X. \tag*{(39)}

Hence, the proof of Proposition 3.1 is complete provided that (38) is established.

Step 1. [Baseline estimate: Improved L2L^2 bound] Suppose that the bootstrap assumption (3.9) holds. Then, for all τ∈[τ0,τ1]\tau\in[\tau_0,\tau_1],

∥Ω(⋅,τ)∥L2≤C(1+Cˉe2ζTmax⁡e(a+δ)(τ−τ0))∥Ω0∥X.(40)\|\Omega(\cdot,\tau)\|_{L^2} \le C(1+\bar{C}e^{2\zeta T_{\max}}e^{(a+\delta)(\tau-\tau_0)})\|\Omega_0\|_X. \tag*{(40)}

Indeed, by Duhamel’s formula:

Ω(⋅,τ)=e(τ−τ0)LssΩ0−∫τ0τe(τ−s)Lss[(W+Ulin)⋅∇Ω+esγΛβΩ](⋅,s) ds.(41)\Omega(\cdot,\tau)=e^{(\tau-\tau_0)L_{\mathrm{ss}}}\Omega_0-\int_{\tau_0}^{\tau}e^{(\tau-s)L_{\mathrm{ss}}}\left[(W+U^{\mathrm{lin}})\cdot\nabla\Omega+e^{s\gamma}\Lambda^\beta\Omega\right](\cdot,s)\,ds. \tag*{(41)}

According to the semigroup estimate (15), we have3

∥Ω(⋅,τ)∥L2≤∥e(τ−τ0)LssΩ0∥L2+∫τ0τ∥e(τ−s)Lss[(W+Ulin)⋅∇Ω−esγΛβΩ]∥L2(⋅,s) ds≤Ce(a+δ)(τ−τ0)∥Ω0∥L2+∫τ0τe(a+δ)(τ−s)[∥(W+Ulin)⋅∇Ω(⋅,s)∥L2(⋅,s) ds+esγ∥ΛβΩ∥L2(⋅,s)] ds.(42)\begin{aligned} \|\Omega(\cdot,\tau)\|_{L^2} &\le\left\|e^{(\tau-\tau_0)L_{\mathrm{ss}}}\Omega_0\right\|_{L^2} +\int_{\tau_0}^{\tau}\left\|e^{(\tau-s)L_{\mathrm{ss}}}\left[(W+U^{\mathrm{lin}})\cdot\nabla\Omega-e^{s\gamma}\Lambda^\beta\Omega\right]\right\|_{L^2}(\cdot,s)\,ds\\ &\le Ce^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_{L^2} +\int_{\tau_0}^{\tau}e^{(a+\delta)(\tau-s)}\left[\|(W+U^{\mathrm{lin}})\cdot\nabla\Omega(\cdot,s)\|_{L^2(\cdot,s)}\,ds\right.\\ &\qquad\left.+e^{s\gamma}\|\Lambda^\beta\Omega\|_{L^2(\cdot,s)}\right]\,ds. \tag*{(42)} \end{aligned}

Next, we estimate the first integrand in the right-hand side of (42). By (34), the smoothness of η\eta in Proposition 2.1, the assumption on WW (36) and the bootstrap assumption (3.9), we obtain that

∥(W+Ulin)⋅∇Ω(⋅,s)∥2≤∥W∥∞∥∇Ω∥2+∥Ulin∥∞∥∇Ω∥2(43)\|(W+U^{\mathrm{lin}})\cdot\nabla\Omega(\cdot,s)\|_2\le\|W\|_\infty\|\nabla\Omega\|_2+\|U^{\mathrm{lin}}\|_\infty\|\nabla\Omega\|_2 \tag*{(43)}
≤eas∥Ω(⋅,s)∥X≲e2sζe(a+δ)(s−τ0)∥Ω0∥X.(44)\le e^{as}\|\Omega(\cdot,s)\|_X\lesssim e^{2s\zeta}e^{(a+\delta)(s-\tau_0)}\|\Omega_0\|_X. \tag*{(44)}

Plugging this estimate in the first integral in (42), we can estimate such integral with CCˉe2ζTmax⁡e(a+δ)(τ−τ0)∥Ω0∥XC\bar{C}e^{2\zeta T_{\max}}e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X.

Next, we consider the dissipation term. First, we interpolate between L2L^2 and H˙1+β2\dot{H}^{1+\frac{\beta}{2}} and by Young inequality we have

(∫τ0τesγ∥ΛβΩ∥L22 ds)1/2≤(∫τ0τesγ∥Λ1+β2Ω∥L22(2−β)2+β∥Ω∥L24β2+β ds)1/2≤C(∫τ0τesγ(∥Λ1+β2Ω∥L22+∥Ω∥L22) ds)1/2≤C∥Ω∥D(τ0,τ)+CCˉ(∫τ0τe2sγe2(a+δ)(s−τ0)∥Ω0∥X2 ds)1/2≤CCˉe(a+δ)(τ−τ0)∥Ω0∥X.(45)\begin{aligned} \left(\int_{\tau_0}^{\tau}e^{s\gamma}\|\Lambda^\beta\Omega\|_{L^2}^2\,ds\right)^{1/2} \le\left(\int_{\tau_0}^{\tau}e^{s\gamma}\|\Lambda^{1+\frac{\beta}{2}}\Omega\|_{L^2}^{\frac{2(2-\beta)}{2+\beta}}\|\Omega\|_{L^2}^{\frac{4\beta}{2+\beta}}\,ds\right)^{1/2} \\ &\le C\left(\int_{\tau_0}^{\tau}e^{s\gamma}\left(\|\Lambda^{1+\frac{\beta}{2}}\Omega\|_{L^2}^2+\|\Omega\|_{L^2}^2\right)\,ds\right)^{1/2} \\ &\le C\|\Omega\|_{D(\tau_0,\tau)}+C\bar{C}\left(\int_{\tau_0}^{\tau}e^{2s\gamma}e^{2(a+\delta)(s-\tau_0)}\|\Omega_0\|_X^2\,ds\right)^{1/2} \\ &\le C\bar{C}e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X. \tag*{(45)} \end{aligned}

where we used the bootstrap assumption both on ∥Ω∥D(τ0,τ)\|\Omega\|_{D(\tau_0,\tau)} and on ∥Ω∥L2\|\Omega\|_{L^2}. Then, using the Cauchy-Schwarz in time and (45),

∫τ0τe(τ−s)(a+δ)esγ/2∥esγ/2ΛβΩ(⋅,s)∥L2 ds≤Ceτγ/2(C‾+1)e(a+δ)(τ−τ0)∥Ω0∥X.(46)\int_{\tau_0}^{\tau} e^{(\tau-s)(a+\delta)}e^{s\gamma/2}\|e^{s\gamma/2}\Lambda^\beta\Omega(\cdot,s)\|_{L^2}\,ds \le Ce^{\tau\gamma/2}(\overline{C}+1)e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X. \tag*{(46)}

This completes the proof of (40).

We notice that from an interpolation inequality, (40) and the bootstrap assumption it follows also an improved L4L^4 bound of the following form:

∥Ω∥L4≤C∥Ω∥L21/2∥∇Ω∥L21/2≤[C(1+C‾e2ζTmax⁡)e(a+δ)(τ−τ0)]1/2C‾1/2ea+δ2(τ−τ0)∥Ω0∥X≤C(1+C‾1/2eζTmax⁡)C‾1/2e(a+δ)(τ−τ0)∥Ω0∥X.(47)\begin{aligned} \|\Omega\|_{L^4}\le C\|\Omega\|_{L^2}^{1/2}\|\nabla\Omega\|_{L^2}^{1/2} \\ &\le[C(1+\overline{C}e^{2\zeta T_{\max}})e^{(a+\delta)(\tau-\tau_0)}]^{1/2}\overline{C}^{1/2}e^{\frac{a+\delta}{2}(\tau-\tau_0)}\|\Omega_0\|_X \\ &\le C(1+\overline{C}^{1/2}e^{\zeta T_{\max}})\overline{C}^{1/2}e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X. \tag*{(47)} \end{aligned}

Notice that this estimate is worse than (40) due to the presence of C‾1/2\overline{C}^{1/2}; however, being its power strictly less than 1, the estimate is still good enough.

Step 2. [Improved LQL^Q bound] Suppose that the bootstrap assumption (3.9) holds. Then for all τ∈[τ0,τ1]\tau\in[\tau_0,\tau_1]

∥Ω(⋅,τ)∥LQ≤C(Cδ0+CC‾e2ζTmax⁡e(a+δ)(τ−τ0))∥Ω0∥X.(48)\|\Omega(\cdot,\tau)\|_{L^Q}\le C(C_{\delta_0}+C\overline{C}e^{2\zeta T_{\max}}e^{(a+\delta)(\tau-\tau_0)})\|\Omega_0\|_X. \tag*{(48)}

We perform an LQL^Q energy estimate on the equation for Ω\Omega, namely we multiply the equation by Q∣Ω∣Q−2ΩQ|\Omega|^{Q-2}\Omega and integrate in space. We obtain

ddt∫∣Ω∣Q dξ≤(−Q+2α)∫∣Ω∣Q dξ+Q∫∣U⋅∇Ω‾∣∣Ω∣Q−1 dξ−Q∫eτγΛβΩ∣Ω∣Q−2Ω dξ(49)\frac{d}{dt}\int|\Omega|^Q\,d\xi\le\left(-Q+\frac{2}{\alpha}\right)\int|\Omega|^Q\,d\xi+Q\int|U\cdot\nabla\overline{\Omega}||\Omega|^{Q-1}\,d\xi-Q\int e^{\tau\gamma}\Lambda^\beta\Omega|\Omega|^{Q-2}\Omega\,d\xi \tag*{(49)}

The last term, related to the dissipation, is known to be negative via the Córdoba-Córdoba inequality (see, for instance, [8 Lemma 2.4]). The main term of interest is the ‘forcing’ U⋅∇Ω‾U\cdot\nabla\overline{\Omega}, which we estimate thanks to (34) and the fact that Ω‾\overline{\Omega} is compactly supported:

∥U⋅∇Ω‾∥LQ≤∥U∥L∞∥∇Ω‾∥LQ≤C∥Ω∥X.(50)\|U\cdot\nabla\overline{\Omega}\|_{L^Q}\le\|U\|_{L^\infty}\|\nabla\overline{\Omega}\|_{L^Q}\le C\|\Omega\|_X. \tag*{(50)}

We conclude, via a Gronwall estimate on the inequality (49), that (48) holds. Here we take advantage of the fact that a0>2αa_0>\frac{2}{\alpha} to see that the term (−Q+2α)∫∣Ω∣Q dξ\left(-Q+\frac{2}{\alpha}\right)\int|\Omega|^Q\,d\xi can be neglected in this estimate, since it does not change the exponential in time behavior of the solution.

Step 3. [Improved ∂θΩ\partial_\theta\Omega bound] Suppose that the bootstrap assumption (3.9) holds. Then for all τ∈[τ0,τ1]\tau\in[\tau_0,\tau_1] and for p=Q,2,4p=Q,2,4, we have

∥∂θΩ(⋅,τ)∥Lp≤C(1+C‾1/2e2ζTmax⁡)C‾1/2e(a+δ)(τ−τ0)∥Ω0∥X.(51)\|\partial_\theta\Omega(\cdot,\tau)\|_{L^p}\le C(1+\overline{C}^{1/2}e^{2\zeta T_{\max}})\overline{C}^{1/2}e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X. \tag*{(51)}

We first observe that

∂θΛβ=Λβ∂θ,(52)\partial_\theta\Lambda^\beta=\Lambda^\beta\partial_\theta, \tag*{(52)}

which follows from differentiating the rotation symmetry Rot⁡θΛβ=ΛβRot⁡θ\operatorname{Rot}_\theta\Lambda^\beta=\Lambda^\beta\operatorname{Rot}_\theta in the angle θ∈R\theta\in\mathbb{R} at θ=0\theta=0. Similarly ∂θcurl⁡W=curl⁡∂θW\partial_\theta\operatorname{curl}W=\operatorname{curl}\partial_\theta W and notice that ∂θ\partial_\theta also commutes through the Biot-Savart law. Commuting ∂θ\partial_\theta into the PDE, we have

∂τ∂θΩ+(−1−ξα⋅∇)∂θΩ+(U‾+W+Ulin)⋅∇∂θΩ+eτγΛβ∂θΩ=F,(53)\partial_\tau\partial_\theta\Omega+\left(-1-\frac{\xi}{\alpha}\cdot\nabla\right)\partial_\theta\Omega+\left(\overline{U}+W+U^{\mathrm{lin}}\right)\cdot\nabla\partial_\theta\Omega+e^{\tau\gamma}\Lambda^\beta\partial_\theta\Omega=F, \tag*{(53)}

where

−F=∂θU⋅∇Ω‾+(∂θW‾+∂θUlin)⋅∇Ω.(54)-F=\partial_\theta U\cdot\nabla\overline{\Omega}+\left(\partial_\theta\overline{W}+\partial_\theta U^{\mathrm{lin}}\right)\cdot\nabla\Omega. \tag*{(54)}

As regards the first term in the force, we estimate it in LpL^p via Calderón-Zygmund as follows

∥∂θU⋅∇Ω‾∥Lp≤∥1∣x∣∂θU∥Lp∥∣x∣∣∇Ω‾∣∥L∞≤C∥∇U∥Lp≤C∥Ω∥Lp.\left\|\partial_\theta U\cdot\nabla\overline{\Omega}\right\|_{L^p}\leq\left\|\frac{1}{|x|}\partial_\theta U\right\|_{L^p}\left\||x|\left|\nabla\overline{\Omega}\right|\right\|_{L^\infty}\leq C\|\nabla U\|_{L^p}\leq C\|\Omega\|_{L^p}.

Using the baseline estimate (40) if p=Qp=Q or p=2p=2 and the improved bound (48) if p=4p=4, we get

∥∂θU⋅∇Ω‾∥L4≤C∥Ω∥L4≤C(1+C‾eζTmax⁡)C‾1/2e(a+δ)(τ−τ0)∥Ω0∥X.\left\|\partial_\theta U\cdot\nabla\overline{\Omega}\right\|_{L^4}\leq C\|\Omega\|_{L^4}\leq C\left(1+\overline{C}e^{\zeta T_{\max}}\right)\overline{C}^{1/2}e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X.

By (3.5), (3.6), (3.7) and the bootstrap assumption we have, for p=2,4p=2,4,

∥(∂θW+∂θUlin)∇Ω∥Lp≤C(∥∂θW∥L∞+∥∂θUlin∥L∞)∥∇Ω∥Lp≤CC‾eaτe(a+δ)(τ−τ0)∥Ω0∥X.(55)\begin{aligned} \left\|\left(\partial_\theta W+\partial_\theta U^{\mathrm{lin}}\right)\nabla\Omega\right\|_{L^p}\leq C\left(\|\partial_\theta W\|_{L^\infty}+\|\partial_\theta U^{\mathrm{lin}}\|_{L^\infty}\right)\|\nabla\Omega\|_{L^p} \\ &\leq C\overline{C}e^{a\tau}e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X. \tag*{(55)} \end{aligned}

For p=Qp=Q we modify the previous computation as follows

∥(∂θW+∂θUlin)∇Ω∥LQ≤C(∥∂θW∥L2Q2−Q+∥∂θUlin∥L2Q2−Q)∥∇Ω∥L2≤C(∥∂θcurl⁡W∥LQ+∥∂θUlin∥L2Q2−Q)∥∇Ω∥L2≤CC‾eaτe(a+δ)(τ−τ0)∥Ω0∥X.(56)\begin{aligned} \left\|\left(\partial_\theta W+\partial_\theta U^{\mathrm{lin}}\right)\nabla\Omega\right\|_{L^Q}\leq C\left(\|\partial_\theta W\|_{L^{\frac{2Q}{2-Q}}}+\|\partial_\theta U^{\mathrm{lin}}\|_{L^{\frac{2Q}{2-Q}}}\right)\|\nabla\Omega\|_{L^2} \\ &\leq C\left(\|\partial_\theta\operatorname{curl}W\|_{L^Q}+\|\partial_\theta U^{\mathrm{lin}}\|_{L^{\frac{2Q}{2-Q}}}\right)\|\nabla\Omega\|_{L^2} \\ &\leq C\overline{C}e^{a\tau}e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X. \tag*{(56)} \end{aligned}

We multiply the equation by p(∂θΩ)p−1p(\partial_\theta\Omega)^{p-1} if p=Q,2,4p=Q,2,4 and integrate by parts. After rewriting the fractional dissipation term and using Holder inequality to estimate the term which includes FF, we get

ddτ∫∣∂θΩ∣p dξ+(2α−p)∫∣∂θΩ∣p dξ+ceτγ∫∣Λβ2∣∂θΩ∣p/2∣2 dξ≤C∥F∥Lp∥∂θΩ∥Lpp−1.(57)\frac{d}{d\tau}\int|\partial_\theta\Omega|^p\,d\xi+\left(\frac{2}{\alpha}-p\right)\int|\partial_\theta\Omega|^p\,d\xi+ce^{\tau\gamma}\int\left|\Lambda^{\frac{\beta}{2}}|\partial_\theta\Omega|^{p/2}\right|^2\,d\xi\leq C\|F\|_{L^p}\|\partial_\theta\Omega\|_{L^p}^{p-1}. \tag*{(57)}

We neglect the last term in the left-hand side, which has the right sign, divide by ∥∂θΩ∥Lpp−1\|\partial_\theta\Omega\|_{L^p}^{p-1} and sum over p=Q,2,4p=Q,2,4.

Step 4. [Improved ∇Ω\nabla\Omega bound] Suppose that the bootstrap assumption (3.9) holds. Then for all τ∈[τ0,τ1]\tau\in[\tau_0,\tau_1] and for p=2,4p=2,4, we have

∥∇Ω∥Lp≤C(1+C‾1/2e2ζTmax⁡)e(a+δ)(τ−τ0)C‾1/2∥Ω0∥X,(58)\|\nabla\Omega\|_{L^p}\leq C\left(1+\overline{C}^{1/2}e^{2\zeta T_{\max}}\right)e^{(a+\delta)(\tau-\tau_0)}\overline{C}^{1/2}\|\Omega_0\|_X, \tag*{(58)}
∥Ω∥D(τ0,τ)≤C(1+C‾1/2e2ζTmax⁡)e(a+δ)(τ−τ0)C‾1/2∥Ω0∥X.(59)\|\Omega\|_{D(\tau_0,\tau)}\leq C\left(1+\overline{C}^{1/2}e^{2\zeta T_{\max}}\right)e^{(a+\delta)(\tau-\tau_0)}\overline{C}^{1/2}\|\Omega_0\|_X. \tag*{(59)}

Let i=1,2i=1,2 and consider the equation for ∂iΩ\partial_i\Omega:

∂τ∂iΩ+(−1−1α−ξ⋅∇α)∂iΩ+(U‾+Ulin+W)⋅∇∂iΩ+∂iU‾⋅∇Ω+∂i(Ulin+W)⋅∇Ω−eτγΛβ∂iΩ=Fi,(60)\partial_\tau\partial_i\Omega+\left(-1-\frac{1}{\alpha}-\frac{\xi\cdot\nabla}{\alpha}\right)\partial_i\Omega+\left(\overline{U}+U^{\mathrm{lin}}+W\right)\cdot\nabla\partial_i\Omega+\partial_i\overline{U}\cdot\nabla\Omega+\partial_i\left(U^{\mathrm{lin}}+W\right)\cdot\nabla\Omega-e^{\tau\gamma}\Lambda^\beta\partial_i\Omega=F_i, \tag*{(60)}

where

−Fi=∂iU⋅∇Ω‾+U‾⋅∇∂iΩ‾.(61)-F_i=\partial_i U\cdot\nabla\overline{\Omega}+\overline{U}\cdot\nabla\partial_i\overline{\Omega}. \tag*{(61)}

We multiply by ∣∇Ω∣p−2∂iΩ|\nabla\Omega|^{p-2}\partial_i\Omega, sum over i=1,2i=1,2 and integrate by parts. The last term in the first line of (60) vanishes, after this computation, due to the transport structure. We have

1pddτ∫∣∇Ω∣p dξ+(2αp−1−1α)∫∣∇Ω∣p dξ−∫eτγΛβ∂iΩ∣∇Ω∣p−2∂iΩ dξ+∑i=12∫∂iU‾⋅∇Ω∂iΩ∣∇Ω∣p−2 dξ+∂i(Ulin+W)⋅∇Ω∂iΩ∣∇Ω∣p−2 dξ=∑i=12∫Fi∂iΩ∣∇Ω∣p−2 dξ.(62)\frac{1}{p}\frac{d}{d\tau}\int|\nabla\Omega|^p\,d\xi+\left(\frac{2}{\alpha p}-1-\frac{1}{\alpha}\right)\int|\nabla\Omega|^p\,d\xi-\int e^{\tau\gamma}\Lambda^\beta\partial_i\Omega|\nabla\Omega|^{p-2}\partial_i\Omega\,d\xi +\sum_{i=1}^{2}\int\partial_i\overline{U}\cdot\nabla\Omega\partial_i\Omega|\nabla\Omega|^{p-2}\,d\xi+\partial_i(U^{\mathrm{lin}}+W)\cdot\nabla\Omega\partial_i\Omega|\nabla\Omega|^{p-2}\,d\xi =\sum_{i=1}^{2}\int F_i\partial_i\Omega|\nabla\Omega|^{p-2}\,d\xi. \tag*{(62)}

We remark that

∂iU‾⋅∇Ω=ζ′(∣x∣)xi∣x∣∂θΩ+ζ(∣x∣)∂ix⊥⋅∇Ω=ζ′(∣x∣)xi∣x∣∂θΩ−ζ(∣x∣)∇⊥Ω(63)\partial_i\overline{U}\cdot\nabla\Omega=\zeta'(|x|)\frac{x_i}{|x|}\partial_\theta\Omega+\zeta(|x|)\partial_i x^\perp\cdot\nabla\Omega=\zeta'(|x|)\frac{x_i}{|x|}\partial_\theta\Omega-\zeta(|x|)\nabla^\perp\Omega \tag*{(63)}

and we exploit the cancellation of the last term in (63) when multiplied by ∇Ω\nabla\Omega. Therefore (64) rewrites as

1pddτ∫∣∇Ω∣p dξ+(2αp−1−1α)∫∣∇Ω∣p dξ−∫eτγΛβ∂iΩ∣∇Ω∣p−2∂iΩ dξ=−∑i=12∫(ζ′(∣x∣)xi∣x∣∂θΩ+∂i(Ulin+W)⋅∇Ω+Fi)∂iΩ∣∇Ω∣p−2 dξ≤C∥ζ′(∣x∣)xi∣x∣∂θΩ+∂i(Ulin+W)⋅∇Ω+Fi∥Lpp+12∫∣∇Ω∣p dξ.(64)\begin{aligned} \frac{1}{p}\frac{d}{d\tau}\int|\nabla\Omega|^p\,d\xi+\left(\frac{2}{\alpha p}-1-\frac{1}{\alpha}\right)\int|\nabla\Omega|^p\,d\xi-\int e^{\tau\gamma}\Lambda^\beta\partial_i\Omega|\nabla\Omega|^{p-2}\partial_i\Omega\,d\xi =-\sum_{i=1}^{2}\int\left(\zeta'(|x|)\frac{x_i}{|x|}\partial_\theta\Omega+\partial_i(U^{\mathrm{lin}}+W)\cdot\nabla\Omega+F_i\right)\partial_i\Omega|\nabla\Omega|^{p-2}\,d\xi\\ &\le C\left\|\zeta'(|x|)\frac{x_i}{|x|}\partial_\theta\Omega+\partial_i(U^{\mathrm{lin}}+W)\cdot\nabla\Omega+F_i\right\|_{L^p}^{p}+\frac{1}{2}\int|\nabla\Omega|^p\,d\xi. \tag*{(64)} \end{aligned}

We estimate each term in the LpL^p norm in the right-hand side. By (51) and since ζ′\zeta' is bounded, we have

∥ζ′(∣x∣)xi∣x∣∂θΩ∥Lp≤∥ζ′(∣x∣)∥L∞∥∂θΩ∥Lp≤C(1+C‾1/2e2ζTmax⁡)e(a+δ)(τ−τ0)C‾1/2∥Ω0∥X.(65)\left\|\zeta'(|x|)\frac{x_i}{|x|}\partial_\theta\Omega\right\|_{L^p}\le\|\zeta'(|x|)\|_{L^\infty}\|\partial_\theta\Omega\|_{L^p}\le C(1+\overline{C}^{1/2}e^{2\zeta T_{\max}})e^{(a+\delta)(\tau-\tau_0)}\overline{C}^{1/2}\|\Omega_0\|_X. \tag*{(65)}

By the explicit expression of Ωlin\Omega^{\mathrm{lin}}, the regularity of η\eta in Proposition 2.1, and (34), we have ∥∇Ulin∥L∞≲eaτ\|\nabla U^{\mathrm{lin}}\|_{L^\infty}\lesssim e^{a\tau}. Using the Gagliardo–Nirenberg interpolation inequality, Calderón–Zygmund and (36), we have

∥∇W∥L∞≤C∥∇2W∥L4θ∥∇W∥LQ1−θ≤C∥∇curl⁡W∥L4θ∥curl⁡W∥LQ1−θ≤eaτ\|\nabla W\|_{L^\infty}\le C\|\nabla^2W\|_{L^4}^{\theta}\|\nabla W\|_{L^Q}^{1-\theta}\le C\|\nabla\operatorname{curl}W\|_{L^4}^{\theta}\|\operatorname{curl}W\|_{L^Q}^{1-\theta}\le e^{a\tau}

for a suitable choice of θ∈(0,1)\theta\in(0,1). We estimate

∥∂i(Ulin+W)⋅∇Ω∥Lp≤(∥∇Ulin∥L∞+∥∇W∥L∞)∥∇Ω∥Lp≤eaτe(a+δ)(τ−τ0)∥Ω0∥X.\|\partial_i(U^{\mathrm{lin}}+W)\cdot\nabla\Omega\|_{L^p}\le\left(\|\nabla U^{\mathrm{lin}}\|_{L^\infty}+\|\nabla W\|_{L^\infty}\right)\|\nabla\Omega\|_{L^p} \le e^{a\tau}e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X.

Finally, as regards the force appearing in (61), we use the baseline estimate (40) and (47). For the first term in the force, also by Calderón–Zygmund, we have

∥∂iU⋅∇Ω‾∥Lp≤∥∇Ω‾(⋅,τ)∥∞∥∇U(⋅,τ)∥Lp≤C∥Ω(⋅,τ)∥Lp≤C(Cδ0+CC‾eζTmax⁡)C‾1/2e(a+δ)(τ−τ0)∥Ω0∥X.\|\partial_iU\cdot\nabla\overline{\Omega}\|_{L^p}\le\|\nabla\overline{\Omega}(\cdot,\tau)\|_\infty\|\nabla U(\cdot,\tau)\|_{L^p}\le C\|\Omega(\cdot,\tau)\|_{L^p} \le C(C\delta_0+C\overline{C}e^{\zeta T_{\max}})\overline{C}^{1/2}e^{(a+\delta)(\tau-\tau_0)}\|\Omega_0\|_X.

For the second term in the force, given that D2Ω‾∈L1D^{2}\overline{\Omega}\in L^{1} since Ω\Omega is smooth and compactly supported, and by the L∞L^{\infty} bound on UU in (34) we obtain

∥U⋅∇∂iΩ‾∥Lp≤∥U∥L∞∥D2Ω‾(⋅,τ)∥Lp≤C(∥Ω∥LQ+∥Ω∥L4)(66)\lVert U\cdot\nabla\partial_i\overline{\Omega}\rVert_{L^p}\leq\lVert U\rVert_{L^\infty}\lVert D^2\overline{\Omega}(\cdot,\tau)\rVert_{L^p}\leq C\bigl(\lVert\Omega\rVert_{L^Q}+\lVert\Omega\rVert_{L^4}\bigr) \tag*{(66)}
≤C(1+C‾1/2eζTmax⁡)C‾1/2e(a+δ)(τ−τ0)∥Ω0∥X.(67)\leq C\bigl(1+\overline{C}^{1/2}e^{\zeta T_{\max}}\bigr)\overline{C}^{1/2}e^{(a+\delta)(\tau-\tau_0)}\lVert\Omega_0\rVert_X. \tag*{(67)}

By the previous estimates, (64) rewrites for p=2p=2 as

12ddτ∫∣∇Ω∣2 dξ−∫∣∇Ω∣2 dξ+∫eτγ∣Λβ/2∇Ω∣2 dξ≤C[(1+C‾1/2e2ζTmax⁡)C‾1/2e(a+δ)(τ−τ0)∥Ω0∥X]2.(68)\begin{aligned} \frac{1}{2}\frac{d}{d\tau}\int|\nabla\Omega|^2\,d\xi-\int|\nabla\Omega|^2\,d\xi+\int e^{\tau\gamma}|\Lambda^{\beta/2}\nabla\Omega|^2\,d\xi\\ &\leq C\bigl[(1+\overline{C}^{1/2}e^{2\zeta T_{\max}})\overline{C}^{1/2}e^{(a+\delta)(\tau-\tau_0)}\lVert\Omega_0\rVert_X\bigr]^2. \tag*{(68)} \end{aligned}

We now observe that, when applying the formula (64) for p=4p=4, the term related to the fractional dissipation (last term in first line) is nonnegative. This is a technical computation that can be seen, for instance, through the Caffarelli extension (∂iΩ)∗(\partial_i\Omega)^* of ∂iΩ\partial_i\Omega in {(ξ,ξ3)∈R2×R:y≥0}\{(\xi,\xi_3)\in\mathbb{R}^2\times\mathbb{R}:y\geq0\} and through the characterization of the fractional laplacian in terms of the extension as follows. By means of the divergence theorem, we compute

−∑i=12∫R2Λβ∂iΩ(ξ,τ)∣∇Ω∣2(ξ,τ)∂iΩ(ξ,τ) dx=−C∑i=12lim⁡ξ3→0+∫R2ξ3b∂ξ3(∂iΩ)∗(x,ξ3,t)∣(∇Ω)∗∣2(ξ,τ)(∂iΩ)∗(ξ,τ) dx=C∑i=12∫R+3div⁡‾(ξ3b∇‾(∂iΩ)∗(x,ξ3,t)∣(∇Ω)∗∣2(ξ,τ)(∂iΩ)∗(ξ,τ)) dx dξ3=C∑i=12∫R+3(ξ3b∇‾(∂iΩ)∗⋅∇‾∣(∇Ω)∗∣2(ξ,τ)2+ξ3b∣∇‾(∂iΩ)∗∣2∣(∇Ω)∗∣2) dξ dξ3=C∫R+3(ξ3b∣(∇Ω)∗∣22∇‾∣(∇Ω)∗∣2+ξ3b(∑i=12∣∇‾(∂iΩ)∗∣2)∣(∇Ω)∗∣2) dξ dξ3≥0,\begin{aligned} -\sum_{i=1}^{2}\int_{\mathbb{R}^2}\Lambda^\beta\partial_i\Omega(\xi,\tau)|\nabla\Omega|^2(\xi,\tau)\partial_i\Omega(\xi,\tau)\,dx \\ &=-C\sum_{i=1}^{2}\lim_{\xi_3\to0^+}\int_{\mathbb{R}^2}\xi_3^b\partial_{\xi_3}(\partial_i\Omega)^*(x,\xi_3,t)|(\nabla\Omega)^*|^2(\xi,\tau)(\partial_i\Omega)^*(\xi,\tau)\,dx \\ &=C\sum_{i=1}^{2}\int_{\mathbb{R}_+^3}\overline{\operatorname{div}}\bigl(\xi_3^b\overline{\nabla}(\partial_i\Omega)^*(x,\xi_3,t)|(\nabla\Omega)^*|^2(\xi,\tau)(\partial_i\Omega)^*(\xi,\tau)\bigr)\,dx\,d\xi_3 \\ &=C\sum_{i=1}^{2}\int_{\mathbb{R}_+^3}\left(\xi_3^b\overline{\nabla}(\partial_i\Omega)^*\cdot\frac{\overline{\nabla}|(\nabla\Omega)^*|^2(\xi,\tau)}{2}+\xi_3^b|\overline{\nabla}(\partial_i\Omega)^*|^2|(\nabla\Omega)^*|^2\right)\,d\xi\,d\xi_3 \\ &=C\int_{\mathbb{R}_+^3}\left(\xi_3^b\frac{|(\nabla\Omega)^*|^2}{2}\overline{\nabla}|(\nabla\Omega)^*|^2+\xi_3^b\left(\sum_{i=1}^{2}|\overline{\nabla}(\partial_i\Omega)^*|^2\right)|(\nabla\Omega)^*|^2\right)\,d\xi\,d\xi_3\geq0, \end{aligned}

where the constant depends only on β\beta and in the last two equalities we omitted to indicate that all integrands are evaluated at (ξ,ξ3,τ)(\xi,\xi_3,\tau). Thanks to the previous estimates, (64) implies then

14ddτ∫∣∇Ω∣4 dξ−(1+12α)∫∣∇Ω∣4 dξ≤C[(1+C‾1/2e2ζTmax⁡)C‾1/2e(a+δ)(τ−τ0)∥Ω0∥X]4.(69)\frac{1}{4}\frac{d}{d\tau}\int|\nabla\Omega|^4\,d\xi-\left(1+\frac{1}{2\alpha}\right)\int|\nabla\Omega|^4\,d\xi\leq C\bigl[(1+\overline{C}^{1/2}e^{2\zeta T_{\max}})\overline{C}^{1/2}e^{(a+\delta)(\tau-\tau_0)}\lVert\Omega_0\rVert_X\bigr]^4. \tag*{(69)}

Integrating the previous inequality and (69) between τ0\tau_0 and τ\tau, and taking advantage of the fact that the second term in the right-hand side does not influence the final estimate via Gronwall, we deduce that (3.30) holds and, with p=2p=2, that also (3.31) holds.

Nonlinear theory

In this section, we complete the proof of Theorem 1.1.

Proof of Theorem 1.1. Let Ωˉ\bar{\Omega}, aa, bb, η\eta, mm, δ0\delta_0 be fixed as in the beginning of § 3.

Step 1. Construction of two distinct solutions via a limiting procedure. For k∈Nk \in\mathbb{N}, consider tk:=e−kt_k := e^{-k}. We solve the Cauchy problem for the fractional Navier–Stokes equations (NSβ)(\mathrm{NS}_{\beta}) with smooth ‘initial’ vorticity u(k)(⋅,tk)=uˉ(⋅,tk)+ulin(⋅,tk)u^{(k)}(\cdot,t_k)=\bar{u}(\cdot,t_k)+u^{\mathrm{lin}}(\cdot,t_k) and smooth forcing term ff defined in Lemma 2.3. For each kk, a short-time solution u(k)u^{(k)} on R2×(tk,Tk)\mathbb{R}^{2}\times(t_k,T_k) exists and is unique in C([tk,Tk];L2∩W1,4)C([t_k,T_k];L^{2}\cap W^{1,4}), among other function spaces, including XX.4 Furthermore, these solutions are regular enough to demonstrate energy conservation:

12∫∣u(k)∣2(x,t′) dx+∫tt′∫∣Λβ2u(k)∣2 dx ds=12∫∣u(k)∣2(x,t) dx+∫tt′∫f⋅u(k) dx ds.(70)\frac{1}{2}\int|u^{(k)}|^{2}(x,t')\,dx+\int_{t}^{t'}\int|\Lambda^{\frac{\beta}{2}}u^{(k)}|^{2}\,dx\,ds=\frac{1}{2}\int|u^{(k)}|^{2}(x,t)\,dx+\int_{t}^{t'}\int f\cdot u^{(k)}\,dx\,ds. \tag*{(70)}

Our desired second solution, violating uniqueness, should arise as k→+∞k\to+\infty. For this purpose, it is necessary to continue the solutions up to some time tˉ\bar{t} independent of kk. In particular, we will demonstrate that there exists a time tˉ>0\bar{t}>0, independent of kk, such that the solutions u(k)u^{(k)} can be continued to R2×(tk,tˉ)\mathbb{R}^{2}\times(t_k,\bar{t}), and writing u(k)=uˉ+ulin+uper,(k)u^{(k)}=\bar{u}+u^{\mathrm{lin}}+u^{\mathrm{per},(k)}, the associated vorticity in self-similar variables Ωper,(k)\Omega^{\mathrm{per},(k)} solves (2.16) in [−k,τˉ][-k,\bar{\tau}] and satisfies

∥Ωper,(k)(⋅,τ)∥X≤eτ(a+δ0),τ∈[−k,τˉ],(71)\|\Omega^{\mathrm{per},(k)}(\cdot,\tau)\|_{X}\le e^{\tau(a+\delta_{0})},\qquad\tau\in[-k,\bar{\tau}], \tag*{(71)}

where τˉ=log⁡tˉ\bar{\tau}=\log\bar{t}.

Once we have the above estimate, we may employ standard weak compactness arguments to take the limit k→∞k\to\infty and obtain a solution u=uˉ+ulin+uperu=\bar{u}+u^{\mathrm{lin}}+u^{\mathrm{per}} satisfying the estimate (71) with Ωper\Omega^{\mathrm{per}} instead of Ωper,(k)\Omega^{\mathrm{per},(k)} on the time interval (−∞,τˉ](-\infty,\bar{\tau}]. In particular, we will have

∥Ω(⋅,τ)−Ωˉ(⋅,τ)∥X≥∥Ωlin(⋅,τ)∥X−∥Ωper(⋅,τ)∥X≥Ceτa−eτ(a+δ0)>0,(72)\|\Omega(\cdot,\tau)-\bar{\Omega}(\cdot,\tau)\|_{X}\ge\|\Omega^{\mathrm{lin}}(\cdot,\tau)\|_{X}-\|\Omega^{\mathrm{per}}(\cdot,\tau)\|_{X}\ge Ce^{\tau a}-e^{\tau(a+\delta_{0})}>0, \tag*{(72)}

when τ≪0\tau\ll0, which implies non-uniqueness. To demonstrate that uu is a Leray solution of (NSβ)(\mathrm{NS}_{\beta}), we use the energy conservation (70) with initial time t=tkt=t_k. The first term on the right-hand side of (70) is simply ∥uˉ(⋅,tk)+ulin(⋅,tk)∥L22/2\|\bar{u}(\cdot,t_k)+u^{\mathrm{lin}}(\cdot,t_k)\|_{L^{2}}^{2}/2, which converges to zero as tk→0+t_k\to0^{+}. Hence, we have the energy equality

12∫∣u∣2(x,t′) dx+∫0t∫∣Λβ2u∣2 dx ds=∫0t∫f⋅u dx ds.(73)\frac{1}{2}\int|u|^{2}(x,t')\,dx+\int_{0}^{t}\int|\Lambda^{\frac{\beta}{2}}u|^{2}\,dx\,ds=\int_{0}^{t}\int f\cdot u\,dx\,ds. \tag*{(73)}

Step 2. The estimate (4.2) via Duhamel’s formula. It remains to verify (71). By continuity in XX, the estimate (71) holds on a small neighborhood of −k-k. We define τˉk>−k\bar{\tau}_k>-k to be the largest non-positive time such that (71) holds and we claim that there exists τˉ\bar{\tau} independent of kk (which will be fixed at the end of the proof) such that τˉk≥τˉ\bar{\tau}_k\ge\bar{\tau}. Therefore, in [−k,τˉk][-k,\bar{\tau}_k] we are in the position to apply Duhamel’s formula and the assumptions of Proposition 3.1 applied with δ=δ0\delta=\delta_0 and W=UperW=U^{\mathrm{per}} satisfying the requirement (3.7). We conclude that for every τ∈[−k,τˉk]\tau\in[-k,\bar{\tau}_k],

Ωper(⋅,τ)=∫−kτS[Uper](τ,s)[−(Uper+Ulin)⋅∇Ωlin+esγΛβΩlin](⋅,s) ds,(74)\Omega^{\mathrm{per}}(\cdot,\tau)=\int_{-k}^{\tau}S[U^{\mathrm{per}}](\tau,s)\left[-(U^{\mathrm{per}}+U^{\mathrm{lin}})\cdot\nabla\Omega^{\mathrm{lin}}+e^{s\gamma}\Lambda^{\beta}\Omega^{\mathrm{lin}}\right](\cdot,s)\,ds, \tag*{(74)}

and that

∥Ωper(⋅,τ)∥X≤∫−κτ∥S[Uper](τ,s)[−(Uper+Ulin)⋅∇Ωlin+esγΛβΩlin](⋅,s)∥X ds≤∫−κτe(a0+δ0)(τ−s)∥−(Uper+Ulin)⋅∇Ωlin+esγΛβΩlin∥X(⋅,s) ds.(75)\begin{aligned} \|\Omega^{\mathrm{per}}(\cdot,\tau)\|_{X} \le\int_{-\kappa}^{\tau} \|S[U^{\mathrm{per}}](\tau,s)[-(U^{\mathrm{per}}+U^{\mathrm{lin}})\cdot\nabla\Omega^{\mathrm{lin}}+e^{s\gamma}\Lambda^\beta\Omega^{\mathrm{lin}}](\cdot,s)\|_{X}\,ds \\ &\le\int_{-\kappa}^{\tau}e^{(a_{0}+\delta_{0})(\tau-s)}\|-(U^{\mathrm{per}}+U^{\mathrm{lin}})\cdot\nabla\Omega^{\mathrm{lin}}+e^{s\gamma}\Lambda^\beta\Omega^{\mathrm{lin}}\|_{X}(\cdot,s)\,ds . \tag*{(75)} \end{aligned}

We claim that the integrands in the right-hand side satisfy

∥(Uper+Ulin)⋅∇Ωlin(⋅,s)∥X+∥esγΛβΩlin(⋅,s)∥X≲es(a+2δ0).(76)\|(U^{\mathrm{per}}+U^{\mathrm{lin}})\cdot\nabla\Omega^{\mathrm{lin}}(\cdot,s)\|_{X}+\|e^{s\gamma}\Lambda^\beta\Omega^{\mathrm{lin}}(\cdot,s)\|_{X}\lesssim e^{s(a+2\delta_{0})}. \tag*{(76)}

Indeed, by the eigenfunction properties in Proposition 2.1 as well as the structure of η\eta in the variable θ\theta and elementary interpolation, we have

∥esγΛβΩlin(⋅,s)∥X≤Ces(a+γ)(∥η∥C3(supp⁡η)+∥∂θη∥C2(supp⁡η))≤Ces(a+γ).(77)\|e^{s\gamma}\Lambda^\beta\Omega^{\mathrm{lin}}(\cdot,s)\|_{X}\le Ce^{s(a+\gamma)}(\|\eta\|_{C^{3}(\operatorname{supp}\eta)}+\|\partial_{\theta}\eta\|_{C^{2}(\operatorname{supp}\eta)})\le Ce^{s(a+\gamma)}. \tag*{(77)}

By (34) and (35) applied to UperU^{\mathrm{per}}, the bound (71) which by assumption holds in [−κ,τˉk][-\kappa,\bar{\tau}_{k}], and the bounds on Ωlin\Omega^{\mathrm{lin}} in terms of ea0se^{a_{0}s}, we estimate all the terms composing ∥(Uper+Ulin)⋅∇Ωlin(⋅,s)∥X\|(U^{\mathrm{per}}+U^{\mathrm{lin}})\cdot\nabla\Omega^{\mathrm{lin}}(\cdot,s)\|_{X} as

∥(Uper+Ulin)⋅∇Ωlin∥Lp≤(∥Uper∥L∞+∥Ulin∥L∞)∥Ωlin∥Lp≤C(∥Ωper∥X+∥Ωlin∥X)∥Ωlin∥C2(supp⁡η)≤Ce2a0s\begin{aligned} \|(U^{\mathrm{per}}+U^{\mathrm{lin}})\cdot\nabla\Omega^{\mathrm{lin}}\|_{L^{p}}\le(\|U^{\mathrm{per}}\|_{L^{\infty}}+\|U^{\mathrm{lin}}\|_{L^{\infty}})\|\Omega^{\mathrm{lin}}\|_{L^{p}} \\ &\le C(\|\Omega^{\mathrm{per}}\|_{X}+\|\Omega^{\mathrm{lin}}\|_{X})\|\Omega^{\mathrm{lin}}\|_{C^{2}(\operatorname{supp}\eta)}\le Ce^{2a_{0}s} \end{aligned}
∥(∂θUper+∂θUlin)⋅∇Ωlin∥Lp≤(∥∂θUper∥L∞+∥∂θUlin∥L∞)∥∇Ωlin∥Lp≤Ce2a0s\|(\partial_{\theta}U^{\mathrm{per}}+\partial_{\theta}U^{\mathrm{lin}})\cdot\nabla\Omega^{\mathrm{lin}}\|_{L^{p}}\le(\|\partial_{\theta}U^{\mathrm{per}}\|_{L^{\infty}}+\|\partial_{\theta}U^{\mathrm{lin}}\|_{L^{\infty}})\|\nabla\Omega^{\mathrm{lin}}\|_{L^{p}}\le Ce^{2a_{0}s}
∥(Uper+Ulin)⋅∂θ∇Ωlin∥Lp≤(∥Uper∥L∞+∥Ulin∥L∞)∥∂θ∇Ωlin∥Lp≤Ce2a0s\|(U^{\mathrm{per}}+U^{\mathrm{lin}})\cdot\partial_{\theta}\nabla\Omega^{\mathrm{lin}}\|_{L^{p}}\le(\|U^{\mathrm{per}}\|_{L^{\infty}}+\|U^{\mathrm{lin}}\|_{L^{\infty}})\|\partial_{\theta}\nabla\Omega^{\mathrm{lin}}\|_{L^{p}}\le Ce^{2a_{0}s}

for p=Q,2,4p=Q,2,4 and

∥(∇Uper+∇Ulin)⋅∇Ωlin∥Lp+∥(Uper+Ulin)⋅∇2Ωlin∥Lp≤(∥∇Uper∥Lp+∥∇Ulin∥Lp)∥∇Ωlin∥L∞+(∥Uper∥L∞+∥Ulin∥L∞)∥∇2Ωlin∥Lp≤Ce2a0s\begin{aligned} \|(\nabla U^{\mathrm{per}}+\nabla U^{\mathrm{lin}})\cdot\nabla\Omega^{\mathrm{lin}}\|_{L^{p}}+\|(U^{\mathrm{per}}+U^{\mathrm{lin}})\cdot\nabla^{2}\Omega^{\mathrm{lin}}\|_{L^{p}} \\ &\le(\|\nabla U^{\mathrm{per}}\|_{L^{p}}+\|\nabla U^{\mathrm{lin}}\|_{L^{p}})\|\nabla\Omega^{\mathrm{lin}}\|_{L^{\infty}}+(\|U^{\mathrm{per}}\|_{L^{\infty}}+\|U^{\mathrm{lin}}\|_{L^{\infty}})\|\nabla^{2}\Omega^{\mathrm{lin}}\|_{L^{p}} \\ &\le Ce^{2a_{0}s} \end{aligned}

for p=2,4p=2,4.

Therefore, since our choice of δ0\delta_{0} satisfies δ0≤14min⁡{γ,a0}\delta_{0}\le\frac{1}{4}\min\{\gamma,a_{0}\}, the previous inequalities imply the validity of (76). We obtain then from (75) and these estimates that

∥Ωper(⋅,τ)∥X≤Ce(a0+δ0)τ∫−∞τeδ0s ds≤Ce(a0+2δ0)τˉ\|\Omega^{\mathrm{per}}(\cdot,\tau)\|_{X}\le Ce^{(a_{0}+\delta_{0})\tau}\int_{-\infty}^{\tau}e^{\delta_{0}s}\,ds\le Ce^{(a_{0}+2\delta_{0})\bar{\tau}}

for every τ∈[−κ,τˉk]\tau\in[-\kappa,\bar{\tau}_{k}]. Choosing finally τˉ\bar{\tau} so that Ceδ0τˉ<1/2Ce^{\delta_{0}\bar{\tau}}<1/2, where CC is the constant appearing in the right-hand side of the previous formula, we find that τˉk≥τˉ\bar{\tau}_{k}\ge\bar{\tau}.

4.0.1. Acknowledgments. DA was supported by NSF Postdoctoral Fellowship Grant No. 2002023 and Simons Foundation Grant No. 816048. MC was supported by the SNSF Grant 182565.

References

School of Mathematics, Institute for Advanced Study, 1 Einstein Dr., Princeton, NJ 08540, USA

Email address: [email protected]

EPFL SB, Station 8, CH-1015 Lausanne, Switzerland

Email address: [email protected]

References

  1. [1]Dallas Albritton, Elia Brué, Maria Colombo, Camillo De Lellis, Vikram Giri, Maximilian Janisch, and Hyunju Kwon. Instability and nonuniqueness for the 2d Euler equations in vorticity form, after M. Vishik, 2021.arxiv.org/abs/2112.04943
  2. [2]Dallas Albritton, Elia Brué, and Maria Colombo. Gluing non-unique Navier-Stokes solutions. arXiv preprint arXiv:2209.03530, 2022.arxiv.org/abs/2209.03530
  3. [3]Dallas Albritton, Elia Brué, and Maria Colombo. Non-uniqueness of Leray solutions of the forced Navier-Stokes equations. Annals of Mathematics, 196(1):415 – 455, 2022.arxiv.org/abs/2112.03116
  4. [4]Claude Bardos, Yan Guo, and Walter Strauss. Stable and unstable ideal plane flows. volume 23, pages 149–164. 2002. Dedicated to the memory of Jacques-Louis Lions.
  5. [5]Alberto Bressan and Ryan Murray. On self-similar solutions to the incompressible Euler equations. J. Differential Equations, 269(6):5142–5203, 2020.DOI
  6. [6]Alberto Bressan and Wen Shen. A posteriori error estimates for self-similar solutions to the Euler equations. Discrete Contin. Dyn. Syst., 41(1):113–130, 2021.arxiv.org/abs/2002.01962
  7. [7]Tristan Buckmaster and Vlad Vicol. Nonuniqueness of weak solutions to the Navier-Stokes equation. Ann. of Math. (2), 189(1):101–144, 2019.arxiv.org/abs/1709.10033
  8. [8]Antonio Córdoba and Diego Córdoba. A maximum principle applied to quasi-geostrophic equations. Comm. Math. Phys., 249(3):511–528, 2004.DOI
  9. [9]Maria Colombo, Camillo De Lellis, and Luigi De Rosa. Ill-posedness of Leray solutions for the hypodissipative Navier-Stokes equations. Comm. Math. Phys., 362(2):659–688, 2018.
  10. [11]Luigi De Rosa. Infinitely many Leray-Hopf solutions for the fractional Navier-Stokes equations. Comm. Partial Differential Equations, 44(4):335–365, 2019.arxiv.org/abs/1801.10235
  11. [15]Jean Leray. Sur le mouvement d’un liquide visqueux emplissant l’espace. Acta Math., 63(1):193–248, 1934.DOI
  12. [16]Zhiwu Lin and Chongchun Zeng. Unstable manifolds of Euler equations. Comm. Pure Appl. Math., 66(11):1803–1836, 2013.arxiv.org/abs/1112.4525
  13. [17]Zhiwu Lin and Chongchun Zeng. Corrigendum: Unstable manifolds of Euler equations [mr3103911]. Comm. Pure Appl. Math., 67(7):1215–1217, 2014.DOI
  14. [18]Frank Merle, Pierre Raphael, Igor Rodnianski, and Jeremie Szeftel. On the implosion of a three dimensional compressible fluid. arXiv preprint arXiv:1912.11009, 2019.arxiv.org/abs/1912.11009
  15. [19]Sung-Jin Oh and Federico Pasqualotto. Gradient blow-up for dispersive and dissipative perturbations of the Burgers equation. arXiv preprint arXiv:2107.07172, 2021.arxiv.org/abs/2107.07172
  16. [20]Misha Vishik. Instability and non-uniqueness in the Cauchy problem for the Euler equations of an ideal incompressible fluid. Part I, 2018.arxiv.org/abs/1805.09426
  17. [21]Misha Vishik. Instability and non-uniqueness in the Cauchy problem for the Euler equations of an ideal incompressible fluid. Part II, 2018.arxiv.org/abs/1805.09440
  18. [22]V.I. Yudovich. Non-stationary flow of an ideal incompressible liquid. USSR Computational Mathematics and Mathematical Physics, 3(6):1407–1456, 1963.DOI

Paper details

vlnc:W_njrm9t8e
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

Not recorded for this paper

View statement
Looking for related papers…
Contents